The Role of Accounting Conservatism in Mitigating Bondholder-Shareholder Conflicts over

Dividend Policy and in Reducing Debt Costs

Author(s): Anwer S. Ahmed, Bruce K. Billings, Richard M. Morton and Mary StanfordHarris
Source: The Accounting Review, Vol. 77, No. 4 (Oct., 2002), pp. 867-890
Published by: American Accounting Association
Stable URL: http://www.jstor.org/stable/3068874
Accessed: 29-11-2016 05:25 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

American Accounting Association is collaborating with JSTOR to digitize, preserve and extend access to
The Accounting Review

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

THE ACCOUNTING REVIEW

Vol. 77, No. 4

October 2002
pp. 867-890

The Role of Accounting

Conservatism in Mitigating
Bondholder-Shareholder Conflicts
over Dividend Policy and in

Reducing Debt Costs

Anwer S. Ahmed
Syracuse University

Bruce K. Billings

Richard M. Morton
Florida State University

Mary Stanford-Hars
Texas Christian University
ABSTRACT: Using both a market-based and an accrual-based measure of

conservatism, we find that firms facing more severe conflicts over dividend
policy tend to use more conservative accounting. Furthermore, we document
that accounting conservatism is associated with a lower cost of debt after
controlling for other determinants of firms' debt costs. Our collective evidence
is consistent with the notion that accounting conservatism plays an important
role in mitigating bondholder-shareholder conflicts over dividend policy, and
in reducing firms' debt costs.

Keywords: accounting conservatism; dividend policy conflicts; cost of debt.

Data Availability: All data used in the study are publicly available from the
sources identified in the text.
We thank Linda Bamber, Christine Botosan, Mark Bradshaw, Tom Dyckman, John Elliott, Dave Harris,
Sriprakash P. Kothari, Charles Lee, Gerry Lobo, Karen Nelson, Terry Shevlin, Doug Stevens, Sam Tiras, Senyo
Tse (the associate editor), Ross Watts, Jerry Zimmerman, two anonymous reviewers, and workshop participants at
the State University of New York at Buffalo, Cornell University, Florida State University, Syracuse University, the
11th Annual Finance and Economics Conference at Michigan, the 2001 AAA Annual Meeting, and the First Annual
Winter Accounting Conference at Utah, for their comments.
Submitted March 2001
Accepted May 2002

867

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

868 The Accounting Review, October 2002

I. INTRODUCTION

W x Tatts (1993) suggests that conservatism likely evolved from the contracting role
of accounting.' He argues that accounting conservatism helps avoid inappropriate
distributions to claim holders. Consistent with Watts (1993), we argue that conservatism mitigates conflicts of interest over dividend policy between shareholders and
bondholders.2 Specifically, we hypothesize that (1) firms experiencing more severe bondholder-shareholder dividend policy conflicts adopt more conservative accounting and (2)
firms that use more conservative accounting incur a lower debt cost. The intuition for our
hypotheses is as follows. Conservative accounting reduces the earnings and retained earn-

ings amounts used in debt contracts to constrain dividends.3 Thus, choosing more conservative accounting is one way the firm can reduce the risk to bondholders that it will pay
excessive dividends to shareholders. This risk reduction is likely to be more important when
bondholder-shareholder conflicts over dividend policy are more severe. Moreover, we expect that if managers choose more conservative accounting, then bondholders will accept
a lower rate of return in light of the reduced risk of dividend overpayment.

We test these hypotheses using both "levels" and "changes" analyses based on data
from two six-year periods: 1993-1998 and 1987-1992. We use two alternative measures

of conservatism: a market-value-based measure and an accrual-based measure. Our first

measure of conservatism, based on Beaver and Ryan (2000), is the component of the bookto-market ratio that reflects biased accounting due to earlier recognition of expenses and

losses, and to deferred revenue recognition. This component captures the extent to which
book value is persistently lower than market value. Beaver and Ryan (2000) demonstrate

(and we confirm) that this component of book-to-market is associated with conservative

accounting method choices.

Our second measure of conservatism is based on Givoly and Hayn (2000), who argue
that conservative accounting leads to persistently negative accruals. In the long run, un-

biased accounting results in the cumulative amount of net income before depreciation and
amortization converging to operating cash flows. Thus, as both positive and negative accruals reverse over time, net cumulative accruals should approach zero. In contrast, con-

servative accounting results in a persistent pattern of negative accruals over time. This
suggests that a firm's mean accrual over a reasonably long period provides an accountingbased, firm-specific proxy for conservatism.

Because we cannot directly observe the extent to which managers voluntarily make
conservative accounting choices, our two proxies for conservatism may reflect GAAPmandated conservatism as well as managers' choices of accounting methods, estimates, and
assumptions. However, GAAP-mandated conservatism likely affects firms in the same industry in similar ways. Thus, to control for GAAP-mandated conservatism, we repeat our

Statement of Concepts No. 2 (FASB 1980) defines conservatism as "a prudent reaction to uncertainty to try to
ensure that uncertainty and risks inherent in business situations are adequately considered." According to Statement of Concepts No. 6, (FASB 1985, para. 44), the motivation for expensing certain intangibles or not recognizing gains is uncertainty about whether the items qualify as assets or liabilities.
2 Conservatism may indirectly affect other conflicts. For example, restrictions on dividend policy may lead borrowers to retain more funds, which would in turn reduce the underinvestment problem. Similarly, conservatism
reduces book value of equity, which places constraints on financing policy via more conservative leverage ratios.
We leave more complete analysis of these other conflicts for future research.
Penman and Zhang (2002) point out that an exception to this generalization occurs when firms that use conser-

vative accounting methods reduce their operating assets. As Penman and Zhang (2002) show, a reduction in

operating assets can increase current period income due to the liquidation of hidden reserves. To the extent that
reductions in operating assets inflate earnings, this biases against our hypotheses, which are based on the assumption that conservative accounting reduces income.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 869

analyses using industry-adjusted measures of conservatism and firm characteristics. Furthermore, by using each firm as its own control, our "changes" analysis controls for any

effects of GAAP-mandated conservatism that vary across firms.

The results of our analyses are for the most part consistent with our first hypothesis,

that firms with more severe bondholder-shareholder dividend policy conflicts, on average,
use more conservative accounting. The results are also strongly consistent with our second
hypothesis: firms that adopt more conservative accounting practices incur lower debt costs,
on average. Our results hold after controlling for other factors (e.g., profitability, firm size,
firm growth) that are potentially correlated with our proxies for conservatism and debt cost.
The study's results contribute to our understanding of the real economic effects of
accounting conservatism in two ways. First, we provide evidence that accounting conservatism mitigates conflicts over dividend policy. Although prior studies have documented
the presence of conservatism in debt contracts, we believe this is the first study that provides
evidence suggesting that conservatism plays an important role in efficient debt contracting.
Second, by documenting that conservatism is associated with a lower cost of debt, our
results suggest that accounting choices can significantly affect firms' cost of capital. In
other words, firms that choose to build a reputation for conservative accounting appear
to reap real economic benefits. Our evidence that use of conservative accounting leads to
lower cost of debt extends Botosan's (1997) and Sengupta's (1998) conclusions that higher
quality disclosures lead to lower cost of equity and lower cost of debt, respectively.
The paper proceeds as follows. Section II builds on prior research to develop the
hypotheses. Section HI describes our empirical proxies for conservatism, debt cost, and
bondholder-shareholder conflicts over dividend policy. Section IV describes our research
design. Section V describes the sample and presents the results, and Section VI concludes
the paper.

II. PRIOR RESEARCH AND HYPOTHESES

Prior Research

Considerable controversy remains whether accounting conservatism yields any real economic benefit to firms (Devine 1963; Sterling 1967). For example, Hendriksen and Van
Breda (1992, 149) state, "conservatism is, at best, a very poor method of treating the
existence of uncertainty in valuation and income. At its worst, it results in a complete
distortion of accounting data." Similarly, Penman and Zhang (2002) suggest that conservatism results in lower quality earnings. Intuitively, conservative accounting practices accelerate the expensing of investment costs, thus reducing income and generating hidden
reserves, as long as the firm is increasing investment expenditures. Consequently, conservative accounting leads to lower cumulative reported earnings. However, when the firm
reduces its investment expenditure levels, those same conservative accounting practices
liquidate hidden reserves, which in turn leads to higher current period income. Thus, conservative accounting leads to lower cumulative reported earnings, but the effect on current
period earnings depends on investment growth.
Although many prior studies examine the relation between accounting choice and debt
covenants, the more general role of conservatism in debt contracting has not received much
attention. Prior research documents that debt contracts often require use of conservative
accounting methods (Leftwich 1983; El-Gazzar and Pastena 1990; Leuz et al. 1998). For
example, Leftwich (1983) finds that modifications to GAAP in debt contracts are typically
conservative. However, there is no direct evidence on the role that accounting conservatism
plays in mitigating bondholder-shareholder conflicts, or on the association between conservatism and a firm's cost of debt.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

870 The Accounting Review, October 2002

Role of Conservatism in Mitigating Dividend Policy Conflicts

Agency theory suggests that fixed and residual claimants in a firm have conflicting

interests over dividend policy (Jensen and Meckling 1976). In particular, overpayment of
dividends can transfer wealth from bondholders to shareholders by reducing the assets
available for meeting bondholders' fixed claims and hence increasing the default risk for
bondholders. Efficient contracting theory suggests that it is in the interests of all parties to
mitigate these conflicts. To address this conflict, bondholders typically include dividend
policy restrictions in debt contracts (Black 1976; Smith and Warner 1979; Kalay 1982;
Leftwich 1983; Healy and Palepu 1990; El-Gazzar and Pastena 1990, 1991).
Kalay (1982) finds that firms use two types of dividend restrictions in their public debt
contracts: direct and indirect. Direct restrictions specify an upper bound on dividends in
terms of cumulative net earnings or retained earnings (Begley 1994). Kalay (1982) documents that 90 percent of his random sample of 150 firms, selected from the Moody's
Industrial Manual, report a direct dividend restriction. Indirect restrictions constrain dividends indirectly by requiring the firm to maintain certain balance sheet ratios. For example,
a cap on the debt-to-asset ratio constrains dividends because dividend payments reduce
assets, and thus move the debt-to-asset ratio closer to its upper bound. We assume that our
sample firms either have or expect to have dividend restrictions in their debt contracts.
Measurement of accounting numbers used in specifying restrictions on dividend policy
plays an important role in debt contracting. Conservative accounting directly affects the

earnings and retained earnings amounts used in debt contracts to constrain dividends, in
that conservative accounting leads to lower (cumulative) reported earnings. Similarly, con-

servative accounting implies tighter restrictions on dividend policy by reducing the measures
of assets in various balance sheet ratios that appear in debt contracts. In short, when debt
contracts impose accounting-based restrictions on dividends, more conservative accounting
reduces the likelihood that the firm will make excessive dividend payments to shareholders.

Debt contracts incorporate conservatism in at least two ways. First, bondholders can
explicitly require the use of conservative accounting. Second, managers might implicitly

commit to consistently use conservative accounting in order to build a reputation for conservative financial reporting. Milgrom and Roberts (1992) argue that reputational concerns

may effectively check managers' willingness to renege on such a commitment by engaging

in ex post opportunistic accounting choices.
In sum, we expect that when bondholder-shareholder conflicts over dividend policy are
potentially more severe, borrowing firms are likely to use more conservative accounting.
The borrowing firms' managers' willingness to choose more conservative accounting will
also depend on the costs associated with increased conservatism. However, holding the
costs of conservatism constant, we expect cross-sectional differences in conservatism to be
related to the severity of bondholder-shareholder conflicts over dividend policy. Our first
hypothesis (stated in alternate form) is as follows:

Hi: Firms that face more severe bondholder-shareholder conflicts over dividend policy
adopt more conservative accounting, ceteris paribus.
Conservatism and the Cost of Debt
As discussed above, managers' choices of more conservative accounting tighten restrictions on dividend policy, decreasing the likelihood of excessive dividend payments. Thus,
bondholders are likely to require a lower rate of return to compensate for the reduced risk
of excessive dividend payments. Therefore, we expect conservatism to reduce a borrower's
cost of debt. Our second hypothesis (stated in alternate form) is as follows:

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 871

H2: Firms that adopt more conservative accounting incur a lower cost of debt, ceteris

paribus.
Our tests of these hypotheses assume that firms' investment opportunities drive their
financing and dividend policies. Given the level of financing required to implement a firm's
investment opportunities, we assume that managers wish to minimize the firm's cost of

debt. Borrowing firms can reduce their cost of debt by accepting tighter restrictions on their
ability to pay dividends, via conservative accounting. We recognize that borrowing firms
might use other mechanisms to address the dividend policy conflict. For example, secured
debt would ensure that specific assets would be available to meet bondholder claims, thereby
obviating the need for conservative accounting to mitigate this conflict. However, secured
debt also has disadvantages, such as reducing managerial flexibility regarding the disposal
of secured assets, and exposing bondholders to the risk of decline in the asset values. If
secured debt is a widely used alternative, then it will most likely bias against finding support
for our predictions. We leave to future research the investigation of other ways borrowers
can mitigate dividend policy conflicts.
The issue then arises as to what prevents ex post departures from conservatism, and
how bondholders detect and respond to such departures. One disincentive to deviate from
conservatism is that building a reputation for conservatism is costly in terms of reporting
relatively lower firm performance. Departing from conservatism risks the loss of reputation
(Milgrom and Roberts 1992). Debt-rating agencies periodically reevaluate firms' creditworthiness, so deviations from conservatism could increase interest costs on future borrowings. Furthermore, once the borrowing firm's reputation is damaged, bondholders are unlikely to view the firm favorably if and when it returns to the bond market. Thus, managers
of firms that expect to issue more debt are likely to find it costly to deviate from
conservatism.

This efficient-contracting perspective contrasts with two alternative hypotheses about

accounting choice. The first alternative is the traditional debt hypothesis, in which managers
have ex post incentives to choose aggressive accounting in order to avoid covenant violations. However, Mohrman (1996) shows that more recent debt contracts generally specify
accounting methods, thereby restricting managers' ability to change accounting methods ex
post. The second alternative hypothesis is that managers could choose aggressive accounting
thereby increasing income in an attempt to appear less risky to prospective bondholders.
This hypothesis assumes that bondholders cannot see through managers' accounting decisions, and thus are fooled by the appearance of less risky financial statements.

Ill. EMPIRICAL PROXIES

In this section, we first describe our proxies for accounting conservatism, as well as
their limitations. We then discuss our proxies for the severity of bondholder-shareholder
conflicts over dividend policy, and we conclude the section with a discussion of our proxy
for the cost of debt.

Accounting Conservatism

Although conservatism broadly affects book value of equity and net income, accountants often focus on narrower forms of conservatism. For example, the choice of accelerated
depreciation as opposed to straight-line, the immediate expensing of research and development expenditures per SFAS No. 2 (FASB 1974), or the lower-of-cost-or-market rule for
inventory valuation are conservative accounting method choices. However, our study relies

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

872 The Accounting Review, October 2002

on a broader conservatism construct that reflects the cumulative effect of managers' accounting choices, including accounting estimates and assumptions, as well as accounting
method choices.

Relying on the notion that conservative accounting generally reduces (1) book value

of equity relative to market value of equity and (2) income relative to operating cash flows,
we use two alternative measures of conservatism: a market-value-based measure following
Beaver and Ryan (2000), and an accrual-based measure following Givoly and Hayn (2000).
These measures reflect the aggregate effects of conservatism at a firm level, as opposed to
an account or line-item level within the firm.4

Market-Value-Based Measure of Conservatism

Beaver and Ryan (2000) argue that variation in book-to-market ratios is a function of
two components: biased accounting recognition and lagged accounting recognition. They
suggest that the bias component of book-to-market reflects persistent differences between
book and market values resulting, in part, from conservative accounting. This notion of
conservatism-book values persistently below market values-is consistent with that in
Feltham and Ohlson (1995). Beaver and Ryan (2000) also suggest that the lag component
reflects temporary differences in book and market values due to the lag with which accounting recognizes unexpected gains and losses.
Beaver and Ryan (2000) empirically measure the bias (conservatism) and lag components by regressing book-to-market ratios onto current and lagged returns with fixed firm
and time effects. The firm-specific effect reflects the persistent bias component primarily
due to conservatism, the time-specific effect captures economy-wide temporal variation,
and the lag component reflects market shocks not yet recognized in book value.
Following Ryan (1995) and Beaver and Ryan (2000), we measure conservatism based

on the firm-specific coefficient, ai, in the following model:

6

BTMit = (x + (Xi + (Xt +1 Pok RETit-k+ eit

k=O

where:

BTMKt = the book-to-market ratio for firm i at fiscal year-end t;

a = the intercept across all firms and years;

ai = the persistent firm-specific bias component of book-to-market ratio over the

sample period;

at= the year-specific component of book-to-market ratio across all firms; and
RETit = the stock return (with dividends) for firm i in year t.
We perform the fixed-effects estimation separately on panel data for the two sample

periods 1993-1998 and 1987-1992. The coefficient, ai, reflects the persistent firm-specific
component of book-to-market over each six-year period relative to the other firms in the

sample. As a component of book-to-market, ai measures conservatism inversely (i.e., ai

I Basu (1997) characterizes conservatism in terms of asymmetric recognition of good news vs. bad news in financial
statements. We do not use Basu's (1997) measure of conservatism for two reasons. First, the relation between
his measure and the conservatism of accounting choices is unclear. Shon (2000) concludes that firms that use
more conservative accounting methods appear to be less conservative based on the Basu measure. Second, Muller
and Riedl (2001) show that inferences based on the Basu (1997) measure could be an artifact of differences in
variances of good news returns vs. bad news returns.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 873

decreases as market value increases relative to book value). For ease of presentation we

multiply oti by -1 and refer to this market-based proxy for conservatism as CONMKT. W
expect more conservative accounting to yield higher measures of CONMKT, thus Hi predicts a positive relation between CONMKT and proxies for bondholder-shareholder conflicts over dividend policy. In addition to reflecting conservatism, CONMKT is also likely
to reflect growth opportunities because market values reflect growth opportunities (unrecorded assets). We discuss controls for these other determinants of CONMKT in Section
IV.

Accrual-Based Measure of Conservatism

Givoly and Hayn (2000) focus on the income-statement effects of conservatism over

extended periods. They argue that conservative accounting leads to persistently negative
accruals, in contrast with the expected pattern of accrual reversals. This suggests that a
firm's mean accrual over a reasonably long period provides a firm-specific proxy for conservatism. Consistent with Givoly and Hayn (2000), we use accruals before depreciation
as a proxy for conservative accounting.5 Specifically, we use the sum of total accruals
excluding depreciation (net income before extraordinary items plus depreciation expense
less operating cash flows) deflated by assets and averaged within each of the two sample
periods, 1993-1998 and 1987-1992, respectively, as our accrual-based proxy for conservatism.6 We also multiply this measure by -1 so it is increasing in the amount of negative
accruals, and refer to it as CONACC. We expect more conservative accounting to yield
higher measures of CONACC.
CONACC includes changes in current assets and current liabilities, deferred taxes,
pension accruals, asset write-downs, gains or losses on sale of assets, bad debt provisions,
restructuring accruals, and other accrued or deferred expenses and revenues. Managers can
exercise discretion over the conservatism with which they record most of these items in
terms of the amount and timing. However, these accruals also reflect economic characteristics unrelated to conservatism, such as growth in operations (Givoly and Hayn 2000). We
discuss controls for these other determinants of CONACC in Section IV.
Discussion of Conservatism Measures

Our two conservatism measures reflect the relative degree of conservatism across individual firms from two unique perspectives. CONMKT is a stock measure that reflects the
cumulative effect of conservative accounting on book value of equity. CONACC, on the

other hand, is a flow measure that reflects the effects of conservative accounting in
the periods over which we calculate the flow measure.
As described in Section II, our hypotheses focus on the extent to which borrowing
firms' managers voluntarily choose conservative accounting methods and estimates. However, our conservatism proxies do not directly measure the conservatism of managers' accounting choices. Thus, our conservatism proxies do not distinguish between the nondiscretionary (GAAP-mandated) component of conservatism and the discretionary component
Depreciation accruals represent the reversal of a positive accrual when the asset is acquired (i.e., net income
exceeds cash flows). Givoly and Hayn (2000) exclude the depreciation accrual because the originating positive
accrual is not captured in the difference between income and operating cash flow. The use of accruals before
depreciation ignores any conservatism in a firm's depreciation calculation, which biases against our hypotheses.

6 Consistent with the implications of Givoly and Hayn (2000), aggregating accruals over an extended time period
should better reveal firm-specific conservatism in accruals. We use six-year estimation periods for comparability
with CONMKT. We repeat our tests with three-year estimation periods for this proxy. All results are qualitatively
similar to those reported in the tables. Periods shorter than three years, in our view, would be less appropriate
for measuring conservatism as persistently negative accruals. We also repeat our analyses using net income after
extraordinary items, and all inferences remain the same.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

874 The Accounting Review, October 2002

of conservatism. There are several reasons why the nondiscretionary component of conservatism is unlikely to significantly affect our inferences.
First, even when GAAP appears to mandate conservatism, managers nonetheless have

considerable discretion in its application. For example, although GAAP requires immediate
expensing of R&D, managers can exercise discretion over the level and timing of R&D
spending. Similarly, GAAP requires firms to use the lower-of-cost-or-market rule to value
inventory, but managers must exercise discretion in estimating "market" value of inventory

and determining the appropriate level of aggregation to apply the test. Thus, managers
exercise discretion over the application of even GAAP-mandated conservative accounting

principles.
Second, to the extent that certain GAAP requirements tend to be conservative, these
standards generally apply to all firms. We have no a priori reason to expect that GAAPmandated conservatism will be higher for firms with potentially more severe bondholdershareholder conflicts over dividend policy, which could confound our analysis of the hypothesized relation between the conservatism of managers' discretionary choices and

dividend policy conflicts.7 Furthermore, it is unlikely that conservative GAAP result from
standard-setters' attempts to ameliorate conflicts among different stakeholders.8
Third, the effect of mandated conservatism likely varies systematically across industries.
For example, the level of R&D expenditures is higher in some industries (e.g., pharmaceuticals) than in others. Inventory write-downs also tend to cluster by industry (e.g., computer
technology). To provide further assurance that our results are not a manifestation of mandated conservatism rather than the conservatism of managers' discretionary choices, we
repeat our analysis using industry-adjusted conservatism measures and explanatory variables. Because results based on these industry-adjusted measures are similar to the results

using unadjusted measures, it is unlikely that cross-sectional variation in mandated conservatism drives our results.

Finally, our "changes" analysis that uses each firm as its own control also controls for
variation in GAAP-mandated conservatism across firms. Our consistent results using both

levels and changes specifications provide further assurance that our results are not driven
by GAAP-mandated conservatism.
Penman and Zhang (2002) conclude that conservative accounting can cause current

period income to increase in periods of declining operations, due to the liquidation of hidden
reserves (e.g., the LIFO reserve). To the extent that a firm has high income (relative to
operating cash flows) due to such liquidation of hidden reserves, our CONACC measure

will not classify the firm as using conservative accounting. This potential error in our
measurement of accounting conservatism is unlikely to affect our inferences. First, our
conservatism measures are averaged over six-year periods. This averaging mitigates any
temporary effects on current income induced by temporary changes in investment levels.
Second, we obtain identical inferences when we repeat our empirical tests after dropping
firms with declining operations, proxied by negative sales growth.
An alternative approach would measure conservatism using specific accounting choices,
such as LIFO vs. FIFO, or accelerated vs. straight-line depreciation methods. We do not
use specific accounting method choices, per se, to measure conservatism, because (1) they

7 To the contrary, high-technology firms likely have more (mandated) conservative accounting (e.g., due to the
writing off of intangibles) and pay lower dividends than low-technology firms with relatively high assets-inplace. This works against our hypothesized positive relation between discretionary conservatism and the level of
dividends.
8 We thank an anonymous referee for suggesting this point.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 875

are difficult to aggregate, (2) they do not reflect the magnitude of the effects of conservatism, and (3) they do not capture the conservatism of many other discretionary choices,
such as estimates of bad debt expense, warranty expense, or impairment losses.9 In Section
V, we discuss the correlation between our proxies for aggregate conservatism and some of
the specific accounting choices prior studies suggest as conservatism measures.

Measuring Bondholder-Shareholder Conflict over Dividend Policy

We use three proxies for bondholder-shareholder conflicts over dividend policy: operating uncertainty, the level of dividends (as a percentage of assets), and leverage. Firms
that face a relatively high degree of operating uncertainty, such as demand or input price
uncertainty, are more likely to experience large positive or negative shocks to their earnings

and asset values. Large positive shocks that are unsustainable will tend to inflate retained
earnings, possibly resulting in overpayment of dividends. In other words, greater uncertainty
about future profits implies a greater risk that current dividends transfer too many resources

to shareholders (Watts 1993). Thus, the higher the operating uncertainty, the more severe
the bondholder-shareholder conflicts over dividend policy. We measure the borrowing firm's

operating uncertainty by the standard deviation of its return on assets (STDROA).

As a second proxy for bondholder-shareholder conflicts of interest over dividend policy,

we use the level of dividends, measured as a percentage of assets. If a firm pays a low
level of dividends, then bondholders are less likely to be concerned about dividend overpayment. Conversely, paying a relatively high level of dividends potentially exacerbates the
conflicts with debt holders, and thus a high level of dividend payments likely indicates
more severe bondholder-shareholder conflicts over dividend policy.10

Our third proxy for the severity of bondholder-shareholder conflicts is leverage, measured as the ratio of long-term debt to assets. Ceteris paribus, higher leverage implies a
relatively larger claim on the firm's assets by bondholders. From a bondholder's perspective,
higher leverage intensifies the conflicts of interest with shareholders and the concern over
excess distributions. Thus, firms with high leverage likely experience greater bondholdershareholder conflicts over dividend policy.

Measuring Firms' Cost of Debt

Our measure of firms' cost of debt is their senior debt rating assigned by Standard and
Poor's (S&P). Prior research has shown that a firm's debt ratings are closely associated
with its eventual payoff of interest and principal obligations. For example, Altman (1992)
finds that of the debt that ultimately defaults, more than 90 percent is rated below investment

grade (BBB) at least a year prior to default. Furthermore, S&P (1986) documents that more
favorable ratings are associated with lower interest rates. Hand et al. (1992) find that
changes in bond ratings are positively related to changes in daily bond prices around ratingchange announcements. Liu et al. (1999) also find similar results. Together, these studies
provide substantial support for using debt ratings as a proxy for firms' cost of debt.
Kaplan and Urwitz (1979) find that debt ratings vary cross-sectionally with firm characteristics, such as profitability, firm size, and equity-risk. To examine whether accounting
9 For example, two firms using the straight-line method for depreciation could still differ in terms of the extent
to which their assets are understated relative to market values, if one firm estimates residual value more conservatively than the other firm.

10 A potential limitation of this proxy is that firms may pay high dividends if they have high, free cash flows
together with a low level of positive net present value investment opportunities. However, these low-growth
opportunities would imply a negative relation between our market-based conservatism proxy and the level of
dividends, biasing against finding the hypothesized positive relation.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

876 The Accounting Review, October 2002

conservatism has an incremental effect beyo

mented in prior work, our analyses control for

in Section IV.

IV. RESEARCH DESIGN

Conservatism Regressions

We employ the following model, estimated in both levels and changes, to explain crosssectional variation in conservatism and test Hi. The model includes our three proxies for
bondholder-shareholder dividend policy conflicts, as well as control variables for profitability and firm size (both proxies for alternative motivations for and costs of conservatism),

sales growth, and growth opportunities. (We discuss the control variables in more detail
below.) Our dividend policy conflict proxies and control variables are measured as six-year
averages, consistent with the estimation of our conservatism proxies:

CONi = 0 + PISTDROAi + 2 DIVi + P3 LEVi + P4ROA1

+ I5 SIZEi + I6 SALESGROj + I7 RNDADVi + si (2)

where:

CONj = one of two measures of conservatism for firm i:

CONMKT = the market-value-based conservatism proxy, defined as

-ai in equation (1), or

CONACC = the accrual-based conservatism proxy, defined as the
mean of total accruals (net income before extraordinary items plus
depreciation expense less operating cash flows) multiplied by -1 and
deflated by assets;

STDROA, = the standard deviation of firm i's ROA;

DIVi = firm i's common dividends divided by its total assets;
LEVI = firm i's long-term debt divided by its total assets;

ROA, = firm i's net income before extraordinary items, divided by its total
assets (a control for profitability);

SIZER = the natural log of firm i's total assets (a control for firm size);

SALESGROj = the annual percentage change in firm i's sales (a control for sales
growth); and

RNDADVj = firm i's R&D plus advertising expense, divided by its sales (a control
for growth opportunities).

Hypothesis 1 predicts positive coefficients on STDROAi, DIVi, and LEVI, our three proxies
for the severity of bondholder-shareholder conflicts over dividend policy.
The additional variables in our tests control for (1) profitability, (2) size, (3) sales
growth, and (4) growth opportunities. Sensitivity analyses discussed in Section V also
control for industry.
Profitability and firm size are potentially associated with alternative motivations for and
costs of conservatism, which Hi assumes constant. For low-profitability firms, the reduction

in profits associated with conservative accounting would be relatively costly. In other words,

more profitable firms can better afford conservative accounting choices. We include ROA
in the model to control for this potential cost of conservatism, and we expect firms with
higher ROAs to use more conservative accounting. However, note that more negative accruals imply both lower ROA and higher CONACC values. Thus, the mechanical, negative

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 877

relation between ROA and CONACC could dominate the expected positive relation between
conservatism and profitability.

Watts and Zimmerman (1978) argue that firms facing high political costs will tend to
use more conservative accounting. Consistent with this argument, Zmijewski and Hagerman
(1981) show that large firms use more conservative accounting methods. Accordingly, we
include the natural log of total assets in the model to control for this alternative motivation
for conservatism. We expect larger firms to make more conservative accounting choices.
An additional motivation to control for total assets is that it helps to ensure that any observed positive relation between conservatism and dividends-to-assets or leverage is not
induced mechanically by lower asset values in the denominator of the ratios. In other words,

the coefficient on DIV is the incremental effect of dividend policy, holding constant the
book value of total assets.

We control for sales growth for three reasons: (1) sales growth is likely to affect accruals
such as changes in receivables and inventory, and hence CONACC, (2) negative sales
growth indicates a decline in operations, under which our CONACC proxy is a poor measure of conservatism (Penman and Zhang 2002), and (3) sales growth may influence the
market's expectations of future growth captured in CONMKT (Dechow and Sloan 1997).11
We include the sum of R&D and advertising expenses as a control for growth opportunities (positive net present value investment opportunities) potentially captured in
CONMKT. However, because our sample is composed of relatively mature firms that are
choosing to obtain a bond rating, our sample firms are likely to have more assets-in-place
than growth opportunities. Thus, this effect may not be significant for our sample firms.
Cost of Debt Regressions
Hypothesis 2 predicts that firms choosing more conservative accounting will incur lower

debt costs. We use the following model, which we estimate in levels and changes, to test
this prediction:

RATINGi = yo + y1CONi + Y2ROA, + Y3LEVj + Y4SIZEj + y5BETAI + y6MSEj + ui (3)

where:

RATINGi = a numerical transformation of S&P's senior debt rating for firm i over the
interval 2 through 21, where larger values correspond to a less favorable

debt rating, averaged over each six-year estimation period;

BETA, = the value-weighted market-model beta, obtained from market-model estimates over the six-year period for firm i, for firms with a minimum of 24

monthly returns; and

MSEj = the mean squared error of the market-model residuals, obtained from market-model estimates over the six-year period for firm i, for firms with a

minimum of 24 monthly returns.

All other variables are as defined above.
As discussed previously, our measure of debt cost is a firm's S&P senior debt rating.

S&P rates a firm's debt from AAA (indicating a strong capacity to pay interest and repay
We repeated our analyses after dropping firms with declining sales (negative sales growth) from our analyses.
None of our inferences are affected by this sensitivity check.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

878 The Accounting Review, October 2002

principal) to CCC- (indicating vulnerability to default and dependency on favorable busi-

ness, financial, and economic conditions to make timely payments of principal and interest).
Compustat codes these ratings on a scale from 2 to 21 such that AAA (the best S&P rating)
corresponds with 2 and CCC- (the worst rating) with 21. Because lower values of RATING
are associated with better debt ratings and lower cost of debt, H2 predicts a negative
coefficient on CON, indicating that more conservative firms pay lower debt costs.

Equation (3) controls for several other determinants of debt ratings identified in prior
research. Kaplan and Urwitz (1979) identify a model of debt ratings often used by subsequent researchers. The firm characteristics they find to be most associated with debt ratings

include profitability, firm size, and equity-risk. More specifically, they find that more profitable, larger, less risky firms have better debt ratings. Thus, we include ROA, a leverage
ratio (LEV), the log of total assets (SIZE), a measure of systematic equity risk (BETA),

and a measure of idiosyncratic equity risk (MSE) in our debt-rating model to control for
these firm characteristics that affect debt ratings.12 We then use this model to assess whether
our measures of the firm's conservatism are incrementally associated with the firm's cost
of debt, as reflected in its debt rating.
V. RESULTS

Sample and Descriptive Statistics

We draw our sample from the 1998 Annual Industrial Compustat (Primary, Secondary,
and Tertiary) files. Our CONMKT conservatism measure derives from the fixed-effects
estimation of Equation (1) that requires that each firm have book and market values of
equity throughout each six-year period, 1993-1998 and 1987-1992. For each book-tomarket firm-year observation, we further require returns (with dividends) for the preceding
six years. We then estimate the fixed-effects model for those firms with an S&P senior debt
rating available on Compustat during the six-year period. The numbers of firms that meet
these data requirements are 730 and 593 in 1993-1998 and 1987-1992, respectively.
Cross-sectional (levels) analyses of conservatism and the cost of debt require additional
accounting and market variables as identified in Section IV. We obtain financial-statement
data from Compustat, and monthly stock returns for a minimum of 24 months of the sixyear sample periods from the 1998 CRSP database. These additional data requirements
reduce our sample of firms to 704 in 1993-1998, and 575 in 1987-1992, for our CONMKT
analysis.13 For our CONACC analysis, additional financial statement data requirements reduce our sample to 632 firms in 1993-1998 and 517 firms in 1987-1992. Because we
estimate our conservatism measures over six-year windows, we similarly use the firmspecific mean value of each explanatory variable over the same six-year window in our
levels specification.
We further investigate these economic relations using a "changes" specification. For
each firm, we calculate changes in all variables as the firm-specific mean value for the
period 1993-1998 less the firm-specific mean value for the period 1987-1992. This differencing results in the loss of additional firms. Specifically, 568 and 484 firms have sufficient data for the changes specification using CONMKT and CONACC, respectively. Thus,
we perform all change analyses on these reduced samples.
12 We use STDROA (an accounting measure of operating uncertainty) in the conservatism regression, and marketbased measures of uncertainty (MSE) in the cost of debt regressions, because the former better reflects the risk
of dividend overpayment whereas the latter better reflects default risk as suggested by option pricing theory.
13 Overall, our fixed-effects estimation results are similar to Beaver and Ryan (2000); most importantly, the lag
coefficients are negative and decline in magnitude monotonically as the lag length increases.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 879

Table 1 reports descriptive statistics for the 1993-1998 period. Untabulated statistics
for the 1987-1992 period are very similar.

CONMKT and CONACC are increasing measures of conservatism. That is, larger values of CONMKT indicate a relatively smaller book-to-market ratio, and greater values of
CONACC indicate more negative accruals. The mean value of CONMKT for our final
TABLE 1

Descriptive Statistics for the Sample of 704 Firms with Conservatism Estimates, S&P Debt
Ratings, and Control Variables 1993-1998
First

Third

Variable Mean Std. Dev. Quartile Median Quartile

CONMKT

0.002

0.283

-0.146

0.028

0.205

CONACC

0.004

0.031

-0.009

0.003

0.016

3.326

7.833

RATING

9.936

STDROA

0.030

DIV

0.017

LEV

0.236

ROA
SIZE

0.041

0.018

9.833

0.007

12.000

0.018

0.037

0.003

0.012

0.027

0.122

0.233

0.318

0.148

0.039

0.043

0.015

0.036

0.059

8.067

1.518

6.985

7.947

9.085

SALESGRO

0.130

0.210

0.037

0.090

0.178

RNDADV 0.023 0.042 0.000 0.002 0.027

BETA
MSE

0.898

0.076

0.439

0.031

0.609
0.054

0.900

1.155

0.067

0.090

CONMKT = a market-based measure of c

effect from a regression of book-toto yield an increasing measure of con
of data, the following variables are li
CONACC = an accrual-based measure of c
total assets, over the six-year period,
vatism (of the 704 firms, 632 had suf
RATING = a numerical transformation of
21, where 2 corresponds to AAA and 21 corresponds to CCC-;
STDROA = the standard deviation of ROA (a proxy for operating uncertainty);
DIV = common dividends divided by total assets (a proxy for dividend policy);
LEV = long-term debt divided by total assets;

ROA = net income before extraordinary items, divided by total assets (a control for profitability);
SIZE = the log of total assets;
SALESGRO = the annual percentage change in sales (a control for growth opportunities);
RNDADV = research and development plus advertising expense, divided by sales (a control for growth
opportunities);

BETA = the value-weighted market model beta estimated over the six-year period using monthly returns (a
control for systematic equity risk); and

MSE = the mean squared error of the market model residuals over the six-year period using monthly
returns (a control for idiosyncratic equity risk).

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

880 The Accounting Review, October 2002

sample is 0.002, which is insignificantly different from zero.'4 CONACC, the accrual-based
measure of conservatism, has a mean of 0.004, consistent with negative accruals on average
because we have multiplied it by -1.
The mean debt rating for our sample is 9.936, corresponding to an S&P rating between
BBB+ and A-. The standard deviation of 3.326 indicates that the sample firms exhibit
reasonable variation in debt ratings. The mean of STDROA, our proxy for operating un-

certainty, is 0.030, whereas the standard deviation of STDROA is 0.041. The average firm
in our sample pays dividends equivalent to 1.7 percent of total assets and has 23.6 percent
of its assets levered. The annual growth in sales is approximately 13 percent over the six

years. Our sample is comprised of relatively large firms (mean total assets of $11,935
million) with low R&D and advertising (mean RNDADV of 2.3 percent) and low equity
risk (mean BETA of 0.898 and MSE of 0.076). Overall, our sample statistics are similar
to those of prior studies.
Beaver and Ryan (2000) document that the bias component of book-to-market ratios,
our CONMKT, is significantly associated with accelerated depreciation methods and R&D
expenditures. In untabulated analysis for both six-year periods, we verify that CONMKT
captures accounting practices associated with lower asset valuations and unrecorded intangibles. Specifically, in both six-year periods CONMKT is significantly positively correlated
with (1) accumulated depreciation deflated by gross property, plant, and equipment; (2)
LIFO reserve deflated by total inventory; and (3) the ratios of R&D expense to sales and
of advertising expense to sales.
We do not expect, nor do we find, similar relations between CONACC and these proxies
for lower-valued or unrecorded assets for the following reasons. First, we measure
CONACC before depreciation, so any relation with accumulated depreciation would be
indirect. Second, any relation to the LIFO reserve is unclear because both increases and
decreases in inventory (i.e., positive and negative accruals) could be associated with increases in the LIFO reserve. Finally, any relation between the cumulative net change in
"accrual" accounts (e.g., the change in accounts payable) captured in CONACC, and levels
of R&D or advertising expense, is not obvious.
As an alternative, we consider the relation between CONACC and signed special items
reported on the income statement. To the extent that special items represent a more discretionary component of net income, we expect CONACC to be associated with more income-

decreasing items.15 The Pearson correlation coefficient between CONACC and special items
is -0.348, significant at < 0.01, consistent with the idea that CONACC reflects managerial
discretion in special items.

Table 2 presents Spearman (above the diagonal) and Pearson (below the diagonal)
correlations among our conservatism measures, debt ratings, bondholder-shareholder di-

vidend policy conflict proxies, and control variables. The two conservatism measures, CONMKT and CONACC, are weakly positively correlated. As discussed in Section
III, CONMKT and CONACC are stock and flow measures, respectively. This fundamental

14 CONMKT is -a1 in Equation (1), and represents an intercept shift for each firm, incremental to the overall
cross-sectional intercept. Thus, it is mean zero by construction in each of the two sample periods. The value
reported in Table 1 differs slightly from zero because we estimate the fixed-effects model on the initial sample
of 730 firms for the 1993-1998 period.

15 For example, Moehrle (2002) documents that firms use restructuring charges, commonly coded as special items
by Compustat, to accrue discretionary reserves, which the firm later reverses to boost income.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 881

c?0 c~ do 00 > N o oo o o N 0 0 -0 o 0 o

f x ~~~~~~~N N (Z 00 00 < Zoo o

-

; >~~~~~~~~~~~~e NO (Z 00 >~ -- Zx ' t (Z -- 0 tn ( O >

Q
o t

I N I n I --q <Z N ,--q 00 (ZI I I I I

^ w N0 ~~~~~~00 en M 0 N00 00 00t 0Y 00 Z N0

@~~ ~ ~ <Z Ot O ?O 00 O (- (O- O en (Z- (O- O ( OO -OO

N

000

000000

X~~~~~~~C O (Z- (oO O <Z q <Z O t> FO

N
"t
C o N
oZ
o <Z
N
(o-~ON
M (Z-tO)
n-O
o
Cu
+ clN0
O0~
O0qt
Ol~0
mc0
'r1's
oO OO N

ZS

- r

~~~
c

~~~I

cm

oo

cQ~oc

cN
0NN
) O 0
IOS<0
O O
O

~s"

n (-) O0
Z- 1O
'- 00'(0'
- >
?ON
e0 o O

~~I
I ~I
X d
O N <~~~
m > o( - ~
en < ~
00 _4'4Nt
"_ I
t (Z)I
(rZ I
-) <tZ~

H @ bus> o< ~~0~ <~Z o~_~N~-_~4 ~aO_8_CN) _oo__(In W) > Ot OZ 0

_

rF-

0
U,~~

020

ei

???o

m?>^

-?

0?'m?

Cu

ko

??

N>o~o

n C N D 0 t 0 0 > 0 n 0 n 0 0 0 N
0'
oo to 'o oot o

oo

>

O m 0 *F~~~~~~~~~n N (-) ) 0 <

0'ro

oo

y i Z ? 10~~~ G 1
Q

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

882 The Accounting Review, October 2002

difference likely contributes to the low univariate correlation between CONMKT and
CONACC.16
Consistent with Hi, CONMKT is significantly positively correlated with two of the
three proxies for bondholder-shareholder dividend policy conflicts (STDROA and DIV).
However, contrary to Hi, CONMKT is not significantly correlated with leverage (LEV),
the third proxy. The Pearson correlations between the accrual-based measure of conservatism, CONACC, and all three proxies-STDROA, DIV, and LEV-are positive, consistent
with Hi. The Spearman correlations between CONACC and two out of the three proxies-STDROA and LEV-are positive, also consistent with HI.
Hypothesis 2 predicts that conservatism reduces the cost of debt. The univariate correlations between CONMKT and RATING support H2, as the Pearson and Spearman
correlations are -0.366 and -0.380, respectively, both with p-values < 0.01. In contrast,

the correlation between CONACC and RATING is positive, contrary to H2. However, these
univariate correlations do not evaluate the incremental effect of conservatism on debt ratings
after controlling for other determinants of debt ratings, including profitability.
Conservatism and Bondholder-Shareholder Conflicts over Dividend Policy
Table 3 reports our tests of Hi. Panel A presents the results of regressions with the
market-based conservatism measure (CONMKT) as the dependent variable, whereas Panel
B presents the results of regressions with the accrual-based measure of conservatism

(CONACC). The first two columns of Panel A present results of estimating Equation (2)
using levels data consisting of the mean firm-specific value for each variable, for the periods
1993-1998 and 1987-1992, respectively. Because the main results are similar for both time
periods, the following discussion focuses on the 1993-1998 period.
Consistent with Hi, the coefficient on STDROA is positive, 2.079 (p = 0.001). This
suggests that firms with greater operating uncertainty adopt more conservative accounting.
Similarly, consistent with Hi, the coefficient on DIV is significantly positive, 3.557
(p = 0.001). LEV, with a coefficient of 0.316, is significant in the more recent sample
period but exhibits a weaker relation in the earlier period.'7 Collectively, these results suggest that conservatism plays an important role in efficient debt contracting.18
Consistent with prior research, firm profitability (ROA) and size (SIZE) are significantly
positively related to our conservatism proxies (p = 0.001) in the later time period. In the
earlier time period, however, the coefficient on SIZE is insignificant.
If CONMKT also reflects growth opportunities, then it should be positively related to
sales growth and to R&D and advertising expenditures. Table 3, Panel A shows that

16 The low univariate correlation is also partially attributable to not controlling for other determinants of CONMK
and CONACC, in particular common determinants with contrasting relations. For example, the Pearson correlation between ROA and CONMKT is 0.330. In contrast, the correlation between ROA and CONACC is -0.342.
After controlling for this common determinant, we find that the partial Pearson correlation between CONMKT
and CONACC increases to 0.259.

17 Beaver and Ryan (2000) find a positive relation between leverage, defined as total liabilities deflated by the
market value of equity, and the bias component of book-to-market ratio. Use of market value of equity as the
denominator of both leverage and the book-to-market ratio can lead to a mechanical, positive relation between
leverage and the book-to-market ratio. To avoid this possibility, we define leverage as long-term debt divided
by total assets. When we use market value of equity instead of total assets as the deflator in defining leverage,
we obtain results consistent with Beaver and Ryan (2000).
18 We consider whether the positive relation between CONMKT and leverage might arise because conservative
accounting decreases the book value of equity (thus increasing CONMKT) and decreases total assets (thus
increasing long-term debt to total assets). However, the insignificant Pearson correlation between CONMKT and
total assets of -0.017 (p = 0.66) suggests that this is unlikely. Further, controlling for the log of total assets
(SIZE) in our models reduces the likelihood of a mechanical relation induced through the effects of conservatism
on assets.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 883

CONMKT is positively related to RNDADV. However, CONMKT is negatively related to

SALESGRO. Although inconsistent with our initial expectations, the negative coefficient
on SALESGRO may be due to a relation between realized growth and book-to-market
ratios. Greater realized growth increases the proportion of new to old assets. This reduces
the disparity between the book and market values for newer assets, which in turn leads to
lower estimated conservatism. This effect may dominate the ability of SALESGRO to proxy
for expected future growth.

TABLE 3

Summary Statistics from Regressions of Accounting Conservatism on Proxies for BondholderShareholder Conflicts and on Control Variables

CON, = (o + (1 STDROAi + (2 DIV + 3 LEVi + 4 ROAi + 5 SIZE,

+ f36 SALESGROi +37 RNDADVi + si
Panel A: Dependent Variable-Market-Based Measure of Accounting Conservatism, CONMKT
All Variables
Measured In

All Variables All Variables Changes from

Measured in Measured in 1993-1998 Period

Predicted Levels Levels vs. 1987-1992

Variablea Sign (1993-1998) (1987-1992) Period
Intercept

-0.644

-0.318

(0.001)

Proxies

for

0.001
(0.975)

Bondholder-Shareholder

STDROA

DIV

(0.005)

Conflic

2.079 1.692 1.395

(0.001) (0.001) (0.001)

3.557

2.563

3.472

(0.001) (0.001) (0.001)

LEV

0.316

0.178

(0.001)

0.309

(0.117)

(0.001)

Control Variables:

ROA

SIZE

2.060
2.900
2.482
(0.001) (0.001) (0.001)

0.045
0.008
0.022
(0.001) (0.227) (0.005)

SALESGRO

-0.125

-0.143

(0.942)
RNDADV

Sample

Adjusted

-0.073
(0.813)

0.790

1.108 1.437
(0.001) (0.001) (0.001)

Size

R2

(0.901)

704

29.2%

575

23.1%

568

25.6%
(Continued on next page)

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

884

The

Accounting

Review,

October

2002

TABLE 3 (Continued)

Panel B: Dependent Variable-Accrual-Based Measure of Accounting Conservatism, CONACC

All Variables
Measured In

All Variables All Variables Changes from

Measured in Measured in 1993-1998 Period

Predicted Levels Levels vs. 1987-1992

VariableY Sign (1993-1998) (1987-1992) Period

Intercept

-0.012 -0.014 0.000

(0.154) (0.156) (0.877)

Proxies for Bondholder-Shareholder Conflict:

STDROA

0.216

0.150

(0.001)
DIV

0.404

0.197

(0.001)
LEV

(0.008)

0.001

(0.001)

0.295

(0.025)

0.000

(0.450)

0.157

(0.001)

-0.014

(0.494)

(0.901)

Control Variables:

ROA

-0.241

-0.278

(0.001)

SIZE

0.001

0.003

(0.040)

SALESGRO
RNDADV

Sample
Adjusted

(0.001)

0.001

(0.001)

(0.064)

-0.003 -0.024 -0.017

(0.246) (0.024) (0.022)
0.051 -0.003 0.053
(0.087) (0.904) (0.102)

Size
R2

-0.270

(0.001)

632

24.9%

517

24.2%

484
22.5%

a See Table 1 for variable definitions.

Variables in the levels specifications are
Variables in the changes specification are the firm-specific mean value for the period 1993-1998 less the mean
value for the period 1987-1992. White (1980) corrected p-values (one-tailed where sign is predicted) appear in
parentheses.

The last column of Table 3, Panel A, presents the results of estimating Equation (2)
using a change specification with all variables calculated as the difference between the firm-

specific mean value for the period 1993-1998 and the firm-specific mean value for the
period 1987-1992. All results are consistent with those reported for the levels specification.
Specifically, firms experiencing increased operating uncertainty, dividends, and leverage are
associated with significantly increasing levels of accounting conservatism.

Overall, even after controlling for profitability, size, sales growth, and R&D and advertising expenditures, we find that accounting conservatism, as measured by CONMKT,

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 885

is significantly increasing in proxies for bondholder-shareholder conflicts over dividend

policy, in both levels and changes specifications.
Table 3, Panel B presents the results of estimating Equation (2) with the accrual-based
conservatism measure, CONACC, as the dependent variable. Consistent with Hi, coefficients on two of the proxies for bondholder-shareholder conflicts, STDROA and DIV, are
significantly positive in both periods and in the changes specification. In contrast with Hi,
LEV is insignificant in both periods and in the changes specification. These results suggest
that accounting conservatism, as captured in cumulative (signed) accruals, is associated
with bondholder-shareholder conflicts stemming from profit volatility and dividend policy,
but not from leverage."9
In Panel B, we expect a negative sign on sales growth, because higher sales growth
likely increases current accruals, which in turn reduces CONACC. Furthermore, we do not
predict the sign of the coefficients on RNDADV because a priori it is unclear how R&D
and advertising might affect CONACC; the fact that a firm expenses R&D or advertising
should not affect the level of its accruals. We also cannot predict the sign of the coefficient
on ROA because there are two offsetting effects of ROA on CONACC. On the one hand,
we expect more profitable firms to make more conservative accounting choices. On

the other hand, more negative accruals would, ceteris paribus, reduce ROA and increase
CONACC values.

Panel B shows that ROA has a negative coefficient, consistent with negative accruals
both reducing ROA and increasing CONACC as discussed above. As expected, SIZE has
a positive coefficient-larger firms record more conservative accruals, on average. The
coefficient on SALESGRO is significantly negative in the earlier period and in the changes
specification. This is consistent with high sales growth leading to large increases in accounts
receivable and/or inventory, which reduce CONACC. The coefficient on RNDADV is not
significant at conventional levels in either period of the levels analysis or in the changes
specification.
In summary, the evidence reported in Table 3 suggests that accounting conservatism
increases with firm characteristics that proxy for greater bondholder-shareholder conflicts
over dividend policy. This result is generally robust to our two distinct measures of conservatism. CONMKT is associated with our proxies for operating uncertainty, level of
dividends (as a percentage of assets), and leverage. CONACC is associated with our proxies
for operating uncertainty and the level of dividends, but not leverage.
Conservatism and the Cost of Debt

Table 4 reports our test of H2, which posits that conservatism is associated with a lower
cost of debt. The first two columns of Table 4, Panels A and B, report results of estimating
Equation (3) using an ordered-logit model based on the firm-specific levels data for the two
time periods. Because RATING is a categorical dependent variable, the basic assumptions
of OLS are violated. The ordinal nature of RATING allows us to estimate Equation (3) as
an ordered-logit model, which assumes that a discrete debt rating corresponds to the value
of an unobservable, continuous, latent index, such as credit risk.
Because the inferences are similar for both periods, we discuss the more recent 19931998 period results. Consistent with H2, the coefficient on CONMKT is negative, -2.348

9 In an earlier version of this paper, we measured CONACC using accruals inclusive of depreciation, which
exhibited a significant positive relation with LEV. Based on unreported analyses, we attribute that relation to
asset intensity increasing both LEV and the depreciation charge. We thank an anonymous reviewer for pointing
this out.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

886 The Accounting Review, October 2002

(p < 0.001), suggesting that firms using more conservative accounting (as evidenced by
book values persistently below market values) have significantly more favorable debt rat-

ings, and in turn, significantly lower debt costs. In addition, all control variables have the
expected signs and are significant. The pseudo-R2 is 67.1 percent, indicating that the model
explains a substantial amount of the variation in firms' debt costs.
The last column in Table 4 presents the results for Equation (3) with all variables
measured in changes. This model tests for a relation between changes in conservatism and

changes in firms' cost of debt. Again, the results are consistent with the levels specification.
The coefficient on the change in CONMKT is -1.233 (p < 0.001) indicating that firms
with higher conservatism also have more favorable debt ratings, and in turn, lower debt
costs. The sign and significance of the control variables are also similar to the results
reported for the levels specification except that the coefficient on BETA is not significant.

TABLE 4

Summary Statistics from Ordered-Logit Tests of Association between the Cost of Debt and
Accounting Conservatism, after Controlling for Other Determinants of Debt Ratings

RATINGi = yo + y1CON1 + y2ROAi + y3LEV, + y4SIZE, + y5BETAi + y6MSE, + si

Panel A: CON Variable-Market-Based Measure of Accounting Conservatism, CONMKT
All Variables
Measured In
All Variables All Variables Changes from

Measured in Measured in 1993-1998 Period

Predicted Levels Levels vs. 1987-1992

Variablea Sign (1993-1998) (1987-1992) Period

Proxy for Accounting conservatism:

CONMKT

-2.348

-1.372

-1.233

(0.001) (0.001) (0.001)

Control Variables:

ROA

-13.725

-23.209

-13.930

(0.001) (0.001) (0.001)

LEV

5.133

5.905

(0.001)
SIZE

0.667

1.217
0.062
(0.001) (0.001) (0.384)

38.123

Sample
Pseudo

(0.001)

-0.565 -0.770 -0.553

(0.001) (0.001) (0.001)

BETA

MSE

4.328

(0.001)

15.233
5.171
(0.001) (0.001) (0.019)

Size

R2

704

67.1%

575

71.3%

568

62.8%
(Continued

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 887

TABLE 4 (Continued)

Panel B: CON Variable-Accrual-Based Measure of Accounting Conservatism, CONACC

All Variables
Measured In

All Variables All Variables Changes from

Measured in Measured in 1993-1998 Period
Predicted Levels Levels vs. 1987-1992

Variable Sign (1993-1998) (1987-1992) Period

Proxy for Accounting conservatism:

CONACC

-6.320 -3.411 -5.486

(0.006) (0.092) (0.003)

Control Variables:

ROA

-18.238 -26.647 -18.003

(0.001) (0.001) (0.001)

LEV

4.786
5.627
3.900
(0.001) (0.001) (0.001)

SIZE

-0.584

-0.687
(0.001)

BETA

MSE

17.680

(0.001)

Pseudo
a

See

Size

R2

Table

(0.001)

0.354 1.051 0.012

(0.031) (0.001) (0.479)

40.974

Sample

-0.564

(0.001)

632

64.6%
for

8.189

(0.001)

517

70.0%

variable

(0.001)

484

59.7%

definitions.

Variables in the levels specification are

Variables in the changes specification ar
value for the period 1987-1992. Models a
tailed where sign is predicted). RATING
with
debt

less

favorable

debt

ratings,

and

neg

ratings.

The coefficients on CONACC in

sistent with H2. The coefficient
at the 10 percent level in the ear
between CONACC and debt costs
that conservative accounting is as
debt

costs,

after

controlling

for

pr

In summary, the evidence prese

mitigate bondholder-shareholder
cost of debt. The collective evide
vatism

plays

an

important

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

role

in

888 The Accounting Review, October 2002

Robustness

In this section we discuss the results of additional robustness checks pertaining to (1

industry controls, (2) analyses of subsamples based on dividend payment or nonpayment
and (3) sensitivity to extreme values.

We repeat our analyses in Tables 3 and 4 using industry-adjusted measures. We measure

all variables as deviations from the industry mean, with industry defined as those firms in
the sample that have the same two-digit Standard Industrial Classification (SIC) code. Industry-adjusted variables help control for GAAP-mandated conservatism and other economic characteristics (such as growth) that likely vary across industries. All of our conclusions about the relation between conservatism and the severity of bondholder-shareholder
conflicts over dividend policy are also supported by the industry-adjusted analyses. Similarly, the industry-adjusted measures confirm our conclusion that firms choosing more conservative accounting incur a lower cost of debt.

Next, we examine subsamples of firms based on whether they paid dividends. Bondholder-shareholder conflicts over dividend policy for firms that do not pay dividends are
likely to be minimal. However, the threat of dividend initiation could still motivate the use
of conservative accounting. Analyzing these subsamples provides evidence on the potential
role of conservatism in controlling for the risk of dividend overpayment associated with
the threat of dividend initiation. We identify 151 firms that did not pay dividends during
the sample period. When we exclude these firms from the analyses, all inferences and
conclusions remain identical to those reported in the paper. When we performed the analysis
on only those firms that did not pay dividends, the results were generally consistent (despite

the much smaller sample size) with those reported for the broader sample, except that the
relation between debt cost and CONACC in the earlier period was not significant at conventional levels. This suggests that even if firms do not currently pay dividends, the threat
that they may initiate dividends provides incentives to mitigate bondholder-shareholder
conflicts over (future) dividend policy.

Finally, to ensure that our results are not driven by extreme values, we estimated the
models in Tables 3 and 4 after deleting observations deemed outliers as defined by a Cook's
D-statistic with a p-value less than or equal to 0.15. Results are consistent with those
reported, indicating that influential observations are not driving the results.
VI. CONCLUSION

This study provides evidence suggesting that accounting conservatism help

gate bondholder-shareholder conflicts over dividend policy, and (2) reduce firm
We use two alternative measures of conservatism: (1) a market-based measure
Beaver and Ryan [2000]), and (2) an accrual-based measure (based on Givoly an
[2000]). We find that both measures are significantly positively correlated with
three proxies for bondholder-shareholder conflicts over dividend policy: stand
of ROA and the dividend-to-asset ratio. The evidence on leverage, our third pro
Overall, we conclude that firms facing more severe bondholder-shareholder co
dividend policy choose more conservative accounting.
Furthermore, we document that firms using more conservative accounting
favorable debt ratings and so incur a lower cost of debt, after controlling for
minants of the debt costs. Taken together, the evidence is consistent with acco
servatism playing an important role in efficient debt contracting.

Our inferences are subject to several limitations. First, because the conserv
managers voluntarily choose is not directly observable, our proxies potentially
GAAP-mandated conservatism as well as managers' choices among more or le
vative accounting methods, estimates, and assumptions. However, it is unlike

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

Ahmed, Billings, Morton, and Stanford-Harris-The Role of Accounting Conservatism 889

results are driven by cross-sectional variation in GAAP-mandated conservatism because:

(1) even GAAP-mandated conservatism requires managers to exercise discretion, (2) our
two different proxies yield similar results even though they are likely to be affected differently by GAAP-mandated conservatism, (3) we find similar results with industry-adjusted

measures that control for the effect of mandated conservatism that varies systematically
across industries, and (4) our results hold in a changes specification, which further controls
for the differential effects of GAAP-mandated conservatism on individual firms.
Second, our conservatism proxies are potentially affected by growth opportunities and
operating asset growth. We control for these factors directly, with control variables, as well
as by using industry-adjusted measures that control for industry-specific growth.
Third, we focus exclusively on the role of accounting conservatism in mitigating bondholder-shareholder conflicts over dividend policy. There are other mechanisms for resolving
these conflicts, such as secured debt. Furthermore, there are other conflicts between bondholders and shareholders (e.g., over investment and financing policies). Conservatism may
also play a role in mitigating these other conflicts. We leave analysis of these other mechanisms and other bondholder-shareholder conflicts for future research.

REFERENCES
Altman, E. 1992. Revising the high-yield bond market. Financial Management 21 (2):
Basu, S. 1997. The conservatism principle and the asymmetric timeliness of earnings
Accounting and Economics 24 (1): 3-38.

Beaver, W., and S. Ryan. 2000. Biases and lags in book value and their effects on the ability of th
book-to-market ratio to predict book return on equity. Journal of Accounting Research 38 (
127-148.

Begley, J. 1994. Restrictive covenants included in public debt agreements: An empirical investigati
Working paper, University of British Columbia.

Black, F. 1976. The dividend puzzle. Journal of Portfolio Management 2 (2): 5-8.
Botosan, C. 1997. Disclosure level and the cost of equity capital. The Accounting Review 72 (3): 323349.

Dechow, P. M., and R. M. Sloan. 1997. Returns to contrarian investment strategies: Tests of naive
expectations hypotheses. Journal of Financial Economics 43 (1): 3-27.
Devine, C. 1963. The rule of conservatism reexamined. Journal of Accounting Research 1 (2): 127138.

El-Gazzar, S., and V. Pastena. 1990. Negotiated accounting rules in private financial contracts. Journal
of Accounting and Economics 12 (4): 381-396.
, and . 1991. Factors affecting the scope and initial tightness of covenant restrictions
in private lending agreements. Contemporary Accounting Research 8 (1): 132-151.
Feltham G., and J. Ohlson. 1995. Valuation and clean surplus accounting for operating and financial
activities. Contemporary Accounting Research 11 (2): 689-731.
Financial Accounting Standards Board (FASB). 1974. Accounting for Research and Development
Costs. Statement of Financial Accounting Standards No. 2. Stamford, CT: FASB.
. 1980. Qualitative Characteristics of Accounting Information. Statement of Financial Accounting Concepts No. 2. Stamford, CT: FASB.
. 1985. Elements of Financial Statements. Statement of Financial Accounting Concepts No. 6.
Stamford, CT: FASB.

Givoly, D., and C. Hayn. 2000. The changing time-series properties of earnings, cash flows and
accruals: Has financial reporting become more conservative? Journal of Accounting and Economics 29 (3): 287-320.

Hand, J. R. M., R. Holthausen, and R. Leftwich. 1992. The effect of bond rating agency anno
on bond and stock prices. Journal of Finance 47 (2): 733-753.
Healy, M., and K. Palepu. 1990. Effectiveness of accounting-based dividend covenants. Journal of
Accounting and Economics 12 (1-3): 97-123.

Hendriksen, E., and M. Van Breda. 1992. Accounting Theory. 5th edition. Homewood, IL: Irwin.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms

890 The Accounting Review, October 2002

Jensen, M. C., and W. H. Meckling. 1976. Theor

ownership structure. Journal of Financial Economics 3 (4): 305-360.
Kalay, A. 1982. Stockholder-bondholder conflict and dividend constraints. Journal of Financial Economics 10 (2): 211-233.

Kaplan, R., and G. Urwitz. 1979. Statistical models of bond ratings: A methodological inquiry. Jo
of Business 52 (2): 231-261.

Leftwich, R. 1983. Accounting information in private markets: Evidence from private lending agree
ments. The Accounting Review 58 (1): 23-42.

Leuz, C., D. Deller, and M. Stubenrath. 1998. An international comparison of accounting-based payout
restrictions in the United States, United Kingdom, and Germany. Accounting and Business Research 28 (2): 111-129.

Liu, P., F. Seyyed, and S. Smith. 1999. The independent impact of credit rating changes: The cas
Moody's rating refinement on yield premiums. Journal of Business Finance & Accounting 2
4): 337-363.

Milgrom, P., and J. Roberts. 1992. Economics, Organization & Management. Englewood Cliff
Prentice Hall.

Moehrle, S. 2002. Do firms use restructuring charge reversals to meet earnings targets? The Accounting
Review 77 (2): 397-414.

Mohrman, M. 1996. The use of fixed GAAP provisions in debt contracts. Accounting Horizons 1
(3): 78-91.

Muller, K., HI, and E. Riedl. 2001. Using stock returns to determine bad versus good news to
conservatism of accounting earnings. Working paper, Pennsylvania State University.
Penman, S., and X. Zhang. 2002. Accounting conservatism, the quality of earnings, and stock re
The Accounting Review 77 (2): 237-264.

Ryan, S. 1995. A model of accrual measurement with implications for the evolution of the book-tomarket ratio. Journal of Accounting Research 33 (1): 95-112.

Sengupta, P. 1998. Corporate disclosure quality and the cost of debt. The Accounting Review 73 (4):
459-474.

Shon, J. 2000. Implications of conservative accounting choice on the asymmetric timeliness of

ings. Working paper, University of Chicago.

Smith, C., and J. Warner. 1979. On financial contracting: An analysis of bond covenants. Journal of
Financial Economics 7 (2): 117-161.

Standard & Poor's, Inc. 1986. Debt Ratings Criteria. New York, NY: Standard & Poor's Corporati
Sterling, R. 1967. Conservatism: The fundamental principle of valuation in traditional account
Abacus 3 (2): 109-132.

Watts, R., and J. Zimmerman. 1978. Towards a positive theory of the determination of accou
standards. The Accounting Review 53 (1): 112-134.

. 1993. A proposal for research on conservatism. Working paper, University of Rochester.

White, M. 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for
heteroskedasticity. Econometrica 40: 817-838.

Zmijewski, M., and R. Hagerman. 1981. An income strategy approach to the positive theory of
accounting standard setting/choice. Journal of Accounting and Economics 3 (2): 129-149.

This content downloaded from 202.43.95.117 on Tue, 29 Nov 2016 05:25:21 UTC
All use subject to http://about.jstor.org/terms