Description
Abstract
We test the relation between financial and social disclosure and the cost of equity capital for a sample of
Canadian firms with year-ends in 1990, 1991 and 1992. We find that, consistent with prior research, the quantity
and quality of financial disclosure is negatively related to the cost of equity capital for firms with low analyst
following. Contrary to expectations, there is a significant positive relation between social disclosures and the cost
of equity capital. This positive relationship is mitigated among firms with better financial performance. We consider
some biases in social disclosures that may explain this result. We also note that social disclosures may benefit the
firm through its effect on organizational stakeholders other than equity investors
Social disclosure, ?nancial disclosure
and the cost of equity capital
Alan J. Richardson*, Michael Welker
School of Business, Queen’s University, Kingston, Canada K7L 3N6
Abstract
We test the relation between ?nancial and social disclosure and the cost of equity capital for a sample of
Canadian ?rms with year-ends in 1990, 1991 and 1992. We ?nd that, consistent with prior research, the quantity
and quality of ?nancial disclosure is negatively related to the cost of equity capital for ?rms with low analyst
following. Contrary to expectations, there is a signi?cant positive relation between social disclosures and the cost
of equity capital. This positive relationship is mitigated among ?rms with better ?nancial performance. We consider
some biases in social disclosures that may explain this result. We also note that social disclosures may bene?t the
?rm through its e?ect on organizational stakeholders other than equity investors. # 2001 Published by Elsevier
Science Ltd.
Regulators have argued that equity markets
require comprehensive and transparent disclosures
of value-relevant information by ?rms in order to
function e?ciently (e.g. Levitt, 1999). Theoreti-
cally, adopting such a ‘‘disclosure position’’ (Gib-
bins, Richardson, & Waterhouse, 1990) should
bene?t ?rms through lower cost of capital for at
least two reasons. First, increased disclosure by
?rms reduces transaction costs for investors
resulting in greater liquidity of the market and
greater demand for the ?rm’s securities (e.g. Dia-
mond & Verrecchia, 1991). Second, increased dis-
closure reduces the estimation risk or uncertainty
regarding the distribution of returns (Clarkson,
Guedes, & Thompson, 1996).
In spite of the regulatory and theoretical sup-
port for increased disclosure by ?rms, direct evi-
dence of a negative empirical relation between
disclosure levels and the cost of capital is limited
(e.g. Botosan, 1997; Botosan & Plumlee, 2000, on
the cost of equity capital, and Sengupta, 1998 on
the cost of debt). In part, the lack of strong
empirical ?ndings on the relationship between
disclosure and cost of capital may be an artifact of
the markets and information set that are used in
empirical tests. If there is little variation in the
information disclosed due to e?ective regulatory
interventions, or if analysts routinely generate
information independently of the ?rms’ own dis-
closures, then the power of empirical tests will be
signi?cantly reduced. For example, Botosan
(1997) documents a statistically signi?cant nega-
tive relation between the level of ?nancial dis-
closure and cost of equity capital for her sample of
USA manufacturing ?rms, but this relation holds
0361-3682/01/$ - see front matter # 2001 Published by Elsevier Science Ltd.
PI I : S0361- 3682( 01) 00025- 3
Accounting, Organizations and Society 26 (2001) 597–616
www.elsevier.com/locate/aos
* Corresponding author.
only for the subset of her sample characterized by
limited analyst following.
A stronger test of the relationship between cor-
porate information disclosures and the cost of
equity capital is possible by choosing markets and
information sets where, ex ante, corporate dis-
closures play a larger role in market valuations. In
this paper we test this relationship in Canadian
markets and with both ?nancial and social dis-
closures. Both of these extensions to the literature
should improve the power of statistical tests as
explained later.
First, we examine a set of Canadian ?rms, pro-
viding an assessment of the bene?ts of expanded
disclosure in an environment other than the Uni-
ted States of America (US). Since the US equity
markets are claimed to be among the most
sophisticated in the world (e.g. Levitt, 1999),
including some of the most stringent disclosure
standards in the world, this extension is poten-
tially important. The generally less comprehensive
required disclosures in Canada create an environ-
ment where variation in voluntary disclosure
could be very important.
1
Second, we examine the relation between both
social and ?nancial disclosures and cost of equity
capital estimates. The past literature has only
examined the relation between ?nancial disclosure
and cost of capital. As summarized in Botosan,
past literature suggests that ?nancial disclosures
could in?uence the cost of capital because the dis-
closures reduce information asymmetry and/or
estimation error. As Richardson, Welker, and
Hutchinson (RWH, 1999) discuss, there are a
number of reasons to suspect a relation between
the cost of equity capital and social disclosure as
well. Past empirical examinations of the relation
between social performance, or social disclosure,
and equity market measures such as realized
equity returns have generally been poorly speci-
?ed, leading to results that are di?cult to inter-
pret. RWH provide an extensive review of this
literature and present a comprehensive model
outlining the ways that social performance and
disclosure about that performance might in?uence
equity market measures. Their analysis focuses on
three distinct ways in which social performance
and disclosure could a?ect the cost of capital.
Social disclosure could play a role similar to
?nancial disclosure and reduce the cost of equity
capital by reducing transaction costs and/or
reducing estimation error. In addition to these
two e?ects, social disclosure could in?uence the
cost of equity capital directly through investor
preference e?ects if investors are willing to
accept a lower expected return on investments
that also ful?ll social objectives. The relationship
between the cost of capital and social disclosures/
performance is one of the issues identi?ed by
Epstein (2000) for future research in his review of
the ?eld.
Consistent with the past literature, we ?nd a
signi?cant negative relationship between the cost
of equity capital and ?nancial disclosure for those
?rms with a small ?nancial analyst following.
Contrary to expectations, we ?nd a signi?cant
positive relationship between social disclosure
and the cost of equity capital. The cost of
equity penalty for ?rms with extensive social
disclosure is mitigated by higher ?nancial
performance.
1. Hypothesis development
The relation between ?nancial disclosure and
the cost of equity capital has been extensively
developed in the past literature (Clarkson et al.,
1996; Diamond & Verrecchia, 1991). Botosan
(1997), for example, argues that ?nancial dis-
closure could result in decreased cost of capital
because expanded disclosure reduces estimation
risk, decreasing the total risk in owning the
equity security, or reduces risk by decreasing
1
The view that US disclosure practices provide more infor-
mation than Canadian practices is apparently widely held.
Nearly 90% of the Canadian analysts surveyed between
November 1994 and January 1995 on behalf of the Toronto
Stock Exchange’s Committee on Corporate Disclosure respon-
ded that disclosure was better in the USA. None of the analysts
felt that disclosure was better in Canada. Reasons provided for
the belief that disclosure is better in the USA included: more
stringent regulation, greater volume of information and more
detailed segmentation of information (Committee on Corpo-
rate Disclosure, 1995).
598 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
information asymmetries and, hence, adverse
selection risk. In either case, an inverse relation
between ?nancial disclosure and cost of equity
capital is predicted.
Richardson et al. (1999) argue that there are at
least three reasons to expect a similar relation
between social disclosure and the cost of equity
capital.
2
They conclude that there may be a direct
in?uence of social disclosure on the cost of equity
capital either through investor preference e?ects,
or through reduced information asymmetry or
estimation risk. The e?ects stemming from
reduced information asymmetry and/or estimation
risk follow directly from the literature on ?nancial
disclosure. If information about social activities is
relevant to assessing the ?rm’s prospects, then
enhanced disclosure of social activities has the
same e?ect as enhanced disclosure of other ?nan-
cial activities.
Investor preference e?ects arise if investors are
willing to accept a lower rate of return on invest-
ments by an organization that supports a social
cause for which some investors have an a?nity.
3
This suggestion is consistent with the emergence of
Green Funds and Ethical Investing (e.g. Reyes &
Grieb, 1998). It also has a direct relationship to
the literature in organizational behavior, manage-
ment, and marketing that suggests that advertising
with a social dimension can be employed to legit-
imate the ?rm in the eyes of consumers and con-
tribute to the ?rms’ product/service market
success (e.g. Garrett, 1987; Menon & Menon,
1997). This literature suggests that consumers
‘‘vote’’ with their dollars and may (rationally)
choose to pay more to both acquire a product or
service and support a social cause for which they
have an a?nity. The extension of this literature to
the capital market is straightforward if investment
decisions are recognized as decisions to forego
current consumption in favor of future consump-
tion. There is also considerable anecdotal evidence
suggesting a link between investor preferences and
social reporting. For example, Downing (1997)
reports that managers of the Canadian Pension
Plan’s $100 billion fund, among other investors,
might be attracted by the information provided by
social reporting.
Our empirical examination does not attempt
to discriminate between these potential e?ects,
but does stand in marked contrast to the past
literature examining the equity market con-
sequences of social disclosure that has tended to
focus on the relation between social performance,
social disclosure and ex-post measures of ?nancial
performance. Ours is the ?rst empirical examina-
tion of social disclosure practices to explicitly
examine the cost of equity capital and the ?rst to
jointly examine the e?ects of both ?nancial and
social disclosure. The speci?c hypotheses we
examine are outlined later, each stated in alter-
native form.
The past literature has suggested that increased
?nancial disclosure reduces information asym-
metry and/or estimation risk. This literature also
suggests that enhanced ?nancial disclosures reduce
the cost of equity capital.
H1a: There is an inverse relation between the
level of ?nancial disclosure and the cost of
equity capital.
The arguments presented earlier suggest that
similar relations exist between the level of social
disclosure and information asymmetry, estimation
risk and the cost of equity capital. Also, if, as has
often been assumed in the past literature, there is a
perceived positive relation between the level of
social disclosure of the ?rm and the ?rm’s social
performance, increased social disclosure may
reduce the cost of capital through investor pref-
erence e?ects.
2
RWH also argue that disclosure about social activities
undertaken by the ?rms could provide investors with informa-
tion about future cash ?ows, or in terms consistent with the
model presented later, future abnormal earnings. This link
might exist, for example, if social activities decrease expected
future regulatory costs, in?uence consumers to acquire the
?rm’s products/services thereby increasing the ?rm’s contribu-
tion margins or market share, or reduce the costs of implicit or
explicit contracting. In short, if social activities have net present
value consequences, then information about social performance
should in?uence investors’ assessments of the abnormal earn-
ings the ?rm can earn in the future (see Scaltegger & Figge,
1998).
3
This linkage is more thoroughly discussed in RWH (1999).
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 599
H2a: There is an inverse relation between the
level of social disclosure and the cost of equity
capital.
Finally, we consider the interaction between
?rms’ own disclosures about their activities and
the information available from third parties. We
hypothesize that the relationship will di?er for
?nancial disclosure and social disclosure. The
literature on ?nancial disclosure suggests that the
bene?ts of ?nancial disclosure may be greater
for ?rms with little analyst following (Botosan,
1997). This assumes that stakeholders are
concerned with a ?rm’s ?nancial performance but
that the richness of the information set available
to them varies depending on the number of
analysts preparing independent reports on the
?rm. In the absence of information about the
?rm from analysts, the ?rm’s own disclosures are
the key source of information. The bene?ts of
better ?nancial disclosure are primarily realized
when other information sources are absent.
Therefore we predict a negative relationship
between the cost of capital ?nancial disclosure
where there is a small number of analysts follow-
ing a ?rm.
H3a(i): The relation between ?nancial dis-
closure levels and the cost of equity capital is
mediated by the level of analyst following.
The e?ect of social disclosure is expected to
follow the same pattern to the extent that social
disclosures inform stakeholders’ expectations of
the ?rm’s ?nancial performance. The importance
of social performance to stakeholders, however,
has been theorized to increase with the size of the
organization. Stinchcombe (1965), for example,
has argued that as a ?rm grows it develops a
structural position (i.e. ties within a network of
resource providers) that contributes to its success.
This structural position changes the demands of
the environment from a demand for short-run
economic e?ciency to a demand for long-run
economic and social e?ciency/legitimacy (see
Hannan, 1998; Oliver, 1991). This prediction is
consistent with the positive accounting theory
literature that suggests that political costs are
greater for larger ?rms (Watts & Zimmerman,
1986). The number of analysts following a ?rm
tends to be correlated with ?rm size (e.g. see
Table 1), thus the importance of social disclosure
and the independent information available from
analysts will increase in parallel. We, therefore,
predict no interaction e?ect between social dis-
closure and the number of analysts following the
?rm.
H3a(ii): The relation between social disclosure
levels and the cost of equity capital is not
mediated by the level of analyst following.
2. Empirical measures of disclosure levels and
cost of equity capital
2.1. Disclosure proxy
Our empirical measures of ?nancial and social
performance are drawn from the joint Society of
Management Accountants of Canada (SMAC)/
University of Quebec at Montreal (UQAM)
sponsored assessments of the annual reports of
a broad cross-section of Canadian companies.
These assessments were conducted and publicly
reported based on 1990, 1991 and 1992 annual
reports, thus providing a limited time-series of
disclosure scores. To our knowledge, these data
are unique in North America because the
scores contain a ranking of ?rms on both the
quality and level of ?nancial disclosure and
social disclosure contained in annual reports.
This provides a signi?cant advantage over
other North American rankings such as the
Association for Investment Management and
Research (AIMR) rankings of US companies.
4
However, the AIMR rankings are prepared
by professional analysts and represent the
rankings of several analysts, while the Canadian
4
The AIMR rankings are compiled annually by the Asso-
ciation for Investment Management and Research. These dis-
closure rankings have been used in empirical studies by
Botosan and Plumlee (2000), Healy, Hutton, and Palepu
(1999), Lang and Lundholm (1993, 1996), and Welker (1995).
600 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
rankings available to us are based on the judg-
ments of less experienced raters and do not re?ect
the same level of averaging across raters as the
AIMR ratings. Nevertheless, the SMAC/UQAM
ratings are the best available source of disclosure
ratings for a broad cross-section of Canadian
?rms. The only alternative measure of disclosure
for Canadian companies would be researcher-
generated measures, as utilized by Botosan (1997).
We choose not to generate our own disclosure
ratings because of the potential for researcher
biases to in?uence the ratings and to avoid the
severe limitations on sample size imposed by this
approach.
For each year from 1990 to 1992, researchers at
UQAM analyzed the annual reports of around
700 Canadian companies coming from nine
industry sectors. The industry sectors reported on
in their publication include: Manufacturing-
Industrial Products, Manufacturing-Consumer
Products; Oil, Gas and Chemicals; Mines, Metals
and Forestry Products; Technology and Com-
munications; Financial Institutions; Retail and
Wholesale Trade; Management and Other; and
Utilities. An extensive checklist of information
related to socially responsible activities was
developed that contained 170 subcategories of
information. Similarly, an extensive checklist
related to ?nancial information was developed
which allocated points across 261 individual
disclosure elements.
Appendix A contains a summary of the 10
categories of social information considered and
the maximum number of points allocated to each
category, as well as the sub-categories of infor-
mation considered within each category. The
Appendix also contains similar information for
the ?nancial disclosure checklist.
Two points should be noted about these check-
lists. First, the social information captured in the
checklist includes a much broader set of dis-
closures than just environmental disclosures that
have been the subject of much of the past social
disclosure literature.
Second, the checklist used to assess ?nancial dis-
closure is similar in many respects to the checklists
utilized by the AIMR and developed by Botosan
(1997). For example, all of the checklists contain
sections devoted to general background informa-
tion that help users to interpret the ?nancial
statements. All three lists contain sections devoted
to assessing the usefulness of the disclosure of
summarized historical results, and all three check-
lists assess the inclusion of forecasted information
within the annual report. In addition to assessing the
types of disclosure made in the annual report, the
UQAM researchers also attempted to make an
assessment of the quality of the disclosure by
awarding more points for disclosures that contained
quantitative data or reported more information.
The extensive nature of the checklists utilized in the
SMAC/UQAM disclosure ratings, combined with
the attempt to discriminate between more and less
informative disclosures, gives us con?dence in the
face validity of these ratings. While they are
undoubtedly noisy measures, they provide us
with some ability to discriminate between ?rms
providing high levels of disclosure and those
providing minimal disclosure. Empirical analysis
reported later in the manuscript also provides evi-
dence corroborating the validity of our disclosure
measures.
3. Empirical measure of cost of equity capital
We follow Botosan (1997), Botosan and Plumlee
(2000) and Gebhardt, Lee, and Swaminathan
(2000), and obtain estimates of the cost of
equity capital using an accounting based valu-
ation model developed in Edwards and Bell
(1961), Feltham and Ohlson (1995) and Ohlson
(1995). Our empirical implementation of the
model is very similar to the method employed
by Gebhardt et al. and interested readers are
referred to their paper for further details on the
estimation procedure and for extensive empirical
analyses of the properties of the estimates. We
provide a brief sketch of the estimation procedure
later.
The valuation model speci?es a relation between
equity values and current book values and future
abnormal earnings. We brie?y sketch a derivation
of the empirical equation we use to estimate equity
cost of capital beginning with the familiar divi-
dend discount model:
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 601
P
it
¼
X
1
t¼1
E
t
d
itþt
È É
1 þ r
i
ð Þ
t
ð1Þ
where d=dividends, r=cost of equity capital, and
i and t are ?rm and time identi?ers, respectively.
Feltham and Ohlson (1995) and Ohlson (1995)
demonstrate that, for clean-surplus accounting
systems, this dividend discount model is algebrai-
cally equivalent to a valuation formula based on
current book value and future abnormal earnings:
5
P
it
¼ BV
it
þ
X
1
t¼1
E
t
x
itþt
À r
i
BV
itþtÀ1
È É À Á
1 þ r
i
ð Þ
t
ð2Þ
where BV=book value of equity, x=earnings,
and all other variables are as previously de?ned.
The ?nite horizon version of Eq. (2) is:
P
it
¼ BV
it
þ
X
T
t¼1
E
t
fFROE
itþt
À r
i
gBV
itþtÀ1
ð Þ
1 þ r
i
ð Þ
t
þTV
it
ð3Þ
where FROE=forecasted return on equity, and
TV=terminal value, or the present value at time t
of the abnormal earnings expected to be earned
after time t+T, and all other variables are as
previously de?ned.
For our estimates that utilize I/B/E/S earnings
forecasts, Eq. (3) is replaced with a version that
utilizes the earnings forecasts for the next three
?scal years from I/B/E/S.
6
The terminal value
involves forecasting return on equity for 12 future
years (i.e. 9 years beyond the latest available
explicit earnings forecast from I/B/E/S). Fore-
casted return-on-equity beyond year 3 is generated
using a linear fade-rate to the industry average
return-on-equity (ROE). Our industry average
ROEs are compiled from data provided by Statis-
tics Canada (Statscan). We acquire industry aver-
age ROE data beginning with 1980 for all
industries as de?ned and tracked by StatsCan. Our
industry average ROE ?gures for year t are com-
puted as the average industry ROE through time
starting in 1980 and ending in year tÀ1. For
example, for our 1990 industry average ROE
measure, we average the industry ROEs over
the 10 year period from 1980 to 1989, and
the 1991 industry ROEs are the average of
the industry ROEs over the 11 year period
from 1980 to 1990. These industry averages have
a mean of 10.75% and range from 4.3%, (the
1992 average for the iron, steel and related
products industry) to 22.7%, (the 1990 average
for the building materials and construction
industry).
Accordingly, the model we utilize to estimate
cost of capital has the following form:
5
A clean surplus accounting system is one in which book
value at time t is equal to book value at time tÀ1 plus earnings
minus dividends net of capital contributions. ‘‘Dirty surplus’’
arises when gains and losses a?ecting book value bypass the
income statement.
P
it
¼ BV
it
þ
X
3
t¼1
E
t
fFROE
itþt
À r
i
gBV
itþtÀ1
ð Þ
1 þ r
i
ð Þ
t
þ
P
11
t¼4
EfFROE
itþ3
þ t À 3 ð Þ½fIROE
it
ÀFROE
itþ3
g,9? À r
i
gBV
itþtÀ1
1 þ r
i
ð Þ
t
þ
IROE
it
À r
i
r
i
1 þ r
i
ð Þ
11
BV
itþ11
ð4Þ
6
About 10% of our ?rm year observations have two-year
ahead EPS forecasts but do not have a 3-year ahead forecast.
When the third year ahead forecast is missing, we forecast ?scal
year 3 (FY3) EPS by assuming that the earnings growth rate
implicit in the ?scal year 2 (FY2) compared with ?scal year 1
(FY1) forecasts applies to ?scal year 3 as well. Speci?cally, we
forecast FY3 EPS as FY2 EPSÂ(FY2 EPS/FY1 EPS). If FY2
EPS/FY1 EPS-0, we do not forecast FY3 EPS and omit the
observation.
602 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
where IROE is the historical industry average
ROE and all other variables are as previously
de?ned.
The exact details of the estimation procedure,
including variable measurement, are described in
the following section. The sensitivity of our results
to variations in the assumptions required to esti-
mate cost of capital are discussed in the section of
the paper entitled ‘‘sensitivity of results to cost of
capital estimation assumptions’’.
4. Empirical analysis
4.1. Disclosure proxy and additional data
Our initial sample consists of all ?rms that
received a disclosure rating based on any or all of
its 1990, 1991 or 1992 annual reports. In each of
these three years, the annual reports of over 700
Canadian companies were collected and analyzed
by a group of research assistants at UQAM. The
results of the ratings are summarized by industry
group and published each year by CMA Canada.
The survey only provides 3 years of data. Since
disclosure policies are probably relatively stable
?rm attributes, this limited time-series should not
severely impact the generality of our results. In
order to perform the initial empirical analysis, we
require ?nancial statement data provided by
Compustat, earnings forecasts and analyst follow-
ing provided by I/B/E/S, and market price and
returns data acquired through Datastream. These
additional data requirements leave us with a
sample of 324 ?rm year observations from 124
di?erent companies with all necessary price data,
?nancial and social disclosure scores, all necessary
Compustat data and at least 1-year ahead EPS
forecast available from I/B/E/S. For this initial
sample, we only require that 1-year ahead ana-
lysts’ forecasts be available, and therefore the
number of analysts making a 1-year ahead fore-
cast, be available from I/B/E/S. In order to
calculate cost of capital estimates, we require
that at least 2-year-ahead earnings forecasts be
obtained from I/B/E/S. We also require that
unambiguous identi?cation with a StatsCan
industry group be possible, further reducing our
sample size to 225 ?rm year observations from 87
di?erent ?rms.
For these observations, we obtain the average
stock price from the 6th month after year-end
from Datastream. We choose a period 6 months
after year-end to ensure that all information con-
tained in the annual ?nancial statements has been
disclosed and is re?ected in market prices.
7
This
choice is also consistent with the empirical analysis
in Botosan (1997). From Compustat, we obtain
the debt-to-equity ratio, return-on-equity, divi-
dend payout ratio, market to book value ratio,
and market value of equity. I/B/E/S provides the
earnings forecasts and the number of analysts fol-
lowing the ?rm, which we measure as the number
of analysts providing 1-year-ahead earnings fore-
cast for the ?rm.
Calculation of forecasted book values requires
an estimated future dividend payout ratio (k). We
estimate this ratio using the following procedure.
First, we obtain the historical 10-year average
payout ratio for each year t from Compustat and
use this to proxy for future payouts if it is avail-
able and positive. If the 10-year average is either
unavailable or negative, we use the 5-year histor-
ical average payout ratio. If the 5-year payout
ratio is either unavailable or negative, we estimate
future payout ratios based on the dividend payout
ratio for year t. Finally, if the year t payout ratio is
negative, we follow a procedure similar to Geb-
hardt et al. (2000) and approximate ‘‘permanent’’
earnings for year t as (BV
tÀ1
ÂIROE), the book
value of equity at the beginning of the year times
the industry average ROE. We then calculate the
dividend payout ratio as DVS
t
/(BV
tÀ1
ÂIROE),
where DVS is the dividends per common share for
year t. Our implementation of Eq. (4) then uses
the ?rm’s estimated dividend payout ratio (k) to
update book values as follows:
BV
it
¼ BV
itÀ1
þ FROE
it
1 À k ð Þ
it
È É
BV
itÀ1
ð4aÞ
7
Our results are qualitatively similar using average stock
prices from the 4th month after ?scal year-end. We use the
average stock price for the month rather than monthly closing
prices to avoid extreme observations that may result from
temporary share price ?uctuations re?ected in a single price
observation such as a closing price.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 603
As Gebhardt et al. point out; the calculation of
an implied cost of capital is the same as determin-
ing the internal rate of return that equates the
stock price to the expected future bene?ts of share
ownership. Gebhardt et al. also provide two
example calculations to which we refer interested
readers. We perform this estimation using Micro-
soft Excel, and our estimate of the cost of capital
is simply the discount rate that equates the book
value and future forecasted earnings to the current
market price.
Tables 1 and 2 provides distributional char-
acteristics and a correlation matrix for the vari-
ables used in the study. Financial disclosure scores
average 32.5 (out of a possible 120), and range
between 4.75 and 63. Social disclosure scores tend
to be lower, averaging 11 (out of 100) and ranging
between 0.25 and 50. The average ?rm size in the
sample is $3.8 billion and the median size is sub-
stantially lower at around $1.6 billion. Nine ana-
lysts follow the average ?rm, and this ranges
between 1 and 37. The average debt to equity ratio
is 75%. The mean ROE for our sample ?rms is
close to zero, re?ecting the poor economic climate
in Canada in the early 1990s.
8
The median ROE is
around 6%. Around one third of our sample ?rms
come from the oil, gas and chemical industry or
the mines, metals and forestry products industry,
industries that have been depicted as environmen-
tally/socially sensitive industries in the past litera-
ture (Deegan & Gordon, 1996; RWH, 1999).
Finally, our cost of capital estimates average close
to 9%, and range between 1.8 and 22.5%.
These estimates of cost of capital appear rea-
sonable in relation to Canadian T-bill rates during
our sample period. Our rates, measured with
8
During the 1990–1992 periods, quarterly growth in gross
domestic product in Canada was 0.3%, far below the average
of 1.1% experienced during the remainder of the 1990s (Cal-
culated based on Datastream data).
Table 2
Correlation matrix (Pearson above diagonal, Spearman below)
a
FDISC SDISC MV NANAL LEV ROE IND COST
FDISC 1 0.662* 0.499* 0.542* 0.067 0.009 0.229* À0.046
SDISC 0.704* 1 0.294* 0.391* 0.080 À0.023 0.298* 0.011
MV 0.555* 0.474* 1 0.413* À0.054 0.132* À0.143* 0.091
NANAL 0.603* 0.509* 0.617* 1 À0.023 0.029 0.321* À0.208*
LEV 0.177* 0.243* 0.024 0.042 1 À0.542* 0.011 0.054
ROE À0.067 À0.096 0.136* 0.009 À0.251* 1 À0.121* 0.074
IND 0.281* 0.278* 0.065 0.346* À0.013 À0.255* 1 À0.362*
COST À0.053 À0.012 0.066 À0.156* 0.131* 0.102 À0.351* 1
a
Variable de?nitions: FDISC=Financial disclosure score for ?rm i, year t; SDISC=social disclosure score for ?rm i, year t;
MV=beginning of year market value of common equity for ?rm i, year t; NANAL=number of analysts making a 1 year ahead EPS
forecast for ?rm i, year t; LEV=debt to equity ratio for ?rm i, year t; ROE=return on equity for ?rm i, year t; IND=a dummy variable
equal to one if ?rm i is a member of the Oil, Gas and Chemicals or Mine, Metals and Forestry Products industry sectors in year t, zero
otherwise; COST= estimated cost of equity capital for ?rm i, year t.
*=signi?cant at the 5% level, two-tailed test.
Table 1
Descriptive statistics
a
Variable name N Mean Median Minimum Maximum
FDISC 324 32.53 31.48 4.75 63.0
SDISC 324 11.04 8.25 0.25 50.0
MV ($000,000) 324 3833 1605 20.2 38,729
ANALFOL 324 9.32 9.0 1 37
LEV (%) 324 75.8 53.5 0 1095
ROE (%) 324 0.19 5.98 À194.0 33.1
IND 324 0.38 0 0 1
COST (%) 225 8.9 8.5 1.8 22.5
a
Variable de?nitions: FDISC=Financial disclosure score
for ?rm i, year t; SDISC=social disclosure score for ?rm i, year
t; MV=beginning of year market value of common equity for
?rm i, year t; NANAL=number of analysts making a 1 year
ahead EPS forecast for ?rm i, year t; LEV=debt to equity ratio
for ?rm i, year t; ROE=return on equity for ?rm i, year t;
IND=a dummy variable equal to one if ?rm i is a member of
the Oil, Gas and Chemicals or Mine, Metals and Forestry
Products industry sectors in year t, zero otherwise; COST=
estimated cost of equity capital for ?rm i, year t.
604 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
forecasted data as of June 1991 (for 1990 annual
reports), June 1992 and June 1993 average 9.85,
9.5 and 8.0%, respectively. The 1-year T-bill rate
in Canada shows similar declines throughout our
sample period. For example, the one-year T-bill
rate in June 1991, 1992 and 1993 was 9.11, 6.63
and 5.79%, respectively. In addition, the risk-
premium inherent in our cost of capital estimates
varies between approximately 0.7 and 2.5%, with
a median of 2.2%. Gebhardt et al. ?nd that
risk premiums implied for their sample (including
over 10,000 observations from 1979 to 1995
and the same method for estimating the cost
of equity capital) vary between À0.7 and 4.9%
with a median of 2.0%. Our cost of capital esti-
mates display a similar relationship to the risk-free
rate.
The univariate correlations reveal no unex-
pected patterns, and show that there is a strong
positive relation between social and ?nancial dis-
closure, as might be expected if both types of dis-
closure are part of an overall disclosure policy.
Cost of capital does not have a signi?cant relation
to either disclosure score. The correlation between
the disclosure variables and other potential deter-
minants of cost of capital (e.g. market value of
equity, analyst following and industry) makes
interpretation of these univariate correlations
problematic. However, these univariate correla-
tions do indicate that our main results are sensitive
to the inclusion of control variables.
Table 3 provides selected descriptive informa-
tion for ?rms in the sensitive, (oil, gas and chemi-
cal or mines, metals and forestry products), versus
non-sensitive industries, and across time for both
groups. Several empirical regularities are revealed
in this descriptive information. First, ?rms in the
sensitive industries have better ?nancial and social
disclosure scores, are followed by more analysts,
have inferior ?nancial performance (except 1992),
and have lower cost of capital estimates than do
their counterparts in the non-sensitive industries.
The two groups appear roughly similar in terms of
?rm size, though the mean ?rm size is typically
larger for the non-sensitive ?rms and the median
?rm size is larger for the sensitive ?rms. The ?rm
size distribution is much more skewed for the
non-sensitive sample. Additionally, cost of capital
estimates, ?nancial performance and the number
of analysts following the ?rm tend to decline
over time for both groups, though the ?nancial
performance for the sensitive ?rms rebounded in
1992.
4.2. Validation of the disclosure proxies
Since the SMAC/UQAM ratings of disclosure
have not been previously used in academic
research, we perform a series of tests that examine
the relationship between these disclosure ratings
and several variables that are related to disclosure
levels in the past literature. Speci?cally, we follow
Botosan (1997) and Lang and Lundhom (1993)
and examine the relation between ?nancial dis-
closure and ?rm size, ?nancial performance,
leverage and analyst following. Past results sug-
gest that a positive relationship should exist
between each of these variables and ?nancial dis-
closure. We also examine the relationship between
social disclosure ratings and ?rm size, industry
membership, ?nancial performance, leverage,
analyst following and the interaction of industry
and ?rm size. Industry membership is coded to
re?ect membership in a socially/environmentally
sensitive industry, namely the oil, gas and chemi-
cals industry or the mines, metals and forestry
products industry. Since larger ?rms and ?rms in
these industries have their social and environ-
mental performance closely scrutinized, past
research suggests a positive relation between
industry membership and the interaction of ?rm
size and industry membership with social dis-
closure (e.g. Deegan & Gordon, 1996).
The relationship between social disclosure and
?nancial performance has been mixed in the past
literature (Pava & Kraus, 1996) while the rela-
tionship between leverage and analyst following
and social disclosures have not been examined in
the past literature. Since social disclosure may
accomplish many of the same objectives as ?nancial
disclosure, we expect to observe positive relation-
ships between these variables and social disclosure.
These tests provide evidence on the validity of our
disclosure ratings as measures of the cross-sec-
tional variation in the level of disclosure provided
in Canadian companies’ annual reports.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 605
Table 3
Descriptive information — across sensitive and non-sensitive industries and time
Variable name N Mean Median Minimum Maximum
Sensitive industry—1990
FDISC 37 36.14 34.75 15.5 56
SDISC 37 13.82 11.25 1.25 45.5
MV ($000,000) 37 2478 1982 69.3 12,543
NANAL 37 13.49 12 3 34
ROE (%) 37 À1.58 6.00 À142.07 26.46
COST (%) 29 8.43 8.12 1.83 15.26
Sensitive industry—1991
FDISC 42 37.15 35.25 16.75 58
SDISC 42 15.31 14.38 2 40.5
MV ($000,000) 42 2840 1994 85.3 11,821
NANAL 42 12.12 10.5 2 36
ROE (%) 42 À8.67 1.95 À158.9 17.54
COST (%) 31 7.56 7.47 3.52 14.15
Sensitive industry—1992
FDISC 43 34.7 34.35 14.25 52.5
SDISC 43 13.95 13.5 2 50
MV ($000,000) 43 2511 2002 88.4 10,554
NANAL 43 10.37 10.0 1 30
ROE (%) 43 À0.81 2.66 À38.38 22.84
COST (%) 34 6.72 6.23 3.89 13.73
Non-sensitive industry—1990
FDISC 63 31.9 29 11.25 63
SDISC 63 10.2 7 0.75 38
MV ($000,000) 63 4675 1364 20.2 32,179
NANAL 63 8.76 8 1 37
ROE (%) 63 5.09 9.98 À97.06 23.68
COST (%) 40 10.84 10.53 6.55 20.34
Non-sensitive industry—1991
FDISC 71 29.7 46.75 8.7 59.5
SDISC 71 8.62 7 0.25 33.5
MV ($000,000) 71 4447 1245 27.5 37,839
NANAL 71 7.82 7 1 34
ROE (%) 71 4.32 7.63 À77.71 33.09
COST (%) 47 10.06 9.62 4.04 19.74
Non-sensitive industry—1992
FDISC 68 29.81 29.28 4.75 63
SDISC 68 8.35 6.38 1 43.25
MV ($000,000) 68 4452 1393 66.1 38,729
NANAL 68 6.77 6.0 1 26
ROE (%) 68 À1.59 6.42 À194.0 31.15
COST (%) 44 8.98 9.1 3.11 22.48
Variable de?nitions: FDISC=?nancial disclosure score for ?rm i, year t; SDISC=social disclosure score for ?rm i, year t; MV=be-
ginning of year market value of common equity for ?rm i, year t; NANAL=number of analysts making a 1 year ahead EPS forecast
for ?rm i, year t; ROE=return on equity for ?rm i, year t; COST=estimated cost of equity capital for ?rm i, year t.
606 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
Additionally, we note that our ratings are based
only on the disclosures contained in annual
reports and do not necessarily re?ect the level of
disclosures made through other media (c.f. Zeghal
& Ahmed, 1990). While this is an acknowledged
limitation of our proxy for disclosure, Botosan
(1997) shares this limitation. We also note that
Lang and Lundholm (1993) and Welker (1995)
examine the AIMR ratings for US companies and
report a high degree of correlation between dis-
closure ratings based on the annual report and
ratings based on other disclosure media. This
?nding provides evidence of convergent validity of
annual report disclosure measures and disclosures
occurring in other media as well.
4.3. Relation of ?nancial disclosure scores with size,
?nancial performance, leverage and analyst following
Past research has demonstrated that ?nancial
disclosure increases with ?rm size, ?nancial per-
formance, leverage, and the number of analysts fol-
lowing the ?rm. One way to validate our empirical
proxy for ?nancial disclosure is to examine these
relationships based on our proxy. Accordingly, we
estimate the following empirical equation:
FDISC
it
¼ o þ[
1
DSIZE
it
þ[
2
ROE
it
þ[
3
LEV
it
þ[
4
DANAL
it
þc ð5Þ
where FDISC=?nancial disclosure score for ?rm
i, year t, DSIZE=a dummy variable equal to one
if the beginning of year market value of equity for
?rm i, year t is above the sample median, zero
otherwise, ROE=return on equity for ?rm i, year
t, LEV=debt to equity ratio, ?rm i, year t,
DANAL=a dummy variable equal to 1 if the
number of analysts providing a 1-year ahead
earnings forecast for ?rm i, year t is above the
sample median, zero otherwise.
The results of estimating Eq. (5) are reported in
Table 4.
9
Consistent with the past literature, each
of the variables in the equation except ROE exhi-
bits a signi?cant and positive relation with ?nan-
cial disclosure. The adjusted R
2
of the regression is
over 30%, indicating that the explanatory vari-
ables are able to explain a reasonable portion of
the cross-sectional variation in ?nancial disclosure
scores. The fact that our ?nancial disclosure scores
are strongly related to variables which the past
literature suggests explain ?nancial disclosure
increases our con?dence that variation in our dis-
closure measure captures the underlying phenom-
enon of interest, variation in ?nancial disclosure.
Table 4
Results of estimating Eq. (5) explaining variation in ?nancial disclosure
FDISC
it
¼ o þ[
1
DSIZE
it
þ[
2
ROE
it
þ[
3
LEV
it
þ[
4
DANAL
it
þc
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 34.113 26.323 0.0001
DSIZE (+) 6.895 5.481 0.0001
ROE (+) 0.032 1.277 0.2026
LEV (+) 0.014 2.535 0.0117
DANAL (+) 8.505 6.776 0.0001
No. of observations=324; adjusted R
2=
32.7%. Variable de?nitions: FDISC=?nancial disclosure score for ?rm i, year t; DSIZE=a
dummy variable equal to one if the beginning of year market value of equity for ?rm i, year t is above the sample median, zero
otherwise; ROE=return on equity for ?rm i, year t; LEV=debt to equity ratio, ?rm i, year t; DANAL=a dummy variable equal to 1
if the number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t is above the sample median, zero otherwise.
9
Our tabulated results provide two-tailed P-values. By
convention, we describe an estimated coe?cient as statistically
signi?cant if the coe?cient is signi?cant at the 0.05 level in a
one-tailed test if we predict the sign of the coe?cient and a two-
tailed test if we do not predict the sign of the coe?cient.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 607
4.4. The relationship of social disclosure scores
with ?rm size, industry membership, ?nancial
performance, leverage and analyst following
Patten (1991) documents that social disclosure is
increasing in ?rm size and is greater in highly
visible and politically sensitive industries. Deegan
and Gordon (1996) ?nd an interaction e?ect
between ?rm size and industry, such that the size
e?ect is particularly pronounced in sensitive
industries. As Richardson et al. (1999) discuss,
results of tests of the relation between social per-
formance/disclosure and ?nancial performance
have been mixed. Leverage and number of ana-
lysts following the ?rm have not been included in
past studies of social disclosure. We include these
variables in our regression since they are related to
?nancial disclosure. Social disclosure accomplishes
similar objectives and may have similar deter-
minants. We examine the relation between the
SMAC/UQAM social disclosure ratings and these
variables by estimating the following equation:
SDISC
it
¼ , þ’
1
DSIZE
it
þ’
2
ROE
it
þ’
3
IND
it
þ’
4
IND
it
ÂSIZE
it
ð Þ þ’
5
DANAL
it
þ
’
6
LEV
it
þu
it
ð6Þ
where SDISC=social disclosure score for ?rm i,
year t, IND=a dummy variable equal to one if
?rm i is a member of the Oil, Gas and Chemicals
or Mine, Metals and Forestry Products industry
sectors in year t, zero otherwise
All other variables are as previously de?ned.
The results of estimating Eq. (6) are presented in
Table 5. Again, the results are encouraging as our
measure of social disclosure appears to be related to
variables that the past literature suggests it should
be. Size, industry, and the interaction of industry and
size are all signi?cantly related to social disclosure,
as suggested by the past literature. Analyst following
is also statistically positively related to social dis-
closure, consistent with our conjectures that similar
factors may in?uence ?nancial and social dis-
closure. ROE is not statistically related to social
disclosure, in keeping with the insigni?cant or mixed
relation documented in the past literature. Lever-
age is also positively related to social disclosure,
again consistent with our conjectures that social and
?nancial disclosures have similar determinants.
We conclude, based on the evidence discussed
earlier, that the ?nancial and social disclosure
scores are valid measures of those disclosures by
the ?rms in our sample. We now turn to the sub-
stantive issue of the relationship between dis-
closure and the cost of capital.
Empirical tests of the relation between disclosure
and the cost of equity capital
Our tests of the relations between ?nancial and
social disclosure and the cost of equity capital are
Table 5
Results of estimating Eq. (6) explaining variation in social disclosure
SDISC
it
¼ , þ’
1
DSIZE
it
þ’
2
ROE
it
þ’
3
IND
it
þ’
4
IND
it
ÂSIZE
it
ð Þ þ’
5
DANAL
it
þ’
6
LEV
IT
þu
it
ð6Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 5.097 6.315 0.0001
DSIZE (+) 3.507 3.021 0.0027
ROE (?) 0.021 1.092 0.2757
IND (+) 2.465 1.957 0.0513
INDÂSIZE (+) 2.958 1.713 0.0876
DANAL (+) 3.931 3.918 0.0001
LEV (+) 0.011 2.596 0.0099
No. of observations=324; adjusted R
2
=26.8%. Variable de?nitions: SDISC=social disclosure score for ?rm i, year t; DSIZE=a
dummy variable equal to one if the beginning of year market value of equity for ?rm i, year t is above the sample median, zero
otherwise; ROE=return on equity for ?rm i, year t; LEV=debt to equity ratio, ?rm i, year t; DANAL=a dummy variable equal to 1
if the number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t is above the sample median, zero otherwise;
IND=a dummy variable equal to one if ?rm i is a member of the Oil, Gas and Chemicals or Mine, Metals and Forestry Products
industry sectors in year t, zero otherwise.
608 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
based on the work of Botosan (1997) and Geb-
hardt et al. (2000). Similar to Botosan, we attempt
to document a negative relation between ?nancial
and social disclosure and the cost of equity capital
that is incremental to the e?ects of other variables
known to in?uence cost of capital. Gebhardt et al.
?nd that the risk premia is negatively related to
the number of analysts following the ?rm and
positively related to leverage. Accordingly, our
empirical tests include these variables as control
variables.
10
In addition, Botosan documents that
disclosure and analyst following has an interactive
e?ect on cost of capital. She ?nds that disclosure
reduces cost of capital only for those ?rms with
low analyst following. We also estimate an equa-
tion that allows for this potential interactive e?ect
and an interaction between analyst following and
social disclosure as well. Speci?cally, our primary
empirical tests come from the estimation of the
following two equations:
COST
it
¼, þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þj
it
ð7Þ
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þ[
5
LANAL
it
ÂFDISC
it
ð Þþ
[
6
LANAL
it
ÂSDISC
it
ð Þ þj
it
ð8Þ
where COST is the cost of equity capital estimated
by Eq. (4); NANAL is the number of analysts
making 1-year ahead earnings forecast; LANAL is
a dummy variable set equal to one if the number
of analysts following ?rm i in year t is below i’s
industry sector median for year t, zero otherwise.
All other variables are as previously de?ned and all
variables are adjusted for industry sector medians.
Gebhardt et al. (2000) document considerable
variation in their risk premia estimates across
industries. Our results (documented in Tables 3
and 5) and the results of previous research
demonstrate that disclosure practices also vary
across industries as well. Accordingly, all variables
utilized in estimating Eqs. (7) and (8) are adjusted
for industry sector (as de?ned in the SMAC/
UQAM disclosure ratings) year medians. This
removes both industry and time e?ects from the
data, avoiding potentially spurious associations.
Since all variables in this speci?cation are industry
adjusted, ?rm size and analyst following directly
enter the regressions without the conversion to
dummy variables performed in earlier tests. We
exclude observations in the top and bottom 1% of
the empirical distribution of cost of capital (i.e.
two observations from each tail of the distribu-
tion) to ensure our results are not unduly a?ected
by extreme observations.
11
In accordance with our hypotheses outlined
earlier, we expect C
1
, C
3
, C
4
and C
5
to be nega-
tive and C
2
to be positive. As discussed in H3a(ii),
we expect C
6
to be statistically insigni?cant.
For comparative purposes, the results of esti-
mating Eqs. (5) and (6), explaining ?nancial and
social disclosure, respectively, using the industry-
adjusted data, are reported in Tables 6 and 7.
Of course, the industry membership variables
10
Other potential risk proxies such as market beta, market
value of equity and book to market ratios could also be con-
trolled for in the empirical analysis. We omit beta from the
analysis because Gebhardt et al. document that they are statis-
tically unrelated to our measure of the cost of equity capital
except in certain multivariate tests. We include size and book to
market ratios in all our equations explaining the cost of capital
and ?nd that these variables have the correct sign but are sta-
tistically unrelated to our industry adjusted cost of capital esti-
mates. Gebhardt et al. also ?nd that the dispersion in analysts’
earnings forecasts is signi?cantly related to the cost of capital.
We choose not to include this variable in our analysis for three
reasons. First, the dispersion in analysts’ forecasts would pre-
sumably be a function of the disclosure variables we include in
our analysis, so the inclusion of this variable would amount to
the inclusion of an alternative disclosure measure and could
hamper our ability to observe a relation between more direct
disclosure measures and the cost of capital. Second, including a
measure of analysts’ forecast dispersion would reduce our
sample size because a measure of dispersion requires that the
?rm be followed by multiple analysts Third, the dispersion in
analysts’ forecasts is related to the number of analysts follow-
ing the ?rm, which we include for consistency with Botosan.
11
Our conclusions are not a?ected by excluding these obser-
vations. In general, there is a slight increase in statistical sig-
ni?cance when the observations are dropped, suggesting these
observations contain measurement error.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 609
Table 6
Results of estimating Eq. (5) explaining variation in ?nancial disclosure—industry adjusted data
FDISC
it
¼ o þ[
1
SIZE
it
þ[
2
ROE
it
þ[
3
LEV
it
þ[
4
NANAL
it
þc ð5Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) À0.276 À0.524 0.6005
SIZE (+) 0.001 5.467 0.0001
ROE (+) 0.015 0.605 0.5458
LEV (+) 0.007 1.472 0.1420
NANAL (+) 0.508 5.291 0.0001
No. of observations=324; adjusted R
2
=23.8%. Variable de?nitions: FDISC=?nancial disclosure score for ?rm i, year t—the industry
sector median for year t; SIZE=beginning of year market value of equity for ?rm i, year t—the industry sector median for year t;
ROE=return on equity for ?rm i, year t—the industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry
sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the industry
sector median for year t.
Table 7
Results of estimating Eq. (6) explaining variation in social disclosure—industry adjusted data
SDISC
it
¼ , þ’
1
SIZE
it
þ’
2
ROE
it
þ’
5
NANAL
it
þ’
6
LEV
it
þu
it
ð6Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 1.148 2.682 0.0077
SIZE (+) 0.002 3.472 0.0006
ROE (?) 0.033 1.687 0.0936
NANAL (+) 0.172 2.196 0.0461
LEV (+) 0.009 2.003 0.0288
No. of observations=324; adjusted R
2
=8.4%. Variable de?nitions: SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t; SIZE=beginning of year market value of equity for ?rm i, year t—the industry sector median for year t;
ROE=return on equity for ?rm i, year t—the industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry
sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the industry
sector median for year t.
Table 8
Results of estimating Eq. (7) explaining variation in cost of capital estimates—industry adjusted data
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þj
it
ð7Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 0.00394 2.244 0.0258
NANAL (À) À0.00101 À3.459 0.0007
LEV (+) 0.00002 1.584 0.1147
FDISC (À) À0.00040 À1.961 0.0512
SDISC (À) 0.00064 2.354 0.0195
No. of observations=221; adjusted R
2
=9.9%. Variable de?nitions: COST=estimated cost of equity capital for ?rm i, year t—the
industry sector median for year t; NANAL=number of analysts providing a one-year ahead earnings forecast for ?rm i, year t—the
industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry sector median for year t; FDISC=?nancial
disclosure score for ?rm i, year t—the industry sector median for year t; SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t.
610 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
originally included in Eq. (6) explaining social
disclosure are omitted from this speci?cation
which already includes industry adjustment. As
the results in Table 6 show, the results of estimat-
ing Eq. (5), which explains ?nancial disclosure
variation, using the industry adjusted data, are
very similar to our earlier results. The only di?er-
ence is that the signi?cance of leverage is dimin-
ished such that it is no longer a signi?cant
explanatory variable at conventional levels. As
Table 7 reveals, the R
2
of the regression explaining
social disclosure falls dramatically (from 27 to
8%) when industry adjusted data are used and
industry related variables are excluded from the
regression. However, the explanatory power of the
remaining variables is very similar to that reported
in our earlier results.
The results of estimating Eqs. (7) and (8) are
reported in Tables 8 and 9, respectively.
12
The
number of analysts following the ?rm has a sta-
tistically reliable e?ect on cost of capital, with
higher analyst coverage resulting in a lower cost of
capital. Financial leverage is positively associated
with the cost of equity capital, but this e?ect is not
signi?cant at conventional levels. Financial dis-
closure is negatively related to cost of capital,
consistent with our predictions. This result is
stronger than the full sample result in Botosan,
perhaps suggesting that ?nancial disclosure plays
a more important role for Canadian ?rms. Sur-
prisingly, social disclosure exhibits a statistically
reliable positive association with the cost of equity
capital. For our sample ?rms and our time period,
enhanced social disclosure results in a higher cost
of capital.
Table 9 reports the results of estimating the
expanded Eq. (8). This speci?cation includes
interaction terms intended to determine if analyst
following modi?es the relation between our dis-
closure variables and the cost of capital. Con-
sistent with Botosan’s (1997) results, we ?nd that
?rms with low analyst following receive bene?ts
from expanded ?nancial disclosure in the form of
a reduction in the cost of equity capital. The
interaction of analyst following and social dis-
closure is not statistically signi?cant.
Table 9
Results of estimating Eq. (8) explaining variation in cost of capital estimates—industry adjusted data
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þ
[
5
LANAL
it
ÂFDISC
it
ð Þ þ[
6
LANAL
it
ÂSDISC
it
ð Þ þj
it
ð8Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 0.00094 0.474 0.6362
NANAL (À) À0.00095 À3.258 0.0013
LEV (+) 0.00002 1.478 0.1408
FDISC (À) À0.00004 À0.169 0.8663
SDISC (À) 0.00078 2.517 0.0126
LANAL*FDISC (À) À0.00098 À2.289 0.0231
LANAL*SDISC (?) À0.00036 À0.567 0.5715
No. of observations=221; adjusted R
2
=12.98%. Variable de?nitions: COST=estimated cost of equity capital for ?rm i, year t—the
industry sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the
industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry sector median for year t; FDISC=?nancial
disclosure score for ?rm i, year t—the industry sector median for year t; SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t; LANAL=a dummy variable set equal to one if the number of analysts following ?rm i in year t is below i’s
industry sector median for year t, zero otherwise.
12
Recall that the sample sizes re?ected in Tables 8, 9, and 10
are 221, not the 324 re?ected in earlier tables. This re?ects the
additional data requirements to estimate the cost of capital
(analysts’ forecasts and industry ROE) and the deletion of
observations with cost of capital below the 1st percentile or
above the 99th percentile of the empirical distribution.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 611
4.5. Speci?cation checks and further analysis
White’s test for heteroskedasticity/model speci-
?cation fails to reject the null hypothesis of
homoskedasticity for all regressions. In addition,
the highest condition index reported in any of our
regressions is 3.6, suggesting that multicollinearity
is not a?ecting our results.
One potential problem with our data is that
each ?rm contributes up to three observations to
the estimation, and these observations may not be
independent. Accordingly, we estimate Eqs. (7)
and (8) on a year-by-year basis rather than on the
pooled sample. The results (not reported) are very
similar to those reported for the full sample. All
variables that are signi?cant in our pooled regres-
sions have consistent signs for each of the 3 years,
and are signi?cant in at least one, often two, of the
yearly regressions. This suggests that our results
are not due to our pooling of sample data.
We explore further our ?nding that social dis-
closure increases the cost of capital. We conjecture
that this result may be related to the poor eco-
nomic conditions that characterize our sample
period. While we do not have data available from
another, more prosperous time period, we conduct
an alternative test to assess this explanation for
our results. If the relationship between social dis-
closure and the cost of capital is mediated by eco-
nomic conditions, then we expect that, within our
sample, ?rms with above average ?nancial perfor-
mance would not experience an increase in the
cost of capital as social disclosure increases, while
those ?rms with below average ?nancial perfor-
mance would. We test this conjecture by estimat-
ing Eq. (9), which replaces the insigni?cant social
disclosure/analyst following interaction from Eq.
(8) with a social disclosure/return on equity inter-
action. For completeness, we also include return
on equity in the equation to test for a direct
impact of ?nancial performance on the cost of
equity capital:
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ
[
3
FDISC
it
þ[
4
SDISC
it
þ
[
5
LANAL
it
ÂFDISC
it
ð Þþ
[
6
DROE
it
ÂSDISC
it
ð Þ þ[
7
ROE
it
þj
it
ð9Þ
where ROE is equal to the return on equity for
?rm i, year t—the industry sector median for year
t, DROE is a dummy variable set equal to one if
?rm i, year t return on equity is above the industry
sector median for year t, zero otherwise, and all
other variables are as previously de?ned.
In keeping with the above discussion, we expect
C
6
to be negative and make no predictions about
C
7
.
13
The results of estimating Eq. (9) are pro-
vided in Table 10. Consistent with our conjectures,
the coe?cient on the social disclosure/?nancial
performance interaction, is reliably negative. This
coe?cient is almost exactly equal in absolute
magnitude (À0.00109) to the coe?cient on social
disclosure (0.00116). This indicates that there is
essentially no relation between social disclosure
and the cost of capital for ?rms with above aver-
age return on equity, but a signi?cant increase
in the cost of capital accompanying better social
disclosure for below average return on equity
?rms.
14
4.6. Sensitivity of results to cost of capital
estimation assumptions
We perform several alternative calculations of
the cost of capital by varying the assumptions
underlying that calculation. The primary results
13
Since this analysis is conducted in light of our earlier
empirical results, we conduct two-tailed tests of statistical sig-
ni?cance for this estimation.
14
The distribution of ROE reveals the existence of some
extreme negative values. There are no negative book value
observations in our sample, so these observations are not
caused by small negative denonimators. Since our ROE and
social disclosure interaction is based on a dummy variable for
ROE, this interaction term should not be impacted by these
observations. However, it is possible that the estimated coe?-
cient for the main e?ect of ROE is impacted by these observa-
tions, and that this a?ects the estimated coe?cient on the
interaction term. We conducted two additional tests to ensure
that our results are not sensitive to extreme observations of
ROE. We repeated the estimation of Eq. (9) after (1) eliminat-
ing all observations with ROE less than À25% (the 5th percn-
tile of the empirical distribution), and (2) retaining all
observations but replacing the continuous version of ROE with
the dummy variable (DROE) to test for the main e?ect of
ROE. Neither change in speci?cation qualitatively alters our
results. In particular, C
6
remains signi?cantly negative and C
7
remains insigni?cantly di?erent from zero in both tests.
612 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
reported above are based on the assumption that
the ROE of each ?rm fades to its historical indus-
try average ROE in a linear fashion between
years +4 and +12. We check for the sensitivity
of our results to this assumption by performing
two new calculations, one in which the linear fade
occurs between periods +4 and +6, and one in
which the fade occurs between +4 and +18. Our
primary empirical results are not sensitive to these
changes in speci?cation. The statistical sig-
ni?cance of most explanatory variables, including
the disclosure variables of primary interest,
increase with the length of the linear fade period,
perhaps suggesting that the cost of capital esti-
mates contain less noise as the linear fade period is
increased.
5. Conclusions and suggestions for future research
This study provides further evidence on the the-
oretical and regulatory premise that improved
corporate disclosure results in a lower cost of
capital. We ?nd that there is a signi?cant negative
relationship between the level of ?nancial dis-
closure and the cost of capital (H1, see Table 8).
We also con?rm Botosan’s (1997) ?nding that
higher levels of ?nancial disclosure can reduce the
cost of capital in cases where there is low ?nancial
analyst following [H3a(i), see Table 9]. Our
results, however, suggest that this relation does
not hold for social disclosures. There is a statisti-
cally signi?cant, positive relation between the level
of social disclosure and the cost of capital, that is,
more social disclosure raises the cost of capital for
the ?rm (H2, see Table 8). The number of analysts
following the ?rm does not a?ect this result
[H3a(ii), see Table 9]. The positive relationship
between cost of capital and social disclosure is
moderated by the return-on-equity of the ?rm
with more successful ?rms being less penalized for
social disclosures.
It is important to recognize that these results are
not based on the content of the disclosures. The
disclosure scores re?ect the completeness and
informativeness of ?nancial and social disclosures
but they do not indicate whether the information
is good or bad news. There are several possible
Table 10
Results of estimating Eq. (9) explaining variation in cost of capital estimates—industry adjusted data
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þ
[
5
LANAL
it
ÂFDISC
it
ð Þ þ[
6
DROE
it
ÂSDISC
it
ð Þ þ[
7
ROE
it
þj
it
ð9Þ
Variable (pred. Sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 0.00130 0.667 0.5058
NANAL (À) À0.00091 À3.221 0.0015
LEV (+) 0.00003 1.863 0.0639
FDISC (À) À0.00007 À0.284 0.7771
SDISC (À) 0.00116 3.462 0.0006
LANALÂFDISC (À) À0.00096 À2.663 0.0083
DROEÂSDISC (À) À0.00109 À2.469 0.0143
ROE (?) 0.00009 1.101 0.2719
No. of observations=221; adjusted R
2
=15.58%. Variable de?nitions: COST=estimated cost of equity capital for ?rm i, year t—the
industry sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the
industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry sector median for year t; FDISC=?nancial
disclosure score for ?rm i, year t—the industry sector median for year t; SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t; ROE=return on equity for ?rm i, year t—the industry sector median for year t; LANAL=a dummy variable
set equal to one if the number of analysts following ?rm i in year t is below i’s industry sector median for year t, zero otherwise;
DROE=a dummy variable set equal to one if return on equity for ?rm i, year t is above the industry sector median for year t, zero
otherwise.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 613
explanations for the results on the relationship
between social disclosure and cost of capital. First,
if there were a consistent bias in social disclosures
where ?rms that experience higher than average
social costs disclose more information, then, on
average, the results reported could hold. The
descriptive literature on social disclosures has
reported that there are severe biases in reporting
(e.g. Guthrie & Parker, 1990; Wiseman, 1982).
The reported bias is that ?rms tend to use social
disclosures for self-promotion. This means that
that ?rms tend to report the positive social con-
tributions that they make but under-report nega-
tive social e?ects. Unfortunately, the relationship
between this bias and the costs incurred and ben-
e?ts received by the ?rm is unclear (Richardson et
al., 1999).
Second, the results could hold if two things are
true. It may be that social responsibility invest-
ments by ?rms are consistently negative present
value projects increasing the overall risk of the
?rm. While proponents of socially responsible
corporate behavior point to the potential cost
savings and long-term strategic advantage of such
behavior (e.g. Porter & van der Linde 1995; Scal-
tegger & Figge, 1998), the market may hold a dif-
ferent view. If this is the case and there is a positive
correlation between social disclosures and social
responsibility actions, then the observed results
could hold.
It is also possible that the results are speci?c to
the data used in this analysis. The time period for
which data were available was an economic reces-
sion. The positive link between social disclosure
and the cost of equity capital we document may be
contingent on these macro economic conditions.
This interpretation is supported by our result that
the e?ect of social disclosure on the cost of capital
is moderated by return on equity (see Table 10).
For ?rms with ROE above the industry median,
there is a negative interaction between social dis-
closure and return on equity that o?sets the main
e?ect between cost of capital and social disclosure.
In other words, for ?rms with above average ?nan-
cial performance, social disclosure may have no
e?ect on the cost of capital. If this is a reasonable
proxy for economic conditions, then the relation-
ship between social disclosure and cost of capital
may be limited to periods of economic downturn.
Future research that expands the data set over a
complete business cycle to test for this e?ect is
clearly called for by our results.
The e?ect of social disclosure on the cost of
equity capital documented in this study should not
be taken to imply that social disclosure has an
overall negative e?ect on the ?rm. Social and
environmental issues have signi?cant distribu-
tional e?ects. Although investors may require a
higher cost of capital for ?rms with a signi?cant
social agenda, other groups such as employees,
customers, regulators and supply-chain partners
may provide greater support to the ?rm because of
these actions. Much of the work on developing a
social accounting agenda has, in fact, been pre-
mised on the assumption that corporate social
disclosure will bene?t a broader community of
stakeholders than the capital providers that are
the primary audience for ?nancial disclosures (e.g.
Gray, 2000). The e?ect of social disclosures on the
expected cost of contributions to the ?rm from
other stakeholders has still to be examined.
Acknowledgements
The authors gratefully acknowledge the ?nan-
cial support of the Certi?ed General Accountants
of Canada Research Foundation and comments
on earlier drafts by Irene Gordon, Marc Epstein
and an anonymous reviewer. The authors also
gratefully acknowledge the contribution of I/B/E/
S International Inc. for providing earnings per
share forecast data, available through the Institu-
tional Brokers Estimate System. These data have
been provided as part of a broad academic pro-
gram to encourage earnings expectations
research.
Appendix A
Panel 1: Social disclosure categories of informa-
tion, maximum number of points allocated to the
category, and number of sub-categories allocated
to each category—information provided in the
1991 SMAC/UQAM Report:
(continued on next page)
614 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
Category of
information
Maximum
points
No. of sub-
categories
Human resources 18 29
Products, services,
and consumers
10 23
Community 18 24
Environment 18 22
Energy resources 6 10
Governments 10 14
Suppliers 6 16
Shareholders 6 9
Competitors 4 9
Miscellaneous 4 14
Totals 100 170
Panel 2: Financial disclosure categories of
information, maximum number of points allo-
cated to the category, and number of sub-cate-
gories allocated to each category—information
provided in the 1991 SMAC/UQAM Report.
Category Maximum
points
No. of sub-
categories
General information
Financial retrospective 10 7
Financial forecasts 10 11
Graphs and tables 10 4
Points of view on
regulations, competition
and economy
5 23
Past performance
and highlights
7 25
Future prospects 5 12
Investment and
disinvestment
5 22
Research, development
and environment
5 14
Risks and uncertainties 5 9
Chairman’s (sic) report 2 4
Financial statements
Sector Information 10 34
Income Statement 10 10
Accounting policies 5 7
Other notes to
?nancial statements
7 13
Pension plans
and leases
3 37
Government aid
and income tax
3 8
Exports and portion
of products
manufactured in Canada
3 8
Information on e?ects
of price ?uctuations
3 8
Special notes on total
quality, environment,
training, and technology
6 2
Miscellaneous 6 3
Totals 120 261
References
Botosan, C. (1997). Disclosure level and the cost of equity
capital. The Accounting Review, 72, 323–349.
Botosan, C., M. Plumlee. 2000. Disclosure level and expected
cost of equity capital: An examination of analysts’ rankings
of corporate disclosure. Working paper, University of Utah,
January, 2000.
Clarkson, P., Guedes, J., & Thompson, R. (1996). On the
diversi?cation, observability, and measurement of estimation
risk. Journal of Financial and Quantitative Analysis (March),
69–84.
Committee on Corporate Disclosure, 1995. Toward Improved
Disclosure: A Search for Balance in Corporate Disclosure.
The Toronto Stock Exchange.
Deegan, C., & Gordon, B. (1996). A study of environmental
disclosure practices of Australian corporations. Accounting
and Business Research, 26, 187–199.
Diamond, D., & Verrecchia, R. (1991). Disclosure, liquidity
and the cost of equity capital. Journal of Finance (Septem-
ber), 1325–1360.
Downing, P. (1997). Upping the stakes. CA Magazine, (June
and July), 41–43.
Edwards, E., & Bell, P. (1961). The Theory and measurement of
business income. Berkeley, CA: University of California Press.
Epstein, M. 2000. The identi?cation, measurement and report-
ing of corporate social impacts—revisited after twenty-?ve
years. Paper presented to the Accounting, Organizations and
Society 25th Anniversary Conference, Oxford, UK, July.
Feltham, G., & Ohlson, J. (1995). Valuation and clean surplus
accounting for operating and ?nancial activities. Con-
temporary Accounting Research, (Spring), 689–731.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 615
Garrett, D. E. (1987). The e?ectiveness of marketing policy
boycotts: environmental opposition to marketing. Journal of
Marketing, 51(April), 46–57.
Gray, R. 2000. The social accounting project and Accounting,
Organizations and Society: privileging engagement, imagin-
ings, new accountings and pragmatism over critique? Paper
presented to the Accounting, Organizations and Society 25th
Anniversary Conference, Oxford, UK, July.
Gebhardt, W., Lee, C. M. C., & Swaminathan, B. (2000).
Toward an implied cost of capital. Working paper, Cornell
University, October 27, 2000. Journal of Accounting Research
(in press).
Gibbins, M., Richardson, A. J., & Waterhouse, J. (1990). The
management of ?nancial disclosure: opportunism, ritualism,
policies and processes. Journal of Accounting Research, 28(1),
121–143.
Guthrie, J., & Parker, L. D. (1990). Corporate social disclosure
practice: a comparative international analysis. Advances in
Public Interest Accounting, 3, 159–175.
Hannan, M. (1998). Rethinking age-dependence in organiza-
tional mortality: logical formalizations. American Journal of
Sociology, 104(1), 126–164.
Healy, P., Hutton, A., & Palepu, K. (1999). Stock performance
and intermediation chanes surrounding sustained increase in
disclosure. Contemporary Accounting Research, 11(2), 485–
520.
Lang, M., & Lundholm, R. (1993). Cross-sectional determi-
nants of analyst ratings of corporate disclosures. Journal of
Accounting Research, 246–271.
Lang, M., & Lundholm, R. (1996). Corporate disclosure
policy and analyst behavior. The Accounting Review, 71,
467–492.
Levitt, A. 1999. Quality Information: The Lifeblood of Our
Markets. The Economic Club of New York, NewYork, N.Y.,
October 18, 1999.http://www.sec.gov/news/spchindx.htm.
Menon, A., & Menon, A. (1997). ‘‘Enviropreneurial’’ market-
ing strategy: the emergence of corporate environmentalism as
market strategy. Journal of Marketing, 6, 51–67.
Ohlson, J. (1995). Earnings, book value, and dividends in security
valuation. Contemporary Accounting Research, 661–687.
Oliver, C. (1991). Strategic responses to institutional processes.
Academy of Management Review, 16(1), 145–172.
Patten, D. (1991). Exposure, legitimacy and social disclosure.
Journal of Accounting and Public Policy, 10, 297–308.
Pava, M., & Krausz, J. (1996). The association between cor-
porate social responsibility and ?nancial performance: the
paradox of social cost. Journal of Business Ethics, 15, 321–
357.
Porter, M., & van der Linde, Claas (1995). Green and com-
petitive: ending the stalemate. Harvard Business Review,
September–October.
Reyes, M. G., & Grieb, T. (1998). The external performance of
socially responsible mutual funds. American Business Review
(January), 1–7.
Richardson, A., Welker, M., & Hutchinson, I. (1999). Manag-
ing capital market reactions to corporate social responsibility.
International Journal of Management Reviews, 1, 17–43.
Scaltegger, S., & Figge, F. ‘‘Environmental shareholder value’’
Centre for economics and business administration, Uni-
versity of Basle WWW-study no. 54, June 1998.
Sengupta, P. (1998). Corporate disclosure and the cost of debt.
The Accounting Review, 73(4), (October), 459–474.
Stinchcombe, A. L. (1965). Social structure and organizations.
In J. G. March (Ed.), Handbook of organizations. Chicago:
Rand-McNally.
Watts, R., & Zimmerman, J. (1996). Positive accounting theory.
Englewood Cli?s: Prentice-Hall.
Welker, M. (1995). Disclosure policy, information asymmetry
and liquidity in equity markets. Contemporary Accounting
Research (Spring), 801–827.
Wiseman, J. (1982). An evaluation of environmental dis-
closures made in annual reports. Accounting, Organizations
and Society, 7(1), 53–63.
Zeghal, D., & Ahmed, S. A. (1990). Comparison of social
responsibility information media used by Canadian ?rms.
Auditing, Accounting and Accountability Journal, 3(1), 38–53.
616 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
doc_345576708.pdf
Abstract
We test the relation between financial and social disclosure and the cost of equity capital for a sample of
Canadian firms with year-ends in 1990, 1991 and 1992. We find that, consistent with prior research, the quantity
and quality of financial disclosure is negatively related to the cost of equity capital for firms with low analyst
following. Contrary to expectations, there is a significant positive relation between social disclosures and the cost
of equity capital. This positive relationship is mitigated among firms with better financial performance. We consider
some biases in social disclosures that may explain this result. We also note that social disclosures may benefit the
firm through its effect on organizational stakeholders other than equity investors
Social disclosure, ?nancial disclosure
and the cost of equity capital
Alan J. Richardson*, Michael Welker
School of Business, Queen’s University, Kingston, Canada K7L 3N6
Abstract
We test the relation between ?nancial and social disclosure and the cost of equity capital for a sample of
Canadian ?rms with year-ends in 1990, 1991 and 1992. We ?nd that, consistent with prior research, the quantity
and quality of ?nancial disclosure is negatively related to the cost of equity capital for ?rms with low analyst
following. Contrary to expectations, there is a signi?cant positive relation between social disclosures and the cost
of equity capital. This positive relationship is mitigated among ?rms with better ?nancial performance. We consider
some biases in social disclosures that may explain this result. We also note that social disclosures may bene?t the
?rm through its e?ect on organizational stakeholders other than equity investors. # 2001 Published by Elsevier
Science Ltd.
Regulators have argued that equity markets
require comprehensive and transparent disclosures
of value-relevant information by ?rms in order to
function e?ciently (e.g. Levitt, 1999). Theoreti-
cally, adopting such a ‘‘disclosure position’’ (Gib-
bins, Richardson, & Waterhouse, 1990) should
bene?t ?rms through lower cost of capital for at
least two reasons. First, increased disclosure by
?rms reduces transaction costs for investors
resulting in greater liquidity of the market and
greater demand for the ?rm’s securities (e.g. Dia-
mond & Verrecchia, 1991). Second, increased dis-
closure reduces the estimation risk or uncertainty
regarding the distribution of returns (Clarkson,
Guedes, & Thompson, 1996).
In spite of the regulatory and theoretical sup-
port for increased disclosure by ?rms, direct evi-
dence of a negative empirical relation between
disclosure levels and the cost of capital is limited
(e.g. Botosan, 1997; Botosan & Plumlee, 2000, on
the cost of equity capital, and Sengupta, 1998 on
the cost of debt). In part, the lack of strong
empirical ?ndings on the relationship between
disclosure and cost of capital may be an artifact of
the markets and information set that are used in
empirical tests. If there is little variation in the
information disclosed due to e?ective regulatory
interventions, or if analysts routinely generate
information independently of the ?rms’ own dis-
closures, then the power of empirical tests will be
signi?cantly reduced. For example, Botosan
(1997) documents a statistically signi?cant nega-
tive relation between the level of ?nancial dis-
closure and cost of equity capital for her sample of
USA manufacturing ?rms, but this relation holds
0361-3682/01/$ - see front matter # 2001 Published by Elsevier Science Ltd.
PI I : S0361- 3682( 01) 00025- 3
Accounting, Organizations and Society 26 (2001) 597–616
www.elsevier.com/locate/aos
* Corresponding author.
only for the subset of her sample characterized by
limited analyst following.
A stronger test of the relationship between cor-
porate information disclosures and the cost of
equity capital is possible by choosing markets and
information sets where, ex ante, corporate dis-
closures play a larger role in market valuations. In
this paper we test this relationship in Canadian
markets and with both ?nancial and social dis-
closures. Both of these extensions to the literature
should improve the power of statistical tests as
explained later.
First, we examine a set of Canadian ?rms, pro-
viding an assessment of the bene?ts of expanded
disclosure in an environment other than the Uni-
ted States of America (US). Since the US equity
markets are claimed to be among the most
sophisticated in the world (e.g. Levitt, 1999),
including some of the most stringent disclosure
standards in the world, this extension is poten-
tially important. The generally less comprehensive
required disclosures in Canada create an environ-
ment where variation in voluntary disclosure
could be very important.
1
Second, we examine the relation between both
social and ?nancial disclosures and cost of equity
capital estimates. The past literature has only
examined the relation between ?nancial disclosure
and cost of capital. As summarized in Botosan,
past literature suggests that ?nancial disclosures
could in?uence the cost of capital because the dis-
closures reduce information asymmetry and/or
estimation error. As Richardson, Welker, and
Hutchinson (RWH, 1999) discuss, there are a
number of reasons to suspect a relation between
the cost of equity capital and social disclosure as
well. Past empirical examinations of the relation
between social performance, or social disclosure,
and equity market measures such as realized
equity returns have generally been poorly speci-
?ed, leading to results that are di?cult to inter-
pret. RWH provide an extensive review of this
literature and present a comprehensive model
outlining the ways that social performance and
disclosure about that performance might in?uence
equity market measures. Their analysis focuses on
three distinct ways in which social performance
and disclosure could a?ect the cost of capital.
Social disclosure could play a role similar to
?nancial disclosure and reduce the cost of equity
capital by reducing transaction costs and/or
reducing estimation error. In addition to these
two e?ects, social disclosure could in?uence the
cost of equity capital directly through investor
preference e?ects if investors are willing to
accept a lower expected return on investments
that also ful?ll social objectives. The relationship
between the cost of capital and social disclosures/
performance is one of the issues identi?ed by
Epstein (2000) for future research in his review of
the ?eld.
Consistent with the past literature, we ?nd a
signi?cant negative relationship between the cost
of equity capital and ?nancial disclosure for those
?rms with a small ?nancial analyst following.
Contrary to expectations, we ?nd a signi?cant
positive relationship between social disclosure
and the cost of equity capital. The cost of
equity penalty for ?rms with extensive social
disclosure is mitigated by higher ?nancial
performance.
1. Hypothesis development
The relation between ?nancial disclosure and
the cost of equity capital has been extensively
developed in the past literature (Clarkson et al.,
1996; Diamond & Verrecchia, 1991). Botosan
(1997), for example, argues that ?nancial dis-
closure could result in decreased cost of capital
because expanded disclosure reduces estimation
risk, decreasing the total risk in owning the
equity security, or reduces risk by decreasing
1
The view that US disclosure practices provide more infor-
mation than Canadian practices is apparently widely held.
Nearly 90% of the Canadian analysts surveyed between
November 1994 and January 1995 on behalf of the Toronto
Stock Exchange’s Committee on Corporate Disclosure respon-
ded that disclosure was better in the USA. None of the analysts
felt that disclosure was better in Canada. Reasons provided for
the belief that disclosure is better in the USA included: more
stringent regulation, greater volume of information and more
detailed segmentation of information (Committee on Corpo-
rate Disclosure, 1995).
598 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
information asymmetries and, hence, adverse
selection risk. In either case, an inverse relation
between ?nancial disclosure and cost of equity
capital is predicted.
Richardson et al. (1999) argue that there are at
least three reasons to expect a similar relation
between social disclosure and the cost of equity
capital.
2
They conclude that there may be a direct
in?uence of social disclosure on the cost of equity
capital either through investor preference e?ects,
or through reduced information asymmetry or
estimation risk. The e?ects stemming from
reduced information asymmetry and/or estimation
risk follow directly from the literature on ?nancial
disclosure. If information about social activities is
relevant to assessing the ?rm’s prospects, then
enhanced disclosure of social activities has the
same e?ect as enhanced disclosure of other ?nan-
cial activities.
Investor preference e?ects arise if investors are
willing to accept a lower rate of return on invest-
ments by an organization that supports a social
cause for which some investors have an a?nity.
3
This suggestion is consistent with the emergence of
Green Funds and Ethical Investing (e.g. Reyes &
Grieb, 1998). It also has a direct relationship to
the literature in organizational behavior, manage-
ment, and marketing that suggests that advertising
with a social dimension can be employed to legit-
imate the ?rm in the eyes of consumers and con-
tribute to the ?rms’ product/service market
success (e.g. Garrett, 1987; Menon & Menon,
1997). This literature suggests that consumers
‘‘vote’’ with their dollars and may (rationally)
choose to pay more to both acquire a product or
service and support a social cause for which they
have an a?nity. The extension of this literature to
the capital market is straightforward if investment
decisions are recognized as decisions to forego
current consumption in favor of future consump-
tion. There is also considerable anecdotal evidence
suggesting a link between investor preferences and
social reporting. For example, Downing (1997)
reports that managers of the Canadian Pension
Plan’s $100 billion fund, among other investors,
might be attracted by the information provided by
social reporting.
Our empirical examination does not attempt
to discriminate between these potential e?ects,
but does stand in marked contrast to the past
literature examining the equity market con-
sequences of social disclosure that has tended to
focus on the relation between social performance,
social disclosure and ex-post measures of ?nancial
performance. Ours is the ?rst empirical examina-
tion of social disclosure practices to explicitly
examine the cost of equity capital and the ?rst to
jointly examine the e?ects of both ?nancial and
social disclosure. The speci?c hypotheses we
examine are outlined later, each stated in alter-
native form.
The past literature has suggested that increased
?nancial disclosure reduces information asym-
metry and/or estimation risk. This literature also
suggests that enhanced ?nancial disclosures reduce
the cost of equity capital.
H1a: There is an inverse relation between the
level of ?nancial disclosure and the cost of
equity capital.
The arguments presented earlier suggest that
similar relations exist between the level of social
disclosure and information asymmetry, estimation
risk and the cost of equity capital. Also, if, as has
often been assumed in the past literature, there is a
perceived positive relation between the level of
social disclosure of the ?rm and the ?rm’s social
performance, increased social disclosure may
reduce the cost of capital through investor pref-
erence e?ects.
2
RWH also argue that disclosure about social activities
undertaken by the ?rms could provide investors with informa-
tion about future cash ?ows, or in terms consistent with the
model presented later, future abnormal earnings. This link
might exist, for example, if social activities decrease expected
future regulatory costs, in?uence consumers to acquire the
?rm’s products/services thereby increasing the ?rm’s contribu-
tion margins or market share, or reduce the costs of implicit or
explicit contracting. In short, if social activities have net present
value consequences, then information about social performance
should in?uence investors’ assessments of the abnormal earn-
ings the ?rm can earn in the future (see Scaltegger & Figge,
1998).
3
This linkage is more thoroughly discussed in RWH (1999).
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 599
H2a: There is an inverse relation between the
level of social disclosure and the cost of equity
capital.
Finally, we consider the interaction between
?rms’ own disclosures about their activities and
the information available from third parties. We
hypothesize that the relationship will di?er for
?nancial disclosure and social disclosure. The
literature on ?nancial disclosure suggests that the
bene?ts of ?nancial disclosure may be greater
for ?rms with little analyst following (Botosan,
1997). This assumes that stakeholders are
concerned with a ?rm’s ?nancial performance but
that the richness of the information set available
to them varies depending on the number of
analysts preparing independent reports on the
?rm. In the absence of information about the
?rm from analysts, the ?rm’s own disclosures are
the key source of information. The bene?ts of
better ?nancial disclosure are primarily realized
when other information sources are absent.
Therefore we predict a negative relationship
between the cost of capital ?nancial disclosure
where there is a small number of analysts follow-
ing a ?rm.
H3a(i): The relation between ?nancial dis-
closure levels and the cost of equity capital is
mediated by the level of analyst following.
The e?ect of social disclosure is expected to
follow the same pattern to the extent that social
disclosures inform stakeholders’ expectations of
the ?rm’s ?nancial performance. The importance
of social performance to stakeholders, however,
has been theorized to increase with the size of the
organization. Stinchcombe (1965), for example,
has argued that as a ?rm grows it develops a
structural position (i.e. ties within a network of
resource providers) that contributes to its success.
This structural position changes the demands of
the environment from a demand for short-run
economic e?ciency to a demand for long-run
economic and social e?ciency/legitimacy (see
Hannan, 1998; Oliver, 1991). This prediction is
consistent with the positive accounting theory
literature that suggests that political costs are
greater for larger ?rms (Watts & Zimmerman,
1986). The number of analysts following a ?rm
tends to be correlated with ?rm size (e.g. see
Table 1), thus the importance of social disclosure
and the independent information available from
analysts will increase in parallel. We, therefore,
predict no interaction e?ect between social dis-
closure and the number of analysts following the
?rm.
H3a(ii): The relation between social disclosure
levels and the cost of equity capital is not
mediated by the level of analyst following.
2. Empirical measures of disclosure levels and
cost of equity capital
2.1. Disclosure proxy
Our empirical measures of ?nancial and social
performance are drawn from the joint Society of
Management Accountants of Canada (SMAC)/
University of Quebec at Montreal (UQAM)
sponsored assessments of the annual reports of
a broad cross-section of Canadian companies.
These assessments were conducted and publicly
reported based on 1990, 1991 and 1992 annual
reports, thus providing a limited time-series of
disclosure scores. To our knowledge, these data
are unique in North America because the
scores contain a ranking of ?rms on both the
quality and level of ?nancial disclosure and
social disclosure contained in annual reports.
This provides a signi?cant advantage over
other North American rankings such as the
Association for Investment Management and
Research (AIMR) rankings of US companies.
4
However, the AIMR rankings are prepared
by professional analysts and represent the
rankings of several analysts, while the Canadian
4
The AIMR rankings are compiled annually by the Asso-
ciation for Investment Management and Research. These dis-
closure rankings have been used in empirical studies by
Botosan and Plumlee (2000), Healy, Hutton, and Palepu
(1999), Lang and Lundholm (1993, 1996), and Welker (1995).
600 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
rankings available to us are based on the judg-
ments of less experienced raters and do not re?ect
the same level of averaging across raters as the
AIMR ratings. Nevertheless, the SMAC/UQAM
ratings are the best available source of disclosure
ratings for a broad cross-section of Canadian
?rms. The only alternative measure of disclosure
for Canadian companies would be researcher-
generated measures, as utilized by Botosan (1997).
We choose not to generate our own disclosure
ratings because of the potential for researcher
biases to in?uence the ratings and to avoid the
severe limitations on sample size imposed by this
approach.
For each year from 1990 to 1992, researchers at
UQAM analyzed the annual reports of around
700 Canadian companies coming from nine
industry sectors. The industry sectors reported on
in their publication include: Manufacturing-
Industrial Products, Manufacturing-Consumer
Products; Oil, Gas and Chemicals; Mines, Metals
and Forestry Products; Technology and Com-
munications; Financial Institutions; Retail and
Wholesale Trade; Management and Other; and
Utilities. An extensive checklist of information
related to socially responsible activities was
developed that contained 170 subcategories of
information. Similarly, an extensive checklist
related to ?nancial information was developed
which allocated points across 261 individual
disclosure elements.
Appendix A contains a summary of the 10
categories of social information considered and
the maximum number of points allocated to each
category, as well as the sub-categories of infor-
mation considered within each category. The
Appendix also contains similar information for
the ?nancial disclosure checklist.
Two points should be noted about these check-
lists. First, the social information captured in the
checklist includes a much broader set of dis-
closures than just environmental disclosures that
have been the subject of much of the past social
disclosure literature.
Second, the checklist used to assess ?nancial dis-
closure is similar in many respects to the checklists
utilized by the AIMR and developed by Botosan
(1997). For example, all of the checklists contain
sections devoted to general background informa-
tion that help users to interpret the ?nancial
statements. All three lists contain sections devoted
to assessing the usefulness of the disclosure of
summarized historical results, and all three check-
lists assess the inclusion of forecasted information
within the annual report. In addition to assessing the
types of disclosure made in the annual report, the
UQAM researchers also attempted to make an
assessment of the quality of the disclosure by
awarding more points for disclosures that contained
quantitative data or reported more information.
The extensive nature of the checklists utilized in the
SMAC/UQAM disclosure ratings, combined with
the attempt to discriminate between more and less
informative disclosures, gives us con?dence in the
face validity of these ratings. While they are
undoubtedly noisy measures, they provide us
with some ability to discriminate between ?rms
providing high levels of disclosure and those
providing minimal disclosure. Empirical analysis
reported later in the manuscript also provides evi-
dence corroborating the validity of our disclosure
measures.
3. Empirical measure of cost of equity capital
We follow Botosan (1997), Botosan and Plumlee
(2000) and Gebhardt, Lee, and Swaminathan
(2000), and obtain estimates of the cost of
equity capital using an accounting based valu-
ation model developed in Edwards and Bell
(1961), Feltham and Ohlson (1995) and Ohlson
(1995). Our empirical implementation of the
model is very similar to the method employed
by Gebhardt et al. and interested readers are
referred to their paper for further details on the
estimation procedure and for extensive empirical
analyses of the properties of the estimates. We
provide a brief sketch of the estimation procedure
later.
The valuation model speci?es a relation between
equity values and current book values and future
abnormal earnings. We brie?y sketch a derivation
of the empirical equation we use to estimate equity
cost of capital beginning with the familiar divi-
dend discount model:
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 601
P
it
¼
X
1
t¼1
E
t
d
itþt
È É
1 þ r
i
ð Þ
t
ð1Þ
where d=dividends, r=cost of equity capital, and
i and t are ?rm and time identi?ers, respectively.
Feltham and Ohlson (1995) and Ohlson (1995)
demonstrate that, for clean-surplus accounting
systems, this dividend discount model is algebrai-
cally equivalent to a valuation formula based on
current book value and future abnormal earnings:
5
P
it
¼ BV
it
þ
X
1
t¼1
E
t
x
itþt
À r
i
BV
itþtÀ1
È É À Á
1 þ r
i
ð Þ
t
ð2Þ
where BV=book value of equity, x=earnings,
and all other variables are as previously de?ned.
The ?nite horizon version of Eq. (2) is:
P
it
¼ BV
it
þ
X
T
t¼1
E
t
fFROE
itþt
À r
i
gBV
itþtÀ1
ð Þ
1 þ r
i
ð Þ
t
þTV
it
ð3Þ
where FROE=forecasted return on equity, and
TV=terminal value, or the present value at time t
of the abnormal earnings expected to be earned
after time t+T, and all other variables are as
previously de?ned.
For our estimates that utilize I/B/E/S earnings
forecasts, Eq. (3) is replaced with a version that
utilizes the earnings forecasts for the next three
?scal years from I/B/E/S.
6
The terminal value
involves forecasting return on equity for 12 future
years (i.e. 9 years beyond the latest available
explicit earnings forecast from I/B/E/S). Fore-
casted return-on-equity beyond year 3 is generated
using a linear fade-rate to the industry average
return-on-equity (ROE). Our industry average
ROEs are compiled from data provided by Statis-
tics Canada (Statscan). We acquire industry aver-
age ROE data beginning with 1980 for all
industries as de?ned and tracked by StatsCan. Our
industry average ROE ?gures for year t are com-
puted as the average industry ROE through time
starting in 1980 and ending in year tÀ1. For
example, for our 1990 industry average ROE
measure, we average the industry ROEs over
the 10 year period from 1980 to 1989, and
the 1991 industry ROEs are the average of
the industry ROEs over the 11 year period
from 1980 to 1990. These industry averages have
a mean of 10.75% and range from 4.3%, (the
1992 average for the iron, steel and related
products industry) to 22.7%, (the 1990 average
for the building materials and construction
industry).
Accordingly, the model we utilize to estimate
cost of capital has the following form:
5
A clean surplus accounting system is one in which book
value at time t is equal to book value at time tÀ1 plus earnings
minus dividends net of capital contributions. ‘‘Dirty surplus’’
arises when gains and losses a?ecting book value bypass the
income statement.
P
it
¼ BV
it
þ
X
3
t¼1
E
t
fFROE
itþt
À r
i
gBV
itþtÀ1
ð Þ
1 þ r
i
ð Þ
t
þ
P
11
t¼4
EfFROE
itþ3
þ t À 3 ð Þ½fIROE
it
ÀFROE
itþ3
g,9? À r
i
gBV
itþtÀ1
1 þ r
i
ð Þ
t
þ
IROE
it
À r
i
r
i
1 þ r
i
ð Þ
11
BV
itþ11
ð4Þ
6
About 10% of our ?rm year observations have two-year
ahead EPS forecasts but do not have a 3-year ahead forecast.
When the third year ahead forecast is missing, we forecast ?scal
year 3 (FY3) EPS by assuming that the earnings growth rate
implicit in the ?scal year 2 (FY2) compared with ?scal year 1
(FY1) forecasts applies to ?scal year 3 as well. Speci?cally, we
forecast FY3 EPS as FY2 EPSÂ(FY2 EPS/FY1 EPS). If FY2
EPS/FY1 EPS-0, we do not forecast FY3 EPS and omit the
observation.
602 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
where IROE is the historical industry average
ROE and all other variables are as previously
de?ned.
The exact details of the estimation procedure,
including variable measurement, are described in
the following section. The sensitivity of our results
to variations in the assumptions required to esti-
mate cost of capital are discussed in the section of
the paper entitled ‘‘sensitivity of results to cost of
capital estimation assumptions’’.
4. Empirical analysis
4.1. Disclosure proxy and additional data
Our initial sample consists of all ?rms that
received a disclosure rating based on any or all of
its 1990, 1991 or 1992 annual reports. In each of
these three years, the annual reports of over 700
Canadian companies were collected and analyzed
by a group of research assistants at UQAM. The
results of the ratings are summarized by industry
group and published each year by CMA Canada.
The survey only provides 3 years of data. Since
disclosure policies are probably relatively stable
?rm attributes, this limited time-series should not
severely impact the generality of our results. In
order to perform the initial empirical analysis, we
require ?nancial statement data provided by
Compustat, earnings forecasts and analyst follow-
ing provided by I/B/E/S, and market price and
returns data acquired through Datastream. These
additional data requirements leave us with a
sample of 324 ?rm year observations from 124
di?erent companies with all necessary price data,
?nancial and social disclosure scores, all necessary
Compustat data and at least 1-year ahead EPS
forecast available from I/B/E/S. For this initial
sample, we only require that 1-year ahead ana-
lysts’ forecasts be available, and therefore the
number of analysts making a 1-year ahead fore-
cast, be available from I/B/E/S. In order to
calculate cost of capital estimates, we require
that at least 2-year-ahead earnings forecasts be
obtained from I/B/E/S. We also require that
unambiguous identi?cation with a StatsCan
industry group be possible, further reducing our
sample size to 225 ?rm year observations from 87
di?erent ?rms.
For these observations, we obtain the average
stock price from the 6th month after year-end
from Datastream. We choose a period 6 months
after year-end to ensure that all information con-
tained in the annual ?nancial statements has been
disclosed and is re?ected in market prices.
7
This
choice is also consistent with the empirical analysis
in Botosan (1997). From Compustat, we obtain
the debt-to-equity ratio, return-on-equity, divi-
dend payout ratio, market to book value ratio,
and market value of equity. I/B/E/S provides the
earnings forecasts and the number of analysts fol-
lowing the ?rm, which we measure as the number
of analysts providing 1-year-ahead earnings fore-
cast for the ?rm.
Calculation of forecasted book values requires
an estimated future dividend payout ratio (k). We
estimate this ratio using the following procedure.
First, we obtain the historical 10-year average
payout ratio for each year t from Compustat and
use this to proxy for future payouts if it is avail-
able and positive. If the 10-year average is either
unavailable or negative, we use the 5-year histor-
ical average payout ratio. If the 5-year payout
ratio is either unavailable or negative, we estimate
future payout ratios based on the dividend payout
ratio for year t. Finally, if the year t payout ratio is
negative, we follow a procedure similar to Geb-
hardt et al. (2000) and approximate ‘‘permanent’’
earnings for year t as (BV
tÀ1
ÂIROE), the book
value of equity at the beginning of the year times
the industry average ROE. We then calculate the
dividend payout ratio as DVS
t
/(BV
tÀ1
ÂIROE),
where DVS is the dividends per common share for
year t. Our implementation of Eq. (4) then uses
the ?rm’s estimated dividend payout ratio (k) to
update book values as follows:
BV
it
¼ BV
itÀ1
þ FROE
it
1 À k ð Þ
it
È É
BV
itÀ1
ð4aÞ
7
Our results are qualitatively similar using average stock
prices from the 4th month after ?scal year-end. We use the
average stock price for the month rather than monthly closing
prices to avoid extreme observations that may result from
temporary share price ?uctuations re?ected in a single price
observation such as a closing price.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 603
As Gebhardt et al. point out; the calculation of
an implied cost of capital is the same as determin-
ing the internal rate of return that equates the
stock price to the expected future bene?ts of share
ownership. Gebhardt et al. also provide two
example calculations to which we refer interested
readers. We perform this estimation using Micro-
soft Excel, and our estimate of the cost of capital
is simply the discount rate that equates the book
value and future forecasted earnings to the current
market price.
Tables 1 and 2 provides distributional char-
acteristics and a correlation matrix for the vari-
ables used in the study. Financial disclosure scores
average 32.5 (out of a possible 120), and range
between 4.75 and 63. Social disclosure scores tend
to be lower, averaging 11 (out of 100) and ranging
between 0.25 and 50. The average ?rm size in the
sample is $3.8 billion and the median size is sub-
stantially lower at around $1.6 billion. Nine ana-
lysts follow the average ?rm, and this ranges
between 1 and 37. The average debt to equity ratio
is 75%. The mean ROE for our sample ?rms is
close to zero, re?ecting the poor economic climate
in Canada in the early 1990s.
8
The median ROE is
around 6%. Around one third of our sample ?rms
come from the oil, gas and chemical industry or
the mines, metals and forestry products industry,
industries that have been depicted as environmen-
tally/socially sensitive industries in the past litera-
ture (Deegan & Gordon, 1996; RWH, 1999).
Finally, our cost of capital estimates average close
to 9%, and range between 1.8 and 22.5%.
These estimates of cost of capital appear rea-
sonable in relation to Canadian T-bill rates during
our sample period. Our rates, measured with
8
During the 1990–1992 periods, quarterly growth in gross
domestic product in Canada was 0.3%, far below the average
of 1.1% experienced during the remainder of the 1990s (Cal-
culated based on Datastream data).
Table 2
Correlation matrix (Pearson above diagonal, Spearman below)
a
FDISC SDISC MV NANAL LEV ROE IND COST
FDISC 1 0.662* 0.499* 0.542* 0.067 0.009 0.229* À0.046
SDISC 0.704* 1 0.294* 0.391* 0.080 À0.023 0.298* 0.011
MV 0.555* 0.474* 1 0.413* À0.054 0.132* À0.143* 0.091
NANAL 0.603* 0.509* 0.617* 1 À0.023 0.029 0.321* À0.208*
LEV 0.177* 0.243* 0.024 0.042 1 À0.542* 0.011 0.054
ROE À0.067 À0.096 0.136* 0.009 À0.251* 1 À0.121* 0.074
IND 0.281* 0.278* 0.065 0.346* À0.013 À0.255* 1 À0.362*
COST À0.053 À0.012 0.066 À0.156* 0.131* 0.102 À0.351* 1
a
Variable de?nitions: FDISC=Financial disclosure score for ?rm i, year t; SDISC=social disclosure score for ?rm i, year t;
MV=beginning of year market value of common equity for ?rm i, year t; NANAL=number of analysts making a 1 year ahead EPS
forecast for ?rm i, year t; LEV=debt to equity ratio for ?rm i, year t; ROE=return on equity for ?rm i, year t; IND=a dummy variable
equal to one if ?rm i is a member of the Oil, Gas and Chemicals or Mine, Metals and Forestry Products industry sectors in year t, zero
otherwise; COST= estimated cost of equity capital for ?rm i, year t.
*=signi?cant at the 5% level, two-tailed test.
Table 1
Descriptive statistics
a
Variable name N Mean Median Minimum Maximum
FDISC 324 32.53 31.48 4.75 63.0
SDISC 324 11.04 8.25 0.25 50.0
MV ($000,000) 324 3833 1605 20.2 38,729
ANALFOL 324 9.32 9.0 1 37
LEV (%) 324 75.8 53.5 0 1095
ROE (%) 324 0.19 5.98 À194.0 33.1
IND 324 0.38 0 0 1
COST (%) 225 8.9 8.5 1.8 22.5
a
Variable de?nitions: FDISC=Financial disclosure score
for ?rm i, year t; SDISC=social disclosure score for ?rm i, year
t; MV=beginning of year market value of common equity for
?rm i, year t; NANAL=number of analysts making a 1 year
ahead EPS forecast for ?rm i, year t; LEV=debt to equity ratio
for ?rm i, year t; ROE=return on equity for ?rm i, year t;
IND=a dummy variable equal to one if ?rm i is a member of
the Oil, Gas and Chemicals or Mine, Metals and Forestry
Products industry sectors in year t, zero otherwise; COST=
estimated cost of equity capital for ?rm i, year t.
604 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
forecasted data as of June 1991 (for 1990 annual
reports), June 1992 and June 1993 average 9.85,
9.5 and 8.0%, respectively. The 1-year T-bill rate
in Canada shows similar declines throughout our
sample period. For example, the one-year T-bill
rate in June 1991, 1992 and 1993 was 9.11, 6.63
and 5.79%, respectively. In addition, the risk-
premium inherent in our cost of capital estimates
varies between approximately 0.7 and 2.5%, with
a median of 2.2%. Gebhardt et al. ?nd that
risk premiums implied for their sample (including
over 10,000 observations from 1979 to 1995
and the same method for estimating the cost
of equity capital) vary between À0.7 and 4.9%
with a median of 2.0%. Our cost of capital esti-
mates display a similar relationship to the risk-free
rate.
The univariate correlations reveal no unex-
pected patterns, and show that there is a strong
positive relation between social and ?nancial dis-
closure, as might be expected if both types of dis-
closure are part of an overall disclosure policy.
Cost of capital does not have a signi?cant relation
to either disclosure score. The correlation between
the disclosure variables and other potential deter-
minants of cost of capital (e.g. market value of
equity, analyst following and industry) makes
interpretation of these univariate correlations
problematic. However, these univariate correla-
tions do indicate that our main results are sensitive
to the inclusion of control variables.
Table 3 provides selected descriptive informa-
tion for ?rms in the sensitive, (oil, gas and chemi-
cal or mines, metals and forestry products), versus
non-sensitive industries, and across time for both
groups. Several empirical regularities are revealed
in this descriptive information. First, ?rms in the
sensitive industries have better ?nancial and social
disclosure scores, are followed by more analysts,
have inferior ?nancial performance (except 1992),
and have lower cost of capital estimates than do
their counterparts in the non-sensitive industries.
The two groups appear roughly similar in terms of
?rm size, though the mean ?rm size is typically
larger for the non-sensitive ?rms and the median
?rm size is larger for the sensitive ?rms. The ?rm
size distribution is much more skewed for the
non-sensitive sample. Additionally, cost of capital
estimates, ?nancial performance and the number
of analysts following the ?rm tend to decline
over time for both groups, though the ?nancial
performance for the sensitive ?rms rebounded in
1992.
4.2. Validation of the disclosure proxies
Since the SMAC/UQAM ratings of disclosure
have not been previously used in academic
research, we perform a series of tests that examine
the relationship between these disclosure ratings
and several variables that are related to disclosure
levels in the past literature. Speci?cally, we follow
Botosan (1997) and Lang and Lundhom (1993)
and examine the relation between ?nancial dis-
closure and ?rm size, ?nancial performance,
leverage and analyst following. Past results sug-
gest that a positive relationship should exist
between each of these variables and ?nancial dis-
closure. We also examine the relationship between
social disclosure ratings and ?rm size, industry
membership, ?nancial performance, leverage,
analyst following and the interaction of industry
and ?rm size. Industry membership is coded to
re?ect membership in a socially/environmentally
sensitive industry, namely the oil, gas and chemi-
cals industry or the mines, metals and forestry
products industry. Since larger ?rms and ?rms in
these industries have their social and environ-
mental performance closely scrutinized, past
research suggests a positive relation between
industry membership and the interaction of ?rm
size and industry membership with social dis-
closure (e.g. Deegan & Gordon, 1996).
The relationship between social disclosure and
?nancial performance has been mixed in the past
literature (Pava & Kraus, 1996) while the rela-
tionship between leverage and analyst following
and social disclosures have not been examined in
the past literature. Since social disclosure may
accomplish many of the same objectives as ?nancial
disclosure, we expect to observe positive relation-
ships between these variables and social disclosure.
These tests provide evidence on the validity of our
disclosure ratings as measures of the cross-sec-
tional variation in the level of disclosure provided
in Canadian companies’ annual reports.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 605
Table 3
Descriptive information — across sensitive and non-sensitive industries and time
Variable name N Mean Median Minimum Maximum
Sensitive industry—1990
FDISC 37 36.14 34.75 15.5 56
SDISC 37 13.82 11.25 1.25 45.5
MV ($000,000) 37 2478 1982 69.3 12,543
NANAL 37 13.49 12 3 34
ROE (%) 37 À1.58 6.00 À142.07 26.46
COST (%) 29 8.43 8.12 1.83 15.26
Sensitive industry—1991
FDISC 42 37.15 35.25 16.75 58
SDISC 42 15.31 14.38 2 40.5
MV ($000,000) 42 2840 1994 85.3 11,821
NANAL 42 12.12 10.5 2 36
ROE (%) 42 À8.67 1.95 À158.9 17.54
COST (%) 31 7.56 7.47 3.52 14.15
Sensitive industry—1992
FDISC 43 34.7 34.35 14.25 52.5
SDISC 43 13.95 13.5 2 50
MV ($000,000) 43 2511 2002 88.4 10,554
NANAL 43 10.37 10.0 1 30
ROE (%) 43 À0.81 2.66 À38.38 22.84
COST (%) 34 6.72 6.23 3.89 13.73
Non-sensitive industry—1990
FDISC 63 31.9 29 11.25 63
SDISC 63 10.2 7 0.75 38
MV ($000,000) 63 4675 1364 20.2 32,179
NANAL 63 8.76 8 1 37
ROE (%) 63 5.09 9.98 À97.06 23.68
COST (%) 40 10.84 10.53 6.55 20.34
Non-sensitive industry—1991
FDISC 71 29.7 46.75 8.7 59.5
SDISC 71 8.62 7 0.25 33.5
MV ($000,000) 71 4447 1245 27.5 37,839
NANAL 71 7.82 7 1 34
ROE (%) 71 4.32 7.63 À77.71 33.09
COST (%) 47 10.06 9.62 4.04 19.74
Non-sensitive industry—1992
FDISC 68 29.81 29.28 4.75 63
SDISC 68 8.35 6.38 1 43.25
MV ($000,000) 68 4452 1393 66.1 38,729
NANAL 68 6.77 6.0 1 26
ROE (%) 68 À1.59 6.42 À194.0 31.15
COST (%) 44 8.98 9.1 3.11 22.48
Variable de?nitions: FDISC=?nancial disclosure score for ?rm i, year t; SDISC=social disclosure score for ?rm i, year t; MV=be-
ginning of year market value of common equity for ?rm i, year t; NANAL=number of analysts making a 1 year ahead EPS forecast
for ?rm i, year t; ROE=return on equity for ?rm i, year t; COST=estimated cost of equity capital for ?rm i, year t.
606 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
Additionally, we note that our ratings are based
only on the disclosures contained in annual
reports and do not necessarily re?ect the level of
disclosures made through other media (c.f. Zeghal
& Ahmed, 1990). While this is an acknowledged
limitation of our proxy for disclosure, Botosan
(1997) shares this limitation. We also note that
Lang and Lundholm (1993) and Welker (1995)
examine the AIMR ratings for US companies and
report a high degree of correlation between dis-
closure ratings based on the annual report and
ratings based on other disclosure media. This
?nding provides evidence of convergent validity of
annual report disclosure measures and disclosures
occurring in other media as well.
4.3. Relation of ?nancial disclosure scores with size,
?nancial performance, leverage and analyst following
Past research has demonstrated that ?nancial
disclosure increases with ?rm size, ?nancial per-
formance, leverage, and the number of analysts fol-
lowing the ?rm. One way to validate our empirical
proxy for ?nancial disclosure is to examine these
relationships based on our proxy. Accordingly, we
estimate the following empirical equation:
FDISC
it
¼ o þ[
1
DSIZE
it
þ[
2
ROE
it
þ[
3
LEV
it
þ[
4
DANAL
it
þc ð5Þ
where FDISC=?nancial disclosure score for ?rm
i, year t, DSIZE=a dummy variable equal to one
if the beginning of year market value of equity for
?rm i, year t is above the sample median, zero
otherwise, ROE=return on equity for ?rm i, year
t, LEV=debt to equity ratio, ?rm i, year t,
DANAL=a dummy variable equal to 1 if the
number of analysts providing a 1-year ahead
earnings forecast for ?rm i, year t is above the
sample median, zero otherwise.
The results of estimating Eq. (5) are reported in
Table 4.
9
Consistent with the past literature, each
of the variables in the equation except ROE exhi-
bits a signi?cant and positive relation with ?nan-
cial disclosure. The adjusted R
2
of the regression is
over 30%, indicating that the explanatory vari-
ables are able to explain a reasonable portion of
the cross-sectional variation in ?nancial disclosure
scores. The fact that our ?nancial disclosure scores
are strongly related to variables which the past
literature suggests explain ?nancial disclosure
increases our con?dence that variation in our dis-
closure measure captures the underlying phenom-
enon of interest, variation in ?nancial disclosure.
Table 4
Results of estimating Eq. (5) explaining variation in ?nancial disclosure
FDISC
it
¼ o þ[
1
DSIZE
it
þ[
2
ROE
it
þ[
3
LEV
it
þ[
4
DANAL
it
þc
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 34.113 26.323 0.0001
DSIZE (+) 6.895 5.481 0.0001
ROE (+) 0.032 1.277 0.2026
LEV (+) 0.014 2.535 0.0117
DANAL (+) 8.505 6.776 0.0001
No. of observations=324; adjusted R
2=
32.7%. Variable de?nitions: FDISC=?nancial disclosure score for ?rm i, year t; DSIZE=a
dummy variable equal to one if the beginning of year market value of equity for ?rm i, year t is above the sample median, zero
otherwise; ROE=return on equity for ?rm i, year t; LEV=debt to equity ratio, ?rm i, year t; DANAL=a dummy variable equal to 1
if the number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t is above the sample median, zero otherwise.
9
Our tabulated results provide two-tailed P-values. By
convention, we describe an estimated coe?cient as statistically
signi?cant if the coe?cient is signi?cant at the 0.05 level in a
one-tailed test if we predict the sign of the coe?cient and a two-
tailed test if we do not predict the sign of the coe?cient.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 607
4.4. The relationship of social disclosure scores
with ?rm size, industry membership, ?nancial
performance, leverage and analyst following
Patten (1991) documents that social disclosure is
increasing in ?rm size and is greater in highly
visible and politically sensitive industries. Deegan
and Gordon (1996) ?nd an interaction e?ect
between ?rm size and industry, such that the size
e?ect is particularly pronounced in sensitive
industries. As Richardson et al. (1999) discuss,
results of tests of the relation between social per-
formance/disclosure and ?nancial performance
have been mixed. Leverage and number of ana-
lysts following the ?rm have not been included in
past studies of social disclosure. We include these
variables in our regression since they are related to
?nancial disclosure. Social disclosure accomplishes
similar objectives and may have similar deter-
minants. We examine the relation between the
SMAC/UQAM social disclosure ratings and these
variables by estimating the following equation:
SDISC
it
¼ , þ’
1
DSIZE
it
þ’
2
ROE
it
þ’
3
IND
it
þ’
4
IND
it
ÂSIZE
it
ð Þ þ’
5
DANAL
it
þ
’
6
LEV
it
þu
it
ð6Þ
where SDISC=social disclosure score for ?rm i,
year t, IND=a dummy variable equal to one if
?rm i is a member of the Oil, Gas and Chemicals
or Mine, Metals and Forestry Products industry
sectors in year t, zero otherwise
All other variables are as previously de?ned.
The results of estimating Eq. (6) are presented in
Table 5. Again, the results are encouraging as our
measure of social disclosure appears to be related to
variables that the past literature suggests it should
be. Size, industry, and the interaction of industry and
size are all signi?cantly related to social disclosure,
as suggested by the past literature. Analyst following
is also statistically positively related to social dis-
closure, consistent with our conjectures that similar
factors may in?uence ?nancial and social dis-
closure. ROE is not statistically related to social
disclosure, in keeping with the insigni?cant or mixed
relation documented in the past literature. Lever-
age is also positively related to social disclosure,
again consistent with our conjectures that social and
?nancial disclosures have similar determinants.
We conclude, based on the evidence discussed
earlier, that the ?nancial and social disclosure
scores are valid measures of those disclosures by
the ?rms in our sample. We now turn to the sub-
stantive issue of the relationship between dis-
closure and the cost of capital.
Empirical tests of the relation between disclosure
and the cost of equity capital
Our tests of the relations between ?nancial and
social disclosure and the cost of equity capital are
Table 5
Results of estimating Eq. (6) explaining variation in social disclosure
SDISC
it
¼ , þ’
1
DSIZE
it
þ’
2
ROE
it
þ’
3
IND
it
þ’
4
IND
it
ÂSIZE
it
ð Þ þ’
5
DANAL
it
þ’
6
LEV
IT
þu
it
ð6Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 5.097 6.315 0.0001
DSIZE (+) 3.507 3.021 0.0027
ROE (?) 0.021 1.092 0.2757
IND (+) 2.465 1.957 0.0513
INDÂSIZE (+) 2.958 1.713 0.0876
DANAL (+) 3.931 3.918 0.0001
LEV (+) 0.011 2.596 0.0099
No. of observations=324; adjusted R
2
=26.8%. Variable de?nitions: SDISC=social disclosure score for ?rm i, year t; DSIZE=a
dummy variable equal to one if the beginning of year market value of equity for ?rm i, year t is above the sample median, zero
otherwise; ROE=return on equity for ?rm i, year t; LEV=debt to equity ratio, ?rm i, year t; DANAL=a dummy variable equal to 1
if the number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t is above the sample median, zero otherwise;
IND=a dummy variable equal to one if ?rm i is a member of the Oil, Gas and Chemicals or Mine, Metals and Forestry Products
industry sectors in year t, zero otherwise.
608 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
based on the work of Botosan (1997) and Geb-
hardt et al. (2000). Similar to Botosan, we attempt
to document a negative relation between ?nancial
and social disclosure and the cost of equity capital
that is incremental to the e?ects of other variables
known to in?uence cost of capital. Gebhardt et al.
?nd that the risk premia is negatively related to
the number of analysts following the ?rm and
positively related to leverage. Accordingly, our
empirical tests include these variables as control
variables.
10
In addition, Botosan documents that
disclosure and analyst following has an interactive
e?ect on cost of capital. She ?nds that disclosure
reduces cost of capital only for those ?rms with
low analyst following. We also estimate an equa-
tion that allows for this potential interactive e?ect
and an interaction between analyst following and
social disclosure as well. Speci?cally, our primary
empirical tests come from the estimation of the
following two equations:
COST
it
¼, þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þj
it
ð7Þ
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þ[
5
LANAL
it
ÂFDISC
it
ð Þþ
[
6
LANAL
it
ÂSDISC
it
ð Þ þj
it
ð8Þ
where COST is the cost of equity capital estimated
by Eq. (4); NANAL is the number of analysts
making 1-year ahead earnings forecast; LANAL is
a dummy variable set equal to one if the number
of analysts following ?rm i in year t is below i’s
industry sector median for year t, zero otherwise.
All other variables are as previously de?ned and all
variables are adjusted for industry sector medians.
Gebhardt et al. (2000) document considerable
variation in their risk premia estimates across
industries. Our results (documented in Tables 3
and 5) and the results of previous research
demonstrate that disclosure practices also vary
across industries as well. Accordingly, all variables
utilized in estimating Eqs. (7) and (8) are adjusted
for industry sector (as de?ned in the SMAC/
UQAM disclosure ratings) year medians. This
removes both industry and time e?ects from the
data, avoiding potentially spurious associations.
Since all variables in this speci?cation are industry
adjusted, ?rm size and analyst following directly
enter the regressions without the conversion to
dummy variables performed in earlier tests. We
exclude observations in the top and bottom 1% of
the empirical distribution of cost of capital (i.e.
two observations from each tail of the distribu-
tion) to ensure our results are not unduly a?ected
by extreme observations.
11
In accordance with our hypotheses outlined
earlier, we expect C
1
, C
3
, C
4
and C
5
to be nega-
tive and C
2
to be positive. As discussed in H3a(ii),
we expect C
6
to be statistically insigni?cant.
For comparative purposes, the results of esti-
mating Eqs. (5) and (6), explaining ?nancial and
social disclosure, respectively, using the industry-
adjusted data, are reported in Tables 6 and 7.
Of course, the industry membership variables
10
Other potential risk proxies such as market beta, market
value of equity and book to market ratios could also be con-
trolled for in the empirical analysis. We omit beta from the
analysis because Gebhardt et al. document that they are statis-
tically unrelated to our measure of the cost of equity capital
except in certain multivariate tests. We include size and book to
market ratios in all our equations explaining the cost of capital
and ?nd that these variables have the correct sign but are sta-
tistically unrelated to our industry adjusted cost of capital esti-
mates. Gebhardt et al. also ?nd that the dispersion in analysts’
earnings forecasts is signi?cantly related to the cost of capital.
We choose not to include this variable in our analysis for three
reasons. First, the dispersion in analysts’ forecasts would pre-
sumably be a function of the disclosure variables we include in
our analysis, so the inclusion of this variable would amount to
the inclusion of an alternative disclosure measure and could
hamper our ability to observe a relation between more direct
disclosure measures and the cost of capital. Second, including a
measure of analysts’ forecast dispersion would reduce our
sample size because a measure of dispersion requires that the
?rm be followed by multiple analysts Third, the dispersion in
analysts’ forecasts is related to the number of analysts follow-
ing the ?rm, which we include for consistency with Botosan.
11
Our conclusions are not a?ected by excluding these obser-
vations. In general, there is a slight increase in statistical sig-
ni?cance when the observations are dropped, suggesting these
observations contain measurement error.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 609
Table 6
Results of estimating Eq. (5) explaining variation in ?nancial disclosure—industry adjusted data
FDISC
it
¼ o þ[
1
SIZE
it
þ[
2
ROE
it
þ[
3
LEV
it
þ[
4
NANAL
it
þc ð5Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) À0.276 À0.524 0.6005
SIZE (+) 0.001 5.467 0.0001
ROE (+) 0.015 0.605 0.5458
LEV (+) 0.007 1.472 0.1420
NANAL (+) 0.508 5.291 0.0001
No. of observations=324; adjusted R
2
=23.8%. Variable de?nitions: FDISC=?nancial disclosure score for ?rm i, year t—the industry
sector median for year t; SIZE=beginning of year market value of equity for ?rm i, year t—the industry sector median for year t;
ROE=return on equity for ?rm i, year t—the industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry
sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the industry
sector median for year t.
Table 7
Results of estimating Eq. (6) explaining variation in social disclosure—industry adjusted data
SDISC
it
¼ , þ’
1
SIZE
it
þ’
2
ROE
it
þ’
5
NANAL
it
þ’
6
LEV
it
þu
it
ð6Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 1.148 2.682 0.0077
SIZE (+) 0.002 3.472 0.0006
ROE (?) 0.033 1.687 0.0936
NANAL (+) 0.172 2.196 0.0461
LEV (+) 0.009 2.003 0.0288
No. of observations=324; adjusted R
2
=8.4%. Variable de?nitions: SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t; SIZE=beginning of year market value of equity for ?rm i, year t—the industry sector median for year t;
ROE=return on equity for ?rm i, year t—the industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry
sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the industry
sector median for year t.
Table 8
Results of estimating Eq. (7) explaining variation in cost of capital estimates—industry adjusted data
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þj
it
ð7Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 0.00394 2.244 0.0258
NANAL (À) À0.00101 À3.459 0.0007
LEV (+) 0.00002 1.584 0.1147
FDISC (À) À0.00040 À1.961 0.0512
SDISC (À) 0.00064 2.354 0.0195
No. of observations=221; adjusted R
2
=9.9%. Variable de?nitions: COST=estimated cost of equity capital for ?rm i, year t—the
industry sector median for year t; NANAL=number of analysts providing a one-year ahead earnings forecast for ?rm i, year t—the
industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry sector median for year t; FDISC=?nancial
disclosure score for ?rm i, year t—the industry sector median for year t; SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t.
610 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
originally included in Eq. (6) explaining social
disclosure are omitted from this speci?cation
which already includes industry adjustment. As
the results in Table 6 show, the results of estimat-
ing Eq. (5), which explains ?nancial disclosure
variation, using the industry adjusted data, are
very similar to our earlier results. The only di?er-
ence is that the signi?cance of leverage is dimin-
ished such that it is no longer a signi?cant
explanatory variable at conventional levels. As
Table 7 reveals, the R
2
of the regression explaining
social disclosure falls dramatically (from 27 to
8%) when industry adjusted data are used and
industry related variables are excluded from the
regression. However, the explanatory power of the
remaining variables is very similar to that reported
in our earlier results.
The results of estimating Eqs. (7) and (8) are
reported in Tables 8 and 9, respectively.
12
The
number of analysts following the ?rm has a sta-
tistically reliable e?ect on cost of capital, with
higher analyst coverage resulting in a lower cost of
capital. Financial leverage is positively associated
with the cost of equity capital, but this e?ect is not
signi?cant at conventional levels. Financial dis-
closure is negatively related to cost of capital,
consistent with our predictions. This result is
stronger than the full sample result in Botosan,
perhaps suggesting that ?nancial disclosure plays
a more important role for Canadian ?rms. Sur-
prisingly, social disclosure exhibits a statistically
reliable positive association with the cost of equity
capital. For our sample ?rms and our time period,
enhanced social disclosure results in a higher cost
of capital.
Table 9 reports the results of estimating the
expanded Eq. (8). This speci?cation includes
interaction terms intended to determine if analyst
following modi?es the relation between our dis-
closure variables and the cost of capital. Con-
sistent with Botosan’s (1997) results, we ?nd that
?rms with low analyst following receive bene?ts
from expanded ?nancial disclosure in the form of
a reduction in the cost of equity capital. The
interaction of analyst following and social dis-
closure is not statistically signi?cant.
Table 9
Results of estimating Eq. (8) explaining variation in cost of capital estimates—industry adjusted data
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þ
[
5
LANAL
it
ÂFDISC
it
ð Þ þ[
6
LANAL
it
ÂSDISC
it
ð Þ þj
it
ð8Þ
Variable (pred. sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 0.00094 0.474 0.6362
NANAL (À) À0.00095 À3.258 0.0013
LEV (+) 0.00002 1.478 0.1408
FDISC (À) À0.00004 À0.169 0.8663
SDISC (À) 0.00078 2.517 0.0126
LANAL*FDISC (À) À0.00098 À2.289 0.0231
LANAL*SDISC (?) À0.00036 À0.567 0.5715
No. of observations=221; adjusted R
2
=12.98%. Variable de?nitions: COST=estimated cost of equity capital for ?rm i, year t—the
industry sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the
industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry sector median for year t; FDISC=?nancial
disclosure score for ?rm i, year t—the industry sector median for year t; SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t; LANAL=a dummy variable set equal to one if the number of analysts following ?rm i in year t is below i’s
industry sector median for year t, zero otherwise.
12
Recall that the sample sizes re?ected in Tables 8, 9, and 10
are 221, not the 324 re?ected in earlier tables. This re?ects the
additional data requirements to estimate the cost of capital
(analysts’ forecasts and industry ROE) and the deletion of
observations with cost of capital below the 1st percentile or
above the 99th percentile of the empirical distribution.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 611
4.5. Speci?cation checks and further analysis
White’s test for heteroskedasticity/model speci-
?cation fails to reject the null hypothesis of
homoskedasticity for all regressions. In addition,
the highest condition index reported in any of our
regressions is 3.6, suggesting that multicollinearity
is not a?ecting our results.
One potential problem with our data is that
each ?rm contributes up to three observations to
the estimation, and these observations may not be
independent. Accordingly, we estimate Eqs. (7)
and (8) on a year-by-year basis rather than on the
pooled sample. The results (not reported) are very
similar to those reported for the full sample. All
variables that are signi?cant in our pooled regres-
sions have consistent signs for each of the 3 years,
and are signi?cant in at least one, often two, of the
yearly regressions. This suggests that our results
are not due to our pooling of sample data.
We explore further our ?nding that social dis-
closure increases the cost of capital. We conjecture
that this result may be related to the poor eco-
nomic conditions that characterize our sample
period. While we do not have data available from
another, more prosperous time period, we conduct
an alternative test to assess this explanation for
our results. If the relationship between social dis-
closure and the cost of capital is mediated by eco-
nomic conditions, then we expect that, within our
sample, ?rms with above average ?nancial perfor-
mance would not experience an increase in the
cost of capital as social disclosure increases, while
those ?rms with below average ?nancial perfor-
mance would. We test this conjecture by estimat-
ing Eq. (9), which replaces the insigni?cant social
disclosure/analyst following interaction from Eq.
(8) with a social disclosure/return on equity inter-
action. For completeness, we also include return
on equity in the equation to test for a direct
impact of ?nancial performance on the cost of
equity capital:
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ
[
3
FDISC
it
þ[
4
SDISC
it
þ
[
5
LANAL
it
ÂFDISC
it
ð Þþ
[
6
DROE
it
ÂSDISC
it
ð Þ þ[
7
ROE
it
þj
it
ð9Þ
where ROE is equal to the return on equity for
?rm i, year t—the industry sector median for year
t, DROE is a dummy variable set equal to one if
?rm i, year t return on equity is above the industry
sector median for year t, zero otherwise, and all
other variables are as previously de?ned.
In keeping with the above discussion, we expect
C
6
to be negative and make no predictions about
C
7
.
13
The results of estimating Eq. (9) are pro-
vided in Table 10. Consistent with our conjectures,
the coe?cient on the social disclosure/?nancial
performance interaction, is reliably negative. This
coe?cient is almost exactly equal in absolute
magnitude (À0.00109) to the coe?cient on social
disclosure (0.00116). This indicates that there is
essentially no relation between social disclosure
and the cost of capital for ?rms with above aver-
age return on equity, but a signi?cant increase
in the cost of capital accompanying better social
disclosure for below average return on equity
?rms.
14
4.6. Sensitivity of results to cost of capital
estimation assumptions
We perform several alternative calculations of
the cost of capital by varying the assumptions
underlying that calculation. The primary results
13
Since this analysis is conducted in light of our earlier
empirical results, we conduct two-tailed tests of statistical sig-
ni?cance for this estimation.
14
The distribution of ROE reveals the existence of some
extreme negative values. There are no negative book value
observations in our sample, so these observations are not
caused by small negative denonimators. Since our ROE and
social disclosure interaction is based on a dummy variable for
ROE, this interaction term should not be impacted by these
observations. However, it is possible that the estimated coe?-
cient for the main e?ect of ROE is impacted by these observa-
tions, and that this a?ects the estimated coe?cient on the
interaction term. We conducted two additional tests to ensure
that our results are not sensitive to extreme observations of
ROE. We repeated the estimation of Eq. (9) after (1) eliminat-
ing all observations with ROE less than À25% (the 5th percn-
tile of the empirical distribution), and (2) retaining all
observations but replacing the continuous version of ROE with
the dummy variable (DROE) to test for the main e?ect of
ROE. Neither change in speci?cation qualitatively alters our
results. In particular, C
6
remains signi?cantly negative and C
7
remains insigni?cantly di?erent from zero in both tests.
612 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
reported above are based on the assumption that
the ROE of each ?rm fades to its historical indus-
try average ROE in a linear fashion between
years +4 and +12. We check for the sensitivity
of our results to this assumption by performing
two new calculations, one in which the linear fade
occurs between periods +4 and +6, and one in
which the fade occurs between +4 and +18. Our
primary empirical results are not sensitive to these
changes in speci?cation. The statistical sig-
ni?cance of most explanatory variables, including
the disclosure variables of primary interest,
increase with the length of the linear fade period,
perhaps suggesting that the cost of capital esti-
mates contain less noise as the linear fade period is
increased.
5. Conclusions and suggestions for future research
This study provides further evidence on the the-
oretical and regulatory premise that improved
corporate disclosure results in a lower cost of
capital. We ?nd that there is a signi?cant negative
relationship between the level of ?nancial dis-
closure and the cost of capital (H1, see Table 8).
We also con?rm Botosan’s (1997) ?nding that
higher levels of ?nancial disclosure can reduce the
cost of capital in cases where there is low ?nancial
analyst following [H3a(i), see Table 9]. Our
results, however, suggest that this relation does
not hold for social disclosures. There is a statisti-
cally signi?cant, positive relation between the level
of social disclosure and the cost of capital, that is,
more social disclosure raises the cost of capital for
the ?rm (H2, see Table 8). The number of analysts
following the ?rm does not a?ect this result
[H3a(ii), see Table 9]. The positive relationship
between cost of capital and social disclosure is
moderated by the return-on-equity of the ?rm
with more successful ?rms being less penalized for
social disclosures.
It is important to recognize that these results are
not based on the content of the disclosures. The
disclosure scores re?ect the completeness and
informativeness of ?nancial and social disclosures
but they do not indicate whether the information
is good or bad news. There are several possible
Table 10
Results of estimating Eq. (9) explaining variation in cost of capital estimates—industry adjusted data
COST
it
¼ , þ[
1
NANAL
it
þ[
2
LEV
it
þ[
3
FDISC
it
þ[
4
SDISC
it
þ
[
5
LANAL
it
ÂFDISC
it
ð Þ þ[
6
DROE
it
ÂSDISC
it
ð Þ þ[
7
ROE
it
þj
it
ð9Þ
Variable (pred. Sign) Coe?cient estimate t-Statistic P-value (two-tailed)
Intercept (?) 0.00130 0.667 0.5058
NANAL (À) À0.00091 À3.221 0.0015
LEV (+) 0.00003 1.863 0.0639
FDISC (À) À0.00007 À0.284 0.7771
SDISC (À) 0.00116 3.462 0.0006
LANALÂFDISC (À) À0.00096 À2.663 0.0083
DROEÂSDISC (À) À0.00109 À2.469 0.0143
ROE (?) 0.00009 1.101 0.2719
No. of observations=221; adjusted R
2
=15.58%. Variable de?nitions: COST=estimated cost of equity capital for ?rm i, year t—the
industry sector median for year t; NANAL=number of analysts providing a 1-year ahead earnings forecast for ?rm i, year t—the
industry sector median for year t; LEV=debt to equity ratio, ?rm i, year t—the industry sector median for year t; FDISC=?nancial
disclosure score for ?rm i, year t—the industry sector median for year t; SDISC=social disclosure score for ?rm i, year t—the industry
sector median for year t; ROE=return on equity for ?rm i, year t—the industry sector median for year t; LANAL=a dummy variable
set equal to one if the number of analysts following ?rm i in year t is below i’s industry sector median for year t, zero otherwise;
DROE=a dummy variable set equal to one if return on equity for ?rm i, year t is above the industry sector median for year t, zero
otherwise.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 613
explanations for the results on the relationship
between social disclosure and cost of capital. First,
if there were a consistent bias in social disclosures
where ?rms that experience higher than average
social costs disclose more information, then, on
average, the results reported could hold. The
descriptive literature on social disclosures has
reported that there are severe biases in reporting
(e.g. Guthrie & Parker, 1990; Wiseman, 1982).
The reported bias is that ?rms tend to use social
disclosures for self-promotion. This means that
that ?rms tend to report the positive social con-
tributions that they make but under-report nega-
tive social e?ects. Unfortunately, the relationship
between this bias and the costs incurred and ben-
e?ts received by the ?rm is unclear (Richardson et
al., 1999).
Second, the results could hold if two things are
true. It may be that social responsibility invest-
ments by ?rms are consistently negative present
value projects increasing the overall risk of the
?rm. While proponents of socially responsible
corporate behavior point to the potential cost
savings and long-term strategic advantage of such
behavior (e.g. Porter & van der Linde 1995; Scal-
tegger & Figge, 1998), the market may hold a dif-
ferent view. If this is the case and there is a positive
correlation between social disclosures and social
responsibility actions, then the observed results
could hold.
It is also possible that the results are speci?c to
the data used in this analysis. The time period for
which data were available was an economic reces-
sion. The positive link between social disclosure
and the cost of equity capital we document may be
contingent on these macro economic conditions.
This interpretation is supported by our result that
the e?ect of social disclosure on the cost of capital
is moderated by return on equity (see Table 10).
For ?rms with ROE above the industry median,
there is a negative interaction between social dis-
closure and return on equity that o?sets the main
e?ect between cost of capital and social disclosure.
In other words, for ?rms with above average ?nan-
cial performance, social disclosure may have no
e?ect on the cost of capital. If this is a reasonable
proxy for economic conditions, then the relation-
ship between social disclosure and cost of capital
may be limited to periods of economic downturn.
Future research that expands the data set over a
complete business cycle to test for this e?ect is
clearly called for by our results.
The e?ect of social disclosure on the cost of
equity capital documented in this study should not
be taken to imply that social disclosure has an
overall negative e?ect on the ?rm. Social and
environmental issues have signi?cant distribu-
tional e?ects. Although investors may require a
higher cost of capital for ?rms with a signi?cant
social agenda, other groups such as employees,
customers, regulators and supply-chain partners
may provide greater support to the ?rm because of
these actions. Much of the work on developing a
social accounting agenda has, in fact, been pre-
mised on the assumption that corporate social
disclosure will bene?t a broader community of
stakeholders than the capital providers that are
the primary audience for ?nancial disclosures (e.g.
Gray, 2000). The e?ect of social disclosures on the
expected cost of contributions to the ?rm from
other stakeholders has still to be examined.
Acknowledgements
The authors gratefully acknowledge the ?nan-
cial support of the Certi?ed General Accountants
of Canada Research Foundation and comments
on earlier drafts by Irene Gordon, Marc Epstein
and an anonymous reviewer. The authors also
gratefully acknowledge the contribution of I/B/E/
S International Inc. for providing earnings per
share forecast data, available through the Institu-
tional Brokers Estimate System. These data have
been provided as part of a broad academic pro-
gram to encourage earnings expectations
research.
Appendix A
Panel 1: Social disclosure categories of informa-
tion, maximum number of points allocated to the
category, and number of sub-categories allocated
to each category—information provided in the
1991 SMAC/UQAM Report:
(continued on next page)
614 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
Category of
information
Maximum
points
No. of sub-
categories
Human resources 18 29
Products, services,
and consumers
10 23
Community 18 24
Environment 18 22
Energy resources 6 10
Governments 10 14
Suppliers 6 16
Shareholders 6 9
Competitors 4 9
Miscellaneous 4 14
Totals 100 170
Panel 2: Financial disclosure categories of
information, maximum number of points allo-
cated to the category, and number of sub-cate-
gories allocated to each category—information
provided in the 1991 SMAC/UQAM Report.
Category Maximum
points
No. of sub-
categories
General information
Financial retrospective 10 7
Financial forecasts 10 11
Graphs and tables 10 4
Points of view on
regulations, competition
and economy
5 23
Past performance
and highlights
7 25
Future prospects 5 12
Investment and
disinvestment
5 22
Research, development
and environment
5 14
Risks and uncertainties 5 9
Chairman’s (sic) report 2 4
Financial statements
Sector Information 10 34
Income Statement 10 10
Accounting policies 5 7
Other notes to
?nancial statements
7 13
Pension plans
and leases
3 37
Government aid
and income tax
3 8
Exports and portion
of products
manufactured in Canada
3 8
Information on e?ects
of price ?uctuations
3 8
Special notes on total
quality, environment,
training, and technology
6 2
Miscellaneous 6 3
Totals 120 261
References
Botosan, C. (1997). Disclosure level and the cost of equity
capital. The Accounting Review, 72, 323–349.
Botosan, C., M. Plumlee. 2000. Disclosure level and expected
cost of equity capital: An examination of analysts’ rankings
of corporate disclosure. Working paper, University of Utah,
January, 2000.
Clarkson, P., Guedes, J., & Thompson, R. (1996). On the
diversi?cation, observability, and measurement of estimation
risk. Journal of Financial and Quantitative Analysis (March),
69–84.
Committee on Corporate Disclosure, 1995. Toward Improved
Disclosure: A Search for Balance in Corporate Disclosure.
The Toronto Stock Exchange.
Deegan, C., & Gordon, B. (1996). A study of environmental
disclosure practices of Australian corporations. Accounting
and Business Research, 26, 187–199.
Diamond, D., & Verrecchia, R. (1991). Disclosure, liquidity
and the cost of equity capital. Journal of Finance (Septem-
ber), 1325–1360.
Downing, P. (1997). Upping the stakes. CA Magazine, (June
and July), 41–43.
Edwards, E., & Bell, P. (1961). The Theory and measurement of
business income. Berkeley, CA: University of California Press.
Epstein, M. 2000. The identi?cation, measurement and report-
ing of corporate social impacts—revisited after twenty-?ve
years. Paper presented to the Accounting, Organizations and
Society 25th Anniversary Conference, Oxford, UK, July.
Feltham, G., & Ohlson, J. (1995). Valuation and clean surplus
accounting for operating and ?nancial activities. Con-
temporary Accounting Research, (Spring), 689–731.
A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616 615
Garrett, D. E. (1987). The e?ectiveness of marketing policy
boycotts: environmental opposition to marketing. Journal of
Marketing, 51(April), 46–57.
Gray, R. 2000. The social accounting project and Accounting,
Organizations and Society: privileging engagement, imagin-
ings, new accountings and pragmatism over critique? Paper
presented to the Accounting, Organizations and Society 25th
Anniversary Conference, Oxford, UK, July.
Gebhardt, W., Lee, C. M. C., & Swaminathan, B. (2000).
Toward an implied cost of capital. Working paper, Cornell
University, October 27, 2000. Journal of Accounting Research
(in press).
Gibbins, M., Richardson, A. J., & Waterhouse, J. (1990). The
management of ?nancial disclosure: opportunism, ritualism,
policies and processes. Journal of Accounting Research, 28(1),
121–143.
Guthrie, J., & Parker, L. D. (1990). Corporate social disclosure
practice: a comparative international analysis. Advances in
Public Interest Accounting, 3, 159–175.
Hannan, M. (1998). Rethinking age-dependence in organiza-
tional mortality: logical formalizations. American Journal of
Sociology, 104(1), 126–164.
Healy, P., Hutton, A., & Palepu, K. (1999). Stock performance
and intermediation chanes surrounding sustained increase in
disclosure. Contemporary Accounting Research, 11(2), 485–
520.
Lang, M., & Lundholm, R. (1993). Cross-sectional determi-
nants of analyst ratings of corporate disclosures. Journal of
Accounting Research, 246–271.
Lang, M., & Lundholm, R. (1996). Corporate disclosure
policy and analyst behavior. The Accounting Review, 71,
467–492.
Levitt, A. 1999. Quality Information: The Lifeblood of Our
Markets. The Economic Club of New York, NewYork, N.Y.,
October 18, 1999.http://www.sec.gov/news/spchindx.htm.
Menon, A., & Menon, A. (1997). ‘‘Enviropreneurial’’ market-
ing strategy: the emergence of corporate environmentalism as
market strategy. Journal of Marketing, 6, 51–67.
Ohlson, J. (1995). Earnings, book value, and dividends in security
valuation. Contemporary Accounting Research, 661–687.
Oliver, C. (1991). Strategic responses to institutional processes.
Academy of Management Review, 16(1), 145–172.
Patten, D. (1991). Exposure, legitimacy and social disclosure.
Journal of Accounting and Public Policy, 10, 297–308.
Pava, M., & Krausz, J. (1996). The association between cor-
porate social responsibility and ?nancial performance: the
paradox of social cost. Journal of Business Ethics, 15, 321–
357.
Porter, M., & van der Linde, Claas (1995). Green and com-
petitive: ending the stalemate. Harvard Business Review,
September–October.
Reyes, M. G., & Grieb, T. (1998). The external performance of
socially responsible mutual funds. American Business Review
(January), 1–7.
Richardson, A., Welker, M., & Hutchinson, I. (1999). Manag-
ing capital market reactions to corporate social responsibility.
International Journal of Management Reviews, 1, 17–43.
Scaltegger, S., & Figge, F. ‘‘Environmental shareholder value’’
Centre for economics and business administration, Uni-
versity of Basle WWW-study no. 54, June 1998.
Sengupta, P. (1998). Corporate disclosure and the cost of debt.
The Accounting Review, 73(4), (October), 459–474.
Stinchcombe, A. L. (1965). Social structure and organizations.
In J. G. March (Ed.), Handbook of organizations. Chicago:
Rand-McNally.
Watts, R., & Zimmerman, J. (1996). Positive accounting theory.
Englewood Cli?s: Prentice-Hall.
Welker, M. (1995). Disclosure policy, information asymmetry
and liquidity in equity markets. Contemporary Accounting
Research (Spring), 801–827.
Wiseman, J. (1982). An evaluation of environmental dis-
closures made in annual reports. Accounting, Organizations
and Society, 7(1), 53–63.
Zeghal, D., & Ahmed, S. A. (1990). Comparison of social
responsibility information media used by Canadian ?rms.
Auditing, Accounting and Accountability Journal, 3(1), 38–53.
616 A.J. Richardson, M. Welker / Accounting, Organizations and Society 26 (2001) 597–616
doc_345576708.pdf