comp sep 2 2008
TRANSCRIPT
-
8/8/2019 Comp Sep 2 2008
1/46
The Benefits of Firm Comparability
Gus De Franco
Rotman School of Management, University of Toronto
105 St. George StreetToronto, Canada, M5S 3E6
Phone: (416) 978-3101
Email: [email protected]
S.P. Kothari
MIT Sloan School of Management50 Memorial Drive E52-325
Cambridge, MA 02142-1347
Phone: (617) 253-0994Email: [email protected]
Rodrigo S. Verdi
MIT Sloan School of Management50 Memorial Drive E52-325
Cambridge, MA 02142-1347
Phone: (617) 253-2956Email: [email protected]
August 27, 2008
ABSTRACT
We develop a new metric of and study the capital market consequences of firm comparability.
Investors, regulators, academics, and researchers all emphasize the importance of comparability.
However, an empirical construct of financial statement comparability is typically not specified.
More importantly, little evidence exists on the benefits of comparability to users. We fill thesegaps. We find that analyst following is increasing in comparability, and that comparability is
positively associated with forecast accuracy and negatively related to bias and dispersion in
earnings forecasts. Our results suggest comparability enhances a firms informationenvironment, a benefit to capital market participants.
______________________________
We appreciate the helpful comments of Stan Baiman, Rich Frankel, Wayne Guay, Thomas Lys, Jeffrey Ng, Ole-
Kristian Hope, Shiva Rajgopal (a discussant), Shiva Shivramkrishnan, Shyam Sunder, and workshop participants atBarclays Global Investors, Columbia University, University of Florida, University of Houston, London Business
School, MIT, and the University of Toronto. We thank I/B/E/S Inc. for the analyst data, available through the
Institutional Brokers Estimate System. I/B/E/S offers access to data as a part of their broad academic program toencourage earnings expectation research. We gratefully acknowledge the financial support of MIT Sloan and the
Rotman School, University of Toronto. Part of the work on this article was completed while Gus De Franco was a
Visiting Assistant Professor at the Sloan School of Management, MIT.
-
8/8/2019 Comp Sep 2 2008
2/46
1
The Benefits of Firm Comparability
1. Introduction
Several factors point toward the importance of comparability of financial information
across firms in financial analysis. According to the Securities and Exchange Commission (SEC)
(2000), when investors judge the merits of investments and comparability of investments,
efficient allocation of capital is facilitated and investor confidence nurtured. The usefulness of
comparable financial statements is underscored in the Financial Accounting Standards Board
(FASB) accounting concepts statement. Specifically, the FASB (1980, p. 40) states that
investing and lending decisions essentially involve evaluations of alternative opportunities, and
they cannot be made rationally if comparative information is not available (our emphasis).1
Financial statement analysis textbooks almost invariably stress the importance of comparability
across financial statements in judging a firms performance using financial ratios, including
ratios for the same firm in prior years, ratios for selected firms in the same industry, or ratios
based on industry averages.
2
For instance, Stickney and Weil (2006, p. 189) conclude that,
Ratios, by themselves out of context, provide little information. Analyst reports routinely
include a list of comparable or peer firms (see evidence in section 2 below). In these
reports, analysts typically evaluate the firms current valuation and/or predicted valuation on the
basis of a comparative analysis of the (past, current, and projected) financial performance of a set
of comparable, peer, or similar firms.
1 As an additional example of the importance of comparability in a regulatory context, comparability is one of
three qualitative characteristics of accounting information included in the accounting conceptual framework (alongwith relevance and reliability). Further, according to the FASB (1980, p. 40), The difficulty in making financial
comparisons among enterprises because of the different accounting methods has been accepted for many years as
the principal reason for the development of accounting standards. Here, the FASB argues that users demand for
comparable information drives accounting regulation.2
See, e.g., Libby, Libby and Short (2004, p. 707), Stickney, Brown, and Wahlen (2007, p. 199), Revsine,Collins, and Johnson (2004, pp. 213-214), Wild, Subramanyam, and Halsey (2006, p. 31), Penman (2006, p. 324),
White, Sondhi, and Fried (2002, p. 112), and Palepu and Healy (2007, p. 5-1).
-
8/8/2019 Comp Sep 2 2008
3/46
2
Despite the importance of comparability: i) The literature lacks an empirical measure of
financial statement comparability; and, ii) Little evidence exists on the benefits of financial
statement comparability to users. This paper fills these gaps. We develop two measures of and
study the benefits of comparability to investors (proxied by sell-side analysts). A key innovation
is the construction of empirical, firm-specific, output-based, quantitative measures of
comparability. Our measures contrast with qualitative, input-based definitions of comparability
such as business activities or accounting methods. Further, our measures are intended to capture
comparability from the perspective of users, such as investors or analysts, who evaluate
historical firm performance, predict future firm performance, or make other decisions using
financial statement information. As a proxy for the users benefits, we study the properties of the
observable outputs (i.e., earnings forecasts) of sell-side analysts, which are publicly available for
a long period.
We measure comparability based on financial statement outputs. In particular, our
comparability measures use (arguably) the primary output of financial reporting: earnings. The
first measure, which we label accounting comparability, is based on the idea that comparable
firms experiencing similar economic events, as proxied by stock returns, should report similar
accounting earnings. The second measure, which we label earnings comparability, is based on
the strength of the historical covariance between a firms earnings and the earnings of other firms
in the same industry, as evidenced by theR2
values. If firms experience similar economic shocks
and account for the economic transactions in a similar way, then such firms earnings should
covary over time. To focus on the similarity in accounting for the events, we control for
similarity of business models and economic shocks when using earnings comparability. While
our primary focus is on creating comparability measures at the firm level, we also produce
-
8/8/2019 Comp Sep 2 2008
4/46
3
measures of relative comparability at the firm-pair level, in which a measure is calculated for
all possible pairs of firms in the same industry. These measures are described in detail below.
Before proceeding to the tests of our hypotheses, because they are new, we study the
construct validity of our earnings comparability measure via an analysis of the textual contents of
a hand-collected sample of analysts reports. We find that the likelihood of an analyst using
another firm in the industry (say, firm j) as a benchmark when analyzing a particular firm (say,
firm i) is increasing in the comparability between the two firms. This shows that our measures of
comparability are correlated with the actual mention of comparable firms in analyst reports,
bolstering the construct validity of our comparability metrics.
We then document the effect of comparability on the properties of analysts outputs.
Given a particular firm, we hypothesize that the availability of information about comparable
firms (as captured by our comparability measures) lowers the cost of acquiring information, and
increases the overall quantity and quality of information available about the firm. Our results are
consistent with the hypothesis. We find that comparability facilitates analyst following.
Specifically, the likelihood that an analyst covering a particular firm (e.g., firm i) would also be
covering another firm in the same industry (e.g., firm j) is increasing in the earnings
comparability between the two firms. Further, firms classified as more comparable are also
covered by more analysts. This result suggests that analysts indeed benefit, i.e., face lower costs,
from higher comparability.
We also find that comparability enables analysts to issue more accurate and less biased
earnings forecasts. Thus, comparability helps analysts more accurately forecast earnings and that
improvement comes, at least in part, through a reduction in forecast bias (i.e., optimism). These
results are consistent with analysts facing lower costs of acquiring information from sources
-
8/8/2019 Comp Sep 2 2008
5/46
4
other than management. This would reduce analysts reliance on managements private
information, and hence decrease their incentive to strategically include optimistic bias in their
forecasts. Last, we document that one of our comparability measures, accounting comparability,
is negatively related to analysts forecast dispersion, consistent with the availability of superior
public information about highly-comparable firms and an assumption that analysts use similar
forecasting models. There is no evidence of a relation between dispersion and the other
comparability measure, earnings comparability.
Our study contributes to the literature in a number of ways. We develop a measure of
comparability that likely captures users notions of comparability and the benefits of
comparability to them. The ability to forecast future earnings is a common task for users such as
investors and analysts, particularly those engaged in valuation. Improved accuracy and reduced
bias, for example, represent tangible benefits to this user group. Further, the results of increased
analyst following, greater forecast accuracy, lower bias, and less dispersion collectively are
consistent with comparability enriching firms information environment, which provides a
tangible benefit for firms with higher comparability. While comparability is generally accepted
as a valuable attribute, there is little evidence beyond this study that would empirically confirm
this widely-held belief.
The paper proceeds as follows. Section 2 outlines our predictions that comparability
provides benefits to analysts. Section 3 defines our comparability measures. We provide
descriptive statistics and construct validity tests of our measures in Section 4. Section 5 presents
the results of our empirical tests. The last section concludes.
2. Hypotheses: The effect of comparability on analysts
In this section, we develop hypotheses about the effect of comparability on analysts and
-
8/8/2019 Comp Sep 2 2008
6/46
5
therefore on the properties of their forecasts. As mentioned above, any lesson on financial
statement analysis emphasizes the difficulty in drawing meaningful inferences about a financial
measure unless there is a comparable benchmark. FASB (1980, p. 40) echoes this point.
Implicit is the idea that better benchmarks enable superior inferences (e.g., better evaluation of
firm performance, better prediction of next years performance, etc.). More comparable firms
constitute better benchmarks for each other and lead to a higher quality information set for these
firms. As an additional consideration, information transfer among comparable firms is likely to
be greater. We thus expect the effort exerted by analysts to understand and analyze the financial
statements of firms with other comparable firms to be lower than for firms without other
comparable firms. As a result of this change in analysts cost of analyzing a firm, we investigate
two dimensions of changes in analysts behavior the number of analysts following a firm and
the properties of analyst forecasts.
Our first hypothesis examines whether comparability enhances analyst coverage. As
discussed in Bhushan (1989) and in Lang and Lundholm (1996), the number of analysts
following a firm is a function of the analysts costs and benefits. We argue that, ceteris paribus,
since the cost to analyze firms with other comparable firms is lower, more analysts should cover
these firms. Our first hypothesis (in alternate form) is:
H1: Ceteris paribus, comparability is positively associated with analyst coverage.
The null hypothesis is that the better information environment associated with higher-
comparability firms will decrease the investor demand for analyst coverage. That is, the benefits
to analysts will decrease as well. However, the literature on analysts suggests that analysts
primarily interpret information as opposed to convey new information to the capital markets
(Lang and Lundholm 1996; Francis, Schipper, and Vincent 2002; Frankel, Kothari, and Weber,
-
8/8/2019 Comp Sep 2 2008
7/46
6
2006; De Franco, 2007). Further, Lang and Lundholm (1996) and others in the literature find
that analyst coverage is increasing in firm disclosure quality. These empirical findings suggest
that an increase in the supply of information results in higher analyst coverage, consistent with
the lower costs of more information outweighing the potentially lower benefit of decreased
demand. These findings in the literature support our signed prediction.
Our second set of hypotheses examines the association between comparability and the
properties of analyst earnings forecasts. The first property we examine is forecast accuracy. The
higher quality information set associated with higher comparability should facilitate analysts
ability to forecast firm i's earnings and lead to improved forecast accuracy. For example, the
existence of comparable firms could allow analysts to better explain firms historical
performance or to use information from comparable firms as an additional input in their earnings
forecasts. Our hypothesis 2a (in alternative form) is:
H2a: Ceteris paribus, comparability is positively associated with analyst forecast
accuracy.
Second, prior research finds analysts long-horizon forecasts are optimistic on average
(e.g., OBrien, 1988, and Richardson et al., 2004).3
Francis and Philbrick (1993), Das et al.
(1998), and Lim (2001) show that part of the bias in analysts forecasts is explained by analysts
adding optimism to their forecasts to gain access to managements private information, which
helps improve forecast accuracy.4
If information from comparable firms serves as a substitute
for information from management, then the incentive to strategically add optimistic bias to gain
access to management is reduced. Further, if more objective information from comparable firms
is available, it is easier to identify when analysts act strategically (i.e., to catch them) regardless
3Analyst optimism has decreased over time and is more-pronounced for longer-horizon forecasts and it seems
to be driven by relatively few observations (Brown 2001; Lim 2001; Gu and Wu 2003; Richardson et al. 2004).4 Recent research by Eames et al. (2002) and Eames and Glover (2003) question these results.
-
8/8/2019 Comp Sep 2 2008
8/46
7
of the reason for the optimism, which hence increases the cost to the analyst of this optimistic
behavior. Therefore analysts forecasts of higher-comparability firms should be less optimistic.
We state this prediction as hypothesis 2b (in alternate form):
H2b: Ceteris paribus, comparability is negatively related to analyst forecastoptimism.
For both these accuracy and optimism predictions, a counter argument is that if forecasts
for the comparable firms are noisy or biased, then increased comparability could lead to less
accurate and more biased forecasts. We expect this effect to reduce the ability of our tests to
support these two predictions.
Third, we investigate the relation between comparability and analyst forecast dispersion.
If analysts have the same forecasting model, and if higher comparability implies the availability
of superior public information, then an analysts optimal forecast will place more weight on
public information and less on her private information. This implies comparability will reduce
forecast dispersion. Our hypothesis 2c (in alternative form) is:
H2c: Ceteris paribus, comparability is negatively associated with analyst forecastdispersion.
We acknowledge that superior public information via higher comparability could
generate more dispersed forecasts, which would support the null of hypothesis 2c. The intuition
here is that if some analysts process a given piece of information differently from other analysts,
then the availability of greater amounts of public information for comparable firms will generate
more highly-dispersed forecasts. A number of theoretical studies predict such a phenomenon.
Harris and Raviv (1993) and Kandel and Pearson (1995) develop models in which disclosures
promote divergence in beliefs. Kim and Verrecchia (1994) allow investors to interpret firm
-
8/8/2019 Comp Sep 2 2008
9/46
8
disclosures differently, whereby better disclosure is associated with more private information
production.
3. Empirical measures of comparability
The term comparability in accounting textbooks, in regulatory pronouncements, and in
academic research is defined in broad generalities rather than precisely. In the first two
subsections, we motivate and explain how we compute our two empirical measures of
comparability, respectively. In the third subsection, we discuss our comparability measures and
contrast it with other measures used in the literature that are indirectly related to ours.
3.1. Measure of accounting comparability
Accounting is the mapping from economic transactions to financial statements. As such,
it can be represented as follows:
Financial Statementsit= it(Economic Transactionsit) (1)
wherefit( ) represents the accounting of firm i during period t.
We define accounting comparability as the similarity with which two firms translate the
same economic events to the financial statements. That is, two firms with comparable
accounting should have similar functionsf( ) such that, given a set of economic transactions X,
firm j produces similar financial statements to firm i. To operationalize this idea, we examine
the relation between earnings, one important summary financial statement measure, and returns,
a proxy for the net economic effect of transactions. For each firm we estimate the following
equation using 16 previous quarters of data:
Earningsit= i + i Returnsit+ it. (2)
-
8/8/2019 Comp Sep 2 2008
10/46
9
where Earnings is the ratio of quarterly net income before extraordinary items (data8) to the
average total assets (data44), taken from the Compustat Quarterly file, andReturns is the stock
price return during the quarter, taken from CRSP.
Under the framework in Equation 1, i and i proxy for the accounting function f( ) for
firm i. Similarly, the accounting function for firmj is proxied by j and j (using earnings andreturns from firmj). Next we calculate the predicted earnings for firm i, given the same set of
economic transactions, using the accounting functions of both firm i andj respectively.
E(Earnings)iit= i + i Returnsit (3)
E(Earnings)ijt= j + j Returnsit (4)
where E(Earnings)iit is the expected earnings of firm i givenfirm is function and firm is returns
in period tand E(Earnings)ijt is the expected earnings of firm i given firm js function and firm
is returns in period t. We restrict the sample to firms whose fiscal year ends in March, June,
September, or December. This ensures that i andj firms earnings are measured at the end of the
same fiscal quarter.
We then define accounting comparability between firm i and j as the negative value of the
average difference between the expected earnings for firm i under firm is andjs functions.
|)()(|*16/115
ijt
t
t
iitijt EarningsEEarningsECompAcct =
(5)
Higher values indicate higher accounting comparability. We estimate accounting comparability
for each firm i firm j combination for J firms within the same SIC 2-digit industry
classification at the end of December for each year. We exclude holding firms. In some cases,
Compustat contains financial statements for both the parent and the subsidiary company, and we
want to avoid matching such two firms. We exclude ADRs and limited partnerships because our
-
8/8/2019 Comp Sep 2 2008
11/46
10
focus is on corporations domiciled in the United States.5
We also exclude firms that have names
highly similar to each other using an algorithm that matches five-or-more-letter words in the firm
names, but avoids matching on generic words such as hotels, foods, semiconductor, etc.
Finally, we restrict the sample to industries with at least 20 firms per year based on the SIC two-
digit classification.
In addition, we provide a firm-year measure of accounting comparability by aggregating
the firm i firm j CompAcctijt for a given firm i. Specifically, after estimating accounting
comparability for each firm i firmj combination, we rank allJvalues ofComp Acctijt for each
firm i from the highest to lowest. We then define Comp4 Acctit as the average CompAcctijt of the
four firmjs with the highest comparability for firm i during period t. (Results are similar if we
use the top ten firmjs instead.) Similarly, we define CompInd Acctit as the average CompAcctijt
for all firms in the same industry as firm i during period t. Firms with high Comp4 Acctand
CompInd Acct are firms for which the accounting function is more similar to a peer group of
firms and to the industry respectively.
3.2. Measure of earnings comparability
If two firms are comparable, they are more likely to experience similar economic shocks.
For instance, a change in input prices or shifts in consumer demand for firms with similar
business models should translate into similar changes in economic profitability. Comparable
firms are also likely to account for economic transactions in a similar way. This leads to an
expectation that earnings for comparable firms will covary over time. While this scenario leads
to positive covariance in earnings, it is also possible that comparable firms could have a negative
earnings covariance over time. For example, if two competitors compete for market share, one
5 Specifically if the word Holding, Group, ADR, or LP (and associated variations of these words) appear in the
firm name on Compustat, the firm is excluded.
-
8/8/2019 Comp Sep 2 2008
12/46
11
firms economic gain could be the other firms economic loss. In contrast, if business models are
different, if firms are sensitive to different types of shocks, or if their accounting policies differ,
then we would expect such firms earnings not to covary over time.
To operationalize this intuition and quantify the degree of similarity between pairs of
firms, we calculate the historical covariance of quarterly earnings among all possible pairs of
firms in the same industry. Ceteris paribus, firms with higher comparability are firms whose
earnings covary more with the earnings of its peers firms. More specifically, using 16 quarters
of earnings data we estimate:
Earningsit= 0ij + 1ij Earningsjt+ ijt. (6)We define our firm i firmj measure of earnings comparability (Comp Earnijt) as the adjusted
R2
from this regression. (Hereafter, we use R2
to mean adjusted R2.) Higher values indicate
higher comparability. In order to avoid the influence of outliers on theR2
measure, we remove
observations in whichEarnings for firm i is more than three standard deviations from the mean
value of the 16Earnings observations for firm i used to estimate Equation 6. Following a similar
procedure and scope to our development of the Comp Acctvariables above we obtain a Comp
Earnijt for firm i firm j pair forJfirms in the same 2-digit SIC industry with available data.
Comp4Earnit is the averageR2
for the four firmjs with the highestR2s. CompIndEarnit is the
average R2
for all firms in the industry. Firms with high Comp4 Earn and CompInd Earn are
firms for which earnings covary more with earnings from a peer group of firms and from the
industry respectively.
An issue with ourComp Earn variable is that theR2s we measure could be mainly due to
similar business models and economic shocks, rather than comparable financial statements, our
primary construct of interest. When we use Comp Earn in our tests, we hence include controls
-
8/8/2019 Comp Sep 2 2008
13/46
12
for similar business model and economic shocks. One control is based on cash flow from
operations, which captures covariation in near-term economic shocks. Comp CFO is created in
an identical manner to Comp Earn except that in Equation 6 we replace Earnings with CFO,
which is the ratio of cash flow from operations quarterly (data108) to the average total assets
(data44). Our second control is stock returns, which capture covariation in economic shocks
related to cash flow expectations over long horizons. Comp Retis also defined in a manner that
parallels the construction ofComp Earn. In Equation 6, instead ofEarnings we use monthly
stock returns taken from the CRSP Monthly Stock file, and instead of 16 quarters we use 48
months.We then calculate pairwise firm i firm j values (Comp CFOij and Comp Retij) and
aggregated firm level values (Comp4 CFO, CompInd CFO, Comp4 Ret, and CompInd Ret) for
these measures.
3.3. Discussion of measures
In developing our two measures, we adopt the perspective of financial statements users
in particular, analysts and focus on earnings, a financial statement output. Other research has
examined comparable inputs such as similar accounting methods, business activities, or industry
membership. As an example, Bradshaw and Miller (2007) study whether international firms that
intend to harmonize their accounting with US GAAP adopt US GAAP accounting methods.
DeFond and Hung (2003) argue that accounting choice heterogeneity (e.g., differences in
inventory methods such as LIFO versus FIFO) increases the difficulty in comparing earnings
across firms.
Other existing measures of comparability are based mainly on similarities in cross-
sectional levels of contemporaneous measures (e.g., return on equity, firm size, price multiples)
at a single point in time. Joos and Lang (1994) study the comparability of accounting data in a
-
8/8/2019 Comp Sep 2 2008
14/46
13
European setting. They expect that improved accounting comparability between countries will
result in smaller differences between accounting measures of profitability (i.e., ROE), between
valuation multiples of accounting data (i.e., Earnings/Price; Book value/Market value of equity),
and between the degree of association between accounting and stock data (i.e., value relevance).
Land and Lang (2002) also focus on comparing valuation multiples across countries. These
measures are typically estimated in aggregate at the country level. Notable exceptions are
studies that examine returns to pairs trading, such as Papadakis and Wysocki (2007). They
identify pairs of similar firms using the average difference in daily normalized price over a 12-
month period. In effect, their measure captures a blend of similar levels andsimilar covariation
over time. Compared to the aggregated measures described above, our measure is dynamic,
capturing similarities over time, and isfirm-specific.
OurComp Earn measure in particular is indirectly related to four others measures. First,
relative performance evaluation theory suggests filtering out of noise caused by factors unrelated
to managements actions on performance (see, e.g., Holmstrm (1979), Antle and Smith (1986),
Banker and Datar (1989), and Dikolli, Hofmann, and Pfeiffer (2007)). One way to identify these
common shocks is to examine the correlation in performance (e.g., annual stock returns) between
a firm and its industry or peer group.
Second, the literature (e.g., Lipe 1990) has established the time-series concept of earnings
predictability in which earnings are regressed on previous-period earnings. OurComp Earn
measure is a cross-sectional version of predictability. While earnings predictability or persistence
measures have been around in the literature for quite some time (see, e.g., Lipe 1990, and
Francis, LaFond, Olsson, and Schipper, 2004), their use in developing a comparability measure
in this study is unique.
-
8/8/2019 Comp Sep 2 2008
15/46
14
Third, Piotroski and Roulstone (2004), and Chan and Hameed (2006), among others,
study stock price synchronicity, which is based on the R2
from a regression of firms stock
returns on market and industry stock returns. They are inherently interested in the type of
information (firm, industry, or market) impounded in stock prices. OurComp Earn measure is
based solely on accounting data, and is hence not sensitive to flows of non-accounting
information, to investors interpretation, or to assumptions about market efficiency. Our
measure could also be applied to private firms.
Fourth, an older literature studies the accounting beta, which captures the covariance
between a firms earnings with the earnings at the industry or market level. Brown and Ball
(1967) show that firm earnings can be explained in part by the earnings of other firms in the
same industry and the earnings of all firms in the market. This research focuses on documenting
that the market beta (i.e., covariance with the market portfolio in the Sharpe-Lintner Capital
Asset Pricing Model) is positively related to the accounting beta (see, e.g., Beaver, Kettler, and
Scholes 1970; Beaver and Manegold 1975; Gonedes 1973, 1975). In contrast, ourComp Earn
measure focuses on the covariance of earnings between i-j firm pairs within an industry.
4. Estimating and validating a measure of comparability
4.1. Estimating comparability
Our sample period spans the years 1993 to 2006. Table 1 presents descriptive statistics
for our measures of comparability. Panels A and B (C and D) present descriptive statistics and
correlations for the pairwise firm i firmj (aggregated firm i) comparability measures. In Panel
A the sample size for the pairwise comparability is 3,592,745 firm ifirm j-year observations.
The mean value forComp Acctij is -1.55 suggesting that the average error in quarterly earnings
-
8/8/2019 Comp Sep 2 2008
16/46
15
between firm i and firmj functions is 1.55% of total of assets. The mean value forComp Earnij
is 11.19 suggesting that on average firmjs earnings explains 11% of firm is earnings. Similar
values forComp CFOij and Comp Retij are 8% and 6%. Panel B presents correlations among
these variables. The correlations are all positive and significant although the magnitudes are
small, ranging from 0.01 between Comp Acctij and Comp Retij to 0.08 between Comp Earnij and
Comp CFOij.
Panels C and D present the descriptive statistics for the firm i comparability measures.
The sample size is 27,972 firm-year observations. The mean value forComp4 Acct is -0.26
suggesting that the average error in quarterly earnings for the top four firms with the highest
accounting comparability to firm i is 0.26% of total of assets. By construction, this value is
higher than the mean value forCompInd Acctwhich is -1.14. The mean value forComp4 Earn
is 52.36 meaning that the earnings of the top four comparable firms explain, on average, 52% of
firm i's earnings. Mean values for Comp4 CFO and Comp4 Ret are 42.83% and 24.76%
respectively. Panel D presents pairwise correlations among these variables. The correlations are
generally positive particularly for the top 4 firm and average industry versions of the same
measure. For example, the Pearson correlation between Comp4 Acctand CompInd Acctis 0.89.
[Table 1]
4.2. Validating our comparability measures
In this section, we test the construct validity of our comparability measures. The test
implicitly assumes that, for any given firm, analysts know the identity of comparable firms (if
any). This seems reasonable because analysts have access to a broad information set about each
firm, which goes beyond the historical financial statements, and includes firms business models,
competitive positioning, markets, products, etc.
-
8/8/2019 Comp Sep 2 2008
17/46
16
We make a testable prediction to fortify our measures construct validity. The prediction
relates to the important assumption underlying our measures, namely, that the relative ranking of
firm i - firmj comparability identifies a set of firms that analysts view as comparable to firm i.
We predict that if an analyst issues a report about firm i, then we expect the analyst to more
likely use firms that are comparable to firm i in her reports. The typical analyst report context
is that the analyst desires to evaluate the current, or justify the predicted, firm valuation multiple,
(e.g., Price/Earnings ratio), using a comparative analysis of peer firms valuation multiples as
benchmarks. Evidence from a test of this prediction suggests our measures of comparability are
reasonable.
The comparable firms that an analyst uses in her analysis are not available in a machine-
readable form in existing databases. We hand collect a sample of analyst reports from Investext
and manually extract this information from the reports. The cost of collecting this information
limits this analysis to one year of data. Reports are chosen as follows. We begin with all firms
(i.e., firm is) in our sample with available data for the year 2005. For these firms, we search
Investext to find up to three reports per firm i, each written by a different analyst and each
mentioning comparable or peer firms (i.e., potential firm js) in the report. We then record
the name and ticker of all firms used by the analyst as a peer or comparable firm for firm i. We
match these peer firms with Compustat using the firm name and ticker. In total, we obtain 1,000
reports written by 537 unique analysts for 634 unique firms. Each report mentions one or more
firms as comparable to the firm for which the analyst has issued the report.6
For our tests, we estimate the following logistic regression:
6 Part of the reason this process is labor intensive is because we do not know ex ante whether Investext covers
firm i, and because not all analysts discuss comparable firms in their analysis. For example, many reports represent
simple updates with no discussion of valuation methods. In other cases, analysts rely more heavily on a discountedflow analysis or use historical valuation multiples to predict future multiples. We exclude reports on Investext that
are computer generated or not written by sell-side analysts.
-
8/8/2019 Comp Sep 2 2008
18/46
17
UseAsCompij = +1 Compijij +Controlsj + ij. (7)
UseAsComp is an indicator variable that equals one if an analyst who writes a reportabout firm i
refers to firmj as a comparable firm in her report, and equals zero otherwise. Compijij is one of
the comparability measures for each firm i firmj pair in our sample (i.e., Comp Acctij, Comp
Earnij, Comp CFOij, orComp Retij). We predict that the probability of an analyst using firmj
in her report is increasing in Compij. We use Size, Volume, Book-Market, ROA, and industry
fixed effects as control variables (results are the same if we use firm i fixed effects). Our choice
of these controls follows their common use by other researchers who match control firms with
treatment firms along these three dimensions (see, e.g., Barber and Lyon, 1996, 1997; Kothari,
Leone, and Wasley, 2005). In addition to the levels of these variables, we control for the
differences in characteristics between firm i and firmj. Differences are measured by the absolute
value of the difference between firm is and firmjs respective variables. The intuition for using
both levels and differences is as follows: An analyst who reports on firm i is more likely to use a
firm as a peer if the firm has similar (comparable) characteristics (e.g., similar size, growth
potential, and profitability) to firm i. This implies the larger the difference between firm i and
firmj, the less likely it is to be covered by the analyst. However, large, high-growth, and highly-
profitable firms are more likely to be covered by an analyst and recognized by investors, which
motivates us to also include the levels of these firm characteristics in the regression. Finally, we
include industry fixed-effects at the 2-digit SIC industry classification and cluster the standard
errors at the firm i level (results are similar if we use firm i fixed-effects instead).
Table 2 presents logistic regressions for model (7). In the first model, we include
accounting comparability (Comp Acctij) and the controls. The coefficient on Comp Acctij is
positive and marginally significant, suggesting that as the Comp Acctij increases, the odds of an
-
8/8/2019 Comp Sep 2 2008
19/46
18
analyst using firmj as a peer firm in a report about firm i increases. In the next three models we
include earnings comparability (Comp Earnij), controls for economic comparability (Comp
CFOij orComp Retij), and the remaining controls. In all cases the coefficient on Comp Earnij is
positive and statistically significant. Similar to Comp Acctij, this result suggests that, as Comp
Earnij increases, the odds of an analyst using firm j as a peer firm in a report about firm i
increases. In addition, Comp CFOij and Comp Retij also increase the likelihood that a firm is
used as a comparable firm in the report.7
In terms of economic significance, an increase from the
10th
to the 90th
percentile in Comp Acctij (Comp Earnij) is associated with an increase in the
probability of a firm being used as comparable firm in the report from 1.27% to 1.39% (1.11% to
1.27%), a relative increase of 10% (14%). For comparison, the same increase forComp Retij
(the strongest predictor in Table 2) is associated with an increase in probability from 0.73% to
2.49% suggesting that the effect is modest but also economically significant. Overall, the results
in Table 2 support the notion that an analyst who writes a report about a firm more likely chooses
benchmark firms that have higher values of comparability. This bolsters the construct validity of
our comparability measures.
[Table 2]
5. Empirical tests
5.1. Conditional analyst coverage
In this section, continuing with our previous analysis in which we use pairwise firm i -
firm j level comparability (as opposed to aggregated firm i level comparability), we provide
7In Table 2 we present three specifications for Comp Earnij depending on the inclusion of Comp CFOij or
Comp Retij. In the subsequent analysis we present only the full model but the results are robust to the other
specifications (i.e., including eitherComp CFOij orComp Retij).
-
8/8/2019 Comp Sep 2 2008
20/46
19
initial evidence of our first hypothesis that higher comparability facilitates analyst coverage.
This test is similar in spirit to the test in the previous section but now we use actual analyst
coverage instead of analyst use of comparable firms in analysts reports. We expect that the
likelihood of an analyst covering a particular firm (e.g., firm i) also covering another firm in the
same industry (e.g., firmj) is increasing in the comparability between these two firms. Hence, we
not only predict that higher earnings comparability leads to more analysts covering the firm (as
we do in the next section), but also specifically predict which other firms the analyst will follow.
We estimate the following logistic regression for each year of our sample:
CondCoverageikj = +1 Compijij +Controlsj + ikj. (8)CondCoverage is an indicator variable that equals one if analyst kwho covers firm i also covers
firmj, and equals zero otherwise. An analyst covers a firm if she issues at least one annual
forecast about the firm. As before, Compijij is one of the four comparability measures for each
firm i firmj pair in our sample (i.e., Comp Acctij, Comp Earnij, Comp CFOij, orComp Retij).
We predict that the probability of covering firmj is increasing in Compijij (i.e., 1 > 0).
In estimating Equation 8, we control for other factors motivating an analyst to cover firm
j by including determinants of analyst coverage previously documented in the literature (e.g.,
Bhushan, 1989, OBrien and Bhushan, 1990, Brennan and Hughes, 1991, Lang and Lundholm,
1996, and Barth, Kasznik, and McNichols, 2001). Size is the logarithm of the market value of
equity measured at the end of the year. Volume is the logarithm of trading volume in millions of
shares during the year. Issue is an indicator variable that equals one if the firm issues debt or
equity securities during the years t-1, t, or t+1, and zero otherwise. Book-Market is the ratio of
the book value to the market value of equity. R&D is research and development expense scaled
by total sales. Depreciation is depreciation expense scaled by total sales. Following Barth et al.
-
8/8/2019 Comp Sep 2 2008
21/46
20
(2001), we industry adjust the R&D and depreciation measures by subtracting the respective 2-
digit SIC industry mean value. Earn Volatility is the standard deviation of 16 quarterly earnings
(deflated by total assets), consistent with the horizon used to estimate earnings comparability.
We also control for earningspredictability. Predictability is the R2
from a firm-specific AR1
model with 16 quarters of data. In addition to the levels of these variables, we control for the
differences in characteristics between firm i and firm j, following the analysis in Table 2. In
particular we control for the differences in size, trading volume, and book-to-market.
The annual sample for this test is quite large. For firm i, there areKanalysts who cover
the firm. For each firm i analyst kpair there areJfirms in the same industry as firm i. Hence,
our sample consists ofIfirms Kanalysts Jfirms. In addition to requiring valid data for all
our measures, we require each analyst kto cover at least five firms. In estimating the model, we
rely on the coverage choice of an analyst within an industry, and therefore require the availability
of at least a few observations per analyst per industry for which CondCoverage equals one. This
restriction should exclude junior analysts, analysts in transition, and data-coding errors. We
exclude analysts who cover more than 40 firms. Covering greater than 40 firms is rare (less than
one percent of analysts) and could be a data-coding error in that the observations could refer to
the firm employing the analyst rather than an individual analyst at the firm.
The large sample size (average annual sample used in our tests consists of 1.2 million
firm i analyst k firmj observations) prohibits us from estimating a panel regression. Thus, in
Table 3, we provide the mean coefficient from the 14 annual logistic regressions. The t-statistic
is based on the distribution of the 14 annual coefficients using the Fama and MacBeth (1973)
procedure. Further, we adjust for potential time-series dependence in the estimates using the
Newey-West (1987) correction with one lag. (In untabulated tests, we find that higher lags lead
-
8/8/2019 Comp Sep 2 2008
22/46
21
to highert-statistics.)
The mean coefficient on Comp Acctij is positive and statistically significant as predicted.
In untabulated analysis we find that the coefficient is positive in all 14 years. Further, the mean
coefficient on Comp Earnij is positive and statistically significant (as well as the coefficients on
Comp CFOij and Comp Retij). The result suggests that the firmjs we identify as comparable
to firm i are more likely to be followed by the analysts who also cover firm i. In terms of
economic significance, using the year of 2000 as a sample year (this year is selected because it
reflects the mean effect over the whole sample period), an increase from the 10th
to the 90th
percentile in Comp Acctij (Comp Earnij) is associated with an increase in the probability of
being covered by an analyst from 1.01% to 1.12% (0.94% to 1.14%), a relative increase of 10%
(21%). For comparison, the same increase forComp Retij (the strongest predictor in Table 2) is
associated with an increase in probability from 0.70% to 1.89%. Overall, we conclude that the
likelihood of an analyst covering firmj, conditional on the analyst covering firm i, increases in
the comparability between firms i and j. This is consistent with higher comparability reducing
the cost of covering the firm. It also suggests the benefits of covering firms with high
comparability (due to their associated higher-quality information set) outweigh the potential
decreased benefit from investors reduced demand for information about highly-comparable
firms.8
[Table 3]
5.2. Firm level comparability
In the previous sections we investigated the consequences of pairwise firm i - firmj level
8This result complements a study by Ramnath (2002), who shows that there is information transfer between
firms covered by the same analyst. He shows that among these firms, the earnings announcement surprises of firms
that announce first are systematically related to forecast revisions for the other firms that the analysts cover.
-
8/8/2019 Comp Sep 2 2008
23/46
22
comparability. In the following sections we examine the benefits to analysts of aggregated firm i
level comparability.
5.2.1. Sample and dependent variables
As seen in Table 1, there are 27,972 firm years with firm-level comparability scores. In
order to test the hypotheses using firm comparability, we further restrict the sample to firms with
available data to compute the dependent variables and the control variables. In particular, the
main restriction is positive analyst coverage, which biases the sample towards larger and more-
frequently-traded firms. These restrictions reduce the sample to a maximum number of 13,037
firm-year observations (this is the sample for the analyst coverage tests; the sample is smaller for
the remaining dependent variables).
The four dependent variables in the tests below are defined as follows. Coverage is the
logarithm of the number of analysts issuing an annual forecast for firm i in year t. Analyst
forecast accuracy is the absolute value of the forecast error:
Accuracy (%)it= |Fcst EPSitActual EPSit|/Priceit-1 -100. (9)
Fcst EPSitis analysts mean I/B/E/S forecast of firm-is annual earnings for yeart. For a given
fiscal year (e.g., December of yeart+1) we collect the earliest forecast available during the year
(i.e., we use the earliest forecast from January to December of year t+1 for a December fiscal
year-end firm). Actual EPSit is the actual amount announced by firm i for fiscal period t+1 as
reported by I/B/E/S. Price is the stock price at the end of the prior fiscal year. Because the
absolute forecast error is multiplied by -100, higher values ofAccuracy imply more accurate
forecasts. We measure optimism in analysts forecasts using the signed forecast error:
Optimism (%)it= (Fcst EPSitActual EPSit)/Priceit-1 100. (10)
-
8/8/2019 Comp Sep 2 2008
24/46
23
Dispersion (%) is the cross-sectional standard deviation of individual analysts annual forecasts
for a given firm, scaled by price, multiplied by 100.
Table 4, Panel A presents descriptive statistics for the dependent variables and the
comparability measures. The mean (median) Coverage (i.e., the logarithm of the number of
analysts covering a firm) is 1.53 (1.61) implying that a median firm is covered by 5 analysts.
Mean forecast accuracy is 4.6% of share price. Mean forecast optimism is 2.6% of share price,
which is consistent with prior research that analysts tend to be optimistic on average. However,
the median is only 0.3%, also consistent with previous research. The mean forecast dispersion is
0.9% of share price. Panel B presents the correlation matrix. Analyst coverage and forecast
accuracy are positively correlated with the comparability measures whereas forecast optimism is
negatively associated with firm comparability. The correlations between forecast dispersion and
the comparability measures are not in a consistent direction.
[Table 4]
5.2.2. Analyst coverage tests
To test whether analyst coverage and comparability are positively related, our first
hypothesis, we estimate the following regression:
Coverageit+1 = +1 Comparabilityit+Controlsit+ it+1. (11)
Comparability is one of the firm-level comparability measures (e.g., Comp4 Acct,
CompInd Acct, Comp4 Earn, or CompInd Earn). Throughout the remaining analysis, for
continuous variables that we do not take the logarithm of, we delete observations if these
variable values fall in the lowest or highest percentile of their respective distributions, calculated
annually (i.e., we trim the data annually at the 1% and 99% percentile). We also include industry
and year fixed effects. Because the estimation of Equation 11 is likely to suffer from time-series
-
8/8/2019 Comp Sep 2 2008
25/46
24
dependence, we estimate the model as a panel and cluster the standard errors at the firm level (in
addition to the year fixed-effects). In estimating Equation 11, we control for other factors
motivating an analyst to cover firmj as described in the prior section - Size, Volume,Issue,Book-
Market,R&D,Depreciation,Earn Volatility andPredictability.
Table 5 presents the regression results. Both of the accounting comparability measures
(Comp4 Acctand CompInd Acct) are positively associated with analyst coverage. In terms of
economic significance, an increase from the 10th
to the 90th
percentile in CompInd Acct is
associated with an increase in the logarithm of analyst following of 0.043 (= 0.0146 x 2.93).
Given that the median firm in our sample is covered by 5 analysts, this effect translates to a
percentage increase in analyst coverage of 4.4%, suggesting that the effect is also modestly
significant on an economic basis. Similarly, Comp4 Earn and CompInd Earn are also positively
associated with analyst coverage (in this case an 80th
percentile increase in Comp4 Earn would
translate to an increase in coverage of 5%). Finally, we note that CFO and return comparability
are also positively associated with analyst coverage. Overall, the regression results in Table 5
confirm the conditional analyst coverage findings in Table 3, and are consistent with hypothesis
1 that predicts a positive association between analyst coverage and comparability.
[Table 5]
5.2.3 Forecast accuracy, optimism, and dispersion tests
To test our hypotheses about whether comparability affects forecast accuracy, optimistic
bias, and dispersion, we estimate the following specification:
Forecast Metricit+1 = +1 Comparabilityit+Controlsit+ it+1. (12)
Forecast Metric is Accuracy, Optimism, orDispersion. Hypothesis 2 predicts that accuracy is
increasing in comparability, and that optimism and dispersion are decreasing in comparability.
-
8/8/2019 Comp Sep 2 2008
26/46
25
We control for other determinants of these forecast metrics as previously documented in
the literature. SUEis the absolute value of firm is unexpected earnings in yeartscaled by the
stock price at the end of the prior year. Unexpected earnings are actual earnings minus the
earnings from the prior year. Firms with greater variability are more difficult to forecast, so
forecast errors should be greater (e.g., Kross, Ro and Schroeder, 1990, and Lang and Lundholm,
1996). Consistent with Heflin, Subramanyam and Zhang (2003), earnings with more transitory
components should also be more difficult to forecast. We include the following three variables
to proxy for the difficulty in forecasting earnings. Neg UEequals one if firm is earnings are
below the reported earnings a year ago, zero otherwise. Loss equals one if the current earnings is
less than zero, zero otherwise. Neg SIequals the absolute value of the special item deflated by
total assets if negative, zero otherwise. We expect these three variables to be positively related
to optimism given that optimism is greater when realized earnings are more negative.
Daysitis a measure of the forecast horizon, calculated as the logarithm of the number of
days from the forecast date to firm-is earnings announcement date. The literature shows that
forecast horizon strongly affects accuracy and optimism (Sinha et al., 1997, Clement, 1999, and
Brown and Mohd, 2003). We also control for Size because firm size is related to analysts
forecast properties (e.g., Lang and Lundholm, 1996). Last, we include industry and year fixed
effects. Similar to the estimation of Equation 11, we estimate the model as a panel and cluster
the standard errors at the firm level.
Table 6 presents the regression results for analyst forecast accuracy. With respect to the
comparability measures, our primary variables of interest, we find that both accounting and
earnings comparability are positively associated with accuracy. In terms of economic
significance, an increase from the 10th
to the 90th
percentile in Comp4 Acct (Comp4 Earn) is
-
8/8/2019 Comp Sep 2 2008
27/46
26
associated with an increase in accuracy of about 0.36% (0.71%), which represents an
improvement in accuracy of about 7% (15%) for the average firm in the sample. This result
supports hypothesis 2a that higher earnings comparability increases the accuracy of analysts
forecasts. Finally, the measures of CFO and return comparability are negatively related with
forecast accuracy in the Column 3 (Comp4 Earn) regression and not significant in the Column 4
(CompInd Earn) regression, contrary to the findings with accounting and earnings comparability.
[Table 6]
Table 7 presents the results for forecast optimism. In support of hypothesis 2b, we find a
consistent negative relation between our measures of comparability and analyst optimism. As
with forecast accuracy, the result is also economically significant suggesting a reduction in
analyst optimism for the average firm in the sample that ranges from 8% with CompInd Earn to
28% with Comp4 Earn. Together with the findings using forecast accuracy, these results suggest
that one way earnings comparability improves forecast accuracy is via a reduction of analyst
optimism.
[Table 7]
The results for forecast dispersion are presented in Table 8. In this case the results are
more mixed. While accounting comparability is negatively associated with forecast dispersion,
we fail to find a significant relation between earnings comparability and forecast dispersion.
Still, the result with accounting comparability suggests a reduction in forecast dispersion
between 11% and 37% for a change from 10th
to the 90th
percentile in Comp4 Acctand CompInd
Acctrespectively.
[Table 8]
-
8/8/2019 Comp Sep 2 2008
28/46
27
In sum, the above results support the hypotheses that analysts accuracy is increasing in
comparability and decreasing in analysts optimism. In addition, they provide some evidence
that forecast dispersion is decreasing in comparability.
6. Conclusion
This paper develops two measures of comparability and then studies the effect of these
comparability measures on analysts. A key innovation is the development of empirical, firm-
specific, output-based, quantitative measures of comparability. The first measure, accounting
comparability, is based on the idea that comparable firms with similar economic events, as
proxied by returns, should report similar accounting earnings. The second measure, earnings
comparability, is based on the strength of the historical covariance between a firms earnings and
the earnings of other firms in the same industry, as evidenced by the R2
values. We first provide
construct validity of our measures. The likelihood of an analyst using firm j as a benchmark
when analyzing firm i in a report is increasing in the comparability between firm i and j, as
defined using our measures. This suggests that our measures are correlated with actual use of
comparable firms in analysts reports.
We then test whether comparability manifests any benefits to the capital markets as
gleaned from the effect on analyst coverage and the properties of analyst forecasts. With respect
to analyst coverage, coverage increases in comparability. Tests also indicate that the likelihood
of an analyst covering firm i also covering firmj is increasingin the comparability between firm
i and j. Hence, we not only show that comparability leads to greater analyst following, but also
specifically predict which other firms an analyst will follow. These results are consistent with
comparability leading to richer information sets, which more than offsets the potential decreased
benefit due to reduced investor demand for information about high-comparability firms. In
-
8/8/2019 Comp Sep 2 2008
29/46
28
addition, analysts who follow firms with higher comparability issue more accurate and less
biased earnings forecasts. These results suggest that earnings comparability helps analysts to
forecast earnings and that the improvement comes, at least in part, through a reduction in
forecast optimism. Last, we document that accounting comparability is negatively related to
analysts forecast dispersion, consistent with the availability of superior public information about
highly-comparable firms and an assumption that analysts uses similar forecasting models.
In sum, we develop a measure of comparability that likely captures users notions of
comparability and the benefits of comparability to them. The ability to forecast future earnings
is a common task for users such as investors and analysts, particularly those engaged in
valuation. Improved accuracy and reduced bias, for example, represent tangible benefits to this
user group. Further, the results of increased analyst following, greater forecast accuracy, lower
bias, and less dispersion collectively are consistent with comparability enriching firms
information environment, which provides a tangible benefit for firms with higher comparability.
While comparability is generally accepted as a valuable attribute, there is little evidence beyond
this study that would empirically confirm this widely-held belief.
We believe our comparability measure could be used in a number of contexts, with
modifications to the measure tailored to suit the needs. Our measure could be used to help assess
the changes in comparability as a result of changes in accounting measurement rules or reporting
standards, accounting choice differences, or of adjustments. For example, according to the
International Accounting Standards Committee Foundation (IASCF), the primary objective of
the International Financial Reporting Standards (IFRS) is to develop a single set of global
accounting standards that require high quality, transparent and comparable information in
financial statements and other financial reporting (our emphasis) (IASCF 2005). Our measure
-
8/8/2019 Comp Sep 2 2008
30/46
29
could be used to assess whether IFRS achieves the intended consequence of enhanced financial
statement comparability (see e.g., Beuselinck, Joos, and Van der Meulen, 2007).
Our measure could also assist practitioners, such as analysts and boards, in their objective
selection of comparable firms. For example, new executive compensation rules issued by the
SEC (2006) advise those companies who engage in compensation benchmarking to identify the
peer companies used as benchmarks. Further, choosing comparables is often considered an art
form (see Bhojraj and Lee, 2002) and the inherent discretion in this choice can lead to strategic
behavior. For instance, Lewellen, Park, and Ro (1996) find firms choices of industry and peer-
company benchmarks are self serving. Thus, our measure could be used internally by firms or
externally by investors to assess or validate this choice.
Notwithstanding the above benefits, some caveats are in order. We do not study the
determinants of comparability. Our analysis is silent on what firms can do to improve cross-
sectional comparability. Certainly, firms could choose to have more comparable accounting
choices. We speculate, however, that economic innovations, which by definition distinguish
firms from their peers, could lead to decreased economic comparability. Further, our results are
silent on the effects of comparability to risk. For example, an investor interested in diversifying
a portfolio might not desire comparability if that means holding securities with a positive return
covariance. Hence from a diversified investors perspective, comparability may lead to negative
risk effects that offset the benefits we document. Last, while earnings are arguably the most
important summary measure of accounting performance, it captures only one financial-statement
dimension. An opportunity exists to create a multi-dimensional financial statement measure. We
leave these issues to future research.
-
8/8/2019 Comp Sep 2 2008
31/46
30
References
Antle, R. and A. Smith. 1986. An empirical investigation of the relative performance evaluation
of corporate executives,Journal of Accounting Research 24, 1-39.
Banker, R.D., and S.M. Datar. 1989. Sensitivity, precision, and linear aggregation of signals forperformance evaluation.Journal of Accounting Research 27 (1): 21-39.
Barber, B.M., and J.D. Lyon.1996. Detecting abnormal operating performance: The empirical
power and specification of test statistics.Journal of Financial Economics 41, 359399.
Barber, B.M., and J.D. Lyon.1997. Detecting long-run abnormal stock returns: The empiricalpower and specification of test statistics.Journal of Financial Economics 43, 341372.
Barth, M.E., R. Kaznick, and M.F. McNichols. 2001. Analyst coverage and intangible assets.
Journal of Accounting Research 39 (1), 134.
Beaver, W., P. Kettler, and M. Scholes. 1970. The association between market determined and
accounting determined risk measures. The Accounting Review 45 (4), 654682.
Beaver, W., and J. Manegold. 1975. The association between market-determined andaccounting-determined measures of systematic risk: Some further evidence. The Journalof Financial and Quantitative Analysis 10 (2), 231284.
Beuselinck, C., P. Joos, and S. Van der Meulen. 2007. International earnings comparability.
Working paper, Tilburg University.
Bhushan, R. 1989. Firm characteristics and analyst following. Journal of Accounting andEconomics 11, 255274.
Bhojraj, S., and C.M.C. Lee. 2002. Who is my peer? A valuation-based approach to the selection
of comparable firms.Journal of Accounting Research 40 (2), 407439.
Bradshaw, M.T., and G.S. Miller. 2007. Will harmonizing accounting standards really harmonizeaccounting? Evidence from non-U.S. firms adopting US GAAP. Forthcoming, Journal ofAccounting, Auditing and Finance.
Brennan, M. and P. Hughes. 1991. Stock prices and the supply of information. The Journal ofFinance 46, 16651691.
Brown, L. D. 2001. A temporal analysis of earnings surprises: Profits versus losses. Journal ofAccounting Research 39 (September), 221241.
Brown, L. D., and E. Mohd. 2003. The predictive value of analyst characteristics. Journal ofAccounting, Auditing and Finance 18 (Fall), 625648.
Brown, P., and R. Ball. 1967. Some preliminary findings on the association between the earningsof a firm, its industry, and the economy.Journal of Accounting Research 5, 5577.
Chan, K., and A. Hameed. 2006. Stock price synchronicity and analyst coverage in emerging
markets.Journal of Financial Economics 80, 115147.
Clement, M.B., 1999. Analyst forecast accuracy: Do ability, resources and portfolio complexity
matter?Journal of Accounting and Economics 27 (June), 285303.
-
8/8/2019 Comp Sep 2 2008
32/46
31
Das, S., C. Levine, and K. Sivaramakrishnan. 1998. Earnings predictability and bias in analysts
earnings forecasts. The Accounting Review 73 (April), 277294.
De Franco, G. 2007. The information content of analysts notes and analysts propensity tocompliment other disclosures. Working Paper, University of Toronto.
DeFond, M.L., and M. Hung. 2003. An empirical analysis of analysts cash flow forecasts.
Journal of Accounting and Economics 35, 73100.
Dikolli, S., C. Hofmann, and T. Pfeiffer. 2007. Efficient benchmarking. Working paper, Duke
Univeristy, University of Tuebingen, and University of Vienna
Eames, M., and S.M. Glover. 2003. Earnings predictability and the direction of analysts
earnings forecast errors. The Accounting Review 78 (July), 707724.
Eames, M., S.M. Glover, and J. Kennedy. 2002. The association between trading
recommendations and broker-analysts earnings forecasts. Journal of AccountingResearch 40 (March), 85104.
Fama, E., and J.D. MacBeth. 1973. Risk, return, and equilibrium: Empirical tests. Journal of
Political Economy 81, 607636.
FASB. 1980. Statement of Financial Accounting Concepts No. 2: Qualitative Characteristics of
Accounting Information. Available at http://www.fasb.org/pdf/con2.pdf
Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2004. Cost of equity and earnings attributes.
The Accounting Review 79 (4), 967-1010.
Francis, J., and D. Philbrick. 1993. Analysts decisions as products of a multi-task environment.
Journal of Accounting Research 31 (Autumn), 216230.
Francis, J., Schipper, K., Vincent, L., 2002. Earnings announcements and competinginformation.Journal of Accounting and Economics 33, 313342.
Frankel, R., Kothari, S., Weber, J., 2006. Determinants of the informativeness of analystresearch.Journal of Accounting and Economics 41, 29-54.
Gonedes, N.J. 1973. Evidence on the information content of accounting numbers: Accounting- based and market-based estimates of systematic risk. The Journal of Financial andQuantitative Analysis 8 (3), 407443.
Gonedes, N.J. 1975. A note on accounting-based and market-based estimates of systematic risk.
The Journal of Financial and Quantitative Analysis 10 (2), 355365.
Gu, Z., and J.S. Wu. 2003. Earnings skewness and analyst forecast bias. Journal of Accountingand Economics 35 (April), 529.
Harris, M., and A. Raviv. 1993. Differences in opinion make a horse race. Review of FinancialStudies 6, 473494.
Heflin, F., K.R. Subramanyam, and Y.Zhang. 2003. Regulation FD and the financialenvironment: Early evidence. The Accounting Review 78 (1), 137.
Holmstrm, B. 1979. Moral hazard and observability.Bell Journal of Economics 10: 74-91.
IASCF. 2005. IASCF foundation constitution. Available at http://www.iasb.org/About+Us/
About+the+Foundation/Constitution.htm
-
8/8/2019 Comp Sep 2 2008
33/46
32
Joos, P., and M. Lang. 1994. The effects of accounting diversity: Evidence from the European
Union.Journal of Accounting Research 32 (Supplement), 141168.
Kandel, E., and N. Pearson. 1995. Differential interpretation of public signals and trade inspeculative markets.Journal of Political Economy 103 (August), 831853.
Kim, O., and R.E. Verrecchia. 1994. Market liquidity and volume around earnings
announcements.Journal of Accounting and Economics 17, 4167.
Kothari, S.P., Leone, A.J., and C.E. Wasley. 2005. Performance matched discretionary accrual
measures.Journal of Accounting and Economics 39, 163197.
Kross, W., B. Ro and D. Schroeder. 1990. Earnings expectations: The analysts information
advantage. The Accounting Review 65 (2), 461476.
Land, J., and M. Lang. 2002. Empirical evidence on the evolution of international earnings. The
Accounting Review 77, 115- 133.
Lang, M.H., and R.J. Lundholm. 1996. Corporate disclosure policy and analyst behavior. TheAccounting Review 71 (4), 467492.
Lewellen, W.G, T. Park, and B.T. Ro. 1996. Self-serving behavior in managers discretionary
information disclosure decisions.Journal of Accounting and Economics 21, 227251.
Libby, R., P.A. Libby, and D.G. Short. 2004. Financial accounting, 4th
edition. McGraw-HillIrwin.
Lim, T. 2001. Rationality and analysts forecast bias. Journal of Finance 56 (February), 369385.
Newey, W., West, K., 1987. A simple, positive semi-definite heteroscedasticity and
autocorrelation consistent covariance matrix.Econometrica 55, 703-708.
OBrien, P.C. 1988. Analysts forecasts as earnings expectations. Journal of accounting and
Economics 10, 5383.
OBrien, P.C. and R. Bhushan. 1990. Analyst following and institutional ownership. Journal ofAccounting Research (Supplement), 5576.
Palepu, K.G., and P.M. Healy. 2007. Business analysis & valuation using financial statements,4
thedition. Thomson South-Western.
Papadakis, G., and P. Wysocki. 2007. Pairs trading and accounting information. Working paper,
Boston University and MIT.
Penman, S.H. 2006. Financial statement analysis and security valuation, 3rd
edition. McGraw-
Hill Irwin
Piotroski, J.D., and D.T. Roulstone. 2004. The influence of analysts, institutional investors, and
insiders on the incorporation of market, industry and firm-specific information into stock
prices. The Accounting Review 79 (4), 11191151.
Ramnath, S. 2002. Investor and analyst reactions to earnings announcements of related firms: Anempirical analysis.Journal of Accounting Research 40 (5), 13511376.
Revsine, L., D.W. Collins, and W.B. Johnson. 2004. Financial reporting and analysis, 3rd
edition.
Prentice Hall, Upper Saddle River, NJ.
-
8/8/2019 Comp Sep 2 2008
34/46
33
Richardson, S.A., S.H. Teoh and P.D. Wysocki. 2004. The walk-down to beatable analysts
forecasts: the roles of equity issuance and insider trading incentives. ContemporaryAccounting Research 21 (Winter), 885924.
SEC. 2000. SEC Concept Release: International Accounting Standards. Available at
http://sec.gov/rules/concept/34-42430.htm.
SEC. 2006. Executive compensation and related person disclosure; final rule and proposed rule.
Available at http://www.sec.gov/rules/final/2006/33-8732afr.pdf.
Sinha, P., L. Brown, S. Das. 1997. A re-examination of financial analysts differential earningsforecast accuracy. Contemporary Accounting Research 14 (Spring), 142.
Stickney, C.P., and R.L. Weil. 2006. Financial accounting: An introduction to concepts,
methods, and uses, 11th
edition. Thomson South-Western.
Stickney, C.P., P.R. Brown, and J.M. Wahlen. 2007. Financial reporting, financial statementanalysis, and valuation, 6
thedition. Thomson South-Western.
White, G.I., A.C. Sondhi, and D.Fried. 2002. The analysis and use of financial statements, 3rd
edition. Wiley.
Wild, J.J., K.R. Subramanyam, and R.F. Halsey. 2006. Financial statement analysis, 9th
edition.
McGraw-Hill Irwin
-
8/8/2019 Comp Sep 2 2008
35/46
34
APPENDIX - Variable definitions
Variable Definition
UseAsComp = Indicator variable that equals one if analyst kwho writes a reportabout firm i refers to firmj as a
comparable firm in her report, and equals zero otherwise.
CondCoverage = Indicator variable that equals one if analyst kwho covers firm i also covers firmj, and equals zerootherwise. An analyst covers a firm if she issues at least one annual forecast about the firm.
Coverage = Logarithm of the number of analysts issuing a forecast for the firm.
Accuracy (%) = Absolute value of the forecast error multiplied by -1, scaled by the stock price at the end of the prior
fiscal year, where the forecast error is the I/B/E/S analysts mean annual earnings forecast less theactual earnings as reported by I/B/E/S.
Optimism (%) = Signed value of the forecast error, scaled by the stock price at the end of the prior fiscal year, where
the forecast error is the I/B/E/S analysts mean annual earnings forecast less the actual earnings asreported by I/B/E/S.
Dispersion (%) = Cross-sectional standard deviation of individual analysts annual forecasts, scaled by the stock priceat the end of the prior fiscal year.
Comp Acctij = The absolute value of the difference of the predicted value of a regression of firm is earnings on
firm is returns using the estimated coefficients for firm i andj respectively. It is calculated for eachfirm i firmj pair, (ij),j = 1 toJfirms in the same 2-digit SIC industry as firm i.
Comp Earnij =R2 from a regression of firm is quarterly earnings on the quarterly earnings of firmj. It is calculated
for each firm i firmj pair, (ij),j = 1 toJfirms in the same 2-digit SIC industry as firm i.
Comp CFOij =R2 from a regression of firm is quarterly CFO on the quarterly CFO of firm j. It is calculated for
each firm i firmj pair, (ij),j = 1 toJfirms in the same 2-digit SIC industry as firm i.
Comp Retij =R2 from a regression of firm is monthly returns on the monthly returns of firmj. It is calculated for
each firm i firmj pair, (ij),j = 1 toJfirms in the same 2-digit SIC industry as firm i.
Comp4Acct = Average of the four highest Comp Acctij for firm i.
Comp4 Earn = Average of the four highest Comp Earnij for firm i.
Comp4 CFO = Average of the four highest Comp CFOij for firm i.
Comp4 Ret = Average of the four highest Comp Retij for firm i.
CompInd Acct = Average Comp Acctij for firm i for all firms in the industry.
CompInd Earn = Average Comp Earnij for firm i for all firms in the industry.
CompInd CFO = Average Comp CFOij for firm i for all firms in the industry.
CompInd Ret = Average Comp Retij for firm i for all firms in the industry.
Book-Market = Ratio of the book value to the market value of equity.
Days = Logarithm of the number of days from the forecast date to the earnings announcement date.
Depreciation = Firms depreciation expense scaled by total sales, less the respective 2-digit SIC industry mean
value of depreciation expense scaled by total sales.
Earn Volatility = Standard deviation of 16 quarterly earnings.
Issue = Indicator variable that equals one if the firm issues debt or equity securities during the preceding,current, or following year, and zero otherwise.
Loss = Indicator variable that equals one if the current earnings is less than zero, and equals zero otherwise.
Neg SI = Absolute value of the special item deflated by total assets if negative, and equals zero otherwise.
Neg UE = Indicator variable that equals one if firm is earnings are below the reported earnings a year ago, andequals zero otherwise.
-
8/8/2019 Comp Sep 2 2008
36/46
35
APPENDIX (Continued)
Variable Definition
Predictability =R2
of a regression of annual earnings on prior-year annual earnings for the same firm.
R&D = Firms research and development expense scaled by total sales, less the respective 2-digit SIC
industry mean value of research and development expense scaled by total sales.
Size = Logarithm of the market value of equity measured at the end of the year.
Size-$ = Market value of equity measured at the end of the year.
SUE = Absolute value of unexpected earnings, scaled by the stock price at the end of the prior year, where
unexpected earnings is actual earnings less a forecast based on a seasonal-adjusted random walk
time-series model.
Volume = Logarithm of trading volume in millions of shares during the year.
Difference = Absolute value of the difference between firm is and firmjs respective variables.
-
8/8/2019 Comp Sep 2 2008
37/46
36
TABLE 1 Comparability: Descriptive statistics
Panels A and B (C and D) provides descriptive statistics of the firm i firm j pair (firm i) level comparability
metrics. Panels A and C present descriptive statistics. Panels B and D present Pearson (Spearman) correlations
above (below) the main diagonal. Variables are defined in the Appendix.
Panel A: Pairwise firm i firmj level comparability - Descriptive statistics (all numbers in %)
Variable No. of Obs Mean STD 10th Percent Median 90th Percent
Comp Acctij 3,592,745 -1.55 2.80 -3.62 -0.65 -0.09
Comp Earnij 3,592,745 11.19 13.85 0.19 5.56 31.26
Comp CFOij 3,592,745 8.24 10.34 0.14 4.10 22.75
Comp Retij 3,592,745 6.16 7.91 0.11 3.04 16.80
Panel B: Pairwise firm i firmj level comparability - Correlations
Comp Acctij Comp Earnij Comp CFOij Comp Retij
Comp Acctij 1.000 0.026 0.013 0.011
Comp Earnij 0.039 1.000 0.082 0.081
Comp CFOij 0.024 0.049 1.000 0.058
Comp Retij 0.033 0.042 0.030 1.000
Panel C: Firm level comparability - Descriptive statistics (all numbers in %)
Variable No. of Obs Mean STD 10th Percent Median 90th Percent
Comp4 Acct 27,972 -0.26 0.95 -0.43 -0.03 0.00
CompInd Acct 27,972 -1.14 1.99 -2.36 -0.53 -0.20
Comp4 Earn 27,972 52.36 14.62 33.23 52.02 72.12
CompInd Earn 27,972 6.69 4.21 2.73 5.58 11.84
Comp4 CFO 27,972 42.83 11.55 27.64 42.69 57.73
CompInd CFO 27,972 4.63 2.15 2.66 4.10 7.17
Comp4 Ret 27,972 24.76 12.25 11.28 22.24 42.29
CompInd Ret 27,972 4.25 4.27 1.01 2.77 9.36
-
8/8/2019 Comp Sep 2 2008
38/46
TABLE 1 (Continued)
Panel D: Firm level comparability - Correlations
Comp4 Acct CompInd Acct Comp4 Earn CompInd Earn Comp4 CFO CompInd CFO
Comp4 Acct 1.000 0.889 0.045 0.030 0.075 0.002
CompInd Acct 0.750 1.000 -0.006 0.051 0.024 0.041
Comp4 Earn 0.200 0.000 1.000 0.564 0.331 0.027
CompInd Earn 0.055 0.084 0.570 1.000 0.055 0.210
Comp4 CFO 0.247 0.040 0.316 0.072 1.000 0.403
CompInd CFO 0.002 0.116 0.047 0.149 0.418 1.000
Comp4 Ret 0.251 0.061 0.306 0.103 0.287 0.029
CompInd Ret 0.077 0.034 0.167 0.132 0.118 0.083
-
8/8/2019 Comp Sep 2 2008
39/46
38
TABLE 2 Use of comparable firm in analysts reports
This table reports an analysis of the relation between the pairwise comparability measures and analysts use in their
reports of firms in the same industry as the sample firm for the year 2005. The sample includes the combination of
analysts reports about sample firms multiplied by the number of firms in each sample-firms industry with availabledata. We estimate various specifications of the following pooled logistic regression:
UseAsCompij
= +1Compij
ij+Controls
j+
ij.
Industry fixed effects are included but not tabulated. Coefficientz-statistics are in italics and are clustered at the firm
level. Significance levels are based on two-tailed tests. ***, **, and * denotes significance at the 1%, 5%, and 10%
levels, respectively. Variables are defined in the Appendix.
Prediction (1) (2) (3) (4)
Comp Acctij + 3.40*
1.89
Comp Earnij + 0.77*** 0.51*** 0.43**
4.77 3.05 2.53
Comp CFOij + 1.10*** 0.86***
5.22 4.16
Comp Retij + 4.99*** 4.96***
18.35 18.00
Size + 0.21*** 0.21*** 0.16*** 0.16***
8.46 8.56 6.26 6.37
Volume + 0.31*** 0.31*** 0.24*** 0.24***
10.20 9.80 8.01 7.64
Book-Market ? 0.33** 0.30** 0.14 0.16
2.54 2.44 1.13 1.30
ROA ? 0.88 0.45 1.54* 1.40*
1.05 0.52 1.86 1.68
Size Difference - -0.27*** -0.28*** -0.23*** -0.22***
-11.31 -11.10 -9.18 -8.78
Volume Difference - -0.14*** -0.14*** -0.11*** -0.12***
-4.51 -4.40 -3.40 -3.61
Book-Market Difference - -0.69*** -0.66*** -0.50*** -0.51***
-4.62 -4.43 -3.48 -3.51
ROA Difference -0.39 -0.51 0.11 0.18
-0.49 -0.59 0.14 0.23
Industry FE Yes Yes Yes Yes
Pseudo R2
3.53% 3.41% 3.67% 3.59%No. of Obs. 139,767 139,767 139,767 139,767
-
8/8/2019 Comp Sep 2 2008
40/46
39
TABLE 3 Conditional Analysts coverage of comparable firms
This table reports an analysis of the relation between the pairwise comparability measures and analyst coverage of
firms in the same industry as the sample firm. For each of the years 1993 to 2006 in our sample, we estimate the
following logistic regression:
CondCoverageikj = +1 Compijij +Controlsj + ikj.
Industry fixed effects are included but not tabulated. The number of observations used in the annual estimation is thecombination of sample firms multiplied by the analysts covering the sample firms multiplied by the number of firms
in each sample-firms industry with available data. The table presents the mean, maximum, and minimum
coefficients, pseudo R2, and number of observations from the 14 annual logistic regressions. Mean coefficient t-
statistics (in parentheses) are based on the distribution of the 14 annual coefficients and adjusted for time-series
dependence using the Newey-West (1987) correction with one lag. Significance levels are based on two-tailed tests.***, **, and * denotes significance at the 1%, 5%, and 10% levels, respectively. Variables are defined in the
Appendix.
Prediction Estimate T-statistic Estimate T-statistic
Comp Acctij + 4.78*** 8.85
Comp Earnij + 0.59*** 22.06
Comp CFOij + 0.63*** 5.98
Comp Retij + 4.97*** 48.12
Size + 0.36*** 15.65 0.32*** 17.85
Volume + 0.28*** 27.72 0.23*** 10.50
Book-Market 0.51*** 5.35 0.32*** 2.71
R&D + 0.65*** 5.21 0.46*** 3.89
Depreciation + 2.01*** 6.14 1.74*** 3.42
Issue + -0.02 -0.58 0.02 1.28
Predictability + -0.23*** -3.08 -0.28*** -3.38Earn Volatility -1.19** -2.08 -2.00*** -4.98Size Difference -0.19*** -10.18 -0.13*** -5.53Volume Difference -0.18*** -24.83 -0.14*** -17.87
Book-Market Difference -0.44*** -6.50 -0.22*** -3.67Industry FE Yes Yes
Pseudo R2 7.22% 6.56%
-
8/8/2019 Comp Sep 2 2008
41/46
40
TABLE 4 Firm comparability: Descriptive statistics
This table reports descriptive statistics for the dependent variables and the comparability metrics. The sample is
restricted to observations with available data to calculate all the variables in this analysis. Panel A presents
descriptive statistics and Panel B reports Pearson correlations.
Panel A: Descriptive statistics (all number are in %)
Variable No. of Obs Mean STD 10th Percent Median 90th Percent
Coverage 13,037 1.53 1.05 0.00 1.61 2.89
Accuracy 11,945 -4.64 13.25 -10.04 -1.23 -0.11
Optimism 11,861 2.61 11.76 -2.04 0.30 7.92
Dispersion 8,292 0.87 1.71 0.06 0.32 2.07
Comp4 Acct 13,037 -0.12 0.37 -0.22 -0.02 0.00
CompInd Acct 13,041 -0.78 1.08 -1.65 -0.45 -0.19
Comp4 Earn 12,550 54.15 14.15 35.36 54.00 73.47
Comp4 CFO 12,550 43.94 11.19 29.08 43.89 58.63
Comp4 Ret 12,550 26.88 12.09 13.14 24.56 44.56
CompInd Earn 12,544 7.12 4.28 2.92 6.01 12.51CompInd CFO 12,544 4.79 2.07 2.80 4.31 7.39
CompInd Ret 12,544 4.69 4.22 1.18 3.31 10.04
-
8/8/2019 Comp Sep 2 2008
42/46
TABLE 4 (Continued)
Panel B: Correlation matrix
Coverage Accuracy Optimism DispersionComp4
Acct
CompInd
Acct
Comp4
Earn
Comp4
CFO
Comp4
Ret
Coverage 1.000 0.190 -0.143 -0.162 0.097 0.181 0.065 0.105 0.264
Accuracy 1.000 -0.921 -0.473 0.112 0.175 0.008 0.010 0.020
Optimism 1.000 0.302 -0.065 -0.101 -0.020 -0.025 -0.053
Dispersion 1.000 -0.146 -0.256 0.026 0.000 0.098
Comp4 Acct 1.000 0.758 0.059 0.080 0.053
CompInd Acct 1.000 -0.029 -0.014 -0.038
Comp4 Earn 1.000 0.268 0.225
Comp4 CFO 1.000 0.227
Comp4 Ret 1.000
CompInd Earn
CompInd CFO
CompInd Ret
-
8/8/2019 Comp Sep 2 2008
43/46
TABLE 5 Firm comparability and analyst coverage
This table reports an analysis of the relation between firm comparability and analyst coverage. The sample is
restricted to observations with available data to calculate all the variables in this analysis. The table reports the
results of various specifications of the following regression:
Coverageit+1 = +1 Comparabilityit+Controlsit + it+1.
Industry and year fixed effects are included for each model but not tabulated. We estimate each model as a panel andcluster the standard errors at the firm level. Coefficient t-statistics are in italics. Significance levels are based on
two-tailed tests. ***, **, and * denotes significance at the 1%, 5%, and 10% levels, respectively. Varia