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

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 p ag es, th e se 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 w ithout permission.

T H R E E E S S A Y S ON F IN A N C IA L A N A LY STS

By

Xi Li

D issertation

S ubm itted to the F aculty o f the

G ra d u a te School o f V anderbilt U niversity

in p artial fulfillm ent o f the requirem ents

for the degree o f

D O C T O R O F P H IL O S O P H Y

in

M anagem ent August, 2002

C opyright 2002 by P ro Q u e s t Inform ation and L earning C om pany

All rights reserv ed T his m icroform edition is p ro te c te d a g a in st unauthorized copying u n d e r Title 17, United S ta te s C o d e.

P roQ uest Inform ation a n d Learning C o m p a n y

30 0 North Z e e b Road

P O Box 1346 Ann A rbor, Ml 48106-1346

C o p y rig h t © 2002 by Xi Li

A ll R ig h ts Reserved

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

To M y Parents, Jiannan Li and Z h u angping Sun

and M y wife, lin g M a

A C K N O W L E D G E M E N T S

T his research project was su ccessfully com pleted th an k s to m any persons who helped m e at various stages I thank m y dissertation co m m ittee m em bers, N ick Bollen, Paul C haney, C raig Lewis, H ans Stoll, and especially my d issertation chairm an, Ronald

M asulis, for providing precious advice and support I also ap p reciate the valuable help of

B ruce C ooil and C hristoph Schenzler I also thank the financial support o f 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

A m 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 M a, 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 P E R FO R M A N C E A N D BE H 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

In tro d u ctio n 1

D ifference from Previous L iterature 3

D ata 8

Experim ental D e sig n 14

Em pirical R e su lts 20

C o n clu sio n s 48

R eferen ces 51

II W IL L P A S T LEA D ER S ST IL L LEA D ? PE R FO R M A N C E PE R SIST E N 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 e th o d o lo g y 61

T w o-P eriod P erform ance P ersistence 70

M ulti-Period Perform ance P ersistence 88

C o n c lu sio n s 95

R eferen ces 98

III C A R EER C O N C E R N S O F EQ U ITY A N A L Y ST S : C O M PEN SA TIO N , T E R M IN A T IO N , A N D P E R F O R M A N C E 101

In tro d u ctio n 101

Related L ite ratu re 108

Sam ple, R ankings, and Perform ance M e asu re m e n t 111

Em pirical A n a ly sis 123

C o n c lu sio n s 153

R eferen ces 156

L IST O F TABLES

C hapter I

1 S u m m a ry Statistics of R ec o m m e n d a tio n s 11

2 J e n s e n ’s A lpha o f Factor R eg ression For Test P o rtfo lio s 18

3 P erfo rm an ce o f Analysts as a G ro u p 22

4 P erfo rm an ce as a Group: O th e r F actor M odels 26

5 P erfo rm an ce as a Group: P ortfo lio s Rebalanced L ater Than R e co m m en d atio n D ate 29

6 C ro ss-sectio n al D istribution o f Individual Analyst P erform ance 37

7 C ro ss-se c tio n a l D eterm inants o f A nalyst Perform ance, R isk T aking Behavior, and A g g ressiv en ess 43

C hapter II 1 S u m m a ry Statistics of R ecom m endation D atabase 63

2 T w o -p e rio d Perform ance P ersisten ce over the W hole Sample P erio d 71

3 P e rsiste n c e T est o f T w o-period Perform ance by P airs o f C o n se c u tiv e Subperiods 75

4 C o n tin g en c y Table of W inners and Losers over the W hole Sample P eriod 78

5 R isk -a d ju ste d Perform ance o f D ecile Portfolios C reated A c c o rd in g to Prior-period R isk A djusted P erform ance 82

6 R isk -a d ju ste d Perform ance o f P ortfolios Created A c c o rd in g to Prior-period R aw R eturn Perform ance 85

7 B u y -a n d -h o ld Returns o f P o rtfolios C reated A c c o rd in g to Previous P eriod Raw R eturn Perform ance 87

8 P e rsiste n c e T est for M ulti-period P erform ance 92

C hapter III

1 S um m ary S tatistics of the IBES R ecom m endation D atabase 113

2 S um m ary S tatistics of A ll-star R an k in g 118

3 P redicting Institutional Investor A ll-stars 126

4 P redicting Institutional Investor A ll-stars G iven A nalyst Status

in the P rio r Y e a r 129

5 T he E ffect o f Past Perform ance and R isk-taking B ehavior

on the In stitu tio n a l Investor A ll-A m erican S ta tu s 134

6 P redicting W all Street Journal A ll-stars 137

7 P redicting W all Street Journal A ll-stars G iven A nalyst Status

in the P rio r Y e a r 140

8 T he E ffect o f Past Perform ance and R isk-taking B ehavior

on the W all S treet Journal A ll-A m erican S ta tu s 144

9 P redicting D eparture from A nalyst P ro fe ssio n 146

10 P redicting D eparture from Profession G iven A nalyst S tatu s 149

11 T he E ffect o f Past Perform ance and R isk-taking Behavior

on Leave from Profession 152

viii

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

o th er researchers find b oth a strong ev e n t-p erio d abnorm al return w hen recom m endations are revised and a sig n ifican t post-event return drift that lasts a m o n th o r even longer [Barber, Lehavy, M cN ichoIs, and T ru e m a n (2001), Elton, G ruber, and G rossm an (1986) and W om ack (1996)] Post-event retu rn drift is evidence against m arket efficiency

A lthough a finding o f abnorm al retu rn s is usually attributed to sam p le lim itations, inaccurate perfo rm an ce m easurem ents, and insufficient risk adjustm ents, evidence in favor of m arket e fficie n cy 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 ev alu ates the perfom iance o f recom m ended buy and sell portfolios o f individual analysts T he study o f individual a n a ly sts’ portfolio

recom m endations is facilitated by a n ew source of data from Institutional B rokers Estim ate System (IB E S ) that includes a m ore com prehensive set o f brokerage firm s and individual financial a n aly sts than p rev io u sly available W ith m ore accurate m easurem ent

o f analyst perform ance a n d ex tensive risk adjustm ents, I find that the eq u ally w eighted

p ortfolios o f individual a n a ly sts’ recom m ended portfolios generate significant abnorm al returns T he abnorm al retu rn s, 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 ben ch m ark s 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 th e abnorm al perform ance reveals that it is generated m ainly

w ithin an event w indow startin g at tw o trading days before the recom m endation dates until about five trading d a y s later, w ith no significant post-event return drift T he

d isappearance o f return d rift is m ostly due to more com plete risk adjustm ents T he 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 stro n g , short-term nature o f abnorm al perform ance by analysts is related to R egulation FD w hich cu rren tly only requires synchronous inform ation release

by co m p an y m anagem ent to all investors If preferred investors o f analysts such as the firm ’s traders do obtain p rio 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 o b jectiv e is to provide new evidence on the cross-sectional

d eterm inants o f analyst p erfo rm an ce and the relationship betw een analyst perform ance and inform ation en v iro n m en t.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 i n g s f o r e c a s ts F r a 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

perform ance differences o f in d iv id u al analysts’ recom m ended portfolios Individual analyst perform ance im proves sig n ifican tly with the n u m b er o f recom m endations issued and w ith the size o f their brokerage firm s The num ber o f sto ck s covered also has a significantly positive, but concave relationship with perfo rm an ce T he optim al num ber o f

stocks is betw een 12 and 13 A d d itio n al evidence suggests that Institutional Investor (II)

A ll-A m erican status and the size o f the companies they c o v e r have little power in predicting analyst perform ance

T h e third goal is to in v estig ate the effect o f an aly st c a re e r concerns on their behavior Scharfstein and Stein (1 9 9 0 ), Prendergast and Stole (1996), and Zw iebel (1995) all suggest that ag en ts’ career c o n c e rn s should affect th eir behavior They predict that som e ag en ts will stay w ith the herd w hile others will be m ore aggressive I find that A ll-

A m erican analysts w ho have m ore reputation capital tend to recom m end more conservative portfolios and d ev iate significantly less often from the portfolios recom m ended by the representative analyst Other ch aracteristics also affect their behavior F or exam ple, analysts c o v e rin g large firms o r m ore sto ck s tend to select less risky p o rtfo lio s and analysts in larg e brokerage firms o r m aking m ore frequent

recom m endations tend to recom m end more risky portfolios

D ifference from Previous L iterature

T h is article is very differen t from previous studies In the first part o f perform ance evaluation, I im prove on all three asp ec ts that are the focus o f the controversy about analyst perform ance: Sam ple lim ita tio n , insufficient risk ad ju stm en ts, and inaccurate

perform ance m easurem ents First, I u s e a com prehensive d a ta set and exam ine the m o st recent tim e period T he IBES datab ase used has much m ore com prehensive c o v erag e o f brokerage firm s and analysts than a n y database used in the p rev iously published research

It includes m any more analysts from sm a lle r brokerage firm s than large databases su c h as First C all and Zacks It also includes im portant brokerage firm s such as M errill L ynch,

G oldm an S achs, and D onaldson, 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

A nalysts” Few previous studies have exam ined this period w hile the influence and bias

of analysts have both increased tre m en d o u sly during this period

S e c o n d , this study provides a n u m b er of im provem ents to the research d esig n for evaluating analyst perform ance O ne im provem ent incorporates the recent advances in long-run perform ance evaluation lite ra tu re [Brav, Geczy, and G om pers (2000), D aniel,

G rinblatt, T itm an , and W erm ers (1 9 9 7 ), and Eckbo et al (2 0 00)] W ith more careful risk adjustm ent, I obtain m ore accurate e v id e n c e on market e fficie n cy as it pertains to lo n g ­term p erform ance I also evaluate the perform ance o f both eq u al- and value-w eighted analyst p o rtfo lio s.4 A second im p rovem ent is to em ploying the m ethodology in the recent mutual fund perform ance literature to ev aluate individual a n a ly s ts ’ recom m endations on

a daily b a sis, w hich allow s more e ffic ie n t coefficient estim ates [B ollen and B usse (2 0 0 1 ) and B usse (1999)] D aily data can a ls o dem and a shorter tim e series for individual

a n a ly s ts , m y e v id e n c e is c o m p le m e n ta r y to t h e p r e 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 a s e T h i s p e r c e n ta g e w ill b e e v e n la r g e r c o m p a r e 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 ffe r th e r e c o m m e n d a tio n fro m a s m a n y s m a ll e r b ro k e ra g e fir m s a s I B 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 f o r m a n c e o f c ith e r value- o r e q u a l - w e ig h t e d p o rtfo lio s S o m e

d is a g r e e m e n t e x is ts a s to w h e th e r v a lu e - o r e q u a l- w e ig h te d p o rtfo lio s a r c th e b e s t c h o ic e fo r te s ts o f

p e r fo r m a n c e o v e r lo n g h o r iz o n s T o test f o r a b n o r m a l p e rfo rm a n c e , a n e q u a l l y w e ig h te d p o r tf o lio is m o re

analysts and th u s reduce potential su rv iv o rsh ip bias M o n th ly d a ta are also used fo r the purpose o f corroboration The third im provem ent is m ore frequent updating o f the

m atching p o rtfo lio s and factors W h ile m ost existing research updates the factor portfolios a n n u ally , I construct b o o k -to -m ark et and e a m in g s/p ric e factors or m atching portfolios q u arterly , and size, m om entum , and liquidity factors o r m atching portfolios

m onthly T h is frequent updating sh o u ld im prove the accu racy o f risk adjustm ent benchm arks

The fo u rth im provem ent is to m easure analyst perfo rm an ce more precisely than existing stu d ies that focus on long-run perform ance B ecau se analysts may revise their recom m endations within weeks or m o n th s after the original recom m endation, I k eep the stocks in the analyst portfolio until an a ly sts revise their recom m endations Previous studies fo llow s recom m endations o f an aly sts for an arb itrary holding period such as 6 or

12 m onths, u su a lly because they lack recom m endation revision dates This type o f assum ption c o u ld m isrepresent an aly st perform ance.5 M y experim ental design is also advantageous com pared to studies ex a m in in g the event effect o f recom m endation revisions b e c au se 1 can exam ine the post-event return drift flexibly, and com pare the

m agnitude o f event-period abnorm al retu rn s and post-event return drift This stu d y o f recom m ended portfolios is also o f in te rest because this is how brokerage houses su g g est that custom ers use their recom m endations [Elton et al (1 9 8 6 ), and Jasen (2001)]

T he m o st im portant im p rovem ent in experim ental design is the study o f individual a n a ly s ts ’ recom m ended p o rtfo lio s The ex istin g research has exam ined

re a s o n a b le T o a s s e s s th e w e alth e ffe c t o n in v e s to r s o f fo llo w in g r e 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 r e 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 firm s and analysts T he inability to identify good analysts sig n ifican tly im pairs 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 sin g le brokerage firm, but even individual investors can generally trade on the recom m ended portfolios o f in d iv id u al analysts at low transaction costs Second, as Barber e t al (2000) point out, an in vestm ent strategy based on

recom m endations w ould be m ore profitable if good perform ers could be identified so that only their recom m endations are fo llo w e d /’ 7 As is said in the m utual fund industry, “Buy the m anagers, not the fund” [C ullen et al (2000)] S tu d y in g the average perform ance of financial interm ediaries such as brokerage firms m ay be m uch less interesting because the

v aluable 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 a ffect the perform ance o f brokerage firm s T he study o f recom m ended portfolios also en a b le s 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 o rtfo lio 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 p erform ance o f frequently used factor

m odels w ith daily data and p ro v id e the first horse race b e tw e en traditional factor m odels and th e m acro factor m odels u se d in Eckbo, M asulis, and N orli (2000) In addition, this article exam ines the cro ss-sectional perform ance differences in an alysts’ recom m ended portfolios and the proportions o f good and bad analysts.

The advantages o f m y sam p le and experim ental d e sig n for the perform ance evaluation naturally extend to th e second and third parts o f th is article In ad d ition, in the seco n d part o f this article, I stu d y the cross-sectional d eterm inants o f analysts’ p o rtfo lio perform ance instead o f event p e rio d abnorm al perform ance P ortfolio perform ance includes both event period ab n o rm al perform ance and any potential post-event abnorm al perform ance generated by a n a ly st recom m endations It sh o u ld be a m ore com prehensive

m easure for overall analyst p erform 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 o n the behavior o f analysts with d iffe re n t age or experience [C h ev alier and Ellison (1998),

H ong, K ubik, and Solom on (1 9 9 9 ), and Lam ont (1995)] In addition, I use investm ent recom m endations rather than e a rn in g s forecast data as in H o n g et al (1999), the on ly

ex istin g study about the im pact o f analyst career concerns o n their behavior

The article is organized as follow s: Section 2 describ es the sam ple S ection 3 discusses the econom etrics o f th e factor m odels and benchm arks the perform ance o f various factor m odels It also g iv es details about the m ethodology used to form the analyst portfolios and their m atching portfolios Section 4 p resen ts the em pirical results

S ection 5 offers concluding rem arks

7 A n e c e s s a r y c o n d itio n fo r id e n tif y in g g o o d a n a ly s ts to b e a really p r o f i ta b l e s tr a te g y is th a t c u r r e n t g o o d

p e r f o r m e r s w ill d o w e ll in the fu tu re

7

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

DataThe prim ary database u se d in this paper comes from IBES Its m ajo r benefit is that it includes recom m endations from a very broad sam ple o f brokerage firm s and financial analysts Even large d a ta b a se s such as Zacks do not include im portant bulge bracket firms such as M errill L y n ch , G oldm an Sachs, and D onaldson, L ufkin, & Jenrette

T he IBES database includes all m ajo r brokerage firms plus a large sam ple o f sm aller brokerage firms A nalysts can a lm o st alw ays be tracked even if they sw itch brokerage firm s Various m arket particip an ts, including professional investors, use this database

IBES has collected buy an d sell recom m endations from the research reports o f financial analysts since the end o f O cto b er 1993.8 The database includes b oth the ratings based on the system s adopted b y individual brokerage firms and a stan d ard ized IBES rating The form er are usually o n a three- to five- level scale T he IB E S -created ratings are on a uniform five-level scale; c h a ra c te r ratings o f “strong b u y ,” “b u y ,” “h o ld ,”

“underperform ,” and “sell” c o rre sp o n d to num eric ratings from I through 5

Recom m endations with num eric ratin g s o f 1 are used to form the buy portfolios o f financial analysts, and the sell p o rtfo lio s are formed using reco m m endations w ith ratings

o f 4 and 5.9 The investm ent reco m m en d atio n data are from the end o f O cto b er 1993 to

D ecem ber 2000 The return and acco u n tin g data are draw n from C R S P and C om pustat, respectively

Panels A and B o f T ab le 1 su m m arize the database T h ere are 241,222 recom m endations by 7,308 fin an cial analysts from 408 institutions in the five buy and

8 B e c a u s e the d a te s o n th e r e s e a r c h r e p o r t s u s u a lly p r e c e d e the d a te s a n a ly s ts a c tu a lly d e l i v e r th e r e p o r ts to

th e p u b lic , I th ere fo re u se “ r e p o r t d a t e s ” o r “ r e 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 r e p o r ts

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 t h 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

T h is is b ecau se 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

sell recom m endation categories In the em pirical analysis, I exclude an aly sts with fewer than 10 recom m endations T h e rem aining sample consists o f 4,383 a n a ly sts before other restrictions are applied Panel B in d icates that favorable reco m m endations are much more prevalent, consistent with o th e r d atabases The ratio o f strong buys to sells for the entire sam ple period is about 1 5 -to -l Panel B also suggests that both the n u m b e r o f negative recom m endations and their p erc e n ta g e o f all recom m endations decline o v e r tim e, despite the grow ing num ber of total recom m endations m ade each year T his su g g ests that the

b u y -to -se ll ratio declines c o n tin u o u sly throughout my sam ple period T o put the changes

in the buy-to-sell ratio in a h isto rical perspective, according to Z acks Investm ent

R esearch, the ratio of “buy” an d “stro n g b uys” to “underperform ” and “se ll” 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 [L aderm an (1998)) This d ram atic m onotonic decline in negative recom m endations independent o f market c o n d itio n s indicates increasing distortions in financial analysts’ incentives in recent years, in stead o f a m ore bullish position by an alysts in the later part

o f my sam ple period

A verage characteristics o f a n alysts are reported in Panel C 10 T h e m ean market cap italizatio n o f stocks analysts c o v e r increases significantly ov er tim e w ith a range o f $3

to $13 billion (Stkcap) T he sh a rp increase in the average m arket cap is m ainly a result o f increased stock price levels d u rin g the sam ple period, as the m ean size decile o f stocks covered by analysts is gen erally b etw een 4 and 5 (Caprk) A nalysts m ak e 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 s u 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 fir m s th at a n a ly s ts c o v e r in a s p e c if ic y e ar 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 issu e

th e ir r e c o m m e n d a tio n s S ec a p p e n d i x f o r d e ta i ls a b o u t how th e siz e 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

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 d a y s (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 c o v e r about 14-15 sto ck s (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 o f absolute increase in size or the significant consolidation in the b ro kerage industry that took place during this tim e p e rio d 11

Panel D o f T ab le 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 o ver the sam ple period It also presents

m ean deciles o f 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 o f the recom m endation, categorized by type o f

recom m endation and y e a r.1" Decile 1 (decile 10) includes stocks w ith the largest (sm allest) m arket cap italization, 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 sp ecific decile The av erage 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 av erage decile would be 5.5 for all characteristics If the average stocks in analyst p ortfolios have characteristic deciles fairly d ifferent from 5.5, this would

11 S o m e s ta tis tic s fo r 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 r ts in O c to b e r 1 9 9 3

P le a s e s e c a p p e n d ix fo r m o r e d e ta ils o f the d e f in itio n o f v a r ia b le s

Table 1

Summ ary Statistics o f Recom m endations

T a b i c 1 r e p o r ts s u m m a r y sta tis tic 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 r y s ta tis tic s r e g a r d i n g the

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 r e tu r n 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 r e fo r m e d q u a rte rly In th e c o lu m n s

u n d e r th e s e ll r e c o m m e n d a tio n s , I r e p o r t 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 t h e s is th a t the

m e a n c h a r a c t e r i s ti c s o f th e b u y and s e ll s a m p le s a r e n o t s ig n if ic a n tly d if f e r e n t *** a n d ** i n d ic a te th a t l-

s t a li s t i c s a r e s i g n i f 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 r e s u lts fo r th e m e d i a n s a re not

r e p o r t e 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 r e fro m O c to b e r 1 9 9 3 th ro u g h

suggest that analysts cover stocks that have very differen t characteristics fro m the overall

m arket

T h e evidence concerning size is consistent w ith previous evidence th a t analysts usually co v e r large-cap stocks T he firm s in the IB E S database are significantly sm aller than the sam ple in W om ack (1996), w ith a m ean d ecile o f about 4.5 The p o ssib le reason

is that W om ack (1996) looks only at th e reco m m endations m ade by the to p 14 All-

A m erican research departm ents ranked by // S m a ller b rokerage firms, su ch as regional firms, usually c o v e r much sm aller stocks

Panel D o f Table 1 also indicates that analysts overall tend to cover gro w th over value stocks, w ith m ean 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 an alysts to cover more g ro w th stocks Panel D also suggests that analysts recom m end sto ck s w ith m om entum c lo s e to overall

m arket as buys and low m om entum sto ck s as sells T he difference in m om entum between buys and sells is sim ilar to w hat is reported in W om ack (1996) The statistics also suggest that analysts co v e r m ore liquid stocks, w hich is to be expected because inv esto rs have

m ore trading interest in those stocks M oreover, analysts also cover more sto c k s w ith

m edian to low earnings/price, a pattern that has becom e m ore significant in recent

i ?years

T o assess the statistical sig nificance 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 m ean a n d m edian ch aracteristics o f buys and sells Since

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 h e average b ook-to-m arket and

m om entum o f analyst buy and sell recom m endations are significantly d ifferen t

C urrent 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 fo r system atic risk If factor m odels are u sed, the models usually include at most the four factors used in Fam a and French (1993) a n d Carhart (1997) B ecause analysts tend to c o v e r stocks with m any differing characteristics and tend to recom m end stocks w ith sig n ifican tly 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 n aly sts’ skills, it is im portant to

m ake sufficient risk or style adjustm ents in evaluating analyst perform ance

E xperim ental D esign

F actor M odel Specification

Portfolio perform ance is m easu red using several specifications o f fac to r models frequently em ployed with m onthly returns P erform ance o f factor m odels is also exam ined using 5 x 5 benchm ark portfolios form ed on the basis o f size an d book-to-

m arket, and 4 x 4 x 4 benchm ark portfo lio s based on size, book-to-m arket, and

m om entum T he specifications include: a single-factor m ark et model; a th ree-factor

m odel that includes size and b ook-to-m arket factors [F am a and French (1 9 9 3 )]; a four- factor m odel that adds m om entum effects [C arhart (1997)]; a five-factor m o d el which

also includes earnings/price factor; and a six -facto r m odel with an ex tra liq u id ity factor.14

A m odified version o f the m acro factor m odels used in E ckbo et al (2000) is also investigated The m odels are m o d ified for use w ith daily data The six -fac to r m odel and the m im icking macro factor m odel are found to m isprice few er test p ortfolios w hile producing higher R -squares am o n g the specifications exam ined

T he factor model is e x p ressed as

; = i

R„ is the excess return on the po rtfo lio o f 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 ‘Je n se n ’s (1 9 6 8 ) alpha’

in a C A P M framework; R is the return o f factor j on day t ; and £lt is th e 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, an d m om entum are popular factors in current asset pricing literature A liquidity factor m easu red by share turnover is tested because B rennan and

S ubrahm anyam (1996) and D atar, Naik, and R adcliffe (1998) find that h ig h er share

tu rn o v er is cross-sectionally rela te d to lower expected stock returns E ckbo and Norli (2000) find this factor is related to the long-run perform ance of seasoned eq u ity offerings

I use an earnings/price factor becau se Fama and French (1992) and Jaffe, K eim , and

W esterfield (1989) find that it ex p la in s the cross-sectional variation in asset returns

14 M y m o m e n tu m facto r is b a s e d o n v a lu e - w e ig h te d r e tu r n s , w h ile C a r h a r t (1 9 9 7 ) w e ig h ts r e tu r n s e q u a lly

E ckbo et al (2000) use a mode! of six m acro factors in th eir stu d y o f long-run perform ance o f SE O I adopt a m odified four-factor version o f th eir m odel because daily data for u n e x p e cte d in flation and real per cap ita co n su m p tio n are un av ailab le T he m acro factors are e x c e ss returns o n the value-w eighted C R S P m arket index; the long-run term spread b etw een T re asu ry b o n d s w ith 30-year and 1-year m aturities (L T S ); the short-run term spread b etw een 180-day and 90-day T reasury b ills (STS); and th e credit spread betw een B A A -rated and A A A -rated corporate bonds (C S ) 16 O f the fo u r factors, only

m arket ex c ess returns are m easured in the form o f returns There can a lso be lim ited variation in th e raw m acro factors To solve the above problem , I c re a te m im icking m acro factors using sto c k returns follow ing Eckbo et al (2 0 0 0 ), except at a d a ily frequency

M im icking fac to rs are in th e form o f returns and sh o u ld include the v ariatio n s in the underlying facto rs P erform ance o f factor m odels u sing raw and m im ic k in g m acro factors

is exam ined

Daily d a ta introduce one notable com plication Scholes and W illia m s (1977) and

D im son (1 9 7 9 ) observe a nonsynchronous trading p roblem in stock retu rn s that hinders regression e stim a tio n for individual securities I address this problem by adding a lagged term for each fac to r in the m o d e l:17

16 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 th e F R E D d a ta b a s e

17 D im s o n ( 1 9 7 9 ) s u g g e s ts i n c l u d i n g a s m a n y as th re e la g s a n d t h r e e le a d te rm s T e s t i n g w ith s e 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 f ir s t la g term is c o n s is te n tly s i g n if ic a n t T h e r e s u l t s u s in g th re e la g an d

th r e e le a d te r m s a r c q u a lita tiv e ly th e s a m e a s u sing o n ly th e f irs t l a g te rm S in c e a n a l y s t p o r tf o lio s ty p ic a lly

in c lu d e se v e ra l s t o c k s , th e p r o b l e 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 fo r

in d iv id u a l s e c u r i t i e s B u s s e ( 1 9 9 9 ) f in d s s im ila r re s u lts f o r d a ily r e tu r n s o f m u tu a l f u n d s F u r th e r m o r e ,

a n a ly s ts u s u a lly c o v e r h i g h - li q u i d it y s to c k s and th e s e s to c k s s h o u l d h a v e less p r o b l e m o f n o n s y n c h r o n o u s

tr a d in g

Although m ost facto r m odels have been proven to work to som e e x te n t for

m onthly stock returns, no ev id en ce exists for d a ily returns Nor has there been a direct com parison o f traditional fac to r m odels and th e m acro factor m odels A ssu m in g stationarity of factor lo adings and risk prem ium s, the m odels imply that or, is zero forpassive portfolios T able 2 presents the num ber o f alpha estimates w ith t-statistics significant at 10% and av erage adjusted R -squares for various regressions R esults for the

25 size and book-to-m arket portfolios and 64 siz e , book-to-m arket, and m o m entum portfolios are reported as in F am a and French (1 9 9 3 ) Since these are pure passive portfolios, the results help estab lish how the v ario u s m odels perform in sta n d a rd tests o f perform ance evaluation u sin g daily data

M easured by the n u m b er o f m ispriced te st portfolios, three-, five- and six-factor

m odels are the best am ong the traditional characteristic-based m odels B oth m acro factor

m odels m isprice far few er test portfolios than m ost traditional ch aracteristic-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 -fac to r m odel and the m im icking m acro factor m odel are used as b en ch m ark models because th ey explain a large portion o f tim e series variations o f stock retu rn s and price analyst p o rtfo lio s more correctly The key results from the other factor m odels are also presented

The finding that facto r m odels m ay m isprice a significant n um 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 th e alpha estim ates o f a nalyst portfolios

m uch as it biases passive p o rtfo lio s with specific characteristics For ex a m p le, if analysts possess no superior ability b u t rather cover and recom m end stocks acco rd in g 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 r tf o lio s w ith s ig n if ic a n t a lp h a a n d t h e a v e r a g e a d ju s te d R - s q u a r c s

a c r o s s a ll te s t 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 r e R:l is the e x c e s s re tu rn o n

e ith e r th e 25 test p o r tf o lio s fo r m 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 r k e t o r th e 6 4 test p o rtfo lio s f o r m e d

o n th e b a s is o f size, b o o k - to - m a r k e t, a n d m o m e n tu m T h e v a r ia b le s in Rf , i n c lu d e th e e x ce ss r e tu r n 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 r k e t in d e x fo r C A P M ( a ) T h e s iz e and b o o k - to - m a r k e t

f a c to r s a rc a d d e d to C A P M to fo rm m o d e l ( b ) , th e F a m a - F r e n e h m o d e l M o d e l (c ), the C a r h a r t ( 1 9 9 7 )

m o d e l, a d d s the re tu rn m o m e n tu m f a c to r A n a d d itio n a l e a r n in g s /p r ic e f a c to r is a d d e d in m o d el (d ) M o d e l (e ) in c lu d e s th e fa c to rs in m o d e l ( d ) p lu s a liq u id ity f a c to r T h e ra w m a c ro f a c t o r m o d e l in ( 0 in c lu d e s t h 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 r k e t 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 y ield c h a n g e s o n b o n d s r a 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 b etw een th e 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 a tu r ity a n d I y e a r to m a tu r ity ; a n d th e d i f f e r e n c e b e tw ee n th e y ie ld s o f

book-to-m arket, and o th er characteristics, a m isspecified factor m odel m ay incorrectly suggest superior perform ance.

I apply a p ro cedure recom m ended by D aniel et al (1997) and E ckbo et al (2000) They recom m end com b in in g m atching tech n iq u e [Barber and Lyon (1997)] and factor

m odel analysis A zero-investm ent p o rtfo lio is created by (1) investing in the analyst recom m ended b u y p ortfolios and shorting the m atching portfolios for buy

recom m endations, o r by (2) investing in the m atching portfolios and shorting the analyst recom m ended sell p ortfolios for sell recom m endations If m atching portfolios are created according to the ch aracteristics o f analyst portfolios, they should exhibit sim ilar tim e series properties, and thus m itigate the p ro b lem o f m isspecified factor m odels Yet, since

it is hard to obtain a perfect m atch and th e 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

A nalyst Portfolios a n d T heir M atching P o rtfo lio s

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 d atabase and for

w hich a buy or sell p ortfolio can be form ed for at least three m onths, I create both avalue-w eighted and an equal-w eighted p o rtfo lio using their recom m ended stocks in the

specific category S tocks enter the a n a ly st portfolios on the recom m endation date and are dropped at the revision date as reco rd ed by IBES The requirem ent o f at least ten

18 T h e sa m e a n a ly s is is c o n d u c t e d u sin g m o n th ly p o r t f o li o 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 r tf o lio s w ith in e ac h m o n th A r e q u ir e m e n t o f a l le a s t

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 r tf o lio s fo r th e c o n s id e ra tio n o f r e g re s s io n a n a ly s is , w h ic h m a y

r e s u lt in so m e s u r v iv o r s h ip b ia s T h e re s u 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 p o ssib ility that a non-financial analyst enters the

d atabase by co-authorship o r assum es the role o f a financial an a ly st tem porarily The short window o f 3 m onths ensures m in im u m survivorship bias R e su lts are largely the

sa m e when the requirem ent is at least six -m o n th or one-year p o rtfo lio returns and at least

P erform ance o f A n a lysts a s a G roup M e a su re d by Alpha

Table 3 presents, for both buys a n d sells, param eter e stim a tes o f the portfolios that place equal w eight on the portfolios o f individual financial a n a ly sts Three types o f

po rtfo lio s are exam ined: analyst portfolios, m atching portfolios, a n d zero-investm ent portfolios, both equally w eighted (EW ) and value-w eighted (V W ) portfolios Panel A presents the results using th e 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 a n d significantly different from zero for both buy and sell portfolios T he factor lo ad in g s on original analyst p o rtfo lio s provide inform ation about the ch aracteristics o f sto c k s covered by analysts and the difference o f characteristics betw een the stocks reco m m en d ed as buys and sells, respectively The

m odel produces significant factor loadings for all six factors for m o st portfolios The only

exception is that analyst sell p o rtfo lio s have an insignificant factor exposure to the earnings/price factor B oth a n a ly s ts ’ buy and sell portfolios have a m arket beta of one The o th er factor loadings are q u ite different in m agnitude for buy and sell portfolios, except for loadings on the size factor Buy portfolios have h ig h er loadings than sell portfolios on return m om entum , eam ings/price, and share tu rnover, 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 po rtfo lio s have positive loadings on m arket factor, book-to-

m arket, eam ings/price, and share turnover and negative loadings on size and return

A lthough Barber et al (2001) and W om ack (1996) use very d ifferen t sam ple, perform ance m easurem ents, risk adjustm ents, portfolio w eighting schem es, and portfolio creation schem es, they find ab norm al returns on analyst reco m m endations sim ilar to my finding For exam ple, W o m ack (1996) finds initial abnorm al returns o f 3.0% for buys and 4.7% for sells and B arber 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 o th er fac to r m odels A possible ex p lan atio n is the inclusion o f recom m endations rated as b oth “underperform ” and “sell” in th e analyst sell portfolios,

21

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

T a b le 3 r e p o rts th e r e s u lts o n th e p e r fo r m a n c e o f th e p o r tf o lio th at e q u a lly w e ig h ts th e p o r tf o 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 r e s e n t r e s u lts

T a b le 3, continued

P an el 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 ) b c s t { b c s) h srs > Adj Rsq.

VW-Buy 0 0 2 4 3.53 0.97 1 06.38 0 0 1 7 6 6 -0.06 -9 6 2 -0 0 2 -3.43 0.92 VW-Match 0 0 0 7 1.79 0.93 1 59.79 0.01 10.76 -0.05 - 1 1.49 - 0.01 -3.90 0.97 VW-Zero( Buy-M atch) 0 0 1 7 4.3 4 0.05 8.61 0 0 0 1.79 -0.01 - 4 0 6 0 0 0 -1.55 0.12 EW-Buy 0 0 2 6 3.43 0.93 9 0 4 5 0 0 2 10.61 -0.09 -1 2 3 0 -0.02 -3.14 0.90 EW-Match 0 0 0 6 1.03 0.85 117.89 0 0 2 14.33 -0.08 -1 5 2 8 -0 0 2 -4.46 0.94 EW-Zero(Buy-Malch) 0.021 6 2 6 0.07 15.35 0 0 0 -0.26 -0.01 -2 0 2 0 0 0 0.41 0.24 VW-Sell -0 0 2 2 -2 4 0 0.70 6 1 1 1 0.0 2 14.11 -0.09 -1 2 0 6 - 0.07 -14.06 0.79 VW-Match -0 0 0 8 -1 3 0 0.73 8 5 9 6 0 0 2 14.13 -0.09 -1 5 0 1 -0 0 6 -15.09 0.90 VW-Zcro(Match-Sell) 0 0 1 4 2.01 0.03 3.4 5 -0.01 -4.78 0.01 1.27 0.01 1.89 0.05 EW-Sell -0 0 2 2 -2.33 0.69 5 7 9 2 0 0 2 15.33 -0.10 -1 2 5 6 - 0.07 -13.59 0.78 EW-Match -0 0 0 8 -1.28 0.71 8 1 9 9 0 0 2 15.98 - 0.10 -1 6 1 7 - 0.07 -15.04 0.89 EW-Zcro(Match-Scl 1) 0 0 1 4 1.94 0.02 2 5 5 -0.01 -4.46 0.00 0.8 1 0.01 1.72 0.04

23

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

w hile previous research ers use only recom m endations rated as “sell” For exam ple, com bining the reco m m en d atio n s rated as 4 and 5, th e abnorm al return is 3.1% in Barber

ct al (2001), very clo se to the abnorm al return o f 3 3% for the same recom m endations in

my sam ple T h e B arb er 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, w hich y ields qualitatively sim ilar results to those from my sam ple

M atching seem s to be fairly effective The fac to r loadings on the original portfolios o f analy sts and on the respective m atching portfolios are o f sim ilar m agnitude, and the factor lo ad in g s on zero-investm ent portfolios have m agnitudes m uch closer to zero M atching alone, how ever, does not seem to e lim in a te the factor exposure of analyst portfolios com pletely F actor loadings are sig nificant for zero-investm ent portfolios, except book-to-m arket for buys, and m arket and sh a re turnover for sells T he adjusted R- squares for buys are about 0.4 B oth results indicate the ineffectiveness o f m atching technique alone for risk adjustm ents Interestingly, the original sell portfolios have insignificant p o sitiv e alphas, but the alphas o f the m atch in g portfolios are positive and significant T his interesting results, com bined w ith the superior perform ance o f zero- investm ent portfolios, yield evidence that the factor m odels alone m ay inflate the alpha estim ates o f original analyst sell portfolios

Panel B rep o rts results for a m im icking m acro factor m odel All the alphas for the buy and sell p o rtfo lio s are again significantly d ifferen t from zero O riginal analyst portfolios have sig n ific a n t 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

spread and short-run term spread T he e x p o su re s to market and short-run term yield are higher for buy recom m endations, and the facto r loadings on the o th e r factors are very sim ilar for buy and sell portfolios Z ero -in v estm en t portfolios all ex h ib it significant

su p e rio r perform ance A bout three o f four facto r loadings are sig n ifican tly different from zero for all the zero-investm ent portfolios A lp h a results are qu alitativ ely the same as for the six -facto r model Interestingly, the m ag n itu d es o f the abnorm al returns suggested by the alphas for the zero -investm ent buy and sell portfolios are alm o st the same, w hichever

m odel is used In contrast, there are m uch g rea ter differences betw een abnormal returns

on original portfolios o f analysts for the tw o m odels This reduction in the extent o f differen ces in abnorm al returns is evidence that focusing on zero-investm ent portfolios pro v id es sufficient risk control rather than adding noise

T he sensitivity analysis appears in T ab le 4, which presents alpha estim ations 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 o d els indicate su p erio r perform ance of financial analysts in both buy and sell p o rtfolios In addition, there is less variation in the

m agnitude o f alphas for zero-investm ent p o rtfo lio s than for the orig in al analyst

p o rtfolios, which is to be ex p ected because the m atching technique should eliminate sizable factor exposures in the original analyst portfolios and thereb y reduce the potential bias and noise that different factor m odels c o u ld introduce into the alpha estimates

1 9 S c h e m e s u s in g size o n ly a n d s iz e a n d b o o k - to - m a r k e t o n ly to m atch y ie ld h i g h e r a b n o rm a l retu rn s, w h ic h

s u g g e s t s th e im p o rta n c e o f m o re s u f f ic ie n t r is k - a d ju s tin g m a tc h in g s c 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 4 p r e s e n ts a lp h a s e s t im a te d w ith v a r io u s fa cto r m o d e ls T h e m odel is /?„ = a , + b l0R F , +bu RF j _, +£ a , w h e r e /?„ a re e x c e s s r e tu r n s o n th e a n a l y s t p o r tf o lio s , th e ir m a tc h in g

p o rtfo lio s b a s e d on s iz 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 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 r e lo n g th e a n a ly s t- r e c o m m e n d e d b u y p o r tf o lio s an d s h o rt th e m a tc h in g p o rtfo lio s fo r th e

b u y list, o r a r c s h o r t th e a n a ly s t- r e c o m m e n d e d se ll 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 se ll list T h e v a r ia b le s in RF , in c lu d e e x c e s s r e tu r n 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

c o e f f ic ie n ts a r e e s tim a te d u s in g o r d in a r y le a s t s q u a re s I o b ta in h e tc r o s c e d a s tic ity - c o n s is tc n t l-sta tistic 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 d a ta are d a ily f r o 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

Value-W eighted Portfolios Equal-W eighted Portfolios

B u y 1M a tc h Z e r o S ell M a tc h Z e ro B u y M a tch Z e r o S e ll M a tc h 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 0 2 0 -0.011 0.002 0.013 4.93 3 0 8 4 8 9 -1.36 0 3 4 1.90 5 47 2.71 6 6 0 -1.27 0.43 1.83 (b) Alpha estim ates using Carhart ( i 997) four-factor model

0.038 0 0 1 7 0.021 0.003 0 0 1 9 0.017 0 0 4 2 0.017 0 0 2 5 0 0 0 3 0.019 0 016 7.34 5.51 6 05 0.32 5 3 0 2.41 7 7 4 4.78 7 7 6 0 3 8 5.24 2.25 (c) Alpha estim ates using five-factor model

0.0 3 2 0 0 1 5 0 0 1 8 0.003 0 0 1 7 0.014 0 0 3 6 0.014 0 0 2 2 0 0 0 4 0.017 0.013 6.76 4 9 3 5 28 0.40 5.01 2.01 7.31 4.13 7 2 9 0.53 4.97 1.78 (d) Alpha estim ates using raw macro factor model

0.0 3 0 0 0 1 2 0 0 1 8 -0.001 0 0 1 2 0.013 0.0 3 4 0.013 0.021 -0 0 0 0 0.013 0.013 4.16 2.71 4 5 8 -0.06 1.62 1.82 3.93 1.95 6 2 3 -0 0 0 1.63 1.80

Event-period A bnorm al R e tu rn s o r P ost-event R etu rn Drift

The results so far su g g e st superior p erform ance of portfolios recom m ended by financial analysts M ight th e abnorm al perform ance result from the even t-p erio d abnorm al returns, the p o st-e v e n t return drift, o r a com bination o f both? T his question is

im portant because the p o st-e v e n t return drift th at B arber et al (2001), E lton et al (1986) and W omack (1996) o b se rv e is a puzzle not e x p la in ed by any type o f m arket efficiency hypothesis It is also o f in te rest because investors m ay not be able to o b tain inform ation about recom m endations e x a c tly on the public announcem ent dates

A crucial elem en t in identifying the so urce o f abnorm al p erform ance is when portfolio formation sh o u ld start T he analysis in 4.1 assum es that analyst portfolios are rebalanced at the closing p rice one day before th e dates appearing on research reports Traditional belief is that rese a rc h reports are d ated several days a fte r analy sts send their reports to data vendors su c h as First Call and m e d ia services such as D ow Jones News Service [Womack (1 9 96)] C h en g (2000) com p ares the two dates using d ata from First

C all, and finds that the d a te s on research reports are no later than the public announcem ent dates In fac t, 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 c a n be obtained by the public should be calcu lated assum ing the rebalance dates to be so m e w h at later than the recom m endation dates in the IBES

database, rather than earlier Furtherm ore, since brokerage firm s m ay distrib u te inform ation to their p refe rre d clients before the inform ation is co n v e y ed to the public, this practice should induce so m e very short-run post-event return drift a fte r the report dates Using report dates to rebalance portfolios m ay inflate the sig n ific a n ce 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 o u ld support a conclusion o f a real drift.

Panels A and B o f T ab le 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 day s corresponds on average to several trading d ay s 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

C arhart (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 sin g le alpha is significant in Panel

B Both the m agnitude and the significance o f alphas d ecrease sharply as the rebalance dates m ove farther aw ay from the recom m endation dates T h ese results support the conclusion that the profit is sm all if investors react after the public announcem ent dates,

w hich is usually several days later than the report dates

T o appreciate the e x act tim e length before abnorm al perform ance evaporates and

to assess w hether analysts have released inform ation g rad u ally and to the most preferred clients earlier and the ex ten t o f early inform ation release F igure 1 plots the abnorm al returns im plied by the six -fac to r model alphas for analyst p o rtfolios created some tim e during the 30 trading days around the report dates T h e abnorm al returns seem to be the highest w hen buy or sell portfolios are rebalanced four tra d in g days before report dates

A considerable drop in p o ten tial profit happens over the sev eral trading days before report dates, w hich su ggests that analysts m ay have given inform ation to the most preferred clients even b efore report dates at w hich they d issem in ate inform ation to o th e r