Do independent research analysts issue more informative recommendation revisions

DO INDEPENDENT RESEARCH ANALYSTS ISSUE MORE INFORMATIVE RECOMMENDATION REVISIONS?. This study examined whether independent research analysts issue more informative stock recommendation r

DO INDEPENDENT RESEARCH ANALYSTS ISSUE MORE INFORMATIVE

RECOMMENDATION REVISIONS?

by Ryan Joseph Casey

A Dissertation Presented in Partial Fulfillment

of the Requirements for the Degree Doctor of Philosophy

ARIZONA STATE UNIVERSITY

May 2009

INFORMATION TO USERS

The quality of this reproduction is dependent upon the quality of the copy submitted Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction

In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted Also, if unauthorized copyright material had to be removed, a note will indicate the deletion

UMI

UMI Microform 3354466 Copyright 2009 by ProQuest LLC All rights reserved This microform edition is protected against unauthorized copying under Title 17, United States Code

ProQuest LLC

789 East Eisenhower Parkway

P.O Box 1346 Ann Arbor, Ml 48106-1346

DO INDEPENDENT RESEARCH ANALYSTS ISSUE MORE INFORMATIVE

RECOMMENDATION REVISIONS?

by Ryan Joseph Casey

has been approved April 2009

Graduate Supervisory Committee:

Mike Mikhail, Chair Artur Hugon Yuhchang Hwang

ACCEPTED BY THE GRADUATE COLLEGE

This study examined whether independent research analysts issue more

informative stock recommendation revisions than non-independent analysts The differential impact on the information content of revisions was examined by regressing abnormal short-term returns on indicator variables for independent classification and recommendation revision direction Results indicated that independent analyst

recommendation upgrades and downgrades were significantly less informative This study then examined whether the identified differences in informativeness were the result

of systematic cross-sectional variation in analyst ability, portfolio complexity, and brokerage firm resources Including analyst and brokerage variables reduced the

information content disparity between independent analysts and non-independent

analysts, however, independent revisions continued to have lower informativeness

Finally, this study evaluated the market reaction before and after the Global Settlement Agreement that was enacted to limit the perceived conflicts in the industry Non-independent upgrades generated a 19.7 percent greater reaction in the post-

regulation period suggesting the Global Settlement Agreement helped mitigate biased research Independent analysts were found to continue to issue less informative

recommendations after the Global Settlement These findings call into question whether the Global Settlement Agreement provided investors with better, timelier research

111

ACKNOWLEDGMENTS

I am grateful to my dissertation committee, Mike Mikhail, Artur Hugon and Yuhchang Hwang for their guidance I would also like to thank workshop participants at Arizona State University including Jim Boatsman, Melissa Martin, Molly Mercer, Steve Kaplan, Rick Laux, Eric Weisbrod and Wan-Ting Wu for their helpful comments and suggestions

IV

Page LIST OF TABLES vi CHAPTER

I INTRODUCTION 1

II RELATED RESEARCH 6

Conflicts 6 Recommendation performance and associated characteristics 8

III DATA AND UNIVARIATE ANALYSIS 10

Data 10 Univariate Analysis 12

IV RESEARCH DESIGN AND EMPIRICAL RESULTS 15

Informativeness of Recommendations 15 Base Model 15 Magnitude 17 Analyst and Brokerage Characteristics 18

Descriptive Statistics: Analyst and Brokerage Characteristics 22 Full Model 22 Information Content Before and After the Global Settlement 24

V ROBUSTNESS TESTS 27

VI CONCLUSION 29

REFERENCES 31

v

LIST OF TABLES

Table Page

1 Descriptive Statistics for Sample Observations 33

2 Transition Matrix of the Distribution of Recommendation Revisions 34

3 Transition Matrix of Event Window Short Term Returns to

Recommendation Revisions 35

4 Regression of Market Reaction to Analyst Revisions 36

5 Regression of Market Reaction to Analysts Revisions Including

Differential Revision Magnitudes 38

6 Comparison of Analyst and Brokerage Characteristics 40

7 Regression of Market Reaction Including Analyst and Brokerage

Characteristics 42

8 Comparison of the Market Reaction Before and After the

Global Settlement Agreement 44

VI

In April 2003, ten large investment banks agreed to settle with the New York State Attorney General and the Securities and Exchange Commission regarding charges of conflict of interest among security analysts.1 The agreement required the sanctioned banks to pay nearly one billion dollars in penalties along with $432 million to fund independent research In addition to these payments, sanctioned banks were ordered to provide three sources of independent research along with their own research reports Thus, the settlement seems to imply that analysts who work for independent research firms are free from the conflicts of interest that cause investment bank analysts to issue biased, presumably inferior, analyst reports This paper studies whether independent research analysts issue more informative stock recommendation revisions than non-independent analysts

Although independent research firm analysts (hereafter, IRAs) are not subject to the

underwriting conflicts faced by investment bank analysts, their research may not be

higher quality for multiple reasons First, analysts from investment banks have access to

a larger pool of resources and additional information channels not available to

independent research firms Second, the profitability and large size of investment banks likely leads to higher pay to retain the best performing analysts Lastly, cross-sectional differences in the number of firms and industries covered may reduce independent

1 Analyst conflicts have been attributed to several factors An analyst's salary and bonus may be linked to quantifiable measures such as his or her firm's underwriting fees (see, e.g., Dugar and Nathan, 1995; Lin and McNichols, 1998) Additionally, brokerages whose analysts issue negative reports on potential or current clients may be excluded from lucrative advisory and underwriting engagements as retribution (see, e.g., Siconolfi, 1995; Solomon and Frank, 2003)

2 research quality

Motivated by the conflict explanation (Michaely and Womack [1999], Lin and McNichols [1998]) versus the resources and ability argument (Mikhail, Walther and Willis [1997], Clement [1999]), the first question my paper empirically tests is whether independent research is more informative than research from analysts at investment banking firms and brokerage firms (hereafter, non-IRAs) I compare the information content of stock recommendation revisions of IRAs with those of non-IRAs by

examining the three-day abnormal return around stock recommendation revisions Each before/after recommendation combination (e.g., a "buy" recommendation upgraded to a

"strong buy", a "buy" downgraded to a "hold" or a "buy" is later reaffirmed as "buy') is partitioned to determine whether the market reacts differently to upgrades, downgrades, and reiterations by analyst firm type

Using a sample of recommendation revisions from 1996 through the end of 2007,1 regress event-period abnormal returns on an indicator variable for analyst firm type, indicators for revision direction and the interaction of the analyst firm type and revision direction I find IRA recommendation upgrades and downgrades significantly less informative than revisions from non-IRAs Consistent with prior research, the market reaction to IRA and non-IRA reiterations is found to be insignificant

In my second analysis, I investigate factors, other than analyst affiliation and revision direction, that could account for the observed differences in recommendation revision informativeness between IRAs and non-IRAs I begin by providing descriptive statistics on the distribution of upgrades, downgrades and reiterations by analyst firm type I find that non-IRAs are less likely to issue revisions that skip recommendation

categories (low magnitude revisions) than IRAs2 This may imply that non-IRA revisions are timelier Additionally, non-IRAs issue relatively more reiterations across all

recommendation levels I also examine the average returns to all before/after

recommendation revision combinations sorted by analyst firm type I find that non-IRA upgrades and downgrades are more informative for both revisions that move one

recommendation category and revisions that skip one or more categories IRA

reiterations are significantly more informative at three of the five reiteration levels Next, I examine whether variations in analyst and brokerage characteristics across firm type explain the information content differences in recommendation revisions To complete this analysis, I introduce proxy variables for analyst ability, brokerage firm resources, and portfolio complexity in my analysis Specifically, I incorporate analyst experience, forecast accuracy, Ail-American status, and a recommendation timeliness measure as proxies for ability The year-specific number of companies and industries followed act as my proxies for portfolio complexity Lastly, I include the size of the brokerage firm as a proxy for firm resources3 I find that my set of explanatory variables helps explain the market reaction to recommendation revisions However, non-IRA revisions remain significantly more informative than IRA revisions even after including these explanatory variables and other controls

Finally, my paper addresses whether the Global Settlement Agreement has affected

2 Mikhail et al (2006) find that the best analysts are less likely to issue revisions that skip recommendation categories

3 The rationale for including these particular variables, along with specific definitions is provided in section 4

4 the market reaction to analyst recommendation revisions I divide my observations into pre- and post-Global Settlement sub-samples with the first month after the agreement was reached as the breakpoint Findings indicate that, after controlling for analyst

characteristics, brokerage characteristics, and company control variables, non-IRA

upgrades are significantly more informative in the post-Settlement period than they were

in the prior period Also, the magnitude of the difference between the informativeness of non-IRA versus IRA upgrades increases in the post-Settlement period Additionally, I find that the information content of downgrades remains constant across time periods The findings from this paper potentially contribute to the literature in several ways I find non-IRA upgrades and downgrades to be significantly more informative, as measured by short-window abnormal returns, than those of IRAs This outperformance extends across all recommendation levels This finding is important given the recent regulatory emphasis on independent firm research Including analyst and brokerage characteristics shrinks the information content difference between non-IRAs and IRAs The difference in the market reaction to upgrades (downgrades) is reduced by 24% (5%)5 Lastly, comparing the pre- and post-regulation period suggests that non-independent upgrades generate a 19.7%6 greater reaction in the post-regulation period suggesting the Global Settlement helped mitigate biased research IRA revisions are less informative both before and after the Global Settlement Agreement

5 The difference in upgrades: from 1.62% to 1.16%, with one-tailed p<.01; the difference

in downgrades: from 1.84% to 1.68%, with one-tailed p = 14

6 The average abnormal returns to pre-regulation (post-regulation) upgrades is 2.36% (2.82%o) The difference is significant with a one-tailed p< 01

The paper proceeds as follows Section 2 discusses other related research In Section 3,1 describe the sample, outline the methodology, define the variables and present descriptive statistics Section 4 reports the results of the empirical tests Section

5 presents the results of additional sensitivity tests Section 6 concludes

II RELATED RESEARCH

Conflicts

Barber, Lehavy, and Trueman (2007) compare the stock recommendation

performance of investment banks and independent research firms by examining returns to portfolios based on the recommendation level and analyst firm type They find that daily abnormal average returns to independent research firm buy recommendations outperform those of investment banks Conversely, they find the investment bank hold/sell

recommendations outperform independent research firm hold/sells Barber et al argue that investment bank analysts' research was compromised due to conflicts related to underwriting relationships

This paper differs from Barber, Lehavy, and Trueman in that my primary focus is the information content of recommendation revisions rather than the long-term

performance of stock recommendation levels While the concepts of information content

of revisions and long-term performance of recommendation levels are related, previous research (Stickel [1995], Womack [1996]) has shown that the strongest market reaction

to recommendations occurs during the short window surrounding the release of the new information resulting from revisions Given that the value of information inherent in

analyst recommendations quickly dissipates, focusing on changes in level rather than the level of a recommendation provides a stronger setting to examine differences across firm

type

Another difference is I separate recommendation revisions into upgrades,

downgrades, and reiterations while Barber et al treat all recommendations within a

general category as equivalent Mikhail, Walther, and Willis (2004) show reiterations of

a recommendation are significantly less informative than upgrades and downgrades The extent that reiterations are more prevalent at non-independent research firms (they are) may, provide a partial explanation for the IRAs superior performance observed in Barber

et al (see footnote 13) Finally, I incorporate explanatory variables for analyst and brokerage characteristics in an effort to better explain abnormal returns around revisions The inclusion of these cross-sectional variables is not feasible in the portfolio analysis as conducted by Barber et al

Agrawal and Chen (2004) investigate conflicts of interest between the investment banking and research arms of investment banks They quantify conflicts by examining the revenue breakdowns of analyst employers They find the level of analysts' stock recommendations is positively related to the magnitude of the conflicts faced (the level of investment banking business) They do not find evidence of long-term mispricing based

on conflicted recommendations My paper extends Agrawal and Chen by explicitly incorporating independent research firms Additionally, I incorporate other analyst and brokerage characteristics as explanatory variables for abnormal returns

This paper is also related to the research of Lin and McNichols (1998), who show that lead and co-lead underwriter analysts' growth forecasts and recommendations are

significantly more favorable than those made by unaffiliated analysts Similar to Barber

et al., Lin and McNichols show that three-day returns to lead underwriter "hold"

recommendations are significantly more negative than those around unaffiliated "hold" recommendations Taken together, these findings suggest that investors are aware of the overly optimistic posture of "conflicted" lead underwriters My paper is similar to Lin

8 and McNichols in that I compare returns around a revision in an analyst recommendation, but I do not partition based on underwriting relationships I compare returns of IRAs, who are assumed to be less conflicted, with returns around investment bank analysts' revisions

Finally, Michaely and Womack (1999) examine analyst recommendations of stocks that their sample firms have recently taken public They show that stocks that

underwriter analysts recommend perform more poorly than "buy" recommendations by unaffiliated brokers around the recommendation date They claim that underwriting conflicts cause analysts at these firms to upwardly bias their recommendations above the level of their peers

Recommendation performance and associated characteristics

In this section, I highlight some relevant papers that link analysts and brokerage characteristics to performance outcomes and information content I follow Stickel (1995) in investigating factors that influence the stock price reaction to recommendation revisions Stickel shows that short-term price reaction is a function of the strength of the recommendation (e.g., upgrade to a strong buy rather than a buy) and the magnitude of the change in recommendation (e.g., upgrade to a strong buy from a buy versus upgrade

to a strong buy from a hold) Stickel also finds that larger brokerage houses have more impact on prices than do smaller brokers, and smaller companies have larger reactions to recommendations than do larger firms This paper incorporates many of Stickel's

findings while comparing differences in price reaction to recommendation revisions Mikhail, Walther, and Willis (1997) investigate the relationship between analyst experience and the profitability of stock recommendations They find that as analysts'

general experience increases, their recommendation revisions are associated with larger abnormal returns Clement (1999) explains variation in analyst forecasting ability with analyst-specific factors He finds that forecast accuracy is positively associated with analysts' experience and employer size, and it is negatively associated with the number of firms and industries followed Mikhail et al (1999) document that analysts with high performance tend to have long tenure at their particular firm

Prior research relates the timeliness of new information, recommendation revisions and forecast revisions, to excess returns and forecast accuracy respectively (Cooper et al [2001], Clement and Tse [2003], and Mikhail et al [2006]) I use revision timeliness as a proxy for analyst ability based on the Mikhail et al (2006) timeliness measure for stock recommendations

Loh and Mian (2006) show that recommendations of accurate forecasters earn

abnormal returns Ertimur, Sunder, and Sunder (2007) relate conflicts of interest with the ability to translate accurate forecasts into profitable buy recommendations Results from these papers suggest that the ability to provide informative stock revisions relates to both the ability and the conditions under which analysts provide relatively accurate earnings forecasts A discussion of specific variable construction is provided in Section 4

III DATA AND UNIVARIATE ANALYSIS

Data

The data for this study comes from the Institutional Brokers' Estimate System (IBES) recommendation detail file The IBES recommendation database records each analyst recommendation as a rating between 1 and 5 A rating of 1 represents a strong buy; 2 represents a buy; 3 is a hold; 4, a sell; and 5 represents a strong sell An upgrade

represents a change to a more positive recommendation category, a downgrade is a change to a more pessimistic category and a reiteration is a recommendation rating that equals the prior rating I also note whether the upgrade or downgrade skips

recommendation categories

Since my design requires a prior outstanding recommendation, I do not include initiations of new firm coverage Each IBES recommendation is compared to the prior recommendation by the same analyst and categorized as an upgrade, downgrade, or reiteration If the prior recommendation in the IBES database occurred over one-year prior to the new recommendation, the observation is dropped under the assumption that the prior recommendation has become stale.7 In the event the analyst changed from one bank type to another between the prior and the current recommendation, I drop the observation8

7 Asquith, Mikhail and Au [2005] report that analysts usually write a minimum of six reports a year on the companies they follow If a prior recommendation is more than a year old, it is likely that recommendations are missing or that the firm is not followed on

a continuous basis by that particular analyst

8 919 of 154,234 revisions or 0.6% come from an analyst that switched between IRA and non-IRA firm types These revisions come from 161 unique analysts Their performance before and after the switch is not examined in this paper

Following Cowen et al (2006), Barber et al (2007) and Jacob et al (2008) I utilize Nelson's Directory of Investment Research (1996-2006) to classify firms Firms that appear in the IBES dataset and also appear in the Nelson's directory were retained in my sample I classify each firm based on its classification in Nelson's: investment bank, brokerage firm, or pure research firm Investment banks participate in underwriting as either a lead underwriter or a distributor of new equity offerings Brokerage firms do not participate in investment banking but receive trading commissions Pure research firms sell research but do not undertake any investment banking or trading activities For ease

of comparison, I group analysts into two groups: IRAs (independent research analysts) and non-IRAs (investment bank analysts and brokerage firm analysts9) Panel A of Table 1 reports summary statistics for the sample by year The data set extends from

1996 to 2007 and contains 153,315 recommendations of 10,733 unique companies As expected, the number of IRAs grows during my sample period at a larger rate than the growth in the number of non-IRAs11 The total sample contains 77,313 revisions that result in a buy rating and 79,603 that result in a hold or sell

9 This grouping was done for expositional purposes In unreported tests, both brokerage and investment bank revisions are found more informative than those of independent research firms

11 The expected growth is partially due to the Global Settlement's funding of independent research and the requirement that sanctioned banks provide this research to their clients

12 The total number of observations is roughly half of the number utilized in Barber et al (2007) due to my requirement that a prior analyst-firm recommendation be available during the prior 12-month period The number of unique brokerage houses, and unique covered firms is very similar to Cowen et al (2006), Barber et al (2007) and Jacob et al (2008)

12

Univariate Analysis

Panel B of Table 1 contains descriptive statistics organized by analyst firm type The average three-day market-size adjusted abnormal returns for non-IRA revisions are significantly larger for both upgrades and downgrades13 Non-IRA upgrades

(downgrades) experience an average abnormal return of 3.45% (-4.86%) compared to IRA upgrades (downgrades) with an average abnormal return of 1.81% (-2.19%) The average number of analysts following a firm receiving non-IRA revision (10.02 analysts)

is significantly greater than the number following IRA revision firms (8.33 analysts) The other firm-level control variables do not significantly differ across analyst firm type Collectively, these statistics provide initial evidence that IRA revisions are of lower relative information content

The distribution of high-magnitude versus low-magnitude revisions, along with the distribution of reiterations, might influence results when comparing the average

performance of recommendation revisions across analyst firm types Table 2 presents a transition matrix of all recommendations included in the sample, organized by bank type

13 The average three-day abnormal returns for buy recommendations (including upgrades

to buy or strong buy and reiterations with a buy or strong buy rating) and hold and sell recommendations (including downgrades to hold, sell, or strong sell along with

reiterations with a hold, sell, or strong sell rating) are also calculated The average return

to the buy group does not significantly vary by analyst firm type (non-IRA: 2.80%, IRA: 2.52%o) It appears that including reiterations with upgrades drives down the average returns to the non-IRA buy category, while driving up the IRA buy category returns The non-IRA hold/sell average returns are significantly greater than IRA hold/sells (non-IRA: -4.01%, IRA: -2.19%)

14 Other included control variables include: firm size (Compustat Datal99 x Data25), market-to-book ratio (Compustat Datal99 x Data25 / Data60) and beta (firm-specific regression of the firm's daily return on the beginning of the year matched size-decile daily return using the 100 days ending 10 days before the revision date)

Each cell of the matrix in Panel A contains two numbers The first number represents the raw number of observations for the particular two-recommendation combination by bank type The second number is the specific bank type percentage of observations within each general recommendation level For example, the investment bank reiterations of strong buy (prior recommendation of 1, subsequent recommendation of 1) occurred 5,714 times and represent 9.03% of the buy (strong buy or buy) of all investment bank

recommendations The highlighted diagonal in the matrix presents reiterations of analyst recommendations

Analyzing data in the "buy" category sheds light on some interesting patterns

Investment bank analysts issue relatively more reiterations of strong buy and buy (9.03%, 17.05%o respectively) than independent research analysts (8.21%, 6.08%>) and brokerage analysts (6.96%>, 8.98%>) Also of note, IRAs issue relatively more buy recommendation revisions that skip a rank (high-magnitude revisions) These revisions may be relatively more "stale" as evidenced by the need to issue an extreme revision due to missing a necessary revision in the interim IRAs issue more recommendation revisions that go from a hold to a strong buy; these represent 39.64%> of their "buy" level revisions This compares to 14.97% for investment bank analysts and 16.87% for brokerage analysts for the same hold-to-strong-buy revision category Additionally, IRAs issue relatively more high-magnitude (revisions that skip a rank) buy revisions then investment banker analysts and brokerage analyst in 4 of the 5 high magnitude buy cells Stickel (1995) finds that short-term price reaction is related to the magnitude of the change in recommendation Last item of note in the buy category is the low number of downgrades from strong buy

to buy for IRAs

14 Focusing on the hold/sell portion of Panel A reveals similar distribution patterns as those in the buy cells IRAs issue relatively more high-magnitude sell revisions In five

of the six high-magnitude hold/sell level recommendation revision cells independents issue relatively more revisions Non-IRAs issue relatively more hold and sell

reiterations, while independents issue relatively more strong-sell reiterations

Table 3 presents the transition matrix with mean abnormal returns represented in each box rather then the distribution of observations Specifically, abnormal returns are calculated as the market-size adjusted return measured over a three-day period starting the day prior to the new recommendation and ending one day after It appears that investment bank and brokerage analyst upgrades and downgrades are more informative then those of independent research analysts in the majority of circumstances In all of the non-reiteration buy cells, and in 9 of 12 non-reiteration hold/sell cells, the investment bank and brokerage analyst revisions are more informative to the market Conversely, independent research analyst reiterations are more informative in both buy reiteration categories and in two of the three hold/sell reiteration cells

Informativeness of Recommendations

As noted above, I measure information content by comparing returns around a revision in

analysts' specific stock recommendations My dependent variable, abnormal return

performance (ABRET), is the market-size adjusted return measured over a three-day

period starting the day prior to the new recommendation and ending one day after My

first regression specification tests whether recommendation revisions by non-IRAs are

more informative then those of IRAs I create an indicator variable, NON-IRA, that takes

the value of one if the given recommendation revision comes from an analyst employed,

during the current year, at an investment bank or brokerage firm This allows me to

directly test the incremental information provided by non-IRA analysts

Base model

To begin, I adopt a simple regression of abnormal returns on indicator variables

for revision, bank type, and interactions as follows:

AB_RET = a 0 +a,NON-IRA + a 2 UP +a 3 DOWN + a 4 NON-IRA*UP + a 5

NON-IRA*DOWN + CONTROLS + e (1)

I examine the market reaction to all revisions simultaneously without segregating based

on the recommendation category15 I include the following firm level control variables:

market-to-book ratio (MKTBK), firm size based on market capitalization (SIZE), firm

period beta calculated using the 100 trading days ending 10 days before the revision date

(BETA), and the number of unique analysts following the firm in a given year

All regressions include standard errors adjusted for both heteroskedasticity and

intra-analyst error correlation Controls for firm and year effects are also included

16 (FOLLOW) These controls are meant to address the possibility that IRAs and non-IRAs cover different types of firms

Results from estimating equation (1) are reported in Table 4 The intercept of this equation captures the abnormal return associated with reiterations for independent

research firms The UP and DOWN indicator variables capture the differential returns to upgrades and downgrades for IRAs Column 1A presents the base regression including a set of firm control variables The bottom portion of Table 4 presents the mean returns associated with IRA and non-IRA reiterations, upgrades and downgrades For the

specification including controls, IRA upgrades (1.0%, one-tailed p<-01) and downgrades (-1.0%, one-tailed p<-01) are significantly different than zero providing evidence that the market finds recommendations by IRAs informative The non-IRA indicator variable is not significantly different from zero (non-IRA reiterations), however interacting the

NON-IRA with the UP and DOWN variables reveals that NON-IRA upgrades and

downgrades have a significantly greater effect on the market reaction The bottom

portion of Table 4 also calculates the difference in mean returns associated with

upgrades, downgrades and reiterations organized by analyst firm type Comparing the

average returns to upgrades (non-IRA upgrades = ao + aj + (X2 + a.4; IRA upgrades: ao + ai) shows that non-IRAs experience significantly larger market reactions (2.6% versus 1.0%, with one-tailed p< 01) Non-IRA downgrades (non-IRA downgrades: ao + a/ + a? + af, IRA downgrades: ao + a?) are also found more informative (-2.8% versus -1.0%,

with one-tailed p< 01) The average abnormal returns to reiterations do not significantly differ between groups These results indicated that non-IRA revisions are more

informative than IRA revisions

Magnitude

In equation (2) I allow the magnitude of the revision to vary by creating single level

versus multiple level revision indicator variables If an analyst revises his/her

recommendation only one level, I categorize these upgrades and downgrades as

UP SINGLE or DOWNSINGLE, respectively If the revision skips at least one level

(for example, an upgrade to a strong buy from a hold), then I classify the upgrade or

downgrade as UP SKIP or DOWN_SKIP, respectively Equation (2) is specified as

follows:

ABRET = oo + aiNON-IRA + a 2 UP_SINGLE + cc 3 UP_SKIP +

a 4 DOWNJINGLE + a 5 DOWN_SKIP + a 6 NON-IRA*UP_SINGLE +

a 7 NON-IRA*UP_SKIP + a 8 NON-IRA*DOWN_SINGLE + a 9

NON-IRA*DOWN_SKIP + CONTROLS + (2)

This specification allows the distribution of "extreme" revisions to vary across bank

type; thus, the result cannot be due to one type of bank issuing relatively more strong

revisions Following equation (1), the revision indicator variables are interacted with the

non-IRA indicator variable to test whether the market finds these revisions incrementally

informative As in table 4, the lower portion of Table 5 presents the mean returns

associated with the various upgrades and downgrades organized by firm type Findings

indicate that both non-IRA single level upgrades (2.5%, one-tailed p< 01) and non-IRA

high-magnitude upgrades (2.7%, one-tailed p< 01) are associated with stronger price

reactions then those from IRAs (0.50%, one-tailed p< 01 and 0.91% p< 01,

respectively) For downgrades, non-IRA high-magnitude revisions (-5.0% one-tailed p<

01) experience a greater reaction than non-IRA single level revisions (-3.3%) one-tailed

p< 01) For both single level and high magnitude downgrades, non-IRA revisions are

18 more informative than those from IRAs (single level: -1.5% one-tailed p< 01; high magnitude: -0.8% one-tailed p< 01) Lastly, high magnitude non-IRA downgrades are significantly more informative than non-IRA single level downgrades (unreported, one-tailed p<.01) while IRA single-level downgrades are more informative than IRA high-magnitude downgrades (unreported, one-tailed p<.01) This result suggests that the market views extreme downgrades as credible only when originating from a non-IRA due

to conflicts or that IRA extreme downgrades are less timely

The power of this model is greater than the model in Table 4 This specification

increases the explanatory power of the model by roughly 10% (R of 13.8% versus 12.6%

in equation 1) In summary, results to this point suggest that IRA revisions are

significantly less informative than those of non-IRA analysts after controlling for revision direction and magnitude

Analyst and Brokerage Characteristics

In this section, I investigate whether the relative informativeness of

recommendation revisions between IRAs and non-IRAs can be explained by analyst, portfolio, or resource considerations I use five measures, based on prior literature, to capture analyst ability or "quality." My first two measures capture experience Mikhail

et al (1997) find experience translates into more profitable stock recommendations Also, more experienced analysts likely possess skills that have helped them remain employed for an extended period of time Mikhail et al (1999) document that analysts with high performance tend to have long tenure at their particular firm My proxy used for general experience is calculated as:

GEXP = number of quarters for which analyst i has supplied at least one

recommendation

I also include a firm-specific experience variable The proxy for firm-specific experience

is calculated based on the analyst's experience following a particular firm as follows:

FEXP = number of quarters for which analyst i supplied at least one

recommendation for the specific firm

My third measure captures forecast accuracy Loh and Mian (2006) show that

recommendations of accurate forecasters earn greater abnormal returns Annual polls of

analyst performance, such as the Institutional Investor and the Wall Street Journal imply

that the ability to forecast earnings and the ability to provide useful recommendations are related skills The association between the quality of earnings forecasts and the quality of recommendations provides a direct quantifiable measure of analyst ability, which I measure as follows:

ACCUR = Analyst i 's forecast accuracy for firm j in year t, calculated as the

absolute forecast error for all analysts who follow firm j in year t minus the absolute forecast error of analyst i following firm j in year t, scaled by the range of absolute forecast errors for all analysts

following firm j in year t

Desai, Liang, and Singh (2000) find recommendations from analysts with "all-star" status outperform other recommendations As another measure of ability, I define all-star status

as follows: