Can companies buy credible analyst research?

Marcus Kirk*
Goizueta Business School, Emory University, 1300 Clifton Road, Atlanta, GA 30322, USA

Version: September 8, 2008

Abstract
The benefits of analyst coverage are well documented but the majority of publicly traded
companies do not have coverage. In this paper, I explore a controversial mechanism recommended by an
SEC advisory committee by which companies can acquire coverage – paying for it.
I find firms that pay for research have greater future performance uncertainty, higher information
asymmetry, and lower visibility. I find that 2-day cumulative abnormal returns are significantly related to
the issuance of paid-for reports suggesting they have information content for investors despite the
inherent conflicts of interest. After the initiation of coverage, companies experience an increase in
liquidity, institutional ownership, and sell-side analyst following.
However, paid-for coverage is neither a perfect substitute for sell-side coverage nor
unconditionally valuable. Paid-for analysts issue relatively less accurate forecasts and more optimistic
recommendations than sell-side analysts. In addition, the results are strongest for fee-based research firms
with ex ante policies that reduce potential conflicts of interest and enhance credibility.

Keywords: Analyst coverage, voluntary disclosure, capital markets, credibility

* I thank Jan Barton, Marty Butler, Stan Markov, Grace Pownall, Greg Waymire and workshop participants at the
Emory Archival Brown Bag for helpful comments. I am grateful for the financial support of the Roberto C. Goizueta
Foundation.

*Contact information: Tel.: +1-404-396-5997; fax: +1-404-727-5337.

Email address: marcus_kirk@bus.emory.edu

Electronic copy available at: http://ssrn.com/abstract=1265709

1. Introduction

This study examines whether paying for analyst research has value to investors and the company

purchasing research. While the existing literature suggests analysts are important information

intermediaries, it is unclear whether the inherent conflicts of interest in paid-for research preclude it as a

viable alternative to traditional sell-side analyst coverage. This question has become particularly

important as economic and regulatory changes over the last decade such as the minimum bid-ask spread,

Reg FD, SOX, and the Global Settlement have accelerated the drop in sell-side analysts’ coverage of

small- and mid-cap stocks (Leone, 2004).1 Lack of analyst coverage is a pervasive empirical regularity

that affects the majority of publicly listed firms. Citing the adverse effects for individual companies from

this lack of coverage, an SEC Advisory Committee recently encouraged the use of paid-for research as a

way for firms to access the benefits of analyst coverage. However, there is no empirical evidence

examining the viability of this controversial equity research model.2

In paid-for research, companies hire a fee-based research firm to prepare one or many reports.

Paid-for and sell-side analysts may perform many similar functions as information intermediaries. For

example, paid-for analysts interpret past events and perform prospective analysis; evaluate business

models; and communicate this analysis by writing reports, which often include a business overview,

earnings forecasts, price forecasts, and a recommendation level. More generally, by providing public

information they can help aid investors’ decisions in allocating resources across securities and make the

market more efficient while also providing benefits for the covered company such as increased liquidity

and lower cost of capital (Healy and Palepu, 2001; Barth and Hutton, 2004; Irvine, 2003).

1
Bushee and Miller (2007) find that investor relations professionals view Reg FD and SOX as decreasing the
analyst following of small and mid-cap firms and that attracting analysts for these firms is unlikely. In particular,
analysts told the IR professionals that their clients lack the trading volume (and thus commissions) to warrant
coverage.
2
For example, Lang et al. (2004) find analysts tend not to cover firms with severe agency problems despite the fact
that these firms may benefit the most from the monitoring effect of analyst coverage. They state “It’s possible some
managers of firms with concentrated ownership might prefer to have greater analyst following but do not have a
ready mechanism with which to attract analysts.”

Electronic copy available at: http://ssrn.com/abstract=1265709

 Despite the potential benefits of analyst coverage, some view paying for research with

skepticism. Critics allege that the inherent conflict of interest robs the research of any value and “is one of

the most scurrilous practices in investor hyping” (Metzger, 2003). Paid-for analysts may feel pressure to

optimistically bias their forecasts and recommendations in order to encourage the client to renew

coverage. Beyond the research funding conflict, paid-for analysts may have neither the ability nor the

experience necessary to provide any new or valuable information.

The goal of this study is to provide evidence on the role of paid-for research. To do this I

examine 10 fee-based firms from 1999 to 2006; including the top five firms in the industry. I hand-collect

over 6,000 analyst reports from the research firms’ websites and archives. I identify over 500 US firms

that purchased coverage over this period, reflecting the burgeoning demand for paid-for research among

firms with little or no analyst coverage.

First, I provide descriptive evidence of what types of companies buy research and model a

company’s decision to pay for analyst coverage. I find that managers are more likely to buy analyst

coverage for firms: that are younger; that are smaller; whose businesses depend on intangible assets;

engaging in mergers or acquisitions; with more growth opportunities; and with more volatile stock

returns. This is consistent with the notion that firms with weaker information environments, lower

visibility, higher information asymmetry between insiders and outsiders, and greater uncertainty about

future earnings and cash flows have the most to gain from analyst coverage and reducing this uncertainty

(Lang et al., 2001; Botosan, 1997) but at the same time are unlikely to attract sell-side analysts as

brokerages are loathe to devote effort and resources to firms with difficult-to-predict future earnings and

cash flows (Bhushan, 1989; Barth et al., 2001). Likewise, I find firms with low share turnover are more

likely to buy analyst coverage as sell-side analysts are unlikely to generate enough commissions to justify

covering small, thinly traded stocks (Irvine, 2003; Bushee and Miller, 2007). Finally, I find that buying

analyst coverage is negatively associated with institutional investor ownership and sell-side coverage and

positively associated with prior stock returns and future financing.

Electronic copy available at: http://ssrn.com/abstract=1265709

 I next examine the consequences on a firm’s stock and information environment after hiring an

analyst. Prior literature concludes that sell-side analysts’ earnings forecasts and recommendations have

information content for investors (Givoly and Lakonishok, 1979). However, both the usefulness and the

credibility of paid-for research are in doubt because of the inherent conflicts of interest. I evaluate

whether investors view paid-for research reports as having information content by examining the 2-day

[0,+1] cumulative abnormal returns (CARs) centered on the report date. I find that these reports provide

information content to the market. Reports initiated with Strong Buy recommendations are related to a

size-adjusted return of 3.1%; a Buy initiation with 1.4%; while initiations at other levels are

insignificantly different from zero. Investor reaction is not limited to the initiation sample suggesting that

the abnormal returns are not solely due to investors interpreting purchasing analyst coverage as a signal of

favorable future performance. Specifically, consistent with intuition and the sell-side literature, upgrades

and reiterations from paid-for analysts are associated with positive and significant abnormal returns while

downgrades are associated with statistically negative abnormal returns. I find a decrease in the bid-ask

spread and increases in institutional ownership and number of sell-side analysts in the quarters following

the initiation of paid-for coverage. However, only the number of institutional investors and sell-side

analysts remain significant after matching to a control company.

I provide evidence on the relative optimism and accuracy of paid-for analyst forecasts and

recommendations versus sell-side analysts using the subset of firms on I/B/E/S that have both sell-side

and paid-for research. I find that critics’ concerns over the quality and independence of paid-for research

are not without merit. While the results show that paid-for earnings forecasts are not significantly more

optimistic relative to sell-side analysts, paid-for analysts tend to issue less accurate forecasts and more

optimistic recommendations than their sell-side counterparts.

Finally, I more closely examine the central issue of credibility by investigating how capital

market reactions and analyst optimism vary across fee-based firms based on which firms are more likely

to have conflicts of interest. I classify fee-based firms into two categories (high credibility and low

credibility) based on ex ante business operations and policies. High credibility firms are pure research

3
firms and do not allow analysts to hold or trade client stock. Low credibility firms have other lines of

business such as investment banking or business consulting and allow analysts to hold and trade client

stock. Consistent with my predictions, I find that high credibility firms have greater abnormal stock

returns to initial recommendations, upgrades and reiterations. I also find that high credibility clients

experience significant decreases in the bid-ask spread and increases in turnover, institutional investor

ownership and sell-side coverage with all these changes remaining economically large and significant

after matching against a control company. On the other hand, low credibility clients saw the same but

smaller increases in turnover, institutional ownership, and sell-side coverage, after initiating paid-for

coverage. However, none of the post-initiation changes remained significant in the low credibility sample

after matching against a control company. The earnings forecasts and recommendations made by high

credibility firms are not significantly less optimistic or more accurate than those made by low credibility

firms; on the contrary, opposite my expectations, low credibility firms produce less optimistic forecasts

than high credibility firms.

My evidence indicates that paid-for analyst research is a viable mechanism for firms with little or

no analyst coverage. The results imply paid-for research reports have information content for investors

and can help firms capture potential benefits from analyst coverage such as increases in liquidity and

institutional ownership. However I hesitate to conclude that paid-for research can be unconditionally a

substitute for sell-side coverage. First, the evidence shows that paid-for analysts issue relatively less

accurate forecasts and more optimistic recommendations. Second, the credibility of the fee-based firm

also plays a central role. Investors appear to consider factors such as other business operations and analyst

trading restrictions when assessing the credibility and information content of fee-based firms.

This study is important for securities regulators, practitioners, and academics. First, I contribute

to the disclosure, visibility, and analyst literature by examining a relatively recent mechanism (i.e.

companies buying analyst research) that has not been systematically examined before. Prior work shows

that only firms with certain characteristics such as size, trading volume, and potential investment banking

business are able to obtain analyst coverage and the resulting benefits. I extend the analyst coverage

4
literature by finding that paid-for analyst research provides a possible avenue to overcome these

problems. Second, I give insight into the viability of the SEC Advisory Committee’s endorsement of

paid-for analyst coverage and whether the well-documented benefits from analyst coverage are lost due to

the inherent conflicts of interest from paying for the coverage.

Lastly, to the extent that this industry continues to grow, it may create methodological issues for

future researchers. Research designs that include analyst following or consensus forecasts may need to

identify whether analysts covering a firm are paid-for or not. For example, I identify between 400 and 450

unique companies in I/B/E/S over 1999–2006 with recommendations and earnings estimates provided by

fee-based research firms in my sample. Without paid-for coverage, most of these companies would be

classified as having no analyst coverage.

In the next section, I offer some institutional background and develop hypotheses. Section 3

describes the sample and research design, section 4 presents the results, section 5 more closely examines

the issue of credibility, section 6 provides robustness tests, and section 7 offers concluding remarks.

2. Background and hypotheses

2.1 Institutional background

Analysts play an important role as information intermediaries in the capital market. By providing

public information they help aid investors’ decisions in allocating resources across securities and can help

make the market more efficient (Barth and Hutton, 2004; Healy and Palepu, 2001). Early research has

concluded that sell-side analyst earnings forecasts and recommendations have information content for

investors (Givoly and Lakonishok, 1979). Firms can also benefit from analyst coverage through enhanced

visibility (Merton, 1987), improved liquidity (Brennan and Subrahmanyam, 1995; Irvine, 2003), and

reduced agency costs through monitoring (Moyer et al., 1989; Lang et al., 2004) which ultimately can

increase the value of the firm (Chung and Jo, 1996).

Yet in the traditional equity research model, coverage depends on the private benefits and costs

that accrue to the brokerage firm and analyst. Sell-side analysts are more likely to follow firms that take

5
less effort to follow (Barth et al., 2001; Bhushan, 1989) and large firms with high trading volume to

generate trading commissions as both retail and institutional investors can buy and sell shares easily

(Irvine, 2003). Investor Relations professionals confirm that attracting sell-side coverage for their small

and mid-cap clients is unlikely as their stocks do not have enough volume to justify coverage of the firm

(Bushee and Miller, 2007).

The 2000s provided numerous economic and regulatory shocks that reduced the likelihood a firm

receives sell-side coverage. The minimum trade size increment to a penny from 1/8th of a dollar ate away

at one revenue stream funding analyst research. In 2003, the SEC sanctioned the 10 largest investment

banks in the Global Research Analyst Settlement. Part of the ruling decoupled investment banking from

research by no longer allowing analysts to be paid for their contribution to the firm’s investment banking

activities. An indirect effect of severing this link is that firms now have less incentive to employ analysts.3

Fewer analysts led to fewer firms covered. The first firms to lose coverage were the small- to

mid-sized firms whose private benefits for the analyst or brokerage firm outweighed the costs of coverage

– 83% of firms with market caps less than $125 million have no coverage (SEC, 2006; Leone, 2004).

Thus many small, thinly traded stocks with difficult to predict revenues and earnings are unlikely to

attract analyst coverage. Ironically, these firms may be the ones who could benefit the most from analyst

coverage because of their lack of visibility, higher agency costs, greater uncertainty, and higher

information asymmetry (Lang, 1991; Botosan, 1997). In response to the paucity and indifference of sell-

side analysts, firms have recently begun to explore changing the structure of the analyst-firm relationship.

For some firms, the potential benefits from analyst coverage are great enough that they are willing to pay

for coverage.

The growth in this demand has led to the emergence of a new industry: paid-for research. In paid-

for research, companies hire an outside research firm to prepare one or more analyst reports. The audience

3
Kadan et al. (2008) report that when eight of the brokerage houses affected by the Global Settlement changed their
ratings system in 2002, they also discontinued coverage on about 12% of their firms.

6
for this new research model is large – about 60% of publicly listed firms have no analyst coverage. I

identify over 500 unique companies that bought at least one analyst report since 1999.

Paid-for and sell-side analysts perform similar functions as information intermediaries. For

example, paid-for analysts interview managers; interpret past events and perform prospective analysis;

evaluate business models; and communicate this analysis by writing reports, which often include a

business overview, earnings forecasts, price forecasts, and a recommendation level.

However, there can be substantial variation between paid-for research firms in terms of analyst

ability, scope of coverage, dissemination, and compensation. On one hand, many analysts have CFAs, are

members of their local AIMR, and have moved across from Wall Street and the buy-side. On the other

hand, people specializing in public and investor relations are also becoming paid-for analysts (Hasset and

Mahoney, 2001). Services range from one-shot company snapshots to longer term contracts where the

analyst provides ongoing services and reports similar to sell-side research. The price for the coverage

ranges from $5,000 for a one-shot report to $50,000 for a year of coverage.4 Paid-for research firms

disseminate their reports through their webpage, subscriber lists, media, and third-parties such as Zacks,

Multex, First Call, and I/B/E/S.5 Some paid-for firms permit accepting stock as part of their

compensation. Other firms explicitly prohibit the paid-for firm or analyst from holding or trading in a

client’s stock, only accept lump sum cash payments up front, and do not allow a firm to withdraw from

coverage before the expiration of the term.

A recent report commissioned by the Security and Exchange Commission (SEC) specifically

highlights the lack of analyst coverage as an existing problem with adverse effects such as reduced firm

valuation, higher financing costs, decreased market efficiency, lower liquidity, and inhibiting investors’
4
Fee-based research firms are required to disclose the compensation received for a report to comply with SEC rule
17(b) which states “It shall be unlawful for any person, by the use of any means of instruments of transportation or
communication in interstate commerce or by the use of the mails, to publish, give publicity to, or circulate any
notice, circular, advertisement, newspaper, article, letter, investment service, or communication which, though not
purporting to offer a security for sale, describes such security for a consideration received or to be received, directly
or indirectly, from an issuer, underwriter, or dealer, without fully disclosing the receipt, whether past or prospective,
of such consideration and the amount thereof.”
5
I find eight paid-for research firms with estimates and/or recommendations in I/B/E/S. These firms cover 400-450
firms on I/B/E/S over the period 1999-2006. As of 2005-2006, approximately 3-4% of the US firms on I/B/E/S had
forecasts or recommendations issued by paid-for analysts that were included in I/B/E/S.

7
resource allocation decisions. In the report, the Advisory Committee on Small Public Companies to the

SEC officially endorses the paid-for research model and recommends the SEC to actively encourage it as

a mechanism for companies to obtain analyst coverage and alleviate this problem (SEC, 2006).

2.2 Hypotheses

One view of paid-for research is that it allows firms unlikely to attract sell-side analysts to

potentially capture the benefits of analyst coverage through reduced information asymmetries and

monitoring. Analyst coverage can have benefits if the information contained in analysts’ reports is

considered useful and credible. The extant literature concludes that sell-side analyst earnings forecasts

and recommendations have information content for investors (Givoly and Lokonishok, 1979), and that

investors also react to other information in the analyst report (Asquith et al., 2005).

However, the alternative view questions both the usefulness and the credibility of paid-for

research. On one hand, many paid-for analyst firms claim to offer in-depth research and analysis: analysts

may visit the company’s site, talk to management, and study how the economic and competitive

environment affects the firm. On the other hand, paid-for analysts may not have the ability needed to

provide new information to the market.

Even if the quality of the paid-for reports approaches that of sell-side reports, the collection and

interpretation of information may be compromised in an attempt not to alienate the firm that supplies their

revenue. For example, one company’s response to a survey on paid-for research was “we feel VERY

strongly that research, in order to be reliable, should be independent, and in order to be independent,

should NOT be paid for” (Metzger, 2002). The inherent conflict of interest from the employment

relationship may taint the perceived objectivity of the report to the point where investors simply ignore

paid-for research.

The ambiguity over whether paid-for research reports have information content to investors

makes it an empirical question. To the extent that paid-for reports provide useful and credible information

8
to investors, I expect there to be a predictable abnormal stock return response to the recommendation

level and change of recommendation provided in the paid-for report.

Hypothesis 1: There will be significant abnormal stock returns during the two-day window (day
0 to day +1) centered on the release of the paid-for research report.

In addition to the information content of individual analyst reports, the extant literature predicts

positive benefits for the firm from increased analyst coverage. Analyst coverage is suggested to reduce

information asymmetry not only between insiders and outsiders but also between informed and

uninformed traders (Brennan and Tamarowski, 2000). According to economic theory, reductions in

information asymmetry decrease the adverse selection cost of trading in a stock and increase liquidity

(Diamond and Verrecchia, 1991; Brennan and Subrahmanyam, 1995). Irvine (2003) finds an incremental

price impact to a sell-side analyst’s initial report and that liquidity improves after initiations.

Paying for analyst coverage could also have an effect on institutional investor ownership by

improving the information environment and visibility (Bushee and Noe, 2000; Merton, 1987). It may also

influence the likelihood of sell-side analysts following the firm through an increase in liquidity and

institutional investor ownership (O’Brien and Bhushan, 1990) and a decrease in the cost of information

acquisition (Bhushan, 1989; Barth et al., 2001). However, many institutional investors remain skeptical

about paid-for research and either dismiss it or “[view] it with derision” (Metzger, 2002). Overall, if paid-

for research is able to improve the information environment of the firm and provide other capital market

participants with meaningful insight, I expect liquidity, institutional investor ownership, and sell-side

analyst coverage will increase after the initiation of paid-for coverage.

Hypothesis 2: There will be an increase in liquidity, institutional investor ownership, and sell-
side analyst coverage after the initiation of paid-for coverage.

The biases in and properties of sell-side analysts’ recommendations and forecasts are well known.

Forecasts tend to be optimistic and recommendations almost exclusively for buys (Brown et al., 1985;

Asquith et al., 2005) although forecast optimism appears to be declining (Matsumoto, 2002). The existing

9
literature identifies numerous factors that influence the optimism of analysts’ forecasts: research funding,

trading incentives and firm reputation (Cowen et al, 2006); investment banking relationships (Lin and

McNichols, 1998; Michaely and Womack, 1999; Dechow et al., 2000); innate ability, company

assignments, and industry specialization (Jacob et al., 1999); and a self-selection bias to cover firms with

favorable outlooks (McNichols and O’Brien, 1997).

Analysts whose firms are affiliated with the covered company (for example, an IPO or

underwriting relationship) tend to issue more optimistic reports. This suggests that paid-for analysts are

similarly influenced to be relatively optimistic because the research is funded by the client who may

discontinue funding if the analyst’s reports are unfavorable. Paid-for analysts may also have less

experience and ability leaving them more susceptible to optimistic suggestions and pressure from

management of the firm under coverage. Cowen et al. (2006) find smaller, pure research firms to be

relatively optimistic in terms of earnings forecasts and recommendations compared to larger, full-service

investment banks.

However, the relative optimism of paid-for analysts may not be straightforward. Cowen et al.

(2006) note that while sell-side research is rarely sold directly, it is paid for indirectly through mark-ups,

commissions, and other funding sources that influence incentives for analyst optimism. As John Dutton,

of the paid-for firm Dutton Associates, argues: “Why all the fuss over issuer-paid-for research? All

research is paid for if you really think about it” (Brooks, 2006). Additionally, the reputation of sell-side

analysts has suffered over the last decade following optimistic recommendations during the dot-com

collapse, failure to detect fraud and accounting failures such as Enron, emails privately disparaging a

stock while publicly recommending it, and the Global Settlement (Cowen et al., 2006).

To the extent that paid-for research is unable to overcome the fundamental conflict of interest

stemming from funding research directly through the client, I predict that paid-for analysts will be

relatively more optimistic and less accurate than sell-side analysts.

Hypothesis 3: Paid-for analysts will have less accurate forecasts and more optimistic forecasts
and recommendations than those of sell-side analysts.

10
3. Sample selection and research design

3.1 Sample selection

I form a sample of fee-based research firms and their clients. I identify ten fee-based research

firms that offer paid-for analyst research; these ten firms include the top five firms that account for 73%

of revenues and 58% of firms covered in the paid-for research industry according to a recent industry

analysis (Badruswamy, 2005). I hand-collect past research reports from the web page archives of the

research firms.6 I construct a recommendation database from the research reports that includes the report

issue date; client name, exchange, and ticker; the fee-based research firm hired; whether the report was

clearly identified as an initial or final report; the recommendation rating; whether the recommendation

rating was an upgrade, reiteration, or downgrade; and the name of the analyst.

Table 1 describes my sample selection. The Full Sample includes all the clients that bought

analyst research. The CRSP/COMPU Sample is the subset of clients with available information on the

CRSP/COMPUSTAT merged database at the initiation of coverage. The ensuing analyses focus on the

CRSP/COMPU Sample because I require stock return data. As this is a new dataset, I include the Full

Sample description to provide a more complete picture of these firms and their clients. Panel A describes

the sample by research firm. The number of firms covered in the Full Sample is concentrated in the top

five firms but fairly evenly distributed amongst them. Overall, I find 619 firms (551 unique firms as some

engaged multiple research firms) bought analyst coverage. However, the two largest firms become more

predominant in the CRSP/COMPU Sample as they tend to cover relatively more firms traded on the

major exchanges. These two firms become even more prevalent in terms of number of reports, likely

6
I cross-checked the reports against Investext and I/B/E/S where available to help mitigate the possibility that the
research firms selectively chose what reports to display in their archive. It was not a straightforward task to hand
collect data from the websites. While some companies appeared to provide a complete archived list, others only
provided the archived reports for their current clients, while others provided an incomplete archived list of past
clients. I used a number of methods to capture the most complete list of clients and research reports possible
including using archives to access the fee-based research firm’s webpage at different dates in the past. My final
sample drawn from internet archives, subscriber lists, and the websites includes all firms that Investext or I/B/E/S
attributed to the fee-based firms and more.

11
reflecting more resources devoted to research as well as more clients continuing coverage for a number of

years. Panel B shows the sample by year in terms of the number of firms initiating coverage in that year

and the number of firms covered during that year. There are an increasing number of both companies

buying analyst coverage over time and the number of firms covered, indicating more firms each year are

initiating coverage than are dropping it. Panel C breaks down the sample firms by exchange listing at the

time they first paid for research. The majority of firms are from the NASDAQ and OTC Bulletin Board or

Pink Sheets. This also shows the major effect of the CRSP/COMPU filter – almost all firms from the

major exchanges are included while those traded on the OTC Bulletin Board or Pink Sheets are dropped.7

I require the following data to test my predictions: stock returns, financial statement data,

institutional holding data, capital market transactions data, and analyst following and forecast data. Stock

return data are from CRSP. Financial statement data are from COMPUSTAT. Analyst data are from

I/B/E/S. Institutional holdings are from Thomson Financial. Equity issuance, debt issuance and merger

and acquisitions activity are from the SDC database.

3.2 Research design

3.2.1 Determinants of buying analyst coverage

I first examine differences between firms that pay for analyst research and firms that do not.

Table 2 describes the characteristics of the Paid-For sample in the fiscal quarter prior to buying coverage

compared to the CRSP/COMPUSTAT universe from 2000–2006 that did not buy analyst coverage. I use

data from 2000–2006 because the frequency of paid-for research is very low prior to 2000. The unit of

observation is firm-fiscal quarter.

On average, firms that buy analyst coverage are smaller, younger, less liquid and have less sell-

side analyst coverage and institutional investor holdings. The mean (median) market value of equity of

the paid-for firms is $98 million ($44 million) contrasted with $2,840 million ($245 million) for firms that

7
This filter likely biases against finding results as stocks traded on the major exchanges are likely to have greater
information environments and paid-for analysts must attain a higher standard to provide new, valuable information.

12
did not pay for analyst coverage. Firms paying for analyst research also have more losses in the prior

quarter, greater stock return volatility, and increased use of intangible assets. This implies they have

greater business uncertainty with more difficult to predict earnings and cash flows.

Overall, firms that pay for analyst research appear markedly different from those that do not in

the univariate setting. To investigate the determinants of paid-for analyst coverage and to provide insights

into a suitable control model for subsequent analysis, I estimate the following logistic regression model

using data from 2000 – 2006:

Prob [PAID FOR = 1] = logit(β0 + β1(Log_MV) + β2(Log_Age) + β3(Loss) + β4(Std Dev of Returns) (3.1)
+ β5(Intangibles) + β6(Turnover)+ β7(Prior Returns) + β8(NIH) + B9(PIH)
+ β10(Number of SS Analysts) + β11(Previous Financing) + β12(Future Financing)
+ β13(M&A Activity) + β14(Book-to-Market) + β15(Leverage)
+ Exchange dummies + Year dummies).

The unit of observation is firm-fiscal quarter. The various and substantial economic and

regulatory changes over the sample period suggest controlling for time effects is necessary. P-values are

based on robust standard errors clustered by firm.

PAID FOR is an indicator variable = 1 if the firm initiated paid-for analyst coverage during the

fiscal quarter and 0 otherwise. Once a company pays for analyst coverage, it is dropped from the sample

for the following fiscal quarters until it drops coverage. I draw from the voluntary disclosure literature to

model a company’s decision to pay for coverage as I believe many of the incentives that drive a

manager’s decision to make a voluntary disclosure will also influence her decision to pay for coverage.

All independent variables, except for financing and M&A activity, are measured based on the

prior fiscal quarter observations.8 Log_MV is the natural log of the firm’s market value of equity at the

end of the fiscal quarter. Log_Age is the number of years the firm has been listed on CRSP. Larger firms

have richer information environments than small firms (Atiase, 1985), are more visible, and face less

structural barriers such as institutional investor constraints. Smaller, younger firms will have greater

uncertainty about future earnings, questions about the viability of the company, and higher information

8
I use the prior fiscal quarter’s observations to avoid possible endogeneity issues.

13
asymmetry between insiders and outsiders. Thus, they will have more to gain from paying for analyst

coverage to reduce this uncertainty (Lang, 1991; Botosan, 1997). I expect firm size and age to be

negatively correlated with paying for analyst coverage.

Loss is an indicator variable = 1 if the fiscal quarter’s net income before extraordinary items

(NIBE) is negative, and 0 otherwise. Losses decrease the relevance of accounting earnings as a predictor

of future earnings as losses cannot be carried forward indefinitely (Hayn, 1995). Std Dev of Returns is

the standard deviation of daily returns over the 250 days prior to the end of the fiscal quarter and with at

least 100 observations. Higher variable returns reflect greater uncertainty about future cash flows.

Intangibles is an indicator variable = 1 if the firm has a non-zero research and development expense for

the fiscal quarter.9 Firms with intangible assets have more information asymmetry between managers and

investors and more uncertainty about firm value and future cash flows than firms whose business have no

intangible assets (Barth et al., 2001). Thus firms with losses, a higher standard deviation of returns, and

intangible assets face a greater level of uncertainty when predicting future earnings and cash flows. I

expect these firms will be the ones to reap greater benefits from paying for analyst coverage and these

variables will be positively related to paying for coverage.

Turnover is the average daily number of shares traded divided by shares outstanding over the

250 days prior to the end of the fiscal quarter and with at least 100 observations. I expect low turnover

firms are more likely to pay for coverage as management has incentives to decrease the cost of capital by

increasing trading volume and liquidity through increased disclosure (Leuz and Verrecchia, 2000) and a

company is more likely to attract sell-side analysts and institutional investors in the future if it has a high

level of trading volume. Prior Returns are the stock returns over the 250 days prior to the end of the

fiscal quarter and with at least 100 observations. Better performance is related to disclosures as managers

have incentives to supply voluntary disclosures to signal good news (Verrechhia, 1983; Dye, 1985;

Miller, 2002). I expect firms with prior strong stock price performance have recently had positive news

and have an incentive to pay for analyst coverage to disseminate this news and increase their visibility.

9
I code observations in COMPUSTAT where the R&D expense are missing as 0.

14
 NIH and PIH are the number and percentage of institutional holdings as of the most recent

institutional ownership report prior to the end of the fiscal quarter.10 I expect institutional ownership to be

negatively related to the probability a company pays for research as institutional investors act as

information intermediaries (Core, 2001) and their presence will decrease the need to pay for coverage.

Number of SS Analysts is the maximum number of non paid-for analysts issuing a quarterly

earnings forecast in the fiscal quarter.11 I include the number of non paid-for analysts to control for sell-

side analyst coverage decisions. I expect Number of SS Analysts will be negatively correlated with

paying for coverage as firms with the benefits of analyst coverage have little incremental benefit to be

gained and thus little incentive to pay for coverage.

Previous Financing (Future Financing) is an indicator variable = 1 if in the 250 days prior to

the end of the fiscal quarter (100 days after the end of the fiscal quarter) the firm issues debt or equity as

reported in the SDC database, and 0 otherwise. Managers have incentives to supply voluntary disclosure

to facilitate access to the capital market and to reduce the cost of capital (Frankel et al., 1995; Lang and

Lundholm, 2000). I expect firms that will be seeking financing in the future will derive a more immediate

benefit from decreasing their cost of capital and be more likely to pay for analyst coverage. M&A

Activity is an indicator variable = 1 if the firm was engaged in mergers or acquisition activity in the

period 250 days prior to the end of the fiscal quarter to 100 days after the end of the fiscal quarter.

Mergers and acquisitions impact a firm’s future operating activities and increase the uncertainty and

complexity of understanding a firm’s future earnings and productive capabilities. I predict firms active in

mergers or acquisitions will be more likely to pay for analyst coverage.

I include Book-to-Market at the end of the fiscal quarter as companies with a high book-to-

market value may feel their stock is undervalued and be motivated to increase disclosure and investor

awareness. On the other hand, low book-to-market firms are considered growth firms with more of their

10
I assume institutional holdings are zero for any period the company is listed but there is not data on the 13f filings.
11
I use the detail forecast file on I/B/E/S to create my analyst following measure. This is important as the summary
statistics in I/B/E/S also include paid-for analysts. I assume sell-side analyst following is zero for any period the
company is listed but there are no forecasts on I/B/E/S.

15
value derived from intangible growth options. Management will be more difficult to monitor and there

will be relatively more uncertainty over future cash flow realizations in these firms (Smith and Watts,

1992) suggesting they will be more likely to buy analyst coverage because of greater potential benefits

from coverage. Finally, I also include Leverage measured as total debt (long-term plus short-term)

divided by total assets at the end of the fiscal quarter.12

3.2.2 Capital market reactions and consequences

I examine the 2-day cumulative abnormal returns (CAR) from the day of and the day after [0, +1]

the report release date.13 I include the day after as the report could have been issued after-hours. I exclude

reports missing a date; reports where the company announced earnings within two trading days of the

report date; and reports issued within two trading days of another report issued for the same company.

I estimate the following OLS regression models:

CAR = β1(Strong Buy) + β2(Buy) + β3(Hold) + β4(Sell) + β5(Not Rated) + ε (3.2)

CAR = β1(Upgrade) + β2(Reit) + β3(Downgrade) + β4(Initiations) + ε (3.3)

where Strong Buy, Buy, Hold, Sell, and Not Rated are the recommendation levels I assign to the

recommendation text.14 P-values are based on robust standard errors clustered by analyst.

I test for changes in liquidity, institutional holdings, and analyst following by examining the

difference between quarters q–4 to q–1 and quarters q to q+3 relative to the quarter the company initiated

paid-for analyst coverage. This allows the company to act as its own control and preserves seasonality

between the pre- and post-period to control for any changes in trading activity related to seasonality in

business cycles, end-of-the-year effects, and others. I require the firm to have non-missing data for the

entire period [q–4, q+3] to ensure the pre- and post-periods are not confounded by IPOs or delistings. I

12
Book-to-Market, Turnover, Std Dev of Returns, and Prior Returns are truncated at 1% and 99% to reduce the
influence of outliers.
13
I use size-adjusted abnormal returns based on assignment to size deciles.
14
Specifically, Strong Buy includes Speculative Strong Buy, Strong Buy, Strong Speculative Buy. Buy includes
Buy and Speculative Buy. Hold includes Hold, Neutral, and Speculative. Sell includes Avoid, Reduce, Sell,
Suspended, and Underperform. Not Rated is used when the rating states “Not Rated.”

16
also create a matched sample based on propensity score matching. Instead of matching dimension by

dimension, I use propensity score matching to collapse the dimensionality of the matching variables to

one and simultaneously match on several characteristics (Rosenbaum and Rubim, 1983; Caliendo 2006).

First, I estimate the logistic regression model from equation 3.1 at the firm-quarter level over

2000 – 2006 but I include industry dummies and exclude year dummies as I match firms based on both

propensity score and calendar quarter. This generates predicted probabilities (i.e. propensity scores) for

each firm-quarter observation. Firms buying coverage are matched with replacement on a one-to-one

basis with the closest non-buying control firm in terms of propensity score proximity within the same

calendar quarter as the buying firm’s initiation of coverage.15 I reduce the potential risk of a poor match

by imposing a restriction that requires the propensity score of the matched non-buying control firm to be

within 0.02 of the propensity score of the buying firm. Compared to the traditional size/industry match,

this method effectively creates a matched counterpart that controls for time, size, age, performance,

growth opportunities, leverage, intangible intensity, capital market activity, prior institutional ownership,

exchange listing, and industry. The liquidity, institutional holdings, and analyst following outcome

variables for the matched non-buying firms’ in each quarter [q–4, q+3]. The difference-in-difference

analysis then compares these adjusted variables before and after the company buys analyst coverage.

3.2.3 Relative optimism and accuracy of paid-for versus sell-side analysts

I exploit that just over 100 firms simultaneously have both paid-for and sell-side analyst coverage

on I/B/E/S to examine the relative optimism and accuracy of fee-based versus sell-side analysts.16 I use a

similar approach to Jacob et al. (1999), Clement (1999), and Cowen et al. (2006), who compare an

analyst’s optimism or accuracy relative to the mean optimism and accuracy of other analysts for that same

company, forecast period, and horizon. This controls for company and time-specific factors that influence

15
The firm that bought coverage is matched with a firm that did not buy coverage in terms of variables measured in
the quarter prior to the initiation of coverage because the independent variables in the logistic model are measured
based on the prior quarter. This match is maintained for the entire period [q-4, q+3].
16
Holding the source of the recommendations (I/B/E/S) constant ensures comparability based on adjustments
I/B/E/S makes either to the forecasts or actual values.

17
optimism or accuracy. Comparing within the same firm is also important as it controls for the potential

self-selection bias that arises when firms choose to initiate paid-for coverage.17

I first delineate three forecast horizon categories relative to the quarterly earnings announcement

date: less than 90 days, 91–180 days, and more than 180 days. I estimate the relative forecast optimism

(RFOPT) of each analyst’s forecast as:

𝑡+𝑘
𝑡+𝑘 𝐹𝑂𝑅𝐸𝐶𝐴𝑆𝑇𝑖𝑡𝑗 − 𝐹𝑂𝑅𝐸𝐶𝐴𝑆𝑇𝑖𝑡𝑡+𝑘
𝑅𝐹𝑂𝑃𝑇𝑖𝑡𝑗 = (3.4)
𝑆𝑇𝐷𝐸𝑉 (𝐹𝑂𝑅𝐸𝐶𝐴𝑆𝑇 𝑖𝑡𝑡+𝑘 )

𝑡+𝑘
where 𝐹𝑂𝑅𝐸𝐶𝐴𝑆𝑇𝑖𝑡𝑗 is analyst j’s forecast of company i's quarterly earnings for period t+k, where the

analyst forecast is made in the horizon category t. The forecast is then compared to the mean forecast of

all analysts’ forecasts for the same company’s (i) quarterly earnings for the same period (t+k) made

within the same horizon category (t). This relative forecast is then scaled by the standard deviation of all

earnings forecasts for the same company, forecast period, and horizon category.

I define relative forecast accuracy (RFACC) in an analogous manner as:

𝑡+𝑘
𝑡+𝑘 |𝐹𝐸𝑖𝑡𝑗 |−|𝐹𝐸𝑖𝑡𝑡+𝑘 |
𝑅𝐹𝐴𝐶𝐶𝑖𝑡𝑗 = (3.5)
𝑆𝑇𝐷𝐸𝑉 (|𝐹𝐸𝑖𝑡𝑡+𝑘 |)

𝑡+𝑘
where 𝐹𝐸𝑖𝑡𝑗 is the forecast error (actual – forecast) of analyst j’s forecast of company i's quarterly

earnings forecast period t+k, where the analyst forecast is made in the horizon period t. For the same

company, forecast period, and horizon, the absolute value of this measure is compared to the mean

absolute forecast errors of all analysts with this difference deflated by the standard deviation of the

absolute forecast errors.

For each company/forecast period/horizon, I include only the first forecast made by an analyst if

there are multiple forecasts made by an analyst and require at least one forecast from a paid-for analyst

and one from a sell-side analyst. Finally, identical forecasts for a given company/period/horizon are

excluded as relative optimism or accuracy is undefined.

17
For example, a predominance of buy recommendations may reflect that only companies with favorable outlooks
are willing to pay for research.

18
 I test whether paid-for analysts are relatively optimistic or accurate versus sell-side analysts by

estimating the following models used in Groysberg et al. (2007):

𝑡+𝑘
𝑅𝐹𝑂𝑃𝑇𝑖𝑡𝑗 = 𝛼 + 𝛽1 𝑃𝐴𝐼𝐷 𝐹𝑂𝑅𝑗𝑡 + 𝛽2 𝐻𝑂𝑅𝐼𝑍𝑂𝑁𝑖𝑗𝑡+𝑘
𝑡 + 𝛽3 𝐴𝐸𝑋𝑃𝑗𝑡 + 𝜀𝑖𝑗𝑡 (3.6)

𝑡+𝑘 𝑡+𝑘
𝑅𝐹𝐴𝐶𝐶𝑖𝑡𝑗 = 𝛼 + 𝛽1 𝑃𝐴𝐼𝐷 𝐹𝑂𝑅𝑗𝑡 + 𝛽2 𝐻𝑂𝑅𝐼𝑍𝑂𝑁𝑖𝑗𝑡 + 𝛽3 𝐴𝐸𝑋𝑃𝑗𝑡 + 𝛽4 𝐴𝐶𝑂𝑀𝑗𝑡 (3.7)

+𝛽5 𝐴𝑆𝑃𝐸𝐶𝑗𝑡 + 𝛽6 𝐹𝑆𝐼𝑍𝐸𝑗𝑡 + 𝛽7 𝐹𝑆𝑃𝐸𝐶𝑖𝑗𝑡 + 𝜗𝑖𝑗𝑡

where PAID FOR is an indicator variable = 1 for a paid-for analyst at time t and 0 otherwise. I control for

horizon differences within the horizon category by including HORIZON as an independent variable

defined as the number of days between the forecast issue date and the related earnings announcement

date. I control for analyst experience, AEXP, by using the log of the number of quarters that an analyst

has an earnings estimate on I/B/E/S up to and including that quarter. The other control variables used in

the relative accuracy model are the number of companies covered by the analyst (ACOM) defined as the

number of other companies an analyst provides earnings estimates for during the calendar year; the

industry specialization of the analyst (ASPEC) defined as the percentage of companies out of the total

other companies covered by the analyst during the calendar year that are in the same I/B/E/S industry

classification as the forecast firm’s industry; firm size (FSIZE) defined as the number of other analysts

working at the forecasting analyst’s firm during the calendar year; and firm industry specialization

(FSPEC) defined as the percentage of other analysts working at the forecasting analyst’s firm who are

covering a firm in the same industry as the forecast firm’s industry.

I measure recommendation optimism using all available stock recommendations by paid-for and

sell-side analysts over 2000–2006. I follow the approach used in Groysberg et al. (2007) and estimate the

ordered logit model:

REC = f(PAID FOR, NUM_BUY, NUM_HOLD, NUM_SELL, AEXP) (3.8)

where REC is the analyst’s recommendation coded 2 for a Strong Buy or a Buy, 1 for a Hold, and 0 for a

Sell. I include the number of strong buy/buy, hold and sell recommendations (NUM_BUY,

NUM_HOLD, NUM_SELL) made by other analysts for the same firm and calendar quarter to control for

19
company and time period effects.18 I also control for analyst experience. In the optimism and accuracy

regression models, I use robust standard errors clustered by analyst.

4 Data and empirical results

4.1 Determinants of buying analyst coverage

Table 3 presents the results from estimating the logistic regression model based on potential

determinants for a company’s decision to buy analyst research. Columns 3 displays the Z-coefficients

from the regression while column 4 interprets the economic significance of the results in terms of the

percentage change in the odds of buying analyst research for a standard deviation increase in the

independent variable.

I find that the likelihood of buying analyst research is positively associated with the standard

deviation of returns, presence of intangible assets, and M&A activity; and negatively associated with size,

age, and book-to-market. This is consistent with the voluntary disclosure literature that argues firms with

high information asymmetry between insiders and outsiders; and greater uncertainty about future

earnings, cash flows, and productive capabilities will have more to gain from reducing this uncertainty

(Lang, 1991; Botosan, 1997). Sell-side analysts are less likely to cover firms where future earnings and

cash flows are more difficult to predict as it requires more effort (Bhushan, 1989; Barth et al., 2001).

Thus, these firms are less likely to attract sell-side coverage and turn to paying for analyst coverage.

Firms are also more likely to initiate paid-for analyst coverage when they have had a recent

strong performance in stock price. This suggests managers look to purchase analyst coverage when they

have had recent positive news and are looking to increase their visibility.

The likelihood of paying for analyst research is negatively associated with the number and

percentage of institutional investors and prior sell-side analyst coverage. This is consistent with their

information intermediary role reducing the benefit of and the need for paid-for analyst coverage. In

18
I group strong buy and buy recommendations into the same category as many sell-side brokerage houses during
this period moved to a 3-tier recommendation system following the Global Settlement (Kadan et al., 2008).

20
particular, a one standard deviation increase in these measures is related to between a 39% and 83%

decrease in the likelihood of buying research.

The results show paying for analyst research is unrelated to previous financing needs but is

positively related to accessing the capital market for financing in the near future. I find firms with high

turnover are less likely to pay for analyst research.

I also find that firms listed on the AMEX or NASDAQ exchange (relative to NYSE) are more

likely to pay for analyst research. The greater probability of firms listed on AMEX and NASDAQ paying

for coverage relative to NYSE likely reflects structural barriers to analyst coverage as NYSE has a greater

reputation and visibility. Prior quarter losses and leverage were insignificantly related to paying for

analyst coverage.

4.2 Capital market reactions and consequences

Table 4, Panel A shows the number of reports within each recommendation level and change in

recommendation level. Consistent with sell-side analyst recommendations, the paid-for recommendations

tend to be predominately buys with about 87% of recommendations in the strong buy or buy category.

I first break out initiations of coverage separately as initiations have incremental impact in sell-

side recommendations (Irvine, 2003). However, I also split the sample to investigate whether the decision

to buy coverage sends a signal to the market about future good performance. In this case, investors could

react to the signal of initiating paid-for coverage yet not be reacting to the information in the report. I

expect two possible outcomes if investors only react to a report’s signal of future performance and not to

the information in the report: (1) the CAR of the initiation will not vary by recommendation, and (2)

subsequent changes in recommendation level will be insignificant.

Table 4, Panel B presents investors’ reactions to the initial reports. The CAR of all initial reports

is 1.7%. I find reports that are initiated with Strong Buy and Buy recommendations are significantly

associated with a CAR of 3.06% and 1.36%. Hold and Not Rated initiations are not statistically different

from zero. The CARs of Strong Buy, Buy and Hold initiations are also statistically different from each

21
other. This implies that investors are reacting to the information content contained in the level of the

recommendation in addition to the initiation of coverage.

Table 4, Panel C, shows that subsequent changes in recommendation level are associated with

abnormal returns: CARs are significantly associated with Upgrades (1.32%), Reiterations (0.47%),

Downgrades (–1.10%), and Initiations (1.74%). Additionally, table 4, panel D, shows that CARs are also

significantly associated with Strong Buys (1.19%), Buys (0.46%), and Holds (–0.80%) while the CARs of

Sells and reports with no recommendation are insignificantly different from zero. Overall, the CARs of

initiations, initiation level, recommendation changes, and recommendation level are consistent with

intuition and the sell-side literature suggesting that investors view paid-for research as credible and react

to the information contained in the report.

Table 5 shows the results for changes in liquidity, institutional holding, and number of analysts

before and after a company pays for analyst coverage. Columns 3 and 4 show the mean quarterly level of

the variable over the four quarters before hiring an analyst [q–4, q–1] and the four quarters after hiring an

analyst [q, q+3]. Columns 5 and 6 show the difference in unadjusted means and associated p-value.

Column 7 shows the difference in means adjusted for the quarterly difference between the paid-for firm

and its matched control firm. Column 8 shows the associated p-value. I define new two variables: Spread

Fqtr and Turnover Fqtr are the mean daily bid-ask spread deflated by price and the mean daily shares

traded divided by shares outstanding both measured over the fiscal quarter. I require both variables to

have at least 25 trading days in the fiscal quarter. Consistent with the sell-side literature (Irvine, 2003), I

find improvements in liquidity after the initiation of analyst coverage. The mean closing bid-ask spread

deflated by price decreased 25% from 4.35% to 3.27%. The increase in daily turnover is not statistically

significant. I find that the average number of institutional investors per fiscal quarter increases 33% from

12.9 to 17.1. The percentage of firm stock owned by institutions also increases from 11.8% to 14.3% (a

21% increase). The number of sell-side analysts following the firm also shows a significant increase and

nearly doubles from an average of 0.31 per quarter to 0.55 per quarter. This implies that paying for

22
coverage may act as a stepping stone to sell-side coverage by increasing liquidity, encouraging

institutional investors, and decreasing the initial information acquisition cost.

However, after comparing the differences against a control sample of firms matched by

propensity score and calendar quarter, the difference for each outcome variable is attenuated and only the

increases in the number of institutional investors and sell-side analysts remain statistically significant. I

explore this further in section 5.

4.3 Relative optimism and accuracy of paid-for versus sell-side analysts

Table 6 presents results from the sample of firms where the companies had both sell-side and

paid-for analysts forecasting the same earnings within the same horizon category. The results indicate that

paid-for earnings forecasts are not significantly more optimistic than those made by sell-side analysts.

However, paid-for analysts appear to be less accurate than their sell-side counterparts in terms of forecast

accuracy even after controlling for differences in horizon, analyst characteristics and analyst firm

effects.19 Paid-for analysts also issued relatively optimistic recommendations over 2000–2006.

Overall, the results imply that paid-for analysts are capable of providing credible

recommendations that have information content for investors and helping firms capture potential benefits

from analyst coverage such as increases in liquidity and institutional investor ownership. However,

critics’ concerns are not without foundation as paid-for analysts tend to issue less accurate forecasts and

more optimistic recommendations than their sell-side counterparts.

5. Credibility and fee-based firms

Conflicts of interest are inherent in paid-for analyst coverage – they are a constant situational

factor of paid-for research that influences credibility. Mercer (2004) argues that despite situational

19
The coefficient on PAID FOR is qualitatively unaffected by the inclusion or exclusion of the other control
variables suggesting the results are independent of differences in analyst experience, analyst specialization, or
brokerage size between paid-for and sell-side analysts.

23
factors, firms can affect credibility through other factors that they have control over such as internal

policies and organizational structure. Fee-based firms vary in terms of their business operations and

policies such as separation of research from other functional areas and restrictions on stock holding and

trading. To the extent that fee-based firms have policies or business operations that make them less likely

to have conflicts of interest, I expect there to be a stronger stock market reaction to their reports; greater

increases in liquidity, sell-side coverage, and institutional ownership for their clients; and more accurate

and less optimistic forecasts and recommendations.

Hypothesis 4: High credibility firms will differ from low credibility in terms of:
a: Greater abnormal stock return on the release of the paid-for research report.
b: Greater increase in liquidity, sell-side analyst coverage, and institutional investor
ownership for their clients after the initiation of paid-for coverage.
c: More accurate and less optimistic forecasts and recommendations

Table 7, Panel A, presents my categorization of fee-based firms into two subsamples (high and

low credibility) based on ex ante firm policies and business operations. First, I determine whether the fee-

based firm has lines of business in addition to its research department and if its policies permit the firm to

have a business relationship besides research coverage with its clients. Of the 10 firms, Taglich Brothers

and Researchstock.com explicitly state they may perform investment banking or other services for their

clients. Cohen Independent Research, EquityNet Research, Howlett Research also may perform and

solicit consulting or other services from their clients. The other five companies do not have brokerage,

investment banking or business consulting operations. The sell-side literature considers analysts whose

brokerage firms have other business relationships with companies under coverage as more likely to suffer

from conflicts of interest that bias their reports and recommendations (Lin and McNichols, 1998;

Michaely and Womack, 1999; Dechow et al., 2000).

Second, I consider whether the fee-based firm allows its employees or analysts to hold and/or

trade in securities issued by clients. The same five firms that had potentially conflicting lines of business

also allow the fee-based firm, its employees, and/or its analysts to trade in the client’s shares. The other

five firms do not allow trading in client’s securities. Permitting trading in client’s securities gives analysts

24
additional incentives to produce buy/sell recommendations based on their holdings instead of independent

evaluation. Overall, I classify the five firms with other related lines of business and that permit trading in

clients’ securities as “Low Credibility” and the five firms without other related lines of business and that

prohibit holding or trading in clients’ securities as “High Credibility.”

The two largest fee-based firms in terms of clients covered and research reports (Dutton

Associates and Taglich Brothers) are divided between the two categories. I define an indicator variable

HIGH CREDIBILITY that = 1 if the fee-based firm is a “High Credibility” firm, and 0 otherwise.

Table 7, Panel B, presents the abnormal 2-day returns around the release of analyst reports

partitioned by credibility. I subset the initiations into only two categories (Strong Buy/Buy and Hold)

because of the limited number of observations per recommendation level when partitioning by credibility.

The results for the high-credibility firms are similar to the previous results in Table 4. The CAR of all

initial reports by high credibility firms is 2.6%, which is significantly higher than the CAR of 0.9% for

low credibility firms. This difference extends to initial reports issued at a Strong Buy/Buy

recommendation where the CAR of 2.5% for high credibility firms is significantly higher than the 0.9%

for low credibility firms. The CAR of Hold initiations is not significantly different from zero for both the

high credibility and the low credibility firms.

The CARs of subsequent changes in recommendation level of high credibility firms are consistent

with intuition and the sell-side literature and statistically significant except for downgrades. For low

credibility firms, the CARs of upgrade, reiterations, and downgrades are all insignificantly different from

zero. The difference between the high and low credibility firms is statistically significant for upgrades,

reiterations, and initiations. Additionally, the CARs of high credibility firms are positive and statistically

significant for Strong Buys and Buy while the CARs of low credibility firms are negative and statistically

significant for Holds and Sells; although the difference between high and low credibility firms is only

significant for the Sell recommendations. Overall, this suggests that while investors view paid-for

research as credible, they also distinguish between paid-for research firms whose ex ante policies either

mitigate or exacerbate potential conflicts of interest.

25
 Table 7, Panel C, shows changes in liquidity, institutional holding, and number of sell-side

analysts before and after a company pays for analyst coverage for both high and low credibility clients.

There is a marked difference between the high and low credibility clients from the quarters before, [q–4,

q–1], to the quarters after [q, q+3] hiring an analyst. High credibility clients experience significant

decreases in bid-ask spread and increases in turnover, institutional investor holdings, and the number of

sell-side analysts. These results hold for both the within-firm and the difference-in-difference analyses

using a matched control firm. The results for the high credibility clients are stronger than for the pooled

sample in Table 5 where turnover was not significant in either analysis and changes in percentage of

institutional holdings and bid-ask spread were not significant in the difference-in-difference analysis. In

contrast, low credibility clients had significant within-firm differences in all variables except turnover but

none of the variables were significant after controlling for time, industry, and other effects by matching

against a control firm. Additionally the difference between high credibility and low credibility clients is

statistically and economically significant in the institutional ownership and analyst following measures.

Table 7, Panel D presents the relative optimistic and accuracy of paid-for analysts split by high

and low credibility. The results indicate that the earnings forecasts and recommendations made by high

credibility fee-based firms are not significantly less optimistic or more accurate than those made by low

credibility firms. On the contrary, low credibility firms produce less optimistic forecasts than high

credibility firms.

Overall, there is a difference between paid-for research firms based on ex ante firm policies that

may enhance or exacerbate the credibility of the fee-based firm. Investors appear to consider these factors

when assessing the credibility and information content of paid-for analysts’ recommendations and the

initiation of coverage. High credibility paid-for analysts produce recommendations that elicit CARs

consistent with intuition and typically of a stronger magnitude than the CARs from the recommendations

of low credibility analysts. The difference between high and low credibility analysts appears most

pronounced for incentive consistent situations such as issuing Strong Buy, Buy, Upgrade or Reiteration

26
recommendations. High credibility clients are also able to capture potential benefits of analyst coverage

such as increased liquidity, institutional ownership, and sell-side coverage.

6. Robustness tests

I apply a number of robustness tests to the various analyses. In the before and after analysis, five

firms change exchanges over the [q–4, q+3] period surrounding hiring an analyst. It is possible that this

may influence changes in liquidity, institutional ownership, and analyst following, although the change is

between major exchanges and the potential impact is relatively smaller than changes between an over-the-

counter exchange and a major exchange. The results are robust to excluding these five firms.

Although my control sample matched by propensity score and calendar quarter should mitigate

concerns, I cannot rule out the possibility that managers decide to simultaneously initiate paid-for

coverage as part of an overall disclosure strategy. Thus, it is difficult to attribute the changes in liquidity,

institutional ownership, and sell-side following after the initiation of paid-for coverage to the coverage or

to changes in disclosure. I investigate this possibility and control for its effects using company-issued

guidelines (CIG) data from First Call. I define a variable Num_mef as the number of management

forecasts a company issues for any period within each fiscal quarter period.20 There is evidence that the

number of management earnings forecasts is higher after initiating paid-for coverage; however, this

difference disappears after matching against a control sample. Controlling for the number of management

forecasts does not affect the conclusions in either the within-firm or difference-in-difference analyses in

Table 5. Splitting the sample by high and low credibility, I find that the high credibility clients show no

change in the number of management forecasts after initiating coverage in either the within-firm or

difference-in-difference analyses. Controlling for the number of management forecasts has no effect on

20
I assume the number of management forecasts is zero for any period the company is listed but there is not data on
the CIG database. The results are also the same using the binary variable Any_mef which = 1 if the company issued
management forecasts within the fiscal quarter period and 0 otherwise. Ajinkya, Bhojraj, and Sengupta (2005)
conclude that the CIG database is a comprehensive source of management forecasts after performing two small
sample tests (in 1997 and 2000) and finding about 2.5 times more CIG forecasts compared to a similar keyword
search in Factiva (formerly Dow Jones News Retrieval Service).

27
the results for the high credibility clients. On the other hand, low credibility clients show a significant

increase in the number of management forecasts in the unadjusted differences; but no significant increases

when matched against a control firm. After controlling for the number of forecasts for low credibility

clients, the unadjusted change in institutional ownership and sell-side coverage become insignificant in

Table 7; only the unadjusted change in bid-ask spread remains statistically significant. All variables in the

difference-in-difference analyses for the low credibility clients remain insignificant.

In the analyst optimism sample, I require at least two analysts per company/fiscal period/horizon

category to compute the relative accuracy and optimism measures. While two analysts is the minimum

necessary, I also run the analyses with a restriction of at least three analysts per company/fiscal

period/horizon category. The results are robust to this restriction. The coefficient on PAID FOR in the

RFOPT model flips sign to become positive as predicted but is still statistically insignificant. The

credibility analysis is qualitatively unchanged. In the optimism sample, I group the recommendations into

three categories (Strong Buy OR Buy, Hold, Sell) to reflect that many firms over that period moved to a

three-tier recommendation system. I also estimate the model separating Strong Buy from Buy. The

coefficients on PAID FOR and HIGH_CRED maintain their sign and statistical significance while the

coefficient on Strong Buy is significantly positive and the coefficient on Buy is significantly negative.

7. Summary

Using a sample of companies that bought analyst coverage during 1999–2006, I provide evidence

on the determinants of the decision to hire an analyst, the capital market reaction to paid-for analyst

reports, the optimism and accuracy of paid-for versus sell-side analysts, and the importance of credibility

within this industry. My empirical analysis shows that firms with weaker information environments,

lower visibility, higher information asymmetry, and greater uncertainty about future performance are

more likely to buy analyst coverage. This is consistent with the notion that these firms have the most to

gain from reducing this uncertainty and exposing themselves to the monitoring effect of analyst coverage

28
but are less likely to attract sell-side coverage because future earnings and cash flows are difficult to

predict and brokerage firms cannot justify covering small firms with low liquidity.

I find investor reaction to paid-for reports as measured by 2-day CARs is economically large and

in the direction consistent with intuition and the extant sell-side literature. The reaction is not limited to

the initiation sample suggesting that investors are not merely reacting to a signal of better future

performance from the initiation of paid-for coverage but to the information content in the report. My

results also show an increase in liquidity, institutional investor holding, and sell-side analyst following

after the initiation of paid-for coverage.

However, paid-for coverage is not a perfect substitute for sell-side coverage or unconditionally

valuable. Using a sample of firms who simultaneously have both paid-for and sell-side coverage in

I/B/E/S, I find that paid-for analysts issue less accurate forecasts and more optimistic recommendations.

The ex ante policies of fee-based firms that reduce potential conflicts of interest and enhance credibility

also play a key role. The stock market reactions to the issuance of paid-for reports from high credibility

fee-based firms are stronger than for low credibility firms. Only high credibility clients show evidence of

increases in liquidity, institutional ownership, and sell-side coverage after the initiation of coverage.

Finally, I also document that the growth in the paid-for research industry has methodological

consequences for future researchers. Paid-for analyst estimates and recommendations are included in

IBES, Zacks, Multex and First Call. I find 400–450 unique companies in I/B/E/S with recommendations

and earnings estimates during my sample period; the majority of which would not have analyst coverage

if it was not bought. Research designs that incorporate analyst following or consensus forecasts may need

to identify whether the analysts covering the firm are paid-for or not.

Or course, I recognize that my findings are based on a small number of fee-based firms; although

my sample includes the top five firms in the industry and I believe I have collected a comprehensive

database of their clients and research reports. A similar concern is the two largest fee-based firms

dominate the sample in terms of number of reports; however, this is less of an issue in the determinants

analysis, the initiations sample, and examining changes after the initiation of paid-for coverage as the

29
number of clients per research firm is more balanced. Also, in the high and low credibility partition each

subset contains one of these firms.

This evidence raises questions for future research and the analyst industry. For researchers, this

study provokes questions about the effectiveness of various forms of analyst research. Regulators view

the lack of sell-side analyst research as a major problem yet solutions such as paid-for coverage are

controversial. The AMEX and NASDAQ exchanges have a high percentage of companies with little or no

analyst coverage and have begun to explore other mechanisms. AMEX is investigating working with an

investment bank to provide information on its listed companies (Friedlander, 2005), while NASDAQ

previously explored a joint venture with Reuters called the Independent Research Network (IRN). My

findings also raise interesting questions about which are the key factors investors assess when evaluating

the differences in credibility among fee-based companies.

For the analyst industry, this study raises questions about whether the traditional equity research

model is the only viable mechanism. Under this structure, about 60% of publicly listed firms have no

analyst coverage. My results imply paid-for coverage may be a model by which companies can achieve

the benefits of analyst coverage despite lacking the characteristics to be attractive to sell-side analysts;

however, I caution that paid-for coverage is neither a perfect substitute for sell-side coverage nor

insensitive to the perceived credibility of the fee-based research firm. These findings reinforce the SEC

Advisory Committee’s endorsement of paid-for research and are consistent with the notion that paid-for

analyst coverage can play a valuable role in the capital markets especially in helping the development of

small and medium cap companies.

30
References

Ajinkya, B., Bhojraj, S., Sengupta, P., 2005. The association between outside directors, institutional investors, and
the properties of management earnings forecasts. Journal of Accounting Research 43, 343–376.
Asquith, P., Mikhail, M., Au, A., 2005. Information content of equity analyst reports. Journal of Financial
Economics 75, 245–282.
Atiase, R., 1985. Predisclosure information, firm capitalization, and security price behavior around earnings
announcements. Journal of Accounting Research 23, 21–36.

Badruswamy, S., 2005. A report on issuer-paid research industry. Aranca US Inc.

Barth, M., Hutton, A., 2004. Analyst earnings forecast revisions and the pricing of accruals. Review of Accounting
Studies 9, 56–96.

Barth, M., Kasznik, R., McNichols, M., 2001. Analyst coverage and intangible assets. Journal of Accounting
Research 39, 1–34.

Bhushan, R., 1989. Firm characteristics and analyst following. Journal of Accounting and Econometrics 121, 255–
274.

Botosan, C., 1997. Disclosure level and the cost of equity capital. Accounting Review 72, 323–350.

Brennan, M., Subrahmanyam, A., 1995. Investment analysis and price formation in securities markets. Journal of
Financial Economics 38, 361–381.

Brennan, M., Tamarowski, C., 2000. Investor relations, liquidity, and stock prices. Journal of Applied Corporate
Finance 12, 26–37.

Brooks, G., 2006. IR to the rescue. Equities Magazine, Spring 2006, 30–34.

Brown, P., Foster, G., Noreen, E., 1985. Security analyst multi-year earnings forecasts and the capital market.
American Accounting Association, Sarasota, FL.

Bushee, B., Miller,G., 2007. Investor relations, firm visibility, and investor following. Working paper, University of
Pennsylvania.

Bushee, B., Noe, C., 2000. Corporate disclosure practices, institutional investors, and stock return volatility. Journal
of Accounting Research 38, 171–202.

Caliendo, M., 2006. Microeconometric evaluation of labour market policies. Springer-Verlag, Berlin Heidelberg.

Chung, K., Jo, H., 1996. The impact of security analysts’ monitoring and marketing functions on the market value of
firms. Journal of Financial and Quantitative Analysis 31, 493–512.

Clement, M., 1999. Analyst forecast accuracy: do ability, resources, and portfolio complexity matter? Journal of
Accounting and Economics 27, 285–303.

Core, J., 2001. A review of the empirical disclosure literature: discussion. Journal of Accounting and Economic 31,
441–456.

Cowen, .A., Groysberg, B., Healy, P., 2006. What types of analyst firms are more optimistic? Journal of Accounting
and Economics 41, 119–146.

31
Dechow, P., Hutton, A., Sloan, A., 2000. The relation between analysts’ forecasts of long-term earnings growth and
stock performance following equity offerings. Contemporary Accounting Research 17, 1–32.

Diamond, D., Verrecchia, R., 1991. Disclosure, liquidity, and the cost of capital. Journal of Finance 66, 1325–1355.

Dye, R., 1985. Disclosure of nonproprietary information. Journal of Accounting Research 23, 123–145.

Frankel, R., McNichols, M., Wilson, P., 1995. Discretionary disclosure and external financing. Accounting Review
70, 135–150.

Friedlander, J., 2005. Better to shine a light? Investment dealers’ digest, March 28, 2005.

Givoly, D., Lakonishok, J., 1979. The information content of financial analysts forecasts of earnings: some evidence
on semi-strong efficiency. Journal of Accounting and Economics 2, 165–186.

Groysberg, B., Healy, P., Chapman, C., Shanthikumar, D., Gui, Y., 2007. Do buy-side analysts out-perform the sell-
side? Working paper, Harvard Business School.

Hassett, J., Mahoney, B., 2001. Research for sale. Shareholder Value, 28–33.

Hayn, C., 1995. The information content of losses. Journal of Accounting and Economics 20, 125–153.

Healy, P., Palepu, K., 2001. Information asymmetry, corporate disclosure, and the capital markets: a review of the
empirical disclosure literature. Journal of Accounting and Economics 31, 405–440.

Irvine, P., 2003. The incremental impact of analyst initiation of coverage. Journal of Corporate Finance 9, 431–451.

Jacob, J., Lys, T., Neale, M., 1999. Expertise in forecasting performance of security analysts. Journal of Accounting
and Economics 28, 51–82.

Kadan, O., Madureira, L., Wong, R., Zach, T., 2008. Conflicts of interest and stock recommendations – the effects
of the global settlement and related regulations. Washington University working paper.

Lang, M., Lins, K., Miller, P., 2004. Concentrated control, analyst following, and valuation: do analysts matter most
when investors are protected least? Journal of Accounting Research 42, 589–623.

Lang, M., 1991. Time-varying stock price response to earnings induced by uncertainty about the time-series process
of earnings. Journal of Accounting Research 29, 229–257.

Lang, M., Lundhom, R., 2000. Voluntary disclosure and equity offerings: reducing information asymmetry or
hyping the stock?, Contemporary Accounting Research 17, 623–662.

Leone, M., 2004. Sell-side analysts dropping small-caps. CFO.com, April 28, 2004, 1.

Leuz, C., Verrecchia, R., 2000. The economic consequences of increased disclosure. Journal of Accounting
Research 38, 93–124.

Lin,H., McNichols, M., 1998. Underwriting relationships, analysts’ earnings forecasts and investment
recommendations. Journal of Accounting and Economics 25, 101–127.

Matsumoto, D., 2002. Management’s incentives to avoid negative earnings surprises. The Accounting Review 77,
483–514.

McNichols, M., O’Brien, P., 1997. Self-selection and analyst coverage. Journal of Accounting Research 35, 167–
199.

32
Mercer, M., 2004. How do investors assess the credibility of management disclosures? Accounting Horizons 18,
185–196.

Merton, R., 1987. A simple model of capital markets equilibrium with incomplete information. The Journal of
Finance 42, 483–510.

Metzger, C., 2002. The cost of coverage/credibility. Investor Relations Update, April 2002, 1–7.

Metzger, C., 2003. A year later: the cost of coverage/credibility. Investor Relations Updates, April 2003, 1–7.

Michaely, R., Womack, K., 1999. Conflict of interest and the credibility of underwriter analyst recommendations.
Review of Financial Studies 12, 653–686.

Miller, G., 2002. Earnings performance and discretionary disclosure. Journal of Accounting Research 40,173–204.

Moyer, C., Chatfield, R., Sisneros, P., 1989. Security analyst monitoring activity: agency costs and information
demands. Journal of Financial and Quantitative Analysis 24, 503–512.

O’Brien, P., Bhushan, R., 1990. Analyst following and institutional ownership. Journal of Accounting Research 28,
55–76.

Rosenbaum, P., Rubin, D., 1983. The central role of the propensity score in observational studies for causal effects.
Biometrika 70, 41–55.

SEC, 2006. Final report of the advisory committee on smaller public companies to the US securities and exchange
commission, April 23, 2006.

Smith, C., Watts, R., 1992. The investment opportunity set and corporate financing, dividend, and compensation
policies. Journal of Financial Economics 32, 763–792.

Verrecchia, R., 1983. Discretionary disclosure. Journal of Accounting and Economics 12, 365–380.

33
Table 1
Sample selection

Panel A: Sample firms and number of reports by fee-based research firm

Full Sample CRSP/COMPU Sample

Firms Percent Firms Percent Reports Percent
Cohen Independent Research 66 10.7% 34 11.7% 76 2.3%
EquityNet Research 14 2.3 4 1.4 6 0.2
Fundamental Research 6 1.0 1 0.3 4 0.1
Howlett Research 13 2.1 0 0.0 0 0.0
Investrend 90 14.5 18 6.2 80 2.4
J.M. Dutton Associates 177 28.6 97 33.4 1,938 59.1
Researchstock.com 19 3.1 11 3.8 105 3.2
SISM Research 13 2.1 1 0.3 10 0.3
Spelman Research Associates 53 8.6 11 3.8 73 2.2
Taglich Brothers 168 27.1 113 39.0 986 30.1
Final Sample 619 100.0% 290 100.0% 3,278 100.0%

Unique firms 551 258

Panel B: Sample firms initiations and coverage by year

Full Sample CRSP/COMPU Sample
Year N Initiated Percent N Covered N Initiated Percent N Covered
1999 23 3.7 23 8 2.8 8
2000 72 11.6 91 53 18.3 60
2001 60 9.7 114 48 16.6 83
2002 65 10.5 120 34 11.7 81
2003 82 13.2 151 41 14.1 91
2004 120 19.4 211 42 14.5 108
2005 103 16.6 243 36 12.4 105
2006 94 15.2 234 28 9.7 104
619 100.0% 290 100.0%

Panel C: Sample firms by exchange listing

Full Sample CRSP/COMPU Sample
Stock exchange N Percent N Percent
NYSE 7 1.1% 7 2.4%
AMEX 84 13.6 84 29.0
NASDAQ 201 32.5 199 68.6
OTC Bulletin Board & Pink Sheets 327 52.8 0 0.0
619 100.0% 290 100.0%

The Full Sample includes all US companies hiring an analyst from one of the fee-based research firms listed. The CRSP/COMPU
Sample is the Full Sample less those firms that did not have data available on the CRSP/COMPUSTAT merged database. Panel
B and Panel C partition the sample by year and exchange listing at the time the company first paid for analyst coverage from a
specific fee-based research firm.

34
Table 2
Descriptive statistics for the fiscal quarter prior to paying for analyst coverage
Paid-For CRSP/COMPUSTAT
Variable Mean Median SD Mean Median SD
###
Market value of equity ($ million) 98.35*** 43.94 251.82 2,840.14 244.58 14,253.06
Firm Age (quarters) 9.41*** 8.00### 7.95 13.94 10.00 13.68
Loss 0.55*** 1.00### 0.50 0.32 0.00 0.47
Std Dev of Returns 0.06*** 0.06### 0.03 0.04 0.03 0.03
Intangibles 0.52*** 1.00### 0.50 0.30 0.00 0.46
Turnover (%) 0.53*** 0.31 0.60 0.69 0.36 0.49
Spread (%) 3.86*** 2.63### 3.36 2.08 1.01 3.36
Prior Returns (%) 25.66* –0.73 95.82 13.33 5.71 69.93
NIH 13.24*** 7.00### 21.16 81.73 36.00 127.62
PIH (%) 11.84*** 5.15### 16.24 35.43 28.45 30.67
Number of SS Analysts 0.32*** 0.00### 1.19 3.39 1.00 5.42
Previous Financing 0.05*** 0.00### 0.22 0.21 0.00 0.41
Future Financing 0.05*** 0.00## 0.21 0.09 0.00 0.28
M&A Activity 0.19*** 0.00### 0.39 0.26 0.00 0.44
Book-to-Market 0.60 0.41### 0.64 0.66 0.53 4.02
Leverage 0.19* 0.12## 0.22 0.21 0.16 0.25
The Paid-For sample consists of 252 firms that initiated paid-for analyst coverage during 2000–2006. Variables in the Paid-For Sample are from the fiscal quarter prior to initiating
paid-for analyst coverage for the first time except for Financing and M&A Activity. All other firms from the CRSP/COMPUSTAT database are assumed to not have paid for
analyst coverage during that period and are in CRSP/COMPUSTAT Sample. The unit of observation is firm-fiscal quarter. Not all firms have complete data to calculate all
variables, so each variable’s results are based on complete observations only for that variable. Market value of equity is the firm’s market value of equity at the end of the fiscal
quarter. Firm Age is the number of years since first listed on CRSP. Loss is an indicator variable = 1 if the net income before extraordinary items (NIBE) is negative, and 0
otherwise. Std Dev of Returns is the standard deviation of daily returns over the 250 days prior to the fiscal quarter end. Intangibles is an indicator variable = 1 if the firm has a
non-zero R&D expense in the fiscal quarter and 0 otherwise. Spread is the average daily closing ask minus closing bid scaled by price over the 250 days prior to the fiscal quarter
end. Turnover is the average daily number of shares trades divided by common shares outstanding over the 250 days prior to the fiscal quarter end. Prior Returns is the buy-and-
hold stock return calculated over the 250 days prior to the fiscal quarter end. I require at least 100 days of trading data for Spread, Turnover, Std Dec of Returns, and Prior
Returns. NIH and PIH are the number and percentage of institutional holdings based on the most recent report issued prior to the end of the fiscal quarter. Number of SS
Analysts is the maximum number of unique sell-side analysts issuing an earnings forecast in the fiscal quarter. I assume that analyst coverage, institutional holdings and
institutional percentage of ownership are zero for any period when the company is listed on an exchange but there is no data available on analyst coverage and institutional
holdings. Previous Financing ( Future Financing) is an indicator variable = 1 if in the last 250 days (future 100 days) relative to the fiscal quarter end, the company issued equity
or debt as captured by the SDC database. M&A Activity is an indicator variable = 1 if the firm engaged in mergers or acquisitions in the prior 250 days or future 100 days relative
to the fiscal quarter end as captured by the SDC database. Book-to-Market is measured as of the end of the fiscal quarter. Leverage is short-term and long-term debt divided by
total assets. ***,**,* ( #, ##, ###) denote a significant difference in means (medians based on Wilcoxon ranked-sum test) between the samples at the .01, 0.05 and 0.10 levels.
Standard errors are clustered at the firm level.

35
Table 3
Logistic regression of the decision to buy analyst research.
Prob [Paid For = 1] = logit( β0 + β1(Log_MV) + β2(Log_Age) + β3(Loss) + β4(Std Dev of Returns) + β5(Intangibles)
+ β6(Turnover) + β7(Prior Returns) + β8(NIH) + β9(PIH) + β10(Number of SS Analysts) + β11(Previous Financing) +
β12(Future Financing) + β13(M&A Activity) + β14(Book-to-Market) + β15(Leverage) + Exchange dummies +Year
dummies.

Variable Pred. Sign Z coef. %STDX

Log(Size) – –0.10 –18.8%
(0.08)
Log(Age) – –0.16 –13.7
(0.03)
Loss + 0.13 6.4
(0.21)
Std Dev of Returns + 3.02 8.4
(0.00)
Intangibles + 0.63 35.2
(0.00)
Turnover – –5.66 –6.7
(0.08)
Prior Returns + 0.11 8.3
(0.00)
NIH – –0.01 –69.3
(0.07)
PIH – –1.64 –39.2
(0.00)
Number of SS Analysts – –0.31 –83.2
(0.00)
Previous Financing +/– –0.11 –4.5
(0.76)
Future Financing + 0.69 22.3
(0.04)
M&A Activity + 0.34 17.0
(0.03)
Book-to-Market +/– –0.00 –1.3
(0.02)
Leverage +/– 0.02 0.4
(0.93)
AMEX +/– 1.20 40.7
(0.00)
NASDAQ +/– 0.89 55.2
(0.02)

Year dummies y
Pseudo R2 10.8%
N 142,475
The logistic regression model examines the decision to buy company research from 2000–2006. Over that period, 190 firms
bought coverage and the necessary data for the regression model. The unit of observation is firm-fiscal quarter. Independent
variables (except for Financing and M&A Activity) are measured the fiscal quarter prior to the one where the decision is made to
buy analyst research. Paid For is an indicator variable = 1 if the firm initiated paid-for coverage during the fiscal quarter. I
assume all firms in the CRSP/COMPUSTAT merged database that are not in my Paid-For sample did not buy analyst coverage.
Other variables are defined in Table 2. %STDX is the percentage change in odds in the likelihood of paying for analyst research
for a standard deviation increase in the independent variable. P-values, in parentheses, are one-way if signed and based on robust
standard errors clustered by firm.

36
Table 4
Abnormal returns around the release of a paid-for analyst report

Panel A: Number of reports issued by recommendation level and change in recommendation

ΔREC
Recommendation Initial Upgrade Reiterate Downgrade Total
Strong Buy 40 56 696 3 795 (28.8%)
Buy 188 79 1,284 56 1,607 (58.1%)
Hold 16 6 198 76 296 (10.7%)
Sell 0 0 22 21 43 (1.6%)
Not Rated 5 0 13 4 22 (0.8%)
Total 249 141 2,213 160 2,763 (100.0%
(9.0%) (5.1%) (80.1%) (5.8%)

Panel B: Abnormal return by initiation recommendation

All initiations (+) Strong Buy (+) Buy (+) Hold (+/–) Sell (–) Not Rated (+/–)
CAR 1.74% 3.06% 1.36% 0.53% – 8.19%
(0.01) (0.01) (0.03) (0.87) (0.13)

Panel C: CAR = β1(Upgrade) + β2(Reit) + β3(Downgrade) + β4(Initiations) + ε

UP (+) REIT (+/–) DOWN (–) INIT (+) N

CAR 1.32% 0.47% –1.10% 1.74% 2,598
(0.08) (0.01) (0.09) (0.00)

Panel D: CAR = β1(Strong Buy) + β2(Buy) + β3(Hold) + β4(Sell) + β5(Not Rated) + ε

SB (+) Buy (+) Hold (+/–) Sell (–) NR(+/–) N

CAR 1.19% 0.46% –0.80% 0.26% 2.16 2,598
(0.00) (0.03) (0.05) (0.90) (0.47)

CAR is the 2-day cumulative abnormal returns from the day of and the day after [0, +1] the report release date. Abnormal returns
are size-adjusted. Specifically, Strong Buy includes Speculative Strong Buy, Strong Buy, Strong Speculative Buy. Buy includes
Buy and Speculative Buy. Hold includes Hold, Neutral, and Speculative. Sell includes Avoid, Reduce, Sell, Suspended, and
Underperform. Not Rated is used when the rating states “Not Rated.” Panel B includes only the first paid-for report issued. N in
panel C and D doesn’t add to the 2,763 in panel A because of missing CARs. P-values, in parentheses, are one-way if signed and
based on robust standard errors clustered by analyst.

37
Table 5
Changes in institutional holding, analyst following, and liquidity after the initiation of paid-for analyst coverage

Pred Before buying After buying

Characteristic sign analyst coverage analyst coverage Difference p-value Diff-in-Diff p-value

NIH + 12.86 17.08 4.22 0.00 2.25 0.01

PIH + 11.82% 14.28% 2.47% 0.01 0.73% 0.18

Number of Analysts + 0.31 0.55 0.24 0.00 0.16 0.05

Spread Fqtr – 4.35% 3.27% –1.08% 0.00 –0.20% 0.26

Turnover Fqtr + 0.54% 0.69% 0.15% 0.13 0.16% 0.11

The sample consists of 152 firms that bought analyst research during 1999–2006 and have non-missing data for quarters –4 to +3 relative to the fiscal quarter the analyst is hired
in. Before buying analyst coverage is the mean level of the variable per fiscal quarter for the period quarter –4 to quarter –1. After buying analyst coverage is the mean level of the
variable per fiscal quarter for the period quarter 0 to quarter +3. NIH and PIH are the number and percentage of institutional holdings based on the most recent report issued prior
to the end of the fiscal quarter. Number of SS Analysts is the maximum number of unique sell-side analysts issuing an earnings forecast in the fiscal quarter. I assume that analyst
coverage, institutional holdings and institutional percentage of ownership are zero for any period when the company is listed on an exchange but there is no data available on
analyst coverage and institutional holdings. Spread Fqtr is the average daily closing ask minus closing bid scaled by price over the fiscal quarter. Turnover Fqtr is the average
daily number of shares trades divided by common shares outstanding over the fiscal quarter. I require at least 25 days of trading data for Spread Fqtr and Turnover Fqtr. The
column Diff in Diff is the difference between before and after buying analyst coverage after the mean level of the quarterly variables have been adjusted relative to a control firm
matched by calendar quarter and propensity score. P-values are one-way if signed and based on robust standard errors clustered by firm.

38
Table 6
Relative optimism and accuracy of paid-for versus sell-side analysts

Panel A: Analyst optimism and accuracy

Independent variable sign RFOPT RFACC REC

PAID FOR + –0.0103 0.1467 1.1095
(0.55) (0.00) (0.00)
HORIZON + 0.0003 0.0006
(0.02) (0.00)
AEXP – –0.0193 0.0109 –0.0025
(0.20) (0.70) (0.00)
ACOM + –0.0016
(0.73)
ASPEC – 0.1485
(0.97)
FSIZE – –0.0003
(0.35)
FSPEC – –0.0627
(0.24)
NUM BUY + 0.5157
(0.00)
NUM HOLD +/– –0.3580
(0.00)
NUM SELL + –0.9332
(0.00)

R2 0.47% 2.43% 13.09%

N 2,504 2,492 324,370
Unique firms 114 114 11,119

The RFOPT and RFACC models consist of 114 firms on I/B/E/S that bought analyst research during 2000–2006 where at least
one paid-for analyst and one sell-side analyst were forecasting earnings for the same company/forecast period/horizon category.
The REC model consists of all available stock recommendations on US companies in I/B/E/S over 2000–2006. Within each
company/forecast period/horizon category, RFOPT is the analyst’s earnings forecast less the mean forecast of all analysts all
scaled by the standard deviation of the forecasts. Within each company/forecast period/horizon category, RFACC is the analyst’s
absolute forecast error less the absolute forecast error of all analysts all scaled by the standard deviation of their absolute forecast
errors. REC is the analyst’s recommendation from I/B/E/S coded 2 for a Strong Buy or Buy, 1 for a Hold, and 0 for a Sell or
Strong Sell. PAID FOR is an indicator variable = 1 if the analyst works for a fee-based firm, and 0 otherwise. HORIZON is the
number of days between the forecast issue date and the related earnings announcement date. AEXP is the log of the number of
quarters than an analyst has an earnings estimate in I/B/E/S up to and including that quarter. ACOM is the number of other
companies an analyst provides earnings estimates for during the calendar year. ASPEC is the percentage of companies out of the
total other companies covered by an analysts during the calendar year that are in the same I/B/E/S industry classification as the
forecast firm’s industry. FSIZE is the number of other analysts working at the forecasting analyst’s firm during the calendar
year. FSPEC is the percentage of other analysts working at the forecasting analyst’s firm who are covering a firm in the same
industry as the forecast firm’s industry. NUM BUY, NUM HOLD, NUM SELL are the number of Strong Buy/Buy, Hold, and
Sell recommendations made by other analysts for the same firm and calendar quarter. P-values, in parentheses, are one-way if
signed and based on robust standard errors clustered by analyst.

39
Table 7
High credibility and low credibility fee-based firms

Panel A: High and low credibility fee-based research firms

Paid-fee research firm: Credibility Research IB Consulting May trade

Cohen Independent Research Low y y y
EquityNet Research Low y y y
Fundamental Research High y
Howlett Research Low y y y
Investrend High y
J.M. Dutton Associates High y
Researchstock.com Low y y y
SISM Research High y
Spelman Research Associates High y
Taglich Brothers Low y y y

Panel B: Abnormal returns around the release of a paid-for analyst report

All initiations (+) Strong Buy OR Buy (+) Hold (+/–)

CAR (high credibility) 2.63% 2.48% 5.11%
(0.01) (0.01) (0.36)
CAR (low credibility) 0.92% 0.90% –4.04%
(0.10) (0.08) (0.13)
Difference 1.71% 1.58% 9.15%
(0.08) (0.09) (0.12)

CAR = β1(Upgrade) + β2(Reit) + β3(Downgrade) + β4(Initiations) + ε

UP (+) REIT (+/–) DOWN (–) INIT (+) N

CAR (high credibility) 2.52% 0.67% –1.25% 2.63% 1,686
(0.01) (0.00) (0.15) (0.01)
CAR (low credibility) –0.33% 0.05% –0.90% 0.92% 912
(0.58) (0.43) (0.20) (0.10)
Difference 2.85% 0.62% –0.35% 1.71% 2,598
(0.06) (0.06) (0.41) (0.08)

CAR = β1(Strong Buy) + β2(Buy) + β3(Hold) + β4(Sell) + β5(Not Rated) + ε

SB (+) Buy (+) Hold (+/–) Sell (–) NR(+/–) N

CAR (high credibility) 1.19% 0.66% –0.84% 4.96% –0.96% 1,686
(0.00) (0.03) (0.22) (0.90) (0.17)
CAR (low credibility) 1.22% 0.24% –0.75% –2.80% 3.60% 912
(0.12) (0.20) (0.04) (0.00) (0.42)
Difference –0.03% 0.42% –0.09% 7.78% –4.56% 2,598
(0.51) (0.17) (0.90) (0.05) (0.28)

40
Table 7 (cont.)
High credibility and low credibility fee-based firms

Panel C: Changes in institutional holding, analyst following, and liquidity after the initiation of paid-for analyst coverage

Pred
Characteristic sign Before After Diff p Diff Diff p

NIH High credibility + 16.35 23.70 7.35 0.00 4.54 0.00

Low credibility + 10.76 13.11 2.35 0.00 0.87 0.22
High – Low + 5.00 0.00 3.67 0.03

PIH High credibility + 15.01% 19.52% 4.51% 0.00 2.92% 0.00

Low credibility + 9.90% 11.14% 1.24% 0.04 –0.30% 0.70
High – Low + 3.27% 0.02 3.22% 0.01

Num of SS High credibility + 0.33 0.77 0.43 0.00 0.38 0.01

Analysts Low credibility + 0.30 0.42 0.12 0.05 0.04 0.39
High – Low + 0.31 0.01 0.34 0.04

Spread Fqtr High credibility – 2.88% 1.92% –1.11% 0.00 –0.63% 0.01
Low credibility – 5.23% 4.17% –1.05% 0.00 0.06% 0.55
High – Low – –0.06% 0.20 –0.69% 0.11

Turnover Fqtr High credibility + 0.47% 0.67% 0.19% 0.02 0.25% 0.01
Low credibility + 0.58% 0.70% 0.12% 0.28 0.11% 0.30
High – Low + 0.07% 0.35 0.14% 0.26

41
Table 7 (cont.)
High credibility and low credibility fee-based firms

Panel D: Relative analyst optimism and accuracy

Independent variable sign RFOPT RFACC REC

PAID FOR + –0.0365 0.1234 1.0293
(0.65) (0.02) (0.00)
HIGH CREDIBILITY – 0.0949 0.0742 0.1560
(0.78) (0.81) (0.70)
HORIZON + 0.0003 0.0006
(0.02) (0.00)
AEXP – –0.0214 0.0074 –0.0025
(0.18) (0.64) (0.00)
ACOM + –0.0006
(0.42)
ASPEC – 0.1392
(0.96)
FSIZE – –0.0004
(0.31)
FSPEC – –0.0487
(0.29)
NUM BUY + 0.5157
(0.00)
NUM HOLD +/– –0.3580
(0.00)
NUM SELL – –0.9332
(0.00)

R2 0.55% 2.47% 9.21%

N 2,504 2,492 324,370
Unique firms 114 114 11,119

This table partitions the fee-based firms into two subsamples (high and low credibility) and replicates the analyses in table 4 to 6
by high and low credibility. Panel A shows the classification of five fee-based firms into high and low credibility based on if it
has other lines of business in addition to its research department and if its policies permit the firm to hold and trade client stock.
Panel B replicates table 4 and shows the cumulative abnormal returns (CAR) from the day of and the day after [0, +1] the report
release date. Panel C replicates table 5 and shows the change in the mean level of the variables per fiscal quarter for the quarters
[–4, –1] (before coverage) to quarters [0, +3] (after coverage). Panel D replicates table 6 and compares whether analysts from
high credibility fee-based research firms are more accurate or less optimistic than analysts from low credibility fee-based firms.
HIGH CREDIBILITY is an indicator variable = 1 if the fee-based firm is classified as a “High Credibility” firm and 0
otherwise.

42