Upload
jonathan-clarke
View
212
Download
0
Embed Size (px)
Citation preview
ARTICLE IN PRESS
Journal of Financial Economics 84 (2007) 713–737
0304-405X/$
doi:10.1016/j
$We wou
Irwin P. Dal
Schwert, Jon
Virginia Tec
European F
contribution
academic pro�CorrespoE-mail ad
www.elsevier.com/locate/jfec
The impact of all-star analyst job changes on theircoverage choices and investment banking deal flow$
Jonathan Clarkea, Ajay Khoranaa,Ajay Patelb, P. Raghavendra Rauc,�
aGeorgia Institute of Technology, Atlanta, GA 30332, USAbWake Forest University, Winston-Salem, NC 27109, USA
cPurdue University, West Lafayette, IN 47907, USA
Received 12 April 2004; received in revised form 8 August 2005; accepted 14 December 2005
Available online 9 February 2007
Abstract
Using a sample of all-star analysts who switch investment banks, we examine (1) whether analyst
behavior is influenced by banking relationships and (2) whether analyst behavior affects investment
banking deal flow. Although the stock coverage decision depends on the relationship with the client
firms, we find no evidence that analysts change their optimism or recommendation levels when joining
a new firm. Investment banking deal flow is related to analyst reputation only for equity transactions.
For debt and M&A transactions, analyst reputation does not matter. There is no evidence that issuing
optimistic earnings forecasts or recommendations affects investment banking deal flow.
r 2007 Elsevier B.V. All rights reserved.
JEL classification: G24; G32
Keywords: All-star analyst; Analyst coverage; Market share; Investment banking relationships; Conflicts of
interests
- see front matter r 2007 Elsevier B.V. All rights reserved.
.jfineco.2005.12.010
ld like to thank an anonymous referee, Bob Bruner, Susan Chaplinsky, Mike Cliff, James Cotter,
ey, Dave Denis, Diane Denis, Paul Irvine, Laurie Krigman, John McConnell, Henri Servaes, Bill
athan Sokobin, Kent Womack, and seminar participants at Boston College, Ohio University,
h, the 2004 Utah Winter Finance Conference, the 2003 FMA Annual Meeting, and the 2003
MA Meeting for helpful comments and suggestions. We would also like to acknowledge the
of I/B/E/S International Inc. for providing earnings per share forecast data, as part of a broad
gram to encourage earnings expectations research.
nding author. Tel.: +1765 494 4488; fax: +1 765 494 9658.
dress: [email protected] (P.R. Rau).
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737714
1. Introduction
Is investment banking deal flow affected by analyst behavior? Anecdotal evidence fromthe popular press suggests that it is:
1Bow2See
David Komansky, former chief executive of Merrill Lynch, and Dennis Kozlowski
discussed ways to improve research coverage of Tyco and hiring an analyst the company
liked, according to an e-mail introduced at the ex-Tyco chief’s trial. After Merrill hired
the analyst, Phua Young, Tyco immediately responded by awarding the investment
bank work on a $2bn bond offering, according to an e-mail sent in 1999 to Mr.
Komansky by Samuel Chapin, Merrill’s vice-chairman. ‘To demonstrate the impact this
hire has on our relationship, Dennis Kozlowski called me on Phua’s first day of work to
award us the lead management of a $2.1bn bond offering,’ Mr. Chapin wrote in the e-
mail of August 31 1999y1
Moreover, is analyst behavior influenced by investment banking relationships betweenthe bank and the firms the analyst covers? The popular press suggests that analysts mightbe pressured to cover firms that they would not otherwise cover, as well as give favorablecoverage to firms that they would otherwise downgrade.2
In this paper, we analyze a sample of 216 cases in which an Institutional Investor All-America Research Team analyst (‘‘all-star’’ hereafter) moves from one investment bank toanother over the 1988 to 1999 period. We investigate two questions. First, we examinewhether the all-star’s behavior changes when he switches investment banks. An all-starwho moves from Goldman Sachs to Merrill Lynch, for example, might choose to continuecovering only those stocks that are likely to generate investment banking business forMerrill. In addition, the analyst might issue more favorable reports for Merrill clients thanwhen at Goldman. Hence, we study whether, in the period following a job change, all-starschoose to continue covering stocks and whether they become more optimistic about thestocks they cover, based on the relationship between the firms being covered and theinvestment bank employing the all-star. Second, we examine whether analyst reputationand coverage affect investment banking deal flow after the all-star joins the new bank.We investigate all-star job changes, instead of job changes across all analysts, because
prior research by Krigman, Shaw, and Womack (2001) and Dunbar (2000) documents thatfirms value all-star research coverage. Specifically, Krigman, Shaw, and Womack find thatthe perceived quality of coverage, as proxied by all-star coverage, is an important driver ina firm’s decision to change the lead underwriter in a follow-on offering. Dunbar (2000)finds a strong positive relation between changes in an investment bank’s Institutional
Investor All-America Research Team ranking and subsequent changes in the bank’smarket share in the initial public offering market. If we find no effect on investment bankmarket share when an all-star analyst moves, it is unlikely we will find an effect for non-allstars. We examine both capital-raising (debt and equity underwriting) and corporatecontrol (M&A) transactions to develop a comprehensive understanding of the relationsamong stock coverage, analyst reputation, investment bank reputation, and deal flow.With respect to our first research question, our results show that an all-star analyst’s
decision to cover a firm is influenced by the investment bank’s relationship with the firm. In
e and Silverman (2004).
, for example, Schroeder and Smith (2002).
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 715
particular, the all-star is significantly more likely to continue covering a stock that is beingcovered at the new bank when that bank also has a prior investment banking relationship(underwriting or M&A advisory) with the firm. We find no evidence, however, thatanalysts change their optimism levels and recommendation ratings for the firms they coverat the new bank. At the median level, all-stars do not become more optimistic after a jobchange, and the difference in their earnings forecasts, before and after the job change, isnot related to the existence of an investment banking relationship with the client firms. Inaddition, recommendation levels, both before and after the analyst changes jobs, do notsuggest that analysts issue significantly more positive recommendations after changingjobs. In a separate sample of non-all-star job changes, we obtain similar findings.
Turning to our second research question, we find that the bank hiring the all-starsignificantly increases its market share in the industry covered by the analyst, relative to thebank losing the all-star. We separately examine the determinants of relative market sharefor bond and equity underwriting and corporate control transactions in a multivariateframework wherein we control for investment bank reputation. Our results show thatproxies for all-star reputation, such as the timeliness and frequency of the all-star’searnings forecasts, have a significantly positive impact on the relative market share of thetwo banks for equity underwriting transactions, but not for debt or M&A transactions. Wefind no evidence that optimism in earnings forecasts (deviation of the analyst’s earningsforecast above consensus) affects relative market share for either capital-raising orcorporate control transactions. Finally, the new business is not generated by clients of theanalyst at the original bank who follow the all-star to the new bank; rather, it comes fromnew firms that the all-star is significantly more likely to cover at the new investment bank.
Our paper contributes to the existing academic literature on analyst behavior in twoways. First, while the extant literature reports some evidence that analysts affiliated withbanks and other financial institutions tend to make more optimistic forecasts andrecommendations than unaffiliated analysts,3 there is no direct evidence that this differencein behavior is due specifically to relationships between the investment bank and the firmsthe analyst chooses to cover. Our analysis of changes in analyst behavior surrounding theirjob changes enables us to examine whether investment bank pressure influences analystrecommendations and forecasts. We find that it does not.
Second, there is no direct evidence in the literature on whether analysts are able toincrease deal flow (underwriting and M&A transactions) for their respective banks. Weshow that analysts are instrumental in winning deal flow for equity underwriting, but notfor debt or M&A transactions. Our results are inconsistent with recent allegations in thepopular press that analysts have helped generate investment banking deal flow by issuingoverly positive recommendations. To the extent that such allegations are true, our resultssuggest that they cannot be generalized across all analysts or types of transactions.
The remainder of the paper is organized as follows. Section 2 discusses the data anddescribes the sample. Section 3 examines both the analyst stock coverage decision andchanges in analyst behavior in the period surrounding job changes by all-stars. Section 4examines the relation between analyst coverage and investment banking deal flow. Finally,Section 5 concludes.
3See Dugar and Nathan (1995), Lin and McNichols (1998), Bradley, Jordan, and Ritter (2003), and Irvine,
Nathan, and Simko (2004). In addition, Michealy and Womack (1999) find that stocks that underwriter analysts
recommend earn lower returns than ‘‘buy’’ recommendations by unaffiliated analysts.
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737716
2. Data, variable construction, and sample description
2.1. Data
We examine a sample of Institutional Investor All-America Research Team analysts whochange investment banks between 1988 and 1999. Following Hong, Kubik, and Solomon(2000), we use the I/B/E/S detail file to determine which analysts change jobs. The detailfile assigns each individual analyst a numerical code, making it possible to track earningsforecasts across time even if the analyst switches investment banks. The I/B/E/S databaseidentifies each analyst and his or her employer by a unique numerical code. We use theBroker Code Key to identify the last name and first initial of each analyst in the databaseand the employer’s identity. This additional information allows us to identify thoseanalysts who were named to Institutional Investor’s All-America Research Team in a givenyear.4 We consider only those cases in which the analyst was an all-star in the year of or theyear prior to the job change. Since we wish to examine analyst behavior (e.g., analystforecasts and recommendations) in the period immediately around the job change, duringwhich time it is not likely that the analyst obtained new information to change his or herforecast, we eliminate cases in which the elapsed time between forecasts is greater than 100trading days.5 We further eliminate cases in which the switch was due to the merging oftwo investment banks. For example, we eliminate four cases where an all-star switchedfrom Kidder Peabody to Paine Webber in 1994. Finally, we eliminate six cases in whichInstitutional Investor named an analyst as a star in the ‘‘multi-industry,’’ ‘‘small growthcompanies,’’ or ‘‘government sponsored enterprises’’ categories. The final sample consistsof 216 cases of analyst job changes.Although many rankings of individual analysts are published each year, Institutional
Investor’s All-America Research Team is appropriate for our analysis because, as Hong,Kubik, and Solomon (2000) note, that sell-side analysts generally aspire to be Institutional
Investor All-Americans. Moreover, Leone and Wu (2002) document that all-star analystsrealize better earnings forecast accuracy, better stock recommendation returns, and smalleroptimism bias than do their non-star counterparts.Our analysis consists of three steps. First, we classify analysts into industries in which they
are rated all-stars. Second, we compare analyst behavior before and after the job change.Third, we examine whether investment bank market share changes after the analyst moves.The following three subsections describe the variable construction for each of these steps.
2.1.1. Analyst industry classification
We assign analysts to an industry based on the firms they follow. Firms are assigned aStandard & Poor’s Global Industry Classification Standard (GICS) industry code,
4Leone and Wu (2002) discuss the selection procedure for the all-American team. To summarize, selection to
the All-American team is based on survey data. Institutional Investor sends a questionnaire to the directors of
research and chief investment officers of money management institutions, and also to other sell-side analysts.
These individuals rank each analyst based on the following six dimensions: accessibility and responsiveness,
earnings estimates, useful and timely calls, stock selection, industry knowledge, and written reports. Scores for
each analyst are calculated by taking the number of votes awarded by each survey respondent and weighting them
by the size of the respondent’s firm. The results are published each year in the October issue of the magazine.5The median length of time between the analyst’s first forecast with his new employer and last forecast with his
previous employer is 24 trading days.
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 717
supplied by COMPUSTAT. These industry codes classify companies into one of fourlevels: ten sectors, 23 industry groups, 59 industries, and 122 subindustries. We use GICScodes in preference to other classification schemes used in the literature, such asStandardized Industry Classification System (SIC) codes, North American IndustryClassification System (NAICS) codes, or the Fama-French (1997) industry groupings,since Bhojraj, Lee, and Oler (2003) show that GICS classifications better explain stockreturn co-movements as well as cross-sectional variations in valuation multiples, forecastedgrowth rates, and key financial ratios. Moreover, Boni and Womack (2006) show thatpartitions on the basis of GICS codes provide a good proxy for how analysts specialize byindustry. As a first pass, we assign each all-star analyst to one of the 59 GICS industriesbased on the industry in which the analyst issued the largest fraction of forecasts in the twoyears before the job change. We then manually check these GICS classifications againstindustry classifications assigned by Institutional Investor and make changes if necessary.6 Ifan analyst is an all-star in more than one industry, we also define a secondary GICSindustry for that analyst. In our sample, all-star analysts issue an average (median) of72.7% (81.8%) of their forecasts in the primary GICS industry in which they are all-stars.
There is substantial cross-sectional dispersion in our sample across industries, with the216 instances of job changes representing 44 unique industries. The industries with themost job changes are chemicals (17 cases), health care providers and services (12 cases),computers and peripherals (11 cases), and oil and gas (10 cases). All-stars are also likely tostay all-stars after they change jobs: 80% remain all-stars in the year following their jobchange and 70% are still classified as all-stars in the second year following their change injobs.
2.1.2. Measuring analyst behavior
Our measures of analyst behavior are based on both earnings forecasts, obtained fromthe I/B/E/S detail files, and analyst recommendations, obtained from the recommendationfiles. We measure analyst behavior along five dimensions: earnings forecast accuracy,optimism, timeliness, frequency of coverage revisions, and recommendation levels. Thesedimensions are meant to capture both analyst reputation and bias. As noted above, someof the measures used by Institutional Investor to rank an analyst include responsiveness,earnings estimates, and timeliness. Our measures of earnings forecast accuracy, frequencyof coverage revisions, and timeliness are proxies for analyst reputation. The other twomeasures, optimism and recommendation levels, capture aspects of analyst behavior thatare likely to proxy for bias.
We measure analyst behavior along each of these five dimensions using a scoringmethodology. Scores are used because measures of analyst accuracy and bias are firm andindustry dependent. For example, as Hong, Kubik, and Solomon (2000) note, simplycomparing the average forecast error of an individual analyst to the average forecast errorof the other analysts who issue earnings estimates that year is problematic, becauseearnings for some firms are more difficult to predict than others. Consequently, we follow
6As an example, Institutional Investor and GICS codes distinguish between the automobile industry and the
automobile components industry. In terms of the number of firms, the automobile components industry
(consisting of firms such as Midas, Cooper Tire and Rubber) is much larger than the automobile industry
(consisting of firms such as GM, Toyota, Nissan, Winnebago). One analyst in our sample issued the majority of
his forecasts in the automobile components GICS industry. However, Institutional Investor ranked the analyst as a
star in the automobile industry. We therefore re-classify this analyst as a star in the automobile industry.
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737718
Hong, Kubik, and Solomon in constructing annual performance scores based on ananalyst’s relative earnings forecast accuracy, forecast frequency, timeliness, and optimismbias.7
Forecast accuracy score: To examine accuracy across all stocks in the analyst’s portfolio,we construct scores by defining Fi,j,t as the most recent forecast of annual earnings pershare issued before the fiscal year-end by analyst i on firm j for year t. Our measure ofanalyst i’s accuracy for firm j in year t is the absolute difference between the earningsforecast and the realized earnings per share of the firm, Aj,t:
Earnings forecast errori;j;t ¼ jFi;j;t � Aj;tj. (1)
We sort the analysts who cover firm j in year t based on their forecast errors given by theabove equation. We then assign a rank based on this sorting, with the most accurateanalyst receiving a rank of one. In the case of ties, we assign each analyst the mean value ofthe ranks that they take up.8 Since the maximum rank an analyst can receive for a firmdepends on the number of analysts who cover the firm, we scale an analyst’s rank by thenumber of analysts who cover the firm. The formula for the forecast accuracy score isgiven by
Forecast accuracy scorei;j;t ¼ 100�Accuracy ranki;j;t � 1
number of analystsj;t � 1
" #� 100, (2)
where number of analystsj,t is the number of analysts who cover firm j in year t. Theaccuracy score ranges from zero for the lowest-ranked analyst covering a firm to a score of100 for the highest-ranked analyst.9
Optimism bias score: We define optimism bias as
Optimism biasi;j;t ¼ Fi;j;t � F�i;j;t (3)
where F�i;j;t ¼ 1=nP
m2 �if gFm;j;t; f�ig is the set of all analysts other than analyst i whoproduce an earnings per share estimate for stock j in quarter t, and n is the number ofanalysts in {–i}. Hence, F�i;j;t is a measure of the consensus forecast made by all otheranalysts except analyst i following stock j in quarter t. We replicate the rankingmethodology for constructing the forecast accuracy score to arrive at an optimism bias
score, which ranges from zero for the least biased analyst covering a firm to a score of 100for the most biased analyst covering the firm in a given year. Intuitively, the optimism biasmeasures how optimistic an analyst is relative to the other analysts covering the stock —the more optimistic the analyst, the higher his or her earnings forecast will be relative to theconsensus.
Frequency of coverage revision score: This score is calculated by ranking analysts basedon the number of times they revise their annual earnings estimates. Like the previousmeasures, the frequency of coverage revision score ranges from zero for the least frequent
7For robustness, we also compute simple measures of accuracy, frequency, timeliness, and bias using quarterly
estimates of earnings per share. For example, we measure earnings forecast accuracy as the difference between the
analyst’s prediction of earnings per share and the realized value, normalized by the stock price. Our results are
qualitatively unchanged using these alternative measures.8Alternative procedures for handling ties, such as employing the median or highest values of the assigned ranks,
produce similar results.9To compute the score, we impose the criterion that at least five analysts cover a security. This requirement
ensures that there will be a meaningful consensus with which to calculate scores.
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 719
forecaster to 100 for the most frequent forecaster. The use of this variable is motivated byKrigman, Shaw, and Womack (2001), who find that dissatisfaction with the frequency ofcoverage is a major reason for switching underwriters.
Timeliness score: To compute a timeliness score, which ranges from zero for the leasttimely forecaster to 100 for the analyst who issues the first earnings forecast on a particularstock in a given year, we rank analysts based on when they issue their first annual earningsforecast for a firm in a given year. We then replicate the ranking methodology describedabove. We include this score because Hong, Kubik, and Solomon (2000) argue that ananalyst who issues the first annual forecast is not likely to be herding with other analysts.Moreover, Clement and Tse (2005) note that analysts who exhibit herd behavior havelower ability, suggesting that timeliness is a useful proxy for reputation.
Recommendation levels: We gather data from I/B/E/S on analyst stock recommendationsfrom October 1993 (the first available date for I/B/E/S data) through the end of oursample, and measure the analyst’s recommendation for each stock relative to theconsensus. Since a strong buy (strong sell) is coded as one (five), a negative relativerecommendation indicates an optimistic recommendation by the analyst. The abnormalrecommendation is computed as the difference between the analyst’s recommendation andthe prevailing consensus, which is calculated as the average recommendation across allother analysts covering the security.
2.1.3. Measuring investment bank market share
We compile a comprehensive database of investment banking deals (capital-raising andcorporate control transactions) between 1986 and 2001 from Thompson Financial
Securities New Issues and Mergers and Acquisitions databases. From the new issuesdatabase, for every initial public offering, seasoned equity offering, and bond offering, weobtain the issuer name and cusip, the filing and issue dates, the identity of the investmentbank retained by the issuer, and the size of the deal. From the mergers and acquisitionsdatabase, we obtain information on the identity of the target and acquiror, theannouncement and effective dates of the transaction, and the size of the deal.
We use our database to calculate the industry market share for the bank the analyst isswitching from (original bank) and the bank the analyst is switching to (new bank).Industry market share is calculated as the gross proceeds raised by an investment bank in aparticular industry divided by total gross proceeds of all deals completed in that particularindustry. Market shares are calculated for the two years before and the two years after theanalyst switches jobs. Industry classifications are based on the 59 GICS industry codesfrom the COMPUSTAT database. For those analysts listed as all-stars in multipleindustries, we add up the gross proceeds across all industries to compute market share.
2.2. Sample description
Table 1 describes the sample. Panel A reports data on analyst turnover by year. Thenumber of analysts issuing earnings forecasts in the I/B/E/S database increases from 2,618in 1988 to 4,543 in 1999. Measured as a percentage of all analysts, the number of all-staranalysts decreases from 12.4% in 1988 to 7.6% in 1999. The decrease is especially sharpover the period 1993–1995 period, when the percentage drops from 15.7% to 8.3%, due toa sharp increase in the total number of analysts and a fairly static number of all-stars.
ARTICLE IN PRESS
Table 1
Sample descriptive statistics
This table presents descriptive statistics on analyst turnover and capital market activity over the sample period.
Panel A reports year-by-year statistics on job changes by analysts. The number of analysts is the number of
analysts submitting forecasts to the I/B/E/S database in a given year. The number of all-stars is the number of
analysts on the Institutional Investor All-America Research Team that issued forecasts in a given year. The number
of institutions is the number of investment banks that had analysts issuing forecasts in a given year. Analyst
turnover (all-star turnover) is the number of analysts (all-star analysts) who moved from one investment bank to
another in a given year. Panel B presents the number of deals between 1986 and 2001 in which an investment bank
served as an advisor. The panel reports both capital-raising transactions (initial public offerings of equity,
seasoned equity offerings, and bond offerings) and corporate control transactions (M&A deals). Deals are
compiled from Thompson Financial Securities New Issues and M&A databases.
Panel A: Analyst turnover by year
Year Number of
analysts
Number of
all-stars
Number of
institutions
Analyst
turnover
Percent
turnover
All-star
turnover
Percent all-
star turnover
1988 2,618 325 172 154 5.88 6 1.85
1989 2,841 368 183 249 8.76 23 6.25
1990 2,648 336 187 149 5.63 9 2.68
1991 2,440 331 191 151 6.19 8 2.42
1992 2,269 353 192 114 5.02 6 1.70
1993 2,479 389 221 166 6.70 16 4.11
1994 2,876 371 226 219 7.61 21 5.66
1995 3,141 262 231 248 7.90 21 8.02
1996 3,528 267 261 282 7.99 20 7.49
1997 3,997 272 308 349 8.73 27 9.93
1998 4,410 322 351 404 9.16 26 8.07
1999 4,543 344 329 367 8.08 33 9.59
Panel B: Capital market activity by year
Capital-raising transactions Corporate control transactions
Year IPOs SEOs Bond offerings Advising acquiror Advising target Total deals
1986 717 772 709 519 655 3,372
1987 544 504 521 529 655 2,753
1988 291 190 402 704 929 2,516
1989 252 313 376 642 923 2,506
1990 215 237 363 454 660 1,929
1991 397 567 845 315 488 2,612
1992 606 624 1,110 365 542 3,247
1993 820 903 1,447 511 750 4,431
1994 642 558 984 605 860 3,649
1995 575 686 1,295 814 1,142 4,512
1996 880 848 1,840 920 1,258 5,746
1997 637 824 2,410 1,085 1,504 6,460
1998 401 640 2,564 1,225 1,730 6,560
1999 573 550 2,011 1,161 1,754 6,049
2000 421 499 2,325 1,248 1,777 6,270
2001 154 627 2,339 900 1,363 5,383
Totals 8,125 9,342 21,541 11,997 16,990 67,995
J. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737720
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 721
The number of institutions employing analysts increases from 172 in 1988 to 329 in 1999.Turnover among all analysts increases from 5.9% (154) in 1988 to 8.1% (367) in 1999.Turnover among all-star analysts increases from 1.9% (6) to 9.6% (33) over the sameperiod. While turnover in general has increased among analysts, the increase is much moredramatic among all-star analysts.10
Panel B reports information on total deal activity when investment banks areinvolved. Of the 67,995 deals in our database, the breakdown is as follows: 8,125 initialpublic offerings, 9,342 seasoned equity offerings, 21,541 bond offerings, and 28,987instances in which either the target or acquiror retained the services of an investmentbank. There is a rapid increase in capital-raising transactions (equity and debt) aswell as corporate control transactions (mergers and acquisitions) in the early part of the1990s.
3. Analyst behavior surrounding the job change
3.1. Does the analyst’s portfolio of covered stocks change? If so, why?
We examine firms that the all-star chooses to continue covering, those he adds, andthose he drops after moving to the new bank. This allows us to investigate whether the all-star’s coverage decision depends on whether the bank has a relationship with the firmunder consideration. Panel A of Table 2 reports descriptive data on stock coverage anddeal flow in the two years before and after the all-star changes jobs. Panel B reports moredetailed descriptive statistics on firm, analyst, and bank characteristics for the samples offirms for which the all-star retains, drops, and add coverage.
The all-star’s workload remains the same after switching investment banks (Panel A).He covers 16 firms (at the median) prior to and 15 firms following the job change11; aWilcoxon rank sum test indicates no difference in the number of stocks covered before andafter the job change. He retains coverage of approximately 65% of the old portfolio at thenew investment bank. Replacing the 35% that he drops, approximately 35.5% of thestocks covered by the all-star at the new investment bank are new firms he did not coverpreviously.
A more relevant question for the purpose of our study is whether all-star stock coveragerelates positively to investment banking deal flow. Panel A shows that at the median, 99unique firms complete deals in the star’s industry in the two years before the job change, incontrast to 106 firms in the two years after the job change. Focusing only on the firms theanalyst covers, around half (seven to eight) complete deals in the two years before and afterthe job change. However, the difference in deal flow between the original and new banks isstriking. At the mean level, a significantly smaller number of firms complete deals with theoriginal bank after the all-star’s departure. Of the firms covered by the analyst after his
10These results contrast with Groysberg and Nanda (2001), who find that in the aggregate, star analysts have
lower turnover than non-stars. Groysberg and Nanda attribute the lower turnover of star analysts not to their
stardom, but to demographic characteristics; stars tend to be older, more experienced, and have greater tenure
(i.e., they move less) than non-stars. We find that while in the earlier part of our sample period, all-stars do
observe a lower turnover rate than non-all-stars, in the latter part, turnover increases dramatically for all-stars and
increases at a slower rate for non-all-star analysts.11This contrasts with Boni and Womack (2005), who find that the average analyst in their sample covers ten
companies. All-stars cover more companies on average.
ARTIC
LEIN
PRES
S
Table 2
All-star analyst coverage before and after a change in jobs
This table examines all-star analyst coverage over the two years prior to (i.e., pre-turnover) departing the original investment bank and the two years after (i.e.,
post-turnover) arriving at the new investment bank. Panel A reports data on overall stock coverage and investment bank deal flow. The number of stocks covered is the
number of stocks covered by the median analyst switching investment banks in the pre- and post-turnover period, respectively. The proportion of stocks retained after
turnover is the fraction of stocks that the analyst continues to cover after arriving at the new investment bank. Proportion of stocks dropped is the fraction of stocks
dropped after arriving at the new investment bank. Proportion of new stocks added is the fraction of new stocks that the analyst begins to cover at the new bank. The
number of unique firms completing deals is computed as the median number of firms that completed deals in the same industry as the all-star in the two years prior to
and after the analyst’s move. Panel B reports data on the median characteristics of stocks retained, added, and dropped. The forecast accuracy score and frequency of
coverage revision score are computed using a scoring methodology as in Hong, Kubik, and Solomon (2000). For both panels, mean values are reported in parentheses.
The p-value for the difference is based on a two-sided Wilcoxon rank-sum test.
Panel A: Data on stock coverage and deal flow pre- and post-turnover
Pre-turnover Post-turnover p-value for difference
Stock coverage
Number of stocks covered 16.00 15.00 0.24
(16.85) (16.76)
Proportion of stocks retained 64.71%
(61.48%)
Proportion of stocks dropped 35.29%
(38.52%)
Proportion of new stocks added 35.50%
(36.39%)
Stock coverage and investment banking deal flow
Unique number of firms completing deals in all-star’s industry 99.00 106.00 0.00
(124.92) (136.05)
Unique number of firms covered by analyst completing deals with any investment bank 7.00 8.00 0.00
(7.97) (9.40)
Unique number of firms covered by analyst completing deals with the original bank 0.00 0.00 0.02
(1.24) (0.98)
Unique number of firms covered by analyst completing deals with the new bank 0.00 1.00 0.00
(1.15) (1.84)
J.
Cla
rke
eta
l./
Jo
urn
al
of
Fin
an
cial
Eco
no
mics
84
(2
00
7)
71
3–
73
7722
ARTIC
LEIN
PRES
SPanel B: Descriptive statistics on stocks retained, added, and dropped
Universe of
stocks covered
before moving
Stocks
retained
Stocks
added
Stocks
dropped
p-value for
difference
(retained vs.
added)
p-value for
difference
(retained vs.
dropped)
p-value for
difference
(added vs.
dropped)
Firm-specific variables
Proportion of firms
with market cap in top 25% of industry 78.57 90.00 50.00 66.67 0.00 0.00 0.01
(76.22) (84.24) (48.41) (59.65)
with trading volume in top 25% of industry 75.00 83.97 12.50 66.67 0.00 0.00 0.00
(74.37) (80.54) (19.30) (59.56)
that completed X2 deals in the 2 years prior to all-star’s move 30.00 33.33 20.00 22.22 0.00 0.00 0.25
(33.23) (33.83) (24.44) (27.90)
that completed X2 deals in the 2 years after the all-star’s move 30.56 34.31 33.33 18.75 0.77 0.00 0.00
(32.45) (34.91) (34.48) (25.96)
with market-to-book ratios above industry average 20.00 12.50 7.69 0.00 0.00 0.01
(24.94) (28.16) (19.30) (21.86)
Analyst-specific variables
Forecast accuracy score 59.11 61.76 55.48 0.03
(58.65) (60.93) (55.12)
Frequency of coverage revision score 54.95 56.90 52.30 0.25
(54.11) (55.65) (51.64)
Relationship-specific variables
Proportion of firms that
complete X1 deals with the new investment bank in the 0.00 0.00 0.00 0.00 0.32 0.03 0.00
2 years prior to the all-star’s arrival (6.08) (7.44) (8.85) (4.54)
are already covered at the new bank by any analyst 46.67 57.74 14.29 33.33 0.00 0.00 0.01
(46.95) (51.23) (23.28) (34.50)
are already covered at the new bank by an all-star 0.00 0.00 0.00 0.00 0.50 0.34 0.72
(16.47) (16.55) (12.95) (13.42)
J.
Cla
rke
eta
l./
Jo
urn
al
of
Fin
an
cial
Eco
no
mics
84
(2
00
7)
71
3–
73
7723
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737724
move to the new bank, a significantly larger number of firms complete deals with the newbank as their advisor than they did before the move.Panel B of Table 2 provides descriptive statistics on changes to the analyst’s port-
folio following the move to the new investment bank. Analysts may be more likely tocover firms that generate investment banking or trading revenue for the new bank. Thesefirms are likely to be larger, have more trading volume, and complete more dealsin the star’s industry before or after the all-star’s move. We therefore compute theproportion of firms covered by the all-star with market capitalization or trading volumein the top 25% in their industry. We also compute the proportion of firms that completeat least two deals in the two years before (following) the analyst’s job change (representingthe 75th percentile of the number of deals done by a firm during the respective two-year window). The results in Panel B indicate that all-stars tend to retain coverage oflarger firms, firms with greater trading volume, and firms that complete a larger numberof deals in the two years before and after the all-star’s move.12 Interestingly, firmsadded to the all-star’s portfolio are significantly smaller and have lower tradingvolumes than stocks dropped from the portfolio. However, a significantly higherproportion of these firms complete investment banking deals over the two years followingthe job change. In other words, analysts add coverage of firms more likely to generateinvestment banking deal flow. They drop coverage of smaller firms, firms with lowertrading volumes, and firms less likely to generate investment banking deal flow in thefuture.Since Jegadeesh, Kim, Krische, and Lee (2004) show that sell-side analysts generally
tend to recommend high growth, high volume, and relatively expensive glamour firms, wealso compute the proportion of glamour firms, defined as firms with market-to-book ratiosabove industry average, in the all-star’s portfolio. Panel B shows that stocks retained oradded by the all-star have higher market-to-book ratios (relative to their industry) thanthose dropped by the all-star, suggesting that, in general, all-stars not only prefer to retaincoverage of glamour stocks, they prefer to cover glamour firms within particularindustries.In terms of analyst-level characteristics, analysts should be more likely to drop coverage
of firms for which they are less accurate and for which they produce less frequent reports.Panel B compares the forecast accuracy and the frequency of coverage revision scores forthe firms the all-star retains to those for the firms he drops. The results show that while all-stars are significantly more likely to drop firms for which they are less accurate, thefrequency of coverage revision is not significantly lower for firms the analyst decides todrop versus those he decides to retain.Finally, we compute the proportion of firms that have a prior relationship with
the new investment bank (firms that complete at least one deal at the new investmentbank in the two years prior to the all-star’s arrival), and the proportion of firmsthat are already covered at the new bank by any analyst or by an all-star. Panel B reportsthat a significantly greater proportion (based on means) of firms the all-star retainsor adds (as opposed to those he drops) had a prior relation with the new bank priorto the all-star’s arrival. Firms already covered by the new bank are significantly morelikely to be retained by the analyst. A significantly smaller proportion of the firms
12The differences (not reported in the table for brevity) are statistically significant.
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 725
added by the analyst are covered at the new investment bank prior to his arrival. Theseresults reinforce our earlier findings that the coverage decision appears to be related to thefirm’s likelihood of generating revenue for the investment bank.
Next, we estimate a multivariate logistic regression to explain the likelihood of ananalyst retaining or adding (versus dropping) coverage of a stock subsequent to movingto the new bank. More specifically, we regress an indicator variable for the decision toretain coverage of a stock (as opposed to adding or dropping it) against firm-specificindicator variables, analyst/firm-specific variables, and firm/bank relationship-specificindicator variables. The firm-specific indicator variables are dummy variables that takea value of one if the firm has a market capitalization or trading volume in the top 25%of its industry, or if it completes two or more deals in the two years prior to or after theall-star’s move, and zero otherwise. The analyst/firm-specific indicator variables are theall-star’s frequency of coverage revision score, the forecast accuracy score, and indicatorvariables that take a value of one if the all-star has a higher frequency of coverage score ora higher forecast accuracy score than the analyst at the new bank covering the firm prior tothe all-star’s move. Finally, the firm/bank-relationship variables are indicator variablesthat take a value of one if the firm is already covered at the new bank by any analyst, ifthe firm is already covered at the new bank by an all-star analyst, or if the firm com-pletes at least one deal with the new investment bank in the two years prior to the all--star’s arrival. We also include interaction terms to examine whether the impact of priorcoverage on the analyst’s decision to retain the stock is enhanced by a prior invest-ment banking relationship with the firm under consideration. These results are reported inTable 3.
Model 1 compares the decision to retain a firm against the decision to add a firm to theall-star’s portfolio, while Model 2 compares the decision to retain a stock against thedecision to drop it. Model 3 examines the stock retention decision conditional on theexistence of coverage at the new bank before the all-star’s arrival and Model 4 comparesthe decision to add coverage of a stock against the decision to drop it.
Consistent with our results in Table 2, all-star analysts are more likely to add coverageof smaller glamour firms with lower trading volume that have a higher potential for futuredeal flow (Model 1). They are more likely to drop coverage of smaller firms that do notcontribute to future deal flow and are more likely to retain coverage of glamour stockswith high market-to-book ratios (Model 2). All-stars are also more likely to retaincoverage of stocks in which they are active, providing accurate earnings forecasts withfrequent revisions over the two years preceding the job change (Model 2). They are morelikely to retain coverage if the stock was covered previously by an analyst at the new bank(Model 2) and more likely to add coverage of stocks that were not previously covered(Model 4). For firms with coverage at the new bank prior to the analyst’s arrival (Model 3),the all-star is more likely to retain coverage of larger firms and those that have a priorrelationship with the new investment bank. The analyst is more likely to drop coverageif the firm has a prior investment banking relationship with the new bank, and isalready being covered by another all-star analyst at the new bank (the interaction term inModel 3).
Overall, our findings suggest that an all-star analyst is more likely to retain or addcoverage if (1) the stock is a large glamour stock in its industry, (2) the analyst has in thepast issued frequent and accurate earnings forecasts on the stock, and (3) the firm has aprior investment banking relationship with the new bank.
ARTIC
LEIN
PRES
S
Table 3
Determinants of stock retention and addition
This table reports the results of logistic regressions of the all-star analyst’s decision to retain, add, or drop a stock on firm-specific indicator variables, analyst/firm-
specific variables, and firm/bank relationship-specific indicator variables. The forecast accuracy score and frequency of coverage revision score are computed using a
scoring methodology as in Hong, Kubik, and Solomon (2000). p-values are reported in parentheses.
Retained vs. added Retained vs. dropped Added vs. dropped
All firms (1) All firms (2) Firms with prior coverage (3) All firms (4)
Intercept �1.43 �0.84 �0.09 1.016
(0.00) (0.00) (0.73) (0.00)
Firm-specific indicator variables
Market capitalization in top 25% of industry 1.10 0.84 1.06 �0.41
(0.00) (0.00) (0.00) (0.00)
Trading volume in top 25% of industry 2.22 0.18 �0.04 �1.80
(0.00) (0.13) (0.89) (0.00)
Firm completed X2 deals with any investment bank in the 2 0.27 0.01 �0.05 �0.28
years prior to all-star’s move (0.02) (0.94) (0.79) (0.03)
Firm completed X2 deals with any investment bank in the 2 �0.25 0.29 0.13 0.59
years after all-star’s move (0.02) (0.01) (0.43) (0.00)
Market-to-book ratio above industry average �1.47 0.25 0.19 1.58
(0.00) (0.02) (0.28) (0.00)
Analyst/Firm-specific variables
Forecast accuracy score 0.003
(0.07)
Frequency of coverage revision score 0.01
(0.00)
All-star has higher forecast accuracy score than analyst 0.05
at new bank prior to move (indicator variable) (0.75)
J.
Cla
rke
eta
l./
Jo
urn
al
of
Fin
an
cial
Eco
no
mics
84
(2
00
7)
71
3–
73
7726
ARTIC
LEIN
PRES
SAll-star has higher frequency of coverage revision score than 0.17
analyst at new bank prior to move (indicator variable) (0.26)
Firm/Bank relationship-specific indicator variables
Firm is already covered at the new bank by any analyst (prior 0.66 0.16 �0.46
coverage) (0.00) (0.09) (0.00)
Firm is already covered at the new bank by an all-star analyst �0.24
(0.15)
Firm completed X1 deals with the new investment bank in the �1.08 �0.71 0.82 0.85
2 years prior to the all-star’s arrival (prior relationship) (0.01) (0.16) (0.03) (0.05)
Prior coverage at new bank � Prior relationship with new bank 0.45 1.08 �0.12
(0.36) (0.05) (0.81)
Prior coverage by an all-star at new bank � Prior relationship �0.94
with new bank (0.07)
Number of observations 3,124 2,517 941 2,366
Percent concordant 84.50 66.30 59.7 76.7
J.
Cla
rke
eta
l./
Jo
urn
al
of
Fin
an
cial
Eco
no
mics
84
(2
00
7)
71
3–
73
7727
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737728
3.2. Do all-star analysts change their behavior after they move?
An analyst who is pressured by an investment bank to provide favorable coverage of aclient firm might change the optimism of his or her earnings forecasts and/or stockrecommendation rating after moving to a new bank that has a different set of clientrelationships. In this section, we investigate changes in analyst optimism bias,recommendation ratings, and earnings forecast accuracy for different types of firmsclassified on the basis of whether they have a prior investment banking relationship withthe new and/or original investment banks.We compute scores for optimism bias, forecast accuracy, timeliness, and frequency of
coverage revision for the all-stars relative to the universe of equity analysts covering thefirms. These scores are computed using annual data for firms the all-star retains, for theyear before and year after the all-star changes jobs. We classify these firms into separatecategories on the basis of whether, in the two years before the job change, the firms haveinvestment banking relationships with either the new or the original bank, with neitherbank, or with both banks. We also compute changes in analyst stock recommendationssurrounding the analyst’s job change by comparing the analyst’s last recommendation on afirm relative to consensus before leaving the original bank with his first recommendationfor the same firm after arriving at the new bank.When the firm has an investment banking relationship with either bank, our results (not
reported for brevity) indicate that there is no significant change in the analyst’s optimismbias scores or abnormal recommendation levels in the period surrounding the job change.There is also no change in earnings forecast accuracy, timeliness, or frequency of coveragerevision across any of the categories.13 The lack of any increase in analyst optimism, ineither earnings forecasts or stock recommendations, is inconsistent with the view in thepopular press that analysts may exhibit extreme optimism in an attempt to win investmentbanking deal flow.While analysts maintain their opinions on stocks that they previously followed, it is
plausible that they are more optimistic about stocks they are covering for the first time andthat have investment banking potential. This may be where we are most likely to find‘‘cheating’’ behavior. We therefore examine newly covered stocks, and separate them intotwo categories based on whether two or more investment banking deals occur in the twoyears following the analyst’s arrival at the new bank. Again, we find no evidence to suggestthat analysts are significantly more optimistic for stocks that have high future deal flow.One explanation for our results is that the reputational concerns of all-star analysts
make them less likely to succumb to pressure from their investment bank to alter theirearnings forecasts and recommendations to increase deal flow. That is, it may be the non-star analysts who are more likely to issue optimistic recommendations or earningsforecasts in an effort to win investment banking deal flow. To test this possibility, wecompile a sample of 1,056 non-star analysts who switch investment banks between 1988and 1999 but continue to cover stocks in the same GICS industry and repeat the aboveanalysis.
13We also obtain similar results in multivariate regressions of the change in analyst bias and reputation scores
against dummy variables proxying for prior relationships with the original and new banks. We find no evidence to
suggest that changes in analyst earnings forecasts and recommendations following the switch in jobs are related to
a prior relationship between the firm and the investment bank.
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 729
Similar to the case of all-star analysts, we find no evidence that non-stars issue moreoptimistic earnings forecasts or recommendations, change timeliness, or experience anysignificant change in their earnings forecast errors after they move to the new bank. Theseresults suggest that analysts without strong reputational concerns do not change theirbehavior either, in an attempt to win investment banking deal flow.
One explanation for the inconsistency between our findings and the conflict-of-interestarguments alleged by regulators and the press is that the incentives of both the new andoriginal investment banks are similar. Therefore, we should not see a change in behavioron average. Since our data do not include any examples of analysts moving frominvestment banks to pure research houses that do not underwrite deals, our analysiscannot fully reveal the conflicts of interest that might exist. The analysis does suggest,however, that the egregious cases of analyst bias alleged in the popular press cannot begeneralized across all analysts.
4. Analyst job changes and investment banking deal flow
4.1. Do the new banks experience increased deal flow following all-star analyst turnover?
Table 4 examines the changes in industry market share across the original and newbanks in the two years before and after the all-star switches jobs.14 The deals are classifiedinto two sub-categories: capital-raising transactions (initial and seasoned equity under-writing and bond underwriting) and corporate control transactions (M&A). To get abroad sense of how deal flow changes surrounding analyst job changes, we do notcondition on stocks being retained or dropped by the analyst or on client relationships atthe original and new investment banks. Industry market share is calculated as the grossproceeds raised in an industry by the analyst’s investment bank, divided by the total grossproceeds of all deals completed in that industry.
Following the analyst’s arrival at the new bank, the difference in market share betweenthe two banks widens significantly for both capital-raising and corporate controltransactions. Across all capital-raising transactions, for example, before the analystmoves to the new investment bank, the market share for the median bank in the sample ofnew investment banks is 2.09% as opposed to a market share of 0.86% for the medianbank in the sample of original banks. The median (mean) difference in relative marketshare is 0.82% (1.35%), significant at the 5% level. After the analyst arrives at the newbank, the median market share at the sample of original investment banks decreases to0.57%, while it increases to 2.35% at the new investment banks. The median (mean)difference in market share is 2.28% (2.27%), significant at the 1% level. Similar increasescan be seen for corporate control transactions. Note, however, that the increase in relativemarket share for capital-raising transactions occurs only for equity underwriting deals. Forbond deals, the zero median market share both before and after an analyst job change forboth the original and new bank is driven by a high concentration of deals done at a small
14We focus on all-star analysts instead of all analysts, since non-all-star analysts do not generate significant
investment banking deal flow. All-star analysts, accounting for only 10% of all sell-side analysts, are involved in
63% of target advisory deals, 64% of SEO deals, 57% of acquiror advisory deals, 76% of bond deals, and 48% of
IPO deals. Non-all-star analysts who switch investment banks are also involved in fewer deals relative to all-stars
who switch investment banks.
ARTICLE IN PRESS
Table 4
Relative industry market shares before and after the all-star analyst job change
This table presents the median (mean) market share for the bank the analyst switches to (the new bank) and the
bank the analyst switches from (the original bank) for deals reported in the all-star’s industry. Industry market
share is calculated as the gross proceeds raised by an investment bank in the all-star’s industry divided by total
gross proceeds for all deals completed in that particular industry. Both pre-turnover and post-turnover market
shares are calculated using two years of data. Industry classifications are based on the 59 GICS industry codes
from the COMPUSTAT database. Capital-raising transactions include seasoned equity offerings (SEOs), initial
public offerings of equity (IPOs), and bond offerings. The p-value for the difference in relative industry market
share is based on a two-sided Wilcoxon rank-sum test.
Original bank (%) New bank (%) p-value for difference
Panel A: Capital-raising transactions
Pre-turnover 0.86 2.09 0.04
(4.42) (5.77)
Post-turnover 0.57 2.35 0.00
(4.11) (5.98)
SEO and IPO deals
Pre-turnover 0.55 0.61 0.30
(3.88) (4.30)
Post-turnover 0.00 1.51 0.00
(4.25) (5.44)
Bond offerings
Pre-turnover 0.00 0.00 0.03
(4.43) (7.06)
Post-turnover 0.00 0.00 0.07
(4.55) (6.47)
Panel B: Corporate control transactions (M&A Deals)
Pre-turnover 0.26 0.77 0.03
(3.18) (4.61)
Post-turnover 0.40 1.75 0.00
(3.92) (4.75)
J. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737730
subset of investment banks. During our sample period, 38% of debt underwritingtransactions are handled by Merrill Lynch, Goldman Sachs, and Lehman Brothers.Our results so far suggest that relative deal flow at the new investment bank increases
following the arrival of the all-star analyst. However, this analysis leaves several questionsunanswered. Is the increase due to the all-star? If so, what characteristics of the all-star areimportant in determining deal flow? Alternatively, is it the case that both deal flow and theall-star are drawn to the new investment bank because of the reputation of the investmentbank? In other words, one cannot examine the impact of the all-star analyst on investmentbanking deal flow without controlling for the reputation of the investment bank.
4.2. Is the increased deal flow following the all-star analyst job change due to the analyst?
We use a multivariate framework to examine whether analyst reputation factors and/oranalyst bias measures affect deal flow after controlling for the investment bank’s
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 731
reputation. Our measures for analyst reputation include indicator variables thattake the value of one if the all-star has been an all-star in each of the three yearsprior to turnover (repeat all-star15), or if the analyst issues earnings forecasts in the75th percentile of all analysts in the sample when ranking the analysts separatelyon accuracy, timeliness, and frequency of revision of forecasts.16 We also includeanalyst forecast optimism as a measure of analyst bias, computed similar to the measuresdiscussed above.
We use two measures of bank reputation. First, we use the relative market shareof the two investment banks in the two years before the analyst’s job change. Therelative market share is defined as the difference in market share between the twobanks in the all-star’s industry. Banks with higher relative market share are likelyto be more reputable since they advise more underwriting and M&A transactions.This measure is used by Megginson and Weiss (1991) and more recently by Ljungqvist,Marston, and Wilhelm (2006). We also include a dummy variable to control forthe trend in relative market shares. This dummy takes the value of one if the relativemarket share difference widens from two years to one year before the job change and zerootherwise.17
Our second proxy for bank reputation is the difference in the total number of all-staranalysts at the new and original banks in the year before the analyst changes jobs. Dunbar(2000) finds a strong relation between the change in an investment bank’s Institutional
Investor All-America Research Team ranking and subsequent changes in their share ofinitial public offerings. Increases in the reputation of an investment bank’s analysts have apositive effect on market share changes.
The dependent variables in the regressions are the relative market shares for the twobanks computed separately for the three types of deals (equity issues, bond issues, andM&A transactions) over the two years after the analyst’s job change.
Our results are reported in Table 5. As expected, the bank’s reputation is important. Therelative market share following the all-star’s arrival at the new bank is significantlypositively related to relative market share before the move for bond and M&A deals. Inaddition, for M&A deals, we find that the relative market share following the move isweakly negatively related to the trend in relative market share over the two years before theanalyst moves, possibly due to mean reversion. For all three types of deals, the differencein the number of all-star analysts between the new bank and the original bank issignificantly positively related to relative market share following an analyst job change.Our results indicate that more reputable investment banks gain a larger market sharefollowing the arrival of an all-star analyst.
15Of the 216 instances of all-star analyst turnover in our sample, 72 are nonrepeat all-star turnovers.16Specifically, to construct the earnings forecast accuracy indicator variable, we first compute the relative
earnings forecast accuracy score for each stock followed by the all-star. Then, for the portfolio of stocks followed
by the all-star, we count the proportion of stocks with a score greater than 50. Finally, we assign a value of one to
earnings forecast accuracy if the proportion of stocks in the analyst’s portfolio is greater than the 75th percentile
of all analysts in our sample, and zero otherwise. Earnings forecast timeliness, earnings forecast frequency, and
earnings forecast optimism variables are computed in a similar manner. The rationale for choosing this approach
is to determine if extremely accurate, optimistic, and timely all-star analysts affect deal flow. Our results are
qualitatively unchanged using alternative cut-offs.17The average value of the trend dummy is 0.36.
ARTICLE IN PRESS
Table 5
Explaining the difference in market share between the new and the original bank following an all-star analyst job
change
This table investigates the determinants of the difference in market share between the investment bank gaining
the all-star analyst (the new bank) and the investment bank losing the all-star analyst (the original bank) following
the all-star’s job change. The dependent variable is the difference in market share between the new bank and the
original bank in the two years following the job change. Relative market share before turnover is the difference in
market share between the new bank and the original bank in the two years before the job change. Trend in relative
market share is an indicator variable that takes the value of one if the market share at the new bank increases
relative to that at the original bank between year -2 and year -1, and zero otherwise. The difference in number of
all-stars is computed as the total number of all-stars at the new investment bank minus the number of all-stars at
the original investment bank in the year preceding turnover. Repeat all-star is a dummy variable that takes the
value of one if the all-star has been an all-star in each of the three years prior to turnover, and zero otherwise. To
construct the earnings forecast accuracy indicator variable, we first compute the forecast accuracy score for each
stock. For the portfolio of stocks followed by the all-star, we count the proportion of stocks with a score greater
than 50. Finally, we assign a value of one to earnings forecast accuracy if the proportion of stocks in the analyst’s
portfolio is greater than the 75th percentile of all analysts in our sample, and zero otherwise. Earnings forecast
timeliness, frequency of coverage revision, and earnings forecast optimism indicator variables are computed in a
similar manner. p-values are reported in parentheses.
Equity Bond M&A
Intercept �3.44 �2.81 2.46
(0.22) (0.42) (0.30)
Bank-specific variables
Relative market share before turnover 0.11 0.41 0.24
(0.16) (0.00) (0.00)
Trend in relative market share 1.66 2.04 �2.44
(0.34) (0.36) (0.08)
Difference in number of all-stars 0.20 0.15 0.16
(0.00) (0.02) (0.00)
Analyst reputation-specific indicator variables
Repeat all-star 1.58 �0.11 �1.81
(0.38) (0.96) (0.22)
Earnings forecast accuracy 0.20 �0.36 0.92
(0.92) (0.89) (0.57)
Earnings forecast timeliness 4.75 1.99 0.22
(0.01) (0.40) (0.89)
Frequency of coverage revision 3.33 1.51 2.22
(0.08) (0.51) (0.16)
Analyst bias-specific indicator variables
Earnings forecast optimism 0.03 0.05 �0.02
(0.57) (0.42) (0.70)
Number of observations 208 198 210
Adjusted-R2 (%) 13.23 25.23 15.82
J. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737732
Controlling for bank reputation, does analyst reputation influence deal flow? For debtunderwriting or M&A transactions, the answer is no. None of our proxies are significant inexplaining the relative market share for the two banks. For equity transactions, theanalyst’s earnings forecast timeliness and frequency of coverage revision are significantlypositively related to the relative market share of the new bank after the job change. Finally,
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 733
contrary to reports in the popular press, we find no evidence that earnings forecastoptimism influences either capital-raising or corporate control transactions.18,19 Inadditional tests not reported in the paper, we examine whether the rank of the analyst(Institutional Investor 1st team, 2nd team, 3rd team, or runner-up) influences deal flow andfind no evidence that this is the case.
In summary, our results suggest that all-stars are more influential in equity underwritingdeals than in debt or M&A transactions. In equity deals, analyst reputation (as measuredby forecast timeliness and frequency of coverage revision) helps increase investmentbanking deal flow. In contrast, in debt underwriting and M&A transactions, investmentbanking deal flow is driven by investment bank reputation rather than analyst reputation.Our results contrast with Ljungqvist, Marston, and Wilhelm (2006), who find that analystsaffect debt, but not equity, deal flow. The evidence in Table 5 is also inconsistent withrecent reports in the media and the press that investment banks use optimistic forecastsand recommendations by analysts to win investment banking business.
4.3. Does the increased deal flow at the new investment bank come from clients at the original
bank who follow the analyst to the new bank?
Table 4 reports that the market share for capital-raising transactions increases(decreases) at the new (original) investment bank following an all-star job change. Inthis section we investigate whether this change in deal flow is driven by firms departing theoriginal investment bank after the all-star leaves.
Of the 12,632 firms that carry out a transaction in the two years after the analyst moved,only 148 are clients of the original investment bank. After the analyst’s move, of these 148firms, 82 stayed with the original investment bank, four firms carried out transactions withthe new investment bank, and the remainder used a third bank. These numbers do notsuggest that analysts are bringing their old clients with them to the new bank when theyswitch jobs.
To analyze the movement of clients more formally, Panel A of Table 6 reports the resultsof a logistic regression that models the probability of losing the client when an all-starleaves. We consider only those firms that do an investment banking deal at the departingall-star’s bank in the two years before the all-star leaves and then complete another deal inthe two years after the all-star’s arrival at the new bank. The dependent variable in thisregression takes the value of one if the individual firm switches investment banks andzero if the firm keeps the same bank in the post-move period. We regress this against
18In an alternative specification, we examine whether recommendation bias influences investment bank relative
market share. To construct recommendation bias, we compute the proportion of stocks in the analyst’s portfolio
with abnormal recommendations less than zero (i.e., recommendation is more positive than the consensus). Then,
we create a dummy variable equal to one if this measure is in the bottom 25th percentile of all analysts in our
sample, and zero otherwise. This variable is not significant. Since the sample size is reduced using recommendation
data (available only after October 1993), we do not report the specification.19At the same time that analysts are moving from one bank to another, it is possible that investment bankers are
also moving, bringing their client contacts and business with them. Thus, the increase in relative market share
might be driven by the concurrent movement of investment bankers, rather than analysts. We therefore track key
bankers for both the bank gaining the all-star and the bank losing the all-star. We obtain our sample of banker
movements from Investment Dealers’ Digest and focus on movements by bankers at the rank of managing
director and above. We find very few cases of departures by such bankers around the departure of our all-star
analysts. Controlling for these departures does not affect our results.
ARTICLE IN PRESS
Table 6
Determinants of increase in the relative market share after all-star analyst turnover
Panel A examines a sample of firms covered by the all-star analyst prior to departure that do an investment
banking deal at the departing star’s bank in the pre-turnover period, and that return to complete a deal in the
post-turnover period. The dependent variable takes the value of one if the individual firm switches investment
banks and zero if the firm keeps the same bank in the post-turnover period. Repeat all-star is a dummy variable
that takes a value of one if the all-star has been an all-star in each of the three years prior to turnover, and zero
otherwise. Forecast accuracy, optimism bias, timeliness, and frequency of coverage revision scores are computed
using a methodology used in Hong, Kubik, and Solomon (2000). Number of all-stars at original bank is the
number of stars at the original bank who also issued at least one forecast in the same GICS industry. Continued
coverage at original bank is a dummy variable taking a value of one if the original bank continues coverage after
the all-star’s departure. Difference in number of all-stars is the difference in the number of all-stars within an
industry at the new investment bank versus at the original investment bank in the year preceding turnover. Panel
B reports median values of firm-specific, bank-specific, and analyst-specific variables for new client firms who do
an investment banking deal for the first time at either the original or the new bank in the two years following all-
star analyst turnover. Market capitalization is the market value of equity measured in millions. The market-to-
book ratio is the ratio of a firm’s market value of equity to its book value of equity measured in the year prior to
turnover. Deal size is measured as gross proceeds for IPOs, SEOs, and bond offerings, and as the size of the
transaction for mergers (in $millions). Percentage of cases in which switching star covers new firm and percentage of
new firms with all-star coverage are computed in the year following turnover. In Panel A, p-values are reported in
parentheses. In Panel B, means are reported in parentheses. The p-value for the difference in Panel B is based on a
two-sided Wilcoxon rank-sum test.
Panel A: Determinants of decision by firms to switch investment banks following the all-star job change
Intercept 1.66
(0.18)
Firm-specific indicator variables
Market capitalization in top 25% of industry 0.28
(0.55)
Market-to-book ratio above industry average 0.46
(0.34)
Analyst-specific variables
Forecast accuracy score �0.005
(0.54)
Optimism bias score 0.006
(0.42)
Timeliness score 0.008
(0.27)
Frequency of coverage revision score �0.009
(0.31)
Repeat all-star �0.18
(0.68)
Firm/Bank-specific variables
Number of all-stars at original bank �0.03
(0.30)
Number of deals completed at original bank �0.28
(0.18)
Continued coverage at original bank �1.43
(0.00)
Difference in number of all-stars 0.006
(0.74)
Number of observations 117
Percent Concordant (%) 73.80
J. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737734
ARTICLE IN PRESS
Panel B: Characteristics of new business after the all-star analyst job change
Original bank New bank p-value for
difference
Firm-specific variables
Market capitalization $985.47 $1,199.52 0.13
($8,620.10) ($7,092.30)
Market-to-book ratio 2.58 2.71 0.19
(3.66) (4.23)
Deal size 123.20 148.60 0.12
(503.64) (519.05)
Percentage of firms in category
Market capitalization in top 25% of industry 68.73% 72.77% 0.18
Market-to-book ratio above industry average 28.03% 30.20% 0.47
Bank-specific variables during year prior to transaction, across all analysts
Forecast accuracy score 50.00 53.33 0.78
(52.84) (53.50)
Optimism bias score 50.00 53.85 0.30
(50.47) (53.21)
Timeliness score 50.00 52.27 0.65
(52.66) (51.24)
Frequency of coverage revision score 58.06 58.62 0.62
(57.95) (56.44)
Analyst-specific variables
Percentage of cases in which switching star covers
new firm (%)
18.58 37.37 0.00
Percentage of new firms with all-star coverage (%) 40.70 82.11 0.00
Table 6 (continued )
J. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 735
analyst-specific and firm-specific control variables that might be expected to influence thefirm’s decision to switch investment banks.
The probability of the firm departing with the all-star is unrelated to the all-star’sreputation specific variables. It is significantly negatively related to continued coverage ofthe firm at the original investment bank. This is consistent with Ljungqvist, Marston, andWilhelm (2006), who find that banks that have underwritten a large share of the firm’s pastdebt and equity offerings are significantly more likely to win future mandates.
Overall, we find no evidence that firms follow a departing all-star analyst to the newbank. Hence, the increase in market share documented in Table 4 does not seem to becoming from clients of the old investment bank that follow the analyst.
4.4. Is all-star analyst coverage important in winning investment banking deal flow from new
clients?
We next examine whether the increase in market share at the new investment bank isrelated to the extent of all-star coverage at the new firms. For the two-year period afterthe all-star arrives at the new bank, we first examine whether new client firms who go to thebank gaining the all-star have different firm characteristics than those who go to theinvestment bank losing the all-star. Second, we investigate whether new client firms are
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737736
influenced by the average optimism bias and earnings forecast accuracy of all analystsemployed at each of the two banks, respectively. Specifically, for each firm that carries outa transaction for the first time with either bank in the two years subsequent to the all-starjob change, we compute the average scores for forecast accuracy, optimism bias, forecast
timeliness, and frequency of coverage revision of all the forecasts issued for the firm in theyear prior to the transaction. Third, we compute the percentage of the new business atboth the original and new banks covered by the switching all-star (and any other all-star)in the year prior to the investment banking deal flow. These results are reported in Table 6,Panel B.The new investment bank attracts firms that have firm characteristics—market
capitalization, market-to-book ratio, and deal size—similar to those that provide newbusiness to the original bank. In addition, the scores for forecast accuracy, optimism bias,forecast timeliness, and frequency of coverage revision are similar across both sets of firms.The only characteristic that distinguishes the two types of firms is that of all-star analystcoverage. The switching all-star covers 37% (19%) of the new business at the new(original) bank in the year before the firm is awarded the deal flow. More interestingly,following an analyst job change, the new (original) bank provides all-star analyst coverageto 82% (41%) of the new business in the year prior to being awarded the deal.20
5. Conclusions
We examine a sample of 216 cases in which an Institutional Investor All-AmericaResearch Team (all-star) analyst moves from one investment bank to another between1988 and 1999 to answer the following two questions: Is analyst coverage influenced byinvestment banking relationships? Does analyst behavior, analyst reputation, and/orinvestment bank reputation influence deal flow?Using a comprehensive data set of investment banking deals (underwriting and
corporate control transactions), we find that all-star coverage choices do indeed depend oninvestment banking relationships between the firm and the all-star’s bank. An all-star ismore likely to retain/add coverage of larger glamour firms that have pre-existinginvestment banking relationships with the bank the all-star is moving to.However, the all-star’s behavior does not change after he changes jobs. There are no
changes in optimism bias, forecast accuracy, or forecast timeliness following the jobchange. The all-star is not significantly more likely to be more optimistic in hisrecommendations, that is, recommendation levels are at consensus, both before andfollowing job change. Our results are inconsistent with recent allegations in the popularpress that analysts have helped generate investment banking deal flow by being extremelyoptimistic in their recommendations. To the extent that these allegations are true, ourresults suggest that they cannot be generalized across all analysts.Finally, even though analyst behavior does not change, the new bank does attract a
significantly larger industry market share of capital-raising and M&A deals after thearrival of the all-star, relative to the bank the analyst leaves. Yet, after controlling for bankreputation, all-star reputation, as measured by earnings forecast frequency and timeliness,influences only equity underwriting transactions. Variables measuring the extent to which
20All-star coverage at the new bank refers to coverage by any all-star at the bank and not just the analyst
experiencing turnover.
ARTICLE IN PRESSJ. Clarke et al. / Journal of Financial Economics 84 (2007) 713–737 737
analysts make optimistic earnings forecasts or recommendations do not influence dealflow. The new business does not come from client firms at the old bank who follow the all-star to the new bank; rather it seems to come from firms that the all-star is more likely tocover than at his original bank. In other words, our results suggest that coverage is moreimportant than the degree of optimism of that coverage.
References
Bhojraj, S., Lee, C.M.C., Oler, D., 2003. What’s my line? A comparison of industry classification schemes for
capital market research. Journal of Accounting Research 14, 745–774.
Boni, L., Womack, K.L., 2006. Analysts, industries, and price momentum. Journal of Financial and Quantitative
Analysis 41, 85–109.
Bowe, C., Silverman G., 2004. Merrill rewarded after hiring analyst Tyco favoured. Financial Times, p. 1,
February 3.
Bradley, D.J., Jordan, B.D., Ritter, J.R., 2003. The quiet period goes out with a bang. Journal of Finance 58,
1–36.
Clement, M.B., Tse, S.Y., 2005. Financial analyst characteristics and herding behavior in forecasting. Journal of
Finance 60, 307–341.
Dugar, A., Nathan, S., 1995. The effects of investment banking relationships on financial analysts’ earnings
forecasts and investment recommendations. Contemporary Accounting Research 12, 131–160.
Dunbar, C., 2000. Factors affecting investment bank initial public offering market share. Journal of Financial
Economics 55, 3–41.
Fama, E., French, F., 1997. Industry costs of equity. Journal of Financial Economics 43, 153–194.
Groysberg, B., Nanda, A., 2001. Does stardom affect job mobility? Evidence from analyst turnover in investment
banks. Unpublished working paper. Harvard Business School.
Hong, H., Kubik, J.D., Solomon, A., 2000. Security analysts’ career concerns and herding of earnings forecasts.
Rand Journal of Economics 31, 121–144.
Irvine, P.J., Nathan, A.S., Simko, P.J., 2004. Asset management and affiliated analysts’ forecasts. Financial
Analysts Journal 60, 67–78.
Jegadeesh, N., Kim, J., Krische, S.D., Lee, C.M.C., 2004. Analyzing the analysts: when do recommendations add
value? Journal of Finance 59, 1083–1124.
Krigman, L., Shaw, W.H., Womack, K.L., 2001. Why do firms switch underwriters? Journal of Financial
Economics 60, 245–284.
Leone, A., Wu, J.S., 2002. What does it take to become a superstar? Evidence from institutional investor rankings
of financial analysts. Unpublished working paper, University of Rochester.
Ljungqvist, A.P., Marston, F., Wilhelm, W.J., 2006. Competing for securities underwriting mandates: banking
relationships and analyst recommendations. Journal of Finance 61, 301–340.
Lin, H., McNichols, M.F., 1998. Underwriting relationships, analysts’ earning forecasts and investment
recommendations. Journal of Accounting and Economics 25, 101–127.
Megginson, W., Weiss, K., 1991. Venture capitalist certification in initial public offerings. Journal of Finance 46,
879–903.
Michaely, R., Womack, K.L., 1999. Conflict of interest and the credibility of underwriter analyst
recommendations. Review of Financial Studies 12, 653–686.
Schroeder, M., Smith, R., 2002. CSFB analysts felt pressured on stock reports. Wall Street Journal, Section C1
(September 6).