Corporate Strategy, Analyst Coverage, and the Uniqueness ... · Corporate Strategy, Analyst...
Transcript of Corporate Strategy, Analyst Coverage, and the Uniqueness ... · Corporate Strategy, Analyst...
Corporate Strategy, Analyst Coverage, and the Uniqueness Paradox
Lubomir P. Litov Olin Business School,
Washington University in St. Louis [email protected]
Patrick Moreton
Olin Business School, Washington University in St. Louis
Todd R. Zenger1 Olin Business School,
Washington University in St. Louis [email protected]
March 18, 2009
Abstract In this paper we argue that managers confront a paradox in selecting strategy. On the one hand, capital markets systematically discount uniqueness in the investment strategy choices of firms. Uniqueness in strategy heightens the cost of collecting and analyzing information to evaluate a firm’s future value. These greater costs in strategy evaluation discourage the collection and analysis of information regarding the firm, and result in a valuation discount. On the other hand, uniqueness in strategy is a necessary condition for creating economic rents and should, but for this information cost, be positively associated with firm value. We find empirical support for both propositions using a firm panel dataset between 1985 and 2007.
1 We thank Garrick Blalock, Richard Frankel, Dirk Hackbarth, Bart Hamilton, Rob Solomon, Bernard Yeung and seminar participants at University of Pennsylvania, INSEAD, London Business School, NYU, USC, Washington University in St. Louis, and Cornell University for helpful comments and suggestions in preparing this manuscript.
2
Corporate Strategy, Analyst Coverage, and the Uniqueness Paradox
Abstract
In this paper we argue that managers confront a paradox in selecting strategy. On the one hand, capital markets systematically discount uniqueness in the investment strategy choices of firms. Uniqueness in strategy heightens the cost of collecting and analyzing information to evaluate a firm’s future value. These greater costs in strategy evaluation discourage the collection and analysis of information regarding the firm, and result in a valuation discount. On the other hand, uniqueness in strategy is a necessary condition for creating economic rents and should, but for this information cost, be positively associated with firm value. We find empirical support for both propositions using a firm panel dataset between 1985 and 2007.
3
Introduction
In an efficient capital market, equity-based rewards encourage managers
to select high quality strategies—strategies that raise the value of the firm. Selecting a
strategy with high future returns involves searching for economic rents—rents that
fundamentally derive from scarcity or uniqueness in the bundles of complementary
assets, activities, and resources that the manager assembles (Bowman, 1974; Rumelt,
1984; Barney, 1986; Montgomery and Wernerfelt, 1988; Brandenburger and Stuart,
1996).2 If the manager selects a common strategy, then competition within product
markets and competition within factor markets for the resources to execute the selected
strategy, eliminate economic rents. Thus, if synergies or complementarities are common
or obvious, the market for business units—a strategic factor market (Barney, 1986)—
incorporates these synergies into the prices paid, leaving no avenue for value creation.
Only firms possessing unique asset positions or managers possessing unique foresight
about the value of asset combinations can predictably obtain complementary assets at less
than their value in their new foreseen use.3 Uniqueness is therefore a necessary condition
of high quality strategy and a necessary condition for value creation.
However, the effectiveness of equity-based rewards in prompting managers to
select value-creating strategies depends on the market’s capacity to efficiently evaluate
the expected future returns of strategies that managers select. In this regard, capital
markets confront a problem of information asymmetry (Akerlof, 1973; Hubbard, 1998);
managers possess knowledge about the quality or value of strategies that the market
2 Indeed, economic rents are defined as “reward[s] for scarcity of superior efficiency that …vanish as [they become] common….” (Becker, 1971:77). 3 Note that a firm’s strategy need not remain unique to generate value. Thus, even if other firms observe the value in a unique strategy and proceed to copy it, these imitating firms are more likely to pay prices for the required assets that are fully reflective of their value in this new, more valuable application.
4
lacks. If information about the “quality” of selected strategies is costly to obtain, the lack
of information elevates uncertainty in the market and diminishes equity values. Implicit
in most models of strategy choice is an assumption that the manager’s task is solely to
maximize the quality of strategy selected where quality is measured as the present value
of its future returns. The magnitude of the information problem that accompanies a
strategy is presumed an exogenous feature of the market. Yet, certainly this assumption
is flawed. Strategies differ significantly in the scope of the information problem that they
impose on the market. Some strategies are rather easily evaluated and examined, while
others are quite difficult and costly to evaluate and assess. Thus, in selecting a particular
strategy, the manager determines both the probable future returns and the scope of the
associated information problem.
Unique strategies are likely to impose the greatest information burden on the
market. Assessing the future value of uncommon combinations of assets and businesses
requires additional effort on the part of participants in the market. While clear incentives
exist to uncover this unique value, these incentives are tempered by the elevated effort
costs. The result is a manager who faces a paradox in strategy choice. Choosing a high
quality, unique strategy maximizes the expected value of future returns, but imposes
higher information costs on the market, resulting in a market discount. Choosing
transparent or common strategy reduces the information problem, but likely involves a
compromise on quality.
Our objective in this paper is to argue and empirically demonstrate that managers
confront a uniqueness paradox in selecting strategy. While unique strategies are more
highly valued, they are also more costly to evaluate. This heightened cost leads to
5
systematically less coverage by securities analysts and results in valuation discount of
what might otherwise be a premium for uniqueness. We empirically test propositions
relating to this uniqueness paradox using a 23-year panel dataset linking approximately
14,000 firms from Compustat’s Industrial and Segment files between 1985 and 2007 with
their corresponding analysts who appear in the I/B/E/S Detail History file. We find
results consistent with a discounted uniqueness premium: while unique strategies trade on
average at a premium to common strategies, more unique strategies receive less analyst
coverage, and this reduced analyst coverage creates a corresponding discount in market
valuation.
Our paper is organized as follows. In the next section we discuss a conceptual
model of the relationship between firm strategy and equity value. Consistent with prior
work in accounting and finance, we argue that analyst coverage positively influences
equity value (Hong, Lim, & Stein, 2000; Elgers, Lo, & Soffer, 2001) and that strategy
choice, particularly uniqueness and complexity, affects the costs of an analyst initiating
coverage. Our third section describes the data and methods used in modeling the effects
of strategy choice on coverage and coverage on valuation. It further presents empirical
results. The fourth section discusses robustness issues. The final section concludes with a
brief summary and some directions for future research.
II. Hypothesis Development
2.1. Firm Strategy, Analyst Coverage, and Equity Values
With incentives to increase the equity value of the firms that employ them (Jensen
& Murphy, 1990; Hall & Leibman, 1998), managers will evaluate the likely impact of
alternative strategy choices on equity values. We view this link between strategy choice
6
and market valuation as occurring in three stages. Managers first choose strategies which
differ in quality as measured by the future cash flows they generate and range from the
rather common or simple to the extremely unique or complex. Common strategies are
those observed frequently among firms in an industry, while unique strategies are those
rarely observed. Simple strategies are strategies constrained to a single business line,
while complex strategies involve a presence in multiple industries. Market participants,
including analysts, then observe the selected strategy and, based on an assessment of
costs and benefits, decide whether to invest in acquiring further information concerning
its value. We operationalize market participants’ choice of whether to invest in
information acquisition using analysts’ decisions regarding the initiation of coverage.
For those firms selected for coverage, analysts then collect and evaluate relevant
information and issue a report available to at least a subset of the investors in the market.
For each analyst, there is a positive probability that he or she will discover some
information that other analysts do not discover. Consequently, the more analysts who
cover a firm, the more informed is the market for that firm’s equity, i.e. the more accurate
is the information about the future prospects of the firm and the lower is the uncertainty
associated with investing in the firm. While a variety of factors may motivate analyst
coverage (see Lin & McNichols, Michaely & Womack, 1999), empirical evidence
generally confirms that increased analyst coverage increases the informational efficiency
of markets (Frankel, Kothari, and Weber, 2006; Hong, Lim, and Stein, 2000; Elgers, Lo,
and Pfeiffer, 2001; Lys & Sohn, 1990). In a final stage, investors in the market review
the available information about the firm, including reports issued by analysts, and then
submit their buy and sell orders to a market maker, who sets a market-clearing price.
7
This process in which publicly traded firms ‘manufacture’ equity investments for
sale to investors through retail brokers is not unlike the process used to sell other
complex, difficult-to-assess products, where managers seek and buyers pay a premium
for third party certification of quality or value. In this setting, investors pay a premium
for securities that receive third party evaluation by analysts. Moreover, these analysts
make decisions on the brokerage firms’ behalf regarding which equities to cover and thus
receive featured “shelf space.” Issuers of equity then have an incentive to cultivate the
attention of analysts who both increase the distribution of the firm’s equity and enhance
its value by reducing the amount of uncertainty that investors have about its performance
prospects.
A manager’s choice of strategy therefore plays two critical roles. First, it defines
the expected financial returns of the firm, including the mean and variance of the firm’s
future cash flows, which in a fully efficient market with no costs of collecting
information defines the firm’s equity price. Second, the firm’s strategy determines the
costs that an investor or market intermediary faces in accurately predicting this mean and
variance. When information is costly to obtain, information is likely to be limited in the
market, thus creating greater uncertainty about a firm’s value. When investors are risk-
averse and unable to eliminate all firm-specific risk through diversification strategies, a
feature of many models of rational trade in financial securities (e.g. Hellwig, 1980), the
market discounts the values of higher risk firms (Merton, 1987). Thus, informational
uncertainty about a firm’s strategy increases the investor’s firm-specific risk from owning
the firm’s stock and thereby diminishes the security’s price. Given this effect, the firm’s
management has an incentive to attract more coverage by analysts in order to increase the
8
amount of information possessed by investors and to increase the equity value of the firm
they manage.4
2.2. Uniqueness, Complexity, and the Information Costs of Coverage
A manager’s choice of strategy, particularly the uniqueness or complexity,
profoundly affects not only the future returns of the firm, but also the costs imposed on
analysts or the brokerage firms that employ them in providing coverage. Our assumption
is that the scope of coverage for a particular security is a function of the costs and
benefits of collecting information about that strategy. On the one hand, the adoption of a
unique or complex strategy could arguably attract analyst coverage, because investors
may place a high value on information precisely in those settings where information is
costly to obtain. However, this argument presumes that analysts and the firms that
employ them are strongly rewarded for satisfying investors’ preferences for information
gathering and analysis. At best, this is only partly true. The alternative hypothesis
which we empirically explore is that the costs of covering more unique or complex
strategy choices generally or frequently outweigh any added benefits to the analyst.
While analysts or their employers bear the full cost of information collection, the benefits
of accurate analysis are quite broadly shared. Consequently, as the costs of collecting
and analyzing information about a firm’s strategy rise, analyst coverage should decline.
The complexity and uniqueness of strategy choices shape the associated costs of
analysis. Thus, the number of distinct businesses in which the firm competes strongly 4 Empirically, Amir, Lev, and Sougiannis (1999) have shown that earnings forecasts by analysts explain between 12 per cent and 40 per cent of the above-normal returns earned by investors from publicly traded stocks, suggesting that analysts do indeed provide information that is valuable to the markets. Zuckerman (1999) has also shown that firms which receive coverage from a larger portion of the analysts focused on the industries represented in their portfolio are more highly valued, as measured by the Berger and Ofek excess value measure.
9
influences the complexity of the analyst’s task. Analysts specialize by industry,
presumably to economize on the costs of information gathering and analysis. If a firm
competes in a range of industries, this increases the costs of coverage, because it either
requires analysts to develop expertise in multiple industries or it requires coverage and
collaboration among analysts across multiple industries. Prior empirical work suggests
that an increase in the complexity of the analysts’ task reduces the accuracy of their
earnings forecasts (cites from Duru and Reeb, 2002; Brown, Richardson, and Schwager,
1987). Our argument here is that diversification increases the complexity of the analysts’
task and the costs of analysis, which in turn causes the level of coverage itself to decline.
Firms pursuing unique or less familiar strategies are also more costly for analysts to
evaluate. Like many other activities, securities analysis is an activity governed by
economies of scale. A firm adopting a common strategy, a strategy involving a
combination of assets and businesses that is common in the industry, imposes a small
incremental cost on the analyst who chooses to cover it, because the analyst has already
made substantial investments in understanding the strategy, as pursued by others. By
contrast, a firm with an unusual strategy may impose a large incremental cost on the
analyst. Assessing the value of a unique collection of businesses not only requires an
understanding of the separate industries in which each competes, but also an
understanding of any complementarities or synergies that are generated through the
combination. The more unique the combination assembled, the less likely it is that any
given analyst will be familiar with these synergies. By simplifying a complex strategy,
or adopting a more common strategy, managers lower analysts’ costs of coverage,
thereby attracting coverage and escalating information available about their security.
10
To summarize, we hypothesize:
H1: Analyst coverage is decreasing in the uniqueness of a corporate strategy.
H2: Analyst coverage is decreasing in the complexity of a firm’s strategy.
2.3. Strategy Choice, Equity Prices, and the Uniqueness Paradox
If, as noted above, uniqueness in strategy is a necessary condition for deliberate
value creation, yet uniqueness diminishes analyst coverage and the informational
efficiency of the market for the firm’s equity, then the manager faces a difficult dilemma.
Pressure by analysts to simplify or conform and thereby lower the costs of analysis may
be quite explicit. For instance, a 1999 analyst report from Paine Webber pushing for
Monsanto’s break-up as a life sciences company provides a surprisingly candid revelation
of these complexity- and uniqueness-related costs:
The life sciences experiment is not working with respect to our analysis or in reality. Proper analysis of Monsanto requires expertise in three industries: pharmaceuticals, agricultural chemicals and agricultural biotechnology. Unfortunately, on Wall Street, particularly on the sell-side, these separate industries are analyzed individually because of the complexity of each. This is also true to a very large extent on the buy-side. At PaineWebber, collaboration among analysts brings together expertise in each area. We can attest to the challenges of making this effort pay off: just coordinating a simple thing like work schedules requires lots of effort. While we are willing to pay the price that will make the process work, it is a process not likely to be adopted by Wall Street on a widespread basis. Therefore, Monsanto will probably have to change its structure to be more properly analyzed and valued.
The suggestion here is clear—that Monsanto should unbundle so as to reduce the
information costs that accompany this complex strategy and thereby promote more
extensive and precise analysis, and ultimately raise the aggregate valuation of these
assets. Empirical evidence supports this prescription, at least as it relates to reductions in
the complexity of a firm’s strategy. Gilson, Healy, Noe, and Palepu (2001) find that
11
focus-increasing transactions (spin-offs, carve-outs, and targeting stock offerings)
increase analyst coverage and increase the accuracy of analysis.5 Bhushan (1989) also
shows that analyst coverage increases as the firm reduces the number of SICs in which a
firm operates. Firms face similar pressures to unbundle unique asset combinations. For
instance, the Financial Times (Roberts, 2005) described Georgia Pacific as “trading at a
discount to the sum of its parts” because it was “an awkward mix of assets that are
difficult to evaluate together.”
Yet, clearly the manager’s primary path to value creation cannot be simply
lowering information costs. After all, economic rents are rewards for scarcity and these
rewards vanish as rent-generating assets become common (Becker, 1971). Thus, value
creation requires an insightful manager who uniquely sees value in particular asset
combinations and thus procures these at prices below their value in their new use
(Barney, 1986). If the value of particular asset combinations is widely known and thus
widely implemented by other firms in an industry, the firm will pay prices for these assets
that are fully reflective of this value. Thus, uniqueness in strategy is key to creating
economic rents and increasing value. As a consequence, the manager necessarily faces a
paradox in selecting strategy: uniqueness in strategy is essential to the discovery of
economic rents, but uniqueness elevates the information costs associated with evaluating
strategy. The simple relationship between uniqueness and firm value is therefore
uncertain. Instead, we expect to observe uniqueness having a positive direct effect on
firm value, but a negative indirect effect through analyst coverage. To summarize:
5 Similarly, Zuckerman (2000) finds evidence that that suggests firms reshape themselves to match the expertise of the analysts providing coverage.
12
H3: While uniqueness in strategy will reduce firm value through reduced analyst coverage, controlling for this effect, firm value is increasing in the uniqueness of strategy choice.
III. Empirical Design
3.1. Data
To analyze these questions, we construct a 23-year panel dataset of firms and their
analyst following between 1984 and 2007. We use the CRSP Monthly Files, the
combined CRSP-Compustat database and the I/B/E/S detailed history datasets in
constructing our dataset.6 We consider an analyst covering a firm in year t if that analyst
has issued annual earnings forecast for that firm’s fiscal period ending in year t.
INSERT TABLE 2
Our sample excludes firms in the financial industry and consists of 59,158
observations on 7,636 unique firms. The mean number of yearly observations per firm is
7.7. Fifty six percent of the firms in our sample are covered by at least one analyst (Table
2). Among covered firms, the mean number of analysts per firm is 8.9, with the
maximum number of analysts covering a firm being 64 (Compaq Computer in 1989).
Our primary interests are three-fold. First, we wish to examine whether a firm’s
strategy choice affects the level of coverage it receives. Second, we seek to confirm that
the amount of coverage that the firm receives affects the stock market’s valuation of the
firm, as well as its subsequent performance. Finally, we seek to examine how a firm’s
strategy choice affects firm value both directly and through analyst coverage.
6 I/B/E/S International Inc. is a unit of Thomson Financial and collects analysts’ estimates and research for institutional investors. It is available to researchers through the Wharton Research Data Services.
13
3.2. Measures
3.2.1. Analyst Coverage
We model the coverage level received by a firm using the following stylized
model of analyst behavior. Define Xi as a vector of firm i traits that are believed to affect
the costs and benefits to the analyst/brokerage firm from covering the firm. Define Dik as
the uniqueness of firm i’s corporate strategy relative to the strategies of the other firms
considered for coverage by analyst k. Finally, define
ikikiik uDU +++≡ 210 βββ X (1)
as the net benefit that analyst k obtains when covering firm i, in which uik is a disturbance
term that incorporates analyst k’s idiosyncratic benefit from covering firm i. With this
specification, analyst k covers firm i if Uik > 0. Assume for simplicity that,
and that the disturbance is uncorrelated both within and across indices. We can then take
the expectation of Uik across all relevant k to obtain the average benefit that an analyst
will obtain from covering firm i:
[ ] iikkii DU εβββ +++≡ E210 X (2)
If we assume that the firm-specific disturbance term εi is i.i.d. across i with some
known distribution F(εi), then the probability that firm i is covered by an analyst is:
[ ]( ) [ ]( )ikkiikkii DFD E1EPr 210210 ββββββε ++−=++> XX
If there are Aj analysts covering industry j, then we can use (2) to generate
[ ]( )( )ikkiji DFAa E1 210 βββ ++−= X , the number of analysts that find it attractive to
cover firm i. Dividing both sides by Aj, we obtain:
[ ]( )( )ikkij
i DFAa
E1 210 βββ ++−= X , (3)
as firm i’s share of the analysts covering industry j.
kiik vu += ε
14
With this simple specification and the additional assumption that F(εi) is the
standard normal distribution, it is relatively straightforward to estimate the parameters in
using the maximum likelihood method and the share of analysts
covering a firm as the dependent variable. Note, however, that there are a small number
of industries each year for which the firms in that industry receive no coverage. For
firms in these industries, we assume that [ ]( ) 0E210 ≅++ ikki DF βββ X so that our
dependent variable for coverage level becomes:
0 if 0 Adj. Coverage
/ if 0j
ii j j
A
a A A
=⎧⎪= ⎨ >⎪⎩
In robustness checks, we will utilize four other measures of analyst coverage.
First, we measure analyst coverage as a simple count of the number of analysts which
cover a particular security. The primary difficulty with this measure is the large
proportion of firms that receive no coverage whatsoever. While we use methods
appropriate for this type of dependent measure, we also present analyses using a simple
dichotomous measure coded as 1 when the firm receives any coverage whatsoever in a
given year, and 0 otherwise. Thirdly, we split adjusted coverage into two components.
Imputed adjusted coverage is the adjusted coverage (sales weighted) that a firm would
have received, if each of its component segments received analyst coverage as though it
were the median pure-play firm. This measure is appealing as we can use it to test if
lower coverage is due to a firm adopting business units in industries that are difficult to
analyze, rather than due to a firm adopting a complex strategy of interaction among these
segments (i.e., the corporate strategy). To capture the latter, we examine the excess
coverage that we define as the difference of the actual and imputed adjusted coverage.
( )210 ,, βββ=β
15
We refer to the imputed coverage as segment-specific coverage, and to the excess
coverage as strategy-specific coverage.
3.2.2. Firm strategy choice proxies
We compute measures of both strategy uniqueness and strategy complexity,
arguing that the choice to pursue either raises analysts’ coverage costs. Analysts
generally specialize by industry and thus consider for coverage only a subset of the
publicly traded firms. Firms which pursue common strategies should be more familiar
and more easily analyzed by industry specialized analysts. For this reason, we measure
the similarity of a firm’s strategy relative to other firms in its primary SIC.
For each firm we define the vector of their sales across all segments, si =
,1, , ,. .i t i n tSales Sales ′⎡ ⎤⎣ ⎦ . We then normalize this vector to unit length, by dividing all
vector elements by , ,j i j t
sales∑ , where i indexes the firm, and j indexes the set of
segment industries (n = 1,107 for 1984-2007), for a given year t.
We then define the primary industry for each firm-year as the industry with the
highest sales of the corporation. For example, GE’s primary industry is SIC 6153 in
2005, since GE Capital had the highest sales within GE in 2005. For each primary
industry, j* each year t, we define the industry vector of sales, as sj*, t =
1, ,. .
i ii n isales sales ′⎡ ⎤
⎣ ⎦∑ ∑ , where i indexes the firms in each of the n=1,107
segment industries in Compustat that are in primary industry j* in that year. We
16
normalize this vector to unit length by dividing with 1,107
,1
i i jj
sales=∑∑ (i indexes the firms, j
indexes 1,107 industries.)
Using the vectors of the firm and its primary industry distribution of sales across all
industries in the economy, we define the firm’s measure of conformity as , , *,i t i t j tconf s s′= .
This measure reflects the proximity of the sales “distribution” for each firm to the one of its
counterparts in the primary industry. Note that ,0 1i tconf≤ ≤ , and higher values of this
measure imply greater conformity to the corporate strategy of other firms in the same
primary industry. To capture the influence of analysts’ coverage on the choices of
corporate strategy we refine our conformity measure further as * *, , *,i t i t j tconf s s′= . The
vector **,j ts is defined as * ** *1, ,
. .i ii n i
sales sales ′⎡ ⎤⎣ ⎦∑ ∑ for primary industry j* in year t
where the vector elements summations include only firms that have any analyst coverage in
primary industry j* in year t. For convenience, we redefine these measures as 1-confi,t and
1-conf*i,t. We refer to these as measures of corporate strategy uniqueness (UNIQUE1 and
UNIQUE2, correspondingly), as higher values indicate lower level of conformity with the
strategy of industry peer group. In our analysis we examine both. For brevity, we show
results with the latter measure only.
The costs of analysis should also rise with the complexity that accompanies increased
diversification. As firms enter related or unrelated industries, analysis of the firm requires
either multiple analysts to collaboratively evaluate the firm or analysts to develop expertise
across multiple industries. In either case, analysis of the firm is more costly than analyzing
a single segment firm. We measure the total number of reported segments, using a series
17
of dummy variables (S1, S2, S3,…, S8) which are coded as 1 if the number of segments in
which the firm competes equals 1, 2, 3, or 8 or more, respectively, and coded as 0
otherwise.
3.2.3. Control variables
As control variables in our analysis, we include several firm-specific factors
identified either empirically or theoretically to affect the costs and benefits of coverage
from the analysts’ perspective. Following Barth, Kaznick and McNichols (2001) and
Bushan (1989) we include the variables for each firm’s 3 to 5 year compound annual
growth rate in sales, the log of its annual trading volume in its shares, the coefficient of
variation of its earnings over the last 3 to 5 years, the log of the number of common
shareholders, and the log common equity for the firm.7 We also control for the
intangible assets of the firm: R&D, advertising, recognized intangible assets, and
depreciation, the first two as a share from total expenses, and the latter two as a share
from total assets (following Barth et al). We also control for the issuance of debt or
equity in the prior year with a dummy variable that has value 1 if the company has issued
either stock or debt.8 Finally, we include variables measuring the firm’s return on average
equity (ROE), the sales-weighted average of the Herfindahl concentration index for all
the SICs in which the firm operates, and a sales-weighted average of its share of the sales
in each of the industries in which it operates, an indicator for regulated industry firms, an
indicator for S&P 600 membership, as well as an indicator for the introduction of
Regulation Fair Disclosure (Reg. F.D.) To confirm that our results are not driven by
7 Barth, Kaznick and McNichols (2001) include the log of market value (which is equivalent to including log(price)+log(shares)) in their regressions as well as several measures of the relative level of intangible assets possessed by the firm. We chose not to use market value in our regressions because the firm’s stock price is endogenous to the level of analyst coverage under the hypotheses we are examining in this paper. 8 In robustness checks we also use the log of amount of debt issued in the prior year. Results are unchanged. Note however that debt issuance is endogenous to analyst coverage, Chang, Dasgupta, Hilary (2006).
18
outliers, we winsorize all variables generated from accounting data at the 2.5 percentiles
in each tail.9 Table 1 provides concise description of the main variables. Table 2 Panel A
provides means, medians, and standard deviations for all variables.
INSERT TABLE 2
From Table 2, we see that 37.9% (100-62.1%) of all firms in our sample have
more than one segment. Moreover, in Panel C we find that fabricated products,
transportation, chemicals, printing and publishing, defense, precious metals and
shipbuilding industries are the industries in which firms have the highest levels of
strategy uniqueness. From Panel D, we also note that corporate strategy uniqueness, as
we measure it, has generally decreased from its 1980s levels.
In Panel E we show correlations of the main variables of interest. We note that
our measures of uniqueness are highly correlated with the number of segments. Hence, to
ascertain robustness of our results, we also examine alternative measures of uniqueness
that are orthogonal to the number of segments. Note further that all of our uniqueness
measures are highly persistent. To correct for the effect of this persistence on our results,
we cluster-adjust our standard errors at the firm level.
3.3. Results
3.3.1. Analyst coverage
Table 3 presents our main multivariate results on analyst coverage. Column 1
presents the estimates of a negative binomial regression of a simple count measure of the
number of analysts covering a firm. We use a conditional fixed effects negative binomial
regression method to model the effects of changes in our independent variables across
9 We also performed all analyses with un-winsorized variables. Our results remain.
19
time and within a firm on the number of analysts covering that firm. We include in this
model both year and industry effects and the full set of control variables described above.
The results support our fundamental hypothesis that costly-to-analyze securities receive
less analyst coverage. Controlling for other factors which may influence analyst
coverage, firms with more lines of business and more unique strategies receive less
analyst coverage. The array of dummy variables measuring diversification shows a
rather linear progression of increasingly negative coefficient values. The estimated
effects of our control variables are also generally consistent with our description of the
analyst coverage decision. Larger firms, as measured by their common equity and firms
that are traded more heavily receive more coverage.
The firm’s performance as measured by accounting profitability (ROE) and sales
growth also have positive effects on the level of coverage. The issuance of either equity
or debt in the prior fiscal year is positively related to the number of analysts providing
coverage. The measure of recognized intangibles in the firm is negatively related to
coverage, in line with Barth et al. (2001). Similarly, the R&D spending and the
depreciation are positively and significantly related to coverage, once again as in Barth et
al. (2001).
Column 2 of Table 3 presents a logit specification with a simple dichotomous
dependent variable, measuring whether or not the firm received analyst coverage of any
magnitude. The results are quite consistent with the negative binomial regression model.
Strategy uniqueness (UNIQUE1) is negatively related to the receipt of analyst coverage.
Similarly, the dummy variables measuring the scope of diversification show the same
very consistent pattern with prior specifications (although now the linear progression in
20
coefficient magnitudes is near monotonic.) Increases in uniqueness or complexity
significantly reduce the likelihood that the firm receives analyst coverage at any level.
Coefficients for the control variables are also similar in sign and significance with the
negative binomial models.
Table 3 also presents models of analyst coverage using the share of analyst
coverage received by the firm (adjusted coverage) as the dependent measure. Column 3
presents the results of an OLS regression of adjusted coverage on the strategy variables,
the diversification dummy variables, as well as the control variables described above.
The results are generally consistent with the results presented so far. The results suggest
that strategy uniqueness (UNIQUE1) and diversification (only at the number of segments
greater than seven) have significant negative effects on adjusted coverage. The sign and
significance of the control variables are also quite consistent across these specifications.
A concern with the OLS specifications of column 3 is that 43.4% of our
observations have a value of 0 for adjusted coverage, suggesting a left censoring
problem. To correct for this, column 4 is a model of adjusted coverage, but estimated by
the Tobit method to account for the left-tail censoring of the dependent variable at zero.10
Consistent with the OLS specification and the negative binomial and logit results, an
increase in corporate uniqueness or diversification of a firm’s strategy is associated with
a decrease in the share of analyst coverage that a firm receives. Note that while the series
of segment dummies do not show the orderly progression of increasingly negative
coefficient values as in columns 1 and 2, the trend of greater diversification diminishing
coverage remains apparent, especially if a firm reports more than three segments.
10Unfortunately a fixed effects Tobit specification would yield biased results (Honoré, 1993). We show it here for consistency, however in robustness we also examine the coefficients of a random effect Tobit method. Our results remain.
21
3.3.2. Strategy-specific and segment-specific analyst coverage
In columns 5 and 6 we attempt to split adjusted coverage into two components:
segment-specific coverage or the portion attributable to the analyst coverage that specific
segments receive and strategy-specific coverage or the portion attributable to the
combination of these segments. We seek to confirm that the effect of strategy
uniqueness in limiting analyst coverage is related to the combination of businesses and
not merely to entering segments that receive less coverage overall.11 We find support for
this hypothesis. Greater strategy uniqueness (UNIQUE1) is associated with less strategy-
specific coverage. However our results on diversification (complexity) are not sustained.
In column 6 we further relate segment-specific coverage to corporate strategy uniqueness.
Our results are similar: firms that have corporate strategies more similar to the centroid
strategy in their primary industry tend to receive more coverage. We note also that the
point estimates of corporate uniqueness are three-fold more negative in the strategy-
specific coverage model as compared to the segment-specific coverage model, indicating
that it is the corporate-strategy that is more difficult to analyze, as opposed to the
component industries for the focal firm.
3.3.3. Robustness
One concern with our measure of corporate strategy uniqueness is that it may
merely reflect the focal firm’s performance co-movement with its primary industry. To
address this, we replicate our results so far using industry fixed effects, instead of the
11 The number of observations in this regression decreases significantly, in order to be able to compute the strategy-specific coverage.
22
firm fixed effects. In columns (7)-(12) we replicate all of our model specifications with
industry fixed effects. Our results remain.
3.3.4. Changes in coverage
Our particular interest is in whether changes in a strategy’s uniqueness or
complexity from year to year are associated with changes in the level of analyst coverage.
We hence examine a differences regression for the models in Table 3. The results are
presented in Table 4.
INSERT TABLE 4
In column 1 we regress log(1+analyst coverage) on the same control variables as in Table
3. We obtain that firms whose strategies become more unique experience less analyst
coverage in the subsequent year. This result is also supported in column 2, where the
dependent variable is the adjusted coverage. When we split this into strategy-specific and
segment-specific coverage, we find that both measures decrease subsequent to increases
in the strategy uniqueness of the focal firm. Note further that the magnitude of decrease is
higher for the strategy-specific coverage, indicating that indeed difficulty to analyzing the
strategy, i.e. the combination of businesses, rather than component industries is
particularly responsible for this relation strategy uniqueness and coverage.
We conclude that there exists a causal link between the uniqueness of the
corporate strategy and the subsequent analyst coverage. We now turn to the firm
valuation implications of this fact.
3.5. Firm Value and Analyst Coverage
3.5.1. Tobin’s q
23
Our interest in examining the link between firm valuation, analyst coverage and
corporate strategy uniqueness is twofold. On the one hand, we have shown that the more
unique is a firm’s corporate strategy, the less coverage by analysts that it receives. We
hence expect that unique strategies will result in lower Tobin’s q, as analyst coverage is
positively associated with it (Doukas, Kim, and Pantzalis, 2005). On the other hand
though, strategy uniqueness is a necessary condition for the generation of economic rents,
and hence we expect the selection of a unique strategy to result in higher Tobin’s q. Thus,
the full effect of selecting a unique strategy on Tobin’s q is theoretically uncertain,
reflective of the fundamental paradox the manager faces in choosing strategy.
To examine the relationship between analyst coverage, corporate strategy
uniqueness and firm value, we regress Tobin’s q on analyst coverage measures, a set of
controls generally known to correlate with Tobin’s q, and measures of corporate
uniqueness. For each firm, our dependent variable is Tobin’s q, which we calculate as
the market value of the firm divided by its book value (Kaplan and Zingales (1997).
We also calculate and examine two industry-adjusted measures of Tobin’s q. The
first one is Tobin’s q measured as market-to-book ratio, less the sales-weighted market-
to-book ratios of its segments (where the segment market-to-book ratios are computed
annually, as the median industry ones based only on single-segment firms; we follow the
methodology of Berger and Ofek, 1995.) The second one is Tobin’s q less the imputed
Tobin’s q for its primary industry centroid strategy (a centroid strategy is the strategy of
business segments combinations at the primary industry level; we define it in detail in the
robustness section.) We compute the latter as the dot product of the primary-industry
normalized sales vector sj*, t and a vector of normalized annual median M/B ratios for
24
single-segment firms in each industry. We refer to that valuation measure as the
uniqueness premium/discount, as firms with more unique strategies may have valuations
that differ from those of their primary-industry centroid strategy.
INSERT TABLE 5
A concern with our analytic method is the joint determination of Tobin’s q,
corporate strategy, and analyst coverage, as outlined above. To address the endogeneity
problem, we run a two-stage least squares (2SLS) regression, where we instrument for
the analyst coverage with variables that capture the visibility of the company in financial
markets. Our choice of instruments is based on the intuition that market visibility would
influence analyst coverage for exogenous reasons, while influencing the choice of
corporate investment strategy. We therefore use as instrumental variables the log of
trading volume and the indicator for S&P 600 index membership.
Column 1 of Table 5 presents estimates from the second-stage equation of the
2SLS system of the market-to-book ratio. The coefficient for adjusted coverage is
positive and significant, as predicted. The result suggests that firms which receive more
coverage trade at a higher premium relative to firms receiving less coverage. In addition,
the coefficient for ROE is positive and statistically significant, while the coefficient for
log sales is negative and significant. These results are broadly consistent with other work
looking at Tobin’s q as a measure of firm value. Consistent with the diversification
discount literature (Lang & Stulz (1994)), our results also suggest that firms that are in
more segments have lower Tobin’s q. Consistent with the strategy literature’s argument
that uniqueness is a necessary condition for competitive advantage and value creation, we
find that the selection of unique strategy is positively related to Tobin’s q.
25
Column 2 presents a specification for the diversification discount—the measure of
firm value in excess of the pure play value of the constituent segments. The results are
generally consistent. The premium for the adoption of a unique strategy remains, though
it diminishes in magnitude. Consistent with the diversification literature we find that the
number of segments is associated with lower valuation, i.e. a diversification discount.
Note that the positive coefficients on the uniqueness measure in these specifications do
not preclude the possibility of a uniqueness discount that accompanies reduced analyst
coverage. Our argument is that uniqueness discourages coverage and reduced coverage
dampens Tobin’s q. Thus, while unique strategies receive a premium, that premium is
lower (i.e. discounted) relative to where it would be with greater analyst coverage.
Column 3 presents results for the uniqueness discount. The dependent variable
here is the difference in the focal firm’s valuation and the valuation of its primary
industry centroid strategy. The coefficient for adjusted coverage remains positive,
significant, and similar in magnitude. The sign, significance, and magnitude of our
control variable coefficients are also similar to the Column 1 results. Finally, the results
on strategy choice, both uniqueness and diversification, remain largely unchanged from
the Column 1 results.
Our results are robust to alternative controls for analyst coverage. When we
instead control for the log(1+analyst coverage) the coefficient on the uniqueness of
corporate strategy measure are:
For model (1): βUNIQUE1= -0.202 (t-stat= 5.18), βCOVERAGE = 1.18 (t-stat=22.4)
For model (2): βUNIQUE1= -0.13 (t-stat = 3.61), βCOVERAGE = 0.98 (t-stat=20.1)
For model (3): βUNIQUE1= -0.144 (t-stat = 4.03), βCOVERAGE = 0.995 (t-stat=20.6)
26
Furthermore, the results obtain when we control for industry instead of firm fixed effects:
For model (1): βUNIQUE1= -0.202 (t-stat= 5.18), βCOVERAGE = 1.18 (t-stat=22.4)
For model (2): βUNIQUE1= -0.13 (t-stat = 3.61), βCOVERAGE = 0.98 (t-stat=20.1)
For model (3): βUNIQUE1= -0.144 (t-stat = 4.03), βCOVERAGE = 0.995 (t-stat=20.6).
3.5.2. Sales growth and dynamics of value-strategy relationship
We have so far established that uniqueness in corporate strategy deters analyst
coverage, but attracts higher valuations. Our explanation for these counterintuitive results
is that there is a tension between the effects of strategy uniqueness on information
production and its direct effects on enterprise value, i.e., that strategy uniqueness in
investment choice should be positively associated with economic rents and firm growth.
While the reduced information production from strategic uniqueness appears to lower
valuations, but for this effect, strategic uniqueness should, in an efficient market, be
associated with higher valuations. We next verify this claim by examining the link of
uniqueness with subsequent sales growth.
INSERT TABLE 6
In Table 6 we study the link between the uniqueness of the corporate strategy and
the subsequent five years sales growth. Here our setup is similar to that adopted in Table
5. We use a 2SLS method, where the endogenous variable (analyst coverage) is once
again instrumented with log trading volume and S&P 600 index membership indicator. In
Model 1 we regress the one-year-ahead sales growth on strategy uniqueness, segment
dummies, adjusted analyst coverage, and controls. Our results are similar to those in
Table 5: firms that maintain unique investment strategies tend to have higher sales
27
growth. In columns 2-5 we further show the coefficient estimates for corporate strategy
uniqueness on sales growth for years two through five. There is almost near monotone
decline in the coefficient magnitude and statistical significance. This is consistent with
our interpretation of strategy uniqueness as a generator of economic rents that dissipate
over time, due to, among other factors, competition in the primary industry or rather
direct imitation of the strategy. The coefficient on analyst coverage has uniformly a
positive sign, indicating that firms that attract more analyst coverage tend to have higher
sales growth in the subsequent five years. Our results on the segment indicators reveal
that one-year sales growth is decreasing in the addition of up to three segments. However,
sales growth over 2-, 3- and 4- years is decreasing also in the addition of up to four
segments.
These results are also robust to using instead industry fixed effects in models (6)-
(10), although the dissipation over time in economic magnitude and statistical
significance of the coefficient on corporate strategy uniqueness is stronger. These results
are also robust to alternative proxies for analyst coverage. When we control for
log(1+analyst coverage) we receive the following estimates:
For model (1): βUNIQUE1= 0.122 (t-stat= 14.5), βCOVERAGE = 0.08 (t-stat=11.3)
For model (2): βUNIQUE1= 0.11 (t-stat = 8.6), βCOVERAGE = 0.11 (t-stat = 10.7)
For model (3): βUNIQUE1= 0.069 (t-stat = 4.62), βCOVERAGE = 0.11 (t-stat=9.1)
For model (4): βUNIQUE1= 0.035 (t-stat = 2.11), βCOVERAGE = 0.093 (t-stat=6.52)
For model (5): βUNIQUE1= 0.022 (t-stat = 1.12), βCOVERAGE = 0.083 (t-stat=4.8)
IV. Robustness
28
4.1. Alternative measures of strategy uniqueness
For robustness we consider an alternative measure of corporate strategy uniqueness.
We construct a measure of the “typical” strategy of all firms that list SIC j as their primary
SIC by creating an s-length vector for all firms listing j as their primary SIC. We refer to
this strategy as a centroid one. The elements 1, 2, 3…, s of this vector for each firm are
equal to the mean sales across all firms in industry j in each of the s SICs that appear
among the segment data for each of the firms in primary industry j. The uniqueness of
each firm’s strategy is then measured as the Euclidian distance, d, between the industry
centroid and an s-element vector constructed from the firm sales in each of the s SICs that
are observed in the industry. This Euclidian distance is then standardized in the following
way:
1
1 1
0 if 0 UNIQUE3
/ if 0
j
j l
iNi
ii i iN N
i i
d
d d d
⎧ =⎪= ⎨
≠⎪⎩
∑
∑ ∑
with Nj equal to the number of firms in industry j. Thus, the variable UNIQUE3i
is a measure of the uniqueness of a firm i’s strategy relative to the average distance of its
peers in primary industry j.
Note that the industry centroid strategy used in the strategy measure is not meant
to portray the most “familiar” strategy in the industry. Rather, it is simply a reference
point against which to compare the strategies of all the firms in the industry. A simple
example will best illustrate this point. Suppose that there are two firms, A and B, in
industry j, each of which has the same sales volume, sj in industry j. Suppose that both
firm A and B are also active in two other SICs, 1 and 2, and that they have sA1, sA2, sB1,
and sB2 in sales in these two industry. To further simplify the example assume that sA1 =
29
sA2 = sB1= sB2 = s so that the industry’s centroid strategy is (sj, s, s). Since this vector is
the same as the vectors characterizing each of the firm’s strategy, the Euclidian distance
between each firm and the centroid is equal to 0, and the mean distance is also equal to
zero, implying that UNIQUE3A = UNIQUE3B = 0. That is, the strategy’s of firms A and
B are commonplace in the extreme.
Now suppose that firm A has sA1 in sales in SIC 1 but none in SIC 2, and that firm
B has no sales in sic 1 but sB2 in sales in industry 2. Again, to simplify the example
assume that sA1 = sB2 = s. The centroid of the industry would then be the vector
(sj,s/2,s/2) and the Euclidian distance of each firm from this centroid would be
( )22/2 sd = . Since both firms are equidistant from the centroid, the mean distance
among all firms would also be d, and our measure of the uniqueness of each firm would
be DIST= d/d = 1. The implication is that both firms are somewhat unique, relative to the
firms in the first example. However, relative to each other, they are equally familiar from
the perspective of an observer of industry j. Therefore, the variable UNIQUE3 provides a
useful summary of the uniqueness of a firm’s strategy both within and across industries.
In Table 2 we present descriptive statistics for UNIQUE3. We note that it has
6.8% correlation with UNIQUE1. Similar to the latter, UNIQUE3 is also highly
persistent, and it has declined from its high levels in the 1980s.
When we use UNIQUE3 as the alternative measure of strategy uniqueness in
models (1)-(3) of Table 3 we obtain the following coefficient estimates:
Model (1): βUNIQUE3= -0.008 (t-stat= -1.70)
Model (2): βUNIQUE3= -0.054 (t-stat = -1.78)
Model (3): βUNIQUE3= -0.012 (t-stat = -3.68)
30
These are consistent with our main findings.
4.3. Multicollinearity
A further concern with UNIQUE1 is its high correlation with the number of
segments (64.9%). There is hence an issue of multicollinearity. To address this we re-
estimate our models in Table 3 without the segment indicators. Our results remain. For
example, the coefficient on the UNIQUE1 is -0.382, highly significant (t-stat is 15.1).
Second, we consider an adjusted measure of uniqueness, UNIQUE1*, which is the
residual of the regression of UNIQUE1 on the number of segments. Using that measure
instead, our results still obtain (for example the coefficient on UNIQUE* in model 1 of
Table 3 is -0.0247, with t-statistic of -7.99). Third, we note that our results using
UNIQUE1 are similar to those using UNIQUE3 and that the correlation of the former
with the number of segments is moderate, at -20.2%.
4.2. Alternative Sample Selection
In our selection procedure we follow the approach by Barth et al (2001). However
our results are robust to including financial companies.
Next, SFAS 131 changed the method of disclosure of segment information in
December 1997. Such change could materially affect our measure of corporate strategy
uniqueness. To verify that our results hold, we split our sample into pre-1997 and post-
1997. The coefficients estimates for model (1) in Table 3 across the two are as follows:
Model (1), pre-Dec 1997 sample: βUNIQUE3= -0.205 (t-stat= -4.71)
Model (1), post-Dec 1997 sample: βUNIQUE3= -0.176 (t-stat= -5.53)
Lastly our results are robust to controlling for the introduction of new regulation
on sell-side analysts’ research, in September 2002 (NASD Rule 2711, NYSE Rule 472,
31
and the Global Analyst Research Settlement) by including an indicator variable for it
(zero pre-Sept 2002 and one otherwise).
V. Discussion and Conclusions
Our empirical results are generally consistent with the relationships we
hypothesized. First, we find that firms which choose more unique strategies receive less
analyst coverage than those which choose more familiar strategies. We can reasonably
conclude that as firms move further away from the strategies of their peers in their
primary industry, they receive less coverage. Alternatively, as they move closer to the
strategy of their peers, analyst coverage increases. Second, this relationship appears to be
causal, as changes in the level of strategy uniqueness drive the changes in levels of
analyst coverage.
Third, firms that receive less coverage also have lower valuations than those that
receive more coverage. Moreover, the effect of a reduction in coverage after controlling
for the likelihood that both coverage and value are correlated with unobserved firm
effects is non-trivial in magnitude. Holding all other variables constant, an increase in
adjusted coverage by one standard deviation increases the firm’s Tobin’s q by
approximately 23%-48% above its mean, depending on the specification used.
Our results thus suggest that managers face a clear paradox in choosing their
investment strategy. While uniqueness in investment strategy choices is a necessary
condition for value creation, as measured by expected future operating performance,
uniqueness in strategy also mandates more costly expenditures by participants in the
market to evaluate such strategy. If analysts (or the investment banks which assign
analysts) are rewarded in ways that overcome these greater costs associated with
32
evaluating unique strategy, then capital markets will correctly (efficiently) evaluate
unique strategies. However, casual observations and our empirical results suggest that
this is not the case. The strong link between investment banking business and analyst
rewards suggests that at best analysts face a multi-tasking problem when allocating effort
to accuracy in analysis of strategies. While accurate and thorough analysis yields some
positive returns to analysts and investment banks, the capacity to draw investment
banking business or trade volume through coverage choices and analysis has been and
continues to be a stronger financial motivation. Because analysts are not directly
rewarded for effort, effort allocated to costly-to-analyze firms is likely to be less than
effort allocated to easily analyzed firms, all else equal. Consequently, capital markets
may systematically discount uniqueness, the most important element required for value
creation.
The result of the above is that managers make strategy choices that are, at the
margin, more common than they would be if the manager were simply choosing
strategies that maximize the discounted present value of expected long term operating
performance. In light of these results, the structure of CEO compensation may have an
important bearing on strategy choice. Rewards that are more weakly linked to present
market value may encourage more unique strategy choices, while rewards linked to
current market valuations encourage strategy choices that are more responsive to effort-
averse analysts and are thus more common and less complex.
33
References
Amihud, Y. and B. Lev, 1981, Risk Reduction as a Managerial Motive for Conglomerate Mergers, Bell
Journal of Economics, 12: 605–17.
Amir, E., B. Lev, and T. Sougiannis, 1999, What Value Analysts? Working Paper, Tel Aviv University.
Barney, J., 1986, Strategic Factor Markets: Expectations, Luck, and Business Strategy, Management
Science, 32: 1231-1241.
Becker, G., 1972, Economic Theory. Alfred A. Knopf: New York.
Berger, P. and Ofek, E., 1995, Diversification’s Effect on Firm Value, Journal of Financial Economics,
37: 39–65.
Berle, A. and G. Means, 1968, The Modern Corporation and Private Property. New York: Harcourt Brace
& World.
Bhushan, R., 1989, Firm Characteristics and Analyst Following, Journal of Accounting and Economics,
11/2/3: 255–74.
Bowman, E., 1974, Epistemology, corporate strategy, and academe. Sloan Management Review. 15: 35-
50.
Brandenburger, A. and J. Stuart, 1996, Value-Based Business Strategy, Journal of Economics and
Management Strategy, 5: 5–24.
Brown, L., G. Richardson, and S. Schwager, 1987, An Information Interpretation of Financial Analyst
Superiority in Forecasting Earnings, Journal of Accounting Research, 25: 49-67.
Donaldson, G., 1984, Managing Corporate Wealth. New York: Praeger.
Duru, A. and D. Reeb, 2002, International Diversification and Analyst Forecast Accuracy and Bias,
Accounting Review, 77: 415-433.
Elgers, P., M. Lo, and R. Pfeiffer, R., 2001, Delayed Security Price Adjustments to Financial Analysts’
Forecasts of Annual Earnings, Working paper, University of Massachusetts - Amherst.
Fama, E., 1970, Efficient Capital Markets: A Review of Theory and Empirical Work, Journal of Finance,
383-417.
Fishman, M. and K. Hagerty, 1989, Disclosure Decisions by Firms and the Competition for Price
Efficiency, Journal of Finance, 44: 633–46.
Frankel, R., S. Kothari, and J. Weber, 2006, Determinants of the Informativeness of Analyst Research,
Journal of Accounting and Economics, Vol. 41: 29-54.
Grossman, S., and J. Stiglitz, 1980, On the Impossibility of Informationally Efficient Markets, American
Economic Review, 70: 393-408.
34
Hall, B. and J. Liebman, 1998, Are CEOs Really Paid Like Bureaucrats?, Quarterly Journal of
Economics, 113: 653–91.
Hellwig, M., 1980, On the Aggregation of Information in Competitive Markets, Journal of Economic
Theory, 22: 477–98.
Holmstrom, B. and S. Kaplan, 2001, Corporate Governance and Merger Activity in the U.S.: Making
Sense of the 1980s and 1990s’. NBER working paper.
Hong, H., T. Lim, and J. Stein, 2000, Bad News Travels Slowly: Size, Analyst Coverage, and the
Profitability of Momentum Strategies, Journal of Finance, 55: 265-295.
Honoré, B., 1993, Orthogonality Conditions for Tobit Models with Fixed Effects and Lagged Dependent
Variables, Journal of Econometrics, Vol. 59: 35-61.
Jensen, M., 1986, Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers, American
Economic Review, 76: 323–9.
Jensen, M., and W. Meckling, 1976, Theory of the Firm: Managerial Behavior, Agency Costs and
Ownership Structure, Journal of Financial Economics, 3: 305–60.
Jensen, M., and K. Murphy, 1990, Performance Pay and Top-Management Incentives, Journal of Political
Economy, 98: 225–64.
Lang, L. and R. Stulz, 1994, Tobin’s Q, Corporate Diversification, and Firm Performance, Journal of
Political Economy, 102: 1248–80.
Lin, H., M. McNichols, 1998, Underwriting Relationships, Analysts’ Earnings Forecasts and Investment
Recommendations, Journal of Accounting and Economics, 25: 101-127.
Lys, T., and S. Sohn, 1990, The Association Between Revisions of Financial Analysts’ Earnings
Forecasts and Security Price Changes, Journal of Accounting and Economics, 13: 341-363.
MacDonald, G. and M. Ryall, 2002, How Do Value Creation and Competition Determine Whether a Firm
Appropriates Value? Working paper, Washington University in St. Louis.
Merton, R., 1987, A Simple Model of Capital Market Equilibrium with Incomplete Information, The
Journal of Finance, 42: 483–510.
Michaely, R., and K. Womack, 1999, Conflict of Interest and the Credibility of Underwriter Analyst
Recommendations, Review of Financial Studies, 12: 573-608.
Montgomery, C. and B. Wernerfelt, 1988, Diversification, Ricardian Rents, and Tobin’s Q. Rand Journal
of Economics, 19: 623-632.
Murphy, K., 1985, Corporate Performance and Managerial Remuneration: An Empirical Analysis,
Journal of Accounting and Economics,7: 11–42.
35
Palepu, K., 1985, Diversification Strategy, Profit Performance and the Entropy Measure, Strategic
Management Journal, 6: 239–55.
Penrose, E., 1959, The Theory of the Growth of the Firm. New York: Oxford University Press.
Rhoades, S., 1973, The Effect of Diversification on Industry Profit Performance in 241 Manufacturing
Industries: 1963, Review of Economics and Statistic, 55: 146-155.
Rhoades, S., 1974, A Further Evaluation of the Effect of Diversification on Industry Profit Performance,
Review of Economics and Statistic, 56: 557–559.
Roberts, D., Georgia Pacific Decides to Leave the Spotlight, The Financial Times, 15 November 2005, p.
29.
Rumelt, R., 1974, Strategy, Structure, and Economic Performance. Cambridge, MA, Harvard University
Press.
Rumelt, R., 1982, Diversification Strategy and Profitability, Strategic Management Journal, 3: 359–369.
Rumelt, R., 1984, Toward a Strategic Theory of the Firm, in R. Lamb (ed.), Competitive Strategic
Management. Englewood Cliffs, NJ: Prentice-Hall.
Schoar, A., 2002, Effects of Corporate Diversification on Productivity, Journal of Finance, 57: 2379–
2403.
Servaes, H., 1996, The Value of Diversification During the Conglomerate Merger Wave, The Journal of
Finance, 51: 1201–25.
Shleifer, A. and R. Vishny, 1989, Management Entrenchment: The Case of Manager-Specific
Investments, Journal of Financial Economics, 25: 123–39.
Teece, D., 1984, Economic Analysis and Strategic Management, California Management Review, 26: 87–
110.
Teece, D., R. Rumelt, S. Winter, and G. Dosi, 1994, Understanding Corporate Coherence: Theory and
Evidence, Journal of Economic Behavior & Organization, 23: 1–30.
Villalonga, B., 2003, The Diversification Discount, Cambridge, MA: Harvard Business School.
Wernerfelt, B., 1984, A Resource-Based View of the Firm, Strategic Management Journal, 5: 171-181.
Zuckerman, E., 1999, The Categorical Imperative: Securities Analysts and the Illegitimacy Discount,
American Journal of Sociology, 104: 1398-1438.
Zuckerman, E., 2000, Focusing the Corporate Product: Securities Analysts and De-Diversification,
Administrative Science Quarterly, 45: 590–621.
36
Table 1. Variables definitions
Key variables UNIQUE1
Measure of the uniqueness of the corporate strategy based on the dot product of the normalized vector of sales for the focal firm across 1,107 available four-digit SIC code defined industries, with the similar normalized vector for the sales of all firms in the primary industry, that are covered in the corresponding annual period by at least one analyst
UNIQUE2
Measure of the uniqueness of the corporate strategy based on the dot product of the normalized vector of sales for the focal firm across 1,107 available four-digit SIC code defined industries, with the similar normalized vector for the sales of all firms in the primary industry
UNIQUE3
Measure of deviation from the typical (centroid) strategy of all firms that list SIC j as their primary SIC code. We first create the centroid strategy as a vector with s elements for each firm, where the elements 1, 2, 3…, s of this vector are equal to the mean sales across all firms in industry j in each of the s SICs that appear among the segment data for each of the firms in primary industry j, in a given year. The uniqueness of each firm’s strategy is then measured as the Euclidian distance between the industry centroid and an s-element vector constructed from the firm sales in each of the s SICs that are observed in the industry.
Adjusted Coverage Analyst coverage scaled by the primary industry coverage for that year.
Analyst Coverage Number of analysts issuing EPS annual forecasts for this fiscal year. We consider the calendar year of fiscal year period end date as the coverage date.
Imputed (segment-specific) Coverage Adjusted coverage of the firm computed based on the median pure play coverage of all its segments. Sales-weighted. Excess (strategy-specific) Coverage
Actual adjusted coverage minus imputed analyst coverage. We require that the included coverage for pure-play median firms is available in order to be able to compute the excess coverage.
Uniqueness Premium
Actual M/B minus the dot product of sj*, t and a vector of annual median M/B ratios for public single-segment firms in each industry. We require at least three single-segment firms within an SIC code in order to compute the median M/B ratio. If less than three single-segment firms are available, we then use the median single segment firm at the 3-digit SIC code level. If we still have less than 3 available firms, we then retrieve the median M/B ratio for single-segment firms at the 2-digit SIC level.
Diversification Discount
Actual M/B minus the dot product of sj*, t and a vector of annual median M/B ratios for public single-segment firms in each industry.
Other control variables Total assets Compustat data # 6, in million US$, logarithm Trade volume Logarithm of the number of shares traded (in million) Firm growth 3- or 5-year compound annual sales growth Issuance Dummy Dummy variable with value of 1 if the firm has issues debt or equity in the preceding fiscal year, in US$ million Earnings variability Coefficient of variation of the earnings over the preceding five years. # of firms in industry Logarithm of the number of firms in the focal firm’s primary SIC in the current fiscal year ROE Return on average equity Number of shares The number of shares, in millions Number of shareholders
Logarithm of the number of common shareholders (#100)
Market equity Logarithm of the common equity of the firm (#25 * #199) Average Herfindahl Index
Sales-weighted Herfindahl index for all SIC codes in which the firm operates. Recorded in percent.
Average market share Sales-weighted average of firm’s share of sales in each of the industries in which it operates. Recorded in percent.
RD_F R&D spending divided by total operating expenses minus the same ratio computed for the firm’s primary industry.
See Barth et al. (2001) for details.
ADV_F Advertising expense divided by total operating expense minus the same ratio computed for the firm’s primary
industry. See Barth et al. (2001) for details.
INTANG_F Ratio of recognized intangibles assets to total assets minus the median such ratio in the firm’s primary industry. See
Barth et al. (2001) for details.
DPT_F Depreciation expense divided by total operating expense minus the median such ratio for the firm’s primary
industry. See Barth et al. (2001) for details.
37
Table 2. Panel A. Key dependent and explanatory variables descriptive statistics. The mean statistics for the segment dummy variables is in percent. Statistics Mean Median Min Max Std. Dev. N Coverage Dummy 0.56 1 0 1 0.50 59,158 Number of Analysts 5.02 1 0 23 7.01 59,158 Coverage Share 0.16 0.02 0.00 1.00 0.27 59,158 Strategy-specific Coverage 0.08 0.01 -0.26 0.83 0.21 26,137 Segment-specific Coverage 0.16 0.10 0.01 0.73 0.17 26,137 UNIQUE1 0.24 0.18 0.78 0 0.78 59,158 UNIQUE2 0.26 0.21 0.8 0 0.78 59,158 UNIQUE3 1.02 0.67 0.00 4.82 1.05 59,155 Log (sales, US$ mln) 5.36 5.40 -6.91 12.56 2.28 59,012 Log (assets, US$ mln) 1.59 1.68 -5.30 2.61 0.50 59,042 Diversification Discount 0.15 -0.03 -1.51 4.98 1.06 58,486 Uniqueness Premium 0.16 -0.04 -1.47 4.87 1.06 58,486 Segment 1 Dummy 62.1% 1 0 1 0.49 59,158 Segment 2 Dummy 19.1% 0 0 1 0.39 59,158 Segment 3 Dummy 11.2% 0 0 1 0.32 59,158 Segment 4 Dummy 4.6% 0 0 1 0.21 59,158 Segment 5 Dummy 1.9% 0 0 1 0.14 59,158 Segment 6 Dummy 0.7% 0 0 1 0.08 59,158 Segment 7 Dummy 0.2% 0 0 1 0.05 59,158 Segment 8+ Dummy 0.2% 0 0 1 0.04 59,158 Table 2. Panel B. Key controls descriptive statistics Statistics Mean Median Min Max Std. Dev. N Log trading volume 16.27 16.33 11.36 19.70 2.01 59,118 Sales Growth (past 3 years) 0.29 0.24 -1.05 1.87 0.53 59,013 Issuances in prior year indicator 0.89 1.00 0.00 1.00 0.32 59,158 Equity Issuances 0.66 0.74 -5.12 5.57 2.57 44,246 Debt Issuances 3.17 3.33 -3.08 7.66 2.56 32,646 Earnings Vol. 0.23 0.21 -3.78 3.90 1.17 59,158 # of firms in industry 49 23 1 437 66 59,158 ROE -0.01 0.09 -2.09 1.06 0.46 59,105 Log number of shareholders 0.83 0.67 -3.22 4.31 1.64 58,297 Log common equity 4.62 4.62 -0.02 8.58 2.05 58,063 Average HHI 0.20 0.16 0.02 0.78 0.16 59,006 Average market share 0.08 0.01 0.00 0.66 0.15 59,006 RD_F (Barth et al., 2001) 0.01 0.00 -0.08 0.28 0.06 59,114 ADV_F (Barth et al., 2001) 0.00 0.00 -0.05 0.05 0.02 59,114 INTANG_F (Barth et al., 2001) 0.04 0.00 -0.11 0.36 0.10 59,157 DPT_F (Barth et al., 2001) 0.01 0.00 -0.08 0.15 0.04 59,063 Number of segments 1.69 1.00 1.00 11.00 1.12 59,158
38
Table 2. Panel C. Fama-French primary industry distribution of key explanatory and dependent variables. Tabulations of main variables across the sample averages for Fama-French industry definitions. The industry tabulations for strategy- and industry- specific coverage are based on the intersection of the samples of available observations for the latter two variables, and are shown in bold. We show tabulates sorted on UNIQUE1.
FF Industry
Code Industry Name Coverage Dummy
Number of
AnalystsCoverage
Share
Strategy-Specific
Coverage
Segment-Specific
CoverageUNIQUE1
UNIQUE3
Number of
segmentsUniqueness
Premium Div.
Discount3 Candy & Soda 0.65 7.16 0.41 0.03 0.47 0.10 1.13 1.43 0.58 0.57
28 Non-Metallic and Industrial Metal Mining 0.58 9.07 0.19 0.11 0.22 0.11 1.41 1.25 0.21 0.25
43 Restaurants, Hotels, Motels 0.61 6.50 0.20 0.10 0.23 0.11 1.08 1.37 0.18 0.19 11 Healthcare 0.51 4.36 0.19 0.08 0.27 0.12 1.16 1.44 0.16 0.16 41 Wholesale 0.59 5.53 0.23 0.09 0.23 0.14 1.09 1.67 0.10 0.10 13 Pharmaceutical Products 0.59 5.68 0.05 0.04 0.03 0.15 1.04 1.27 0.41 0.35 9 Consumer Goods 0.62 4.50 0.32 0.14 0.29 0.16 1.08 1.54 0.21 0.21
35 Computers 0.55 4.54 0.11 0.07 0.10 0.16 1.03 1.48 0.16 0.16 36 Electronic Equipment 0.52 4.68 0.07 0.04 0.08 0.17 0.99 1.32 0.15 0.14 48 Other 0.09 0.47 0.02 -0.02 0.13 0.17 1.14 1.68 -0.07 -0.12 10 Apparel 0.67 5.26 0.36 0.08 0.45 0.17 1.01 1.57 0.22 0.22 5 Tobacco Products 0.86 8.82 0.65 0.18 0.37 0.18 1.03 2.18 1.90 1.94
16 Textiles 0.66 4.25 0.37 0.11 0.39 0.19 1.06 1.78 0.13 0.14 34 Business Services 0.70 4.96 0.49 0.04 0.56 0.20 1.04 1.73 0.09 0.07 12 Medical Equipment 0.56 4.22 0.06 0.05 0.06 0.21 0.98 1.28 0.24 0.25 7 Entertainment 0.51 4.95 0.21 0.16 0.24 0.22 1.21 1.54 0.05 0.06
30 Petroleum and Natural Gas 0.69 5.26 0.45 0.02 0.43 0.23 0.84 2.28 0.30 0.31 2 Food Products 0.65 6.40 0.38 0.11 0.40 0.24 1.03 1.80 0.28 0.28
18 Construction 0.50 3.55 0.20 0.10 0.25 0.24 1.02 1.98 0.08 0.08 42 Retail 0.55 3.92 0.23 0.12 0.23 0.25 1.12 1.90 0.12 0.12 1 Agriculture 0.67 4.84 0.59 0.01 0.68 0.25 0.81 2.11 0.94 0.95
32 Telecommunication 0.49 4.68 0.09 0.05 0.13 0.26 0.99 2.14 0.02 0.02 37 Measuring and Control
i0.53 5.06 0.07 0.06 0.07 0.27 0.87 1.45 0.16 0.13
38 Business Supplies 0.51 3.33 0.08 0.06 0.08 0.27 0.97 1.48 0.07 0.08 6 Recreation 0.46 2.84 0.17 0.14 0.19 0.27 1.1 1.68 0.01 -0.03
33 Personal Services 0.51 5.81 0.09 0.05 0.12 0.27 1.25 1.95 0.03 0.05 31 Utilities 0.46 5.66 0.07 0.05 0.08 0.27 0.88 1.75 0.15 0.14 23 Automobiles and Trucks 0.55 4.36 0.16 0.09 0.20 0.28 1.28 1.48 0.22 0.25 4 Beer & Liquor 0.45 5.40 0.29 0.30 0.34 0.30 1.13 2.07 0.08 0.11
19 Steel Works Etc 0.69 6.41 0.29 0.13 0.19 0.31 1.02 2.03 0.18 0.17 24 Aircraft 0.69 5.75 0.17 0.12 0.11 0.34 1.06 2.09 0.17 0.17 21 Machinery 0.59 4.59 0.19 0.11 0.16 0.34 1.05 1.89 0.14 0.10 22 Electrical Equipment 0.52 3.12 0.20 0.08 0.21 0.35 1.02 1.93 0.14 0.12 39 Shipping Containers 0.68 6.73 0.28 0.09 0.25 0.37 1.04 2.08 0.15 0.16 15 Rubber and Plastic
d0.47 2.25 0.17 0.13 0.15 0.37 1 1.87 0.26 0.21
17 Construction Materials 0.61 4.30 0.34 0.12 0.37 0.38 1.02 2.18 0.23 0.22 29 Coal 0.81 7.11 0.62 0.37 0.37 0.39 0.83 2.49 0.40 0.37 20 Fabricated Products 0.49 2.34 0.25 0.07 0.43 0.4 1.06 1.96 0.07 0.05 40 Transportation 0.63 5.05 0.44 0.31 0.36 0.4 0.97 2.33 0.11 0.13 14 Chemicals 0.67 6.86 0.23 0.11 0.17 0.42 1 2.14 0.08 0.04 8 Printing and Publishing 0.67 7.42 0.25 0.17 0.18 0.43 1.04 2.35 0.11 0.11
26 Defense 0.72 5.36 0.33 0.17 0.18 0.46 1.17 2.41 0.09 0.08 27 Precious Metals 0.67 4.95 0.54 0.36 0.41 0.52 0.97 3.01 0.65 0.64
25 Shipbuilding Equipment 0.59 7.17 0.21 0.15 0.15 0.55 1.18 2.71 0.05 0.04
39
Table 2. Panel D. Key explanatory and dependent variables over time The industry tabulations for strategy- and industry- specific coverage are based on the intersection of the samples of available observations for the latter two variables, and are shown in bold.
Year Coverage Dummy
Number of Analysts
Coverage Share
Strategy-Specific
Coverage
Segment-Specific
Coverage UNIQUE1 UNIQUE3 Number of segments
Uniqueness Premium
Div. Discount
1985 0.32 3.64 0.15 0.11 0.26 0.31 1.12 1.9 -0.02 -0.01 1986 0.37 4.01 0.16 0.09 0.24 0.27 1.1 1.9 0.02 0.03 1987 0.39 4.21 0.16 0.10 0.22 0.26 1.09 1.8 0.08 0.10 1988 0.39 3.99 0.15 0.11 0.21 0.26 1.06 1.7 0.06 0.07 1989 0.47 4.62 0.16 0.11 0.18 0.25 1.05 1.7 0.08 0.09 1990 0.48 4.62 0.16 0.10 0.19 0.24 1.03 1.7 0.09 0.11 1991 0.49 4.27 0.17 0.10 0.19 0.23 1.03 1.6 0.11 0.12 1992 0.51 4.23 0.16 0.11 0.18 0.23 1.04 1.6 0.12 0.13 1993 0.52 4.52 0.16 0.10 0.17 0.23 1.06 1.6 0.08 0.08 1994 0.52 4.42 0.15 0.08 0.16 0.22 1.05 1.6 0.09 0.09 1995 0.53 4.44 0.15 0.09 0.15 0.21 1.07 1.5 0.07 0.08 1996 0.55 4.51 0.14 0.09 0.14 0.21 1.04 1.5 0.10 0.11 1997 0.57 4.70 0.14 0.08 0.13 0.2 1.05 1.5 0.18 0.20 1998 0.58 4.79 0.14 0.07 0.13 0.27 1 1.8 0.27 0.27 1999 0.59 4.82 0.16 0.07 0.14 0.27 1.02 1.8 0.18 0.20 2000 0.60 4.87 0.16 0.07 0.14 0.26 1.01 1.8 0.27 0.27 2001 0.60 5.08 0.17 0.07 0.15 0.25 1 1.7 0.25 0.26 2002 0.63 5.44 0.17 0.06 0.15 0.25 0.98 1.8 0.21 0.22 2003 0.66 5.75 0.18 0.06 0.15 0.25 0.97 1.7 0.19 0.20 2004 0.71 6.22 0.18 0.06 0.14 0.25 0.97 1.7 0.19 0.21 2005 0.75 6.69 0.19 0.06 0.14 0.22 0.95 1.7 0.20 0.22 2006 0.77 7.31 0.20 0.07 0.14 0.23 0.95 1.6 0.18 0.19 2007 0.83 8.36 0.23 0.06 0.17 0.22 0.94 1.6 0.28 0.29
40
Table 2. Panel E. Pairwise correlations. The p-values for each of the pair wise correlations are shown below the estimate, in parentheses. The estimates in bold are computed based on the sample of available data of strategy-specific analyst coverage and segment-specific analyst coverage.
Coverage Dummy
Coverage Count Adj. Coverage
Strategy-specific
Coverage
Segment-specific
Coverage UNIQUE3 UNIQUE1
Coverage Count 68.2% 100.0% (0.00)
Adj. Coverage 52.2% 52.0% 100.0% (0.00) (0.00)
Strategy-specific Coverage12 - 39.7% 76.6% 100.0%
- (0.00) (0.00) Segment-specific
Coverage 3.1% 2.5% 41.5% -3.7% 100.0% (0.00) (0.00) (0.00) (0.00)
UNIQUE3 -9.4% -28.5% -16.9% -28.3% -4.2% 100.0% (0.00) (0.00) (0.00) (0.00) (0.00)
UNIQUE1 0.4% 4.3% 1.3% -11.2% -13.7% 6.8% 100.0% (0.14) (0.00) (0.00) (0.00) (0.00) (0.00)
Number of segments 5.7% 12.1% 14.9% 18.8% 6.6% -20.2% 64.9% (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Persistence of the measures of uniqueness: autocorrelations.
Variable Rho1 Rho2 Rho3
UNIQUE1 0.85 0.77 0.70 UNIQUE2 0.82 0.75 0.68 UNIQUE3 0.90 0.84 0.78
12 We cannot compute correlation of the coverage dummy with strategy-specific coverage as the former does not vary when the latter is available.
41
Table 3. Analyst coverage regressions. This table provides estimates of the effects of corporate strategy choice on the level of analyst coverage. Models (1) and (7) are negative binomial regressions of a simple count measure of analysts covering the firm analyst on strategy choice and various industry and firm controls. Models (2) & (8) examine coverage as a dichotomous variable using a logit specification. The dependent measure is coded as 1 if the firm receives any coverage in a given year and 0 otherwise. Models (3) and (9) examine adjusted coverage measured as the percentage of all analysts in an industry which cover the focal firm, by the OLS method. Models (4) and (10) do the same allowing for truncation of the dependent variable in a tobit model, correcting for the left truncation in the dependent measure. Models (5) and (11) examine the strategy-specific analyst coverage. Models (6) and (12) examine the segment-specific analyst coverage. All variables are defined in Table 1. The t-statistics (in parentheses) are based on robust standard errors cluster-adjusted at the firm level. The ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Regressions include year fixed effects, industry firm effects (models (7)-(12)) and firm fixed effects (models (1)-(12)).
Coverage Coverage Dummy
Adj. Coverage
Adj. Coverage
Strategy-specific
Coverage
Segment-specific
Coverage Coverage Coverage Dummy
Adj. Coverage
Adj. Coverage
Strategy-specific
Coverage
Segment-specific
Coverage Variable
(end of prior fiscal year) Neg. Bin. Logit OLS Tobit OLS OLS Neg. Bin. Logit OLS Tobit OLS OLS (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Log Sales 0.171*** 0.208*** -0.001 0.015*** 0.007** -0.003 -0.001 0.414*** -0.001 0.019*** 0.000 0.000 (19.29) (13.82) (1.23) (7) (2.26) (1.55) (0.13) (8.62) (0.8) (7.92) (0.32) (0.25) Log trading volume 0.349*** 0.435*** 0.032*** 0.072*** 0.004* 0.002 0.193*** 0.684*** 0.017*** 0.05*** 0.006*** 0.001 (67.46) (37.17) (39.52) (44.96) (1.87) (1.61) (33.53) (25.06) (18.3) (27.57) (3.18) (0.76) Sales Growth (past 3 years) 0.189*** 0.21*** 0.005*** 0.022*** 0.019*** -0.001 0.138*** 0.059 0.004*** 0.01*** 0.032*** -0.001 (17.77) (10.59) (3.69) (7.78) (11.93) (0.91) (16.31) (1.5) (2.86) (4.03) (32.05) (1.12) Issuances in prior year 0.334*** 0.495*** 0.013*** 0.073*** 0.002 0.009*** 0.113*** 0.289*** 0.006*** 0.038*** 0.009** 0.009*** (16.35) (14.36) (5.33) (13.92) (0.37) (3.05) (6.21) (4.53) (2.65) (8.07) (2.2) (3.46) Earnings Vol. 0.005 0.013 -0.002*** -0.001 -0.001 0.0001 0.002 0.031* 0.0001 0.001 -0.002* -0.001** (0.93) (1.51) (3.91) (0.76) (0.65) (0.67) (0.54) (1.96) (0.07) (1.17) (1.93) (2.27) # of firms in industry -0.002*** 0.000 0.0001*** 0.0001*** 0.0001*** 0.0001*** -0.001*** -0.005*** 0.0001*** 0.0001*** 0.0001*** 0.0001*** (6.81) (0.04) (8.9) (5.08) (5.5) (3.53) (5.67) (4.9) (9.43) (10.28) (2.78) (5.34) ROE 0.294*** 0.331*** 0.016*** 0.045*** 0.004 0.0001 0.191*** 0.194*** 0.003* 0.019*** 0.005 0.0001 (15.35) (10.7) (9.49) (10.65) (1.26) (0.06) (12.85) (3.76) (1.74) (5.03) (1.55) (0.10) Log number of shareholders -0.084*** -0.108*** -0.003*** -0.011*** 0.0001 0.001 -0.087*** -0.021 -0.001 -0.006*** 0.000 0.000 (21.11) (11.02) (4.01) (9.3) (0.16) (0.6) (16.15) (0.63) (0.91) (3.66) (0.45) (0.54) Segment 2 -0.121*** -0.114*** 0.0001 -0.005 0.017*** 0.001 -0.082*** -0.034 0.003 0.003 0.015*** 0.003*** (7.94) (3.54) (0.16) (1.12) (7.08) (0.49) (5.93) (0.48) (0.99) (0.62) (9.75) (2.66) Segment 3 -0.12*** -0.081* 0.016*** 0.013** 0.13*** 0.078*** -0.098*** -0.107 0.008** 0.01* 0.147*** 0.13*** (6.04) (1.91) (4.81) (2.4) (10.49) (9.17) (5.55) (1.16) (2.31) (1.75) (14.37) (20.38) Segment 4 -0.219*** -0.258*** -0.007 -0.017** 0.307*** 0.072*** -0.116*** -0.014 0.000 0.000 0.526*** 0.056*** (7.54) (4.24) (1.35) (2.18) (19.24) (6.62) (4.92) (0.12) (0.08) (0.03) (49.71) (8.46) Segment 5 -0.286*** -0.40*** 0.001 -0.015 -0.003 0.003 -0.148*** -0.393** 0.007 0.002 -0.008** 0.008*** (7.54) (4.53) (0.13) (1.43) (0.69) (1.23) (4.81) (2.29) (1.19) (0.22) (2.5) (3.98) Segment 6 -0.389*** -0.418*** -0.002 -0.022 0.001 0.007* -0.236*** -0.699*** -0.002 -0.017 -0.002 0.012***
42
(6.72) (2.94) (0.17) (1.32) (0.15) (1.77) (5.38) (2.67) (0.17) (1.15) (0.45) (4.16) Segment 7 -0.726*** -1.38*** -0.062*** -0.134*** -0.025*** -0.001 -0.555*** -0.94*** 0.027** -0.005 -0.022*** 0.014*** (6.25) (6.65) (2.85) (4.8) (2.90) (0.10) (6.87) (2.78) (1.97) (0.21) (2.89) (2.89) Segment 8 -1.085*** -1.854*** -0.084*** -0.202*** -0.001 0.006 -0.866*** -1.369*** 0.029 -0.016 0.018 0.019** (7.23) (7.56) (3.08) (6.07) (0.08) (0.64) (7.43) (2.73) (1.66) (0.51) (1.36) (2.33) Log common equity 0.288*** 0.446*** 0.016*** 0.057*** 0.09*** -0.034* 0.156*** 0.393*** 0.012*** 0.041*** -0.010 0.024 (30.71) (26.54) (15.15) (24.58) (3.54) (1.94) (18.39) (9.93) (8.2) (16.43) (0.39) (1.57) Average HHI 0.436*** 0.517*** 0.201*** 0.384*** 0.074** -0.061*** 0.023 -0.638*** 0.148*** 0.293*** -0.008 -0.031 (8.86) (5.07) (18.64) (27.54) (2.18) (2.66) (0.5) (2.83) (18.05) (21.7) (0.22) (1.42) Average market share -0.133*** 0.939*** 0.762*** 0.82*** 0.173** -0.196*** 0.192*** 1.932*** 0.545*** 0.765*** 0.023 -0.08* (2.86) (7.5) (57.96) (55.65) (2.55) (4.22) (3.89) (5.93) (50.14) (46.17) (0.33) (1.78) RD_F (Barth et al., 2001) 1.774*** 2.462*** -0.027*** 0.177*** 0.021 0.018 0.42*** 2.495*** 0.031 0.207*** 0.022 -0.023** (17.18) (11.64) (2.82) (6.13) (0.65) (0.81) (3.56) (4.08) (1.42) (5.91) (1.21) (2.01) ADV_F -0.164 0.723 0.16*** 0.171* 0.125 0.028 -0.099 2.949 0.123* 0.174* -0.11* 0.061 (0.57) (1.09) (3.11) (1.94) (1.3) (0.43) (0.29) (1.62) (1.91) (1.67) (1.85) (1.63) INTANG_F -0.296*** -0.164 -0.033*** -0.085*** 0.017 -0.017* 0.333*** 2.055*** 0.026*** 0.010 -0.018* -0.011* (6.06) (1.41) (4.15) (5.79) (1.23) (1.81) (7.09) (7.35) (2.68) (0.67) (1.83) (1.82) DPT_F 0.617*** 0.922*** 0.028 0.135*** -0.061 0.033 -0.209 -0.549 -0.023 0.011 0.076** -0.05*** (3.75) (2.94) (1.44) (3.17) (1.36) (1.06) (1.28) (0.7) (0.85) (0.22) (2.52) (2.65) UNIQUE1 -0.249*** -0.588*** -0.084*** -0.144*** -0.043*** -0.029*** -0.248*** -0.695*** -0.054*** -0.111*** -0.069*** -0.026*** (6.64) (7.78) (13.53) (14.15) (5.17) (5.08) (8.02) (4.81) (10.26) (12.23) (9.07) (5.39) S&P 600 Index Membership 0.229*** 0.788*** 0.014*** 0.042*** -0.004 -0.002 0.041*** 0.249** -0.015*** -0.014*** -0.009*** 0.005*** (18.46) (17.36) (5.01) (8.93) (1.05) (0.65) (2.81) (2.41) (4.93) (2.92) (3.0) (2.7) Regulated industry Indicator -0.244* -0.552** -0.018 -0.07*** -0.1*** 0.029 -1.504*** -1.057* -0.047*** -0.066*** -0.007 -0.001 (1.73) (2.51) (1.44) (2.88) (2.73) (1.16) (34.33) (1.87) (2.93) (5.18) (0.22) (0.05) Reg F.D. dummy -0.768*** 1.239*** -0.038*** -0.013 -0.088*** -0.08*** -17.702 1.696*** 0.005 -0.045*** -0.11*** -0.078*** (17.76) (14.69) (5.99) (1.2) (9.79) (12.95) (0.05) (11.39) (1.15) (4.8) (11.88) (13.46) Observations 59,158 59,158 59,158 59,158 26,140 26,140 59,158 59,158 59,158 59,158 26,140 26,140 Chi-squared - 15030.8 - 24506.5 - - 11239.6 5629.8 - 10415.8 - - R-squared - - 53.4% - 70.7% 78.4% - - 78.7% - 49.5% 68.7% Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effects No No No No No No Yes Yes Yes Yes Yes Yes
Firm Fixed Effects Yes Yes Yes Yes Yes Yes No No No No No No
Table 4. Analyst coverage and corporate strategy conformity: first differences regressions. OLS regressions of various measures of coverage as in Table 3. All control variables are defined in Table 1. The t-statistics (in parentheses) are based on robust standard errors cluster-adjusted at the firm level. The ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.
∆Log (1+Coverage
Count) ∆Adj. Coverage ∆Strategy-specific
Coverage ∆Segment-specific
Coverage ∆Log Sales 0.081*** 0.005*** 0.011*** -0.005** (11.64) (3.06) (3.53) (2.47) ∆Log trading volume 0.042*** 0.002** 0.004** 0.002 (9.36) (2.39) (2.03) (1.67) ∆Sales Growth (past 3 years) 0.053*** 0.004*** 0.005*** -0.001 (17.82) (5.58) (2.81) (0.77) ∆Issuances in prior year indicator 0.006 0.002 0.003 0.001 (1.3) (1.68) (0.79) (0.32) ∆Earnings Vol. 0.001 0.000 0.000 0.000 (1.15) (0.39) (0.66) (0.1) ∆(# of firms in industry) -0.001*** 0.001*** 0.001** 0.001*** (2.84) (6.6) (2.26) (6.42) ∆ROE -0.055*** -0.005*** -0.011*** 0.001 (10.25) (4.35) (3.92) (0.69) ∆Log number of shareholders 0.009* 0.002** 0.003* -0.001 (1.95) (2.45) (1.85) (0.84) ∆Segment 2 0.109*** 0.011*** 0.019*** -0.003 (16.37) (7.13) (7.52) (1.53) ∆Segment 3 -0.003 0.13*** 0.106*** 0.09*** (0.12) (9.65) (3.74) (5.25) ∆Segment 4 0.157*** 0.374*** 0.237*** 0.085*** (3.61) (14.17) (5.78) (3.46) ∆Segment 5 -0.012 0.019*** 0.03*** 0.002 (1.24) (6.66) (5.73) (0.5) ∆Segment 6 -0.012 0.029*** 0.046*** 0.003 (0.93) (7.51) (5.37) (0.51) ∆Segment 7 -0.021 0.038*** 0.055*** 0.006 (1.33) (6.72) (4.16) (0.79) ∆Segment 8 -0.038** 0.042*** 0.072*** 0.004 (1.98) (5.41) (3.85) (0.29) ∆Log common equity -0.066** 0.06*** 0.079*** 0.02* (1.99) (4.72) (2.87) (1.75) ∆Average HHI -0.084* 0.034 0.045** 0.044*** (1.8) (1.58) (2.11) (3.77) ∆Average market share -0.123* 0.023 0.050 0.054*** (1.76) (0.6) (0.96) (3.39) ∆RD_F 0.223*** 0.011 0.027* -0.008 (3.12) (1.07) (1.81) (1.08) ∆ADV_F 0.492** 0.099* -0.057 0.148 (2.18) (1.89) (0.45) (1.5) ∆INTANG_F 0.144*** 0.035*** 0.029* -0.019* (4.08) (3.75) (1.95) (1.74) ∆DPT_F 0.011 -0.011 0.013 0.011 (0.13) (0.61) (0.35) (0.52) ∆UNIQUE1 -0.037** -0.124*** -0.126*** -0.069*** (2.12) (16.33) (9.46) (9.6) ∆S&P 600 Index Membership Indicator 0.033** -0.002 -0.002 -0.001
44
(2.21) (0.52) (0.39) (0.18) ∆Regulated industry Indicator 0.057 -0.038* -0.040 0.013 (1.32) (1.71) (1.01) (0.63) ∆Reg F.D. dummy -0.02** 0.004 -0.007 0.012*** (2.11) (1.42) (1.61) (4.23) Observations 56,979 56,979 23,426 23,426 R-squared 3.1% 4.8% 2.7% 1.4%
45
Table 5. Corporate value, analyst adjusted coverage, and corporate strategy uniqueness. The table presents the second stage of the two-stage least squares (2SLS) of the effects of analyst coverage and corporate strategy uniqueness on Tobin’s q, the diversification discount, and the uniqueness premium. We measure analyst coverage as adjusted coverage. We treat adjusted analyst coverage as endogenous variable and instrument it with the log trading volume and a dummy variable for S&P 600 membership. All variables are defined in Table 1. All regressions include year- and firm fixed effects. The t-statistics (in parentheses) are based on robust standard errors cluster-adjusted at the firm level. The ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.
M/B
Adj. M/B #1 (Diversification
Discount)
Adj. M/B #2 (Uniqueness
Premium) (1) (2) (3) Log Sales -0.107*** -0.11*** -0.101*** (6.68) (7.82) (7.17) Earnings Vol. -0.008 -0.009* -0.008 (1.31) (1.71) (1.38) ROE 0.158*** 0.128*** 0.135*** (7.67) (7.06) (7.41) Log number of shareholders -0.051*** -0.031*** -0.035*** (4.14) (2.87) (3.18) Log common equity -0.404*** -0.061** -0.065** (20.24) (2.41) (2.57) Average HHI -1.841*** -0.17*** -0.174*** (12.47) (4.92) (5.02) Average market share -5.36*** -0.172*** -0.181*** (13.06) (3.86) (4.04) Segment 2 -0.119*** -0.232*** -0.243*** (4.15) (3.79) (3.94) Segment 3 -0.239*** -0.27*** -0.279*** (6.08) (3.03) (3.11) Segment 4 -0.256*** -0.464*** -0.483*** (5.04) (3.4) (3.52) Segment 5 -0.338*** -0.491*** -0.436** (4.83) (2.74) (2.42) Segment 6 -0.449*** -0.343*** -0.351*** (4.41) (19.53) (19.84) Segment 7 -0.685*** -1.393*** -1.41*** (4.38) (10.75) (10.84) Segment 8 -0.64*** -4.383*** -4.537*** (3.12) (11.96) (12.32) Instrumented Adjusted Coverage 9.294*** 7.8*** 7.92*** (14.39) (13.6) (13.75) UNIQUE1 1.183*** 0.938*** 0.964*** (12.03) (10.9) (11.16) Regulated industry Indicator 0.072 0.234 0.298* (0.42) (1.55) (1.96) Reg F.D. dummy 0.467 0.685*** 0.669*** (0.52) (12.06) (11.72) Observations 59,158 59,158 59,158 Model Chi-squared statistic (p-value) 82542.8 (0.0) 1634.0 (0.0) 1717.5 (0.0)
46
Table 6. Panel A. Corporate value, analyst adjusted coverage, and corporate strategy conformity. The table presents the second stage of the two-stage least squares (2SLS) of the effects of adjusted coverage and corporate strategy uniqueness on measures of real performance: sales growth over the subsequent five years). We treat adjusted analyst coverage as endogenous and instrument it with the log trading volume and an indicator S&P 600 membership. We show only the main variables of interest, but control for all variables as in Table 5, Panel A. All variables are defined in Table 1. All regressions include year fixed effects. Models (1) through (6) include firm fixed effects, models (7) through (12) include industry effects (four-digit SIC). The t-statistics (in parentheses) are based on robust standard errors cluster-adjusted at the firm level. The ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.
Sales growth: 1-year 2-years 3-years 4-years 5-years 1-year 2-years 3-years 4-years 5-years (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Segment 2 -0.016*** -0.041*** -0.058*** -0.062*** -0.05*** -0.012*** -0.043*** -0.062*** -0.072*** -0.063*** (3.12) (5.49) (6.4) (6.37) (4.34) (3.53) (6.58) (7.55) (7.64) (5.42) Segment 3 -0.018** -0.048*** -0.06*** -0.051*** -0.032* -0.02*** -0.044*** -0.058*** -0.053*** -0.034** (2.57) (4.72) (4.7) (3.62) (1.82) (4.39) (5.16) (5.17) (4.07) (2.01) Segment 4 -0.008 -0.03** -0.053*** -0.063*** -0.036* -0.012* -0.041*** -0.068*** -0.085*** -0.065*** (0.86) (2.26) (3.31) (3.61) (1.69) (1.83) (3.55) (4.72) (5.14) (3.12) Segment 5 -0.016 -0.041** -0.072*** -0.061** -0.048 -0.028*** -0.052*** -0.086*** -0.083*** -0.066** (1.32) (2.28) (3.23) (2.5) (1.61) (3.08) (3.24) (4.24) (3.59) (2.28) Segment 6 -0.006 -0.022 -0.035 -0.047 -0.031 -0.015 -0.021 -0.033 -0.049 -0.032 (0.35) (0.85) (1.13) (1.38) (0.75) (0.99) (0.88) (1.1) (1.47) (0.77) Segment 7 0.024 0.025 0.006 -0.026 -0.014 0.06** 0.035 0.017 -0.023 0.004 (0.88) (0.63) (0.12) (0.49) (0.22) (2.59) (0.95) (0.36) (0.45) (0.06) Segment 8 0.026 0.085* 0.133** 0.176*** 0.278*** 0.038 0.045 0.079 0.121** 0.21*** (0.77) (1.74) (2.33) (2.98) (4.04) (1.43) (0.98) (1.42) (2.03) (2.98) Instrumented Adjusted Coverage 0.914*** 1.274*** 1.387*** 1.057*** 1.121*** 0.559*** 0.972*** 1.066*** 0.873*** 0.839*** (8.06) (7.48) (6.5) (4.38) (3.81) (15.59) (8.86) (6.87) (4.38) (3.26) UNIQUE1 0.086*** 0.114*** 0.121*** 0.065* 0.066 0.039*** 0.059*** 0.052* 0.010 -0.007 (4.97) (4.37) (3.62) (1.72) (1.46) (4.05) (2.91) (1.88) (0.31) (0.17) Observations 59,158 54,615 48,592 43,153 38,138 59,158 54,615 48,592 43,153 38,138 Quasi R-squared 32.4% 46.5% 56.1% 63.2% 70.2% 4.3% 4.3% 5.7% 6.6% 7.8% Industry Fixed Effects No No No No No Yes Yes Yes Yes Yes Firm Fixed Effects Yes Yes Yes Yes Yes No No No No No