Three essays of financial analysts

... distortions in financial analysts incentives in recent years, instead o f a m ore bullish position by analysts in the later part o f my sam ple period A verage characteristics o f analysts are... permission of the copyright owner Further reproduction prohibited without permission C hapter III Sum m ary Statistics of the IBES R ecom m endation D atabase 113 Sum m ary Statistics of All-star... Analyst P rofession 146 10 Predicting D eparture from Profession G iven A nalyst Status 149 11 The Effect o f Past Performance and R isk-taking Behavior on Leave from Profession

INFORMATION TO USERS This manuscript has been reproduced from the microfilm m aster. UMI films the text directly from the original or copy submitted. Thus, som e thesis and dissertation copies a re in typewriter face, while others may be from any type of computer printer. The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, th ese will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. Oversize materials (e.g., m aps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand corner and continuing from left to right in equal sections with small overlaps. ProQ uest Information and Learning 300 North Z eeb Road, Ann Arbor, Ml 48106-1346 USA 800-521-0600 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R e p ro d u ce d with p erm ission of th e copyright ow ner. F u rth er reproduction prohibited w ithout perm ission. T H R E E ESSA Y S ON FIN A N C IA L A N A LY STS By Xi Li Dissertation Subm itted to the Faculty o f the G raduate School o f V anderbilt U niversity in partial fulfillm ent o f the requirem ents for the degree o f D O C TO R O F PH IL O SO PH Y in M anagem ent August, 2002 Nashville, T ennessee A pprove^: Date: iI /f l 1/c-2^ f h ‘ ) hL 9/jf/o2 < /JJj% 1 ^ ■ f Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3058711 C opyright 2002 by Li, Xi All rights reserv ed . ___ ® UMI UMI M icroform 3058711 C opyright 2002 by P ro Q u e st Information and L earning C om pany. All rights reserved. T his m icroform edition is p ro tected ag a in st unauthorized copying u n d e r Title 17, United S ta te s C ode. P roQ uest Inform ation a n d Learning C o m p an y 300 North Z e e b Road P .O . Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C o p y rig h t © 2002 by Xi Li A ll R ights Reserved Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. To M y Parents, Jiannan Li and Z huangping Sun and My wife, lin g M a iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A C K N O W LED G EM EN TS This research project was successfully completed thanks to m any persons who helped m e at various stages. I thank m y dissertation com m ittee m em bers, N ick Bollen, Paul C haney, C raig Lewis, Hans Stoll, and especially my dissertation chairm an, Ronald M asulis, for providing precious advice and support. I also appreciate the valuable help of Bruce C ooil and Christoph Schenzler. I also thank the financial support of the D issertation Enhancem ent G rant from Vanderbilt U niversity and the 2001 AAH A ccepted D issertation Proposal G rant o f the Financial M anagem ent A ssociation and Am erican A ssociation of Individual Investors. I am also grateful to the support and encouragem ent o f m y father, Jiannan Li, my m other, Z huangping Sun, and my lovely w ife, Jing Ma, through this long effort. W ithout them , I cannot im agine that I can finish this long adventure. iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T A B L E OF C O N T E N T S Page D E D IC A T IO N .......................................................................................................................................iii A C K N O W L E D G E M E N T S ............................................................................................................... iv LIST O F T A B L E S .............................................................................................................................. vii LIST O F F IG U R E S ............................................................................................................................. ix C hapter I. PER FO R M A N C E A N D BEH A V IO R O F IN D IV ID U A L FIN A N C IA L A N A L Y S T S ....................................................................................................1 Introduction.......................................................................................................................... 1 Difference from Previous Literature.............................................................................3 Data........................................................................................................................................ 8 Experim ental D esign ....................................................................................................... 14 Em pirical R esults.............................................................................................................20 C onclusions....................................................................................................................... 48 R eferences......................................................................................................................... 51 II. W ILL PA ST LEA D ER S STILL LEA D ? PER FO R M A N C E PER SISTEN C E O F FIN A N C IA L A N A L Y S T S ........................................................... 55 Introduction....................................................................................................................... 55 Sam ples and M ethodology ............................................................................................ 61 Tw o-Period Perform ance Persistence........................................................................ 70 M ulti-Period Perform ance Persistence......................................................................88 C onclusions....................................................................................................................... 95 R eferences......................................................................................................................... 98 III. CA REER C O N C E R N S O F EQUITY A N A LY STS: C O M PEN SA TIO N , T E R M IN A TIO N , A N D PE R FO R M A N C E ...................................................................101 Introduction ..................................................................................................................... 101 Related L iteratu re........................................................................................................... 108 Sam ple, R ankings, and Perform ance M easurem ent.............................................. 111 Em pirical A nalysis........................................................................................................ 123 C onclusions..................................................................................................................... 153 R eferences........................................................................................................................ 156 V Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendix A. C R E A T IN G EM PIRICA L F A C T O R S ..........................................................................159 vi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST O F TABLES C hapter I Table Page 1. S u m m ary Statistics of R ecom m endations...........................................................................11 2. J e n s e n ’s A lpha o f Factor R egression For Test P ortfolios............................................... 18 3. P erform ance o f Analysts as a G ro u p .................................................................................... 22 4. P erform ance as a Group: O th er Factor M odels................................................................ 26 5. Perform ance as a Group: Portfolios Rebalanced L ater Than R ecom m endation D ate............................................................................................................. 29 6. C ross-sectional Distribution o f Individual Analyst Perform ance.................................. 37 7. C ross-sectional D eterm inants o f A nalyst Perform ance, R isk T aking Behavior, and A ggressiveness....................................................................... 43 C hapter II 1. S um m ary Statistics of R ecom m endation D atabase...........................................................63 2. T w o -p erio d Perform ance Persistence over the W hole Sample P eriod.........................71 3. P ersistence T est of T w o-period Perform ance by Pairs o f C onsecutive Subperiods........................................................................................................... 75 4. C on tin g en cy Table of W inners and Losers over the W hole Sample P eriod................78 5. R isk-adjusted Performance o f D ecile Portfolios C reated A ccording to Prior-period R isk A djusted Perform ance...................................................82 6. R isk-adjusted Performance o f P ortfolios Created A ccording to Prior-period R aw Return Perform ance...................................................... 85 7. B uy-and-hold Returns o f P ortfolios Created A ccording to Previous Period Raw Return Perform ance............................................... 87 8. P ersistence T est for M ulti-period Perform ance..................................................................92 vii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C hapter III 1. Sum m ary Statistics of the IBES R ecom m endation D atabase........................................ 113 2. Sum m ary Statistics of All-star R anking..............................................................................118 3. P redicting Institutional Investor A ll-stars........................................................................... 126 4. Predicting Institutional Investor A ll-stars G iven Analyst Status in the Prior Y ear.......................................................................................................................129 5. The E ffect o f Past Performance and R isk-taking B ehavior on the Institutional Investor A ll-A m erican S ta tu s...........................................................134 6. Predicting W all Street Journal A ll-stars..............................................................................137 7. Predicting W all Street Journal A ll-stars G iven A nalyst Status in the P rior Y e a r....................................................................................................................... 140 8. The E ffect o f Past Performance and R isk-taking Behavior on the W all Street Journal A ll-A m erican S ta tu s .............................................................144 9. Predicting D eparture from Analyst P rofession ..................................................................146 10. Predicting D eparture from Profession G iven A nalyst Status........................................ 149 11. The Effect o f Past Performance and R isk-taking Behavior on Leave from Profession...................................................................................................... 152 viii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST O F FIGURES C hapter I Figure 1. Page A bnorm al Perform ance o f A nalysts around R ecom m endation date........................... 32 C hapter II !. CD Fs Illustrating O ne-sided Tw o-sam ple K-S T e st.......................................................90 ix Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C H A PT E R I PE R FO R M A N C E A N D BEHAV IO R O F IN D IV ID U A L FIN A N C IA L ANALYSTS Introduction A cadem ic researchers are div id ed on the question o f w hether following recom m endations o f analysts generates superior returns. Evidence follow ing the first study by C ow les (1933) suggests th at analysts do not exhibit superior perform ance. Yet. o ther researchers find both a strong event-period abnormal return w hen recom m endations are revised and a significant post-event return drift that lasts a m onth or even longer [Barber, Lehavy, M cN ichoIs, and T ruem an (2001), Elton, G ruber, and G rossm an (1986). and W omack (1996)]. Post-event return drift is evidence against m arket efficiency. Although a finding o f abnorm al returns is usually attributed to sam ple lim itations, inaccurate perform ance m easurem ents, and insufficient risk adjustm ents, evidence in favor of m arket efficiency is subject to the sam e problem s.1 To shed new light on the research on analyst recom m endations, this article pursues three lines o f inquiry. It first evaluates the perfom iance o f recom m ended buy and sell portfolios o f individual analysts. T he study of individual an aly sts’ portfolio recom m endations is facilitated by a new source of data from Institutional Brokers Estim ate System (IB E S ) that includes a m ore com prehensive set o f brokerage firm s and individual financial analysts than p reviously available. W ith m ore accurate m easurem ent 1 S e e D im so n a n d M a rsh (1 9 8 4 ) an d W o m a c k (1 9 9 6 ) fo r a c o m p re h e n s iv e r e fe r e n c e o n e a rly lite ra tu re . P ra c titio n e rs a ls o p ro v id e e v id e n c e to th is c o n tr o v e r s y . F o r e x a m p le , a re c e n t s tu d y b y R isk M e tr ic s G ro u p s h o w s that a n a ly s ts p e r fo r m b a d ly o n a r is k - a d ju s te d b a s is (B ro w n (2 0 0 1 )). 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. o f analyst perform ance an d extensive risk adjustments, I find that the equally w eighted portfolios o f individual an aly sts’ recom m ended portfolios generate significant abnorm al returns. T he abnorm al returns, for both buy and sell recom m endations, are insensitive to the various factor m odels and risk adjustm ents used. Individually, about 10% o f analysts significantly outperform benchm arks in their buy portfolios, and 6% o f analysts significantly outperform in their sell portfolios. About 3% o f analysts significantly underperform benchm arks for buys o r sells. D ecom position o f the abnorm al perform ance reveals that it is generated m ainly w ithin an event window starting at tw o trading days before the recom m endation dates until about five trading days later, w ith no significant post-event return drift. The disappearance o f return d rift is m ostly due to more complete risk adjustm ents. The gradual disappearance o f the inform ation content in recom m endations also highlights gradual inform ation release to a w ider group o f investors, a com m on industry practice. T his practice and the strong, short-term nature o f abnormal perform ance by analysts is related to R egulation FD w hich currently only requires synchronous inform ation release by com pany m anagem ent to all investors. If preferred investors o f analysts such as the firm ’s traders do obtain prio r inform ation about recom m endations and their public release tim e and front-run less preferred clients such as individual investors, R egulation FD may need to be extended to financial analysts. T he second objective is to provide new evidence on the cross-sectional determ inants o f analyst perform ance and the relationship betw een analyst perform ance and inform ation environm ent.2 1 find that analyst characteristics can predict the 2 C le m e n t ( 19 9 9 ) a n d J a c o b , L y s , a n d N e a le ( 1 9 9 9 ) e x a m in e the d e te r m in a n ts o f a c c u r a c y o f a n a ly s t e a r n in g s f o re c a s ts . F ra n c is a n d S o f f e r ( 1 9 9 7 ) a n d S tic k e l (1 9 9 5 ) in v e s tig a te c r o s s - s e c tio n a l d e te r m in a n ts o f 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. perform ance differences o f individual analysts’ recom m ended portfolios. Individual analyst perform ance improves significantly with the num ber o f recom m endations issued and w ith the size o f their brokerage firm s. The num ber o f stocks covered also has a significantly positive, but concave relationship with perform ance. T he optimal num ber of stocks is betw een 12 and 13. A dditional evidence suggests that Institutional Investor (II) A ll-A m erican status and the size o f the companies they cover have little power in predicting analyst performance. T h e third goal is to investigate the effect of analyst career concerns on their behavior. Scharfstein and Stein (1 9 9 0 ), Prendergast and Stole (1996), and Zwiebel (1995) all suggest that agents’ career c o n cern s should affect their behavior. They predict that some agents will stay with the herd w hile others will be m ore aggressive. I find that AllA m erican analysts who have m ore reputation capital tend to recom m end more conservative portfolios and deviate significantly less often from the portfolios recom m ended by the representative analyst. Other characteristics also affect their behavior. For exam ple, analysts co v erin g large firms or m ore stocks tend to select less risky portfolios and analysts in larg e brokerage firms or m aking m ore frequent recom m endations tend to recom m end more risky portfolios. D ifference from Previous L iterature T his article is very different from previous studies. In the first part of perform ance evaluation, I im prove on all three aspects that are the focus o f the controversy about analyst perform ance: Sam ple lim itation, insufficient risk adjustm ents, and inaccurate e v e n t r e tu r n s a n d lo n g -ru n p e rfo rm a n c e o f in d iv id u a l re v isio n s a n d r e c o n f ir m a tio n s o f re c o m m e n d a tio n s a n d e a r n in g s fo re c a s ts . S in c e th is a rtic le e x a m in e s the p e rfo rm a n c e o f p o r tf o lio s re c o m m e n d e d b y fin a n c ia l 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. perform ance m easurem ents. First, I u se a comprehensive d ata set and examine the m ost recent tim e period. The IBES database used has much m ore com prehensive coverage o f brokerage firm s and analysts than a n y database used in the previously published research. It includes m any more analysts from sm aller brokerage firm s than large databases such as First C all and Zacks. It also includes im portant brokerage firm s such as Merrill Lynch, Goldm an Sachs, and Donaldson, L u fk in , & Jenrette that are not in Zacks. R ecom m endations from these three firm s compose about 10% o f all the recom m endations.3 A nother advantage is that its time period is the 1990s, “The A ge o f Analysts” . Few previous studies have exam ined this period w hile the influence and bias of analysts have both increased trem endously during this period. S econd, this study provides a n um ber of im provem ents to the research design for evaluating analyst perform ance. O ne im provem ent incorporates the recent advances in long-run perform ance evaluation literatu re [Brav, Geczy, and G om pers (2000), D aniel, Grinblatt, T itm an, and W erm ers (1 9 9 7 ), and Eckbo et al. (2000)]. W ith more careful risk adjustm ent, I obtain m ore accurate ev id e n c e on market efficiency as it pertains to lo n g ­ term perform ance. I also evaluate the perform ance o f both equal- and value-w eighted analyst p o rtfo lio s.4 A second im provem ent is to employing the m ethodology in the recent mutual fund perform ance literature to evaluate individual a n a ly sts’ recom m endations on a daily basis, which allow s more e fficien t coefficient estim ates [Bollen and Busse (2001) and B usse (1999)]. Daily data can a lso dem and a shorter tim e series for individual a n a ly sts, m y e v id e n c e is c o m p le m e n ta ry to t h e p re v io u s studies. ! 10% is a c c o r d in g to th e IB E S d a ta b ase. T h is p e rc e n ta g e will be e v e n la r g e r c o m p a re d to th e Z a c k s d a ta b a s e b e c a u s e Z a c k s d o e s n o t o ffer the r e c o m m e n d a tio n from a s m a n y s m a ll e r b ro k e ra g e firm s a s IB E S . 4 P re v io u s lite r a tu r e in v e s tig a te s the p e r fo r m a n c e o f c ith e r value- o r e q u a l- w e ig h te d p o rtfo lio s. S o m e d is a g re e m e n t e x is ts a s to w h e th e r valu e- o r e q u a l- w e ig h te d p o rtfo lio s a rc th e b e s t c h o ic e for te s ts o f p e rfo rm a n c e o v e r lo n g h o riz o n s . T o test fo r a b n o r m a l p e rfo rm a n ce , a n e q u a lly w e ig h te d p o rtfo lio is m o re 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. analysts and thus reduce potential survivorship bias. M onthly data are also used for the purpose o f corroboration. The third im provem ent is m ore frequent updating o f the m atching p ortfolios and factors. W hile m ost existing research updates the factor portfolios annually, I construct book-to-m arket and eam in g s/p rice factors or m atching portfolios quarterly, and size, m om entum , and liquidity factors o r m atching portfolios monthly. T his frequent updating should im prove the accuracy o f risk adjustment benchm arks. The fourth improvement is to m easure analyst perform ance more precisely than existing studies that focus on long-run perform ance. B ecause analysts may revise their recom m endations within weeks or m onths after the original recom m endation, I keep the stocks in the analyst portfolio until analysts revise their recom m endations. Previous studies follow s recom m endations o f analysts for an arbitrary holding period such as 6 or 12 months, usually because they lack recom m endation revision dates. This type o f assum ption co u ld m isrepresent analyst perform ance.5 M y experim ental design is also advantageous com pared to studies ex am in in g the event effect o f recom m endation revisions b ecau se 1 can exam ine the post-event return drift flexibly, and com pare the magnitude o f event-period abnorm al returns and post-event return drift. This study o f recom m ended portfolios is also o f interest because this is how brokerage houses suggest that custom ers use their recom m endations [Elton et al. (1986), and Jasen (2001)]. T he m o st important im provem ent in experim ental design is the study of individual an a ly sts’ recom m ended portfolios. The existing research has exam ined re a so n a b le . T o a s s e s s th e w ealth e ffe c t o n in v e s to r s o f fo llo w in g re c o m m e n d a tio n s , v a lu e -w e ig h te d p o rtfo lio s a re m o r e a p p ro p ria te . 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. recom m endations only at the aggregate level, or at best, at the brokerage firm level, partly because they lack a database with com prehensive coverage o f brokerage firms and analysts. The inability to identify good analysts significantly impairs the value o f this research. First, it is im possible even for institutional investors to hold portfolios o f all the stocks recom m ended by a single brokerage firm, but even individual investors can generally trade on the recom m ended portfolios o f individual analysts at low transaction costs. Second, as Barber et al. (2000) point out, an investm ent strategy based on recom m endations would be m ore profitable if good perform ers could be identified so that only their recom m endations are fo llow ed/’ 7 As is said in the mutual fund industry, “Buy the m anagers, not the fund” [C ullen et al. (2000)]. S tudying the average perform ance of financial interm ediaries such as brokerage firms m ay be m uch less interesting because the valuable elem ent o f a sell-side research departm ent is its analysts, as in the m utual fund industry. The hiring or losing o f good analysts can affect the perform ance o f brokerage firm s. The study o f recom m ended portfolios also enables us to investigate for the first tim e a w ide range o f interesting questions such as cross-sectional distribution o f perform ance, determ inants o f individual analysts’ p ortfolio perform ance and behavior, and perform ance persistence. Lastly, I study several new questions in the first part o f this article. I exam ine characteristics o f individual analysts and their recom m ended portfolios in detail for a large sam ple o f analysts. I also investigate the perform ance o f frequently used factor 5 S tu d ie s o f lo n g -ru n p e rfo rm a n c e c o u ld u n d e re s tim a te a n a ly s t p e r fo rm a n c e if a n a ly s ts h a v e re v is e d th e ir r e c o m m e n d a tio n s e a r lie r o r la te r th a n th e c u t- o f f p e rio d . H o w e v e r , th e s e stu d ie s c o u ld a ls o o v e re s tim a te a n a ly s t p e rfo r m a n c e in th o se r e c o m m e n d a tio n s m a d e u n d e r b ia s e d in c e n tiv e s for th e s a m e re a s o n . 6 A n o th e r d ir e c tio n o f re se a rc h is in v e s tig a tio n o f w h e th e r s o m e ty p e s o f re c o m m e n d a tio n s a re m o re in f o r m a tiv e th a n th e o th e r re c o m m e n d a tio n s . L in a n d M c N ic h o ls (1 9 9 8 ) a n d M ic h a e ly a n d W o m a c k (1 9 9 9 ) a re s o m e g o o d e x a m p le s o f th is d ir e c tio n o f re se a rc h . 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. m odels w ith daily data and p ro v id e the first horse race b etw een traditional factor m odels and the m acro factor m odels u sed in Eckbo, Masulis, and N orli (2000). In addition, this article exam ines the cross-sectional perform ance differences in analysts’ recom m ended portfolios and the proportions o f good and bad analysts. The advantages o f m y sam ple and experimental d esign for the perform ance evaluation naturally extend to the second and third parts o f this article. In addition, in the second part o f this article, I stu d y the cross-sectional determ inants o f analysts’ portfolio perform ance instead o f event p eriod abnorm al perform ance. Portfolio perform ance includes both event period abnorm al perform ance and any potential post-event abnorm al perform ance generated by analyst recom m endations. It sh o u ld be a more com prehensive m easure for overall analyst perform ance. In the third part o f this article, I investigate for the first tim e the im pact o f c a re e r concerns on the behavior o f analysts with different reputation. Previous literature has only exam ined the im pact o f career concerns on the behavior o f analysts with different age or experience [C hevalier and Ellison (1998), Hong, Kubik, and Solom on (1999), and Lam ont (1995)]. In addition, I use investm ent recom m endations rather than earnings forecast data as in H o n g et al. (1999), the only existing study about the im pact o f analyst career concerns on their behavior. The article is organized as follows: Section 2 describes the sample. Section 3 discusses the econom etrics o f the factor m odels and benchm arks the perform ance o f various factor m odels. It also g ives details about the m ethodology used to form the analyst portfolios and their m atching portfolios. Section 4 presents the em pirical results. Section 5 offers concluding rem arks. A n e c e s s a ry c o n d itio n fo r id e n tify in g g o o d a n a ly s ts to b e a really p r o f ita b le s tra te g y is th a t c u r re n t g o o d p e rfo r m e rs w ill d o w e ll in the fu tu re . 7 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Data The prim ary database u sed in this paper comes from IBES. Its m ajor benefit is that it includes recom m endations from a very broad sample o f brokerage firm s and financial analysts. Even large d atab ases such as Zacks do not include im portant bulge bracket firms such as M errill L ynch, G oldm an Sachs, and D onaldson, Lufkin, & Jenrette. The IBES database includes all m ajor brokerage firms plus a large sam ple o f sm aller brokerage firms. Analysts can a lm o st alw ays be tracked even if they sw itch brokerage firm s. Various m arket participants, including professional investors, use this database. IBES has collected buy and sell recom m endations from the research reports o f financial analysts since the end o f O ctober 1993.8 The database includes both the ratings based on the system s adopted by individual brokerage firms and a standardized IBES rating. The form er are usually on a three- to five- level scale. T he IB E S-created ratings are on a uniform five-level scale; character ratings of “strong b u y ,” “buy,” “hold,” “underperform ,” and “sell” c o rresp o n d to numeric ratings from I through 5. Recom m endations with num eric ratings o f 1 are used to form the buy portfolios o f financial analysts, and the sell portfolios are formed using recom m endations w ith ratings o f 4 and 5.9 The investm ent recom m endation data are from the end o f O ctober 1993 to D ecem ber 2000. The return and accounting data are drawn from C R SP and C om pustat, respectively. Panels A and B o f T able 1 sum m arize the database. T here are 241,222 recom m endations by 7,308 financial analysts from 408 institutions in the five buy and B e c a u s e the d a te s o n th e r e s e a rc h r e p o r ts u s u a lly p re c e d e the d a te s a n a ly s ts a c tu a lly d e li v e r th e re p o rts to th e p u b lic , I th ere fo re u se “ re p o rt d a t e s ” o r “ re c o m m e n d a tio n d a te s ” f o r th e d a te s o n th e r e s e a r c h re p o rts a n d “ p u b lic a n n o u n c e m e n t d a te s ” f o r th e a c tu a l p u b lic a n n o u n c e m e n t d a te s . 8 T h is is b ecause th e re a re m a n y f e w e r n e g a tiv e r e p o rts . 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. sell recom m endation categories. In the empirical analysis, I exclude analysts with fewer than 10 recom m endations. T he rem aining sample consists o f 4,383 analysts before other restrictions are applied. Panel B indicates that favorable recom m endations are much more prevalent, consistent with o th er databases. The ratio o f strong buys to sells for the entire sam ple period is about 15-to-l. Panel B also suggests that both the n u m b er o f negative recom m endations and their percentage o f all recom m endations decline over tim e, despite the grow ing num ber of total recom m endations made each year. T his suggests that the b u y -to -sell ratio declines co n tinuously throughout my sam ple period. T o put the changes in the buy-to-sell ratio in a historical perspective, according to Zacks Investm ent R esearch, the ratio of “buy” and “strong buys” to “underperform ” and “sell” is 0.9 to 1 in 1983, 4 to 1 by the end o f the 1980s, 8 to 1 in early 1990s, and 48.2 to 1 in 1998 [Laderm an (1998)). This dram atic m onotonic decline in negative recom m endations independent o f market conditions indicates increasing distortions in financial analysts’ incentives in recent years, instead o f a m ore bullish position by analysts in the later part o f my sam ple period. A verage characteristics o f analysts are reported in Panel C .10 T h e m ean market capitalization o f stocks analysts co v er increases significantly over tim e w ith a range of $3 to $13 billion (Stkcap). T he sh arp increase in the average m arket cap is m ainly a result o f increased stock price levels during the sam ple period, as the m ean size decile o f stocks covered by analysts is generally betw een 4 and 5 (Caprk). A nalysts m ake betw een 11 to 10 S tk c a p is th e s iz e o f the s to c k s c o v e r e d b y a n a ly s ts and is m e a su re d a s th e m e a n m a r k e t v a lu e o f c o m m o n s to c k s o f th e firm s th at a n a ly sts c o v e r in a s p e c if ic y ear. I o b ta in the m a rk e t c a p w h e n th e a n a ly s ts issue th e ir r e c o m m e n d a tio n s . S ec a p p e n d ix fo r d e ta ils a b o u t how th e size d e c ile s fo r C a p r k a r e c r e a te d . D u ra is th e n u m b e r o f d a y s betw een th e f irs t a n d th e la s t re c o m m e n d a tio n in a y e a r d iv id e d b y th e to ta l n u m b e r o f re c o m m e n d a tio n s . B rk sz, o r the s iz e o f b r o k e r a g e h o u se , is m e a su re d a s th e n u m b e r o f a n a ly s ts th at b e lo n g s to it in th e IB E S d a ta b a s e w ith in a y e a r. I f a n aly sts sw itc h firm s w ith in a y e a r , th e y a re a ssig n e d th e a v e ra g e s iz e o f th e tw o b r o k e ra g e firm s . 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16 recom m endations a year (N rec), and the average tim e betw een tw o recom m endations ranges from 23 to 30 days (D ura). The statistics indicate that analysts are now m aking few er recom m endations a year, and it takes longer for them to m ake recom m endations, perhaps because analysts sim ply tend not to dow ngrade their previous positive recom m endations. A nalysts on average cover about 14-15 stocks (Nstk). The average brokerage firm em ploys betw een 30 and 50 analysts. Firm size increases over tim e (B rksz), which could be a result of absolute increase in size or the significant consolidation in the brokerage industry that took place during this tim e p e rio d .11 Panel D o f T able 1 show s that analysts have an average o f 6.3 stocks in their buy portfolios and 2.2 stocks in their sell portfolios over the sam ple period. It also presents m ean deciles of several com m on characteristics o f com panies covered by analysts, including size, book-to-m arket ratio (BM), m om entum (M O M ), share turnover (TO ), and earnings/price (EP) at the tim e of the recom m endation, categorized by type of recom m endation and y e a r.1" Decile 1 (decile 10) includes stocks w ith the largest (sm allest) m arket capitalization, highest (low est) book-to-m arket, price m om entum , trading volume, and share turnover. At the tim e o f recom m endation, each recom m ended stock is placed into a specific decile. The average decile o f each characteristic is calculated for stocks recom m ended as buys and sells each year. If all NYSE stocks were w eighted equally, the average decile would be 5.5 for all characteristics. If the average stocks in analyst portfolios have characteristic deciles fairly different from 5.5, this would 11 S o m e s ta tis tic s for 1993 a r e m is s in g b e ca u se th e d a ta b a s e s ta rts in O c to b e r 1993. P le a s e se c a p p e n d ix fo r m o r e d e ta ils o f the d e fin itio n o f v a ria b le s . 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1 Summary Statistics of Recomm endations T a b i c 1 r e p o rts s u m m a r y sta tistic s o f r e c o m m e n d a tio n s . P a n e l A p r e s e n ts s u m m a ry s ta tis tic s re g a r d in g the s iz e o f IB E S re c o m m e n d a tio n d a ta b a s e . P a n e l B re p o rts n u m b e r o f re c o m m e n d a tio n s in e a c h c a te g o r y as p e r c e n ta g e s o f all re c o m m e n d a tio n s b y y e a r, e x c e p t th e la s t c o lu m n w h ic h re p o rts th e to ta l n u m b e r ol re c o m m e n d a tio n in e a c h y e a r. P anel C re p o rts th e a v e ra g e a n a ly s t c h a r a c te ris tic s , w h ic h i n c lu d e m ark et c a p a n d d e c ile r a n k o f m a rk e t cap o f th e s to c k s th e y c o v e rs ( S tk c a p a n d C a p r k , re s p e c tiv e ly ), th e av erag e d u r a tio n b e tw e e n tw o re c o m m e n d a tio n s (D u ra ), th e siz e o f th e i r b r o k e ra g e firm (B rk s z ), th e n u m b e r o f re c o m m e n d a tio n s th e y issu e (N re c), a n d th e n u m b e r o f s to c k s th e y c o v e r (N s lk ). P a n e l D s h o w s th e m ean o f th e n u m b e r s o f s to c k s , siz e , b o o k -to -m a rk e l (B M ), m o m e n tu m (M O M ), sh a re lu rn o v e r /T O (m o n th ly v o lu m e d iv id e d b y n u m b e r o f sh a res o u ts ta n d in g in th e p r e v io u s m o n th ), a n d e a rn in g s /p ric e ( e a r n in g s per s h a re s ta n d a r d iz e d b y p ric e in the fisc al q u a rte r e n d in g tw o q u a r te r s b e fo re ) (E P ) d e c ile s o f th e s to c k s that a n a ly s is r e c o m m e n d . M o m e n tu m is b u y -a n d -h o ld re tu rn o v e r th e p a s t 12 m o n th s , e x c lu d in g th e p re v io u s m o n th . A ll th e d e c ile s a re b a se d o n N Y S E c u to ff s . D e c ile 1 ( 1 0 ) c o n ta in s th e la rg e st ( s m a lle s t) sto ck s, s to c k s w ith h ig h e s t (lo w e s t) b o o k -to -m a rk e t, p r ic e m o m e n tu m , a n d s h a re tu rn o v e r. D e c ile s a r c re fo rm e d m o n th ly e x c e p t f o r b o o k -lo -m a rk e t a n d e a r n in g s /p r ic e , w h ic h a re re fo rm e d q u a rte rly . In th e c o lu m n s u n d e r th e se ll r e c o m m e n d a tio n s , I re p o rt th e r e s u lts o f th e tw o -s a m p le t- te s t o f the h y p o th e s is th a t the m e a n c h a r a c te r is tic s o f th e b u y and se ll s a m p le s a r e n o t s ig n if ic a n tly d if fe re n t. *** a n d ** in d ic a te th a t ls ta lis tic s a re s ig n if ic a n t a t 1% and 5 % le v e ls, r e s p e c tiv e ly . A lth o u g h th e re s u lts for th e m e d ia n s a re not r e p o rte d in th e ta b le , th e y a re very s im ila r to th a t o f m e a n s. T h e d a ta a re fro m O c to b e r 1 9 9 3 th ro u g h D ecem ber 2000. _____________________ P a n e l A : S u m m a ry S ta tis tic s o f IB E S R e c o m m e n d a tio n D a ta b a se N u m b e r o f A n a ly s ts : 7 3 0 8 N u m b e r o f B ro k e rs : 4 0 8 N u m b e r o f A n a ly s ts w ith > 10 re c o m m e n d a tio n s : 4 3 8 3 N u m b e r o f A n a ly s ts w ith > 100 re c o m m e n d a tio n s : 5 6 3 _______________________________ P a n e l B : B re a k d o w n o f th e R e c o m m e n d a tio n C a te g o r ie s b y Y ear IB E S R a tin g s Sell T o ta l 3% 15337 2 3 29521 36 2 3 30854 33 32 2 2 29734 31 37 29 1 2 30350 29 39 30 1 1 35445 40 28 ■> 1 37318 31 40 27 1 1 32663 29 36 32 2 2 241222 S tro n g B uy 2 3 4 1993 2 6% 30% 39% 2 1 994 25 33 37 1995 27 32 1996 30 1997 1998 1999 30 2 0 0 0 A v e ra g e % Panel C : M e a n C h a ra c te ris tic s o f A n a ly s ts Y ear C a p rk S tk c a p 1993 4 .3 9 3 .8 4 D u ra B rk sz N rec N s tk 1994 4 .7 3 3 .1 0 2 3 .2 8 3 0 .7 7 16.15 1 4 .4 9 1995 1996 4 .7 3 3.24 2 5 .3 5 3 2 .7 4 14.81 15.3 4 4 .8 8 3 .7 6 2 6 .7 6 3 5 .3 6 12.79 1997 5 .0 2 1 5 .5 6 4 .4 5 2 7 .5 5 4 0 .2 4 11.67 14.71 1998 4 .7 9 5 .9 4 1999 4 .3 4 2 6 .8 2 4 5 .5 1 12.04 1 3 .8 3 8 .2 1 28.61 4 7 .4 0 1 1 .6 8 1 4 .0 2 2 0 0 0 3 .7 7 12.59 3 0 .7 9 4 6 .1 0 10.74 1 4 .3 6 A v e rag e 4 .5 8 5 .6 4 2 4 .1 5 3 7 .8 6 12.7 9 1 4 .1 7 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. Table 1, continued P a n e l D . M e a n D e c ile s o f C h a ra c te ris tic s o f S to c k s A n a ly s is R e c o m m e n d B uy R e c o m m e n d a tio n s __________________________ Y ear N stk 1993 S ize BM MOM TO EP 4 .6 0 6 .9 8 5 .2 4 3.65 5 .5 9 S e ll R e c o m m e n d a tio n s N stk S iz e BM MOM TO EP 4 .2 2 5 .4 4 * * 6 .3 4 * * * 3 .8 8 5 .9 0 1994 6 .9 4.81 6 .9 6 4 .8 9 3 .5 6 5 .6 8 2.5 4 .2 3 5 .8 0 * * 6 .5 1 * * * 3.91 5 .5 7 1995 6 .9 4 .7 4 7 .0 0 4 .3 5 3 .4 3 5 .9 0 2.3 4 .0 8 5 .9 7 * * 3 .6 8 5 .9 6 1996 6 .7 5 .1 6 7 .0 6 4 .4 5 3.31 6 .2 3 2.5 4 .3 3 5 .9 9 * ’ 5 .8 2 * * * 5 7 3 * ., 3 .8 0 5 .7 9 1997 6.4 5 .2 7 7.11 4 .9 9 3 .3 7 6 .4 2 2 .0 4 .7 5 5 .9 4 * * 6 .6 6 *** 3 .8 5 6 .2 3 1998 6 .0 4 .8 9 7 .0 3 4 .6 6 3 .4 9 6 .4 3 2 .0 4.71 5 .9 1 * * 6 .7 2 * * * 3 .7 3 6 .2 6 1999 5 .8 4.21 7 .3 6 4 .2 6 3.14 6 .5 8 2 .0 4 .7 4 6 .2 4 * * 6 .4 8 * * * 3 .4 8 6.61 2 0 0 0 5 .6 3 .5 9 7 .9 2 3.85 2 .8 6 6.9 5 2 .1 4 .2 3 6 .6 6 6 .9 2 * * * 3 .2 4 6 .2 3 A v e ra g e 6.3 4 .6 6 7 .1 8 4 .5 9 3.35 6 .2 2 2 .2 4 .4 1 3 .7 0 6 .0 7 ** 5 .9 9 6 .4 0 suggest that analysts cover stocks that have very different characteristics fro m the overall market. T he evidence concerning size is consistent w ith previous evidence th at analysts usually cover large-cap stocks. The firm s in the IBES database are significantly sm aller than the sam ple in W omack (1996), w ith a m ean decile o f about 4.5. The p o ssib le reason is that W om ack (1996) looks only at th e recom m endations m ade by the top 14 AllAm erican research departm ents ranked by //. S m aller brokerage firms, such as regional firms, usually co v er much sm aller stocks. Panel D o f Table 1 also indicates that analysts overall tend to cover grow th over value stocks, w ith mean book-to-m arket decile alw ays above that of overall m arket for both buys and sells. They also recom m end stocks w ith higher book-to-m arket ratios as purchases. T his could reflect a tem poral trend by analysts to cover more g ro w th stocks. Panel D also suggests that analysts recom m end stocks w ith m om entum c lo se to overall m arket as buys and low m om entum stocks as sells. T he difference in m om entum between buys and sells is sim ilar to what is reported in W om ack (1996). The statistics also suggest that analysts cover more liquid stocks, w hich is to be expected because investors have more trading interest in those stocks. M oreover, analysts also cover more sto ck s with m edian to low earnings/price, a pattern that has becom e m ore significant in recent years. i? T o assess the statistical significance o f difference betw een the characteristics of the analyst buy and sell recom m endations, I use a tw o-sam ple t-test and a nonparam etric rank-sum test to com pare the mean and m edian characteristics o f buys and sells. Since 13 T h e a b o v e re s u lts a b o u t the c h a r a c te ris tic s o f s to c k s c o v e r e d b y a n a ly s ts a re c o n siste n t w ith Jc g a d e e s h , K im , K ris c h c , a n d L e e (2 0 0 1 ). 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the results for the median tests are sim ilar, I only report the results for the m eans in the colum ns for the sell recom m endations in Panel D. T he average book-to-m arket and m om entum o f analyst buy and sell recom m endations are significantly different. Current literature on perform ance evaluation o f financial analysts m atches recom m ended stocks only by size and industry indexes [W om ack (1996)] o r by beta [Elton et al. (1986)] to control for system atic risk. If factor models are used, the models usually include at most the four factors used in Fam a and French (1993) an d Carhart (1997). Because analysts tend to c o v er stocks with m any differing characteristics and tend to recom m end stocks w ith significantly different characteristics for th e ir buy and sell portfolios and because these characteristics can be related to system atic risk or investm ent styles that are not related to contribution o f a nalysts’ skills, it is important to m ake sufficient risk or style adjustm ents in evaluating analyst perform ance. E xperim ental Design Factor M odel Specification Portfolio perform ance is m easured using several specifications o f factor models frequently em ployed with m onthly returns. Perform ance o f factor models is also exam ined using 5 x 5 benchm ark portfolios formed on the basis of size and book-tom arket, and 4 x 4 x 4 benchm ark portfolios based on size, book-to-m arket, and m om entum . T he specifications include: a single-factor m arket model; a three-factor m odel that includes size and book-to-m arket factors [F am a and French (1993)]; a fourfactor model that adds m om entum effects [Carhart (1997)]; a five-factor m odel which 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. also includes earnings/price factor; and a six-factor m odel with an extra liquidity factor.14 A m odified version of the m acro factor m odels used in Eckbo et al. (2000) is also investigated. The models are m odified for use w ith daily data. The six-factor m odel and the m im icking macro factor m odel are found to m isprice fewer test portfolios while producing higher R-squares am o n g the specifications exam ined. The factor model is expressed as K =a. +£„ (1) ;=i R„ is the excess return on the portfolio of analyst i on day t; a i m easures the abnorm al return o f the portfolio o f analyst i and plays a role analogous to ‘Jen sen ’s (1968) alpha’ in a C A PM framework; R . is the return o f factor j on day t ; and £lt is the idiosyncratic return o f the portfolio o f analyst i on day t . T he risk-free rate of return is based on the daily U.S. 90-day Treasury bill. 15 Size, book-to-m arket, and m om entum are popular factors in current asset pricing literature. A liquidity factor m easured by share turnover is tested because B rennan and Subrahm anyam (1996) and D atar, Naik, and R adcliffe (1998) find that higher share turnover is cross-sectionally related to lower expected stock returns. Eckbo and Norli (2000) find this factor is related to the long-run perform ance of seasoned equity offerings. I use an earnings/price factor because Fama and French (1992) and Jaffe, K eim , and W esterfield (1989) find that it ex p lain s the cross-sectional variation in asset returns. 14 M y m o m e n tu m factor is b a s e d o n v a lu e -w e ig h te d re tu r n s , w h ile C a r h a r t (1 9 9 7 ) w e ig h ts re tu rn s e q u a lly . S e e A p p e n d ix fo r e x p la n a tio n o f th e f a c to r s a n d th e ir fo rm a tio n . 15 R is k - fre e re tu rn s are c a lc u la te d u s in g th e F e d e ra l R e s e r v e ’s c o n s ta n t-m a tu rity in te re s t r a te s c rie s . R e tu rn s a rc c a lc u la te d fro m the p u b lis h e d y ie ld s u s in g a h y p o th e tic a l b o n d w ith the sta te d m a tu rity a n d a c o u p o n e q u a l to th e y ie ld , thus tr a d in g a t p a r o r fa c e v a lu e . A n e n d - o f- p e r io d p r ic e on th e b ill u s in g th e n e x t d a y ’s y ie ld is first c alc u la te d . T h e p ric e is t h e n u se d to o b ta in th e ris k - fre e re tu rn s. 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. E ckbo et al. (2000) use a mode! of six m acro factors in their study o f long-run perform ance o f SEO . I adopt a modified four-factor version o f their m odel because daily data for unexpected inflation and real per capita consum ption are unavailable. The macro factors are e x cess returns o n the value-w eighted C R S P m arket index; the long-run term spread betw een T reasury bonds with 30-year and 1-year m aturities (L T S ); the short-run term spread betw een 180-day and 90-day T reasury b ills (STS); and the credit spread betw een B A A -rated and A A A -rated corporate bonds (C S ).16 O f the four factors, only m arket excess returns are m easured in the form o f returns. There can also be lim ited variation in th e raw m acro factors. To solve the above problem , I c re a te m im icking macro factors using sto c k returns follow ing Eckbo et al. (2000), except at a d a ily frequency. M im icking factors are in the form o f returns and should include the variations in the underlying factors. P erform ance o f factor m odels using raw and m im icking m acro factors is exam ined. Daily d a ta introduce one notable com plication. Scholes and W illiam s (1977) and D im son (1979) observe a nonsynchronous trading problem in stock returns that hinders regression estim atio n for individual securities. I address this problem by adding a lagged term for each factor in the m o d e l:17 K = a, + X 16 + A A - i 1+ £" (2> T h e d a ta fo r y i e l d s o n T -b iils , T - b o n d s , a n d c o rp o ra te b o n d s a r c fro m the F R E D d a ta b a s e . D im s o n ( 1 9 7 9 ) s u g g e s ts in c lu d in g a s m a n y as th re e la g s a n d th r e e lea d te rm s. T e s ti n g w ith se v e ra l c o m b in a tio n s s h o w s th a t o n ly th e firs t lag term is c o n s is te n tly s ig n if ic a n t. T h e r e s u l ts u s in g th re e la g and th re e lea d te rm s a r c q u a lita tiv e ly th e s a m e a s using o n ly th e first l a g te rm . S in c e a n a l y s t p o rtf o lio s ty p ic a lly 17 in c lu d e se v e ra l s to c k s , th e p r o b le m a s s o c ia te d w ith n o n s y n c h r o n o u s tra d in g is n o t a s s e r io u s a s for in d iv id u a l s e c u ritie s . B u ss e ( 1 9 9 9 ) fin d s sim ila r re su lts fo r d a ily r e tu r n s o f m u tu a l f u n d s . F u rth e rm o re , a n a ly s ts u s u a lly c o v e r h ig h - liq u id ity s to c k s and th ese sto c k s s h o u ld h a v e less p r o b l e m o f n o n s y n c h ro n o u s tra d in g . 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Although m ost factor m odels have been proven to work to som e e x ten t for m onthly stock returns, no evidence exists for d a ily returns. Nor has there been a direct com parison of traditional factor m odels and the m acro factor models. A ssum ing stationarity of factor loadings and risk prem ium s, the m odels imply that or, is zero for passive portfolios. Table 2 presents the num ber o f alpha estimates w ith t-statistics significant at 10% and average adjusted R -squares for various regressions. R esults for the 25 size and book-to-m arket portfolios and 64 size, book-to-m arket, and m om entum portfolios are reported as in Fam a and French (1993). Since these are pure passive portfolios, the results help establish how the various m odels perform in stan d ard tests o f perform ance evaluation using daily data. M easured by the nu m b er o f m ispriced test portfolios, three-, five- and six-factor m odels are the best am ong the traditional characteristic-based m odels. Both m acro factor m odels misprice far few er test portfolios than m ost traditional characteristic-based factor m odels, although R -squares indicate that the raw m acro factors explain less o f the time series variation o f stock returns. In the em pirical analysis, both the six-factor model and the m im icking m acro factor m odel are used as benchm ark models because they explain a large portion of tim e series variations o f stock returns and price analyst portfolios more correctly. The key results from the other factor m odels are also presented. The finding that factor m odels may m isprice a significant num ber o f test portfolios raises an issue the literature on analyst recom m endations does not address. T hat is, a m isspecified factor m odel may bias the alpha estimates o f analyst portfolios m uch as it biases passive portfolios with specific characteristics. For exam ple, if analysts possess no superior ability but rather cover and recom m end stocks according to size, 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2 Jensen’s alpha of factor regression for test portfolios T a b le 2 re p o rts the n u m b e r o f te s t p o rtfo lio s w ith s ig n if ic a n t a lp h a a n d th e a v e ra g e a d ju sted R - s q u a rc s a c ro s s all test p o rtfo lio s. T h e m o d e l is Rit = a t +bi0R F , +btiR Fj_l + £ „ w h e re R:l is the e x c e ss re tu rn o n e ith e r th e 25 test p o rtfo lio s fo rm e d o n th e b a s is o f s iz e a n d b o o k -to -m a rk e t o r th e 6 4 test p o rtfo lio s fo rm e d o n th e b a s is o f size, b o o k -to -m a rk e t, a n d m o m e n tu m . T h e v a ria b le s in Rf , in c lu d e th e e x cess r e tu rn o n th e C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t in d e x fo r C A P M ( a ). T h e siz e and b o o k -to -m a rk e t fa c to rs a rc a d d ed to C A P M to fo rm m o d e l (b ), th e F a m a -F re n e h m o d e l. M o d e l (c ), the C a rh a rt (1 9 9 7 ) m o d e l, a d d s the. re tu rn m o m e n tu m fa c to r. A n a d d itio n a l e a rn in g s /p ric e fa c to r is a d d e d in m odel (d ). M o d e l (e ) in c lu d e s the fa cto rs in m o d e l (d ) p lu s a liq u id ity fa c to r. T h e ra w m a c ro f a c to r m o d el in ( 0 in c lu d e s th e e x c e s s re tu rn on the C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t in d e x ; the d iffe re n c e b e tw e e n th e m o n th ly y ield c h a n g e s o n b o n d s ra te d B A A a n d A A A b y M o o d y ’s; th e d i f f e r e n c e betw een th e y ie ld s o f T r e a s u ry b o n d s w ith 3 0 y e a rs to m a tu rity a n d I y e a r to m a tu rity ; a n d the d i f f e r e n c e b etw een th e y ie ld s o f 1 8 0 -d a y a n d 9 0-day T r e a s u ry b ills. In th e m im ic k in g m a c ro fa c to r m o d e l (g ), the F a m a -F re n c h ( 1 9 9 3 ) m e th o d is u se d to c re a te m im ic k in g p o r tf o lio r e tu r n s fo r th e th re e m ac ro fa c to rs . A lp h a is the in te rc e p t o f th e m o d e ls e x p re sse d in p e rc e n ta g e s . T h e c o e f f ic ie n ts a re e s tim a te d u sin g o r d in a r y least sq u a re s. I o b ta in h c le ro s c e d a s tic ity -c o n s is tc n t t- s ta tis tic s to m e a s u re th e s ig n ific a n c e o f th e a lp h a s [W h ite (1 9 8 0 )]. T h e d a ta a re d a ily fro m O c to b e r 1 9 9 3 th ro u g h D e c e m b e r 2 0 0 0 . T w e n ty - f iv e te s t p o rtfo lio s A lp h a (a ) (b ) (c ) (d ) CAPM F a m a -F re c h (1 9 9 3 ) th r e e -f a c to r m o d e l C a rh a rt (1 9 9 7 ) fo u r-fa c to r m o d e l F iv e -fa c to r m odel 15 A lp h a 31 26 32 27 26 10 0 .6 9 8 0 .4 8 7 26 0 .6 1 4 ( 0 R a w M a c ro fa cto r m o d e l 2 0 .8 1 4 0 .8 2 4 0 .8 3 0 0 .8 3 4 0.601 (g ) M im ic k in g M a c ro fa c to r M o d e l 6 0 .7 6 3 (e ) S ix - f a c to r m odel 10 14 12 13 S ix ty -fo u r test p o rtfo lio s A d j. R sq. 0 .5 9 9 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A d j. R sq 0 .4 8 6 0 .6 5 7 0 .6 8 8 0 .6 9 4 book-to-m arket, and other characteristics, a m isspecified factor model m ay incorrectly suggest superior perform ance. I apply a procedure recom m ended by Daniel et al. (1997) and E ckbo et al. (2000). They recom m end com bining m atching technique [Barber and Lyon (1997)] and factor m odel analysis. A zero-investm ent portfolio is created by (1) investing in the analyst recom mended buy portfolios and shorting the m atching portfolios for buy recom m endations, or by (2) investing in the m atching portfolios and shorting the analyst recom mended sell portfolios for sell recom m endations. If m atching portfolios are created according to the characteristics of analyst portfolios, they should exhibit sim ilar tim e series properties, and thus m itigate the problem o f m isspecified factor m odels. Yet, since it is hard to obtain a perfect match and the m atching technique alone m ay not elim inate the factor exposure o f analyst portfolios, the additional factor regressions provide m ore effective risk adjustm ents. Analyst Portfolios a n d Their M atching P ortfolios For each financial analyst who m ade at least ten recom m endations between O ctober 1993 and D ecem ber 29, 2000 in the IBES recom m endation database and for w hich a buy or sell portfolio can be form ed for at least three m onths, I create both a value-weighted and an equal-w eighted portfolio using their recom m ended stocks in the • 18 specific category. Stocks enter the an aly st portfolios on the recom m endation date and are dropped at the revision date as recorded by IBES. The requirem ent o f at least ten 18 T h e sam e a n a ly s is is c o n d u c te d u sin g m o n th ly p o r tf o lio re tu rn s c a lc u la te d a s b u y -a n d -h o ld re tu rn s fro m th e d a ily re tu rn s o n th e a n a ly s t a n d th e ir m a tc h in g p o rtfo lio s w ith in e ac h m o n th . A r e q u ire m e n t o f a l le a st 1 2 m o n th ly re tu rn s is a p p lie d to a n a ly s t p o rtfo lio s fo r th e c o n sid e ra tio n o f re g re s s io n a n a ly s is , w h ic h m a y re su lt in som e s u r v iv o r s h ip b ia s . T h e re su lts a rc q u a lita tiv e ly th e sam e a n d re th u s n o t r e p o rte d . 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. recom m endations is to m inim ize the possibility that a non-financial analyst enters the database by co-authorship o r assum es the role o f a financial analyst tem porarily. The short window of 3 m onths ensures m inim um survivorship bias. R esults are largely the sam e when the requirem ent is at least six-m onth or one-year p ortfolio returns and at least 20 recom mendations. A benchmark portfolio is created for individual Financial analysts by matching the recom m ended stocks in their “buy list” and “sell list” by size, book-to-m arket, and m om entum with the 4 x 4 x 4 quartile portfolios at the time o f recom m endation. The benchm ark portfolios are rebalanced w h en ev er the original recom m endations change. E m pirical Results P erform ance o f A nalysts a s a G roup M ea su red by Alpha Table 3 presents, for both buys and sells, param eter estim ates o f the portfolios that place equal weight on the portfolios o f individual financial analysts. Three types o f portfolios are exam ined: analyst portfolios, m atching portfolios, a n d zero-investm ent portfolios, both equally w eighted (EW ) and value-weighted (V W ) portfolios. Panel A presents the results using the six-factor m odel. Panel B reports the estim ates from the m im icking macro factor m odel. In Panel A, alphas are uniform ly and significantly different from zero for both buy and sell portfolios. T he factor loadings on original analyst portfo lio s provide inform ation about the characteristics o f sto ck s covered by analysts and the difference o f characteristics between the stocks recom m ended as buys and sells, respectively. The m odel produces significant factor loadings for all six factors for m o st portfolios. The only 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. exception is that analyst sell portfolios have an insignificant factor exposure to the earnings/price factor. B oth a n a ly sts’ buy and sell portfolios have a m arket beta of one. The other factor loadings are quite different in m agnitude for buy and sell portfolios, except for loadings on the size factor. Buy portfolios have higher loadings than sell portfolios on return m om entum , eam ings/price, and share turnover, and low er loadings on book-to-m arket. T he signs o f all factor loadings are the sam e for buy and sell portfolios. Both kinds o f portfolios have positive loadings on m arket factor, book-tom arket, eam ings/price, and share turnover and negative loadings on size and return m om entum . T he zero-investm ent portfolios show superior perform ance for both buys and sells. As a group, the value-w eighted portfolios o f analysts’ buys generate 4.6% annualized returns (alpha = 0.018) and the equal-w eighted portfolios 5.7% (alpha = 0.022) annualized abnorm al returns. The same portfolios for analyst sells produce 3.6% (alpha = 0.014) and 3.3% (alp h a = 0.013) annualized abnorm al returns, respectively. A lthough Barber et al. (2001) and W om ack (1996) use very different sam ple, perform ance m easurem ents, risk adjustm ents, portfolio w eighting schem es, and portfolio creation schemes, they find abnorm al returns on analyst recom m endations sim ilar to my finding. For example, W om ack (1996) finds initial abnorm al returns o f 3.0% for buys and 4.7% for sells and Barber et al. (2001) find 4.2% for buys and 5.0% for sells. Interestingly, the abnorm al returns generated by buy recom m endations are higher than those for sell recom m endations. This apparent difference from som e prior studies is robust to the usage o f other factor m odels. A possible explanation is the inclusion o f recom m endations rated as both “underperform ” and “sell” in the analyst sell portfolios, 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. Table 3 Performance of analysts as a group T a b le 3 re p o rts th e re s u lts o n the p e rfo rm a n c e o f the p o rtfo lio th at e q u a lly w e ig h ts th e p o rtfo lio s o f in d iv id u a l a n a ly s ts . P a n e ls A a n d B p re s e n t re s u lts fo r the s ix - fa c to r m o d e l a n d fo r th e m im ic k in g m a c ro fa c to r m o d e l, re s p e c tiv e ly . T h e d e p e n d e n t v a ria b le s a re e x c e s s re tu rn s o n th e a n a ly s t p o rtfo lio s , th e ir m a tc h in g p o rtfo lio s b a se d o n siz e , b o o k -to -m a r k e t, a n d m o m e n tu m , a n d th e re tu rn s o n the z e ro - in v e s tm e n t p o rtfo lio s th a t a re lo n g th e a n a ly s tre c o m m e n d e d b u y p o rtfo lio s a n d sh o rt the m a tc h in g p o rtfo lio s fo r th e b u y list, o r a re sh o rt th e a n a ly s t- re c o m m e n d e d sell p o r tfo lio s a n d lo n g the m a tc h in g p o rtfo lio fo r the sell list. In P a n e l A , the v a ria b le s in RFl in c lu d e e x c e s s re tu rn o n th e C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t in d e x , s iz e (S M B ), b o o k -to -m a r k e t (H M L ), re tu rn m o m e n tu m (M O M ), e a rn in g s /p ric e (E P ), a n d liq u id ity (T O V ) fa c to rs. T h e m a c r o fa c to rs in P a n e l B in c lu d e the m im ic k in g p o rtf o lio re tu rn s fo r c re d it s p re a d , w h ic h is the d iffe re n c e b e tw e e n the m o n th ly y ie ld c h a n g e s on b o n d s ra te d B A A an d A A A by M o o d y ’s (C S ); th e lo n g -ru n te rm s p re a d , w h ic h is th e d iffe re n c e b e tw e e n the y ie ld s o f T re a s u ry b o n d s w ith 3 0 y e a rs to m a tu rity a n d 1 y e a r to m a tu rity (L T S ); a n d th e s h o rt-r u n te rm s p re a d , w h ic h is th e d iffe re n c e b e tw e e n th e y ie ld s o f 1 8 0 -d a y a n d 9 0 -d a y T re a s u r y b ills (S T S ). T h e m im ic k in g m a c ro fa c to rs a re c re a te d u sin g the F a m a -F re n c h (1 9 9 3 ) m e th o d . T h e e x c e s s re tu rn o n th e C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t in d ex is a ls o o n e fa c to r in th e m a c ro fa c to r m o d e ls. A lp h a is the in te rc e p t o f the m o d e ls e x p re s s e d in p e rc e n ta g e s . T h e c o e f fic ie n ts a re e s tim a te d u sin g o rd in a ry lea st sq u a re s . 1 o b ta in h e te ro s c e d a s iic ity -c o n s is te n l t-s la tis tic s to m e a su re the s ig n ific a n c e o f th e a lp h a s [W h ite (1 9 8 0 )]. A d j. R sq . is th e a d ju s te d R -s q u a re s o f th e re g re s s io n . T h e d a ta a re d a ily fro m O c to b e r 1993 th ro u g h D e c e m b e r 2 0 0 0 . P a n e l A : R e s u lts o f S ix -F a c to r M o d e l M o d e ls a t{a ) 1.04 7 5 .5 7 0 .0 5 0 .0 4 W hm l V W -B u y 0 .0 3 2 V W -M a tc h V W -Z e ro (B u y -M a tc h ) 0 .0 1 5 4 .9 3 0 .9 9 110.69 0 .0 1 8 5 .2 8 0 .0 5 5 .9 6 0 .0 1 0 .0 3 6 0 .0 1 4 1.05 7 3 .4 5 9 4 .0 4 0 .1 1 1 .0 0 0 .1 1 4 .2 6 5 .4 9 0 .0 2 2 7.31 4 .1 3 7 .2 9 0 .0 6 7 .5 4 0 .0 0 V W -S c ll 0 .0 0 3 0 .4 0 1 .0 0 4 6 .8 2 V W -M a tc h 0 .0 1 7 5.01 0 .9 9 9 7 .0 2 V W - Z c r o ( M a tc h -S e ll) 0 .0 1 4 2 .0 1 0 -0 . 2 0 E W -S e ll 0 .0 0 4 0 .5 3 0 .9 9 4 5 .9 0 E W - M a tc h E W - Z e ro (M a tc h - S e ll) 0 .0 1 7 4 .9 7 0 .9 9 9 6 .3 5 0 .0 1 3 1.78 - 0 -0 .0 5 -0 .1 6 E W -B u y E W -M a tc h E W - Z c r o ( B u y -M a tc h ) 6 .7 6 ) ^H M L ^M KT -0 . 0 .0 0 ) ^M O M ^ SMB ^^M O M ) ^E P t{b (rp ) ^TOV I ^ T O V ) A d j. R sq. 8 .0 8 0 .2 2 13.22 0 .9 6 8.71 11.94 0 .9 8 0 .1 2 13.14 8 .8 0 0 .1 1 12.08 0 .9 6 0 .9 7 0 .4 5 -0 .3 3 -1 6 .7 8 -0 .1 8 -9 .2 4 2 .1 1 -0 .2 4 -1 9 .4 3 -0 . 1 1 -8.81 0 .0 8 4 .7 8 0 .0 9 0 .6 7 -0 .0 9 -7 .9 9 -0 .0 7 -5 .3 2 0 .1 2 7 .8 3 0 .1 2 -2 5 .1 0 -3 2 .2 6 -4 .8 5 -0 .1 7 -0 . 1 1 -8 .3 9 -7 .3 9 -5 .2 8 0 .1 9 7 .3 8 5 .2 5 6 .2 4 0 .2 3 0 .1 2 -0.51 -0 .4 6 -0 .0 5 0.41 9 .9 4 -0 .4 9 -1 7 .0 4 -0 .2 7 -9 .4 5 0 .0 3 0 .7 4 0 .1 3 5 .4 2 0 .8 5 0 .2 3 12.87 -0.41 -2 8 .5 6 -0 .3 3 -2 5 .4 6 0 .0 9 4 .6 4 0 .1 1 8 .0 9 0 .9 7 -0 .1 7 -4 .4 9 0 .0 8 3.41 -0 .0 7 -2 .7 9 0 .0 6 1.97 -0 .0 3 - 1 .3 0 0 .0 7 0 .4 3 1 0 .2 2 -0 .2 6 -9 .1 6 0 .0 1 0 .1 4 0 .1 3 5 .3 2 0 .8 4 14.78 -0 .5 3 -0 .4 8 -1 8 .3 1 0 .2 7 -3 3 .3 5 -0 .3 2 -24.21 0 .0 9 4 .4 4 0 .1 1 -4 .1 8 0 .0 5 2 .1 8 -0 .0 5 -2 .1 4 0 .0 8 2 .5 0 8 .0 7 -1 .1 6 0 .0 5 1 .8 8 -0 .0 6 0 .2 0 0 .1 1 0 .0 8 -0 .0 2 0 .3 9 0 .9 7 T able 3, continued P anel B: R e s u lts o f M im ic k in g M a c ro F a c to r M o d e l Models a t(a ) ^MKT t& M K T ) ^L T S l (^L T S ) bcs t{ b c s ) h srs > Adj. Rsq. 0.92 0.97 VW-Buy VW-Match VW-Zero( Buy-M atch) 0 .0 2 4 0 .0 0 7 0 .0 1 7 3.53 1.79 4.34 0.97 0.93 0.05 106.38 0 . 0 1 159.79 0.01 8.61 0 .0 0 7 .6 6 -0.06 10.76 -0.05 1.79 -0.01 -9 .6 2 -0 .0 2 - 1 1.49 - 0.01 -4 .0 6 0 .0 0 -3.43 -3.90 -1.55 0.12 EW-Buy EW-Match EW-Zero(Buy-Malch) 0 .0 2 6 0 .0 0 6 0.021 3.43 1.03 6 .2 6 0.93 0.85 0.07 9 0 .4 5 0 .0 2 117.89 0 .0 2 15.35 0 .0 0 10.61 -0.09 14.33 -0.08 -0.26 -0.01 -1 2 .3 0 -0.02 -1 5 .2 8 -0.02 -2 .0 2 0 .0 0 -3.14 -4.46 0.41 0.90 0.94 0.24 VW-Sell -0 .0 2 2 -2 .4 0 -0 .0 0 8 -1 .3 0 VW-Match VW-Zcro(Match-Sell) 0 .0 1 4 2.01 0.70 0.73 0.03 61.11 0.02 8 5 .9 6 0 .0 2 3.45 -0.01 14.11 -0.09 14.13 -0.09 -4.78 0.01 -1 2 .0 6 - 0.07 -15.01 -0 .0 6 1.27 0.01 -14.06 -15.09 1.89 0.79 0.90 -0 .0 2 2 -2.33 EW-Sell -0 .0 0 8 -1.28 EW-Match EW-Zcro(Match-Scl 1) 0 .0 1 4 1.94 0.69 0.71 0.02 5 7 .9 2 0 .0 2 8 1 .9 9 0 .0 2 2 .5 5 -0.01 15.33 -0.10 15.98 - 0.10 -4.46 0.00 -1 2 .5 6 -1 6 .1 7 0.81 0.07 0.07 -13.59 -15.04 0.01 1.72 0.78 0.89 0.04 - 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0.05 while previous researchers use only recom m endations rated as “sell” . For example, com bining the recom m endations rated as 4 and 5, th e abnorm al return is 3.1% in Barber ct al. (2001), very close to the abnormal return o f 3.3% for the same recom m endations in my sample. T he B arber et al. abnorm al return o f 3.1 % is also lower than both the 5.0% abnorm al returns on recom m endations rated as 5 a n d the 4.2% abnorm al returns for recom m endations rated as 1 in their sample, which yields qualitatively sim ilar results to those from my sam ple. M atching seem s to be fairly effective. The factor loadings on the original portfolios of analysts and on the respective m atching portfolios are o f sim ilar magnitude, and the factor loadings on zero-investm ent portfolios have m agnitudes m uch closer to zero. M atching alone, how ever, does not seem to elim in ate the factor exposure of analyst portfolios com pletely. Factor loadings are significant for zero-investm ent portfolios, except book-to-m arket for buys, and market and share turnover for sells. The adjusted Rsquares for buys are about 0.4. Both results indicate the ineffectiveness o f matching technique alone for risk adjustm ents. Interestingly, the original sell portfolios have insignificant positive alphas, but the alphas o f the m atching portfolios are positive and significant. This interesting results, combined with the superior perform ance of zeroinvestm ent portfolios, yield evidence that the factor m odels alone may inflate the alpha estim ates o f original analyst sell portfolios. Panel B reports results for a mim icking m acro factor model. All the alphas for the buy and sell portfolios are again significantly different from zero. O riginal analyst portfolios have significant factor loadings on all the m acro factors, with positive factor loadings on m arket factors and long-run term spread, and negative loadings on credit 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. spread and short-run term spread. The exposures to market and short-run term yield are higher for buy recom m endations, and the factor loadings on the o th er factors are very sim ilar for buy and sell portfolios. Z ero-investm ent portfolios all exhibit significant superior perform ance. A bout three o f four factor loadings are significantly different from zero for all the zero-investm ent portfolios. A lpha results are qualitatively the same as for the six-factor model. Interestingly, the m agnitudes o f the abnorm al returns suggested by the alphas for the zero-investm ent buy and sell portfolios are alm ost the same, w hichever m odel is used. In contrast, there are m uch greater differences betw een abnormal returns on original portfolios o f analysts for the tw o m odels. This reduction in the extent of differences in abnorm al returns is evidence that focusing on zero-investm ent portfolios provides sufficient risk control rather than adding noise. The sensitivity analysis appears in T able 4, which presents alpha estimations for analyst portfolios, the m atching portfolios, and the zero-investm ent portfolios for various factor m odel regressions.19 A gain, all the m odels indicate superior perform ance of financial analysts in both buy and sell portfolios. In addition, there is less variation in the m agnitude o f alphas for zero-investm ent portfolios than for the original analyst portfolios, which is to be expected because the m atching technique should eliminate sizable factor exposures in the original analyst portfolios and thereby reduce the potential bias and noise that different factor m odels co u ld introduce into the alpha estimates. 19 S c h e m e s u sin g size o n ly a n d siz e a n d b o o k -to - m a rk e t o n ly to m atch y ie ld h ig h e r a b n o rm a l returns, w h ic h s u g g e s ts th e im p o rta n c e o f m o re s u ff ic ie n t r is k - a d ju s tin g m a tc h in g sc h e m e s. 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4 Performance as a group: Other factor models T a b le /?„ = a , 4 p re s e n ts a lp h a s e s tim a te d w ith v a rio u s fa cto r m o d e ls . The m odel is +bl0R F , +bu R F j _, +£a , w h e r e /?„ a re e x c e s s re tu r n s o n the a n a ly s t p o rtfo lio s , th e ir m atc h in g p o rtfo lio s b a s e d on s iz e , b o o k - to -m a rk e t, a n d m o m e n tu m , a n d the r e tu r n s o n th e z e ro -in v e s tm e n t p o rtfo lio s th a t a re lo n g th e a n a ly s t- re c o m m e n d e d b u y p o r tf o lio s and sh o rt th e m a tc h in g p o rtfo lio s for th e b uy list, o r a r c sh o rt th e a n a ly s t- r e c o m m e n d e d sell p o rtfo lio s a n d long th e m a tc h in g p o r tf o lio for the sell list. T h e v a ria b le s in R F , in c lu d e e x c e s s r e tu rn on th e C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m ark e t in d e x , s iz e a n d b o o k -to -m a rk e t fa c to rs (a ). M o d e l ( b ) adds th e r e tu r n m o m e n tu m factors. A n a d d itio n a l e a m in g s /p r ic e fa c to r is a d d e d in m o d e l (c ). T h e ra w m a c ro f a c t o r m o d e ls in clude e x c e ss re tu rn o n th e C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m ark et in d e x ; th e d iff e re n c e betw een the m o n th ly y ie ld c h a n g e s o n b o n d s ra te d B A A a n d A A A b y M o o d y ’s; th e d if f e r e n c e b e tw e e n the yields o f T re a su ry b o n d s w ith 3 0 y e a rs to m a tu rity a n d I y e a r to m a tu rity ; and th e d if f e r e n c e b e tw e e n the y ie ld s o f 1 8 0 -d a y a n d 9 0 -d a y T r e a s u ry b ills . A lp h a is th e in te rc e p t o f th e m o d e ls e x p r e s s e d in p e rce n ta g e s. T h e c o e ffic ie n ts a r e e s tim a te d u sin g o r d in a r y le a s t sq u a re s . I o b ta in h e tc ro s c e d a s tic ity -c o n s is tc n t l-statistic s to m e a su re th e s ig n ific a n c e o f the a lp h a s [W h ite ( 1 9 8 0 )]. T h e d a ta are d a ily f ro m O c to b e r 1993 th ro u g h D ecem ber 2 0 0 0 . Value-Weighted Portfolios____________________________ Equal-W eighted Portfolios M atch Z ero S e ll M a tch Z e ro (a) Alpha estim ates using Fama and French (1993) three-factor model 0.027 0 .0 1 0 0.0 1 7 -0.011 0 .0 0 2 0.013 0.031 0 .0 1 0 0 .3 4 1.90 5.47 2.71 3 .0 8 4 .8 9 4.93 -1.36 0 .0 2 0 6 .6 0 -0.011 -1.27 0.002 0.43 0.013 1.83 (b) Alpha estimates using Carhart ( i 997) four-factor model 0 .0 1 7 0.021 0.017 0 .0 4 2 0.038 0 .0 1 9 0.003 2.41 7.7 4 7.34 5.51 6.05 0.32 5 .3 0 0.017 4.78 0 .0 2 5 7 .7 6 0 .0 0 3 0.38 0.019 5.24 0.016 2.25 (c) Alpha estim ates using five-factor model 0 .0 1 7 0.032 0 .0 1 5 0.0 1 8 0.003 6.76 4 .9 3 5.28 5.01 0.40 B uy 1M a tc h Z ero Sell M a tc h Z ero B uy 0.014 2.01 0 .0 3 6 7.31 0.014 4.13 0 .0 2 2 7 .2 9 0 .0 0 4 0.53 0.017 4.97 0.013 1.78 (d) Alpha estimates using raw macro factor model 0.030 0 .0 1 2 0.0 1 8 0 .0 1 2 0.013 -0.001 4.16 2.71 4 .5 8 1.62 1.82 -0.06 0.0 3 4 3.93 0.013 1.95 0.021 6.2 3 -0 .0 0 0 -0 .0 0 0.013 1.63 0.013 1.80 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Event-period A bnorm al R e tu rn s o r Post-event R etu rn Drift The results so far su g g est superior perform ance of portfolios recom m ended by financial analysts. M ight th e abnorm al perform ance result from the event-period abnorm al returns, the post-event return drift, or a com bination o f both? T his question is im portant because the p o st-event return drift th at B arber et al. (2001), Elton et al. (1986) and W omack (1996) o b serv e is a puzzle not explained by any type o f m arket efficiency hypothesis. It is also of in terest because investors m ay not be able to obtain information about recom mendations e x a ctly on the public announcem ent dates. A crucial elem ent in identifying the source o f abnormal perform ance is when portfolio formation should start. T he analysis in 4.1 assumes that analyst portfolios are rebalanced at the closing price one day before th e dates appearing on research reports. Traditional belief is that research reports are dated several days after analysts send their reports to data vendors su ch as First Call and m ed ia services such as D ow Jones News Service [Womack (1996)]. C heng (2000) com pares the two dates using data from First C all, and finds that the dates on research reports are no later than the public announcem ent dates. In fact, they are on average four days earlier than the public announcem ent dates; the actual difference depends on the brokerage house. I conclude that portfolio returns that can be obtained by the public should be calculated assum ing the rebalance dates to be so m ew h at later than the recom m endation dates in the IBES database, rather than earlier. Furtherm ore, since brokerage firm s m ay distribute inform ation to their p referred clients before the inform ation is conveyed to the public, this practice should induce som e very short-run post-event return drift after the report dates. Using report dates to rebalance portfolios m ay inflate the significance o f the 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. trading profits and result in a false conclusion o f short-run post-event return drift. If the post-event return drift w ere long-term, the evidence w ould support a conclusion o f a real drift. Panels A and B o f T able 5 docum ent the investm ent perform ance achievable on portfolios rebalanced 5 and 15 trading days after the actual report dates. A lag o f 5 trading days corresponds on average to several trading days after the public announcem ent dates. A lag o f 15 trading days is used to test for long-run post-event return drift. In Panel A, no alpha is significant, except for the alphas produced by the Carhart (1997) m odel. T h e m agnitude o f all the alpha estim ates is significantly reduced. W hen the rebalance dates are 15 trading days later, no single alpha is significant in Panel B. Both the m agnitude and the significance o f alphas decrease sharply as the rebalance dates m ove farther aw ay from the recom m endation dates. T hese results support the conclusion that the profit is sm all if investors react after the public announcem ent dates, which is usually several days later than the report dates. T o appreciate the exact time length before abnorm al perform ance evaporates and to assess w hether analysts have released inform ation gradually and to the most preferred clients earlier and the extent o f early inform ation release. Figure 1 plots the abnorm al returns im plied by the six-factor model alphas for analyst portfolios created some tim e during the 30 trading days around the report dates. T he abnorm al returns seem to be the highest w hen buy or sell portfolios are rebalanced four trading days before report dates. A considerable drop in potential profit happens over the several trading days before report dates, which suggests that analysts m ay have given inform ation to the most preferred clients even before report dates at which they dissem inate information to o th er 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T a b le 5 Performance as a group: Portfolios rebalanced later than recommendation date P a n e ls A an d B o f T a b le 5 re p o rt the re s u lts w h e n th e p o rtfo lio s a rc a s s u m e d to b e re b alan c e d 5 a n d 15 tra d in g d a y s, r e s p e c tiv e ly , la te r th a n th e r e c o m m e n d a tio n d a te s . The m odel is R,t = cr, +hjQR Fl +bii R F l_i +£„, w h ere Rlt a re e x c e s s re tu rn s o n th e a n a ly s t p o rtfo lio s , their m a tc h in g p o rtfo lio s b a se d o n s iz e , b o o k -io -m a rk c l, a n d m o m e n tu m , a n d th e r e tu r n s o n the z e ro -in v e s tm e n t p o rtf o lio s th a t a re lo n g th e a n a ly s t-re c o m m e n d e d b u y p o rtfo lio s a n d s h o r t th e m a tc h in g p o rtfo lio s fo r th e b u y list, o r a re s h o r t th e a n a ly s t-re c o m m e n d e d se ll p o rtfo lio s a n d l o n g th e m a tc h in g p o rtfo lio fo r th e se ll lis t. T h e v a ria b le s in R F , in clu d e e x c e ss r e tu rn o n th e C R S P v a lu e - w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t in d ex , s iz e , b o o k -to -m a rk e t, an d r e tu rn m o m e n tu m fa c to rs in m o d e l (a ). M odel (b) a d d s th e c a rn in g s /p ric e fa c to rs . A n a d d itio n a l sh a re tu r n o v e r fa c to r is a d d e d in m o d e l (c ). T h e raw m ac ro f a c to r m o d e ls in clu d e e x c e s s re tu rn on the C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m arket in d e x ; th e d iffe re n c e b e tw e e n th e m o n th ly yield c h a n g e s o n b o n d s rated B A A a n d A A A b y M o o d y 's ; the d if f e r e n c e b e tw e e n the y ie ld s o f T r e a s u r y b o n d s w ith 3 0 y e a rs to m atu rity a n d 1 y e a r to m a tu rity ; and the d if f e r e n c e b e tw e e n the y ie ld s o f 1 8 0 -d a y and 9 0 -d a y T r e a s u r y b ills. T h e m im ic k in g m a c ro fa c to rs arc c re a te d u s in g th e F a m a -F re n c h ( 1 9 9 3 ) m eth o d . A lp h a is th e in te rc e p t o f th e m o d e ls e x p re s s e d in p e rce n ta g e s. T h e c o e ffic ie n ts a rc e s tim a te d u sin g o rd in a ry le a s t s q u a re s . I o b ta in h e te ro s c c d a s tic ity -c o n s is te n t t-s ta tis tic s to m e a s u r e the s ig n if ic a n c e o f the a lp h a s [ W h ite (1 9 8 0 )]. T h e d a ta a r e d a ily fro m O c to b e r 1993 th r o u g h D ecem ber 2 0 0 0 . Panel A: Results Form ing Portfolios from 5 Trading Days after Report Dates. _____________ Value-W eighted Portfolios____________________________ Equal-W eighted Portfolios_____________ B uy M a tch Z ero Sell M a tc h Z e ro B uy M a tc h Z ero S e ll M atch Z ero (a) Alpha estimates using the modified Carhart (1997) four-factor model 0 .0 2 4 0 .0 1 7 0 .0 0 7 0 .0 1 4 0 .0 1 4 0 .0 0 6 0 .0 2 6 0 .0 1 8 0 .0 0 8 0 .0 1 4 0 .019 0 .0 0 5 4 .6 2 5 .4 6 2 .1 0 1.74 1.74 0 .8 7 4 .7 3 4 .8 6 2 .5 3 1.75 5.35 0 .7 6 0 .0 0 3 0 .4 3 0 .0 2 2 4 .1 4 0 .0 1 6 4 .3 6 0 .0 0 6 1.97 0 .0 1 7 2 .0 8 0 .019 5.18 0 .0 0 2 0 .2 5 0 .0 0 3 0 .4 9 0 .0 1 9 3.91 0 .0 1 4 4 .1 4 0 .0 0 5 1.64 0 .0 1 5 1.87 0.017 5.03 0 .0 0 2 0 .3 1 0 .0 1 7 1.95 0 .0 1 3 1.91 0 .0 0 4 1.22 0 .0 1 1 1.06 0.013 1.67 0 .0 0 2 0 .2 4 (b) Alpha estimates using the five-factor model 0 .0 2 1 3 .9 6 0 .0 1 6 4 .9 9 0 .0 0 5 1.38 0 .0 1 6 1.99 0 .0 1 9 5.21 (c) Alpha estimates using the six-factor model 0 .0 1 8 3 .7 0 0 .0 1 4 4 .8 2 0 .0 0 3 1.02 0 .0 1 4 1.77 0 .0 1 7 5 .0 6 (d) Alpha estimates using the raw macro factor model 0 .0 1 5 2 .1 0 0 .0 1 2 2 .6 0 0 .0 0 3 0 .8 9 0.011 1.04 0 .0 1 2 1.66 0 .0 0 2 0 .2 3 (e) Alpha estimates using the mimicking macro factor model 0 .0 0 9 0 .0 0 7 0 .0 0 2 - 0 .0 1 1 ".0.008 0 .0 0 3 0 .0 0 9 0 .0 0 5 0 .0 0 4 -0 .0 1 1 -0.008 0 .0 0 3 1.32 1.67 0 .5 9 -1.17 -1 .3 3 0 .4 0 1.20 0 .9 9 1.14 -1 .1 4 -1.30 0 .3 7 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T ab le 5. continued Panel B: Results F orm ing P ortfolios from 15 T rading Days alter Report Dates. V alue-W eighted Portfolios______________________________ E q ual-W eighted Portfolios B uy Z e ro Z e ro S e ll M a tc h Z e ro B uy M a tc h M a tc h (a) Alpha estim ates using m odified Carhart ( 1997) four-factor model 0 .0 0 5 -0 .0 1 7 -0 .0 2 0 -0 .0 0 4 0 .0 2 3 0 .0 2 2 0 .0 1 7 0 .0 1 8 0 .0 0 5 -0 .5 2 1.50 4 .0 9 5 .4 4 1 .29 -2 .1 4 -5 .5 9 4 .1 5 4 .8 6 (b) Alpha estim ates using five-factor model 0 .0 1 8 0 .0 1 6 0 .0 0 2 -0 .0 1 9 -0 .0 2 0 0.000 S ell -0 .0 1 7 -2 .1 6 M atch Z ero -0 . 0 2 0 -5.48 - 0 .0 0 3 - 0 .3 6 0 .0 0 3 -0 . 0 2 1 4 .3 5 0 .9 2 -2 .5 3 -0 . 0 2 0 -5.34 0 .0 0 1 -0 .0 5 0 .0 1 9 3.54 0 .0 1 6 -5 .4 2 (c) Alpha estim ates using six-factor model 0 .0 1 5 0 .0 1 5 0 .0 0 0 -0 .0 1 8 -0 .0 1 8 -0 .0 0 1 0 .0 1 6 0 .0 0 1 3 .4 0 4 .9 7 0 .5 3 -2 .4 2 0 .1 8 -0 .0 1 9 -0.018 3.25 0 .0 1 5 4 .1 2 0 .0 0 2 .1 0 0 .5 4 -2 .3 2 -5.16 0.001 0 .0 1 3 0 .0 1 3 0.000 1.55 1.93 0 .1 4 -0 .0 1 5 -1 .3 6 -0.014 0 .1 4 -1.71 0 .1 7 (e) Alpha estim ates using m im icking macro factor m odel 0 .0 0 6 0 .0 0 7 -0 .0 0 1 0 .0 0 7 0 .0 0 7 0 .0 0 0 -0 .0 4 0 .8 3 1 .6 5 -0 .2 7 0 .8 2 1 .20 0 .0 0 5 0 .0 0 5 0.000 0 .0 0 7 0.0 0 7 0 .0 0 0 0 .7 0 1 .0 0 0 .7 8 1 .2 0 0 .0 1 3 .0 9 4 .7 9 0 .1 3 -2 .2 0 -5 .2 5 (d) Alpha estim ates u sing raw m acro factor m odel 0 .0 1 2 0 .0 1 2 0 .0 0 0 -0 .0 1 4 -0 .0 1 3 1 .6 8 2 .6 0 0 .0 9 -1 .3 4 -1 .7 3 -0 -0 . 0 1 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 .1 4 0 .0 0 1 relatively less preferred clients. T he m o st significant drops in potential profits o ccu r within the four trading days after the report dates for buy portfolios and one trading day after the report dates for sell portfolios. Beyond these periods, the e stim a ted alphas becom e insignificant, and all drop to less than 0.5 basis point. This ev id en ce reaffirm s that public announcem ents of recom m endations occur in these periods, and that the investor reactions result in quick incorporation o f the information co n ten t o f recom m endations into stock prices. T he gradual disappearance o f abnorm al perform ance starting ev en before the report date suggests gradual inform ation release. This type o f inform ation release has im plications for R egulation FD, w h ich requires corporate m anagers release inform ation to all the investors at approxim ately th e sam e tim e. Gradual inform ation release could result in exploitation o f less preferred investors, especially individual investors, by m ore preferred investors o f analysts or trad ers in the sam e brokerage firms. T h is is a p articular concern since there is on average a stro n g event period return at public release o f analyst recom m endations. For exam ple, if a co m p an y is upgraded to “strong b u y ,” traders in the analyst’s brokerage firm can profit at the expense o f other investors by buying the stock im m ediately prior to public releases a n d selling the stocks shortly th ereafter, bearing little risk. If there is evidence o f front-running behavior, Regulation FD m ay n eed to be extended to cover financial analysts’ reports. Interestingly, the results also su g g est that there may not be as m any layers o f preferred clients for sell recom m endations as there are for buys since the form er recom m endations are m ade public m u ch m ore quickly. This difference is understandable because a m ajor determ inant of an aly st com pensation is trading co m m issio n s generated 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. % Daily Abnormal R eturns O.O' VW-Buy 0.015 EW-Buy •VW-Sell 0 . 0 1 ■EW-Sell - 0 .0 0 5 15 -10 ■5 0 5 10 15 Figure 1. Abnormal performance of analysts around recommendation date 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. by their recom m endations. Since sells generate a sm all fraction of trading volum e com pared to buys, there is less incentive for analysts to distribute sell inform ation gradually. These results are different from those in Barber e t al. (2001), Ellon et al. (1986), and W om ack (1996). For exam ple, Barber et al. (2001) find a post-event return drift o f up to 15 days. Elton et al. (1986) observe a significant post-event return drift of 2.1% in the tw o m onths after announcem ent. The return drift found b y W om ack (1996) is the largest and persists the longest, with m agnitudes o f 2.4% and 9 .1% for time intervals o f o n e and six m onths for buys and sells, respectively. Since my p aper uses a m ore com prehensive sam p le o f brokerage firm s and applies different portfolio creation schem e, more extensive risk-adjustm ents, and a m ore accurate perform ance m easurem ent, there are several possible explanations for my different results. Instead o f creating analyst portfolios first and then pooling all the analyst portfolios, I pool all the recom m endations into o n e aggregate recom m endation portfolio. The results are qualitatively the sam e as above. I create individual analyst portfolios by assum ing a six-m onth holding period for e a ch recom mendation and find a significant but sm aller alpha than w hat is reported in sectio n 5.1 on analyst portfolio rebalanced at report release dates. I also find a shorter post-event drift. Most o f the reduction in post-event return drift is due to my more ex ten siv e risk-adjustment benchm arks. For exam ple, if I only use either the m atching m ethodology or a factor m odel regression, th e drift is m uch longer. T his highlights the importance of the riskadjustm ent benchm arks in the studies o f long-run perform ance evaluation. 33 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C ross-sectional Distribution o f Individual Analyst P erform ance H enceforth, I present o n ly analysis based on the a lp h a estim ates of zeroinvestm ent portfolios for the six -facto r model, because the results for the other factor m odels and original analyst portfolios are qualitatively the sam e. Table 6 presents the cross-sectional distributions o f the t-ratios for the six-factor m odel alphas on the zeroinvestm ent portfolios. Notice th a t the t-statistics o f the alp h as are actually information ratios, or alphas divided by the am ount o f specific risks tak en by analysts.20 Table 6 also show s the m axim um and m inim um t-statistics for each p o rtfo lio , together with p-values based on the B onferroni inequality. Results for binom ial t-tests, norm ality tests, and ttests o f the m eans o f alphas are presented. Panel A reports the results for portfolio rebalanced the day before report dates; Panels B and C are for 5 and 15 trading days after the report dates, respectively. The Bonferroni p-value is a conservative m easure that gives the upper bound on the p-values o f a jo in t test. It is a one-tailed test w ith the nu ll hypothesis that all the alphas are zero against the alternative that at least one is po sitiv e (maximum t-ratio) or negative (m inim um t-ratio). If a n y o f the N statistics for a test o f size p rejects the hypothesis, given dependent ev e n ts, the joint probability is less than or equal to the sum o f the individual probabilities, pN . It is obtained by m ultiplying the smallest of the N pvalues for the individual tests b y N. The hypothesis underlying the binomial tests, whose statistic is t-distributed, is that 5 0 percent o f the alphas are p o sitiv e.21 The normality test 20 C o m p a re d to a lp h a s , th e re arc s o m e d ra w b a c k s o f u sing in fo rm a tio n r a tio s to m e a su re in d iv id u a l a n a ly st p e rfo rm a n c e b e c a u s e th e re su lts m a y b e b ia s e d a g a in st a n aly sts w h o h a v e f e w e r sto c k s in th e ir re c o m m e n d e d p o rtfo lio s a n d thus h a v e g r e a te r s p e c ific risks. T h e re a r c n o s im ila r sta tistica l tests th at a p p ly to a lp h a , h o w e v e r. T h e b in o m ia l l-s ta tis tic is c a lc u la te d a s (0 .5 -x )/[0 .5 * 0 .5 /N )l/:, w h e re N is th e n u m b e r o f th e c ro sss e c tio n s in th e s a m p le a n d x is the f r a c tio n o f th e N alpha e s tim a te s t h a t a re n e g a tiv e . T his sta tistic ig n o re s th e c o r re la tio n a c ro s s th e a n a ly st p o r t f o li o s , th o u g h . 21 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. is based on the K olom ogorov-Sm im ov D -stalistic, and is used to determ ine w hether the distribution o f perform ance is highly skew ed o r fat-tailed, and thus significantly different from norm al. T he distributions o f t-statistics should be close to normal, g iven the sample sizes involved. T h is test is less inform ative than the above tests. Even if th e norm ality hypothesis is rejected, one cannot tell w hether it is because o f too m any m ediocre analysts or too m any good perform ers. The t-test o f the m eans of the alphas is based on the cross-sectional distribution o f estim ated alphas. Panel A o f Table 6 show s that the extrem e t-statistics are never significant for the recom m ended sell portfolios, but the m axim um t-statistics of both value-w eighted and equal-w eighted buy portfolios are significant at high confidence levels. B inom ial tstatistics and t-statistics for the m eans o f estim ated alphas are all positive an d significant for both buys and sells. Although a m ajority o f analysts are m ediocre, a sm all group o f analysts do outperform : about 10% for the buy portfolios, and 6% for the sell portfolios. To com pare, bad perform ers constitute only 3.7% o f the analyst population for both buys and sells. Thus, there are at least tw ice as m any good perform ers as bad perform ers. Interestingly, the norm ality tests indicate that only the distribution o f t-statistics for equal-w eighted buy portfolios is significantly different from the normal distribution. T hese results indicate on the one hand, the im portance of a carefully executed analyst selection procedure. A lthough the m ajority o f analysts are m ediocre, the small group o f superior analysts generates significant outperform ance for the w hole analyst group. On the o th er hand, however, it is unclear from the above results that selecting the good analysts will guarantee significant abnorm al returns. T o achieve abnorm al returns, good analyst perform ance has to be persistent. 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Panels B and C of Table 6 present the results fo r portfolio rebalancing d a te s delayed by 5 an d 15 trading days, respectively. Perform ance deteriorates more m aterially the longer the rebalancing dates are after the recom m endation dates. For a 5-trading day delay, no extrem e t-statistics are significant for any portfolios, and the binomial t-statistics are significant only for the buy recom m endations. In addition, normality test statistics are m arginally significant for only value-w eighted sell portfolios (p = 0.09). A lthough all the t-statistics for means rem ain positive and significant, they are much sm aller. M oreover, the percentage of analysts w ho significantly outperform drops sharply. T h e sam e pattern o f scarce significance is observed for a 15-trading day delay. The only statistics that are significant, how ever m arginally, are the t-statistics for the m eans for buy portfolios. Overall, analysts exhibit superior perform ance when portfo lio s are rebalanced at the recom m endation dates, but outperform ance is lim ited to a relatively sm all group o f analysts. Perform ance worsens significantly the farther the portfolio rebalancing date is m oved from the reporting date, which indicates that th e superior perform ance is mainly attributable to event-period abnorm al returns. T he results are consistent w ith the results for the equally w eighted portfolios of all individual analyst portfolios. C ross-sectional D eterm inants o f R ecom m endation Perform ance A few researchers have investigated th e announcem ent effects o f recom m endations cross-sectionally using sam ples with less com prehensive coverage [Francis and Soffer (1997) and Stickel (1995)]. T h eir results suggest that analyst characteristics can be useful in predicting d ifferences in the perform ance o f their 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. Table 6 Cross-sectional distribution of individual analyst performance T a b le 6 re p o rts the c ro s s -s e c tio n a l d is trib u tio n o f the t-s ta tis tic s o f J e n s e n ’s a lp h a s fro m th e s ix -f a c to r m o d e l. P a n e ls A , B , a n d C re p o rt th e re s u lts w h e n th e p o r tf o lio r e b a la n c e d a te s a re th e re p o rt d a te s, 5 a n d 10 tra d in g d a y s a fte r re p o rt d a le s , r e s p e c tiv e ly . T h e s ix -fa c to r m o d e l is Rjj = a i +blQR F , + bh R F + £„ , w h e re Rn is e x c e s s re tu rn on th e z e ro -in v e s tm e n t p o rtf o lio s th at a rc lo n g th e a n a ly s t- re c o m m e n d e d b u y p o rtfo lio s a n d sh o rt th e m a tc h in g p o rtfo lio s fo r the b u y list, o r a re sh o rt the a n a ly s t- re c o m m e n d e d sell p o rtfo lio s a n d lo n g th e m a tc h in g p o r tfo lio fo r th e se ll list. T h e v a ria b le s in RFl in c lu d e e x c e s s re tu rn o n th e C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t in d e x , s iz e , b o o k -to -m a r k e t, re tu rn m o m e n tu m , e a rn in g s /p ric e , a n d liq u id ity fa c to rs. T h e c o e ff ic ie n ts a re e s tim a te d u s in g o rd in a r y le a s t sq u a re s . I o b ta in h e te r o s c e d a s tic ity -c o n s is te n t ts ta tis tic s to m e a s u re th e s ig n if ic a n c e o f th e a lp h a s [W h ite (1 9 8 0 )]. T h e n u m b e rs in c o lu m n 2 -9 a re th e n u m b e r o f a n a ly s t p o rtf o lio s fo r w h ic h th e ts ta tis tic s fo r th e a lp h a s fell w ith in th e ra n g e o f v a lu e s in d ic a te d in the to p ro w . T h e B o n fe rro n i p -v a lu e is th e m a x im u m o f m in im u m o n e -ta ile d p -v a lu e fro m th e t-d is tr ib u tio n , a c ro s s all th e a n a ly s t p o rtfo lio s , m u ltip lie d by th e n u m b e r o f a n a ly s t p o rtfo lio s , w h ic h is n. * * * , **, a n d * in d ic a te th a t B o n fe rro n i p -v a lu e is a t 1%, 5 % , a n d 10% s ig n ific a n c e le v e ls, re s p e c tiv e ly , an d is re p o rte d u n d e r th e c o lu m n (1 ) a n d (1 0 ). T h e b in o m ia l t-s ta tis tic is ( 0 . 5 - . v ) / [ ( 0 . 5 ) ( 0 . 5 ) / ; i ] '/ : , w h e re x is the fra c tio n o f th e a lp h a e s tim a te s th a t a re n e g a tiv e . N o rm a lity test is K o lo m o g o ro v -S m irn o v test. T h e t-sta tistic fo r th e m e a n is b a s e d o n th e d is trib u tio n s o f th e e s tim a te d a lp h a s. N is th e to ta l n u m b e r o f a n a ly s ts in th e sa m p le . T h e d a ta a re d a ily fro m O c to b e r 1993 th ro u g h D e c e m b e r 2 0 0 0 . M odel M in t-sta t -2 .3 2 6 -1 .9 6 0 > t- 1 .6 4 5 > t 1,6 4 5 > t 1 .9 6 0 > t 2 .3 2 6 > t t> M a x 0 >t >t > -2 .3 2 6 > -1 .9 6 0 > -1 .6 4 5 i >0 > 1 .6 4 5 > 1 .9 6 0 2 .3 2 6 t-stat B in o m ia l t-s ta tis tic t-s ta tis tic fo r M e a n N o rm a lity T est n P a n e l A : R e s u lts F o rm in g P o rtfo lio s at R e p o rt D a tes. (6 ) (2 ) (3 ) (4) V W -Z e ro ( B u y -M a tc h ) (1) -2 .9 9 17 37 80 (5 ) 1414 1788 (7) 148 (8 ) 107 (9 ) 93 4 .4 6 * * * (10) E W - Z c r o ( B u y -M a tc h ) -3 .2 3 29 37 70 1369 1755 154 109 161 4 .1 6 * * 1 1 .1 0 V W - Z e r o ( M a te h -S e ll) E W -Z e ro (M a tc h -S c ll) -3 .7 7 * 8 14 27 536 6 6 8 34 11 12 34 530 6 6 8 31 17 19 3.25 -3 .1 9 31 30 4 .5 2 4.41 13 24 E W -Z e ro ( M a tc h -S e ll) -2 .8 3 -3 .4 8 -3.11 -3 .4 3 V W - Z e ro ( B u y -M a tc h ) -3 .0 7 24 46 6 6 1563 1765 106 71 39 E W -Z e r o (B u y -M a tc h ) -3 .6 2 -3.41 31 40 85 1545 107 63 54 11 28 13 31 29 61 7 -3 .1 8 12 11 1755 592 609 597 33 16 14 3.25 (11) 9 .6 9 ) (1 3 ) (1 4 ) 10.67 1 1 .9 0 ( 0 .1 5 0 .0 1 3684 3684 4 .9 0 4 .9 1 0 .5 8 0.41 1335 1335 3 .8 3 3 .9 5 1.78 1.71 0 .1 5 0 .1 5 0 .0 9 0 .2 7 3684 3684 1331 1331 1 .8 8 0 .1 5 0 .1 5 3680 12 P a n e l B: R e su lts F o rm in g P o rtfo lio s fro m 5 T ra d in g D a y s a fte r R e p o rt D a tes. V W - Z c r o ( B u y -M a tc h ) E W -Z e ro (B u y -M a tc h ) V W - Z e r o (M a tc h -S e ll) 10 10 50 40 13 11 68 79 18 24 1501 1488 621 615 1817 1794 609 612 119 126 28 26 71 73 19 19 45 60 13 14 3 .5 0 3 .6 6 3 .0 0 3 .0 0 6 .9 2 6 .9 5 -0 .1 9 -0 .3 0 P aneJ C: R e s u lts F o rm in g P o rtfo lio s fro m 15 T ra d in g D a y s a fte r R e p o rt D a te s . V W -Z e ro ( M a tc h - S e ll) E W -Z c r o (M a tc h -S c ll) 3 .5 2 3 .7 0 4 .6 5 4 .5 8 12 3 .0 8 13 3 .0 4 -0 .6 3 -0 .1 4 1.59 0 .5 9 0 .4 7 0 .5 6 0 .4 7 3680 1319 1319 recom m endations. My study com plem ents their research by extending the analysis to a m uch more com prehensive set o f analysts. Another distinction is that I study the determ inants o f individual a n a ly st performance. A nalysts’ portfolios include stocks recom m ended in the current period as well as those recom m ended in any previous periods, but have not been revised. Portfolio perform ance is a more accurate m easure because although I find that an aly st portfolios on average d o not show post-event return drift, the portfolios o f som e g o o d analysts appear to exhibit drift. My com prehensive controls for risks are preferable to traditional adjustm ents because early m ethods are not alw ays adequate for controlling for risks as show in Brav et al. (2000) and E ckbo et al. (2000) and because investors are interested in the risk levels o f recom m ended stocks. Brokerage firms such as M errill Lynch, Morgan Stanley, and Salomon Sm ith B arney issues risk ratings for all the sto ck s they recom mend [B row n (2001)]. I use the following m odel: A lpha, = « 0 +fl,AA l,., + n 2A42,_, + alAA3l_l + a 4A 4 4 r_, + a $LN BRK SZt (3) +a(LNSTKCAP,+a1LNSTKCAPSQ, +a„LNNREC, + avLN NSTK,+a]tlLNNSTKSQ, +e, where Alpha is the intercept o f the six-factor model estim ated w ith the portfolio returns o f analysts within a specific y ear. To be included, an analyst m ust have at least three m onths o f return data within a year.22 A A l to A A 4 are m easures o f analyst reputation from 1 through 4 for first-team , second-team , third-team , and runners-up of the A ll-A m erican Research T eam as ranked by //. They are dum m ies equal to one for each corresponding group of analysts and zero R e su lts for a re q u ire m e n t o f 8 m o n th s o f re tu rn d a ta arc q u a lita tiv e ly th e s a m e , and th u s n o t re p o rte d . R e su lts u sin g in fo rm a tio n ra tio a s p e r f o r m a n c e m easu re is sim ila r. 22 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. otherw ise.23 E ach October, / / issues a ranking o f A ll-A m erican financial analysts based on a survey o f m ore than 2,000 m oney managers conducted betw een M arch and M ay to evaluate the perform ance of a n a ly sts during the previous year on the basis o f stock picking, earnings forecasts, w ritten reports, industry know ledge, accessibility and responsiveness, and useful and tim ely calls.24 Several studies com pare the perform ance of All-A m erican and non-A ll-A m erican analysts. Stickel (1992) finds that A ll-A m erican analysts perform better on earn in g s forecasts o f a sam ple o f stocks that they cover. Stickel (1995) show s that the b u y recom m endations o f the first-team A ll-A m erican generate abnorm al announcem ent returns in the first tw o m onths across all stocks. I expect these four dum m y variables to be positively correlated w ith analyst perform ance. L N B R K S Z is the logarithm o f the num ber o f analysts in the brokerage firm. Logue (1986) first hypothesizes that n u m b er of analysts m ay be positively correlated with brokerage firm perform ance b ecau se large brokers have larger sales forces, and therefore can have m ore price impact. In som e sense, this is a m easure o f analyst recognition, since analysts in larger brokerage h o u ses gather more recognition from investors and thus have more im pact. L arge brokerage firm s also have more resources available to analysts and easier access to com pany m anagem ent. In addition, large brokerage firm s usually have other com m ercial relationships w ith the firms covered, and presum ably their analysts have m ore ch an n els to access inform ation. Stickel (1995) show s that brokerage size is 23 T he A ll- A m c ric a n lis t is p u b lis h e d in O c to b e r. So I u se th e re la tio n b e tw e e n th e a lp h a s in th e c u r re n t year and the A ll-A m e ric a n ra n k in g in th e p r e v io u s y e ar to in v e s tig a te th e p re d ic tiv e a b ility o f A ll-A m e ric a n ran k in g . T o a c c o u n t fo r th e p o s s ib ility th a t s o m e in v e sto rs m ay h a v e k n o w n A ll-A m c ric a n r a n k in g in the first s e m ia n n u a l p e r io d , th e r e la tio n s h ip b e tw e e n a lp h a s in th e c u r r e n t y e a r a n d th e c u r r c n l- y e a r A llA m e ric a n ra n k in g is a ls o e x a m in e d . T h e r e s u lts are v e ry sim ila r. - 4 In 1993 an d 1 9 9 4 , / / b a se d its r a n k in g o n a n a ly st a b ilitie s in s to c k p ic k in g , e a rn in g s e s tim a te s , w ritte n re p o rts, a n d o v e ra ll se rv ic e . 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. positively related to the event-period abnormal return, w hich supports both th e marketing ability hypothesis and the resource hypothesis. L N ST K C A P is the logarithm o f the mean m arket capitalization of the com panies covered by analysts. T his is also a m easure of recognition because analysts covering larger firm s could gain more recognition from investors. M arket cap could also be a proxy for firm s’ inform ation environm ent. Previous literature often suggests that large firm s are less subject to inform ation asymmetry with m ore intensive inform ation collection by analysts and other m arket participants, so a n a ly sts’ impact should be greater for sm all firm s because their contribution of inform ation is greater. The converse argum ent is that large firms m ay suffer more from inform ation asym m etry because of m ore com plicated operations and corporate structures. O r, analysts may have m ore distorted incentives because larger firm s generate m ore com m ercial business fo r a brokerage. Stickel (1995) presents evidence that event-period abnorm al returns are negatively correlated w ith the size o f com panies covered. T o control for a possible nonlinear relationship, I also include LN STK C APSQ , w hich is the square o f LN STK C AP. This variable is expected to negatively affect perform ance. L N N R E C and LN NSTK b o th m easure the efforts o f analysts. LN N REC is the logarithm o f the num ber of recom m endations made by an analyst within a year. Cooper, Day, and Lew is (2001) and Stickel (1992) both find that analysts who outperform in earnings estim ates usually issue reports more frequently. If analyst perform ance is determ ined m ainly by event-period abnorm al returns, given the sam e m agnitude o f eventperiod abnorm al returns, the m ore recom m endations being m ade, the better their perform ance. I expect LN N REC to be positively correlated w ith perform ance. 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LN N ST K is the logarithm o f the number o f sto c k s that an analyst covers during a year. Previous study has n o t exam ined this variable. S ince analysts usually concentrate on related industries, broader sto ck coverage may tra n sla te into a better understanding o f the industry trends, which sh o u ld m ake them more able to outperform . How ever, a positive relation betw een LN N STK a n d perform ance may beco m e w eaker or even negative as the num ber o f stocks covered increases further because perform ance may deteriorate if analysts take on too m uch w ork. To account for this possibility. I also include in the m odel LN N STK SQ , which is the square o f LN NSTK. T h is variable is hypothesized to have a negative coefficient. Panel A o f Table 7 reports m ean coefficients and their t-statistics based on tim eseries distributions of coefficients from calendar y ear regressions that attem pt to explain the cross-sectional distribution o f the alphas on analyst portfolios [Fama and M acB eth (1973)]. Both value- and equal-w eighted portfolios are exam ined for buy and sell recom m endations. For buy recom m endations, analyst perform ance im proves with their brokerage firm size and the num ber o f stocks covered. As expected, the positive correlation betw een perform ance and the num ber o f stocks covered w eakens as the number o f stocks covered increases beyond a certain level. For example, the o p tim al num ber of stocks im plied by the coefficient estim ates on buy portfolios is betw een 12 and 13. In com parison, a typical analyst in m y sam ple covers about 14 to 15 stocks, suggesting a bit higher than optim al w orkload for analysts. Perform ance also improves w ith the size of the com panies covered for value-w eighted portfolios and the num ber of recom m endations made for equalw eighted portfolios, although the coefficients are o n ly m arginally significant. 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Interestingly, som e All-American analysts make less inform ative buy recom m endations than other analysts. For sell recom m endations, perform ance im proves with the num ber of recom m endation m ade and w orsens with the size of the com panies covered. S o m e AllA m erican analysts have more inform ative sell recom m endations than other analysts. The signs o f the coefficient estim ates generally confirm the hypotheses ex cep t for the reversed signs on the 11 A ll-A m erican analyst dum m y variables. How ever, the results that som e II A ll-A m erican analysts have m ore inform ative sell recom m endations and that some other A ll-A m erican analysts have less inform ative buy recom m endations could suggest that the II All-American analysts have more d istorted incentives than the other analysts. They m ay have made m any m ore buys than the stocks actually m erit, and they may have tried harder to avoid sells if possible. The m ore distorted incentives m ay be attributable to the facts that brokerage firm s frequently use All-A m erican analyst coverage to m aintain or attract investm ent banking business [Hahn (2000)] and that AllA m erican analyst coverage is an im portant factor in a firm ’s decision on an investm ent bank for business such as IPOs and SEO s [Krigman, S haw , and W om ack (2001)]. Since it is not II All-American analysts with the sam e rank who produce buys and sells with extrem ely different inform ation content, another possibility is that perform ance o f the II A ll-A m erican analysts as a group may be close to that o f other analysts. T o test this possibility, I classify all the A ll-A m erican analysts into a single group rather than by the rank o f their team , and find that their perform ance is indistinguishable from non-AUAm erican analysts for both buy and sell recom m endations. Thus, I conclude th at AllAm erican analysts m ay not have more distorted incentives. Instead, their status ju st does not have predictive pow er for the future perform ance o f analysts. 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 7 Cross-sectional determinants of analyst performance, risk taking behavior, and aggressiveness T a b l e 7 re p o rts th e re s u lts on the c r o s s - s e c tio n a l d e te r m in a n ts o f in d iv id u a l a n a ly s t p e rfo rm a n c e , risk t a k in g b e h a v io r, a n d d e v ia tio n fro m th e risk le v e l ta k e n b y m e d ia n a n a ly sts. In P a n e l A , th e d e p e n d e n t v a r ia b le is th e p e rfo rm a n c e m e a su re of a n a ly s ts , w h ic h is th e a lp h a of th e s ix - f a c to r m o d e l: R„ = « , + b ’l0R F i +bllR F l_l +e„ . R, is e x c e s s re tu r n o n th e z e ro -in v e s tm e n t p o r tfo lio s th a t a re lo n g the a n a ly s t- re c o m m e n d e d b u y p o rtfo lio s a n d s h o r t th e m a tc h in g p o rtfo lio s for the b u y lis t, o r a re sh o rt the a n a ly s t- re c o m m e n d e d se ll p o rtfo lio s a n d lo n g th e m a tc h in g p o r tf o lio f o r the sell list. A n a ly s t p o rtfo lio s are m a tc h e d b y siz e , b o o k -to -m a rk e t, a n d m o m e n tu m . T h e v a r ia b le s in v a lu e - w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t in d e x , s iz e , R F , in clu d e e x c e s s r e tu r n o n th e C R S P b o o k -to -m a rk e t, re tu r n m o m e n tu m , e a rn in g s /p ric e , a n d liq u id ity factors. In P a n e l B, th e d e p e n d e n t v a ria b le s are risk m e a s u r e s s u c h a s the s ta n d a r d d e v ia tio n o f th e re sid u a ls fro m th e a b o v e s ix - f a c to r m o d e l e s tim a tio n an d th e s ta n d a r d d e v ia tio n o f th e o rig in a l a n a ly s t p o r tf o lio returns. In P a n e l C , th e d e p e n d e n t v a ria b le s a re the a n a ly s t d e v ia tio n s fro m th e ris k le v e l ta k e n b y m e d ia n a n aly sts. A A I , A A 2 . A A 3 . a n d A A 4 a re d u m m ie s th a t e q u a l to o n e if the a n a ly s ts b e lo n g to Institutional Investor A l l - A m e r i c a n first, s e c o n d , third te a m , and ru n n e r-u p s , re s p e c tiv e ly , an d z e ro o th e rw is e . A ll d e p e n d e n t v a r ia b le s a re in p e rc e n ta g e s. L N B R K S Z , L N S T K C A P , L N N R E C , a n d L N N S T K a rc the lo g a rith m o f th e a v e r a g e n u m b e r o f c o h o rt a n a ly s ts in th e ir b ro k e ra g e h o u s e ; lo g a rith m o f th e m ean m a rk e t c a p o f th e firm s a n a ly s ts c o v e r; lo g a rith m o f th e n u m b e r o f re c o m m e n d a tio n s a n a n a ly s t m ak e s; lo g a rith m o f th e n u m b e r o f sto c k s a n a n a ly st c o v e r s ; re s p e c tiv e ly . L N N S T K S Q an d L N S T K C A P S Q a rc the s q u a re o f L N N S T K a n d L N S T K C A P , r e s p e c tiv e ly . I u se F a m a M a c B e th (1 9 7 3 ) a p p r o a c h to e stim a te th e c o e f f ic ie n ts b y y e a r. T h e m ea n c o e ffic ie n t a c r o s s th e s a m p le p e r io d a n d the t-s ta tis tic b a se d on th e d is tr ib u tio n o f e s tim a te d c o e ffic ie n ts , are re p o rte d . ***, **, a n d * in d ic a te th a t t-s ta tis lic s a re sig n ific a n t a t 1% , 5 % , a n d 10% le v e ls, re sp e c tiv e ly . T h e d a ta a re d a ily fro m J a n u a ry 1994 th ro u g h D e c e m b e r 2 0 0 0 . P a n e l A . C ro s s - s e c tio n a l D e te rm in a n ts o f A n a ly s t P e rfo rm a n c e B u y P o rtf o lio s ______________________________ Sell P o rtfo lio s In d e p e n d e n t V a ria b le In te rc e p t V a lu e -W e ig h te d M ean 1 C o e ft. -0 .0 4 1 - 1 .8 8 * AAI AA2 - 0 .0 0 9 AA3 AA4 -0 .0 1 5 -0 . 0 0 1 LNBRKSZ LN STK CA P LN STK CA PSQ LNNREC LNNSTK LNNSTK SQ 0 .0 1 0 -1.62 3 9 - 0 .0 4 2 - 0 .0 0 6 0 .0 1 0 1 .1 0 .7 E q u a lly W e ig h te d M ean •I C o e ft. *** -0.09 -0 .0 1 2 0 .0 0 1 0 .0 0 2 2 .0 0 ** 0 .0 0 4 -0 . 0 0 1 1.73* -0.95 - 1 .8 6 * - 1 .1 2 1.07 - 2 .0 1 *** 0 .1 7 ** V a lu e -W e ig h te d M ean t C o e ft. 0 .0 2 5 1.44 0 .0 2 0 -0 .0 0 5 0 .0 1 4 0 .0 1 9 -0.54 1.49 1.79* 0 .0 0 1 0 .3 0 0 .0 0 3 0 .0 0 3 2 .0 2 1.44 -0 .0 0 3 0 .0 0 0 -0 .6 7 0 .0 0 0 0 .0 0 4 0 .0 4 0 0.99 2 .3 6 * * * 0 .0 0 7 0 .0 3 9 2 .2 0 * * 0 .0 0 5 - 0 .0 1 3 -0 .0 0 8 -2.9 3 * * * -0 .0 0 8 -2 .7 3 * * * 0 .0 0 1 1.82* 2.78 * * * E q u a lly W e ig h te d M ean C o e ft. 0 .0 2 2 0 .0 2 2 -0 .0 0 2 0 .0 1 5 0 .0 1 8 0 .0 0 1 t 1 .2 0 3 .1 3 * * * -0 .2 4 1.72* 1.72* 0 .5 5 - 2 . 6 8 *** -0.14 2.19*** - 0 .0 0 4 -1 .5 2 0 .2 4 -0 .0 1 2 -1 .4 3 0 .0 0 0 0 .1 2 0 .0 0 0 0 .0 0 6 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. -2 .6 4 * * * -0 .3 6 7 35*** C areer C oncerns o f Analysts a n d Tlieir B ehavior In this section I explore system atic variations in analyst behavior. It is possible that analyst characteristics can predict the risk levels o f their recom m ended portfolios. A lthough any such variations m ay be o f independent interest, my focus is on w hether or not differences in behavior reflect analyst career concerns. Scharfstein and Stein (1990), Prendergast and Stole (1996), and Zw iebel (1995) all predict that agents’ career concerns can lead them to herd on a com m on action. Li (2002) shows som e evidence that analysts are less likely to be ranked as AllA m erican analysts and are m ore likely to be term inated from their career w hen their recom m ended portfolios are relatively more risky controlling for p erform ance.25 He also finds that analyst career outcom es such as com pensation and career term ination are affected by the performance of their investm ent recom m endations. A ll-A m erican status offers m uch higher com pensation, som etim es m illions m ore. Analysts w ho are AllA m erican analysts in the previous y ear could be m ore aggressive than o th er analysts and recom m end stocks with higher portfolio risk in the hope o f higher returns because they have m ore reputation capital. A lternatively, they m ay be m ore conservative because investors sim ply dislike overly risky recom m endations or because a high-risk level can result in significantly poor perform ance. I estim ate the regression P ortfolio _ Riskiness, = aQ+ ct,AAlM + rz,A42,_, + a 3AA3t_{ + a 4AA4M +ai LN BRK SZl + a bL N S T K C A P ,+ a 1LNNREC, + asLN N STK , +£, 25 (4) S o m e m a y a rg u e th a t in v e s to rs o n ly c a re a b o u t th e r is k in e s s o f a ll th e sto c k s r e c o m m e n d e d b y all a n a ly s ts a s a g ro u p o r all a n a ly sts fro m a b r o k e ra g e firm s a s a g ro u p . H o w e v e r, a s e v id e n c e th a t in v e s to rs a re in te r e s te d in th e risk in e ss o f e a c h r e c o m m e n d e d s to c k s , b r o k e ra g e firm s such as M e r rill L y n c h , M o rg a n S ta n le y , a n d S a lo m o n S m ith B a rn e y is s u e s r is k ra tin g s fo r a ll th e s to c k s th e y re c o m m e n d [B ro w n (2 0 0 1 )]. 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T he m easures o f portfolio riskiness exam ined are the residual standard e rro r from the sixfactor m odel regression an d the square root of return variance o f analyst portfolios. I use A A I-A A 4 to measure a n aly st reputation and the other inform ation available from the recom m endation database to control for factors not related to reputation. Panel B of T ab le 7 reports m ean coefficients and their significance levels based on Fam a-M acBeth regressions. A lm ost all variables are highly significant and are im portant determ inants o f analyst risk-taking behavior. T he sign o f coefficients are fully consistent across buy and sell recom m endations and different risk m easures. T he primary observation is that m ost co efficien t estim ates on the A ll-A m erican analyst dum m y variables are significantly negative, suggesting that existing A ll-A m erican analysts recom m end significantly less risky portfolios. Since A ll-A m erican analysts are usually m ore experienced, som e m ay argue that their conservatism m ay be an artifact o f analyst experience or age. H ow ever, the previous literature has show n that younger or less experienced analysts are less aggressive in their earnings forecasts [C hevalier and Ellison (1998), Hong, Kubik, a n d Solom on (1999), and Lam ont (1995)]. A lthough inform ation on experience is unavailable due to the short time span o f the recom m endation database, the effects o f age and ex p e rien c e should only bias the results against finding a significantly negative effe c t o f A ll-A m erican status on risk-taking behavior. The table also su g g ests that analysts covering larger firm s or m ore stocks have less risky portfolios, w h ich is intuitive since large firms usually have low er stock return volatility and covering m o re stocks helps analysts diversify th eir portfolios. Interestingly, those analysts from larger brokerage firms or m aking more recom m endations have more risky portfolios. It is p o ssib le that w hen some analysts underperform , th ey w ork harder to 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. make m ore recom m endations. T h e y m ay also increase their p o rtfo lio s’ risk level to make more aggressive bets in hopes o f h ig h er returns. T h e second set of test aim s to exam ine the relationship betw een analyst characteristics and the extent to w h ich analysts deviate from the risk level o f a typical analyst’s portfolio. In the context o f analyst reputation, if A ll-A m erican analysts are less aggressive in the risk o f their recom m ended portfolios, I expect that they to deviate less from riskiness o f the average an aly st portfolio. That is, they m ay choose to follow the herd. T o exam ine this hypothesis, I estim ate the regression A ggressiveness, = a 0 + a,AAl,_| + a 2AA2M + «3AA3M + o,A A 4,_l (5) +a5L N B R K SZ ,+ aflLN STK C A P l + ^L N N R E C , + a, L N N ST K , +£, Aggressiveness measures how aggressively analysts deviate from a typical an aly st’s portfolio. T he dependent variables are the absolute values o f the differences betw een the level o f residual risk or total risk o f analyst portfolios and their average levels across all analysts. C oefficient estimates and th eir significance levels are presented in Panel C o f Table 7. A ll-A m erican analysts’ d ev iatio n in terms o f residual risk is not significantly different from other analysts and A ll-A m erican analysts tend to deviate significantly less from average analyst portfolio in term s o f total risk. Both results are consistent w ith my prior expectation. The other co efficien t estimates indicate that analysts covering large firms alw ays stick closer to the risk level o f average analyst portfolio. T he results also suggest that analysts in large b ro k erag e firms or covering m ore stocks deviate less in their recom m ended buy portfolios, but m ore in their recom m ended sell portfolios. A nalysts 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. Table 7. continued P a n e l B. C r o s s - s e c t i o n a l D e t e r m i n a n t s o f A n a l y s t s ’ R isk T a k i n g B uy R e co m m en d a tio n s R esidual R isk In tercep t AAI AA2 AA3 AA4 LNBRKSZ LNSTK CA P LNNREC LN N STK VW 0 .2 6 2 * * -0 .0 1 0 -0.013 -0 .0 1 1 * * -0 .009 0 .0 0 3 * * -0.0 1 1 * * 0 .0 1 8 * * -0 .0 5 8 * * EW 0.263*** -0.010 -0.013* -0.010*** -0.007 0.003*** -0.010*** 0.017*** -0.061*** Sell R e c o m m e n d a t i o n s T o t a l R isk VVV 3.349*** -0.142** -0.206*** -0.182*** -0.108*** 0.0 1 8 -0.096*** 0.157*** -0.504*** EW 3 .3 3 5 * * * -0 .1 4 5 * -0 .2 0 3 * * * -0 .1 7 6 * * * -0 .0 8 6 * * * 0 .0 3 6 * * * -0 .0 8 8 * * * 0 .1 5 3 * * * -0 .5 6 0 * * * R esidual R isk VW 0.31 4 * * * -0.044*** -0.047 -0.044* 0.008 0.0 1 7 * * * -0.0 3 2 * * * 0 .01 9 ** * -0.048*** EW 0 .3 1 5 * * * -0 .0 4 3 * * * -0 .049 -0 .0 4 7 * * 0.005 0 .0 1 7 * * * -0 .0 3 2 * * * 0 .0 2 0 * * * -0 .0 4 9 * * * T otal R isk VW 3 .9 6 9 * * * -0 .5 0 3 * * * -0 .4 8 8 * * * -0 .4 5 0 * 0.128 0 .1 2 4 * * * -0 .3 6 7 * * * 0.029 -0 .3 1 5 * * EW 3 .9 6 1 * * * -0 .4 9 3 * * * -0 .5 0 7 * * * -0 .4 7 0 * * 0.097 0 .1 3 1 * * * -0 .3 7 1 * * * 0.037 -0 .3 3 0 * * * P a n e l C. C r o s s - s e c t i o n a l D e t e r m i n a n t s o f A n a l y s t s ’ D e v i a t i o n fr o m M e d i a n A n a l y s t R i s k T a k i n g ^ Buy R e co m m en d a tio n s In tercep t AAI AA2 AA3 A A4 LNBRKSZ LNSTK CA P LNNREC LNN STK R esid u al R isk VW EW 0.09 3 5 * * * 0 .0919*** 0.0 0 0 9 0.0 0 0 6 -0.0 0 1 3 -0.0002 0 .0 0 1 6 0.0022 0 .0 0 4 6 0.0 0 5 0 -0.0010** -0.0010** -0.0 0 3 0 * * * -0.0033*** 0.00 2 6 * * * 0 .0027*** -0.0 1 0 8 * * * -0.0098*** Sell R e c o m m e n d a t i o n s T o tal R isk VW 1.0089*** -0.0648* -0.0764** -0.0639*** -0.0037 -0.0122*** -0 .0 387*** 0 .02 9 0 * * * -0 .0 673*** EW 0 .9 7 3 5 * * * -0 .0 6 1 5 * -0.0587 -0.0521 0.0098 -0 .0 1 4 8 * * * -0 .0 3 4 9 * * * 0 .0 1 7 6 * * * -0 .0 3 8 5 * * * R esidual R isk EW VW 0 .1201*** 0.1 2 0 4 * * * -0.0048 -0.0057 -0.0156 -0.0128 -0 .0099 -0.0105 -0.0041 -0.0051 0.0064* 0 .0 0 6 6 * * -0.0096*** -0 .0 0 9 8 * * * -0 .0 0 2 6 -0.0025 0.0 0 4 8 0.0055 T otal VW 1.3594*** -0.1965 -0.2880** -0.0537 -0.0464 0.0 9 0 5 * * * -0.1 2 4 4 * * * -0.1 0 5 8 * * * 0 .1 0 7 0 * R isk EW 1.3460*** -0 .1 8 9 0 -0.3 0 6 1 * * -0.0834 -0.0783 0.0 8 9 8 * * * -0 .1 2 1 7 * * * -0 .1 0 7 7 * * * 0 .1 1 7 6 * making m ore recom m endations are m ore aggressive in buy portfolio recom m endations and are less aggressive in their sell recom m endations. The theoretical models in the previous literature predict that agents’ career concerns can lead to herding behavior. M y results show that more reputable analysts behave m ore conservatively, which supports this prediction. Previous em pirical studies on herding hypothesis have focused on the behavior o f agents of different age or experience levels. T hey find that younger fund m anagers or analysts are less aggressive [Chevalier and E llison (1998). Hong, K ubik, and Solom on (1999), and Lam ont (1995)]. My research com plem ents this finding by show ing a sim ilar effect o f career concerns on the behavior o f analysts with different reputation capital at stake. C onclusions Using a new database with m ore com prehensive coverage of analyst recom m endations, I exam ine the perform ance o f portfolios recommended by individual analysts w ith m ore accurate perform ance m easurem ent and more com prehensive risk adjustm ents. T he analysis shows that portfolios that equally weigh the portfolios recom m ended by individual analysts generate significant abnormal returns for both buy and sell recom m endations. Individually, about 10% o f analysts significantly outperform benchm arks in their buys; 6% o f analysts significantly outperform in their sells. In com parison, about 3% o f analysts significantly underperform the benchm arks for buys or sells. The superior perform ance and significant num ber o f outperform ing analysts indicates strong inform ation content in their recom m endation, but the extent o f abnorm al returns does not provide trading profits once transaction costs are taken into account. 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Interestingly, abnorm al perform ance is lim ited to a short w indow starting two trading days before the recom m endation date a n d ending five trading d ay s after the recom m endation date. T he lack o f a post-event return drift is prim arily due to the more com plete risk-adjustm ents than previous literature. Since analysts often publicly release recom m endations several days afte r the initial recom m endation date or the date on the report, the rapid evaporation o f superior perform ance supports m arket efficiency. The gradual dissipation and the strong and short-term nature o f recom m endations’ investment value are also related to R egulation FD, which requires synchronous inform ation release by com pany m anagem ent to all investors. If trad ers in the same brokerage firm obtain inform ation earlier and front-run less preferred cu sto m ers o f analysts, this regulation may need to be applied to analysts too. I also find that analyst perform ance is pred ictab le on the basis o f their characteristics. Analyst perform ance improves sig n ifican tly with the n um ber of recom m endations issued and w ith their brokerage firm size. The num ber o f stocks covered by analysts has a significant positive, b u t concave relationship w ith performance. The optim al num ber o f stocks is betw een 12 and 13, a bit lower than the typical analyst responsibility to cover 14-15 stocks. II A ll-A m erican ranking cannot predict analyst perform ance. In addition, analyst behavior is affected b y their career concerns. T hose with more reputation capital at stake m ake m ore conservative portfolio recom m endations and deviate significantly less from the portfolio recom m endations of a typical analyst. Other analyst characteristics also affect analyst behavior. For exam ple, analysts covering large firms or m ore stocks tend to recom m end less risk y portfolios and analysts in large 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. brokerage firms or m ak in g m ore recom m endations tend to recom m end m o re risky portfolios. Tw o questions are suggested for future research agenda. First, if analysts outperform , but their p erfo rm an ce is not persistent, investors cannot p ro fit from trading on the recom m ended p o rtfo lio s, and the m arket efficiency prevails [G ru b er (1996)]. An im portant future research topic would be to ask w hether analyst p e rfo rm a n c e is persistent over time. Second, it is im portant to investigate w hether the traders a n d m arket makers in the sam e brokerage firm s as analysts obtain inform ation and tim ing a b o u t public release o f recom m endations e a rlie r than other investors and whether they e x p lo it this information to front-run other investors. 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R eferences B arber. Brad M., Reuven Lehavy, M aureen M cN ichols, and B rett Truem an, 2001, Can investors profit from the prophets? C onsensus analyst recom m endations and stock returns, Journal o f Finance 56, 531-563. B arber, Brad M., and John D. Lyon, 1997, D etecting long-run abnorm al stock returns: T he empirical pow er and specification o f test statistics. Jo u rn a l o f Financial Econom ics 43, 341-372. Bollen, Nicolas P.B., and Jeffrey A. B usse, 2001, On the tim ing ability o f mutual fund m anagers, Journal o f Finance 56, 1075-1094. Brav, A lon, Christopher G eczy, and Paul A. G om pers, 2000, Is the abnorm al return follow ing equity issuance anom alous? Journal o f Financial E conom ics 56, 209-249. B rennan, Michael J., and A vanidhar Subrahm anyam , 1996, M arket m icrostructure and asset pricing: On the com pensation for illiquidity in stock returns, Journal o f F inancial Econom ics 41, 3 4 1 -3 6 4 . B row n, Ken, 2001, A nalysts’ top stock picks get failing grade on risk m eter, Wall Street Jou rn a l, August 15, 2001. Busse, Jeffrey A., 1999, V olatility tim ing in m utual funds: E vidence from daily returns. R eview o f Financial Studies 12, 1009-1041. C arhart, M ark M., 1997, On Persistence in m utual fund perform ance, Journal o f Finance 52, 57-82. C heng, Yingmci, 2000, Inform ativeness o f analyst recom m endations, W harton School W orking Paper. C hevalier, Judith A. and G lenn D. E llison, 1999, Career concerns o f m utual fund m anagers, Q uarterly Journal o f E conom ics 114, 389-432. C lem ent, Michael B., 1999, A nalyst forecast accuracy: Do ability, resources, and portfolio com plexity m atter? J o u rn a l o f Accounting and E conom ics 27, 285-303. C ooper, Rick A., Theodore E. Day, and C raig M. Lewis, 2001, Follow ing the leader: A study of individual analysts earnings forecasts. Journal o f F inancial Economics 61, 383-416. C ow les, Alfred, 1933, Can stock m arket forecasters forecast? E conom etrica 1, 309-324. 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C ullen, Lisa Reilly, Pablo Galarza, Brian P. M urphy, Sarah R ose, Penelope Wang and Suzanne W oolley, 2000, Ten For 2000, M o n ey , February 1, 2000. Daniel, Kent, M ark G rinblatt, Sheridan T itm an, and Russ W erm ers, 1997, M easuring m utual fund perform ance with characteristic-based benchm arks, Journal o f Finance 52, 1035-1058. Datar, V., N. Naik, and R. Radcliffe, 1998, Liquidity and stock returns: An alternative test. Journal o f F inancial M arkets 1, 2 0 3 -2 1 9 . D im son, Elroy, 1979, Risk m easurem ent w hen shares are subject to infrequent trading, Journal o f F inancial Economics 7, 197-226. D im son, Elroy, and Paul M arsh, 1984, An analysis o f brokers’ and analyst unpublished forecasts of U K stock returns, Journal o f F inance 39, 1257-1292. Eckbo, B. Espen, R onald W. Masulis, and O yvind Norli, 2000, Seasoned public offerings: R esolution o f the ‘new issues p u z z le ,’ Journal o f F inancial Economics 56, 251-291. Eckbo, B. Espen, and O yvind Norli, 2000, L everage, liquidity and long-run IPO returns, D artm outh C ollege W orking Paper. Elton, Edwin J., M artin J. Gruber, and Seth G rossm an, 1986, D iscrete expectational d ata and portfolio perform ance, Journal o f Finance 41, 699-713. Fam a, Eugene F., and K enneth R. French, 1992, T h e cross-section o f expected stock returns. Journal o f Finance 47,427-465. Fam a, Eugene F., and K enneth R. French, 1993, C om m on risk factors in the returns on stocks and bonds, Journal o f Financial E conom ics 33, 3-56. Fam a. Eugene F., and Jam es D. M acBeth, 1973, R isk, return and equilibrium : Em pirical tests, Journal o f P olitical Economy 81, 607-636. Francis, Jennifer, and Leonard Soffer, 1997, The relative inform ativeness o f analyst stock recom m endations and earnings forecast revisions. Journal o f A ccounting Research 35, 193-211. G ruber, M artin J., 1996, A nother puzzle: The grow th in actively m anaged mutual funds, Journal o f F inance 60, 783-810. H ahn, Avital Louria, 2000, Fallout from A T & T W ireless IPO M ay Be Costing Solly M andates, Investm ent D e a le r’s D igest, O ctober 16, 2000 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Hong, H arrison, Jeffre y D. Kubik, and A m it Solom on, 2000, S ecurity analysts’ career concerns and the herding o f earnings forecasts, R A N D Jo u rn a l o f Economics 31, 121144. Jacob, John, T hom as Z Lys, and M argaret A. N eale, 1999, E xpertise in forecasting perform ance o f security analysts. Journal o f A ccounting E conom ics 28, 51-82. Jaffe, Jeffrey, D onald B. Keim , and R andolph W esterfield, 1989, Earnings yields, m arket values and sto ck returns, Journal o f F inance 44, 135-148. Jasen, G eorgette, 20 0 1 , Raym ond Jam es W as T op Picker O f S tocks in the Fourth Quarter, W all S tre e t Journal, February 12 2001 Jegadeesh, N arasim han, Joonghyuk Kim, Susan D. Krische, and C harles Lee, 2001, Analyzing the analysts: when do recom m endations add value? W orking Paper o f U niversity o f Illinois at U rbana-Cham paign. Jensen, M ichael C ., 1968, T he performance o f m utual funds in the period 1945-1964, Journal o f F inance 23, 389-416. Krigman, Laurie, W ayne H. Shaw, and Kent L. W om ack, 2001, W hy do firm switch underw riters? J o u rn a l o f Financial E conom ics 60, 245-284. Laderman, Jeffrey M ., 1998, Wall street’s spin gam e: Stock analysts often have a hidden agenda, B usiness W eek, O c to b e rs, 1998. Lamont, O w en, 1995, M acroeconom ic forecasts and m icroeconom ic forecasters, N B E R W orking Paper 5284. Lin, Hsiou-w ei, and M aureen F. M cNichols, 1998, U nderw riting relationships, analyst earnings forecasts and investment recom m endations, Jo u rn a l o f Accounting and Economics 25, 101 - 127. Li, Xi, 2002, C a re er concerns o f analysts: C om pensation, term ination, and perform ance, Vanderbilt U niversity W orking Paper. Logue, Denis E., 1986, Discussion, Journal o f Finance 41, 713-714. M ichaely, Roni, and K ent L. W omack, 1999, C onflict o f interest and the credibility o f underw riter analyst recom m endations, R eview o f F inancial S tu d ies 12, 653-686. Prendergast, C anice and Lars Stole, 1996, Im petuous youngsters and jaded old-tim ers: Acquiring a reputation for learning. Journal o f Political E conom y 104, 1105-34. Scharfstein, D avid S., and Jerem y C. Stein, 1990, Herd behavior and investment, Am erican E co n o m ic R eview 80, 465-479. 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Scholes, M yron, and Joseph T. W illiam s, 1977, Estimating b e ta s from nonsynchronous data, Journal o f F inancial Econom ics 5 , 309-328. Stickel, Scott E., 1992, Reputation and perfo rm an ce among secu rity analysts, Journal o f Finance 4 7 , 1811-1836. Stickel, Scott E., 1995, T he anatomy o f th e perform ance of b u y and sell recom m endations, F inancial A nalysts Journal, Septem ber-O ctober 1995, 25-39. W hite, H albert, 1980, A heteroscedasticity-consistent covariance matrix estim ator and a direct test for heteroscedasticity, E co n o m etrica 48, 817-838 W om ack, Kent L., 1996, D o brokerage a n a ly st recom m endations have investm ent value? Journal o f F inance 5 1,137-167. Z w iebel, Jeffrey, 1995, C orporate co nservatism and relative com pensation. Journal o f P olitical E conom y 103, 1-25. 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C H A P T E R II W IL L PAST L E A D ER S S T IL L LEAD? PER FO R M A N C E P ER SISTEN C E O F FIN A N CIA L A N A LY STS Introduction A lthough perform ance persistence o f various m arket participants has been extensively studied, there is no evid en ce on the performance persistence for sell-side financial an aly sts.26 Analyst perform ance persistence is im portant for several reasons. First, analysts are playing an increasingly prom inent role in the investm ent process through their traditional influence on institutional investors and rising influence on individual investors. Second, evidence on perform ance persistence is relevant to our continuing study o f m arket efficiency. Several recent studies suggest that analyst recom m endations exhibit abnorm al returns at event dates and a significant post-event return drift that could last for m onths. [Elton, Gruber, and G rossm an (1986), and W om ack (1996)]. Li (2001a) show s that the investment value o f analyst recom m endations is limited to a short w indow around public release dates, and resolves a puzzle against m arket efficiency hypothesis. Yet, it is unknow n w hether analyst perform ance is persistent over tim e. R andom distribution o f abnorm al perform ance over tim e would support market efficiency. "6 Studies su g g estin g p e rsisten t p erfo rm an ce o f m u tu al funds include B ollen an d B usse (2 0 0 2 ), B row n and G oetzm ann (1 9 9 5 ), C a rh a rt (1997), G rin b la tl and T itm a n (1992), G ru b e r (1 9 9 6 ), and M alk icl (1 9 9 5 ). R esearchers also find e v id en c e o f p e rfo rm an c e p e rsiste n c e o f pension funds [C h rislo p h e rso n , F crso n , and G lassm an (1998)1, h ed g e funds [A garw al a n d N a ik (2 0 0 0 ), and B row n, G o etzm a n n , and Ib b o tso n (1 9 9 9 )], b ut not in v estm en t n ew sletters (Jaffe and M a h o n e y (1999)1 or brokerage firm s [B arb er, L e h av y , and T ruem an (2 0 0 1 ) a n d E lton, G ruber, and G ro ssm a n (1986)1. 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T hird, evidence on perform ance persistence is essential in providing guidance on how investors should choose analysts to follow. As Barber. Lehavy, M cN ichols, and Trueman (2001) and Li (2001a) point out, we can im prove the profitability o f trading strategies based on recom m endations if we could identify good analysts and follow only their recom m endations. Although Li (2001a) shows that m any analysts significantly outperform risk-adjusted benchm arks and there are m ore good perform ers than bad ones, the viability o f a trading strategy based on past perform ance depends on perform ance persistence, particularly persistence o f good performers. In addition, the perform ance persistence of sell-side analysts is related to that o f buy-side m oney m anagers. M any buy-side money m anagers report them selves that they actively search for inform ation in the research of sell-side analysts [Nelson Inform ation (2001)]. Since research shows that the perform ance o f buy-side m oney m anagers is persistent, we can infer that they m ay not rely so heavily on sell-side research if analyst perform ance persistence is less significant than money m anagers. However, if we find that analysts actually dem onstrate m ore significant perform ance persistence, an interesting question w ould be w hy the information in their recom m endations is not fully exploited or how this inform ation g ets lost in the investm ent process.27 M oreover, m y research is related to the literature that investigates system atic differences in analyst perform ance and the ranking system s that attem pt to identify good analysts on the basis o f past perform ance. For example, Li (2001a) and Stickel (1995) ~7 T o com pare the p e rfo rm a n c e p ersisten ce o f m oney m anagers to that o f se ll-sid e analysts, w c have to also consider m any o th e r d iffe re n c e s betw een th e se tw o professions. F or ex a m p le , m o n ey m an ag e rs incur significant tra n sa c tio n c o s ts and usually h a v e to allocate som e p art o f the fu n d s as cash if th ey are op en ended funds. M oney m a n a g e rs are also s u b je c t to stricter regulations and c a n n o t short-sell se cu ritie s in general. In ad d itio n , th e re sea rc h on m o n e y m a n ag e rs rarely has the d ata to track individual m an ag ers. It can only track in d iv id u al fu n d s, which c a n b e m an ag ed by differen t m a n ag e rs o v e r tim e. M o re o v e r, the 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. exam ine whether analyst characteristics are related to their performance in investm ent recom m endations. C lem ent (1999) and Jacob, Lys, and Neale (1999) assess the determ inants of accuracy o f analyst earnings forecasts. C ooper. Day, and L ew is (2001) explore new m ethodology to identify leaders in earn in g s forecasts. Institutional Investor and the W all Street J o u rn a l select a group of all-star analysts each year an d analysts who becom e all-stars are rew arded by m illions more in com pensation. Yet if analyst perform ance is not persistent, both the research resu lts and the rankings sh o u ld be considerably discounted, since investors would not be able to profit from follow ing past leaders. T his paper co n tributes to the understanding o f analyst perform ance in several ways. First, the (IB ES) database used has much m o re com prehensive c o v erag e of brokerage firms and financial analysts than any o th e r databases in the previous research. It includes m any m ore analysts from sm aller brokerage firms than large databases such as First Call and Zacks. It also includes important brokerage firms such as M errill Lynch, G oldm an Sachs, and D onaldson, Lufkin & Jenrette that are not included in Zacks. The data is virtually free o f survivorship bias. The more com prehensive database also enables m ore accurate m easurem ent of analyst perform ance. O th e r researchers follow analyst recom m endations fo r an arbitrary holding period such as 6 or 12 m onths, usually because they lack recom m endation revision dates. This ty p e o f assum ption could m isrepresent analyst perform ance, because analysts m ight revise th eir recom m endations w ithin weeks or months after an original research on m oney m an ag e rs u su ally uses m onthly return d a ta at m ost. S ince I track a n a ly s t reco m m en d atio n s at daily fre q u e n c y , m y risk adjustm ents c o u ld be m ore accurate. 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. recom m endation. W ith the exact revision d a te , I keep stocks in the analyst portfolio until analysts revise their recom m endations. Second, I study both tw o-period an d m ulti-period performance persistence over the quarterly, sem iannual, and annual intervals.28 I im prove on the m ulti-period persistence test developed by Agarwal and N aik (2000), which offer m ore pow er to discrim inate betw een persistence arising from chance and persistence due to an aly sts’ superior ability. The m ulti-period test also d o e s not require analysts to survive during both ranking and evaluation period, and thus introduces no bias as the tw o-period tests. I use daily data that offers m ore efficient perform ance estim ates compared to the m onthly data usually available in other persistence studies. Since I track analyst recom m endations at daily frequency, m y risk adjustm ents according to portfolio holdings should be m ore accurate than using less frequently updated portfolio holdings in other persistence studies. In addition, perform ance is m easured using both raw returns and risk-adjusted returns. I incorporate recent advances in the long-run perform ance evaluation literature and provide extensive risk adjustm ents to analyst perform ance [Brav, G eczy, and G om pers (2000), D aniel, G rinblatt, Titm an, and W erm ers (1997), and Eckbo, M asulis, and Norli (2000)]. C areful risk adjustm ents a re im portant because Brown and G oetzm ann (1995) suggest that persistence m ay appear sim ply because all managers use a com m on strategy not captured by standard stylistic categories or risk adjustment procedures. C arhart (1997) shows that perform ance persistence o f mutual funds disappears after controlling for m om entum . 28 G iv en (he lim ited histo ry o f a n a ly s ts’ portfolios, it is no t p o ssib le to exam ine three- to fiv e -y e a r p erfo rm an c e as in m utual fu n d literatu re. A lso Li (2 0 0 2 ) show s th at analyst career co n c e rn s are re la te d only to p erfo rm a n c e in the p ast o n e o r tw o years. 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I show that less com prehensive stylistic/risk adjustm ents produce stronger persistence. In addition to adjustm ents to system atic risks, I further adjust analyst abnorm al perform ance for specific risks in their recom m ended portfolios. A nalyst perform ance may be persistent sim ply because individual analysts are exposed to different levels o f specific risks consistently over tim e. An exam ination o f raw return persistence provides an upper bound for perform ance persistence. As noted by B row n and G oetzm ann (1995), raw returns include all the possible persistence, even spurious cases. In addition, while m utual fund m anagers and academ ics are interested in risk-adjusted returns, individual investors and hedge fund m anagers m ay be more interested in raw returns. M oreover, a widely recognized analyst ranking o f The Wall Street Jo u rn a l ( W SJ) is based on raw returns o f analysts’ recom m ended portfolios. This ranking is useless to investors if raw returns are not persistent. M y results indicate one-period-ahead perform ance persistence for financial an aly sts’ buy recom m endations that is invariant to testing m ethodologies, portfolio w eighting schem es, return m easurem ent intervals, or risk adjustments. Perform ance o f a n aly sts’ sell recom m endations, how ever, is almost nev er persistent, except for raw returns at the annual and sem iannual intervals. Perform ance of buy recom m endations is also persistent in a majority o f subsam ple periods w hen two-period persistence tests are conducted for each pair o f consecutive subperiods. In addition, risk-adjusted perform ance o f an alysts’ buy portfolios is m ore persistent at the annual and sem iannual intervals, and raw returns at the semiannual and q u arterly intervals. T h e performance persistence is m ore pronounced for raw returns than for risk-adjusted returns, confirm ing the 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. im portance o f sufficient risk adjustm ents. M oreover, past winners and losers both exhibit perform ance persistence. And the overall perform ance persistence o f financial analysts is largely attributable to past w inners rather than losers. Thus, we can use th e perform ance persistence o f losers to avoid bad analysts, and, m ore im portant, the w in n e rs’ perform ance persistence to create profitable investm ent strategies. I also exam ine o f the econom ic value o f tw o-period perform ance persistence by analyzing the profitability o f a trading strategy that trades the recom m endations o f each analysts. T he results verify the above findings. Follow ing the buy recom m endations of past w inners based on risk-adjusted and raw returns generates significantly positive riskadjusted and raw returns, respectively. Follow ing the buy recom m endations o f past w inners in term s of risk-adjusted returns generates annualized risk-adjusted returns as high as 15.7%. Portfolios of past w inners ranked by raw returns could produce annualized buy-and-hold returns o f 45.2% . Follow ing past winners in term s o f raw returns does not generate superior risk-adjusted returns, though. Overall, the WSJ ail-star rankings could be a starting point to form trading strategies if investors are interested only in raw returns. Finally, the perform ance persistence o f both risk-adjusted returns and raw returns tends to disappear in a m ulti-period fram ew ork, except for the raw returns o f buy recom m endations at the sem iannual and quarterly intervals. Raw return persistence is thus m ore robust and relevant in potential trading strategies. Yet raw return persistence does not challenge m arket efficiency. The evaporation o f perform ance persistence in the face o f m ore sufficient risk adjustm ents supports m arket efficiency and confirm s the suggestion by Brown and G oetzm ann (1995) and Carhart (1997) that perform ance 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. persistence m ay appear sim ply because m anagers follow a comm on strategy not captured by standard stylistic categories or risk adjustm ent procedures. The rest o f this paper is organized as follow ing. Section 2 d escrib es the sample and discusses the econom etrics o f the factor m odels and the tests o f tw o-period perform ance persistence. It also gives details about the m ethodology used to form the analysts’ portfolios and their m atching portfolios. Section 3 exam ines persistence in the traditional tw o-period fram ew ork. Section 4 presents the tests and results in the m ulti­ period fram ew ork. Section 5 offers concluding rem arks. Sam ples and M ethodology Sam ple D escription T he prim ary database used in this paper com es from IBES. Its m ajor benefit is that it includes recom m endations from a very broad sample of brokerage firm s and financial analysts. Even large databases such as Z acks do not include im portant bulge bracket firm s such as M errill Lynch, G oldm an Sachs, and Donaldson, Lufkin & Jenrette. The IBES database includes all m ajor brokerage firm s plus a large sam p le o f sm aller brokerage firm s. Analysts can alm ost always be tracked even if they sw itch brokerage firms. V arious market participants, including professional investors, use this database. IBES has collected buy and sell recom m endations from the research reports of financial analysts since the end o f O ctober 1993. The database includes both brokerage firm -specific ratings and a standardized IBES rating. The former are usually on a threeto five- level scale. The IB ES-created ratings are on a uniform five-level scale; numeric 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ratings from 1 through 5 correspond to “strong buy,” “ buy," “hold,” “underperform ,” and “sell” . Buy portfolios o f analysts are form ed using recom m endations with ratings o f 1, and sell portfolios are form ed using recom m endations with ratings o f 4 and 5 .29 The investm ent recom m endation data are from the end o f October 1993 through D ecem ber 2000. The return and accounting data are drawn from CRSP and C om pustat, respectively. Panel A o f T ab le 1 sum m arizes the database. T here are 241,222 recom m endations by 7,308 financial analysts from 408 institutions in the five buy and sell recom m endation categories. In the em pirical analysis, I exclude analysts with fewer than 10 recom m endations. T hat sam ple consists o f 4,383 analysts before other restrictions are applied. Panels B and C o f T able 1 show , for buy and sell recom m endations, respectively, the estim ated effect o f survivorship bias on risk-adjusted performance o f alpha. The panels present percentage daily excess returns on the overall sample and portfolios o f surviving and disappearing analysts by y e a r.30 D isappearing analysts have significantly w orse perform ance than surviving analysts for m ost o f the sample period, w hatever the type of recom m endations or portfolio weighting schem es. Yet survivorship bias does not have a pronounced effect on the perform ance distribution of the overall sam ple because its m ean perform ance is not significantly different from the perform ance o f surviving analysts in most years. Even if the effect o f survivorship bias may be slight, it is fortunate that the IBES database does not have any survivorship bias. Analyst recom m endations '9 I com b in e ratings 4 a n d 5 b e c a u se th ere are m any fewer “S e ll” reports. 30 I w ill describe the risk a d ju s tm e n ts an d a n a ly st portfolios in d e tail in the next tw o sectio n s. 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1 Summary Statistics of Recommendation Database T ab ic 1 p resen ts su m m a ry statistics regarding th e IB E S recom m endation d ata b a se . P anel A show s the size o f th e d atab ase. P an els B and C re p o rt, fo r buy and sell reco m m en d atio n s each y ear, average a lp h a o f all an a ly sts and average a lp h a o f a n aly sts w ho survive o r d is a p p e a r the n ex t year. A lp h a is the in tercep t o f the six -facto r m o d e l R it = or, + bl0R F , + b u R F w here | +£„ . Ru is e x c e s s return on the zero -in v estm en t p o rtfo lio s that are long th e a n aly st- recom m ended b u y p o rtfo lio s and sh o rt the m atch in g p o rtfo lio s for the buy list, o r a re sh o rt the an aly st-rec o m m e n d ed sell p o rtfo lio s and long th e m a tc h in g p ortfolio for the sell list. T h e variables in /? f , in clu d e ex cess retu rn on the C R S P v alu e-w e ig h te d N Y S E /A M E X /N A S D A Q m arket index, siz e , b o o k -to -m a rk e t, return m o m e n tu m , e a rn in g s/p rice, and liq u id ity factors. T h e coefficients a re estim a ted using ordinary least sq u a re s. VW and EW are v a lu e-w e ig h te d and equally w eig h te d p o rtfo lio s. D ata are from Ja n u a ry 1994 through D ecem ber 2CKX). _____________ P an el A: S u m m ary S tatistics o f IB E S R e c o m m en d a tio n D atabase N um ber o f A n a ly sts: 7308___________________________________________________ N um ber o f B ro k ers: 4 9 8_____________________________________________________ N um ber o f A n a ly sts w ith > 10 R eco m m en d atio n s: 4 3 8 3 ______________________ N um ber o f A n a ly sts w ith > 100 R eco m m en d atio n s: 5 6 3 ______________________ Y ear 1994 1995 1996 1997 1998 1999 A verage Panel B. R ecom m ended B u y P o rtfo lio s W h o le S am p le S u rv iv a ls_______ VW EW EW VW 0 .0 0 4 0 .0 0 2 0 .0 0 8 0 .0 0 4 -0 .0 0 4 -0.001 -0.002 0 .0 0 2 0.021 0 .0 2 8 0 .0 2 9 0 .0 2 2 0 .0 1 3 0 .0 1 8 0 .0 1 8 0.015 0 .0 1 9 0 .0 2 9 0 .0 2 2 0 .0 3 3 0.021 0 .0 1 6 0 .0 2 6 0 .0 1 9 0.011 0 .0 1 7 0 .0 1 3 0 .0 1 9 Panel C. R ecom m ended S ell P o rtfo lio s W h o le S am p le S u rv iv a ls VW EW EW VW 1994 1995 1996 1997 1998 1999 A verage -0 .0 0 4 0 .0 2 8 0 .0 2 9 0 .0 3 5 0 .0 0 4 0 .0 1 5 0 .0 1 8 -0.005 0 .0 2 8 0 .0 3 3 0 .0 3 6 0 .0 0 2 0 .0 1 0 0 .0 1 7 -0.002 0 .0 2 7 0 .0 3 0 0 .0 4 7 0.001 0.021 0.021 -0 .0 0 4 0 .0 2 7 0 .0 3 5 0 .0 4 7 -0 .0 0 2 0 .0 1 5 0 .0 2 0 D isap p earan ce EW VW -0.025 -0.015 -0 .0 3 2 -0.029 0 .0 1 4 0.016 0 .0 1 6 0.003 -0.004 -0.008 -0 .0 1 2 -0.007 -0.0 07 -0.007 D isap p earan ce EW VW -0.014 0.043 0.015 -0.059 0.030 -0.020 -0.001 -0.013 0 .0 3 3 0 .0 2 0 -0 .0 4 9 0 .0 3 6 -0 .0 2 2 0.001 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. are kept in the d atab ase even if an analyst disappears. Perform ance M ea su res I use average returns and buy-and-hold returns o f original analyst portfolios to m easure raw return perform ance. Risk-adjusted perform ance is m easured on the basis o f a six-factor m odel estim ated with daily returns. T h e factor model is expressed as +'LPo,R„+£„ (•) /=i where Rit is the excess return on the portfolio o f analyst /' on day /; or, m easures the abnorm al return o f the portfolio o f analyst i ; R Jt is the return o f factor j on day t ; and £„ is the idiosyncratic return o f the portfolio o f an aly st i on day t . The risk-free rate o f return is based on the daily U.S. 90-day Treasury b ill.31 The six factors are the excess return on the C R S P value-w eighted N Y SE/A M EX /N A SD A Q m arket index, the size and book-to-m arket factors o f Fam a and French (1993), the return m om entum o f Carhart (1997), the earnings/price factor, and a liquidity facto r.32 1 use a six -facto r m odel because Li (2001a) finds that analysts cover stocks w ith m any differing characteristics from the overall m arket and tend to recom m end stocks with significantly different characteristics for their buy and sell portfolios. Because these characteristics can be related to system atic risk or investm ent styles that have nothing to do w ith the co n trib u tio n o f analysts' skills, it is im portant to m ake the appropriate risk or style adjustm ents in evaluating analyst perform ance. Brow n and G oetzm ann (1995), 31 R isk-free re tu rn s a rc c a lc u la te d using the Federal R e se rv e ’s c o n stan t-m atu rity interest rate series. R etu rn s arc calcu lated from th e p u b lish e d yields using a hypothetical b o n d w ith the stated m aturity an d a co u p o n eq u al to the yield , thus tra d in g at p a r o r face value. An en d -o f-p erio d p ric e on the bill using the n ex t d a y ’s yield is first c alcu lated . T he p ric e is th en used to obtain the risk-free retu rn s. 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Carhart (1997), and Fama an d French (1993) su g g est that sufficient sty lish and risk adjustm ents are necessary to prevent drawing sp u rio u s conclusions o f su p erio r perform ance and perform ance persistence. In addition, Li (2001a) com pares the perform ance o f various em pirical asset pricing m odels using daily data an d he suggest that the six-factor model ex p lain s a large portion o f the time series variation o f stock returns and prices analyst portfo lio s more correctly. As a robustness check, I also measure perform ance using several o th er frequently used factor m odels: a single-factor market m odel; a three-factor m odel th at includes size and book-to-m arket factors [F am a and French (1993)]; a four-factor m odel that adds m om entum effects [Carhart (1997)]; and a five-factor model that adds an earnings/price factor. I also use a modified version of the m acro factor model o f E ckbo, et al. (2000) as in Li (2001a). Persistence resu lts are similar and are thus not reported. The use o f daily d ata introduces one com plication. Dimson (1979) and Scholes and W illiam s (1977) observe a nonsynchronous trad in g problem in sto ck returns that hinders regression estim ation for individual securities. I address this p ro b lem by adding a lagged term for each factor in the m odel:13 /-I Li (2001a) reports that the six-factor m odel still misprices a significant num ber o f test portfolios. I thus apply a procedure recom m ended by Daniel, et al. (1 9 9 7 ) and Eckbo 32 T he a p p en d ix d escrib es the fa c to rs an d their form ation. 33 D im son (1 9 7 9 ) suggests in clu d in g a s m any as three lags a n d th ree lead term s. T e stin g w ith several co m b in a tio n s sh o w s that only the first lag term is c o n siste n tly significant. T h e re su lts u s in g th ree lag and three lead term s are m uch the sa m e a s u sin g only the first la g term . Since a n aly st p o rtfo lio s typ ically include several sto ck s, the n o n sy n c h ro n o u s trading p ro b le m is n o t as serious as for in d iv id u a l securities. B ussc (1 9 9 9 ) fin d s sim ilar results fo r d a ily returns o f m u tu a l fu n d s. F urtherm ore, a n a ly s ts u su ally cover h ig h -liq u id ity sto c k s, stocks that s h o u ld not have much o f a p ro b le m o f n o n sy n ch ro n o u s trad in g . 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. et al. (2000) to more accurately m easure the risk-adjusted perform ance o f analysts. They recom m end com bining a m atching technique [B arb er and Lyon (1997)] an d factor m odel analysis. A zero-investm ent portfolio is created by ( I ) investing in the analystrecom m ended buy portfolios an d shorting the m atching portfolios for buy recom m endations, or by (2) investing in the m atching portfolios and sh o rtin g the analyst recom m ended sell portfolios for sell recom m endations. Matching portfolios created according to the characteristics o f analyst portfolios should exhibit sim ilar tim e series properties, and thus m itigate the problem of m isspecified factor m odels. B ecause it is hard to obtain a perfect m atch, and the m atching technique alone m ay not elim inate the factor exposure of analyst portfolios, the additional factor regressions p ro v id e m ore effective risk adjustments. A nalyst Portfolios and Their M atching Portfolios For each financial analyst w ho made at least ten recom m endations betw een O ctober 1993 and D ecem ber 29, 2000, in the IB E S recom m endation database and for whom a buy or sell portfolio can be formed for at least three months, I create both a value-w eighted and an equal-w eighted portfolio using the recom m ended sto ck s in the specific category. Stocks enter the analyst portfolios on the recom m endation date and are dropped at the revision date as recorded by IB ES. T he empirical tests use returns from January 1994 through D ecem ber 2000. Because I study analyst perform ance at the quarterly, sem iannual, and annual intervals, I require the length o f analyst portfolios to be at least tw o m onths, three m onths, and three m onths at the respective intervals. R esults using other length 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. requirem ents are sim ilar. T he requirem ent o f at least ten recom m endations is to m inim ize the possibility that a non-financial analyst en ters the database by co-authorship or assum es the role o f a Financial analyst tem porarily. The short length requirem ents ensure m inim um survivorship bias. There is disagreem ent as to the choice o f value- or equal-w eighted portfolios for tests o f perform ance over long horizons. If the purpose is to test for abnorm al perform ance, an equally w eighted portfolio w ill be more reasonable and tends to m agnify the significance o f abnorm al returns. To assess the wealth effect on investors o f follow ing recom m endations, value-w eighted portfolios are m ore appropriate. B ecause the w eights placed by analysts on their recom m ended stocks are unavailable, I present results for both value-w eighted and equally weighted portfolios. A benchm ark portfolio is created for individual financial analysts by m atching the recom m ended stocks in their “buy list" and “sell list” by size, book-to-m arket, and m om entum with the 4 x 4 x 4 quartile portfolios at the time o f recom m endation. The benchm ark portfolios are rebalanced w henever the original recom m endations change. Tw o-Period P erform ance Persistence Tests I use a w ide range o f two-period persistence tests to ensure robustness o f results. Perform ance-ranked portfolio tests sort analysts each period into decile portfolios on the basis of their perform ance over the preceding ranking period and then study the perform ance o f equally w eighted decile portfolios o f individual analyst portfolios over the subsequent evaluation period. The t-statistics o f the alphas on the difference portfolio betw een the zero-investm ent portfolios of the best- and worst-perform ing deciles in the 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ranking period determ ine the statistical significance o f perform ance persistence. C arp en ter and Lynch (1999) find this test to be the m ost pow erful one when there is no su rv iv o r bias. Chi-square tests, cross-product ratios (CPR), and Z -tests are nonparam elric tests based on a contingency table. They count the num bers o f w inners and losers separately relative to median perform ance in both the ranking and the evaluation periods, and ex am in e the independence o f rankings across two consecutive periods. For example, persistent performers are analysts w ho are winners (losers) in tw o consecutive periods and are denoted by W W (LL). T he perform ers who are w inners (losers) in the ranking periods but losers (w inners) in the evaluation period are thus denoted as W L (LW). The Chi-square statistic o f one degree o f freedom is [(VVW - /V /4 ): + (WL - N / 4 ) 2 +(LW - N / 4 ) 2 + ( L L - N / 4 ) 1] / N , w here N is the num ber o f analysts w ho survive in both periods. Carpenter and Lynch (1999) suggest that this test to be the m ost robust to su rv iv o r bias. The C P R equals (W W x L L ) / ( W L x L W ) , and the null hypothesis of independent perform ance corresponds to an odds ratio o f one. W ith independent observations, the standard error o f the natural log o f the odds ratio is J —— + — + —!— + — asym ptotically. Thus the log odds ratio divided by the standard V WW WL LW LL erro r has a standard norm al distribution. The Z-test for repeated w inners or losers is from a binom ial distribution. The Z -statistic equals (Y - np) / ^Jnp( 1 - p ) , where Y is the nu m b er o f winners or losers in the evaluation period, and n is the total num ber of w inners or losers in the ranking period, p is expected to be 1/2 if there is no persistence. W hen n > 20, Z is approxim ately standard norm al. This test is interesting because it 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. could distinguish betw een the persistence o f w inners and losers. A final test is somewhat related to contingency table tests. It regresses evaluation-period perfo rm an ce on rankingperiod perform ance cross-sectionally and tests the significance o f th e coefficient on ranking period perform ance. M issing Data, S u rv iv o r Bias, a n d Look-Ahead Bias Carpenter and Lynch (1999) suggest that m issing data c o u ld bias persistence test results. Fortunately, my sam ple does not have this problem . B row n, G oetzm ann, Ibbotson, and R oss (1992) show that when perform ance is c ro ss-sectio n ally heteroscedastic, and survival depends on single-period p erform ance, conditioning on survival biases tw o-period contingency table tests to indicate p erfo rm a n c e persistence w hen none exists. Y et if the survival criterion is based on m u lti-p erio d performance, conditioning on survival will bias results towards reversals [B row n e t al. (1992), G rinblatt and T itm an (1992), and Hendricks, Patel, and Z eck h au ser (1993)]. Li (2002) show s that an aly sts’ survival depends on their perform ance in the p rio r year, which suggests a survival criterion based on multi-period perform ance, g iv en the short time intervals used in this study. T his bias from the experim ental d esign o f two-period persistence tests should bias against finding perform ance p ersistence. Carhart, C arpenter, Lynch, and Musto (2000) carefully d istin g u ish survivor bias and look-ahead bias, both related to survivorship bias. S urvivor bias characterizes the effect o f including only analysts w ho survive till the end o f a sa m p le period. It is usually a limitation o f a data set itself and does not apply in my sam ple. L o o k -ah ead bias appears when we require a m inim um length o f an analyst portfolio d u ring ran k in g and evaluation 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. periods to estim ate performance. It is p articularly a problem for contingency table tests and cross-sectional regression tests. F o r exam ple, the length requirem ents during ranking periods exclude som e analysts and change both analyst perform ance distributions and ranks o f individual analysis. Length requirem ent in evaluation p erio d s is even more problem atic, b ecau se analysts who fail the length requirem ent and disappear early often underperform surviving analysts as seen in T able I. Li (2002) also finds that bad perform ance increases the risk of an a n a ly st leaving the profession. To reduce potential look-ahead b ias, I require short portfolio lengths o f three m onths (annual frequency), three m onths (sem iannual frequency), and two months (quarterly frequency) in each ranking o r evaluation period for c o n tin g en cy table tests and cross-sectional regressions. For the d ecile portfolio tests, the po rtfo lio length requirem ent pertains only in ranking periods. The d e c ile portfolios formed in ev aluation periods include all analysts in ranking period, w h atev er the length of their portfolios in evaluation periods, and decile portfolios are rebalanced w henever analysts disap p ear. The short length requirem ent should impose m inim um look-ahead bias on th e results. Tw o-Period Perform ance Persistence Statistical Significance o f Pooling Sam ple Panels A and B o f Table 2 report for tw o-period perform ance persistence results including cross-sectional regression, C hi-square, and cross-product ratio tests for valueand equal-w eighted analyst portfolios o f b u y and sell recom m endations, at annual, sem iannual, and quarterly intervals. T e sts using the overall sam ple pool all analyst- 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. Table 2. Two-period performance persistence over the whole sample period T ab le 2 rep o rts on tw o-period p erfo rm an ce p ersisten ce test results for the w hole sam p le p erio d from Jan u ary 1994 through D ecem b er 2000. 1 apply p aram etric (reg ressio n -b a sed ) and non p aram etric (C h i-sq u are and C P R ) tests at the quarterly, the sem ian n u al, and annual intervals for risk-adjusted p erfo rm an ce m easu res such as a lp h a and inform ation ratio and raw return m easures such as b u y -an d -h o ld retu rn s and av erage returns. A lpha is intercept o f the six -fa cto r m odel Rtl = a t + b t{JR F t + b h R r t_| + £ „ . w here Rn is ex cess return on the zero -in v estm en t p o rtfo lio s that are long the analystreco m m en d ed buy p o rtfo lio s and short the m atch in g p o rtfo lio s for the buy list, or arc sh o rt the an aly st-reco in m en d ed sell p o rtfo lio s and long the m atc h in g p o rtfo lio for th e sell list. T h e v ariab les in R F , include ex cess retu rn on the C R S P v alu e-w eig h ted N Y S E /A M E X /N A S D A Q m arket index, size, b o o k -to -m ark et, retu rn m o m en tu m , earn in g s/p rice, and liq u id ity factors. T he co efficien ts are estim ated using ord in ary least squares. 1R is alpha divided by the resid u al stan d ard d ev iatio n . B H R and A R are b u y -an d -h o ld returns and average re tu rn s on original a n a ly s ts’ p o rtfo lio s d u rin g specific intervals. P aram etric tests rep o rt the co efficien t on p rio r-p erio d p erfo rm an ce and h e tero sced asticity -co n sisten t t-statistics in reg ressio n s o f current period p erfo rm a n ce on p rev io u s p erio d p erfo rm an ce. T he C h i-sq u are test is d efin ed as [(VPW - N / 4 ) 2 + (\VL- N / A)' +(LW —N / 4 ) ‘ + ( L L - N / 4)2] / N , w h ere N is the n u m b er o f an aly sis a p p earin g in both cu rren t and p rev io u s periods, and W and L represent w inners and lo sers (individual an aly sis w ith p erfo rm an ce better o r w orse than m edian analyst p erfo rm an ce in each period). W W and LL d en o te w inners and losers in tw o co n secu tiv e periods; LW d e n o te s losers in p rev io u s p erio d s and w inners in cu rren t p erio d s and W L d en o tes the rev erse. T he C h i-sq u are statistic h as one d eg ree o f freed o m and eq u als 2.71 (3 .8 4 ) at the 10% (5 % ) sig n ifican t level. CPR is cro ss-p ro d u ct ratio and equals ln [ ( in v x LL)/(WL x L\V)\I JMWW + \IWL+\I LW + \/ LL . It has a standard norm al distrib u tio n. P an el A. Buy R eco m m en d atio n s __________ A n nual R eturns___________ Param etric N o n p aram etric __________ S em iannual R eturns_____________ P aram etric N o n p aram etric A lp h a-V W A lp h a-E W C ocff. 0.041 0 .0 5 0 t-stat. 4.0 b 4.93 C hi-sq. 20.18 9.25 CPR 4.49 3.04 C oeff. 0 .1 0 0 0.1 0 9 t-stat. 9.83 10.62 1R-VW IR -E W 0.034 0.032 3.39 3.22 13.71 3.44 3.70 1.85 0.064 0.071 B H R -V W B H R -E W 0.004 0.0 1 2 0.32 1.03 11.38 4.21 3.37 2.05 A R -V W A R -E W 0.0 0 2 0.003 0.1 6 0 .3 0 21.28 16.98 4.61 4.12 Q u a rte rly R eturns______________ Param etric N onparam etric C PR -0.38 2.39 C hi-sq. 13.82 28.00 C PR 3.72 5.29 C oeff. 0.016 0.0 3 0 t-stat. 3.03 5.39 C hi-sq. 0.15 5.72 6.77 7.49 12.06 20.04 3.47 4.47 0.021 0.034 4 .0 0 6.43 0.0 0 6.41 0.02 2.53 0.177 0.201 14.90 16.98 60 .7 6 74.9 0 7.79 8.64 0 .1 1 0 0.141 18.38 23.46 60.67 132.07 7.79 11.48 0.154 0 .1 7 0 13.85 15.41 67.79 89.70 8.23 9.46 0.093 0.122 16.07 21.04 65.69 131.14 8.10 11.44 Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. Table 2, continued P anel B. S ell R eco m m en d atio n s A nnual R eturns P aram etric N on p aram etric C oeff. t-stat. C hi-sq. CPR 0 .0 8 0 0 .0 0 2.85 0.02 0 .0 8 9 3.23 0 .0 0 0.06 S em ian n u al R eturns_____________ P aram etric N o n p aram etric C oeff. t-stat. C hi-sq. C PR 0.091 3.51 2.61 1.62 0.093 3.63 3.55 1.88 IR -V W IR-ENV 0 .0 3 6 0 .0 4 9 1.52 2.07 0.04 0.12 0.21 -0.35 0.037 0.037 1.53 1.56 1.06 2.76 1.03 1.66 0.013 0.016 1.06 1.28 1.03 0.00 -1.02 -0.06 B H R -V W B H R -E W 0 .1 1 9 0 .1 1 6 4.15 4 .1 0 5.82 5.27 2.41 2.29 0.222 0.237 8.53 8.90 27.58 18.55 5.24 4.30 0 .0 9 0 0.094 6 .5 0 6 .7 0 0.03 0.78 0.18 0.89 A R -V W A R -E W 0 .0 0 4 0.008 0.15 0 .3 0 2.97 1.08 1.72 1.04 0.089 0.097 3.48 3.75 4.41 5.23 2.10 2.28 -0.011 -0.007 -0.83 -0.4 9 2.89 0 .0 0 -1.70 0.01 A lp h a-V W A lpha-E W -J N) Q u arterly R etu rn s P aram etric N onparam etric C hi-sq. CPR C oeff. t-stat. -1.40 0.019 1.44 1.95 0.014 1.06 0.05 -0.23 interval observations and sh o u ld have greater power than perform ance persistence tests for each pair o f consecutive subperiods. I use zero-investm ent portfolios to m easure riskadjusted perform ance such as alpha and inform ation ratio and the analysts’ original portfolios to m easure the raw return perform ance. Generally, the results in Panel A suggest tw o-period perform ance persistence for both risk-adjusted and raw retu rn s for buy recom m endations at all intervals w ith few exceptions. At the annual interval, all test statistics are significant at the 5% level, except that the C hi-square test and C P R suggest persistent perform ance o f equally w eighted analyst portfolios on the basis o f inform ation ratios at 10% level, and cross-sectional regression suggests no perfo rm an ce persistence for raw returns. All the test statistics are also significant at the 5% level for the sem iannual and quarterly intervals w ith few exceptions. R isk-adjusted retu rn s seem to be the m ost persistent at the sem iannual intervals and the least p ersistent at the quarterly intervals, and raw returns are the most persistent at the quarterly intervals and the least persistent at the annual interval. Raw returns are m ore p ersistent than risk-adjusted returns at the sem iannual and quarterly intervals. Test sta tistic s are m uch higher for raw returns, indicating greater significance. For exam ple, at the sem iannual interval, the C hi-square test statistic on the basis o f value-w eighted b u y -an d -h o ld returns is 60.76. The sam e statistics on the basis o f alpha and inform ation ratio are 13.82 and 12.06, respectively. Although the result is not reported, I find perform ance persistence to be negatively correlated with e x te n t o f risk adjustm ents through factor m odels and the m atching m ethodology, w h ich co n firm s both the suggestion o f Brown and G oetzm ann (1995) that persistence m ay a p p e a r sim ply because m arket participants follow a com m on 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. strategy not captured by standard style categories or risk adjustm ent procedures and the observation of C arhart (1997) that perform ance persistence o f m utual funds disappears after controlling for m om entum . T his negative correlation also supports the relevance of exam ining the perform ance persistence o f raw returns: it provides an up p er bound for perform ance persistence tests. In addition, this result underscores the im portance of sufficient risk adjustm ents. Interestingly, perform ance persistence is robust to the adjustm ents o f both system atic and idiosyncratic risks o f individual analyst portfolios because almost all the tests indicate persistence on the basis o f the inform ation ratio. In Panel B for sell recom m endations, risk-adjusted perform ance show s little persistence. Few test statistics are significant at the 5% confidence level. Raw returns still appear to be som ew hat persistent, since buy-and-hold returns are persistent at the 5% significance level at both the annual and semiannual intervals and average returns are persistent at the 5% level at the sem iannual interval. Statistical Significance o f Persistence Tests by Consecutive Subperiods Although pooling the overall sam ple increases the pow er o f tests, results could be potentially affected by a sm all num ber o f consecutive subperiods w here there is strong perform ance persistence, and we w ould not be able to see perform ance reversals in some consecutive subperiods. Thus, I also conduct perform ance persistence tests for each pair o f consecutive subperiods. Panels A and B o f Table 3 report the percentage o f cases w here perform ance persistence is significant at a 10% confidence level. R esults are reported for value- and equal-w eighted analyst portfolios and for both buy and sell recom m endations at annual, sem iannual, and quarterly intervals. The zero-investm ent 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3. Two-period persistence test by pairs of consecutive subperiods T a b ic 3 rep o rts th e p ercen ta g e o f su b p e rio d s ex h ib itin g p erfo rm an ce p ersiste n c e significant at the 10% level fro m January 1994 th ro u g h D ecem ber 2 0 0 0 . I ap p ly p aram etric (re g re ssio n -b a se d ) and n o n p aram etric (C h is q u a re and C P R ) tests at the quarterly, th e sem ian n u al and a n n u al in te rv a ls fo r risk-adjusted p e rfo rm a n c e m e a su re s such a s a lp h a and inform ation ra tio and raw return m e a su re s su ch as buy-and-hold re tu rn s and av e ra g e return s. A lp h a is intercept o f the six -fa c to r m odel R tl = a , +b,0R F , +bh R F t l + £ i r , w h ere Rlt is e x c e ss return o n th e zcro -in v cstm cn t p o rtfo lio s that are long the an aly st-reco m m en d ed buy p o rtfo lio s an d sh o rt the m atch in g p o rtfo lio s for the b u y list, o r are short the an a ly st-re c o m m e n d e d sell p o rtfo lio s a n d lo n g the m atching p o rtfo lio for the sell list. T h e v ariab les in R Fl in c lu d e ex c e ss return on the C R S P v a lu ew eig h ted N Y S E /A M E X /N A S D A Q m ark et index, size, b o o k -to -m a rk e t, retu rn m om entum , e a rn in g s/p ric e , and liquidity facto rs. T h e coefficients a re e stim a te d using ord in ary least sq u ares. IR is alpha d iv id e d b y the re sid u a l stan d ard d e v ia tio n . B H R and A R a rc buy -an d -h o ld retu rn s an d av erag e returns on o rig in al a n a ly s ts ’ p o rtfo lio s d u rin g sp e c ific intervals. P a ra m e tric tests report the p e rc e n ta g e o f significant l-sta tislic s in reg ressio n s o f c u rre n t p erio d p erform ance m easu res on p rio r-p erio d p erfo rm an ce. The C h i-sq u are te st is d e fin e d as [ ( W W - N / 4 ) : + ( W L - N / 4 ) 2 + (L W - N / 4 ) 2 + ( L L - N / 4 ): ] / /V , w here N is th e n u m b e r o f a n a ly s ts ap p earin g in both cu rren t and p re v io u s periods, and W an d L re p re se n t winners and lo sers w h o are in d iv id u al a n aly sts w ith perform ance b e tte r o r w orse than m edian a n a ly s t p erform ance in e a c h p e rio d . W W an d L L den o te w in n ers and losers in tw o co n se c u tiv e periods; L W d e n o te s lo sers in previous p e rio d s and w in n e rs in c u rre n t p e rio d s and W L d en o tes th e rev erse. All n u m b ers are in percentage. A lp h a-V W A lp h a-E W Panel A . R eco m m en d ed Buy P ortfo l ios Sem iannual R etu rn s P aram etric C hi + Total 43 0 71 71 71 86 0 86 A nnual R eturns P aram etric C hi + T otal 0 33 33 50 50 0 50 50 Q uarterly R etu rn s C hi Param etric + T o ta l 38 24 52 62 52 19 38 71 IR -V W IR -E W 50 33 17 17 67 50 33 67 7! 57 29 14 100 71 43 71 29 43 24 29 53 72 52 33 B H R -V W B H R -E W 83 67 17 17 100 84 100 83 57 71 0 0 57 71 86 86 57 62 19 19 76 81 57 67 A R -V W A R -E W 50 50 17 17 67 67 100 83 71 71 0 0 71 71 71 100 52 62 24 19 76 81 62 76 Panel B. R eco m m en d ed Sell P o rtfo lio s Q uarterly R etu rn s C hi Parametric + T o tal 24 10 24 34 19 14 29 33 A nnual Returns P aram etric C hi + T otal 0 33 33 0 33 0 33 0 Sem iannual R etu rn s Param etric C hi + Total 14 29 14 43 14 29 14 43 IR -V W IR -E W 0 17 0 0 0 17 0 0 29 29 14 14 43 43 0 29 19 24 10 10 29 34 14 33 B H R -V W B H R -E W 33 33 0 0 33 33 33 33 71 71 0 0 71 71 71 57 33 38 5 5 38 43 29 24 A R -V W A R -E W 0 0 0 0 0 0 33 17 57 57 14 14 71 71 29 57 29 29 14 14 43 43 29 29 A lp h a-V W A lpha-E W 75 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. portfolios are used to m easure risk-adjusted perform ance and the original analyst portfolios to m easure raw return perform ance. Buy portfolios show persistent perform ance for both raw returns and risk-adjusted returns. Risk-adjusted returns are persistent at the annual and semiannual intervals, as test statistics are usually significant in the majority o f consecutive subperiods. T hey are the m ost persistent at the sem iannual intervals and the least persistent at the quarterly intervals. Raw returns are persistent at all intervals; here the annual intervals show the m ost persistent perform ance and the quarterly intervals the least persistent. T here are som ew hat perform ance reversals, as dem onstrated by a significantly negative slope in the cross-sectional regression o f analyst perform ance o n lagged value. T he percentage of significantly negative t-statistics is higher particularly at the quarterly intervals. Interestingly, the Chi-square test, w hich is the m ost robust to survivorship bias, suggests the highest percentage o f subperiods w ith persistent performance for both risk-adjusted and raw returns. Com pared to buy recom m endations, sell recom m endations show no riskadjusted perform ance persistence in Panel B. Raw return performance o f sells is persistent only at the sem iannual intervals. The results for pooled and consecutive subperiods are overall consistent with each other. The perform ance o f buy recom m endations is persistent, whatever the testing m ethodologies, return m easurem ent intervals, w eighting schemes o f portfolios o f analysts, and system atic and specific risk adjustm ents. Raw returns are m ore persistent than risk-adjusted returns. T he perform ance of sell recom m endations is persistent only for raw returns at the annual and sem iannual intervals. Interestingly, the perform ance persistence o f analysts does not a p p e ar to be fleeting, as analyst perform ance seem s to be 76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the m ost persistent at the annual and sem iannual intervals, rather than at the shortest quarterly interval. This finding em phasizes the im portance of the m ore pow erful m ulti­ period persistence tests below . Sources o f Performance P ersistence: W inners o r Losers T he two-way contingency tables in T able 4 show the num ber o f analysts classified as winners or losers in evaluation periods conditioned on their classification as w inners or losers in the preceding ranking periods. T he table also reports Z -statistics that separately test whether the perform ance o f w inners or losers is persistent. I report results for value- and equal-w eighted analyst portfolios at annual, semiannual, and quarterly intervals. Panels A and B present results for buy and sell recom m endations. Z-statistics in Panel A suggest that w in n ers’ performance is significantly persistent for the risk-adjusted return perform ance o f buy recom m endations. T his result is insensitive to portfolio w eighting schem es and return m easurement intervals, except for value-w eighted portfolios at the quarterly intervals. W inners’ perform ance persistence is overall m ore significant at the annual and sem iannual intervals. L osers’ risk-adjusted perform ance is not persistent at the quarterly intervals and for equally w eighted portfolios at the annual intervals. Turning to raw return perform ance, both w inners and losers show significantly persistent performance. W in n ers have more persistent performance than losers again. Raw return performance o f b oth w inners and losers is generally at least as persistent as risk-adjusted perform ance. T hese results are largely invariant to portfolio w eighting schem es and return m easurem ent intervals. 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4. Contingency table of winners and losers over the whole sample period T a b le 4 re p o rts the c o n tin g e n c y ta b le s o f w in n e rs a n d lo s e rs fro m Jan u a ry 1 9 9 4 th ro u g h D e c e m b e r 2 0 0 0 d u r in g th e w hole s a m p le p e r io d . R e s u lts a re re p o rte d a t th e q u a rte rly , the s e m ia n n u a l a n d ann u al in te rv a ls fo r risk -a d ju s te d p e rfo rm a n c e m e a s u re s s u c h a s a lp h a a n d in fo rm a tio n ra tio a n d r a w r e tu rn m ea su re s su c h a s b u y -a n d -h o ld re tu rn s a n d a v e r a g e re tu rn s . W a n d L re p re s e n t w inners a n d lo s e r s w h o are in d iv id u a l a n a ly s ts w ith p e rfo rm a n c e b e tte r o r w o rs e th a n m e d ia n a n a ly s t p e rfo rm a n c e in e a c h p e rio d . W -R d e n o te s th e w in n e r in the p re c e d in g ra n k in g p e rio d s an d W - E d e n o te s w in n e rs in the fo llo w in g e v a lu a tio n p e rio d s . L -R and L -E a re lik e w is e d e f in e d for lo s e rs . A lp h a is in te rce p t of th e s ix -fa c to r m o d el Rj, = or, +bl0R F , +hh R f f | +£ „. w h e re /?„ is e x c e s s r e tu r n o n the z e r o -in v e s tm e n t p o rtfo lio s th a t a re lo n g th e a n a ly s t-re c o m m e n d e d b u y p o r tfo lio s a n d s h o r t th e m a tc h in g p o rtfo lio s f o r th e b u y list, or a re sh o rt th e a n a ly s t-re c o m m e n d e d sell p o r tf o lio s a n d lo n g th e m a tc h in g p o rtfo lio fo r th e se ll list. T h e v a ria b le s in / ? f ( in clu d e e x c e ss re tu rn o n th e C R S P v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t index, siz e, b o o k -to -m a rk c t, re tu rn m o m e n tu m , e a rn in g s /p ric e , a n d liq u id ity factors. T h e c o e f f ic ie n ts are e stim a te d u s in g o rd in a ry lea st s q u a re s . IR is a lp h a d iv id e d b y th e re s id u a l sta n d ard d e v ia tio n . B H R a n d A R arc b u y a n d -h o ld retu rn s an d a v e r a g e r e tu r n s o n o rig in a l a n a l y s t s ’ p o rtfo lio s d u rin g s p e c if ic in terv als. T h e Z s ta lis tic e q u a ls (L and - np) / Jnp(\ - p) w h e re n is th e to ta l n u m b e r o f w in n ers o r lo s e r s in ra n k in g p e rio d s Y is n u m b er o f w in n e rs o r lo s e rs in the e v a lu a tio n p e rio d , p is e x p e c te d to b e p e rs is te n c e . W hen 1/2 if there is n o n > 2 0 a s in th is s a m p le , Z is a p p r o x im a te ly sta n d a rd n o rm al. P a n e l A . R e c o m m e n d e d B u y P o r tf o lio s A n n u a l R e tu rn s ____________S e m ia n n u a l R e tu rn s VW A lp h a W -E L -E Z -s ta t IR Z -s ta t L -R W -R 2706 2428 3 .88 2553 2732 2 .4 6 2681 2466 2691 2511 2 .5 0 2708 2559 L -E 2462 3 .4 2 2690 1.81 W -E 2690 2578 L -E 2460 Z -s ta t AR W -R W -E Z -s ta t BHR EW 3 .2 0 W -E L -E 2697 2415 3 .94 2691 1.56 2560 2747 2 .5 7 3 .0 0 2692 2508 2 .5 5 2708 2437 3 .7 8 VW L -R 2589 2683 1.29 W -R 2880 2634 3.31 L -R 2722 2867 1 .94 L-R 2 9 1 4 2676 2 5 9 7 2916 4 .2 7 3.21 2703 2876 2604 2885 2737 2911 2632 3.41 2849 1 .5 0 2613 4.01 2603 3021 2497 2597 2622 0 .3 5 2563 2711 2 .0 4 2992 2538 6 .1 1 2970 4 .9 2 2979 2599 2519 3006 5 .4 4 6 .2 0 VW W -R 2613 0 .1 2 Q u a r te rly R etu rn s EW 7 .0 5 3012 2475 7 .2 5 2.32 EW W -R 8610 L -R 8641 8565 0 .3 4 8525 0 .8 9 8642 8621 8548 0 .7 2 8530 0 .6 9 2599 8987 8298 2986 5.18 8151 6 .3 9 8905 4 .6 3 2578 3038 6.14 8974 8298 8947 4 .9 4 8122 6 .5 2 W -R L -R 8725 8 5 3 3 8416 86 6 7 2.36 1 . 0 2 8747 85 2 3 8413 8 6 5 8 2.55 1.03 9154 8 1 3 2 7974 9081 9.02 7 .2 3 8152 7958 9 1 0 3 8.95 7 .2 4 9128 78 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4, continued Panel B. R ecom m ended Sell P o rtfo lio s A n n u a l R e tu rn s ___________ S e m ia n n u a l R e tu r n s VW A lp h a W -E L -E Z -s ta t IR L -R W -R L -R W -R 473 503 474 469 502 506 L -R 494 445 479 456 1.61 515 507 454 0 .6 6 0 .9 3 0 .7 9 0 .7 3 W -R VW L -R 490 W -R 520 0 .9 4 1583 1.71 1527 1 .0 0 EW L -R 1639 W -R L -R 1554 1622 1584 0 .9 7 1554 1603 0 .3 3 0 .0 0 478 503 470 501 1539 1613 46 2 506 513 1576 1631 1587 1562 449 509 456 493 473 507 469 502 441 1556 1 .2 2 0 .9 6 0 .6 9 1 .2 2 1.29 0 .1 6 1.71 0 .6 3 0 .6 6 0 .7 8 0 .1 1 1602 0 .1 9 516 414 482 508 487 3 .34 412 488 3.1 7 0 .0 3 544 424 445 558 3.86 3 .5 7 535 434 3 .2 4 456 546 2 .8 4 1548 1555 0 .1 3 1604 1626 0 .3 9 1561 483 0 .0 3 W -E 498 486 512 481 514 1652 1545 1616 486 480 458 520 452 483 522 1511 425 2 .4 0 489 430 496 L -E 1588 1.95 0.51 1.73 1.23 1 .99 1.23 1550 0 .0 9 1622 0 .0 0 1582 1.28 W -E Z -s ta t Z -s ta t W -R 445 0 .9 2 Q u a rte rly R e tu rn s EW L -E L -E AR VW W -E Z -s ta t BHR EW 1 .1 2 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1538 0.41 1593 1641 0 .84 0 .1 1 Panel B indicates that although w inners in term s of sell recom m endations do not have persistent risk-adjusted perform ance, their raw return perform ance is persistent at the annual and sem iannual intervals. L osers’ perform ance is som ew hat persistent only for raw returns at the sem iannual intervals. T hese results suggest that the raw return persistence o f sells show n in Tables 2 and 3 is m ainly due to w inners. In addition to consistent results w ith previous sections, I also find in this section that perform ance persistence is largely attributable to winners, a pattern especially pronounced for risk-adjusted returns. T his finding suggests that we can use w inners’ perform ance persistence as an investment strategy to make a profit and lo sers’ persistence as a useful indicator to av o id som e analysts w ith poorer ability. E conom ic Significance a n d D ecile Portfolio Tests Som e clarification is im portant to distinguish three different types o f econom ic significance exam ined in previous research. C arhart (1997), and M alkiel (1995) sort mutual funds into deciles by raw return perform ance in ranking periods and study riskadjusted performance o f decile portfolios created in subsequent evaluation periods. Agarw al and Naik (2000) and Brown and G oetzm ann (1995) exam ine the relation betw een raw returns across ranking and ev alu atio n periods. Bollen and B usse (2002), Brown and Goetzmann (1995), and C hristopherson, Ferson and G lassm an (2001) investigate the relation betw een risk-adjusted perform ance across ran k in g and evaluation periods. Although all three types o f research study two-period perform ance persistence, the interpretations are quite different. The first one examines the predictability o f risk- 80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. adjusted perform ance on the basis of raw return perform ance in the previous ranking periods. The latter tw o study the correlations of raw returns and risk-adjusted returns, respectively, across tw o consecutive periods. I test all three types o f perform ance persistence, using the zero-investm ent portfolios to m easure risk-adjusted perform ance and the original analyst portfolios for raw return perform ance. Table 5 investigates the econom ic significance of persistence in risk-adjusted perform ance. I present annualized risk-adjusted returns on the decile portfolios created by investing an equal am ount in the zero-investm ent portfolios o f individual analysts w ithin each decile ranked on the basis o f perform ance in ranking periods. T he best- and the w orst-perform ing deciles are coded 9 and 0, respectively. T he annualized risk-adjusted returns are calculated as (1 + or)252 - 1, where alpha is the intercept o f the six-factor m odel whose dependent variable is the decile portfolio returns. For the buy recom m endations in Panel A, excess returns on decile portfolios decline nearly m onotonically w ith portfolio rank, except som etim es for the two w orstperforming decile portfolios, which suggests persistent risk-adjusted perform ance at all intervals despite som ething o f a perform ance reversal for the tw o w orst-perform ing decile portfolios. In addition, m any decile portfolios o f buy recom m endations exhibit significantly positive risk-adjusted perform ance. Buying the best-perform ing decile portfolios could earn returns as high as 16.4% per year if analyst portfolios are equally weighted and if the ranking is on the basis o f alphas at the annual intervals. These results corroborate the m ore persistent perform ance o f winners previously observed, and suggest that the significant reversal o f the tw o worst-perform ing deciles portfolios m ust have contributed 81 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5. Risk-adjusted performance of decile portfolios created according to prior-period risk adjusted performance T a b le 5 re p o rts the re la tio n b e tw e e n ris k -a d ju s te d p e rfo rm a n c e a cro ss the p r e c e d in g ra n k in g p e r io d s a n d th e s u b s e q u e n t e v a lu a tio n p e r io d s fro m Ja n u a ry 1 9 9 4 th ro u g h D e ce m b er 2 0 0 0 . R e s u lts fo r a n a ly s t b u y a n d se ll p o rtfo lio s arc in P a n e ls A a n d B . 0 (9 ) c o r r e s p o n d s to th e d e c ile p o rtfo lio fo rm e d in e v a lu a tio n p e rio d s th a t e q u a lly w eight p o rtfo lio s o f in d iv id u a l a n a ly s ts w h o p e r fo rm the w orst ( b e s t) in r a n k in g p e r io d s . 9 - 0 is th e d iffe re n c e p o rtfo lio b e tw e e n th e h ig h e s t- a n d lo w e s t- r a n k e d d ecile p o rtfo lio s . R e t. is th e a n n u a liz e d re tu r n s b a se d on in te rc e p t o f th e s ix -fa c to r m o d e l w ith d e c ile p o rtfo lio ex cess re tu r n s a s d e p e n d e n t v a ria b le . Thus R et. e q u a ls (I + or ) 152 - 1 an d is /?„ = o r , +bl0RFl +bljR Fl_l + £ t l , w h e r e /? ,., in p e rc e n ta g e . in c lu d e ex cess return The on s ix -fa c to r th e CRSP m odel is v a lu e -w e ig h te d N Y S E /A M E X /N A S D A Q m a r k e t in d e x , siz e , b o o k - to - m a r k e t, return m o m e n tu m , e a rn in g s /p r ic e , a n d liq u id ity factors. T o o b ta in a lp h a a n d IR , th e d e p e n d e n t v a ria b le o f the s ix -fa c to r m o d e l is e x c e s s r e tu r n on th e z e ro -in v e s tm e n t p o rtfo lio s th a t a re lo n g th e a n a ly s t- r e c o m m e n d e d buy p o rtfo lio s a n d s h o r t th e m a tc h in g p o r tf o lio s for the b u y list, o r a re sh o rt th e a n a ly s t- r e c o m m e n d e d sell p o r tf o lio s a n d lo n g th e m a tc h in g p o r tf o lio for the sell list. A lp h a is th e in te rc e p t o f th e s ix - f a c to r m odel w h e re e x c e s s re tu rn s o f in d iv id u a l a n a ly s t p o rtfo lio s a re d e p e n d e n t v a ria b le s. IR is a lp h a d iv id e d b y residual s ta n d a rd d e v ia tio n . T h e s ix - f a c to r m o d e l is a lw ay s e s tim a te d u s in g o rd in a ry le a st s q u a r e s . H c te ro s c e d a s tic ity -c o n s is te n t t-s ta tis tic s m e a s u r e th e s ig n ific a n c e o f e x c e s s r e tu r n s [W h ite (1 9 8 0 )]. P a n e l A . R e c o m m e n d e d B uy P o r tf o lio s (a) A n n u a liz e d R e tu rn s o n P o rtfo lio s S o rte d b y A lp h a in the P r e v io u s P e rio d A nnual R e tu rn s VW 0 1 R e t. t-stat R et. 7 .2 2.5 2 .8 1 .6 7 .2 5 .0 2.5 2 1.4 0 .9 3 4 1 .2 0 .8 1 .8 1 .2 5 3.4 3.4 6 7 S e m ia n n u a l R e tu rn s EW t -stat t-s ta t VW EW R et. 4 .7 t-sta t 1.9 R et. 3 .6 4 .8 0 .7 1.3 2.7 0 .5 I-sta t R et. 2.7 t-s la t 4.1 3 .4 2 .5 2.5 2 .6 1 .6 0 .6 2 .8 2.4 0.3 2.7 3.0 1.5 1 .6 1 .0 0 .2 3.3 3 .0 2.4 2 .6 1 .2 0 .8 2 .3 1.9 2 .8 2 .0 2 .6 2 .2 3 .2 3.4 5 .8 2.5 2.7 2.5 2 .7 2 .0 2 .6 1.7 1.9 4.1 5.5 8.4 3.3 3.7 1 .6 1.1 3.5 2.4 2.7 2.4 4 .7 4 .0 3 .6 3 .0 2 .8 6 .2 2 .6 5.3 4 .4 3.7 3.3 5.1 4.3 6.7 11.7 2 .9 3 .6 8 .6 4 .5 2.5 3.4 5 .4 16.2 5 .3 9 .8 2.7 2 .9 11.7 4 .8 3 .9 1 0 .0 2 .2 10.9 2 .6 6 .0 1.3 8 .8 2 .1 0 .8 2 .6 2 .0 9 5 .2 7 .7 14.8 4 .2 8.9 16.4 3 .9 4 .3 5 .6 9 -0 7.1 1.7 8 .6 2 .2 8 R et. Q u a rte rly R e tu rn s EW VW 4 .9 2 .0 2 .4 3 .4 2 .3 (b ) A n n u a liz e d R e tu rn s o n P o r tf o lio s S o r te d b y IR in the P re v io u s P e rio d 0 1 *? 3 .9 4 .5 1.7 2 .6 1 .6 4.8 4.4 2 .0 1 .6 0 .8 2 .1 1 .0 2.9 1 .2 1 .6 2 .8 0.7 0 .4 3 .6 2.4 2 .1 1 .2 1.7 1 .1 3 .0 0 .0 1.7 3.7 1.1 2.4 1.4 1 .8 -0 . 1 4.5 2 .6 2 .8 1.7 3 .0 2 .9 3.8 3.7 4 .0 6.7 2.5 1 .6 4 .5 4 .2 3.7 4 .3 2 .8 3.4 4 .2 4 .3 2.5 2.4 6 .6 2.5 4 .4 1.9 2 .4 7 .3 7 .8 1 .0 6 .1 3 1 .8 1 .0 4 2 .6 1.7 4 .8 2.7 5.4 5 3.3 5.5 6 5 .3 4 .4 2.9 7 5 .6 3.1 7.3 5.5 8 5 .3 9 1 1 .8 2.9 3.7 7 .6 1 .8 9 -0 1 .0 2.7 2 .7 4 .8 4 .2 5 .5 3.4 3.8 4 .7 5.7 3.5 8 .2 5 .6 6 .2 5.9 7.1 3 .9 4 .0 4 .9 6 .9 3.5 2 .9 2 .3 6 .0 5.8 12.3 3 .6 3.4 1 0 .8 4 .3 4 .0 6 .9 7 .2 1.9 5.2 1.2 8 .6 2 .4 3 .9 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 .4 2.7 3 .9 4 .0 3.3 1.8 T able 5, continued Panel B. R ecom m ended Sell P o rtfo lio s (a) A n n u a li/.e d R e tu rn s o n P o r tfo lio s S o rte d by A lp h a in th e P re v io u s P eriod A n n u al R e tu rn s VW S e m ia n n u a l R e tu rn s EW VW Q u a rte rly R eturns VW EW Ret. 18.0 t-s ta t 2 .3 - 1 .8 -0 . 2 1.5 -5.1 7.8 -2 . 6 - 1 .0 4 .9 1.1 2.9 0 .7 1.9 -5 .2 -4 .9 - 1 .2 -1 .4 - 1.1 -0.4 -0 .3 -0 . 1 -1 .5 6 .4 -0.5 -3.8 - 1 .2 3 .0 2 .4 0 .8 1.9 0 .5 - 1 .1 -0 . 1 -1.3 -5.6 0.7 - 1 .0 0 .1 -0 . 8 -0 . 1 R et. 8.4 t-stat 0 .9 R et. 12.9 I-sta t 1.4 R e t. 15.4 t-s ta t 0 1 -7 .9 -1.5 -4 .7 -0 .9 -1 .3 2 1.1 0 .2 -3 .4 -0 . 8 -6 .5 4 .0 3 4 4 .6 -2.7 1.0 8 .2 1 .8 -0.7 -0 .4 5 2.7 -0 .9 0 .7 -1.5 0 .7 -0.3 -4.1 - l.l -0 .9 -0 . 2 -2 .1 -7 .6 -0 . 1 - 1.2 -0 .4 0 .2 2 .1 -1 .3 0 .4 0 .8 - 1 .2 -0 . 1 0 .0 0 .0 0 .1 0 .0 1 .2 0 .1 -5 .4 -0 .9 -0.7 -1 1 .5 -0 . 8 - 1 .2 -14.2 - 1 .2 1 .0 6 7 8 -5.5 -5 .6 9 -2.3 9 -0 -9 .9 0 .2 0 .3 -0 .5 -0 .7 -2 .4 -5 .8 -1 3 .3 2 .0 0 .9 0 .1 -0 . 2 -0 . 2 -0 .7 EW R e t. 0.3 -0 . 8 0 .1 -2 . 0 -0 .4 -4.3 - 1 .2 t-sta t Ret. 1 .8 6 .1 t-s ta t 0 .2 1 .8 0 .5 -0 .3 0 .1 (b ) A n n u a liz e d R e tu rn s o n P o r tf o lio s S o rte d by IR in th e P r e v io u s P e rio d 0 .4 0 .4 0 .1 0 .7 -0 .7 0 .1 4.3 - 1.1 0 .3 0 .7 -2 . 8 -0.5 -0 .9 -0 . 2 0 .8 0 .8 0 .0 3 .6 -4 .4 -0 . 8 8 3.6 -0.3 1.4 -7.7 1.9 -4.1 -0 . 8 0 .3 -1.5 -2 .6 -2 . 2 -0.5 -0 .4 9 -8 . 2 - 1 .2 -6 . 2 9 -0 -9 .9 - 1 .2 -6 . 6 0 1 2 3 4 5 6 7 1.9 -6 . 0 1.7 6 .1 -0 . 1 1.1 4 .5 6 .5 1.9 -4 .2 1 .0 1.3 0 .4 - 1 .0 0 .4 5.2 10.5 -0 . 1 -1.4 1 .2 2 .1 -2 .7 -3 .9 - 1 .u -0 .7 -0 . 8 -2.5 -1.9 2 .6 - 1 .2 3.2 -0 .3 0 .7 -6 . 2 1.4 - 1 .1 0 .3 6 .1 -0 . 2 -0 . 2 -0 .3 1.3 11.9 2 .1 3 .6 -5.1 -0 .9 -0 . 1 0 .7 - 1 .0 5.6 -3.8 -0 . 8 -4 .5 -0 .7 -0 .7 -8 . 6 - 1.1 0 .0 -1 .5 -0 .6 1.7 -0 .4 0 .5 0 .3 0 .2 0 .0 6 .6 1.4 4.4 -0 . 2 0 .8 1 .0 2 .2 -1 .4 0 .4 -0 .3 -3.0 -0 .7 -6.3 -0 . 8 - 1 .0 - 1 0 .0 -2 . 0 -7.1 -1 .5 -10.9 -1 .4 -7 .6 - 1 .2 -4.7 -0 . 8 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 0 .0 to the less persistent perform ance o f losers. M oreover, risk-adjusted returns are almost alw ays greater when d e c ile portfolios are created according to past alphas o r when analyst portfolios are equal-w eighted. T o assess the statistical significance o f the econom ic profits, I focus on alphas of the performance d ifference in evaluation periods betw een the top a n d bottom perform ance analysts ranked in preceding ranking periods (9-0). T e st statistics overall suggest persistence in risk-adjusted returns at the annual and sem iannual intervals. I find persistent perform ance at the quarterly intervals only when analyst portfolios are equally w eighted, which su g g ests perform ance persistence to be the least sig n ifican t at the quarterly intervals. Turning to P anel B, perform ance o f sell recom m endations o f analysts does not seem to be persistent a t all. There is no m onotonic relation betw een rank and perform ance, and alp h as o f the difference portfolios are never statistically significant. Results in T ab le 6 suggest that raw return perform ance in ranking periods cannot predict risk-adjusted perform ance in evaluation periods. No m onotonic relation exists betw een rank and p erform ance, and no alpha o f the difference portfolios is significant for either buy or sell recom m endations. Table 7 exam ines the econom ic significance o f the observed raw returns persistence. B uy-and-hold returns on decile portfolios created by investing an equal am ount in the original portfolios of individual analysts within each decile are ranked on the basis o f raw return perform ance in ranking periods. Perform ance persistence is almost alw ays highly significant for buy recom m endations, except for som e perform ance reversal for the w orst-perform ing decile portfolio at the annual intervals. Annualized buy- 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6. Risk-adjusted performance of portfolios created according to prior-period raw return performance T a b le 6 rep o rts the p r e d ic ta b ility o f r is k -a d ju s te d p e r f o r m a n c e in the s u b s e q u e n t e v a lu a tio n p e rio d s o n th e b a s is o f raw re tu rn p e r fo r m a n c e in th e p r e v io u s ra n k in g p e r io d from J a n u a ry 1 9 9 4 th r o u g h D e c e m b e r 2 0 0 0 . R e s u lts fo r a n a ly st b u y a n d s e ll p o r tf o lio s a r e in P a n e ls A a n d B. 0 (9 ) c o r re s p o n d s to th e d e c ile p o rtfo lio fo rm e d in e v a lu a tio n p e rio d s th a t e q u a lly w e ig h t p o r tf o lio s o f ind iv id u al a n a ly s ts w h o p e rfo rm the w o rst ( b e s t) in ra n k in g p e rio d s . 9 - 0 is th e d iffe re n c e p o r tf o lio b etw een th e h ig h e s t a n d lo w e s t ran k ed d e c ile p o rtfo lio s . Ret. is th e a n n u a liz e d re tu r n s b a s e d o n in te r c e p t o f the s ix - fa c to r m o d e l w ith c rea te d d e c ile p o r tf o lio ex ce ss re tu rn s as d e p e n d e n t v a ria b le . T h u s R e t. e q u a ls ( 1 + a ) 252 f a c to r m o d el is Afir = + bl0R F , +b uR F - 1 a n d is in p e rc e n ta g e . T h e six- +£ tl, w h e r e RfJ in clude e x c e s s r e tu r n o n th e C R S P v alu e- w e ig h te d N Y S E /A M E X /N A S D A Q m a rk e t index, s iz e , b o o k -to -m a rk e t, re tu rn m o m e n tu m , e a rn in g s/p ric e , a n d liq u id ity fa cto rs. B H R a n d A R a re b a s e d on o rig in a l a n a ly sts’ p o rtfo lio s d u r i n g th e sp e c ific in terv als. T h e six -fa c to r m o d e l is a lw a y s e s tim a te d u sin g o r d in a r y least sq u a re s . H e le r o s c e d a s tic ity - c o n s is lc n t ts ta tis tic s m easu re th e s ig n if ic a n c e o f e x c e s s re tu rn s ( W h ite ( 1980)|. P a n e l A . R e c o m m e n d e d B u y P o r tf o lio s (a ) A n n u a liz e d R e tu rn s o n P o rtfo lio s S o r t e d b y B H R in th e P r e v io u s P e rio d A n n u a l R e tu rn s Ret. 0 1 0 .2 t-sta t 3.4 R e t. 14.1 VW l-s ta t 4 .6 R e t. 8.9 4 .9 4 .7 4.1 1 5.8 2 .8 6 .1 3.5 2 2 .8 2 .5 1 .6 3 4 5 .0 1.7 3 .0 1.1 5 .5 4 .2 3 .6 1 .6 5 2 .0 6 2.9 3.8 7 2 .0 8 2.4 9 1 0 .6 9 -0 3.1 3.1 3 .2 3 .0 2.3 EW VW i -stat R et. t-stal 3.1 9.3 5 .9 3.3 3.5 4 .0 2 .6 2 .8 2 .9 2.3 5 .9 2 .7 3.5 4 .5 5.1 t -sta t 2 .2 1 .0 0 .7 2 .8 2 .1 3.4 3 .0 2 .1 4.5 2.5 2.7 2.4 1.3 2.3 1.7 0 .9 1.4 3.3 4.3 0 .8 4.5 1.9 2 .2 1 .2 1.3 3 .3 1 .8 1.9 2.4 1 .2 3.5 3.1 3.7 2.5 1.7 3.3 1 1 .8 4 .0 7.5 2.3 10.5 3.6 6 .6 0 .2 -2 .3 -0 . 8 -1 .4 -0 . 1 1 .2 0.4 -2 .4 2 .9 2 .0 1 .6 1 .2 EW R e t. 9 .0 4 .6 5 .7 4 .3 3 .3 4 .6 2.3 0.4 _________Q u a r te r ly R e tu rn s S e m ia n n u a l R e tu rn s EW VW 2 .6 2 .1 3 .2 1.5 -0 .6 R et. 8.7 4.5 1 0 .6 1.9 t-stat 2.7 2.4 2 .9 1.9 2 .2 1.9 2.7 3.1 2.4 3.4 0 .0 (b ) A n n u a liz e d R e tu rn s o n P o rtfo lio s S o r te d b y AR in th e P re v io u s P e rio d 0 1 2 3 4 11.5 3.4 3.7 1 3 .4 4 .3 8 .8 3.2 8 .8 1.8 6 .7 3.8 4 .6 4 .4 2 .7 2.5 5.3 5.1 2.7 7.7 2.5 5.4 2 .9 3.5 2 .2 1.9 2 .5 1.3 2.5 4.4 3.5 2.5 3.4 4.4 2 .2 1 .6 2.4 1 .8 2 .9 2 .0 2.3 1 .8 1.1 0 .8 3.4 3.9 2.7 2 .6 1 .8 1 .2 2 .2 5 .4 3.7 3 .8 3 .0 2 .9 2 .1 1 .6 2 .9 2 .1 4.1 2 .8 3 .0 2 .5 1 .6 4 .5 6 .5 2.5 2.5 2 .6 1 .6 2 .8 1.9 2.3 1.1 1.9 1.5 3.1 0 .9 1.4 5.4 2 .8 4.2 6 .5 1.9 1 0 .8 3.5 0.4 -2 . 2 3.1 0 .2 2 .8 2 .0 2 .0 1.5 2.4 1 .6 6 2.3 2.7 1.4 2 .6 9 5.6 10.9 4 .8 4 .3 5 .0 3.3 13.3 4 .3 9 .3 2 .8 3.7 12.4 9 -0 -0 . 6 0.1 -0 . 1 -0 .4 0 .5 0.3 3 .6 8 2 .6 2 .4 4.6 3.5 5 7 8.7 5 .2 4 .2 3.3 3.4 1.4 - -0 .6 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 .6 2 .8 Table 6, continued Panel B. R e c o m m e n d e d Sell P o rtfo lio s (a ) A n n u a liz e d R e tu rn s o n P o rtfo lio s S o rte d b y B H R in th e P re v io u s P erio d A n n u a l R e tu rn s VW 0 R e t. - 9 .0 1 - 1.1 S e m ia n n u a l R e tu rn s EW t-s ta t -0 .9 -0 . 2 R et. t-sta t -8 .9 -0 . 8 -3 .6 -0 . 6 0 .4 3.4 0 .7 1.5 Q u a rte rly R e tu rn s VW EW EW VW t s ta t -1 .7 R e t. t -stat - 1.1 R e t. -17.1 4 .3 t-sta t - 1.6 0 .7 3 .9 0 .6 1.5 0 .3 0 .7 -4 .2 -0 .9 0 .0 -1 .9 -0 .9 -5 .7 R e t. -1 5 .7 -9 .5 R e t. -1 0 .9 t- s ta t - 1 .2 0 .1 2 .4 1.4 0 .4 6 .8 2 .3 0 .6 -0 . 1 5 .9 1.4 -8 . 1 -3 .7 2 .0 0 .6 0 .0 8 .1 2 .2 2 .0 0 .6 4 .2 1 .2 -0 . 2 1.5 -0 .3 5 .8 1.4 1 .1 0 .3 0 .1 1 .6 -0 . 8 1.7 -0 .4 -0 . 1 5 .5 4 .8 -5 .0 7 3 .0 0 .8 0 .8 0 .0 8.7 2 .2 7 .3 1 .8 8 .9 2 .1 -3 .5 -2 . 6 0 .0 8 3.3 7 .0 2 .2 1.7 0.4 0 .6 0 .5 7 .3 2 .4 2 .6 2 .8 -0 . 6 -2 . 8 -0 . 1 9 0 .5 - 1 .0 -0 . 6 0 .4 0 .3 -0 .3 0.5 6 - 0 .9 6 .4 -0 . 2 - 1 .0 4 .5 0 .7 15.4 1 .2 -4 .3 -0 .5 -0 .3 -2 . 6 -4 .4 -0 . 6 - 1 .2 2 3 4 5 9 -0 1 1 .8 0 -0 .5 -2 . 0 -2 . 2 1 2 3 4 5 1.5 0 .4 2 .1 0 .6 1 .2 1.4 11.9 1.3 1 1 1 14.3 1 .2 17.8 (b ) A n n u a liz e d R e tu rn s on P o rtfo lio s S o rte d b y A R in th e P re v io u s P erio d 0 .0 -0 .3 -0 .4 -3 .7 -2 . 2 -0 . 8 -0 .3 -0 .3 -0 . 1 .6 -0 .7 -6 .9 -0 .7 2 .2 0 .4 0 .1 - 1.1 0 .3 -0 .3 0 .9 2 .5 0 .1 1 .6 0 .5 0 .4 0 .9 2 .9 0 .8 0 .5 -3 .7 1 .1 -0 .4 1.1 1 .0 - 1 .2 4 .6 1.1 2.4 - 1 .1 0.3 0.7 -2 . 6 -3 .0 -6 .3 -0 .7 -0 .7 -0 .9 5 .6 9.1 0 .6 0 .3 -6 2 .1 0 .5 0 .0 0 .0 4 .3 -0 .3 4 .0 1.4 1 .0 8 .8 2 .2 0 .8 0 .3 0 .3 -0 . 8 1.4 3 .5 3 .9 4 .3 -2 .4 -3 .0 -0 . 1 1 .2 2 .0 9 7 .7 -3 .6 -0 . 2 0 .4 0 .3 -0 .4 -2 . 2 -0 .3 -2.3 -0 . 6 -0 . 6 -0 .3 9 -0 -3.1 -0 .3 1.5 -0 .5 4 .3 0 .2 6 7 8 -0 .5 1.1 -7 .0 4 .6 -6 . 8 -0 .9 0 .9 - 1 .6 0 .1 -3 .2 -1 .3 -0 . 8 1 .6 1.4 -1 .5 0 .0 0 .5 0 .9 -0 .7 1.4 1.9 1.5 3.3 -2 .3 5 .2 4 .7 -1 .9 -0.3 -0 .9 -0 . 1 5.1 0.3 3.4 0 .3 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.3 1 .0 Table 7. Buy-and-hold returns of portfolios created according to previous period raw return performance T a b le 7 re p o rts th e r e la ti o n b e tw e e n ra w re tu rn p e r fo r m a n c e a c ro ss th e p re v io u s ran k in g p e r io d s a n d th e s u b s e q u e n t e v a lu a tio n p e r io d s fro m J a n u a ry 19 9 4 th r o u g h D e c e m b e r 2 0 0 0 . R e s u lts for analyst b u y a n d sell p o rtfo lio s are in P a n e ls A a n d B. T h e ra w r e tu r n m e a s u re s in c lu d e b u y -a n d -h o ld returns a n d a v e ra g e re tu rn s . T h e r e p o rte d n u m b e r s a re a n n u a liz e d ra w re tu rn s . 0 (9 ) c o r re s p o n d s to th e d ecile p o rtfo lio fo rm e d in e v a lu a tio n p e rio d s t h a t e q u a lly w eig h t p o r tf o lio s o f in d iv id u a l a n a ly s ts w h o p e rfo rm the w o rs t (b e st) in r a n k in g p e rio d s. Panel A . R e c o m m e n d e d B uy P o rtfo lio s S e m ia n n u a l R e tu rn s A n n u a l R e tu r n s BHR 0 1 ~t 3 4 AR Q u a rterly R e tu rn s BHR AR BHR EW VW EW VW EW VW EW VW EW VW EW 30.1 2 4 .6 3 2 .9 2 5 .0 33.5 2 3 .0 19.0 1 9 .9 2 0 .3 18.4 17.1 16.5 17.0 16.7 16.5 16.0 2 3 .3 23.7 2 6 .5 2 3 .0 26.1 3 0 .5 2 1 .9 2 3 .4 2 1 .8 24.5 24.6 24.1 2 2 .4 2 1 .0 2 1 .7 16.1 17.5 16.5 17.0 2 0 .8 2 1 .4 2 0 .6 21 2 3 .3 2 4 .9 26.1 1 9 .0 2 3 .9 2 6 .9 2 6 .7 2 2 .1 2 4 .8 2 5 .7 2 2 .9 26.1 7 2 6 .7 2 6 .0 27.1 2 7 .4 2 6 .4 2 7 .7 8 2 7 .2 3 4.3 2 7 .9 2 6 .3 3 4 .7 2 7 .6 2 8 .0 2 9 .2 3 5 .7 27.9 36.3 2 7 .7 4 0 .4 3 2 .4 2 9 .8 3 9 .4 3 2 .2 4 1 .3 17.1 4 .7 18.2 6.5 8 .6 2 15.5 14.6 12.7 3 4 1 0 .0 10.3 11.5 7 .3 13.7 16.7 1 1 .1 11.9 17.9 2 0 .0 14.5 12.9 11.7 7 .3 15.1 1 0 .2 5 13.3 8 .7 1 1 .8 14.7 18.6 9 .6 2 2 .4 9 2 7 .0 26.5 2 0 .1 19.7 6 25.1 2 4 .8 2 5 .8 2 4 .6 2 5 .8 2 0 .6 19.6 19.0 25.1 2 4 .5 2 3 .2 5 AR VW 4 0 .9 2 0 .6 19.3 20.9 2 2 .9 2 3 .2 2 4 .8 23.2 24.9 26.4 3 1 .7 3 9 .6 33.4 4 5 .2 15.3 14.9 14.0 14.7 14.3 15.7 15.4 2 18.3 19.7 2 0 .8 2 3 .9 2 3 .6 2 3 .7 3 3 .4 3 7 .9 19.0 18.3 2 0 .9 2 4 .3 2 2 .9 28.1 3 3 .4 4 4 .4 Panel B . R e c o m m e n d e d S ell P o rtfo lio s 0 1 6 7 8 9 12.3 13.9 15.2 1 7 .0 1 1.7 17.0 11.9 12.4 24.7 14.5 5 .0 4 .6 11.5 7 .7 2 0 .9 12.4 8 .2 7 .5 4 .5 4 .7 5 .2 5 .8 6 .9 9 .4 1 0 .1 9.1 8.9 9.1 19.9 1 6 .0 18.5 16.9 15.8 19.9 16.1 2 3.4 2 8 .0 13.5 19.6 2 7 .7 17.2 2 1 .1 17.3 18.6 14.7 16.6 16.6 2 0 .6 2 8 .9 8 .2 8.7 13.4 14.3 15.5 16.0 1 2 .8 16.9 14.3 18.1 15.8 18.5 2 1 .5 16.2 1 1 .0 1 2 .1 1 0 .6 1 1.5 9.1 4 .5 17.5 10.4 17.7 11.5 13.1 7 .9 18.4 1 0 .0 1 0 .0 9.2 87 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 .8 16.1 18.5 and-hold rates o f returns on the best-perform ing decile portfolios c a n be as high as 36.3% , 41.3% , and 45.2% at annual, sem iannual, and quarterly intervals, respectively. By com parison, the sam e rates of returns are 19.7%, 20.0% , and 2 1 .3 % on the dividends reinvested S& P 500 index. C om pared to buy recom m endations, an aly st perform ance in sell recom m endations never seem s to be persistent. In addition to results consistent w ith previous section, I fin d that it would be a highly profitable investm ent strateg y for investors to hold, during the evaluation period, the stocks recom m ended as buys by the best-perform ing analysts in the ranking period. Investors can earn significant risk-adjusted returns in evaluation periods if we use riskadjusted returns to classify analysts in ranking periods and earn significant raw returns if we rank analysts on the basis o f raw returns. Even with high p o rtfo lio turnover, profit should still substantially outw eigh potential transaction costs. T h e s e results also suggest that the W SJ all-star ranking m ay be an inexpensive list for in v esto rs interested only in raw returns to create trading strategies, because better raw return perform ance persists at the annual intervals. M ulti-Period Perform ance Persistence T o investigate m ulti-period perform ance persistence, I use an improved m ulti­ period test based on the original w o rk o f A garw al and Naik (20 0 0 ). T he m ulti-period persistence test should have greater po w er to differentiate betw een persistence attributable to chance or attributable to superior ability. For e x a m p le , there is m uch less chance o f w inning by luck over five periods than over two periods. A nother advantage o f this test is that it does not require a very long sam ple period as trad itio n al tests do for 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. tests of long-run perform ance persistence. It thus does not require survival in both ranking and evaluation periods, and should not introduce bias into test results like the traditional tw o-period persistence tests. Agarwal and Naik (2000) first construct a series o f w ins and losses for each hedge fund in their sam ples and then com pare observed frequency distributions with the theoretical frequency distributions o f tw o and m ore consecutive wins and losses. For exam ple, under the null hypothesis o f no persistence, the theoretical probability o f observing W W and LL is equal to one-fourth and that o f observing W W W and LLLL equals one-sixteenth, and so on. They use a tw o-sided tw o-sam ple K olom ogorovSm im ov (K -S) test to compare the observed distribution o f w ins and losses w ith the theoretical distribution. The null hypothesis is F = G against an alternative that F is not equal to G, w here F and G are both distribution functions. Let Fn and Gm be distribution functions corresponding to independent sam ples X,, X 2 X n and Y{, Y2 Ym , respectively. T h e K-S statistic is: £L= S«P \ Fn(t) - Gm(t)\ (3) — *»> G F< G VW EW EW VW W in Loss 0.04 0.03 0.05 0.05 EW VW W in Loss W in Loss 0.02 0.02 0.0 0 0.01 0.0 6 0.06 0 .0 2 ** 0 .0 2 * * * W in 0.01 0.02** L oss 0.01 W in 0 .0 0 Loss 0 .0 0 W in 0 .0 0 Loss 0.00 0 .03*** 0.02*** 0.03*** 0.02*** 0 .0 2 * * * IR F > G F< G 0.05 0.03 0.0 6 0 .02 0.02 0 .06 0.02 0.00 0.07 0 .02 ** 0.01 0 .0 3 * * * 0.0 0 0.02** 0 .0 0 0.03** * 0.00 0.02*** 0 .0 0 0.02*** 0 .0 0 0.0 1*** 0 .0 0 0 .0 2 * * * BHR F > G F< G 0.00 0.04 0 .0 0 0.07 0 .02 0.04 0.00 0 .02 *** 0.07 0.01 0 .0 2 * * 0.02 0.03*** 0.01 0.03*** 0.01 0.01 0 .0 0 0.01** 0 .0 0 0.01** 0.0 0 0 .0 2 * * * 0 .0 0 AR F > G F< G 0.01 0.04 0.0 0 0.04 0.02 0.05 0.0 0 0.0 2 ** 0.05 0.01 0 .0 2 * 0.01 0.02*** 0.01 0.03*** 0.01 0.01 0.00 0.01* 0 .0 0 0.01 0 .0 0 0 .0 2 * * * 0.00 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 8, continued P a n e l B. R e c o m m e n d e d S ell P o rtfo lio s A n n u a l R e tu rn s ______________________ S e m ia n n u a l R e tu rn s _____________________________ Q u a rte rly R e tu rn s VW A lte rn a tiv e A lp h a IR BHR AR VW EW L o ss 0 .0 3 0 .0 0 F < G 0 .1 0 0 F> G F < G 0 .0 0 .11* 0 .0 3 0 .0 0 0 .1 0 0 0 .1 0 0 F > G F < G 0 .0 3 0.01 0 .0 8 0 .0 4 0 .0 0 0 .0 2 0 .0 2 0 .0 2 0 .0 1 0 .1 0 0 .1 0 0 .0 7 0 .0 3 0 .0 3 0 .0 4 * * 0 .0 3 F > G 0 .0 3 0 .0 9 0 .0 0 0 .0 9 0 .0 4 0 .0 0 0 .0 1 0 .0 0 0 .0 2 0 .0 0 0 .1 0 0 .1 0 0 .0 4 * 0 .0 4 ** 0 .0 5 * * * 0 .0 4 * * 0 .0 3 .1 1 * W in L o ss 0 .0 4 0 .0 0 0 .1 1 * 0 .1 0 * .1 1 * W in L o ss W in EW VW W in F < G so w EW H y p o th e s is F > G L o ss W in L o ss W in L o ss 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 4 ** 0 .0 5 *** 0 .0 5 * * * 0 .0 6 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 4 * * 0 .0 6 *** 0 .0 4 * * 0 .0 6 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 4 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 4 * * * 0 .0 0 0 .0 0 0 .0 0 0 .0 0 0 .0 5 * * + 0 .0 5 * * * 0 .0 5 * * + 0 .0 5 * * * Table 8 reports one-sided tw o-sam ple K-S statistics and their asym ptotic significance levels.36 I conduct both tests using the null hypothesis F = G against the alternatives F > G and F < G. There are 127, 16,383, and 268,435,455 observations at annual, sem iannual, and quarterly intervals, respectively, in the theoretical distribution sam ple. For the buy recom m endations, theoretical distributions are not significantly different from em pirical distributions o f perform ance persistence at the annual intervals for any perform ance measures. At the sem iannual intervals, risk-adjusted analyst perform ance is significantly less persistent than w hat the theoretical distribution im plies, but raw return perform ance is more p ersistent than what would be im plied at a high significance level. Raw return perform ance is also highly persistent at the quarterly interval, although at a relatively low er significance level than at the sem iannual intervals. For sell recom m endations, analyst perform ance persistence is sim ilar to what the theoretical distribution implies at the annual interval, but is w eaker at the sem iannual and quarterly intervals for all perform ance m easures. O verall, perform ance persistence tests suggest that analysts’ risk-adjusted perform ance is not persistent in the m ulti-period framework, while raw return perform ance o f buy recom m endations is m ore persistent at the annual and sem iannual intervals. This raw return persistence provides an upper bound for analyst perform ance persistence. It also suggests that investors can use it to develop profitable trading strategies if they care only about raw returns, like som e individual investors and hedge funds. 36 M o n te C a rlo e s tim a te s o f sig n ific a n c e le v e ls fo r th e e x a c t te s t are a lso o b ta in e d fo r a n n u a l a n d th e s e m ia n n u a l in te rv a ls . T h e y su p p o rt m u ch th e s a m e c o n c lu s io n s , and th u s a re n o t r e p o rte d . I d o n o t o b ta in 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C onclusions The very com prehensive 1BES recom m endation database allow s examination o f perform ance persistence o f analyst recom m endations at different frequencies in both a tw o-period and m ulti-period framework and w ithout a survivorship bias. The results suggest one-period-ahead performance persistence of analyst buy recom m endations at the quarterly, sem iannual, and annual frequencies over the w hole sam ple period; the results are invariant to portfolio testing m ethodologies, portfolio w eighting schemes, return m easurem ent intervals, and systematic and specific risk adjustm ents. The performance o f sell recom m endations, however, is alm ost never persistent e x c ep t for raw returns at the annual and sem iannual intervals. The perform ance o f buys is also persistent in the m ajority o f .s ubperiods according to persistence tests for pairs o f consecutive subperiods. R isk-adjusted perfonnance is m ore persistent at the annual and sem iannual intervals, and raw return performance is m ore persistent at the semiannual and quarterly intervals, suggesting that analyst perform ance persistence is not a transitory phenom enon. Persistence is also more pronounced for raw returns than for risk-adjusted returns, indicating the importance of appropriate risk adjustm ents in persistence tests. I also find that the above persistence is largely attributed to w inners rather than losers, which is especially obvious for risk-adjusted perform ance. T he performance of both winners and losers is generally persistent, suggesting that we can use losing persistence as a useful indicator to avoid bad analysts, and m ore importantly, that we can use winning persistence as a profitable investm ent strategy. F o llo w in g the buys o f past w inners in term s o f risk-adjusted returns generates annualized risk-adjusted returns as M o n te C a rlo e s tim a te s f o r th e q u a rte rly in te rv al b e c a u s e o f e n o rm o u s r e q u ir e m e n t o n co m p u tin g fa cility . 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. high as 15.7%. P ortfolios o f past w inners in term s of raw returns yield annualized buyand-hold raw retu rn s as high as 45.2% , although raw return perfonnance in the ranking period does not p red ict risk-adjusted perform ance in evaluation periods. In addition, I im prove on the m ore pow erful m ulti-period perfonnance persistence test innovated by A garw al and N aik (2000) and show that analyst perform ance persistence largely disappears in the m ulti-period fram ew ork except for raw returns of buy recom m endations at the sem iannual and quarterly intervals. This finding reinforces the robustness o f raw return persistence and its im portance in potential trading strategies. Yet raw return persisten ce in m ulti-period fram ew ork does not contradict m arket efficiency. On the contrary, the evaporation o f perform ance persistence after m aking the appropriate risk adjustm ents provides strong support for m arket efficiency. T h is w ould confirm the suggestion by Brown and G oetzm ann (1995) and C arhart (1997) that performance persistence may appear because m arket participants use a com m on strategy not captured by standard style categories or risk adjustm ent procedures. O verall, the results both support m arket efficiency and suggest that follow ing past leaders is a profitable trading strategy at least in the short term. The annual perform ance persistence results provide fundam ental support for identification o f leaders among financial analysts and for the all-star rankings by the / / and WSJ. Since the prev io u s literature on buy-side fund m anagers focuses on one-period-ahcad perform ance persistence, we cannot so far com pare the m ulti-period perform ance persistence o f fund m anagers and sell-side analysts. Although fund m anagers are show n to exhibit persistent perform ance, the econom ic significance o f this persistence is not as great as w hat I find for financial analysts. One reason for the less persistent perform ance results 96 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for fund m anagers is in the significant transaction costs such as trading com m issions and marketing fees. A nother explanation is that open-ended funds must have to allocate funds to cash or liquid securities that earn lim ited returns. C onsider also that fund m anagers m ay not have fully exploited the valuable inform ation in analyst recom m endations. An interesting question for future research is exam ination o f the relation between analyst recom m endations and mutual fund holdings. 97 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. References Agarwal, Vikas, and Narayan Y. Naik, 2000, M ulti-period performance persistence analysis o f hedge funds. J o u rn a l o f Financial a n d Q uantitative Analysis 35, 327-342. Barber, Brad M ., R euven Lehavy, M aureen M cN ichols, and Brett Trueman, 2001, Can investors profit from the prophets? Consensus analyst recom m endations and stock returns. Journal o f Finance 56, 531-563. Barber. Brad M ., Reuven Lehavy, and Brett T ruem an, 2001, Are all brokerage houses created equal? Testing for system atic differences in the perform ance o f brokerage house stock recom m endations, University of C alifornia at Berkeley W orking Paper. Barber, Brad M ., and John D. Lyon, 1997, Detecting long-run abnormal stock returns: The em pirical pow er and specification o f test statistics, Journal o f Financial Economics 43, 341-372. Bollen, N icolas P.B ., and Jeffrey A. Busse, 2002, Short-term persistence in m utual fund perform ance, V anderbilt U niversity W orking Paper. Brav, Alon, C hristopher Geczy, an d Paul A. Gom pers, 2000, Is the abnormal return following equity issuance anom alous? Journal o f F in ancial Economics 56, 209-249. Brown, Stephen J., and W illiam N. Goetzmann, 1995, Perform ance persistence. Journal o f Finance 50, 679-698. Brown, Stephen J., W illiam N. G oetzm ann, and R oger G. Ibbotson, 1999, O ffshore hedge funds: Survival and perform ance 1989-1995, Journal o f B usiness 72, 91-1 17. Brown, Stephen J., W illiam N. G oetzm ann, Roger G. Ibbotson, and Stephen A. Ross, 1992, S urvivorship bias in perform ance studies. R eview o f Financial Studies 5, 553580. Busse, Jeffrey A., 1999, Volatility tim ing in mutual funds: Evidence from daily returns. Review o f F inancial Studies 12, 1009-1041. Carhart, M ark M ., 1997, On persistence in mutual fund perform ance, Journal o f Finance 52, 57-82. Carhart, M ark M ., Jennifer N. C arpenter, Anthony L ynch, and David K. M usto, 2000, Mutual fund survivorship, N ew York University W orking Paper. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C arpenter, Jennifer N., and A nthony Lynch. 1999, Survivorship b ias and attrition effects in m easures o f performance persistence. Journal o f F in a n cia l E conom ics 54, 337374. C hristopherson, Jon A., W ayne E. Ferson, and Debra A. G lassm an, 1998, Conditioning m anager alphas on econom ic inform ation: A nother look at the persistence of perform ance, Review o f F inancial Studies 11, 111-142. C lem ent, M ichael B., 1999, A nalyst forecast accuracy: Do ability, resources, and portfolio complexity m atter? J o u rn a l o f Accounting a n d E conom ics 27, 285-303. C ooper, R ick A., Theodore E. D ay, and C raig M. Lewis, 2001, F ollow ing the leader: A study o f individual analysts’ earnings forecasts, Journal o f F in a n cia l Econom ics 61, 383-416. D aniel. K ent, M ark Grinblatt, S heridan Titm an, and Russ W erm ers, 1997, M easuring m utual fund performance w ith characteristic-based benchm arks, Journal o f Finance 52, 1035-1058. D im son, E., 1979, Risk m easurem ent w hen shares are subject to infrequent trading. Journal o f Financial E conom ics 7, 197-226. Eckbo, B. Espen, Ronald W. M asulis, and Oyvind N orli, 2000, Seasoned public offerings: Resolution o f the “n ew issues puzzle,” Journal o f F in ancial Econom ics 56, 251-291. Elton, E dw in J., Martin J. G ruber, and Seth Grossman, 1986, D iscrete expectational data and portfolio performance, J o u rn a l o f Finance 41, 699-713. Fam a, E ugene F., and Kenneth R. French, 1993, Com m on risk factors in the returns on stocks and bonds. Journal o f F in a n cia l Economics 33, 3-56. G ail, M itchell H., and Sylvan B. G reen, 1976, Critical values for the one-sided twosam ple K olom ogorov-Sm irnov Statistic, Journal o f A m erican Statistical Association 71, 757-760. G rinblatt, M ark, and Sheridan T itm an, 1992, The perform ance, Journal o f F inance 47, 1977-1984. persistence of m utual fund G ruber, M artin J., 1996, A nother puzzle: The growth in actively m anaged mutual funds. Jo u rn a l o f Finance 60, 783-810. H endricks, D., J. Patel, and R. Z eckhauser, 1993, Hot hands in m utual funds: Short-run persistence o f relative perform ance, 1974-1988, Journal o f F inance 48, 93-130. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Jacob, John, Thom as Z. Lys, and M argaret A. N eale, 1999, E xpertise in forecasting perform ance of security analysts. Jou rn a l o f A ccounting E conom ics 28, 51-82. Jaffe, Jeffrey F., and Jam es M. M ahoney, 1999. T he perform ance o f investm ent new sletters, Journal o f F inancial E conom ics 53, 289-307. Li, X i, 2001a, Perfonnance evaluation o f recom m ended portfolios o f individual financial analysts, Vanderbilt U niversity W orking Paper. Li, Xi, 2002, Career concerns o f financial analysts: C om pensation, term ination, and perform ance, Vanderbilt U niversity W orking Paper M alkiel, Burton G., 1995, Returns from investing in equity m utual funds 1971 to 1991, Jou rn a l o f Finance 50, 549-572. Scholes, M yron, and Joseph T. W illiam s, 1977, Estim ating betas from nonsynchronous data. Journal o f Financial E conom ics 5, 309-328. Sm irnov, N.V., 1939, On the estim ation o f the discrepancy betw een empirical distribution curves for tw o independent sam ples. Bulletin o f U niversity o f M oscow, International Series (M athem atics) 2, 3-14. Stickel, Scott E., 1995, The anatom y o f the perform ance of buy and sell recom m endations. F inancial A n a lysts Jou rn a l, Septem ber-O ctober 25-39, 1995. W hite, H albert, 1980, A heteroscedasticity-consistent covariance m atrix estim ator and a direct test for heteroscedasticity, E conom etrica 48, 817-838. W om ack, Kent L., 1996, Do brokerage an aly sts’ recom m endations have investm ent value? Journal o f Finance 51, 137-167. 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C H A PT E R HI C A R EE R C O N C ER N S O F AN ALY STS: C O M PEN SA TIO N , TER M IN A TIO N , AN D PE R FO R M A N C E Introduction D uring the 1990s, sell-side financial analysts have played a prom inent role in the investm ent process through their traditional influence on institutional inv esto rs and rising influence on individual investors. However, sin ce the m eltdow n of Internet and technology stocks in 20 0 0 and 2001, the biased incentives o f analysts have received more scrutiny from law m akers, regulators, and investors. Because their com pensation depends not only on perform ance, but also on the am ount o f investm ent banking b u sin e ss and trading com m issions th at they attract, they are su b ject to serious conflicts o f interest w hen issuing investm ent recom m endations and earnings forecasts.37 38 Congress has held several hearings regarding analysts’ conflicts o f interest. The Securities and Exchange C om m ission (SEC) and the National Association o f Securities Dealers (NASD) have conducted investigations into these conflicts. The H o u se Financial Service Com m ittee (H FSC ), SEC , and NASD have announced proposals to address analysts’ conflicts of interest [Schroeder (2002)]. C elebrity analysts such as Henry 37 A n a ly s ts ’ e a rn in g s fo re c a s ts a re k n o w n in g en eral to b e o v e r o p tim is tic a n d above th e a c t u a l e arn in g s [A b a rb a n c ll (1 9 9 1 ) and S tic k e l (1 9 9 0 ) ]. A n a ly sts w h o s e b r o k e ra g e h o u se h a s u n d e rw ritin g re la tio n s h ip w ith a firm a lw a y s issue m o r e p o s itiv e p re d ic tio n s th a n t h o s e fro m n o n -a ffilia te d h o u s e s . E a r n in g s a c c u ra c y b e c o m e s le s s im p o rta n t w h e n a n a ly s ts c o v e r sto c k s u n d e r w r itte n by th e ir o w n b ro k e ra g e f i r m s [H ong an d K u b ik (2 0 0 2 )] . T h e b u y s o f th e f o rm e r a n a ly s ts arc less in f o r m a tiv e w h ile th e ir sells a re m o r e in fo rm ativ e [D e c h o w , H u to n , an d S lo a n ( 1 9 9 7 ) , D u g a r a n d N a th an ( 1 9 9 5 ) , L in a n d M c N ic h o ls ( 1 9 9 8 ) , a n d M ichacly a n d W o m a c k (1 9 9 9 )]. I n d iv id u a l in v e s to rs a re p a r tic u la r ly v u ln e ra b le to a n a ly s t b ia s a n d a rc o n ly c o n c e rn e d a b o u t a n aly st p e r fo rm a n c e s u c h as e a rn in g fo re c a s t a c c u ra c y and e x c e s s r e tu r n s g e n e ra te d b y th e ir re c o m m e n d a tio n s 38 101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Blodget, Jack Grubm an, a n d M ary M eeker face in v esto r suits for overly optim istic recom m endations about c o m p a n ies in w hich their em ployers have investm ent banking business. Tw o key issues stan d o u t in the current d ebate on reducing an aly sts’ bias. The first is w hether analysts s h o u ld be allowed to ow n the stocks they cover and if so, what kind o f disclosure should b e required. T he second issue, w hich is my focus, is how to strengthen the relation b e tw e en analyst perform ance and career concerns such as com pensation and career term ination. This issue has been a focus o f law m akers, regulators, and industry associations. For exam ple, the guidelines o f the Association for Investm ent M anagem ent a n d Research, a professional organization for buy-side and sellside analysts, rely on peer p ressu re to end analyst bias [Etzel (2001)]. T he guidelines o f the Securities Industry A sso ciatio n also suggest changes in analyst com pensation. The proposals o f HFSC, SEC, a n d NA SD bar analysts from earning bonuses for luring investm ent banking business o n individual deal basis, yet their bonuses can still depend on the overall perform ance o f investm ent banking business o f their firms. Practitioners such a s research directors in securities firms argue that performance is already im portant in a n a ly s ts ’ career concerns. T h ey usually point out that analyst status on Institutional In v e sto r (II) A ll-A m erican team s is used as one o f the three most im portant determ inants o f an a ly st com pensation, along w ith the investm ent banking fees and trading com m issions th e y g en erate.39 T heir argum ent h ig h lig h ts an im portant but little explored question so far in this policy debate. That is, how e ffectiv e are the current c areer concerns such as perform ance- b c c a u s c th e y d o n o t c a rc a b o u t i n v e s tm e n t b a n k in g a s c o r p o ra tio n s d o a n d a re n o t in th e s a m e p o sitio n as in s titu tio n a l in v e s to r s to tap a n a l y s t s ’ k n o w le d g e th ro u g h p r iv a te ta lk s . 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. based compensation evaluation system s in inducing better perform ance. For exam ple, if current evaluation system s do not offer enough incentives for good perform ance in earnings forecasts and investm ent recom m endations, they need to b e reformed. If they already strongly m otivate analysts, we only need to strengthen the dependence o f analysts’ total com pensation o n the current system s, and the task o f proposed reform w ill be less formidable. Thus, evidence on the effectiveness o f career co n cern s provides both a better understanding o f the current relation betw een analyst’s p e rfo n n an ce and their career concerns and im portant im plications fo r any proposed reform and ultimately the appropriate level o f regulation in this area. Direct data on perform ance-based or total com pensation o f analysts are not publicly available. Previous research uses jo b turnover and career term ination as com pensation proxies and then exam ines th eir effectiveness to increase earnings forecast accuracy.40 I instead study how effective II a ll-star ranking and W all Street Journal (WSJ) all-star ranking are to m otivate better investm ent recom m endation perform ance. I also investigate the effectiveness o f career term ination possibility. T he perform ance m easure that I exam ine is the returns generated by investm ent recom m endations, instead of earnings forecast accuracy. R ecom m endations and earnings forecasts contain inform ation independent from each other and making recom m endations is the m o re important part o f analysts’ daily job com pared to offering earnings estim ates [D orfm an (1991). Francis and 39 T o re d u c e u n n e ce ssa ry te rm in o lo g y , I w ill re fe r II A ll- A m e r ic a n ran k in g a s II a ll- s ta r ra n k in g th e re a fte r. 40 C a r e e r te rm in a tio n is n o t th e m o st a c c u r a te d e s c rip tio n . A lth o u g h b u y -sid e m o n e y m a n a g e rs m a k e less m o n e y o n a v e ra g e , som e m a n a g e rs , e s p e c ia lly th o se o f h e d g e fu n d s, can b e c o m e v e r y w e alth y a n d still e n jo y a g o o d life style. S o m e a n a ly s ts m a y e x is t the p r o f e s s io n to g o to the b u y - s id e b e c a u se th ey a rc th e b e st p e rfo rm e rs and are v e ry c o n f id e n t a b o u t th e ir a b ility a n d b e c a u s e they e n jo y a b e lte r life sty le. T e r m in a tio n is used h e re to k e e p th e c o n s is te n c y w ith p r e v io u s lite ra tu re . I la te r fo r m a lly test th is h y p o th e s is b y stu d y in g th e te r m in a tio n - p e r fo rm a n c e r e la tio n fo r th e best a n d w o r s t p e rfo rm e rs s e p a ra te ly in s e c tio n 4 .3 . In a ddition, a n a ly s ts m a y b e p ro m o te d to m a n a g e m e n t p o sitio n s w i t h in a b ro k e ra g e firm s o r th e y c a n d is a p p e a r b e ca u se o f d e a th a n d illn e ss o r b e c a u s e th e y feel th at th e y a re o v e r w o rk e d [L i (2 0 0 1 a)J. 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Soffer (1997) and W om ack (1996)]. M ore im portantly, bias in recom m endations is the focal point o f the current debate on conflicts o f interest. Although avoiding career term ination is an obvious and extrem e incentive, the incentives provided by all-star rankings m ay be less clear. Previous literature and research directors o f brokerage firms suggest that m ost brokerage firm s use II ranking as the single proxy for analysts’ perform ance when they determ ine the perform ance-based com ponent of analyst com pensation and II A ll-star status increases analyst com pensation enorm ously, som etim es by m illions o f dollars [D orfm an (1988), D orfm an (1997). H ong and Kubik (2002), K essler (2001), L aderm an (1998), M ichaely and W om ack (1999), and Stickel (1992)]. So II ranking is a m uch m ore direct com pensation proxy than job turnover or c areer term ination. The ranking is the m ost im portant because it is based on the opinions o f a large num ber o f m oney m anagers each year and because it has a m uch longer history than any other rankings. However, the ranking is regarded as a “beauty contest” or “popularity contest” by m any on the Street, including many buy-side institutions for w hich the list is com piled. Sell-side analysts are know n for their creativ e and persistent efforts to intensively lobby m oney m anagers before the survey [D orfm an (1988), Ip (1998), and K essler (2001)]. This is why W57 provides an annual ranking solely based on the raw returns on the portfolios recom m ended by analysts w ithin an industry. However, because W57 have various restrictions for analyst to be considered in the ranking, the ranking is based on less than one third o f the overall analyst population, which raises the question w hether the ranking is (purely) based on perform ance over the whole analyst population, even if it is solely based on perform ance o f the gro u p o f analysts meeting its restrictions. 104 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Com pared to II ranking, W SJ ranking is m uch less used by brokerage firm s to determ ine analyst compensation. I investigate its effectiveness to reward better perform ance over the whole analyst population and its results also provide a direct contrast to the II ranking. Some may argue that perform ance-based com pensation structures are different in brokerage firms with different-status. For exam ple, Mikhail, W alther, and W illis (1999) and Hong and Kubik (2002) both use brokerage status as a m easure o f analyst com pensation. W ith the fierce com petition in the labor market for ranked analysts, how ever, perfom iance-based com pensation and even total com pensation are likely to be very sim ilar across brokerage firm s for th em .41 Another benefit o f studying the determ inants of different c areer concerns is that it also helps practitioners to know the real m eaning of the rankings. For exam ple, II keeps the survey and ranking process secret. As the ranking becomes such a pow erful weapon in the com pensation bargain, m any want to know its true m eanings. H ow ever, even the request for an explanation from research directors in major brokerage firm s was rebuffed [Dorfm an (1988)]. This study helps people like research directors, institutional and individual investors, and regulators to understand the determ inants o f different career concerns. It also helps financial analysts to understand the m ost effective ways to obtain good career outcom es and to avoid bad ones. Empirically, although I find that th e probability of becom ing II all-stars is significantly improved by perform ance m easured as risk-adjusted returns on their F o r th e reaso n s w hy b ro k e ra g e firm s c o m p e te f o r ra n k e d a n a ly s ts and a d e s c rip tio n o f th e in te n sity o f th is c o m p e titio n , p lease se e D o rfm a n ( 1 9 8 8 ) , L a d c rm a n ( 1 9 9 8 ) , a n d K essler (2 0 0 1 ). K rig m a n , S h a w , an d W o m a c k (2 0 0 1 ) sh o w th at o b ta in in g II a ll-sta r c o v e r a g e is o n e o f the m o st im p o rta n t m o tiv e s for is s u e rs to 41 s w itc h u n d e rw rite rs. A m o n g th e m o s t fre q u e n t h e a d lin e s a b o u t a n aly sts in in d u s try n e w s le tte r s su c h a s 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. recom m ended portfolios and efforts m easured as the num ber o f recom m endations m ade and the n u m b er o f stocks covered, this probability im proves far greater with recognition measured as the size o f their brokerage firms and the com panies they cover, and w hether their brokerage firm s are top ten IPO underw riters or have a top 300 money m anagem ent operation in the U.S. T his probability also im proves far greater with their reputation m easured as all-star status in the prior year. Studying the determ inants of this probability for analysts w ho are and are not II all-stars in the prior year, I find that perform ance is only relevant for non-all-stars to become all-stars, but not for the reelection o f past all­ stars. The m o st effective w ay for past II all-stars to be reelected or for the non-all-stars to become all-stars is to sw itch to a large brokerage firms. Focusing on past II all-stars, the probability o f m oving up, and especially m oving dow n, different team of II ranking depends m u ch m ore on perform ance and efforts than the probability o f reelection, although brokerage firm size and IPO underw riting ranking still play a role o f sim ilar im portance to perform ance and efforts. II ranking punishes analyst aggressiveness measured as the deviation o f riskiness from the average analyst portfolios, but not the riskiness o f their recom m ended portfolios. In co ntrast, I find that perfonnance is the m ost im portant for WSJ all-star ranking. Although recognition and reputation effects exist, they appear because of the positive correlation betw een these factors and the various restrictions o f WSJ ranking on analyst eligibility. C onditional on past WSJ all-star status, perform ance is the only factor that significantly im proves the reelection probability for past W SJ all-stars. W hile recognition and reputation are still significant for non-all-stars to becom e all-stars, improving W all S tre e t L e t t e r a re th a t b r o k e ra g e firm s a ttra c t II A ll-s ta rs w ith h e fty p a y s a n d th e se a n a ly sts j u m p s h ip s v ery re g u la r ly . I a ls o fin d th a t II A ll-s ta rs fre q u e n tly s w itc h b ro k e ra g e f ir m s fro m o n e y e ar to th e n e x t. 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. perform ance and efforts is o f sim ilar im portance. In addition, the probability o f m oving up or down the different levels o f a ll-star ranking depends solely on perform ance fo r past W SJ all-stars. A ggressiveness significantly reduces a n a ly sts’ chances o f becom ing W SJ all-stars. For career term ination, reputation and efforts seem to be the m ost im portant, although the effects o f perform ance and recognition are statistically significant too. Conditional on past II all-star status, perform ance has no effect on the term ination probability of past II all-stars, but has significant effect on that o f non-all-stars. In contrast, although perform ance is im portant for the term ination probability of both a ll­ stars and non-all-stars o f WSJ, its im pact is far more im portant for past WSJ all-stars. Term ination probability also increases w ith the risk level o f p ortfolio recom m endations. For all the career concerns, I also find the prior o n e-y ear perform ance is m uch m ore important than earlier perform ance. In addition, raw return perform ance has som ew hat less im pact than risk-adjusted perform ance, suggesting greater im portance o f relative performance and that investors care about risk-adjusted returns even for the portfolios of individual financial analysts. M oreover, the ev idence suggests asym m etric effect o f performance as H ong and K ubik (2002). T hat is, analysts are more likely to be punished for bad perform ance than being rew arded for good perform ance. For the policy debate on reducing analyst bias through changes in their com pensation structure, the evidence suggests that II ranking, currently the most frequently used in com pensation determ ination, does not o ffer enough incentive for analysts to improve perform ance o r efforts. Instead, W SJ ranking have done a m uch better jo b in achieving this objective. T o further im prove its ability to reward better 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. perform ance, W SJ ranking needs to carefully modify their restrictions to reduce the effect o f recognition and reputation and strives to m ake more analysts eligible for the ranking. Because investors are more interested in risk-adjusted perform ance, it would also be helpful if WSJ ranks analysts on the basis o f risk-adjusted perform ance in the future, instead o f raw return perform ance. T he rest o f the article is organized as follows: Section 2 com pares my paper to the previous literature. Section 3 describes the sam ple and the econom etrics o f the factor models. It also discusses the various career concerns exam ined. In addition, this section gives details about the m ethodology used to form analyst portfolios and the m atching portfolios. Section 4 presents the em pirical results. Section 5 offers concluding rem arks. R elated Literature G iven considerable evidence o f an alysts’ distorted incentive, improved com pensation evaluation system s that rew ard performance can trem endously benefit investors by inducing better perform ance. Surprisingly, little evidence exists about the relation betw een analyst perform ance and their career concerns even though the relation between perform ance and com pensation and other career concerns has been extensively studied for m any other agents such as C E O s and fund m anagers. M ikhail, W alther, and W illis (1999) are the first to study the relation between analysts’ career concerns and perform ance. They find significant im pact o f relative but not absolute perform ance o f earnings forecasts on analysts’ jo b turnover.42 43 E xtending A b so lu te p e r fo r m a n c e is th e m a g n itu d e o f d if f e r e n c e s b e tw ee n th e a c tu a l e a r n in g s a n d the e a rn in g s fo re c a sts o f a n a ly s ts . R e la tiv e p e r fo r m a n c e is th e a b s o lu te p e rfo rm a n c e o f in d iv id u a l a n a ly sts c o n tr o lle d b y th e a b so lu te p e rfo rm a n c e o f a v e ra g e a n a l y s t s c o v e r in g th e sam e sto c k s. 42 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. their work, Hong and Kubik (2001) show that analysts with relatively m ore accurate forecasts are able to transfer to m ore reputable brokerage houses and are rew arded by the coverage o f more important stocks w ithin their brokerage firms. They also find that optim istic forecasts benefit jo b prospects. In addition, they illustrate that accuracy is m uch less im portant when analysts cover stocks underwritten by their ow n brokerage houses and during the stock m arket boom o f the late 1990s, while optim istic forecasts benefit jo b prospects even more in the above tw o situations. Separately, Hong, K ubik and Solom on (2000) dem onstrate that young analysts are more likely to leave analyst profession for poor or bold earning forecasts and they are less bold than o ld er analysts in a test o f herding hypothesis. M y study differs from the prior w ork in several im portant ways. First, the II ranking is a more direct com pensation proxy than job turnover, because it is well know n to be one o f the three determ inants o f analyst com pensation, in addition to the investm ent banking fee and trading com m issions they generate (Dorfman (1988), D orfm an (1997), H ong and Kubik (2002), Kessler (2001), Laderm an (1998), M ichaely and W om ack (1999), and Stickel (1992)]. C om pensation is also constantly relevant to all the analysts w hile jo b turnover is more extrem e. In addition to being more relevant, a m ore direct proxy can im prove the power to d etect a relation between com pensation and perform ance (M ikhail et al. (1999)]. 43 A lth o u g h M ik h a il c t al. (1 9 9 9 ) e x a m in e b rie fly re c o m m e n d a tio n p e rfo rm a n c e a n d fin d it u n r e la te d to j o b tu rn o v e r , m y stu d y d iffe rs in se v e ra l a s p e c t s . F irs t, m y d a ta in clu d e s a c o m p re h e n s iv e s e t o f b ro k e ra g e firm s w h ile th e ir d a ta b a s e d o c s not i n c lu d e s o m e o f th e m o st p ro m in e n t b ro k e ra g e firm s s u c h a s D U , G o ld m a n S a c h s, a n d M e rrill L ynch th a t m a k e m o re th a n 10% o f all re c o m m e n d a tio n s . M y s a m p le p e rio d c o v e rs th e m o st re c e n t d e c a d e w hen th e b ia s e d in c e n tiv e s h a v e b e co m e a w id e sp re a d p r o b le m . In a d d itio n , a s th e y s u g g e s te d , th e fin d in g o f n o r e le v a n c e o f re c o m m e n d a tio n p e rfo rm a n c e m ay be a ttr ib u ta b le to th e ir u s a g e o f le s s d ire c t c a r e e r co n ce rn m e a s u r e s . M o r e o v e r , I m ea su re a n a ly s t r e c o m m e n d a tio n p e rfo rm a n c e m o re a c c u r a te ly a n d p r o v id e m ore s u f f i c ie n t ris k a d ju s tm e n t to a n aly st p e rfo rm a n c e . 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Second, I exam ine the perform ance measured as abnorm al returns generated by investm ent recom m endations instead o f accuracy of earn in g s forecasts. Existing literature has alm ost n e v e r studied recom m endation performance. R ecom m endations and earnings forecasts contain independent inform ation and m aking recom m endations is m ore im portant as a n a ly sts’ daily job [D orfm an (1991), Francis an d Soffer (1997) and W om ack (1996)]. M ore im portantly, bias in recom m endations is the focal point o f the current debate on conflicts of interest. Third, although I also study jo b turnover as previous research, I focus clearly on career term ination. M ix in g jo b turnover o f different directions can yield am biguous results [Hong and Kubik (2001)]. A nother related paper by Stickel (1992) shows that analysts who are subsequently ranked as / / all-stars produce more accurate earnings forecasts than non-all-stars on the stocks covered by II all-stars. Except for the obvious d ifference that I examine recom m endation perform ance, my stu d y investigates the im pact o f other analyst characteristics such as reputation, recognition, and efforts o n all-star status, in addition to perform ance, in the context of m ultiple regressions, w hich reduces the possibility that perform ance is a proxy for another factor. For example, as I show later, the ranking process or th eir eligibility restrictions are positively correlated with the market c a p o f the stocks covered and the size of their brokerage firm. The fact that past all-stars usually cover large stocks could significantly increase their chance o f rem aining all-stars. In addition to univariate analysis, focusing on the stocks covered by all-stars produces serious selection bias. Since the freedom to choose coverage should increase w ith seniority, all-stars have more freedom to choose stocks that are easier for them to forecast. 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Sam ple, R ankings, and Perfonnance M easurem ent The A nalyst Sam ple and C a re e r Term ination T he prim ary database used in this paper comes fro m IBES. Its m ajor benefit is the inclusion o f recom m endations from a very broad sam ple o f brokerage firms and financial analysts. Even large databases such as Zacks do not include important bulge bracket firm s such as M errill Lynch, G o ld m an Sachs, and D onaldson, Lufkin & Jenrette. The fBES database includes all m ajo r brokerage firms plus a larger sample o f sm aller brokerage firm s than Zacks. A nalysts can alm ost always b e tracked even if they sw itch brokerage firm s. V arious m ark et participants, including professional investors, use this database. [BES has collected b u y and sell recom m endations from the research reports o f financial analysts since the e n d o f O ctober 1993. The datab ase includes both brokerage firm -specific ratings and a stan d ard ized IBES rating. T he form er are usually on a threeto five- level scale. The IB E S -created ratings are on a un ifo rm five-level scale; num eric ratings from 1 through corresp o n d to “strong buy,” “buy,” “ hold,” “underperform ,” and “sell” . A nalyst portfolios are form ed using recom m endations with numeric ratings o f 1 and 5 because these are the rankings wilh the strongest signal and with no am biguity in analyst opinion. A nalysts are notorious for their am biguous term inology and clarification o f term inology is one goal o f th e recent proposals by H F S C , NASD, and SEC. For exam ple, although “ H old” m ay appear to be a neutral ratin g , it actually m eans “S ell” for m any analysts because they d o not w ant to displease c o m p an y management. It is also ill Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. difficult to assign arbitrary w eights to different recom m endation categories. In addition, as we will see, the co m b in ed recom m endations with ratings o f 1 and 5 com prise alm ost half o f all recom m endations excluding “ Hold”. I use the investm ent recom m endation data from O ctober 1993 to D ecem ber 2000. The return and accounting data are draw n from C R SP and C om pustat, respectively. Panels A and B o f T able 1 sum m arize the database. T here are 241,222 recom m endations by 7,308 financial analysts from 408 institutions in the five b u y and sell recom m endation categories. In the empirical analysis, I exclude analysts w ith fewer than 10 recom m endations. T hat sam ple consists of 4,383 analysts before other restrictions are applied. Panel B indicates that favorable recom m endations are m uch more prevalent, consistent w ith w hat we see in other databases such as First Call and Zacks. The ratio o f strong buys to sells for the entire sample period is about 15-to-l. Panel B also suggests that both the num ber o f negative recom m endations and these recom m endations as a percentage o f all recom m endations decline over time, despite the grow ing num ber o f total recom m endations made each year. T his suggests that th e buyto-sell ratio declines co n tinuously throughout the sam ple period. To put the changes o f buy-to-sell ratio in a historical perspective, according to Zacks Investment R esearch, the ratio o f “buy” and “strong buys” to “underperform ” and “sell” is 0.9 to 1 in 1983, 4 to 1 by the end of the 1980s, 8 to 1 in early 1990s, an d 48.2 to 1 in 1998 [Laderm an (1998)]. T his constant monotonic decline o f negative recom m endations is consistent w ith m ore distorted incentives for financial analysts in recent years instead o f a purely m ore bullish position taken by analysts in the later portion o f the sam ple period. 112 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1: Summary Statistics of the IBES Recommendation Database T a b ic 1 r e p o rts s u m m a ry sta tistics o f IB E S r e c o m m e n d a tio n d a ta b a s e . Panel A p r e s e n ts s u m m a ry sta tistic s r e g a r d in g th e siz e o f IB E S re c o m m e n d a tio n d a ta b a s e . P a n e l B re p o rts nu m b er o f r e c o m m e n d a tio n s in e a c h c a te g o r y a s p e r c e n ta g e s o f all re c o m m e n d a tio n s b y y e a r , e x c e p t th e last c o lu m n , w h ic h re p o rts the to ta l n u m b e r o f re c o m m e n d a tio n in e a c h y e a r. P a n e l C s h o w th e n u m b e r o f a n a ly sts w h o m e e t m y se le c tio n c r ite r ia a n d th e p ro p o rtio n o f a n a ly s ts w h o d is a p p e a r fr o m IB E S d a ta b ase e a c h y e a r. It a ls o re p o rts, s e p a ra te ly fo r a n a ly s ts w h o su rv iv e a n d w h o d is a p p e a r a n d fo r th e o v e ra ll a n a ly st s a m p le , th e risk -a d ju s te d p e r fo rm a n c e a n d re s id u a l risk lev el o f th e p o rtfo lio s r e c o m m e n d e d b y in d iv id u a l a n a ly s ts . R isk -a d ju s te d p e r fo r m a n c e a n d r is k level are b o th in p e rc e n ta g e . T h e d a ta a re fro m Jan u a ry 19 9 4 th ro u g h D e c e m b e r 2000 . ______________________ P a n e l A: S u m m a ry S ta tis tic s o f I B E S R e c o m m e n d a tio n D a ta b a s e ______________________ N u m b e r o f A n a ly s ts : 7 3 0 8 N u m b e r o f B ro k e rs: 4 0 8 N u m b e r o f A n a ly s ts w ith > 10 re c o m m e n d a tio n s : 4 3 8 3 N u m b e r o f A n a ly s ts w ith > 100 re c o m m e n d a tio n s : 5 6 3 ______________________ P a n e l B: B re a k d o w n o f th e R e c o m m e n d a tio n C a te g o rie s by Y e a r______________________ IB E S R a tin g s S tro n g Buy 4 3 2 1994 25% 33% 37%) 2 1995 27 32 36 2 S e ll % T o ta l 3%. 29521 3 30854 1996 30 33 32 2 1997 31 37 29 1 2 30350 1998 29 39 30 1 1 35 4 4 5 1 37318 29734 1999 30 40 28 2 2 0 0 0 31 40 27 1 I 32 6 6 3 A v e ra g e 29 36 32 t ? 225885 Panel C . S u m m a r y S ta tis tic s a b o u t C a r e e r T erm in atio n Y ear S iz e R ate 1994 1204 0.11 1995 1502 0 .10 1996 1638 1997 S u rv iv a ls W h o le S a m p le D is a p p e a ra n c e R isk A lp h a 0 .0 0 4 0 .1 1 2 0 .0 0 7 0 .1 1 2 0 .1 1 7 0 .1 1 5 0.118 0 .0 8 6 0 .1 1 4 0 .10 0 .0 9 9 0 .1 3 0 0 .1 0 0 0 .1 3 0 0 .0 9 0 0 .1 2 4 0 .1 3 7 A lp h a Risk A lp h a R isk 0.112 - 0 .0 2 6 0 .1 1 5 1894 0.11 0 .1 1 2 0 .1 3 5 0 .1 1 5 0.135 0 .0 9 3 1998 2156 0.12 0 .0 7 7 0 .1 7 2 0 .0 8 5 0 .1 7 0 0 .0 1 6 0 .1 8 8 1999 2323 0.15 0 .1 3 5 0 .1 7 7 0 .1 4 6 0 .1 7 6 0 .0 7 4 0 .1 8 5 A v e ra g e 1786 0.11 0 .0 9 0 0 .1 4 1 0 .0 9 5 0 .1 4 0 0 .0 5 5 0 .1 4 4 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Although 1 do not directly observe w h eth er analysts have left the profession, it is very unlikely that an analyst switched to a firm not covered by 1BES, if analysts stop subm itting recom m endations to EBES. T h is is because the vast majority o f brokerage houses subm it earnings forecasts and in vestm ent recom m endations to IBES. Hence, 1 follow Hong, et al. (2000) and M ikhail e t al. (1999) and define that analysts experience career term ination in year t if analysts m ade recom m endations in year l-1 but stopped recom m endation stocks som etim e in y ear t and forw ard. Although sell-side analysts usually strive to be / / all-stars, there is a sm all but realistic possibility that som e analysts m ay have left sell-side to become m utual fund or hedge fund m anagers or buy-side analysts. S ince com pensation on the buy-side depends solely on perform ance, only the best perform ing analysts would dare to m ove to buy-side. I later formally test this possibility by studying the term ination-perform ance relation for the best and w orst perform ers separately in section 4.3. If the probability o f career term ination is larger for the worst perform ers, career term ination provides incentives for analysts to outperform . Yet, evidence suggesting larger termination probabilities for the best perform ers w ould indicate that the best perform ers exit analyst profession for better careers. Panel C o f T able 1 show sum m ary statistics about career term ination. Analyst attrition rate is alw ays som ew hat higher than 10% each year. The table also com pares the risk-adjusted perform ance and specific risks on the portfolios of surviving analysts, disappearing analysts, and all the analysts. R isk-adjusted performance is m easured as the percentage daily excess returns on analyst portfolios or the alpha of the m arket model. Specific risk is the residual standard e rro r o f the m arket m odel regression. Panel C 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. suggests that disappearing analysts consistently underperform surviving analysts. The difference indicates a potentially significant relation between perform ance and term ination. Yet, analyst disappearance does not have a marked effect on the perform ance distribution o f the overall sam ple in m ost o f the sam ple period, because m ean performance o f the overall sam ple is not significantly different from perform ance o f surviving analysts in m ost years. D iversifiable risk does not seem to be noticeably different between surviving and disappearing analysts. A ll-sta r Rankings Each year, II sends questionnaires for d ifferen t industries to the research directors, c h ie f investm ent officers, analysts and portfolios m anagers at m ajor m oney m anagem ent institutions including the largest 300 funds in the U.S. ranked by II and o th e r key U.S., European, and Asian investors. F o r exam ple, in 2001, II sent ballots to m ore than 780 institutions and received the o pinions of more than 3,200 individuals from about 400 institutions [Dini (2001)]. T hese investors are asked to give analysts a num eric score. Industry classification changes ev ery year in consultation with W all Street research directors and buy- and sell-side analysts. II determinates ranking using the average o f the numerical scores analysts receive, by w eighting the votes for individual analysts on the basis of the size o f voting institutions. In the ranking published in its O ctober issue, //se le c ts the first, second, and the third teams, and runner-ups, w ho are the analysts receiving the best av erag e numeric scores am ong those in each industry. Although usually there is only one an aly st in the 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. first or second team , there could be several analysts in the third team and runner-up group within an industry. II keeps the survey and ranking process secret. It has been noted that analysts in small brokerage firm s are disadvantaged because their firms d o not have a large sales force directed to institutional investors to publicize their work [D orfm an (1997)]. In addition, because votes are w eighted by the size o f voting institutions and large brokerage firms usually have large asset m anagem ent business, analysts in sm all brokerage firms face another significant hurdle because the money m anagers o f large brokerage firms are likely to vote for their own analysts. M oreover, because sm all brokerage firm s usually focus on covering small cap and m id ­ cap stocks, analysts from these firms are hurt because size o f institutions investing in small cap and m id-cap stocks is much sm aller. F or exam ple, even if the average num eric sores of tw o analysts are the same, the analysts covering relatively sm aller stocks are less favorably considered in a value-weighted scale. The im portance o f the ranking in analyst com pensation, the criticism o f ranking as a “beauty sh o w ” , and the secrecy o f the ranking process all w arrant a careful analysis o f its determ inants. T hese factors are also reasons w hy the W SJ provides the second m ost influential ranking solely based on perform ance.44 W SJ ranking identifies the top five stock pickers in term s o f raw return perform ance o f recom m ended portfolios in each industry. W hen creating analyst portfolios, for a stock with a “ Strong Buy," rating, the return is m ultiplied by 2; for a stock rated a "Buy," the return is m ultiplied by 1.5; for a stock with a "H old," rating, the return is m ultiplied by 0; for a stock rated a "Sell," the return is m ultiplied by -0.5; for a stock rated a "Strong Sell," the return is m ultiplied by 1. So buys receive m ore w eight than sells. Industry classifications are based on D ow 116 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Jones Global Industry C lassification Structure, which changes significantly each year. Analysts have to cover at least five stocks and at least one o f the largest five stocks in an industry. Eligible com panies in an analyst’s portfolio have to have prices above a minimum and som etim es are also required to have active earnings estimates from the analysts. These various restrictions exclude m any analysts from participation and m any recom m endations are excluded if analysts m aking those recom m endations are not eligible for those industries. They also put analysts covering sm all stocks or analysts from sm aller brokerage firms in a particularly unfavorable position, sim ilar to II ranking. For exam ple, in 2001,433 all-stars were selected from m ore than 4,000 analysts at some 220 firms, w ith only 1,370 analysts being eligible [W iegold (2001)]. T he restrictions raises the issue about what the ranking is really based on w hen we exam ine it over the whole analyst population, even if it is solely based on perform ance over the group o f analysts w ho m eet its restrictions. I also use this ranking as a contrast to 11 ranking and as a raw return m easure of analyst portfolios. Table 2 reports the sum m ary statistics o f all-star rankings. Panel A shows analyst status changes from one year to another for II and WSJ rankings, respectively. II all-star status seems to be highly correlated from one year to the next, because 80 to 90% o f all­ stars are still all-stars in the next year except for the fourth team . About 40% o f the fourth team becomes non-all-stars in the next year. V ery little m obility exists even among different ranks o f all-star team s because the diagonal elem ents o f the status transition m atrix are alm ost alw ays the largest in each colum n and analyst status rarely m oves by 44 IV57 a lso o ffe rs ra n k in g b a s e d o n p e r fo r m a n c e m e a s u re d a s the a c c u r a c y o f e a rn in g s forecasts. 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2: Summary Statistics of All-star Ranking Institutional Investor (II) a n d Wall Street Journal (WSJ) fo r a n a ly s ts w ith d iffe re n t a ll- s ta r s ta tu s in the p r e v io u s y e a r. Panel A sh o w s , fo r T a b le 2 p re se n ts d e ta ile d in f o r m a tio n a b o u t m o v e m e n ts o f a ll-s ta r status o f a n a ly s ts w ith d iffere n t s ta tu s in th e c u r re n t y e ar, th e ir s ta tu s in th e next y e a r. R a n k 1-4 re p re se n ts F irst, S e c o n d , T h ird T eam , a n d R u n n e r - u p s o f Institutional Investor All-American r e s e a r c h a n a ly sts, o r th e a lls ta r a n a ly s ts ranked a s th e b e st th r o u g h th e fo u rth b e s t p e rfo r m in g analysts b y Wall Street Journal. R a n k 5 m e a n s th a t a n a ly sts a re n o t s e le c te d b y e ith e r ra n k in g . P a n e l B r e p o rt the n u m b e r o f a ll-sta rs o f II a n d WSJ a s p ro p o rtio n s o f o v erall a n a ly s t p o p u la tio n , p e r c e n ta g e o f a n a ly s ts w ho b e c o m e a ll- s ta rs in th e n e x t y e a r g iv e n th e ir all-star sta tu s in th e p re v io u s y e ar, r a n k in g c h a n g e s for an aly sts w ith a ll-sta r s ta tu s in th e p re v io u s y e a r, and p ro p o rtio n s o f a n a ly s ts w h o le a v e p ro f e s s io n g iv en their a l l - s t a r sta tu s in th e p re v io u s y e a r. P a n e l A . C h a n g e s in A ll-s ta r R a n k in g s G iv e n A n a ly s ts ’ A ll- s ta r S ta tu s in the P r e v io u s Y e a r c 2 « w z C3 Institutional Investor R a n k in g Wall Street Journal R a n k in s C u rre n t Y e a r R a n k C u r r e n t Y e a r R ank 1 2 3 4 1 63.4 19.1 6 .3 3.5 0.1 2 17.3 39.1 18.4 9 .0 0 .3 1 2 3 4 1 58.2 2 1 .7 5.4 2.2 0.2 2 19.4 3 6 .2 12.9 4 .4 0 .0 0 .3 5 3 8.1 20.6 2 8 .3 12.4 0 .7 4 2.8 9 .4 2 8 .3 35.1 2.1 5 8.4 11.8 18.6 40.1 9 6 .7 Sum 385 33 9 389 624 1 1226 n £_ «c 5 3 11.2 10.1 32.3 14.8 4 4.1 1 5 .9 22.6 3 4.8 1.9 5 7.1 1 5 .9 2 6 .9 4 3 .7 9 7 .5 Sum 189 182 208 215 12169 x C4 z P a n e l B. M o v e m en ts o f A ll-s ta r S ta tu s a n d T e r m in a tio n „ „ , O verall Sam ple ' II Analysts with More Than ' v Three Years of Experience WSJ WSJ P e rc e n ta g e o f A ll-star A n a ly s ts 1 3 .4 0 6 .1 3 17.89 8 .8 2 P e rc e n ta g e o f A ll-stars C o n d itio n e d o n A ll-s ta r S ta tu s in th e P rio r Y ear P a st S ta tu s A ll-sta rs 7 6 .6 8 16.12 7 9 .7 9 17.99 3 .2 4 5 .9 5 3 .6 6 7.67 Up 1 7 .9 0 6 .3 0 17.32 6 .6 9 D ow n 4 1 .6 2 8 9 .9 2 4 1 .1 3 8 9 .3 3 S ta y al R a n k 1 1 4 .1 0 0 .8 8 13.51 1.05 S ta y a t O th e r R a n k s 2 6 .3 7 2 .9 0 2 8 .0 4 2.93 N o n -a ll-s ta rs P e rc e n ta g e o f A ll-S tars W h o C h a n g e R a n k s P e rc e n ta g e o f A ll-stars W h o L e a v e P ro fe s s io n P a st S ta tu s A ll-s ta rs N o n -a ll-s ta rs 3.21 8 .1 0 2 .9 5 8.04 11.27 10.32 13.24 11.79 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. m ore than one level. For exam ple, 63.4% o f the First team stay in the first team and 80.7% o f the First team belong to the First a n d second team in the next year. In contrast to II ranking, WSJ all-star status seem to be m uch less autocorrelated. A bout 85% of all-stars becom e non-all-stars in the next year for all the all-star team s. Since WSJ ranking should be relative to raw return performance o f investm ent recom m endations, th e less frequent changes in II all-star status across years suggests that II ranking may not b e as objective. In addition, the moderate autocorrelation is consistent w ith the evidence on perform ance persistence in Li (2001b). M oreover, since Li (2001b) Finds that raw returns are the most persistent perform ance m easure o v er various return m easurem ent intervals, the much greater autocorrelation in II ranking should be m ostly due to something o th er than performance. Panel B reports several key proportions, or the unconditional probability, for both rankings and for b o th the overall sample an d the analysts with m ore than three-year experience. Focusing on the overall sam ple, the num ber of W SJ all-stars is about h a lf o f II all-stars, because about 13% of analysts are II all-stars com pared to 6% for WSJ ranking. Conditioned on all-star status in the prior year, about 77% o f II all-stars stay as all-stars com pared to about 16% for WSJ all-stars. About 3% o f non-all-stars becom e all­ stars in II, while tw ice as many non-all-stars becom e WSJ all-stars. T his difference is m ore striking than it appears, since the n u m b er o f WSJ all-stars as a proportion o f all analysts is only h a lf o f that o f II all-stars. For all-stars in the prior year, about 18% o f II all-stars m ove up by at least one team and 14% o f the First team stay in the First team , suggesting that about 32% o f a ll­ stars have positive career experience in the next year. About 42% o f II all-stars 11 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. experience negative career outcom e, because they move dow n by at least one team . A bout 26% o f II all-stars slay at the sa m e team in the next year, w ith no change in career outcom e. In contrast, it is much more d ifficult for past W SJ all-stars to avoid being dow ngraded. Total percentage of all-stars w ho m ove up by at least one team or stay in the first team is about 8%. A bout 3% o f all-stars stay in their prior team and about 90% o f all-stars move dow n at least one team. The percentage o f analysts leaving the profession is quite close betw een W SJ all­ stars (8.10%) and non-all-stars (10.32% ). In contrast, there is a huge difference in the percentage of analysts leaving the profession conditional on their II all-star status. 3.21% o f II all-stars leave the profession in th e subsequent year and 11.27% o f non-all-stars leave the profession in the next year. T h e se statistics support our discovery in Panel A. In contrast, WSJ all-star status is not as correlated across years and it provides m uch less security against being term inated. Turning to analysts with at least three years of experience, the percentage o f all­ stars are somewhat higher than the overall sam ple for both rankings, suggesting that experience is positively correlated w ith the probability o f becom ing all-stars as M ikhail, W alther, and W illis (1997). Since the experience requirem ent excludes analysts with relatively worse perform ance, it should be against me of finding a significant relation betw een the career concerns and perform ance. The other statistics arc quite sim ilar betw een the overall sam ple and analysts w ith at least three-year experience. 120 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Analyst Portfolios For each financial analyst w ho m ade at least ten recom m endations betw een October 1993 and D ecem ber 29, 2000 in the IBES recom m endation database and for whom a portfolio can be form ed for at least three months, I create both a value-w eighted and an equal-w eighted portfolio by purchasing stocks recom m ended as “Strong B uy” and selling stocks recom m ended as “S ell” . Stocks enter the analyst portfolios on the recom m endation date and are dropped at the revision date as recorded by IBES. R eturns from January 1994 th rough D ecem ber 2000 are used. I require the length o f analyst portfolios to be at least three months w ithin a year for estim ation purposes. Results with other length requirem ents are similar. The requirem ent o f at least ten recom m endations is to m inim ize the possibility that nonfinancial analysts e n te r the database by co-authorship or assum es the role o f a financial analyst tem porarily. R esults with a requirem ent o f m inim um tw enty recom m endations are similar. I also require analysts have at least three years o f experience so that I can examine the im pact o f m ulti-year perform ance on career concerns. For exam ple, analyst career concerns m ay n ot be related to prior one-year perform ance but are related to performance over several previous years. It is useful to exam ine different horizon to determine w here the im pact o f perform ance is the most significant. The final sam ple consists 5421 analyst-year observations. Because the results from using equal-w eighted and value-w eighted portfolios are quite sim ilar, I only present results for equally w eighted an aly st portfolios in the rest o f this paper. Since stocks w ith “Strong Buy” rating will dom inate the overall portfolio o f individual analysts, I a lso study the relation betw een career concerns and perform ance for 121 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. buy and sell recom m endations, respectively. The results for either buy or sell recom m endations are q u ite sim ilar to those for the overall portfolio and are unreported. Performance M easures I use average returns on analyst portfolios w ithin a year to m easure raw return performance. I also use the W SJ ranking to capture extrem ely good perform ers in raw returns. But because it restricts analyst eligibility in various w ays, it may not be as relevant as other raw return perform ance measures. R isk-adjusted perform ance is measured as the intercept o f a single-factor market m odel estim ated with daily returns. W hen the t-statistic o f th e intercept, or information ratio, is used to m easure perform ance, the results are sim ilar. T h e factor m odel is expressed as =a,+Po,Rml+£„ ( 1) R-, is the excess return on the portfolio o f analyst i on day t: or, m easures the abnorm al return of the portfolio o f analyst i ; Rmt is the return on the C R SP value-weighted N Y SE/A M EX /N A SD A Q m arket index on day t ; and eit is the idiosyncratic return o f the portfolio o f analyst i on day t . T he risk-free rate o f return is based on the daily U.S. 90day Treasury bill.45 As robustness tests. I also m easure performance using several other frequently used factor m odels: a th ree-facto r m odel that includes size and book-to-m arket factors [Fam a and French (1993)]; a four-factor model that adds m om entum effects [Carhart (1997)]; a five-factor m odel also including earnings/price factor; and a five-factor m odel R isk -free re tu rn s a re c a lc u la te d u s in g th e F e d e ra l R e se rv e ’s c o n s ta n t- m a tu r ity in te re s t ra te sc rie s . R e tu r n s a rc calcu la te d fro m th e p u b lis h e d y ie ld s u s in g a h y p o th e tic a l b o n d w ith th e s ta te d m a tu rity a n d a c o u p o n 45 122 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. that adds a liquidity factor o f share turn o v er [Li (2001a)]. R esults are m uch th e sam e and are not reported. The use o f daily data introduces one notable com plication. Dimson (1979) and Scholes and W illiam s (1977) observe a nonsynchronous trading problem in sto ck returns that hinders regression estim ation for individual securities. I address this problem by adding a lagged m arket factor:46 = a , + Pa, Rm, + Pi, K .,-1 + E„ (2 ) E m pirical Analysis The II A ll-sta r Ranking A probit model is used to study the relation betw een II all-star status and perform ance and other analyst characteristics. The m odel also includes yearly dum m ies for each year from 1996 through 1999: P r( A A H , = 1) = d> a () + P e r f o r m a n c e Ef fe ct s + E f f o r t E f f e c t s + R e p u t a t i o n E f f e c t s + R e c o g n i t i o n Ef f e c ts an = + « , P E R F O R M + ( a zL N N R E C ,_ l + a , L N N S T K , _ t ) (3) +a i A A H ,_ i + ( a ^ L N B R K S Z , ^ + a ,L N S T K C A P ,_ t fr a .I P O R A N K + a„FUND300) where P E R F O R M is perform ance m easures in Section 3.5. For both risk-adjusted returns and raw returns, I inspect the them o v e r prior 1-, 2- and 3-year periods because II ranking e q u a l lo th e y ie ld , th u s trad in g at p a r o r fa ce v a lu e . A n e n d -o f- p c r io d p ric e o n the hill u s in g th e n e x t d a y ’s y ie ld is first c a lc u la te d . T h e price is th e n u s e d to o b ta in the r is k -f re e re tu rn s . ^ D im s o n ( 1 9 7 9 ) s u g g e s ts in c lu d in g a s m a n y a s th re e lag s a n d th r e e lea d te rm s. T e s tin g w ith s e v e ra l c o m b in a tio n s s h o w s th at o n ly the first la g te r m is c o n s is te n tly s ig n ific a n t. T h e re su lts u s in g th r e e la g an d th re e le a d te rm s a rc q u a lita tiv e ly th e s a m e a s u s in g o n ly the first la g te rm . S in c e a n aly st p o r t f o li o s ty p ic a lly in c lu d e s e v e ra l s to c k s , th e n o n s y n c h ro n o u s t r a d i n g p ro b le m is n o t a s s e r io u s a s for in d iv id u a l s e c u ritie s . 123 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. may depend on m ulti-year perform ance. Because raw return and risk-adjusted perform ance are correlated, I exam ine them in separate regressions. AAII, equals 1 if analysts are all-stars in a specific year and 0 otherw ise. LNNREC and LN N ST K are the logarithm of the num ber of recom m endations analysts m ake and the num ber o f stocks they cover. They are used as proxies for analyst efforts. represents II all-star in the prior year and is used as a proxy for past reputation. LNBRKSZ and L N ST K C A P are the logarithm o f a n a ly sts’ brokerage firm size m easured as the num ber o f analysts it employs and the logarithm of the market capitalization o f the com panies analysts cover. They are both proxies for analyst recognition by investors. IP O R A N K is a dummy variable that equals 1 if analysts belong to a top 10 IPO underw riting brokerage firm and F U N D 300 is a dum m y variable th at equals 1 if analysts belong to a brokerage firm with a top 300 m oney m anagem ent operations in the U.S. I obtain the inform ation on IPO underw riting ranking and m oney m anagem ent ranking from Investm ent D ealers' Digest and Institutional Investor, respectively. I include these two variables as proxies for analyst recognition. The FUND3Q0 variable is particularly interesting. B ecause II sends survey to the top 300 U.S. funds and m oney m anagers in brokerage firm s with a top 300 fund should vote for the sell-side analysts in the same brokerage firm s, being in such a brokerage firm should significantly increases analysts’ probability o f becom ing all-stars. Ideally, I w ould have analyst experience. Because of the short span o f the recom m endation database, it cannot be obtained from the recom m endation database. Because IBES uses different codes to identify analysts in earnings forecast and F u rth e rm o re , a n a ly s ts u su a lly c o v er h ig h - liq u id ity s to c k s and th e s e s to c k s s h o u ld h a v e le s s p ro b le m o f n o n s y n c h ro n o u s tr a d in g . 124 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. investm ent recom m endation database, it is also hard to m atch analysts across these two databases to obtain analyst experience from the earnings forecast database. T he difficulty of m atching the tw o databases also prevents me from including earnings forecast accuracy variable in the above regression. In future research, it would be interesting to match these tw o databases and com pare the effect o f earnings forecast accuracy and investment recom m endation perform ance. T hese co n tro l factors are very im portant because they provide insights about the effect o f factors o th er than perform ance. They also control for bias in estim ation due to the proxy effect o f one factor for another. For exam ple, since the ranking process m akes analysts from large brokerage firm s m ore likely to becom e all-stars and Li (2001a) shows contem poraneous relation betw een brokerage firm size and performance, I m ay find a spurious relation betw een the rankings and perform ance or a spurious relation betw een the ranking and brokerage firm size if I exam ine the relation in a univariate fashion. C oefficient estim ates, their significance levels, and the effects o f different factors on m arginal probability are presented in Table 3. The effect of continuous independent variables on m arginal probability equals the change in the probability that an analyst becomes an II a ll-star for a one standard deviation change in the independent variable with other independent variables held constant at their m eans. The effect o f dum m y variable on m arginal probability equals the change in probability w ith the dum m y variable changing from 0 to 1, w hen other variables are held constant at th eir means. R isk-adjusted and raw return perform ance in the prior year seem s to have significant im pact on the probability o f becom ing all-stars. For exam ple, a one standard deviation im provem ent in analysts’ prior-year risk-adjusted perform ance increase the 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. Table 3: Predicting Institutional Investor All-stars 11 a ll-sta rs T a b le 3 re p o rts th e re s u lts o f p ro b it m o d e l E q u a tio n (3 ) w h e re the d e p e n d e n t v a ria b le is a 0-1 v a ria b le th a t e q u a ls 1 if a n a ly s ts a rc ra n k e d as fro m 1996 to 2 0 0 0 . In c o lu m n ( l ) - ( 3 ) , ris k -a d ju s te d re tu rn s on a n a ly st re c o m m e n d e d p o rtfo lio s a re u se d to m e a su re p e rfo rm a n c e . In c o lu m n (4 ), 1 use in d ic a to r v a ria b le s r a th e r th a n c o n tin u o u s v a ria b le s to m e a s u re a n a ly s t p e rfo rm a n c e . In c o lu m n (5 )-(8 ), ra w re tu rn s o r a v e ra g e re tu rn s o n a n a ly st p o r tfo lio s a re u se d to m e a s u re p e rfo rm a n c e . S in c e Wall Street Journal ra n k s a n a ly s ts b y th e ir p e rfo rm a n c e w ith in e a c h in d u s try , I a ls o u se it a s a ra w re tu rn p e rfo r m a n c e m e a s u re . F o r b o th ris k -a d ju s te d p e rfo rm a n c e a n d ra w re tu rn p e rfo rm a n c e , 1 c a lc u la te d re tu rn s o n in d iv id u a l a n a ly s t p o rtfo lio s o v e r the p r io r o n e , tw o , a n d th re e y e ars. I re p o rt e s tim a te d c o e ffic ie n t a n d th e ir sig n ific a n c e le v e ls. * **, **, a n d * in d ic a te th a t t-s ta tis tic s a re sig n ific a n t at 1% , 5 % , a n d 10% le v e ls, re s p e c tiv e ly . T h e e n trie s in the b ra c k e ts a re th e e ffe c t o f o n m a rg in a l p ro b a b ility . T h e d a ta a re d a ily fro m J a n u a ry 1994 th ro u g h D e c e m b e r 2 0 0 0 . V a ria b le R a w R e tu rn P e rfo rm a n c e R is k -a d ju s te d P e rfo rm a n c e P a s t 2P a st 1-p e rio d P e rfo rm a n c e 0 .3 6 * * [1.1 1 ] p e rio d [0 . 1 0 P a s t 1-p e rio d P a st 2 -p e rio d P a st 3 -p e rio d -0 .0 5 0.0 2 ] WSJ R a k in g A s y m . E ff. P a st 3 -p e rio d [- 0 .2 2 0 .3 0 * ] T O P 10% [0 .9 8 ] -0 . 0 2 0.01 [0 .0 3 ] 0 .1 0 [-0 .0 8 ] [1 .6 5 ] 0 .0 7 [1 .1 8 ] B O T T O M 10% -0 . 1 2 [- 2 . 1 0 J E ffo rts LNNREC 0 .0 9 * 11 LNNSTK .1 A A //0 -1 ) R e c o g n itio n LNBRKSZ ] 0 .0 0 [0 .0 R e p u ta tio n 2 2 2 2 1 ] *** [3 7 .8 5 ] 0.51 *** [ 8 .7 7 ] LN STK CA P 0 .1 2 *** [ 3 .0 1 ] 1PO R A N K 0 .1 6 * * [ 2 .7 6 ] FU N D 300 # o f O bs. P seu d o -R S Q 0 .1 5 * * 0 .1 0 * 0 .1 0 * 0 .1 0 * [1 .1 7 ] [1 .1 7 ] [1 .1 6 ] 0 .0 0 0 .0 0 0 .0 0 [ 0 .0 0 2.21 ] *** [3 7 .7 5 ] 0.51 *** [ 8 .7 6 ] 0 .1 2 *** [ 3 .0 1 ] 0 .1 6 * * [ 2 .7 8 ] 0 .1 5 * * [ 0 .0 0 2.21 ] *** [ 3 7 .7 5 ] 0.51 *** [ 8 .7 7 ] 0 .1 2 *** [ 3 .0 4 ] 0 .1 6 * * [ 2 .7 8 ] 0 .1 5 * * 0 .0 0 [-0 .0 3 ] 2.21 *** [3 7 .7 7 ] 0 .5 1 * * * [8 .7 7 ] 0 0 .0 9 * [1.12] . 1 2 *** [3 .0 0 ] 0 .1 6 * * [2 .7 9 ] 0 .1 5 * * [0 .0 3 ] 2 .2 1 *** [3 7 .8 2 ] 0 .5 1 * * * [8 .7 2 J 0 . 1 2 *** [3 .0 0 ] 0 .1 6 * * [2 .8 1 ] 0 .1 5 * * 0 .1 0 * 0 .1 0 * 0 .0 9 * [1 .1 7 ] [1 .1 7 ] [1 .1 4 ] 0 .0 0 0 .0 0 0 .0 0 [ 0 .0 0 2.21 ] *** [3 7 .7 5 ] 0 .5 1 * * * [8 .7 6 ] 0 . 1 2 *** [3 .0 1 ] 0 .1 6 * * [2 .7 8 ] 0 .1 5 * * [- 0 .0 1 2.21 ] *** [3 7 .7 5 ] 0.51 *** [ 8 .7 7 ] 0 .1 2 *** [ 3 .0 3 ] 0 .1 6 * * [ 2 .7 7 ] 0 .1 5 * * [- 0 .0 2 1 ] . 2 0 *** [3 7 .6 6 ] 0 .5 1 * * * [8 .7 4 ] 0 . 1 2 *** [2 .9 8 ] 0 .1 6 * * [2 .7 6 ] 0 .1 5 * * [2 .6 0 ] [2 .6 1 ] [2 .6 0 ] [2 .5 8 ] [2 .6 2 ] [2 .6 1 ] [ 2 .6 0 ] [2 .6 0 ] 5421 5421 5421 5421 5421 5421 5421 5421 0 .5 8 0 .5 8 0 .5 8 0 .5 8 0 .5 8 0 .5 8 0 .5 8 0 .5 8 probability o f becom ing I I all-stars by 1.11%, about equal to raw return perform ance’s im pact o f 0.98% . R ecall from T able 2 that the unconditional probability o f becom ing II all-stars is about 17.89% . If the cross-sectional d istribution o f perform ance is approxim ately normal, bein g a top 5% best risk-adjusted performer, or a tw o standard deviation change in perform ance, increases the unconditional probability by about 13%. In contrast, perform ance beyond the prior one-year period seem s not to affe c t the ranking significantly, since the m arginal effect o f perform ance over the prior 2 o r three years are sm aller than that of the p rio r one-year perform ance. I also exam ine individual-year perform ance and co efficien t estim ates for perform ance beyond the first y e a r are not significant. The sim ilar and som etim es slightly stronger effect o f risk-adjusted perform ance in both statistical and eco n o m ic significance supports the anecdotal evidence that investors concern m ostly about risk-adjusted returns, even in the context o f individual analyst portfolios [Brown (2001)]. Since the results of risk-adjusted and raw return performance are largely similar, I o n ly present the results for risk-adjusted perform ance in the rest of this section. Analyst efforts a lso im prove the probability o f being selected. An increase in the num ber o f recom m endations they m ake significantly im proves the probability o f becom ing II all-stars by sim ilar m agnitude as perform ance does. Yet, II ranking seems to depend m ore on reputation and recognition. B rokerage firm size, size o f com panies analyst cover, and past II all-star status all have m uch m ore significant im p act than perform ance or efforts. F o r exam ple, the boost in m arginal probability is about 9% , 3%, 3% , and 3%, respectively, w hen brokerage firm size and the size of com panies analysts 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. cover increase by one standard deviation and w hen analysts are em ployed by brokerage firm s with top 10 IPO underw riting ranking and to p 300 fund m anagem ent in the U.S. T he increase in marginal probability is about 38% w hen analysts w ere all-stars in the previous year. I also exam ine the im pact o f analyst characteristics o th er than performance in the prior tw o and three years and find that only the prior one-year characteristics m atter. A sym m etric Effect o f Perform ance. Hong and Kubik (2002) show s, for jo b turnovers, that the reward o f being the most accurate earnings forecaster is not as great as the punishm ent o f being the worst perform ers. T o exam ine potential asym m etric effect o f recom m endation performance, I estim ate Equation (3) by using tw o indicator variables for analysts with the best and the w orst 10% perform ance. This specification also makes the im pact o f perform ance more com parable to that o f past all-star status. Although coefficient estim ates for perform ance reported in th e fourth column o f Table 3 suggest that being the top 10% perform ers increases the c h an ce o f becom ing II all-stars by 1.18%, while being the w orst 10% perform ers red u ces the chance by 2.10% , they are statistically insignificant. II A ll-stars vs. N on-all-stars in the Prior Year. Because o f the dom inant effect of past II all-star status on future II all-star ranking, it m ay be more inform ative to examine the relation betw een the ranking and analyst characteristics separately for past II ail-star and non-all-star analysts, which should help us understand the effectiveness o f the //a llstar ranking to m otivate analyst perform ance for non-all-stars and all-stars, respectively, 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without perm ission. T a b le 4 Predicting Institutional Investor All-stars Given Analyst Status in the Prior Year T a b le 4 e x a m in e s th e e ffe c t o f p a st re tu rn s o n in d iv id u a l a n a ly st p o rtfo lio s o n th e lik e lih o o d th a t th e y b e c o m e a ll-s ta rs o r m o v e u p a n d d o w n d iffe re n t lev el o f a ll-s ta r te a m s, g iv e n th e ir a ll-s ta r s ta tu s in th e p r io r y e a r a s s h o w n in e q u a tio n (4 ). A n a ly s ts w h o h a v e a t le a st th re e p r io r y e a rs o f e x p e rie n c e a re in c lu d e d . T h e d e p e n d e n t v a ria b le is a 0-1 v a ria b le th at e q u a ls 1 i f a n a ly s ts a re ra n k e d a s 11 a ll- s ta rs fro m 1996 to 2 0 0 0 in c o lu m n ( l ) - ( 2 ) , a 0-1 v a ria b le th at e q u a ls 1 if p rio r- y e a r a ll-s ta r a n a ly s ts m o v e u p a lls ta r te a m s b y at le a st o n e le v e l in c o lu m n (3 ), a 0-1 v a ria b le th a t e q u a ls 1 if m o v e d o w n a ll-s ta r te a m s by a t le a s t o n e le v e l in 11 a ll-s ta rs m o v e d o w n th e a ll-s ta r te a m s b y at le a st o n e II a ll- s ta rs w h o d o not b e lo n g to the first te a m sta y in th e ir o rig in a l te a m s , tw o i f p a st first te a m II a ll-s ta rs sta y in th e first te a m , a n d th re e i f p a s t 11 a ll-s ta rs m o v e u p th e a ll-s ta r te a m s b y a l le a s t o n e le v e l. In c o lu m n (4 ). In c o lu m n (5 ), th e d e p e n d e n t v a ria b le e q u a ls z e ro i f p a st lev e l o r b e c o m e n o n -a ll-s ta rs , o n e i f p a st c o lu m n ( l ) - ( 2 ), 1 e s tim a te the re la tio n b e tw e e n fu tu re a ll-s ta r sta tu s a n d c u rre n t a n a ly s t c h a ra c te ris tic s fo r p r io r y e a r a ll-s ta rs a n d n o n -a ll- s ta r s , s e p a ra te ly . In c o lu m n (3 )-(4 ), I in v e s tig a te , fo r a ll-s ta r a n a ly s t o n ly , th e re la tio n b e tw e e n th e d ire c tio n o f c h a n g e s in le v e l o f a ll-s ta r s ta tu s a n d a n a ly s t c h a ra c te ris tic s . 1 re p o rt e s tim a te d c o e ffic ie n t a n d th e ir s ig n ific a n c e le v e ls. * * * , **, and * in d ic a te th a t l-s ta lis tic s a re sig n ific a n t at 1% , 5 % , a n d 10% le v e ls, re s p e c tiv e ly . T h e e n tr ie s in th e b ra c k e ts a re th e e ffe c t o f o n m a rg in a l p ro b a b ility . T h e d a ta a re d a ily fro m J a n u a ry 1994 th ro u g h D e c e m b e r 2 0 0 0 . P e rfo rm a n c e R A JR E T E ffo rts LNNREC N o n - a ll-s ta r A ll-s ta r E le c tio n R e e le c tio n 0 .5 0 * * [0 .3 4 ] 0 .1 9 * * * [0 .5 0 ] LNNSTK R e c o g n itio n LNBRKSZ P s e u d o -R S Q 0 6 ] .12 * [2 .7 0 ] -0 . 1 1 -0 . 1 2 [-1 .7 2 ] 0 .6 5 * * * 0 .1 6 * * * 0 .3 8 * * * 0 .0 7 [0 .2 6 ] # o f O b s. [2 .3 0 ] [ 2 .6 [-4 .0 2 ] [1 .3 8 ] FU N D 300 0 .0 4 0 .5 5 * 0 .0 3 [0 .8 7 ] 1PO R A N K [ 3 .0 0 ] M ove Up [0 .0 7 ] [ 2 .3 6 ] LN STKCA P 0 .2 4 B in o m ia l P ro b il 4451 0.21 O rd e r e d P ro b it M ove D ow n -0 .4 5 * [- 2 . 8 8 ] -0 . 1 2 * 0 .2 1 * 0 .0 4 [0 .6 1 ] [-4 .6 0 ] 0 .0 3 0 .0 6 -0 .0 5 [3 .0 9 ] 11.99] [-2 .6 5 ] [-2 .6 5 ] 0.11 [9 .6 8 ] 0.1 2 [ 8 .8 7 ] [3 .5 9 ] 970 970 0.01 0 .0 2 .1 1 * -0 .1 8 * [3 .8 8 ] 0 .1 9 * 0 .3 1 * * 0 [-3 .4 2 ] [ 6 .8 4 ] -0 .0 3 0 .4 9 * * -0 .2 5 * * 0 .1 7 0 .0 5 * -0 . 0 1 0.11 [-0 .3 3 ] -0 . 0 1 0 .0 2 [ 0 .6 6 970 0.01 ] 970 and the different career c o n c ern s facing ranked and unranked analysts. It also helps clarify whether the im portance o f reputation and recognition is due to som e unknow n restrictions of / / on analyst eligibility that are related to these factors. F o r exam ple, since past II all-stars are likely to m eet those restrictions, performance m ay be m ore im portant to their reelection, while reputation and recognition factors m ay becom e m uch less im portant. To exam ine this issue, I estim ate Equation (3) separately for past II all-star and non-all-star analysts the past II all-star statu s and report the results in the First and second colum n o f Table 4. The coefficient on perform ance for non-all-stars suggests that a one standard deviation improvement in perform ance for non-all-stars could increase their probability o f becom ing II all-stars by 0 .3 4 % . Since the unconditional probability o f prom otion to II all-stars is 3.66%, a one sta n d a rd deviation im provem ent in perform ance accounts for about 9% of the unconditional probability. A nalyst efforts help too. A one standard deviation increase in the n u m b er of recom m endations analysts m ake im proves the unconditional probability by about 14%. H ow ever, the impact of perform ance and efforts is lim ited compared to reco g n itio n factors, sim ilar to the overall analyst sam ple. For exam ple, a one standard d e v ia tio n increase in brokerage firm size accounts for about 65% o f the unconditional p robability, and a sim ila r increase in the size o f the com panies that analyst cover accounts for ab o u t 24%. B eing in the top 10 IPO underw riting brokerage firm s improves the unconditional probability by 38% . For past II all-stars, th e only factor th at has statistically significant effect o f rem aining all-stars is the size o f analysts’ brokerage firms. Since the unconditional probability of past II a ll-star rem aining all-stars is about 79.79% , a one standard deviation 130 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. increase in the size o f past II all-stars’ brokerage firm size accounts for about 9% o f the unconditional probability. T hese results suggest that perform ance and effort are only im portant w hen nonall-stars w ant to becom e II all-stars. O nce analysts become I I all-stars, perform ance is not im portant for reelection. In addition, recognition are still far m ore im portant than perform ance or efforts for both II all-stars and non-all-stars, excluding the possibility that the dom inant im pact of reputation and recognition is due to eligibility restrictions related to these factors. It appears that for non-all-stars to become II all-stars or for existing a ll­ stars to secure status, the most effective w ay is not to im prove perform ance or effo rts, but to sw itch to larger brokerage firms o r to cover larger stocks. P erform ance and Rank C hanges across A ll-star Team s f o r P ast All-stars. A lthough the above results suggest that past II all-stars are sim ilar to a clique and are very unlikely to be voted off the ranking because of perform ance, it is still possible that they have to com pete with each o th e r to stay in the same team or m ove up the team raking. It is also possible that they have to w ork hard to prevent m oving down the team ranking. D ifference in com pensation can be m illions of dollars even between the first and second team s o f II all-stars. To explore this possibility, I investigate whether the d irectio n o f m ovem ents am ong different team s provides incentive for existing II all-stars to im prove perform ance. Specifically, I estim ate a probit model P r ( D O M , = 1) = [...]... conditions indicates increasing distortions in financial analysts incentives in recent years, instead o f a m ore bullish position by analysts in the later part o f my sam ple period A verage characteristics o f analysts are reported in Panel C 10 T h e m ean market capitalization o f stocks analysts co v er increases significantly over tim e w ith a range of $3 to $13 billion (Stkcap) T he sh arp increase... Zacks R ecom m endations from these three firm s compose about 10% o f all the recom m endations.3 A nother advantage is that its time period is the 1990s, “The A ge o f Analysts Few previous studies have exam ined this period w hile the influence and bias of analysts have both increased trem endously during this period S econd, this study provides a n um ber of im provem ents to the research design... evidence in favor of m arket efficiency is subject to the sam e problem s.1 To shed new light on the research on analyst recom m endations, this article pursues three lines o f inquiry It first evaluates the perfom iance o f recom m ended buy and sell portfolios o f individual analysts T he study of individual an aly sts’ portfolio recom m endations is facilitated by a new source of data from Institutional... aggressive I find that AllA m erican analysts who have m ore reputation capital tend to recom m end more conservative portfolios and deviate significantly less often from the portfolios recom m ended by the representative analyst Other characteristics also affect their behavior For exam ple, analysts co v erin g large firms or m ore stocks tend to select less risky portfolios and analysts in larg e brokerage... permission of the copyright owner Further reproduction prohibited without permission C H A PT E R I PE R FO R M A N C E A N D BEHAV IO R O F IN D IV ID U A L FIN A N C IA L ANALYSTS Introduction A cadem ic researchers are div id ed on the question o f w hether following recom m endations o f analysts generates superior returns Evidence follow ing the first study by C ow les (1933) suggests th at analysts. .. ore profitable if good perform ers could be identified so that only their recom m endations are fo llow ed/’ 7 As is said in the mutual fund industry, “Buy the m anagers, not the fund” [C ullen et al (2000)] S tudying the average perform ance of financial interm ediaries such as brokerage firms m ay be m uch less interesting because the valuable elem ent o f a sell-side research departm ent is its analysts, ... m arket participants, including professional investors, use this database IBES has collected buy and sell recom m endations from the research reports o f financial analysts since the end o f O ctober 1993.8 The database includes both the ratings based on the system s adopted by individual brokerage firms and a standardized IBES rating The form er are usually on a three- to five- level scale T he IB... Journal A ll-A m erican S ta tu s .144 9 Predicting D eparture from Analyst P rofession 146 10 Predicting D eparture from Profession G iven A nalyst Status 149 11 The Effect o f Past Performance and R isk-taking Behavior on Leave from Profession 152 viii Reproduced with permission of the copyright owner Further reproduction prohibited without permission LIST O F FIGURES... in their sell portfolios over the sam ple period It also presents m ean deciles of several com m on characteristics o f com panies covered by analysts, including size, book-to-m arket ratio (BM), m om entum (M O M ), share turnover (TO ), and earnings/price (EP) at the tim e of the recom m endation, categorized by type of recom m endation and y e a r.1" Decile 1 (decile 10) includes stocks w ith the... f Table 1 also indicates that analysts overall tend to cover grow th over value stocks, w ith mean book-to-m arket decile alw ays above that of overall m arket for both buys and sells They also recom m end stocks w ith higher book-to-m arket ratios as purchases T his could reflect a tem poral trend by analysts to cover more g ro w th stocks Panel D also suggests that analysts recom m end stocks w ith