using fundamental analysis to assess earnings quality- evidence from the center for financial...

7/27/2019 Using Fundamental Analysis to Assess Earnings Quality- Evidence From the Center for Financial Research and Anal

1/32

Using Fundamental Analysis to Assess Earnings Quality:

Evidence from the Center for Financial Research and Analysis

by

Patricia M. FairfieldAssociate Professor

Georgetown UniversityThe McDonough School of Business

Washington, DC 20057

(202) 687-4583

J. Scott Whisenant*Assistant Professor

Georgetown UniversityThe McDonough School of Business

Washington, DC 20057(202) 687-4386

October 2000

*corresponding author

Georgetown University

The McDonough School of Business37 th & O Streets, NW

Washington, D.C. 20057email: [email protected]

______________________* This study has benefitted from the comments of Jim Ohlson, Prem Jain, Stephen Penman, Katherine Schipper, Srinivasan

Sankaraguruswamy, Teri Yohn, Richard Sweeney, Jim Bodurtha, Rob Schilit, Howard Schilit, and participants at the 2000 KPMG/JAAF

Conference. We would like to especially thank John Core, the discussant, for his many helpful suggestions.


2/32



ABSTRACT: We document post-event negative abnormal returns to the (implicit) sell recommendations ofa group of fundamental analysts. We also find statistically significant deterioration in the financialperformance of the identified firms in the year after the recommendations. Together the results areconsistent with the claim of fundamental analysts that they are able to identify firms that are successfullymasking operational problems with aggressive accounting. The sample in this study comprises 373 firmsidentified over a four-year period by the Center for Financial Research and Analysis (CFRA). The CFRAoffers to subscribers a monthly report identifying approximately ten firms which CFRA claims areexperiencing operational problems and particularly those that employ unusual or aggressive accountingpractices to mask the problems. The CFRA analysts rely on traditional techniques of fundamental analysis,including mechanical screens and more time-consuming analyses of footnotes and other public disclosures.Their data sources include only publicly available information, primarily SEC filings. We conclude thatCFRAs apparent success in identifying firms with deteriorating performance provides evidence about the

usefulness of traditional financial statement analysis. The results also provide a strong rationale for futureresearch to identify specific techniques of fundamental analysis that can be employed to detect operationalproblems masked by aggressive accounting practices.

Key Words: fundamental analysis, market efficiency, contextual analysis, off-financial-statement data

Data Availability: All data are available from public sources. The list of firms was obtained from theCenter for Financial Research and Analysis, Inc.


3/32

1Thornton OGlove, author ofThe Quality of Earnings, says about public filings: Between the lines of these bland, legalistic, and

too often overlooked filings are to be found stores of information. Kathryn Staley, author ofThe Art of Short Selling, puts such

emphasis on the disclosures in the SEC filings that she titles a chapter of her book, If You Cant Read It, Short It.

2Analysts selling such advice include David Tice (Behind the Numbers), and Kellogg Associates (Financial Statement Alert).

3CFRA was founded in 1994 by Howard Schilit, a former accounting professor. The firms primary product, The CFRA Monthly

Research Compendium, was, during the period covered by this study, a monthly list of approximately five (in its first year) to ten (in later

years) research reports on publicly held companies. During the period covered by our sample, subscribers to the CFRA research report

received the information on the same date every month via overnight delivery. Subscribers to the CFRA research report include mutual

funds, money managers, hedge funds, insurance companies, banks, CPA firms, law firms, and individual investors. CFRA claims that it

does not offer brokerage services, manage client money, or engage in short selling. CFRAs research reports are discussed in If the

Numbers Look Fishy, Heres the Man to Call (Fortune , April [1999]), Analysts Pursuit of Shenanigans Rocks Lots of Boats

(Silicon Street, March [1999]), and The Sherlock Holmes of Accounting (Business Week, September [1994]). CFRA moved to an

internet-based product in November 1999.



1. INTRODUCTION

Financial statement analysis as commonly understood encompasses more than computer generated

analyses of quantitative financial statement data. Any textbook outlining the techniques of fundamental

analysis points out the importance of reading actual documents filed with the Securities and Exchange

Commission (SEC), not just the financial statement extracts. A number of accountants and analysts have

developed reputations as practitioners of fundamental analysis, and have espoused the usefulness of

fundamental analysis in detecting overvalued stocks.1

Unfortunately, with the exception of two studies

investigating the returns to firms identified by Abraham Briloff in Barrons (Foster [1987], [1979]), we

know of no study investigating the claims of fundamental analysts to uncover operational problems that may

be masked by aggressive accounting practices.2 A key reason has been the absence of a public record of

dated investment recommendations generated by analysts relying exclusively upon fundamental analysis.

Our study contributes to research in this area by investigating the analytical ability of the principals

associated with the Center for Financial Research and Analysis (hereafter CFRA).

3

CFRA publishes

research reports identifying firms with quality of earnings problems. According to CFRA analysts, their

recommendations are based on information disclosed in publicly available data. Their analytical procedures


4/32

2

4See Lucent Net Trails Expectations, Confirming Recent Warnings (The Wall Street Journal, January 21, 2000) and Is

Healtheon/WebMD Pushing the Limits on Revenue? (The Wall Street Journal, February 7, 2000) for illustrations of the type of issues

covered in CFRAs research reports.

include a combination of mechanical screens on financial statement data and subsequent analysis of

disclosures in SEC filings to identify firms that may be attempting to mask operational problems with

unusual or aggressive accounting.4 CFRA analysts claim that the mechanical and textual analyses of

financial disclosures enables them to detect poor earnings quality in identified firms. For those firms

highlighted in a CFRA research report, CFRA implicitly makes a dual prediction as a consequence of poor

earning quality: earnings and prices will fall for the identified firms.

We test the claims of CFRA analysts by examining operating performance and market returns of firms

in the period subsequent to their identification in a research report. The sample includes 373 firms identified

over a four-year period. We present two types of evidence consistent with the claims that fundamental

analysis can be used to detect overpricing attributable to aggressive accounting. First, we document that

financial performance deteriorates significantly in the year following the CFRA report. The median

(average) percentage change in ROA decreases by almost 2 (3) percentage points in the four quarters after

the CFRA report compared to the four quarters before the CFRA report.

Second, we find negative abnormal returns of approximately one percent over a two-day announcement

period around CFRA (research report) publication dates, and negative abnormal returns of approximately

ten percent over the year following publication of CFRA research reports. Both the deterioration in

operating performance and stock returns are consistent with the claims of CFRA analysts. The decrease

suggests that operational problems surface subsequent to CFRA research reports. The evidence of negative

abnormal stock returns is important in that it demonstrates that CFRA reports appear to reveal new

information about either operational problems or aggressive reporting practices to market participants.


5/32

3

5 Barrons invites a group of eight to twelve participants to their Roundtable each year. The participants are described as Wall

Street Superstars byBarrons and include successful mutual fund managers, stock analysts, and private investors. The participants

recommend their buys and sells and describe the rationale for their picks.

2. PRIOR RESEARCH

Prior research on the information content of investment advice has most often focused on the

informativeness of analysts recommendations by testing for price reactions to those recommendations

(Foster [1987], [1979]; Bjerring et al. [1983]; Lee [1986]; Desai and Jain [1995]; Womack [1996]). Such

research has investigated the payoffs to the use of investment advice, but provides no evidence on the

usefulness of fundamental analysis to detect poor earnings quality. For example, Barber and Loeffler

(1993) analyze recommendations published in the Dartboard column ofThe Wall Street Journal. They

find significant publication day returns, but also document that the abnormal returns reverse in the days

immediately following the publication. No evidence is provided on the subsequent financial performance of

the recommended firms.

Desai and Jain (1995) report that investors do not benefit from using the investment advice from

another source (Barrons Annual Roundtable) from 1968 to 1991.5 They document that abnormal returns

are essentially zero for one to three-year post-publication holding periods. In contrast, in a comprehensive

study of the recommendations from fourteen large brokerage firms, Womack (1996) reports that the

analysts from the brokerage firms have market timing as well as stock picking abilities. Thus there is mixed

evidence suggesting that some public recommendations by analysts are potentially useful to market

participants.

With the exception of two studies on Abraham Briloffs published accounting analyses (Foster [1987],

[1979]), prior studies do not discriminate between the returns to recommendations based upon fundamental

analysis and those based upon other sources of information. In the two studies by Foster, the evidence of

negative market reactions to Briloffs critiques published inBarrons is the closest we come to finding any

evidence justifying the claims of fundamental analysts to detect poor quality earnings. Yet, because Foster

provides no information on the subsequent financial performance of the firms, one cannot distinguish


6/32

4

between a negative market reaction based on Briloffs reputation and one based on valid claims of poor

earnings quality.

While the evidence in this study has relevance to the issue of earnings management, it does not relate

directly to the current literature on the issue. In the academic research on earnings management, the term

encompasses accounting policies or entries made by management to effect certain outcomes; motivation is

a key part of the definition. More importantly, most of this research depends on a questionable premise.

Earnings management is said to exist when the accrual component of earnings is something other than what

the researcher believes it should have been. When ex-post validation of earnings management is sought,

the researcher typically tests whether subsequent accruals are of the opposite sign and in excess of what

he(she) believes they should have been. To be fair, the researchers expectations have some objective

reality as they are generated by a time series model of accruals. Tests on the sign and magnitude of

deviations from expectations (abnormal errors) are presented as evidence that earnings management does or

does not exist. However, the relation of this evidence to anything that financial analysts and investors

actually do or care about is not clear.

In this study we take a different approach. The phenomenon of interest is not earnings management,

but is instead earnings quality. Whether or not management has intervened to produce a particular

accounting outcome is not a consideration. Poor quality earnings produced by correct and consistent

application of GAAP are also of interest. We look for subsequent decreases in operating performance and

earnings per share as evidence of poor earnings quality. For those identified firms, we investigate whether

negative abnormal stock returns are associated with the decreases in operating performance and earnings per

share. We interpret these two forms of evidence as confirming the ability of skilled analysts to use financial

statements to detect poor quality earnings, and the failure of the market as a whole to do so. However,

because CFRA analysts do not ascribe motivation to the management in reporting earnings, we cannot claim

that we provide any evidence on whether or not earnings has been managed or any violation of GAAP

has occurred. It is an open question whether techniques practiced by CFRA analysts would be of interest to


7/32

5

6CFRA principals claim that their analysts contact target firms (of research reports) as well as its competitors to ask a series of

questions without revealing which firm is the actual target. They state that their analysts rarely ask specifics about the content of the

actual report. Instead, the discussions and interviews are simply to confirm their own understandings of revenue recognition or cost

capitalization issues in the industry. The CFRA principals did state that firms sometimes decline to speak to them.

regulatory authorities responsible for enforcing compliance with GAAP. On the other hand, the earnings

and price declines we document are likely to be of interest to analysts and investors.

Central to the research issue is CFRAs claim, which we take at face value, that their analysts use

fundamental analysis to identify target firms. A key limitation of this research is that we do not identify any

of the methods used by CFRA analysts, nor do we present evidence demonstrating the usefulness of any

specific subset of techniques or disclosures used to detect poor quality earnings. Because CFRA is a

private, for-profit entity, its principals were understandably reluctant to discuss any specific screens or

procedures they use to assess earnings quality. They did confirm that they use a combination of data

screens (using CompustatandLexis-Nexis databases) followed by an analysis of numerical and textual data

in SEC filings. They also state that they do not use private information from management in identifying

firms, and that discussions with management are not used to initiate reports and that no such discussions

stopped an investigation into a firm during the sample period.6

CFRA promotional literature identifies seven specific accounting maneuvers that its analysts can detect

through fundamental analysis: (1) recording revenue too soon, (2) recording bogus revenues, (3) boosting

income with one-time gains, (4) shifting current expenses to a later period, (5) failing to record or disclose all

liabilities, (6) shifting current income to a later period, and (7) shifting future expenses to the current period.

CFRA analysts also explicitly state that data beyond the financial statements are central to the analysis of

earnings quality. A list of disclosures scrutinized by CFRA includes press releases, footnotes, proxy

statements, the Presidents Letter, Managements Discussion and Analysis of Operations, and auditors

reports.


8/32

6

3. METHODOLOGY AND DATA

Our approach to investigate the payoffs to fundamental analysis offers evidence that extends existing

research by examining the ability of the CFRA to use, as claimed, fundamental analytical tools to detect

poor quality earnings. We compare the earnings changes prior to the CFRA research report dates with

earnings changes subsequent to the report dates. In addition to testing the ability of the CFRA analysts to

detect deteriorating firm performance, we also investigate returns to a buy-and-hold trading strategy for the

373 firms with available data for the periods after their identification in any of the CFRA research reports

over the four-year period, 1994 to 1997.

3.1 CFRA Research Report and Financial Statement Data

CFRA began reporting its monthly research report in January 1994. The first research report was a

summary of prior findings by CFRAs lead analyst over several years, and is excluded from our sample

evidence. The second CFRA research report began identifying firms in real time. The sample period in our

study includes the 47 monthly reports from February 1994 to December 1997. The financial disclosure

data are taken from the quarterly database of Standard & Poors 1999 CompustatIndustrial and Research

files (Primary, Supplemental, Tertiary; Full-Coverage; and Industrial Research Files). The stock price

and shares data are obtained from the Center for Research in Security Prices (CRSP) 1999 daily

(NYSE/AMEX/NASDAQ Combined) database. The sample consists of firms listed on the New York

Stock Exchange, American Stock Exchange, and NASDAQ.

During 1994, the research reports identify at least five and no more than seven firms per month without

regard to the differences in sizes of the identified firms. Beginning with the first report in 1995, CFRA

reports use a size classification of small to mid and mid to large which effectively doubled the number

of firms on which reports were written, as each classification continues to include approximately five

alerts (i.e., a firm identified in CFRA research report) per month. Of the four-year sample from 1994 to

1997, three of the 413 on which CFRA reported in the 47 research reports are friendly alerts. We

exclude those three friendly alerts, another 25 observations due to missing financial statement, and 12


9/32

7

7 We include firms in our sample with at least two quarters of financial statement data in either the pre- and post-publication periods.

Of the final sample of 373 observations, 40 firms have incomplete financial statement data in either one or two quarters of the pre- or

post-publication periods. The results of the tests of differences for ROA and)EPS are unchanged when those 40 firms are removed.

Also, 17 firms have incomplete stock price data across all event windows in our study. When we delete those observations, the

conclusions are also unchanged on the tests of abnormal stock returns. Therefore, the final sample excludes firms with missing data in

eitherCompustatorCRSP, but includes those firms with incomplete (as defined herein) data.

8Of the final sample of 373 observations, 69 comprise the group with more than one inclusion in a CFRA report. Although we report

on the full sample as though each is an independent observation, if those 69 observations are removed, the conclusions from the tests of

differences for ROA and )EPS as well as stock price behavior for each of our event windows are unchanged.

other observations on which missing stock price data across one or more event periods exists. 7 Thus, the

final sample size is 373 firms, excluding a total of 40 of the 413 firms in the 47 research reports. 8

An important point is worthy of mention since our analysis uses quarterly financial statement data from

Compustat. Compustatquarterly files provide financial statement data only on a restated basis (unlike

Compustats annual database of financial statement data where restated data are alternative data items).

Restatements may include balance sheet or income statement reclassifications, post-merger purchase price

adjustments, pooling effects, and/or voluntary or mandatory restatements filed in amended Form 10-K

(annual reports) or 10-Q (quarterly reports) filings at the SEC. Table 1 shows the number of CFRA firms

for which Compustatcoded one or more restatements over the fiscal quarters from which we obtain

quarterly financial statement data to measure operating performance. Approximately 21 percent of the

CFRA sample has a restatement during the seven-year period. A seven-year period is used since we may

obtain data up to two years before and one year after the period in which a firm in included in CFRA

research reports. Although the seven-year period exceeds the data requirements for any CFRA firm, we

present the entire period for in order to compare the incidence of restatements across the sample and control

groups. The latter group does not have any event date such as CFRA research report publication dates.

The 21 percent restatement rate over the seven-year period for the CFRA sample is higher than the rate

for the control group ofCompustatfirms, all of which are in the same industries as the CFRA sample.

Approximately 11 percent of the control group data has a restatement, giving it a restatement rate of about

half the CFRA samples. This difference in the incidence of restatements in the two groups may suggest

that CFRA is successfully identifying firms whose financial results are so compromised that they are forced


10/32

8

9For the four quarters before and after firms are identified in CFRA research reports, we find almost identical restatement rates to

those reported for the CFRA sample. This suggests that it is unlikely that CFRA analysts use restatements as a screen to initiate an

investigation.

to restate prior results. However, without determining the nature and timing of each of the restatements in

the two populations, we offer that only as an untested conjecture.9

In our study, we include all CFRA firms regardless of whether or not a restatement occurred in the pre-

or post-CFRA report periods. To investigate whether our results are compromised by inclusion of

observations with restatements, we repeated the tests of differences in financial performance in the pre- and

post-CFRA report periods. After removing any CFRA firm with restated data in the fiscal quarters from

which we obtain financial statement data prior to any CFRA research report, the conclusions are unchanged

on the tests of abnormal operating performance and abnormal stock returns. Perhaps more importantly, we

repeat the tests after purging the sample of any firm with a restatement across the seven-year period shown

in table 1. Our conclusions are also robust to this data requirement.

Table 2 classifies the firms in the sample by year and by industry. Approximately 50 percent of firms

reported in CFRA research reports operate in fifteen different industries (based on two-digit SIC codes).

The remaining are distributed across 30 other industries. Additionally, table 2 shows that only three

industries account each for more than 10 percent of the sample. Business services (SIC codes 7300-7399),

electrical equipment (3600-3699), and commercial machinery/computers (3500-3599) make up

approximately 14, 12, and 12 percent, respectively, of the sample. Thus, the CFRA analysts appear to

report on firms across a broad group of industries.

Table 3 shows that inclusion of a firm in a CFRA report does not appear to be strongly related to a

particular fiscal quarter. Although the second and fourth fiscal quarters are somewhat over-represented as

the last fiscal quarter of data that is publicly available prior to inclusion in CFRA research reports, the

distribution of the reports spans all four fiscal quarters prior to inclusion in CFRA reports. Panels B and C

of table 3 document that CFRA analysts report on firms in both NYSE and NASDAQ/other markets, as


11/32

9

10In some cases this assumption biases against our finding evidence of CFRA analysts ability to detect poor quality earnings, since

they claim to work only from SEC documents and not from press releases or earnings announcements. That is, in the event CFRA issued

a report between the release of earnings and the filing of the Forms 10-Q or 10-K, we assume they had access to financial statement

disclosures in the SEC filing which were in fact unavailable to them. Although perhaps overly conservative on our part, this assumption

does insure, however, that all earnings data classified as post-report data were in fact unavailable to the CFRA analysts at the date of

their report.

well as large and small firms (as defined by CFRAs two size designations,small to mid-cap and mid to

large-cap, which CFRA claims are separated at approximately $1 billion market capitalization level).

For our analysis of CFRA advice, we first examine whether firm performance subsequent to the

investment advice has deteriorated as compared to the operating performance prior to the publication date of

the research reports. For our tests of differences between the periods before and after publication dates of

the investment advice, we document differences in operating performance and earnings per share over the

four fiscal quarters prior to and after publication of the research report.

It is important to determine the last financial statement data available to CFRA analysts prior to each

report. For the first 43 months of our sample period, the publication date of the CFRA report was the

fifteenth of each month, thereby arriving to subscribers the following day; the publication date was changed

to the twentieth of each month beginning with the September 1997 report. If the earnings announcement

date occurs on or before the fifteenth (twentieth beginning September 1997) of the month, we assume the

CFRA analysts had access to the underlying financial disclosures, including those filed later in the 10-Q or

10-K.10 That is, the most recent announcement date, which we obtain from Compustat, on or before the

fifteenth (or twentieth) of the month in which a company is included in the CFRA research report is used as

the last fiscal-quarter for our pre-publication period. The subsequent four fiscal quarters are considered the

post-publication period.

One variable of interest is change in earnings, )EPS, which we define as:

)EPSt = , (1)[ ][ ]EPS EPS

ABS Average EPS EPS EPS EPS

t t

t t t t

+ + +

4

1 2 3 4( ) ( ) ( ) ( )


12/32

10

11Shares are adjusted by the adjustment factor at the ex-dividend date. Also, firm subscripts are not shown, but should be assumed.

where EPSt is defined as income available to common equity using common shares for primary earnings

per share at the end of quartert.11 The seasonally-differenced firm performance measure is scaled by the

average value of EPS over the four previous fiscal quarters relative to quarter t. If the average value of

EPS across the four fiscal quarters is less than zero, we use the absolute value of the average. We calculate

)EPSfor each of the four quarters in the pre- and post-publication periods.

To determine expected performance, we match the pre-publication data on )EPSandROA (description

follows) according to the methodology shown in Barber and Lyon (1996). They show that test statistics are

well specified only in those cases where matching is done on pre-event characteristics. Performance

matching adjusts for the mean reversion in accounting data that reflects transitory components of operating

income (Penman [1991]). We also test abnormal operating performance by investigating differences in pre-

and post-publication return on assets. Thus, another operating performance variable of interest is:

ROAt= . (2)[ ]( )

( ) ( )

Operating Income

Average Total Assets Total Assets

t

t t+ 1

For ROA, we define performance as operating income before depreciation at the end of quarter tdivided by

average total assets at the end of quartert. Thus, in conjunction with )EPS as defined, we are able to test

the sensitivity of our results to special items and other non-operating charges. Average total assets is

calculated using beginning and ending values of total assets at the end of quartert.

As with )EPS, we define expected performance in accordance with the Barber and Lyon (1996)

methodology (hereafter, BL). We match on pre-event characteristics using ROA from the four quarters

prior to a firms inclusion in a CFRA research reports. The matching process used by BL first attempts to

match firms to two-digit (SIC code) industry groups and then to pre-event performance within those

industries. The performance matching is done on the basis of a 90-110% range. We use alternative rules

shown in BL when we cannot find a matching firm using both industry and similar performance criteria in


13/32

11

12To test the robustness of our conclusions to alternative measures of firm performance, we use, in addition to )EPS, (1) change in

EPS scaled by EPS lagged by two fiscal quarters, (2) unscaled changes in EPS, (3) change in primary EPS including extraordinary items

scaled by the absolute value of seasonally-lagged primary earnings per share including extraordinary items, and (4) scaled seasonal

differences in earnings by taking the absolute value of each prior quarter instead of the absolute value of the average. In each case, the

results of tests of differences between fiscal quarters before and after the CFRA publication date are similar to those reported in the

study.

the pre-CFRA report period (approximately 40% of the ROA matching process uses alternative rules). The

first alternative procedure is to match within the 90-110% pre-event performance for one-digit (SIC code)

industry groups. Second, the same filter is applied without regard to industry. If no match is found, the

final step is to use the firm with performance closest to the remaining unmatched firms.

Although we feel that inclusion of all firms is justified regardless of which step finds an appropriate

match, the evidence of abnormal operating performance is robust to the sample matched by two-digit SIC

code and pre-event performance (step one of the matching process). However, we include all firms since

the exclusion of some has the potential to bias test statistics. Finally, the two measures of operating

performance used in our study, )EPS and ROA, are expectation models that can be characterized as a

changes and levels model, respectively. BL show that both are well specified when matched to pre-

event performance.

To further analyze abnormal operating performance of the CFRA sample, we decompose ROA into

profit margin and asset turnover. The tests of differences in the components of ROA use the ROA

matching process to determine expected performance in the two components. That is, profit margin and

asset turnover are not independently matched on pre-event performance of each. We believe the evidence

from both a levels and changes model, as well as adherence to the BL matching process, is preferable to

assuming each CFRA firms past performance is its expected future performance. However, as

documented in BL, if we assume expected performance is simply each firms past performance, we find

much stronger evidence of abnormal operating performance for each measure employed. Table 4 offers

further descriptions of the variables of interest to this study.12


14/32

12

13Since outliers are present in the data, trimmed and Winsorized means are estimated to produce more robust estimators of the

population mean which are relatively insensitive to the outlying values. The trimmed mean is computed after the ksmallest and klargest

observations are deleted from the sample. The Winsorized mean is computed after theksmallest observations are replaced by the (k+1)

smallest observation, and the k largest observations are replaced by the (k+1) largest observation. In other words, the observations are

Winsorized at each end. A 5 percent Winsorization method (where k = 18) is used for the reported data in the study. Tests were

reperformed using a 1 percent Winsorization method, a trimming of means at 5 and 1 percent, and without any controls for the effects

3.2 The Behavior of Stock Prices Around CFRA Publication Dates

Our comparison of firm performance before and after CFRA research reports offers no evidence as to

whether the deterioration in performance anticipated by CFRA analysts is also anticipated by the market and

thus already impounded into the stock price at the date of CFRA report. Therefore, we also examine stock

prices coincident with and subsequent to CFRA publication dates. Evidence of abnormal returns around the

publication dates of CFRA reports is consistent with the data offered by CFRA being informative to market

participants and, thus, not previously impounded into stock prices. However, market participants might

also speculate on the information contained in CFRA research reports and thus long window returns are also

examined to provide evidence that any publication date (windows) returns do not reverse.

We measure the behavior of stock prices by estimating buy-and-hold abnormal returns (BHAR). The

estimation techniques for abnormal returns reported in the study use two alternative specifications of return

indices. A size (market capitalization) portfolio return index and a value-weighted return index are used to

estimate CFRA sample abnormal returns. Appendix 1 provides further descriptive information on the

abnormal return estimation procedures.

4. EMPIRICAL RESULTS

To evaluate the ability of CFRA analysts to forecast deteriorating firm performance, we investigate the

pre- and post-CFRA report periods for)EPS, as well as ROA and its components. The expected

performance is the result of industry and/or pre-event performance matching. The expected and actual

performance statistics, as well as results of test of differences, are reported in table 5. In Panels A and B,

the actual and expected performance measures are shown, respectively. The table reports the mean,

standard deviation, median, and percentage of means greater than zero.13 To test whether CFRA analysts


15/32

13

of outliers. The conclusions are unchanged, and we report Winsorized means since a few extreme observations skew the mean and

standard deviation for some variables.

14In all cases, t-tests on the differences in means of)EPS and ROA are consistent with the Wilcoxon signed rank tests reported here.

If using parametric t-tests of differences in the variables, we find evidence of significant differences in)EPS(t = -3.07),ROA (t = -5.74),

PM(t = -3.13), andATO (t = -1.43).

appear to be able to forecast deteriorating firm performance, we use a nonparametric Wilcoxon signed rank

test on the paired differences in )EPS, ROA, and its components.14 In panel C, the differences between

expected and actual measures of firm performance are reported. In each case, the variable of interest is

lower in the post-publication period (i.e., actual performance) compared to the expected performance (based

on control group).

The evidence suggests that both measures of operating performance exhibit abnormal behavior for the

CFRA sample. Post-event mean (median) ROA deteriorates by approximately 300 (200) basis points

compared to the control firms. For the ROA component analysis, asset turnover is marginally different in

means and medians. We find that profit margin is significantly lower in both means and medians. This

result suggests the CFRA analysts use triggers that predict deteriorating (marginal) returns to revenues based

on the evidence of significantly decreased profit margins. Thus, the evidence seems consistent with the

claims of CFRA analysts as they appear to be able to predict deteriorating firm performance.

Indications of deteriorating signals of firm performance offer no evidence as to whether the CFRA

analysts offer new information to the market. Therefore we also examine stock price behavior coincident

with and subsequent to CFRA publication dates. Table 6 presents summary results for the 373

observations. For the sample, we report the average and median buy-and-hold abnormal returns (BHARs)

for each interval. We report significance tests for buy-and-hold returns using both parametric and

nonparametric procedures. Table 6 documents that the publication two-day event period [0, +1] shows a

negative announcement effect with mean (median) CARs of -1.15 (-0.76) percent using size decile

portfolios for market index returns. Table 6 also reports abnormal returns using a CRSPvalue-weighted


16/32

14

15We also specify the abnormal returns based on standardized abnormal returns by estimating the market model over a 255-day period

ending both thirty and zero days prior to the beginning of the event periods of interest, as well as by estimating the model equally over

the pre- and post-event periods. We further estimate comparison period abnormal returns by subtracting the mean return of the common

stock over a 255-day period ending both thirty and zero days prior to the beginning of the event periods. Finally, we alternatively estimate

abnormal returns using an equally-weighted market index. The results obtained from each alternative abnormal return specification are

consistent with those reported and have no effect on our conclusions. We also perform nonparametric tests based on the Wilcoxon signed

rank test and a rank test described in Corrado (1989). The results are also robust to the use of these alternative (nonparametric) test

procedures .

benchmark.15 The results of either specification of abnormal returns suggests that market participants react

to CFRA research reports.

Additional negative abnormal returns over other buy-and-hold periods suggests that new unfavorable

information is released to the market in periods after publication of CFRA research reports. Subsequent to

the publication of CFRA research reports in which a firm is included, the firm continues to produce

consistently negative abnormal returns. For example, the abnormal return intervals of one, two, three, and

four quarters show mean (median) CARs of -3.46 (-2.10) percent, -7.89 (-7.48) percent, -8.51 (-11.01)

percent, and -11.71 (-14.44) percent, respectively, for size-indexed returns.

The evidence of negative BHARs is robust, and in most cases much stronger for alternative

specifications of a market index (e.g., CRSPvalue-weighted, also reported; CRSPequally-weighted or

comparison period returns, not reported) or alternative estimation procedure for abnormal returns. The

results suggest that speculation is not a reasonable explanation for the negative abnormal returns to the

CFRA research reports as the announcement period abnormal returns do not reverse. Additionally, we find

evidence of abnormal returns over two-year return window subsequent to inclusion in CFRA reports. The

evidence of decreasing BHARs beyond CFRA publication dates suggests that investors under-react to both

CFRA reports and subsequent earnings announcements for the set of CFRA firms.

We present graphical evidence of the behavior of abnormal returns surrounding CFRA reports in figure

1. The evidence in figure 1 shows the daily abnormal returns from 250 trading days prior to the

announcement date up to 250 trading days after the announcement. The graphical evidence shows that the

portfolio of CFRA firms experiences positive cumulative abnormal returns prior to the report. However,

abnormal returns turn negative on or about the CFRA announcement date and are consistently negative


17/32

15

16Womack (1996) documents a similar drift in his portfolio of sell recommendations made by analysts at fourteen major U.S.

brokerage firms. Ball and Brown (1968) were the first to document such a drift (see also Bernard and Thomas [1989], [1990]).

throughout the remaining period shown. Figure 1 also suggests some leakage of the information, or

alternatively suggests that market participants find other sources for the CFRA data, at approximately ten

days prior to the report. Although not reported, we test the ten daily abnormal returns during the period

from -10 to -1 trading days. The results show significant negative abnormal returns for days [-5] and [-7] at

a 10 percent significance level. Two days of significant negative abnormal returns over ten trading days

may also simply be due to chance. Thus, evidence suggests that the information in CFRA research reports

is new to market participants that is not fully impounded into stock prices at announcement dates. 16

We also document the BHARs by year. Figure 2 shows graphically the results of those tests. The first

year of the CFRA research report does not yield significant negative abnormal returns. Tests confirm that

neither announcement date nor subsequent period returns are significantly negative. However, for each of

the remaining years, the graphical evidence suggests a change in the direction of abnormal returns on or

about the release of CFRA research reports. Tests confirm that both announcement date and subsequent

period returns are significantly negative for each of the remaining years. Equally interesting is the increasing

magnitude of the negative returns. Apparently, investors either are assigning a reputation to CFRA research

reports that leads to increasingly negative returns and/or CFRA analysts are becoming more skilled at

forecasting deteriorating operating performance.

5. SUPPLEMENTAL TESTS

5.1 Sensitivity of Results to Pre-Event Stock Price Performance

The sample of CFRA firms fall disproportionately into the high ROA ()EPS) deciles in periods prior to

inclusion in CFRA research reports. As such, we employ the Barber and Lyon (1996) methodology to test

the sensitivity of the tests on abnormal operating performance by evaluating whether CFRA analysts simply

anticipate mean reversion in )EPS or ROA. The reported results offer evidence that CFRA analysts claims


18/32

16

17A size (market capitalization) portfolio return index (SI) is used, and reported in table 7, to estimate abnormal returns. The results

are robust to alternative measures used (see footnote 15).

of deteriorating earnings are robust to the mean reversion evident in accounting data due to transitory

components. However, pre-event (i.e., CFRA publication dates) stock price performance may also be a

simple screen which CFRA analysts use without regard to signals of fundamental analysis. Figure 1 offers

graphical evidence of significantly positive pre-event BHARs for the CFRA sample.

To test whether the results are robust to controls for pre-event stock price behavior, we match each of

the CFRA firms to a control sample based on pre-event abnormal returns. Control firms are sorted into

deciles based on pre-event twelve-month BHARs.17 Each CFRA firm is matched to the appropriate decile

based on its pre-event abnormal return. For both control and CFRA firms, the 12-month accumulation

period ends in the month before the CFRA report. CFRA firms were matched to one of 470 different

control portfolios, depending on the month of the CFRA report and the magnitude of the pre-event CFRA

return.

We subtract the post-event abnormal return of each CFRA firm from the median post event abnormal

return of the control firms (essentially all CRSP firms in the same 45 industries as the CFRA sample) in the

same decile and in the same period. Thus, the stock price reactions subsequent to the CFRA reports are

adjusted for the average mean reversion exhibited by the firms in the price-performance control group.

Table 7 documents the results of those tests on the 354 CFRA sample firms with complete stock price data

in the twelve-month pre- and post-publication periods. For the entire pooled sample, we can reject the null

of no difference at less than 1% significance level. On average, the CFRA sample yield approximately 15

percentage points more negative BHARs in the post-event period compared to the control group.

In four of the ten price-performance deciles (representing approximately 62 percent of the total CFRA

firms), CFRA sample firms exhibit statistically significant negative abnormal returns in excess of the control

sample using both parametric and nonparametric test procedures. For the remaining six deciles (or

approximately 38 percent of the sample), all but one exhibit lower BHARs compared to the median of the


19/32

17

control group, although the differences are not statistically significant. We conclude that the evidence of

CFRA abnormal returns is robust to controls for pre-event stock price performance.

5.2 Small versus Large Firm Returns

In table 8, we report BHARs for the 373 CFRA firms partitioned on a size variable as designated in

CFRA reports. Beginning with the first report in 1995, CFRA reports use a size classification of small to

mid and mid to large which effectively doubled the number of firms on which reports were written, as

each classification continued to include approximately five alerts per month. Negative abnormal returns are

observed in both size portfolios. For the larger firms, however, the returns are somewhat smaller and do

not appear until the second quarter after the release of the CFRA report. This result is consistent with prior

evidence that has shown that the marginal information content of financial disclosures decreases with firm

size. It may also be consistent with evidence showing that analyst following increases with firm size (e.g.,

see Bhushan [1989], Freeman [1987]) thereby producing a richer information set, and perhaps fewer

surprises, for larger firms.

6. SUMMARY AND CONCLUDING REMARKS

This study tests whether there is evidence supporting the claims of fundamental analysts to be able to

forecast deteriorating firm performance. We also test whether the information in the analysts research

reports is impounded into the market price of subject firms at the date of the report. The evidence

supports the claims that the analysts are able to anticipate deteriorating firm performance. That is, we find

the firms identified by the analysts have deteriorating firm performance in the year following the report.

Deteriorating performance is evident from changes in earnings per share, returns on assets, and profit

margins.

Additionally, we document significantly negative abnormal returns to the portfolio of firms identified in

the analysts reports. Not only is the announcement date return significantly negative, but the abnormal

returns to a buy and hold strategy are also significantly negative for up to two years subsequent to the date


20/32

18

of the research report. The results suggest that investors may under react to CFRA reports, as well as

subsequent earnings announcements, for the set of CFRA firms. The evidence also suggests that the

speculation explanation is not a reasonable explanation for the negative abnormal returns to the CFRA

research reports as the announcement period abnormal returns do not reverse as would be suggested by the

speculation explanation.

Central to our interpretation of the results is the claim of the analysts associated with the CFRA use only

publicly available information to generate their research reports. With that caveat noted, we conclude that

fundamental analysis can be used to detect signals of deteriorating firm performance, and that these signals

in publicly available data are not priced by the market. We do not address the question of which forms or

methods of financial statement analysis are used by the CFRA analysts, or whether, in fact, their results

could be replicated by other analysts. Future research may identify a combination of screens and procedures

which could be used to replicate the success of the CFRA analysts.


21/32

19

REFERENCES

Ball, R., and P. Brown. An Empirical Evaluation of Income Numbers.Journal of Accounting Research(Autumn 1968): 159-178.

Barber, B.M., and D. Loeffler. 1993. The Dartboard Column: Second-hand Information and Price

Pressure. Journal of Finance and Quantitative Analysis 28 (2): 273-284.

______, and J.D. Lyon. 1996. Detecting Abnormal Operating Performance: The Empirical Power andSpecification of Test Statistics. Journal of Financial Economics (41): 359-399.

Bernard, V., and J. Thomas. 1989. Post-earnings Announcement Drift: Delayed Price Response or RiskPremium. Journal of Accounting Research (27): 1-36.

________, and J. Thomas. 1990. Evidence that Stock Prices do not Fully Reflect the Implications ofCurrent Earnings for Future Earnings.Journal of Accounting and Economics (13): 305-340.

Bjerring, J.h., J. Lakonishok, and T. Vermaelen. 1983. Stock Prices and Financial Analysts

Recommendations. Journal of Finance (March): 187-204.

Bushan, R. 1989. Collection of Information about Publicly Traded Firms: Theory and Evidence.Journal of Accounting and Economics (11): 183-206.

Corrado, C.J. 1989. A Nonparametric Test for Abnormal Security-price Performance in Event Studies.Journal of Financial Economics (23:2): 385-396.

Desai, H., and P.C. Jain. 1995. An Analysis of the Recommendations of the Superstar MoneyManagers atBarrons Annual Roundtable. Journal of Finance (September): 1257-1273.

Foster, G. 1979. Briloff and the Capital Market. Journal of Accounting Research (Spring): 262-274.

_____. 1987. Rambo IX: Briloff and the Capital Market. Journal of Accounting, Auditing & FinanceNew Series v2(4): 409-430.

Freeman, R. 1987. The Association Between Accounting Earnings and Security Returns for Large andSmall Firms. Journal of Accounting & Economics (9): 195-228.

Lee, C.J. 1986. The Information Content of Financial Columns. The Journal of Economic and Business(May): 27-40.

Penman, S. 1991. An Evaluation of Accounting Rate of Return.Journal of Accounting, Auditing &Finance (6): 233-255.

Schilit, H. 1993. Financial Shenanigans: How to Detect Accounting Gimmicks and Fraud in FinancialReports. Boston: McGraw-Hill.

Sprent, P. 1989.Applied Nonparametric Statistical Methods. London: Chapman and Hall.

Womack, K.L. 1996. Do Brokerage Analysts Recommendations Have Investment Value? The Journalof Finance (March): 137-167.


22/32

20

APPENDIX 1

The abnormal return (AR) for CFRA firms is computed as follows:

, (1)AR R Rjt jt mt=

where is the rate of return on the common stock of the jth firm on day t, is the observed returnRjt Rmton the market index m (i.e., either a size portfolio return or a return on a broader market index) for day t.We refer the abnormal return at the publication date (denoted as two-day event window) of the CFRAreport or abnormal returns over multiple trading days as a buy-and-hold abnormal return (BHAR). Using abuy-and-hold returns strategy to estimate sample abnormal returns over an interval of two or more tradingdays beginning with T1, and ending with T2, we define each BHAR as:

. (2)( ) ( )BHAR R RT T jt t T

T

j

N

mt

t T

T

N1 21

2

1

2

1 1 1 1 1

1

, = +

+

== =


23/32


24/32

22

TABLE 2

Descriptive Statistics on Industry Followings for the

Center For Financial Research and Analysis (CFRA) Research Reports: 1994 to 1997

Industry Classification (SIC codes)

Full

Sample

% of

Sample 1994 1995 1996 1997

Business Services (7300-7399) 52 13.94% 6 9 17 20

Commercial Machinery & Computers (3500-3599) 43 11.53% 4 11 14 14

Electrical Equip. & Household Appliances (3600-

3699)

43 11.53% 6 11 14 12

Health Services (8000-8099) 20 5.36% 0 5 7 8

Misc Retail (5900-5999) 19 5.09% 0 9 7 3

Durable Goods - Wholesale (5000-5099) 17 4.56% 2 3 4 8

Measuring & Photo Equipment (3800-3899) 16 4.29% 2 1 4 9

Chemicals (2800-2899) 12 3.22% 1 2 4 5

Eng, Acctg, & Other Mgmt Services (8700-8799) 11 2.95% 1 3 1 6

Transportation Equipment (3700-3799) 10 2.68% 4 2 3 1

Restaurants (5800-5899) 10 2.68% 6 3 0 1

Misc. Manufacturing (3900-3999) 9 2.41% 0 1 6 2

Apparel (5600-5699) 8 2.14% 1 5 1 1

Home Stores (5700-5799) 7 1.88% 3 2 1 1

Insurance Carriers (6300-6399) 6 1.61% 0 0 2 4

Classifications with five or fewer firms 185 49.60% 23 57 58 47

TOTALS 373 100% 49 104 112 108

The table provides descriptive statistics of the 373 observations contained in CFRA reports with available data from February 1994 to

December 1997. Thirty industries are represented in the last group of 185 observations with five or fewer observations from any

other industry classification. Industries are classified by use of Standard Industrial Classification (SIC) codes and code names. The

1997 SIC manual, which defines industries in accordance with the composition and structure of the economy and attempts to cover

the entire field of economic activities, is the reference source for SIC code assignments.


25/32

23

TABLE 3

Descriptive Statistics on the

Center For Financial Research and Analysis (CFRA) Sample

Panel A: Fiscal Quarter Prior to CFRA

Report

Full

Sample

% of

Sample 1994 1995 1996 1997

First Fiscal Quarter 69 18.50% 9 19 22 19

Second Fiscal Quarter 110 29.49% 17 30 28 35

Third Fiscal Quarter 89 23.86% 9 26 30 24

Fourth Fiscal Quarter 105 28.15% 14 29 32 30

TOTALS 373 100.00% 49 104 112 108

Panel B: Stock Exchange Listings

Full

Sample

% of

Sample 1994 1995 1996 1997

New York 183 49.06% 23 54 55 51

NASDAQ / American / other 190 50.94% 26 50 57 57

TOTALS 373 100.00% 49 104 112 108

Panel C: Firm Size Proxy

Full

Sample

% of

Sample 1994 1995 1996 1997

CFRA Size Designation:Small to Mid cap stocks 159 49.07% n/r 53 56 50

CFRA Size Designation:Mid to Large cap stocks 165 50.93% n/r 51 56 58

TOTALS 324 100.00% n/r 104 112 108

The table provides additional descriptive information on the 373 observations contained in CFRA research reports with available data

from February 1994 (first research report by CFRA using the alert format) to December 1997. In panel A, the fiscal quarter of

the 373 observations prior to inclusion the CFRA report are reported. In panel B, the stock exchange listings are reported. In panel

C, the distributions across a firm size proxy (CFRA Report size designations) are reported. CFRA did not report size designations

for the reports in 1994. Those without size designations in 1994 are shown as n/rfor not reported. Of the 373 observations,

approximately 57 percent have calender year-end fiscal years.


26/32

24

[ ][ ]EPS EPS

ABS Average EPS EPS EPS EPS

t t

t t t t

+ + +

4

1 2 3 4( ) ( ) ( ) ( )

TABLE 4

Variable Definitions

Variable Definition

ROAt [ ]( )

( ) ( )

Operating Income


t

t t+ 1

PMt( )

( )

Operating Income

Sales

t

t

ATOt [ ]( )

( ) ( )

Sales


t

t t+ 1

)EPSt

The table defines the variables used to measure operating performance. The financial statement data are taken from the quarterly

database of Standard & Poors 1999 CompustatIndustrial and Research files (Primary, Supplemental, Tertiary; Full-Coverage;

and Industrial Research Files). The operating performance measures are calculated for the four quarters in the pre-publication

period for the control groups matched on two-digit SIC codes and pre-publication performance as well as in the post-publication

period for the CFRA sample of firms. Tests of differences are then performed on the average values of the four operating

performance measures in both the pre- and post-publication periods.

ROAt = operating income before depreciation at the end of quartertdivided by average total assets at the end of quartert. Theaverage is the calculated using beginning and ending values of total assets at the end of quartert.

PMt = operating income before depreciation divided by sales(net) at the end of quartert.

ATOt = sales (net) at the end of quartertdivided by average assets at the end of quartert.

)EPSt = the change in earnings per share for quartert, defined as (EPSt - EPSt-4), to yield a seasonally differenced change in

earnings per share. The change is deflated by (absolute value) of the average earnings per share in the four quarters

prior to quartert. We define the earnings as income available to common equity before extraordinary items and

discontinued operations. We scale earnings by common shares for primary earnings per share (adjusted for stock

dividends and splits) at the end of quartert.


27/32

25

TABLE 5

Tests of Abnormal Operating Performance on the

Center For Financial Research and Analysis (CFRA) Sample

Panel A: Operating Performance of CFRA Sample: Post-Publication Date Performance

Variable Calculation Method N Mean Std. Dev. % Mean > 0 Median

ROA annualized 373 0.1522 0.1420 93.03% 0.1421

PM average 373 0.1038 0.6606 91.42% 0.1230

ATO annualized 373 1.3989 0.8683 100.00% 1.2841

)EPS average 373 -0.4882 1.9293 58.73% 0.0740

Panel A: Operating Performance of Control Sample: Two-digit SIC and Pre-Publication Performance Matched

Variable Calculation Method N Mean Std. Dev. % Mean > 0 Median

ROA annualized 373 0.1826 0.1215 96.25% 0.1617

PM average 373 0.1435 0.1821 94.91% 0.1320

ATO annualized 373 1.4276 0.8272 100.00% 1.3795

)EPS average 373 -0.1051 1.5298 62.47% 0.0875

Panel C: Test of Differences (prediction: CFRA Sample < Control Sample)a

Variable Calculation Method

Difference

in Means

test

statistic

Difference

in Medians

test

statistic

ROA annualized -0.0304 -4.20 *** -0.0196 -3.15 ***

PM average -0.0397 -3.69 *** -0.0090 -3.29 ***

ATO annualized -0.0287 -1.53 * -0.0954 -1.33 *

)EPS average -0.3831 -2.54 *** -0.0135 -1.43 *

The firm performance variables are tested for differences between the CFRA sample and the control group matched on pre-CFRA

publication operating performance and industry. See table 4 for variable definitions. To better approximate annual firm performance

measure, ROA and ATO are reported as annualized amounts. The table shows Winzorized means and standard deviations using a 2.5

percent Winsorization process in each tail since a few outliers skew the data. Tests of differences in means and medians are

performed by calculating Wilcoxon (for means) and median scores tests.

a ***, **, * denote significance at the 1%, 5%, and10% levels, respectively, using the Wilcoxon signed rank tests for differences in

means and median scores tests for the differences in medians. Significance levels are based on one-tailed tests.


28/32


29/32

27

TABLE 7

Tests of Differences in Buy-and-Hold Abnormal Returns (BHAR) in the Pre- and Post-CFRA

Report Periods after Controlling for Regression to the Mean

Average 12-Month

Buy-and-Hold Abnormal

Returns

(Pre-Event)

CFRA Sample

Statistics

(Pre-Event

Matching)

Differences in Post-

Event BHAR Between

CFRA Sample and

Control Group

Test: CFRA - Control

Deciles

Control

Group

CFRA

Sample N %

10 (high) 1.1482 0.9523 84 23.73% -0.313 ***

9 0.3945 0.3899 59 16.67% -0.138 **

8 0.2110 0.2232 52 14.69% -0.157 **

7 0.0943 0.0911 35 9.89% -0.032

6 0.0065 0.0063 21 5.93% -0.050

5 -0.0671 -0.0704 21 5.93% -0.062

4 -0.1516 -0.1449 23 6.50% 0.053

3 -0.2756 -0.2590 30 8.47% -0.117

2 -0.4745 -0.4470 23 6.50% -0.269 ***

1 (low) -0.9761 -0.8217 6 1.69% -0.036

Full

Sample 354 100% -0.155 ***

The control sample consists ofCRSPfirms (other than CFRA sample firms) in the same industries as the CFRA sample. For the

CFRA sample, we calculate twelve-month buy-and-hold abnormal returns (BHARs) ending the month before inclusion in a CFRA

research report. CFRA sample firms are matched to a pre-event price performance control group consisting of firms in the same

industry. The event is defined as inclusion in a CFRA research report. Subsequent BHARs over the twelve months beginning with

inclusion in CFRA research reports are compared against the stock price performance of the control group. A size (market

capitalization) portfolio return index (SI) is used to estimate abnormal returns.

*, **, *** denote significantly different from zero in a one-tailed test at 10%, 5%, 1%, respectively.


30/32

28

TABLE 8

Stock Price Effects Around the Publication Dates for the Center For Financial Research and

Analysis (CFRA) Research Reports: Evidence on CFRA Size Designations in Research Reports

Cumulative Abnormal Returns (CAR) for Windows in Event Timea

Panel A: CFRA Size Designation: Small to Mid Cap Stocks

Event Windows

Relative to CFRA

Publication Dates

No. of

Obs.

MAR-VW

Average

CAR (%) t-Statistic a

SI

Average

CAR (%) t-Statistic

(0, +1) 165 -2.30 -4.58 *** -2.13 -3.21 ***

(0, +62) Q1 165 -7.65 -2.71 *** -6.59 -1.77 **

(0, +125) Q1 & Q2 165 -14.10 -3.53 *** -11.71 -2.22 **

(0, +187) Q1, Q2, & Q3 165 -14.24 -2.92 *** -9.94 -1.54 *

(0, +250) one year 165 -23.81 -4.22 *** -17.11 -2.30 **

(0, +500) two years 151 -48.28 -5.78 *** -26.61 -2.44 ***

Panel B: CFRA Size Designation: Mid to Large Cap Stocks

Event Windows

Relative to CFRA

Publication Dates

No. of

Obs.

MAR-VW

Average

CAR (%) t-Statistic

SI

Average

CAR (%) t-Statistic

(0, +1) 159 -0.39 -0.75 -0.41 0.88

(0, +62) Q1 159 -1.00 -0.63 -0.62 -0.94

(0, +125) Q1 & Q2 159 -6.89 -1.77 ** -5.71 -1.53 *

(0, +187) Q1, Q2, & Q3 159 -11.23 -2.20 ** -9.13 -2.00 **

(0, +250) one year 159 -14.83 -2.51 *** -11.93 -2.27 **

(0, +500) two years 153 -26.41 -3.87 *** -15.13 -2.65 ***

Panels A and B report the results for firms partitioned by size based on CFRA research report size designations. Beginning with the

first report in 1995, CFRA reports use a size classification of small to mid and mid to large cap (i.e., market capitalization levels) of

the stocks. Median and average CARs are reported for each event window and market index used to estimate the abnormal returns to

the sample. The estimation techniques for abnormal returns reported in the study utilize two alternative specifications of return indices.

A size (market capitalization) portfolio return index (SI) and a value-weighted return index (VW) are used to estimate CFRA sample

abnormal returns.

a The specifications for the abnormal returns and the test statistic for parametric procedures are described in Appendix 1. We also

perform nonparametric tests based on the Wilcoxon signed rank test and the rank test described in Corrado (1989). Using the

nonparametric test procedures, the conclusions are qualitatively similar to those using the results of parametric tests.

*, **, *** denote significantly different from zero in a one-tailed test at 10%, 5%, 1%, respectively.


31/32

29

-250

-125 0

125

250

(250 days approximates one year)

Trading Days Relative to CFRA Report

0%

5%

10%

15%

20%

25%

30%

CumulativeAbnormalReturn

CAR (SI)

Figure 1

Cumulative Abnormal Returns From One Year Prior to One Year After Publication of the

Center For Financial Research and Analysis (CFRA) Research Reports

The figure shows the cumulative (average) abnormal returns of the portfolio of firms reported in CFRA reports during the sample

period. The estimation techniques for abnormal returns reported in the study utilize two alternative specifications of return indices. In

this figure we document abnormal returns using a size (market capitalization) portfolio return index (SI). We assume that a period

covering 250 trading days is approximately equal to one fiscal year. The sample of CFRA reports span February 1994 to the December

1997 report dates. The market adjusted (average) abnormal returns are cumulated beginning with one year before inclusion

(approximated by the 250 trading days before CFRA publication date) in a CFRA report up to one year after. The vertical line at

trading date zero represents an approximation of the CFRA report release dates.


32/32

30

-250

-125 0

125

250



-10%

0%

10%

20%

30%

40%

50%


1994

-250

-125 0

125

250



-10%

0%

10%

20%

30%

40%

50%


1995

-250

-125 0

125

250



-10%

0%

10%

20%

30%

40%

50%


1996

-250

-125 0

125

250



-10%

0%

10%

20%

30%

40%

50%


1997

Figure 2

Cumulative Abnormal Returns From One Year Prior to One Year After Publication of the

Center For Financial Research and Analysis (CFRA) Research Reports

For Each Year in the Sample Period

The figure shows the cumulative (average) abnormal returns of the portfolio of firms reported in CFRA reports during the sample

period. Each year of the sample period is shown. See Figure 1 for additional descriptive information on the market benchmark used.

The market adjusted (average) abnormal returns are cumulated beginning with one year before inclusion (approximated by the 250

trading days before CFRA publication date) in CFRA reports and ending with one year after inclusion in the CFRA reports. The

vertical line at trading date zero represents an approximation of the CFRA report release dates.

using fundamental analysis to assess earnings quality- evidence from the center for financial...

Documents