Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
Wiley and Accounting Research Center, Booth School of Business, University of Chicago are collaborating with
JSTOR to digitize, preserve and extend access to Journal of Accounting Research.
http://www.jstor.org
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
Journal of Accounting Research
Vol. 28 Supplement 1990
Printed in U.S.A.
1. Introduction
In this paper, we examine analysts' decisions to follow firms, along
with institutional investors' decisions to hold these same firms in their
portfolios. Since the two decisions are interrelatedthrough institutions'
and brokers'customer/supplierrelationship, we develop a simultaneous
statistical model of the joint behavior. Taking account of the simulta-
neous decision context leads to different inferencesthan we wouldobtain
if we consideredthe determinantsof each decision separately.
A growing literature cites analyst coverage as a component of firms'
informationenvironments. For example, in Shores' [1990] study of the
"preemption"of annual earnings surpriseby interim disclosures,analyst
followingis a proxy for the level of interim information available about
a firm. Skinner [forthcoming]interprets an increase in analyst following
after a firm's common stock is listed for options trading as an indication
that more information is being gathered and disseminated about the
firm. Brennan and Hughes [1990] and Bhushan [1989b] equate analyst
following with the economy-wide amount of information gathering in
their models of stock splits and disclosure,respectively. Our motivation
for studyinganalyst followingis to identify characteristicsthat influence
* University of Michigan; t Massachusetts Institute of Technology. The authors are
grateful to Robert Egan and Melinda Su for research assistance, and to Lynch, Jones &
Ryan for providing the analyst data. We are indebted to Jeff Abarbanell, Stan Kon,
Maureen McNichols, Cindy Schipani, Dennis Sheehan, Doug Skinner, the referee, and
participants in the 1990 Journal of Accounting Research Conference for helpful comments
and discussions. We received valuable comments on an earlier paper from seminar partic-
ipants at Cornell University, Duke University, the University of Michigan, M.I.T., and
Washington University. Remaining errors and omissions are our responsibility.
55
Copyright (?, Institute of Professional Accounting 1991
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
56 JUDGMENT ISSUES IN ACCOUNTING AND AUDITING: 1990
successfully demonstrated its care as trustee by documenting its use of analyst research. It
is common for courts to accept trustees' search for and reliance on professional advice as
evidence of prudence. See, e.g., Chase v. 1pevear[419 N.E.2d 419 (1981)], Coulthard v. Speirs
[603 P.2d 1074 (1979)], and Estate of Scheuer [94 Misc.2d 538 (1978)]. Conversely, trustees
can be surcharged for lack of attention to the value of investments in their trust, e.g.,
Whitfield v. Cohen [S.D.N.Y., 682 F.Supp. 188 (1988)], Killey Trust [326 A.2d 372 (1974)],
and Wilmington Trust Co. v. Coulter [200 A.2d 441 (1964)].
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 57
and firms' decisions to retain or switch auditors. For example, auditors
decide on a level of audit testing based on such factors as the likelihood
of error, the importance of potential misstatement, and the adequacy of
internal controls. Management's design of internal controls is also based
on the chance of error, the amount of possible misstatement, and an
interest in reducing audit fees and receiving a clean opinion. Thus,
management's decision about internal controls depends on the audit
environment, and the auditor's decision about audit testing depends on
the internal control environment. Across firms, neither can be taken as
strictly exogenous to the other.
Using single-decision models in multiple-decision contexts creates a
statistical misspecification known as simultaneous equations bias, which
makes coefficient estimates unreliable. Our results indicate this effect is
present, since we find different inferences from single-equation models
and a two-equation simultaneous system. In particular, the association
between firm size variables and analyst following may be artifacts of this
bias, rather than a causal relation.
To focus on analysts' and institutions' joint endogenous decisions, we
examine year-to-year changes in analyst following and institutional
ownership, rather than levels of these two variables. The reason for this
decision is illustrated by considering firm size, a widely noted character-
istic associated with the information environment. Bhushan [1989b] and
Shores [1990], among others, note that large firms generally are more
heavily followed by analysts.3 Our question is: if a firm grows, do more
analysts decide to follow (or more institutions to hold) the firm? Note
that the cross-sectional association documented in previous work does
not imply an affirmative answer to our time-series question.
Year-to-year changes in variables provide a stronger test of causal
relations than do levels of those variables, since the levels of many
economic and noneconomic variables are cross-sectionally correlated
without any direct causal link. While correlations in changes do not
imply causality either, a failure to find correlation in changes can help
to distinguish between meaningful and spurious associations, because
they provide a necessary condition for the existence of a correctly
specified causal relation.4
We investigate both firm and industry characteristics as determinants
of analyst following and institutional ownership. Our results show that
analyst following increases more in firms with smaller prior analyst
following and in firms whose return volatility has declined, and that
3 Studies that suggest more information is generated for larger firms include Grant
[1980], Atiase [1985], Collins, Kothari, and Rayburn [1987], Freeman [1987], and Bhushan
[1989a].
4Measurement problems and leads or lags in relations between variables may hinder
attempts to find correlations empirically. We refer here to "correctly specified" causal links
to abstract from these empirical considerations which are not discussed until later in the
paper.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
58 P. C. O'BRIEN AND R. BHUSHAN
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 59
stock-related variables as of trading day -45, approximately two calendar
months prior to the year-end. This timing gives us roughly contempora-
neous data from all three sources, allowing for a small time lag in the
collection and publication of information in the IBES Summary and
S&P Stock Guide.5
We used exploratory data analysis on the data from fiscal 1985-87 to
specify the variables and their functional forms. We examined several
functional forms for analyst following and institutional ownership, lead-
lag relations among variables, and several variables that do not appear
in the paper, such as the firm's number of exchange listings, its S&P
stock rating, and membership in the S&P 500. We selected variables and
functional forms based on preliminary regression results and residual
plots, and on our interest in exploring hypotheses advanced in previous
work.
Though it is common in empirical work, exploratory data analysis for
model specification can impair inference, because classical statistical
methods presume that the data are drawn after the model and hypotheses
are specified. Technically, reusing data reduces the degrees of freedom
for statistical tests, and as a practical matter it can result in an overfit
model, descriptive within the sample but not outside it. To protect against
this potential, we retain fiscal years 1981-84 as a holdout sample. Our
results, reported in section 4, are reasonably robust.
Our final selection of variables appears in table 1. The jointly endog-
enous variables are the annual change in the number of analysts fore-
casting the firm's current-year earnings per share, D/$iANLST, and the
annual change in the number of institutions holding the firm's common
stock, D/$iINSTN. These are proxies for the change in analyst following
and the change in institutional ownership, respectively.
'The IBES data base is updated continuously as information arrives at Lynch, Jones &
Ryan, and the Summary is produced in the third week of each month. The monthly S&P
Security Owners' Stock Guide is "revised through the last business day of the prior month."
Therefore, data in the November issue are current as of the end of October.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
60 P. C. O'BRIEN AND R. BHUSHAN
TABLE 1
Definitionsof Variables
Variable Definition
Name
#ANLST*t The number of analysts with estimates of current-year EPS, as
reported in the IBES Summary File in the month prior to
the fiscal year-end.
#INSTN*t The number of institutions holding the stock, as reported in
the S&P Security Owners' Stock Guide in the month prior to
the fiscal year-end.
SIC The three-digit SIC code listed in CRSP for the firm.
SICGRO The net entry of firms to the industry over the five-year period
ending in the month prior to the fiscal year-end. Net entry is
the number of new firms listed in CRSP for the industry
minus the number of firms exiting the industry, divided by
the number of firms that were in the industry at the begin-
ning of the period.
REGIND Regulated industry: takes the value 1 if the industry is in SIC
421, 483, 493, 612, 621, 633, 805, or 809; and 0 otherwise.
Industry names are in table 3.
LNMVEQ* The natural logarithm of the market value of equity (in
$1,000s), 45 trading days prior to the fiscal year-end date.
Data are from the CRSP files.
MKTARET Market-adjusted return: the continuously compounded return
on the firm's common stock over trading days -294 to -45,
minus the value-weighted market return over the same
period (xIOO). Day 0 is the fiscal year-end date.
LNSHRS* The natural logarithm of the number of shares outstanding, as
listed in the CRSP files (cross-checked against data from the
S&P Guide).
BETA * The estimated systematic risk of the stock from a market
model regression using a value-weighted market index, esti-
mated over trading days -244 to -45. Day 0 is the fiscal
year-end date.
RESIDSE* The residual standard error (xOO) from the market model
regression described above in the definition of BETA.
* Variablenames markedwith an asteriskare used in first-differenced
form.
t The regressorsincludelagged,undifferenced#ANLSTand #INSTN.
6
on commissions. We assume an implicit link between commissions
revenue and analyst compensation, so analysts expect to derive more
benefit from following firms where more trades can be generated.
Analysts associated with full-service brokers may also be rewarded for
generating corporate finance fees, if the brokerage house is connected
with an investment bank. For example, an analyst who maintains good
relations with the management of firms he or she follows may attract
underwriting business from those firms to the broker's investment bank.
6 Stickel [1990] discusses the link between analysts' research activities and compensation.
Konrad and Greising [1989] and Torres [1989] discuss the controversy surrounding com-
mission-linked compensation for analysts.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 61
7 Our rationale for each exogenous variable is meant ceteris paribus with respect to the
Preliminary results did not differ when an equally weighted index from all markets was
used. Results differed when only the NYSE/AMEX index was used.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
62 P. C. O BRIEN AND R. BHUSHAN
'A precise decomposition of changes in firm size can be obtained from the sum of
DLNSHRS and the log of relative prices (in [Pt/P,_1]). MKTARET differs from this simple
price relative by the dividend payout, stock split adjustments, and the market adjustment.
However, MKTARET measures firm-specific performance, in line with the rationale for its
association with analysts' decisions. We have replicated our results using changes in the
natural log of the market value of equity, instead of DLNSHRS and MKTARET, without
altering our conclusions.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 63
analyst following, and the prior level of institutional ownership, all of
which have been used to establish the prudence of investments in legal
cases.10
If firm size is important to institutions because of prudence standards,
both price performance (MKTARET) and changes in shares outstanding
(DLNSHRS) should be associated with changes in institutional owner-
ship. However, the two components may differ if other factors come into
play. To the extent that size reflects an institutional preference for
"winners," there will be a positive association between stock performance
and changes in institutional ownership, but no link with changes in
shares outstanding is implied. If liquidity increases with the number of
shares outstanding, institutions will find it less costly to trade large
blocks in companies with more shares,1"implying institutional ownership
will increase along with shares outstanding, with no implication about
stock performance.
Institutions typically pay their managers based on the dollar value of
assets under management. More explicitly performance-based compen-
sation arrangements were prohibited until November 1985 by section
205(1) of the 1940 Investment Advisers Act. The prohibition, intended
to prevent investment managers from making overly speculative invest-
ments, was relaxed by the SEC's Release No. IA-996, which allows fees
based on performance. In late 1986, the Department of Labor opened the
door for pension funds to use incentive fee structures by ruling that
certain performance-based fee arrangements would not violate ERISA's
self-dealing prohibition.12 However, in August 1988, Institutional Investor
reported that incentive fees were still uncommon at corporate pension
funds, perhaps because of the complexity of establishing appropriate
performance benchmarks.13 Therefore, for the 1981-87 period covered by
this study, it is probably reasonable to assume institutional money
managers' fees were of the traditional percent-of-assets type.
Fees based on a percentage of assets under management provide a
weak link between performance and pay, but no extraordinary reward
for superior performance. On the other hand, the possibility of lawsuits
for lack of prudence is clearly greater for large losses than for gains.
Further, although ERISA requires trustees of employee benefit plans to
diversify plan holdings, modern portfolio theory is not universally ac-
cepted by the courts in evaluating fiduciaries, and many statutes and
case law precedents stipulate that each investment, evaluated separately,
10See, e.g., Chase v. Pevear [419 N.E.2d 419 (1981)] and Estate of Scheuer [94 Misc.2d
538 (1978)].
" See Ring [1987], Blanc [1986], and Priest [1985] for discussions of liquidity costs faced
by institutions because of their large holdings.
12 Investment Advisors Act [15 U.S.C.A. ? 80b-5(1)] and ERISA [29 U.S.C.A. ? 1106(b)
and 1108(c)].
13
Elliott [1988]. Cf. Paustian [1987], Fisher [1986], and Lipton [1986].
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
64 P. C. O'BRIEN AND R. BHUSHAN
and:
DY$<INSTNit= bo + bi DY$ANLSTit+ b2 SICGROit+ b3REGINDit
+ b4 MKTARETit + b5 DLNSHRSit + b6 DBETAit (2)
+ b7DRESIDSEit + b8LY$ANLSTit
+ bgLYI$INSTNit
+ e2it
However, to the extent that the two dependent variables are jointly
determined by similar factors and equations (1) and (2) capture these
relations imperfectly, the unexplained portions of analyst following and
institutional ownership, e1itand e2it, will be correlated. Since DYI$INSTN
appears on the right-hand side of equation (1) (similarly for DYI$ANLST
14
ERISA [29 U.S.C.A. ? 1104(a)(1)(C)]. Cases where the court required each investment
to pass a prudence test include Chase v. Pevear [419 N.E.2d 1358 (1981)] and Estate of
Beach [542 P.2d 994 (1975)]. See Young and Lombard [1985] for a discussion of fiduciary
responsibility.
15
Badrinath, Gay, and Kale [1989] make a similar argument.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 65
in equation (2)), the correlation between the error terms implies corre-
lation between the regressor DXIUNSTNand the error term elit, violating
OLS assumptions. This correlation causes the single-equation coeffi-
cients to be biased and inconsistent.
We use two-stage least squares (2SLS) to eliminate the simultaneity.
That is, in the first stage, we regress each of the jointly endogenous
variables against all variables other than DYI$INSTNand DI$ANLST.
The fitted values from these regressions, which by construction are
independent of their respective error terms, are used as instrumental
variables in the second stage.
To avoid creating a singular covariance matrix in the second stage, we
identify the two-equation system by specifying regressors that influence
only one of the two endogenous variables. Based on our argument that
institutions form a clientele with distinct preferences about systematic
risk, DBETA should influence analyst following only through its effect
on institutions. It is therefore a natural candidate for exclusion from the
analyst equation. Our premise for SICGROis that analysts find it costly
to become informed about industry products and processes, so SICGRO
should affect institutional ownership only via analyst following. Exclud-
ing one variable from each equation makes the following system just-
identified:
D1*$ANLSTit
= + C2 SICGROit+ C3REGINDit
Co + C1DYI$INSTNit
(3)
+ C4MKTARETit + c5 DLNSHRSit + C7DRESIDSEit
+
+ c8 LY$ANLSTit c9 LYI$INSTNit
+ e3it
and:
D/I$INSTNit
= do + d1 D/IANLSTt + d3 REGINDit
+ d4 MKTARETit+ d5 DLNSHRSit + d6 DBETAit (4)
+ d7 DRESIDSEit + d8 LY$ANLSTit
+ d9 LY$<INSTNit
+ e4it
In the second stage of 2SLS we estimate (3) and (4), substituting the
instruments constructed in the first stage for DYI$INSTN in equation (3)
and DY$<ANLSTin equation (4). Since both equations are just-identified,
2SLS gives identical results to other simultaneous equations procedures
such as maximum likelihood or three-stage least squares. For comparison,
we also estimate equations (3) and (4) using ordinary least squares (OLS).
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
66 P. C. O'BRIEN AND R. BHUSHAN
TABLE 2
Sample Selection
Panel A; Intersection of Analyst and Stock Market Data Sets
IBES NYSE/AMEX NASDAQ
Firms in data set ............. ............. 5,811 5,019 7,659
Firms matched by CUSIP1......... .......... 4,920 2,545 2,854
Matched firms with no industry change1 ....... 4,254 2,267 2,466
AMEX and those matchedon NASDAQ,becausesome firms changedfrom OTC to one of the major
exchanges(or the reverse)between 1976 and 1986, and so appearon both. We used CRSP'sIPERM
identificationnumberto verifythat these were identicalfirms.
the regression results. In 1984, Energy Management Co. (CUSIP 29270110) had a measured
change in volatility of 13.84, more than twice the range in volatility of all other observations.
In 1985, Medical Care Int. (CUSIP 58450510) was the only sample firm in the Health and
Allied Services Industry (SIC 809) which had a five-year industry growth rate of 1300%,
more than quadruple the next-highest observation. Each of these observations influenced
one regression coefficient substantially (for volatility and industry growth, respectively)
but had little effect on other coefficients.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 67
Tables 3 and 4 give information on the sample's industry and time
composition. In table 3, the 38 industries are reported by name, along
with the number of sample firms in each. SIC 671, Holding Offices,
contains 31% of the sample firms. About one-third of the firms in SIC
671 are bank holding companies, but it includes many insurance holding
companies, several of the regional telephone companies, and other hold-
ing companies whose operating subsidiaries are in diverse industries (e.g.,
TABLE 3
Sample Industries
Numberof
SIC IndustryName Sample
Firms
131 Crude Petroleum and Natural Gas 61
201 Meat Products 6
205 Bakery Products 3
221 Broadwoven Fabric Mills, Cotton 8
232 Men's and Boys' Furnishings 14
243 Millwork, Plywood & Structural Members 7
264 Misc. Converted Paper Products 11
275 Commercial Printing 15
283 Drugs 46
287 Agricultural Chemicals 4
301 Tires and Inner Tubes 9
324 Cement, Hydraulic 5
332 Iron and Steel Foundries 8
343 Plumbing and Heating, Except Electric 7
349 Misc. Fabricated Metal Products 18
354 Metalworking Machinery 13
358 Refrigeration and Service Machinery 11
364 Electrical Lighting and Wiring Equipment 14
369 Misc. Electrical Equipment and Supplies 12
379 Misc. Transportation Equipment 5
384 Medical Instruments and Supplies 40
394 Toys and Sporting Goods 9
421 Trucking and Couriers Services, Except Air 17
483 Radio and Television Broadcasting 13
493 Combination Utility Services 32
503 Lumber and Construction Materials 3
512 Drugs, Proprietaries and Sundries 10
533 Variety Stores 6
566 Shoe Stores 5
591 Drug Stores and Proprietary Stores 14
612 Savings Institutions 21
621 Security Brokers and Dealers 11
633 Fire, Marine and Casualty Insurance 15
671 Holding Offices 224
721 Laundry, Cleaning and Garment Service 4
751 Automotive Rentals, No Drivers 4
805 Nursing and Personal Care Facilities 8
809 Health and Allied Services, NEC 3
716
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
68 P. C. O'BRIEN AND R. BHUSHAN
TABLE 4
Sample Distribution Information
Panel A: Number of Observations by Fiscal Year-End and Fiscal Year
Fiscal Numberof Fiscal Numberof
Year-End Observations Year' Observations
1 79 81 338
2 41 82 411
3 64 83 431
4 44 84 473
5 41 85 499
6 115 86 455
7 35 87 393
8 71 3,000
9 124
10 71
11 59
12 2,256
3,000
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 69
significance is based on assuming that all observations are independent
draws. This assumption is clearly violated in some cases. For example,
the number of institutions holding a firm's stock is correlated from year
to year.
The correlations in table 5 are important for comparing inferences
from our first-differenced model with the levels-form correlations re-
ported in other studies, particularly since our variable definitions differ
from those in other studies. Even in first-differenced form, our dependent
variables DY$<ANLSTand DY$<INSTNare apparently strongly correlated
with each of the size-related variables, MKTARET and DLNSHRS, as
well as with other regressors. Some of these associations disappear or are
reversed in the simultaneous equations estimation in section 4, but the
table 5 correlations should reduce concerns that our later results are
simply due to misspecification of, for example, the functional form or
the timing of the association. Moreover, our exploratory data analysis of
the 1985-87 subsample did not in general reveal stronger associations
with other specifications of functional form or between the levels of these
variables and lagged levels.
4. Results
As discussed in section 2, we model analyst following and institutional
ownership as simultaneous equations and make initial estimates using
the 1985-87 subperiod. In table 6, we present results for single-equation
regressions as well as simultaneous estimation of the two-equation system
on this subsample.
Within each subsample, we pool time-series and cross-section obser-
vations. Since the data are annual changes, the possibility that we are
pooling correlated observations is not as great as when variables are
measured in levels, but this still may be a concern. For example, if
changes in analyst following or institutional ownership are serially cor-
related, consecutive years' observations for a single firm are not inde-
pendent draws. We tested for this possibility by sorting the data by year
within firm and computing the first-order residual autocorrelation and
Durbin-Watson statistic. The estimated residual autocorrelations in both
2SLS equations are about -0.04, insignificant at the 5% level.18
Two single-equation OLS regressions are reported for each endogenous
variable in table 6, the first including all possible regressors, and the
second in the same form as the just-identified 2SLS model of equations
(3) and (4). The results reported in the 2SLS column for each endogenous
18 Under the null hypothesis of no autocorrelation, the standard error on this estimate
is approximately 1/ln or 0.027. The Durbin-Watson statistics are 2.083 for the analyst
equation and 1.925 for the institutions equation. Tabled values for the statistic of which
we are aware (King [1983] and Savin and White [1977]) extend only to N = 300, while our
sample size is 1,347. For N = 300, k' = 8, the region [da, 4 - dJ] in which the statistic fails
to reject the null of no autocorrelation at the 5% level is [1.859, 2.141].
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
70 P. C. O'BRIEN AND R. BHUSHAN
Ccl)
CI CI -- cq c o C) cq C
o Cl)
C*o m C
H
o) -- cq q co )
Cl)
t - C:) cq C CD 00 C) c
o C C C oo q
4k I~~~~~~~~~~~~~~~~~-
- 1-- LO 00 m q "It ) cq C) C) C. C) C
Cl * LO* * co *C*l)
o * o * *l)
co ** * * * co
Ct
CR C C) C) 00 .t D>sOR . . .
CDOHHOHO . .
0
Cl) .. m C
t O H
o
.C
N H CS o
ClC
t ~0. O cOO CD o- o 2..2) . HHC
q * * * * * CI * * * * Cl * * * * * U
2.C.
C2 L O LO Lo 00 0 Cl) v co 00 CD . 0
q q~~~~~~~~~~~C q C0
C-0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
II o
z ) F N*c HoF *t *H * * F * *O * * *
Cl) ~ ~ ~ ~ ~ ~~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~~Q0
C O) CgL C) g12 o CZ
0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~4
:)2 Z E~~~
666~~~~-~~ ?~~~ ?~~~
Z 666~~~~-~~ 00
u Cl) ~ Cl) )
DC Lo 12
I I 22~~~~~~~~~
z) Q~~~~~~~ I~~~~i~
2co
* * U) Cl) p-)
* Q~~~ CID Cla)
m * *
- ~~~~~~~00
Cd ~~~~~~~~~~~Cl) Cd 1
P-4 P-4 P-4~~~~~~~~~~~~~~~~~~~.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 71
TABLE 6
OLS and 2SLS Regressions of Analyst Following and Institutional Investor Ownership on
Firm and Industry Characteristics, 1985-87
variable are from joint estimation of the two-equation system. The 2SLS
estimation explains about 8% of the variation in changes in analyst
following and about 39% of the variation in changes in institutional
ownership.
From the 2SLS results, changes in analyst following appear positively
associated with net entry to the industry and regulation, and negatively
associated with preexisting analyst following, as expected from our dis-
cussion in section 3. Our expectation that analysts would prefer volatility
is refuted by the significantly negative coefficient on that variable.
Changes in institutional ownership are positively associated with stock
performance, increases in shares outstanding, lagged analyst following,
and lagged institutional ownership, as expected. Institutions do not
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
72 P. C. O'BRIEN AND R. BHUSHAN
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 73
TABLE 7
OLS and 2SLS Regressions of Analyst Following and Institutional Investor Ownership on
Firm and Industry Characteristics, 1981-84
Coefficient Estimates (t-statistics in parentheses)
D#ANLST D#INSTN
OLS OLS 2SLS OLS OLS 2SLS
INTERCEPT 0.39 0.38 0.29 0.87 0.75 2.13
(5.29) (5.19) (2.48) (1.31) (1.17) (1.03)
D#INSTN 0.02 0.02 0.09
(8.71) (8.80) (1.60)
D#ANLST 1.85 1.84 -1.08
(8.71) (8.68) (-0.26)
SICGRO 0.32 0.32 0.34 -0.93
(2.31) (2.25) (2.13) (-0.75)
REGIND 0.34 0.35 0.36 -0.91 -0.81 0.06
(2.80) (2.87) (2.58) (-0.85) (-0.76) (0.04)
MKTARET 0.32 0.35 -0.19 7.37 7.59 8.89
(2.31) (2.55) (-0.39) (6.08) (6.44) (3.99)
DLNSHRS 0.31 0.34 -0.54 12.73 12.67 14.63
(1.83) (1.99) (-0.69) (8.56) (8.53) (4.60)
DBETA 0.17 2.28 2.32 2.96
(1.32) (2.05) (2.08) (1.99)
DRESIDSE -0.22 -0.19 -0.19 -0.11 -0.11 -0.79
(-2.58) (-2.32) (-1.96) (-0.15) (-0.15) (-0.63)
L#ANLST -0.03 -0.03 -0.05 0.33 0.33 0.26
(-2.91) (-2.87) (-2.46) (3.51) (3.51) (1.84)
L#INSTN1 0.41 0.41 0.39 -0.51 -0.51 0.70
(6.21) (6.25) (5.11) (-0.87) (-0.88) (0.38)
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
74 P. C. O'BRIEN AND R. BHUSHAN
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
ANALYST FOLLOWING AND INSTITUTIONAL OWNERSHIP 75
Other coefficient values are similar in magnitude and statistical signifi-
cance to those in the first column of table 6.
5. Conclusions
We find evidence of a behavioral link between analysts' decisions to
follow firms and differential costs and benefits of gathering information.
We find that analysts prefer industries with increasing numbers of firms
and regulated industries and tend to avoid volatility and competition
from preexisting analyst following. We find no evidence of a causal link
between changes in analyst following and changes in firm size, once the
simultaneity with institutions' decisions is eliminated. Our evidence
supports a behavioral association between institutions' decisions to hold
firms' common stock and changes in firm size (both price changes and
changes in shares outstanding) and prior analyst following. Somewhat
paradoxically, we find that institutions seem to prefer firms whose risk
has increased. Overall, our results are reasonably robust when the model
is tested in an earlier time period.
We examine analyst following from a perspective different from that
of most prior information environment studies. By examining changes
in analyst following along with changes in institutional ownership, we
produce more demanding tests of the causal relations between firm and
industry characteristics and analysts' and institutions' decisions. The
lack of evidence that increasing firm size increases analyst following,
once feedback effects between institutions and analysts are considered,
calls into question some conventional assumptions about how size affects
information production. If analyst following is a reasonable proxy for
information production, our results fail to support a direct causal link
between it and size, instead suggesting an indirect link via institutions'
simultaneous decision.
Multiple-decision settings are probably the rule, not the exception, in
the real world, yet accounting researchers often consider decisions as
single endogenous events. Cross-sectional variation in firm characteris-
tics such as capital structure or the existence of executive bonus plans
are probably jointly endogenously determined, not exogenous to such
accounting decisions as audit qualifications or accounting method
choices. In our study, taking account of simultaneity between analysts'
and institutions' decisions alters the empirical results and therefore alters
implications for modeling information production. Similar results may
pertain in other accounting decision contexts.
REFERENCES
ATIASE, R. "Predisclosure Information, Firm Capitalization, and Security Price Behavior
Around Earnings Announcements." Journal of Accounting Research 23 (1985): 21-36.
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions
76 P. C. O'BRIEN AND R. BHUSHAN
This content downloaded from 128.192.114.19 on Sat, 30 May 2015 03:56:02 UTC
All use subject to JSTOR Terms and Conditions