Professional Documents
Culture Documents
Givoly2000 PDF
Givoly2000 PDF
Abstract
This paper documents changes in the patterns of earnings, cash #ows and accruals over
the last four decades. In the absence of a generally accepted de"nition of conservatism,
a number of measures of reporting conservatism are identi"ed and examined. These
measures rely on the accumulation of nonoperating accruals, the timeliness of earnings
with respect to bad and good news, characteristics of the earnings distribution and the
market-to-book ratio. The patterns are consistent with an increase in conservative
"nancial reporting over time. The "ndings have implications for accounting standard
setting, regulation of "nancial information and "nancial statement analysis. 2000
Elsevier Science B.V. All rights reserved.
夽
We thank Jerry Zimmerman (the editor), Ray Ball, Jack Hughes, Katherine Schipper, and
participants in the accounting workshops at the University of California at Berkeley, Duke
University, University of Michigan, New York University, Tulane University and Tel Aviv Univer-
sity for their helpful comments on this paper.
* Corresponding author. Tel.: #1-310-206-9225; fax: #1-310-825-3165.
E-mail address: chayn@agsm.ucla.edu (C. Hayn).
0165-4101/00/$ - see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 5 - 4 1 0 1 ( 0 0 ) 0 0 0 2 4 - 0
288 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
1. Introduction
Many FASB statements result in an earlier recognition of expenses or losses. These include
SFAS 106, Employer's Accounting for Postretirement Bene"ts Other Than Pensions (1992), SFAS
114, Accounting by Creditors for Impairment of a Loan (1993), SFAS 68, Research and Develop-
ment Arrangements (1982), SFAS 123, Accounting for Stock-Based Compensation (1995) and SFAS
121, Accounting for the Impairment of Long-Lived Assets and for Long-Lived Assets to be Disposed
Of (1995). More conservative rules by the AICPA include AICPA SOP 97-2 (1997) and SOP 98-5
(1998) on software revenue recognition and the costs of start-up activities.
See, for example, Berton (1992), Kothari et al. (1988), Minow (1984) and Walters (1992).
Examples include managers' tendency to write o! acquired in-process R&D, as well as the
election by most "rms to report the full impact of SFAS 106 in the year of adoption rather than
amortize it over future years. Patterns of management response to the litigious environment are
modeled by Hughes and Sankar (1998) and documented by Kasznik and Lev (1995) and Skinner
(1994, 1997), among others.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 289
Krishnan and Krishnan (1997) and Stice (1991) suggest that client characteristics signi"cantly
a!ect auditors' litigation risk. The former study "nds that auditors are more likely to resign from
jobs that have a higher probability of resulting in litigation. Anecdotal evidence is provided in &More
accounting "rms are dumping risky clients', (The Wall Street Journal, 23 April 1997). Some studies
even suggest that the increased legal liability of auditors inhibits reporting innovation and dis-
closures (Minow, 1984; Kothari et al., 1988).
See, for example, Cole et al. (1996), Dreman (1998) and Wyatt (1996).
See, for example, &On the books, more facts and less "ction', The New York Times, 16 February
1997.
290 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
setters and regulators, the empirical "ndings regarding the extent of, and, in
particular, the trend in, conservatism are scant. Stober (1996), using the market-
to-book ratio as a proxy for the degree of conservatism, "nds a conservative bias
in accounting book values relative to market values. Leftwich (1995) examines
the content of the history of the FASB pronouncements and agenda decisions.
He concludes that they manifest a conservative tilt and a predominant concern
by the FASB that management is more likely to use its measurement discretion
and disclosure options to report higher earnings and net worth over some
horizon.
Basu (1997) devises several empirical measures of conservatism, among them
a stronger association of stock price movements and earnings in periods of bad
news. His results are consistent with the presence of reporting conservatism and
with the level of conservatism varying with changes in auditors' legal liability
exposure. Holthausen and Watts (2000), employing Basu's methodology, pro-
vide indications of conservative accounting even in the pre-standard setting era
and an increase in its degree during the FASB's regime. Similar market-based
measures are used by Ball et al. (1999) and by Pope and Walker (1999) to assess
cross-country di!erences in conservatism. Results that could be interpreted as
indicating reporting conservatism are also reported by McNichols (1988). She
"nds that returns during earnings announcement periods are skewed more
negatively than non-announcement returns, which is consistent with earnings
re#ecting bad news in a more timely manner than good news.
The studies cited above, with the exception of Holthausen and Watts (2000),
have not focused on the intertemporal variations in the degree of conservatism.
This study extends this line of research by addressing the variation in conserva-
tism over time. Further, rather than relying primarily on market-based
measures of conservatism (such as the market-to-book ratio), this paper recog-
nizes that conservatism is essentially an issue of the timing and sequencing of
revenues and expenses relative to the associated cash #ows. Therefore, both
time-series and distributional properties of earnings, cash #ows and accruals are
examined.
The "ndings show persistent, non-trivial changes in the relation between
earnings, cash #ows and accruals. In line with the results of previous studies (e.g.,
Hayn, 1995; Jan and Ou, 1994; Collins et al., 1997), we "nd that reported
See, for instance, EITF 94/3 (FASB, 1994) which attempts to limit the amounts written o! as
restructuring charges before actual cost commitments are made.
Arthur Levitt, the SEC Chairman, cited an assortment of accounting practices commonly
used by public corporations to manage earnings. Most of the practices (such as overstatements
of restructuring and loss provisions) are income-decreasing (see www.sec.gov/news/speeches/
spch220.txt).
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 291
pro"tability over the last four decades has generally declined. More revealing,
however, is the "nding that this decline is not accompanied by a corresponding
decline in cash #ows. In particular, the evidence points to a massive accumula-
tion of negative nonoperating accruals over this period. In addition, the earnings
distribution has become more dispersed and negatively skewed relative to that
of cash #ows. The evidence also suggests a more timely recognition of &bad news'
relative to &good news'. These results are consistent with an increase over time in
the degree of reporting conservatism.
The "ndings have implications for "nancial statement analysis, accounting
standard setting, securities regulation and research. The changes in the time-
series patterns of earnings and cash #ows suggest that an analysis of "nancial
statements should take into account this trend rather than relying on historical
benchmark ratios. In particular, the indications of a greater conservatism in
"nancial reporting suggest that the higher earnings multiples and market-
to-book ratios of recent years do not necessarily imply equity overpricing.
The results also call into question the assumption, often made by researchers
and regulatory bodies, regarding management's motivation to accelerate
earnings.
The remainder of the paper is organized as follows. Section 2 discusses the
nature of conservatism and introduces measures to assess its extent. The data
and sample are described in Section 3. Section 4 contains the "ndings concern-
ing conservatism, based on the measures introduced in Section 2. Section
5 provides a summary and concluding remarks.
months beginning nine months prior to the end of "scal year t, and DR is
GR
a dummy variable set equal to 1 if R is negative and 0 otherwise.
GR
Various measures of conservatism can be developed from the results of
regression (1):
One of the limitations of the above measures is that they rely on the stock
price movements to identify good and bad news. The fourth measure requires, in
addition, the estimation of the risk-free rate and further assumes that it is
constant over time.
The notion that, under conservative reporting, bad news is reported on
a more timely basis than good news serves as the basis for yet another set of
conservative measures. Note that if conservatism leads to an immediate and
complete recognition of negative events and a delayed and gradual recognition
of positive events, it is likely to result in a negatively skewed earnings distribu-
tion. Second, to the extent that increased conservatism takes the form of either
a more immediate (rather than gradual) recognition of bad news, or a greater
tendency to provide for anticipated future costs or losses, such an increase will
be associated with increased variability of the earnings series. Accordingly, two
additional measures of conservatism are the skewness and variability of the
earnings distribution.
A "nal proxy for conservatism is based on the balance-sheet-oriented de"ni-
tion of conservative accounting suggested by the theoretical framework
As in Basu (1997) and Beaver et al. (1980), the reverse regression is used because the OLS
standard errors and test statistics are better speci"ed when the leading variable is independent and
the lagging variable is dependent.
The average downward bias is shown by Pope and Walker (1999) to equal (under certain
assumptions): (1/k!b ) R Pr(good) } (b #b !1/k) R Pr(bad), where R (R ) is the
mean return over a good news (bad news) period, de"ned as a period with positive (negative) returns
and Pr(good) (Pr(bad)) is the relative frequency of good news (bad news) periods. The parameter 1/k,
the risk-free rate, is estimated by the intercept of regression (1).
294 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
3. Sample
The sample consists of all "rms on the 1999 Compustat database contained in
the primary, secondary and tertiary, full coverage and research "les. The
resulting sample, referred to as the &full sample', spans the 49-year period from
1950 to 1998. Since this paper addresses patterns in the relation between
accounting numbers and cash #ows over time, regulated "rms (i.e., utilities) for
which this relation is a!ected by unique institutional and regulatory factors were
excluded from the sample. The full sample increases in size from 593 "rms in
Note that the very existence of "rms implies a positive economic rent to creating "rms and
hence a market-to-book ratio that exceeds unity, independent of conservatism. Yet, the ratio is still
meaningful in gauging changes over time in the degree of conservatism.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 295
1950 to about 9,000 in the late 1990s, as new "rms are added to the Compustat
database. To ensure comparability of the results over time, most of the analyses
are conducted on a constant sample of 896 "rms that exist for the period from
1968 to 1998 (hereafter, the &constant sample'). Selection bias issues associated
with both samples are discussed in the following section. Return information
was retrieved from the Center for Research in Security Prices (CRSP) database.
4.1. Proxtability
The pattern of "rm pro"tability over the last four decades for both the full and
constant samples of "rms is shown in Table 1. In line with previous results
(Hayn, 1995; Jan and Ou, 1995; Collins et al., 1997), the percentage of "rms in
the full sample reporting losses has increased signi"cantly, from 2}3% in the
early years to over 35% in the late 1980s and the 1990s (see Panel A). The
constant sample, shown in Panel B, shows a similar trend.
This increase in the frequency of losses re#ects the drop in reported pro"tabil-
ity over the years. For the full sample, the accounting rate-of-return, de"ned as
the ratio of net income-to-total assets (hereafter, ROA), shows a continuous
decline over the years, falling fairly steadily from a mean of about 7% in the
1950s to !8% in the 1990s. Some portion of this decline is due, undoubtedly,
to the changing composition of "rms on the Compustat database, which
Net income (or net loss) is de"ned throughout this paper as the &bottom line' net income (i.e.,
income after gains or losses from discontinued operations, extraordinary items and the cumulative
e!ect of changes in accounting principle). Very similar results are obtained for all of the analyses in
the paper when income from continuing operations is used instead of &bottom line' net income.
The "nding that earnings performance over the last two decades is rather lackluster runs
contrary to the common belief that vigorous earnings are one of the causes for the surge in the stock
market in recent years. This belief is re#ected in many analyses in the "nancial press such as
Bongiorno, L., &Hot Damn, What a Year!' Business Week, 3414 (March 6, 1995) 98}100; Clark, K.,
&These are the Good Old Days', Fortune, 135 (11) (June 9, 1997) 74}86; and Spiers, J., &The Pro"ts
Flowed', Fortune, 133 (8) (April 29, 1996) 260}261.
296 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
Table 1
Frequency of losses and net income-to-total assets, ROA
Freq. of ROA Freq. of ROA
No. of losses losses
Year "rms (%) Mean Median Subperiod (%) Mean Median
Table 1 (continued)
Freq. of Freq. of
ROA ROA
No. of losses losses
Year "rms (%) Mean Median Subperiod (%) Mean Median
To test for signi"cance, nine subperiods (generally "ve years long) are formed as shown in Table
1. T-tests are performed to assess whether the di!erence in the mean ROA between adjacent
subperiods is signi"cant. All t-values indicate a signi"cant decline (at the 1% signi"cance level).
This argument is reinforced by the fact that the mean annual stock return over the 1950}1998
period is much higher for the "rms in the constant sample (9.8%) than for the other "rms (2.0%).
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 299
Table 2
Alternative pro"tability measures, by subperiod (constant sample of 896 "rms)
The most extreme (0.5%) of the cases at either end of the distribution each year were
truncated.
The ratio of net income-to-book value of equity is computed only for cases with positive book
values. The percentage of "rms with negative book values is negligible in the "rst four subperiods. It
rises steadily in the last "ve subperiods, with about 5% of the "rms having negative book values by
the latest subperiod.
in pro"tability, mirroring the ROA measure. So too in recent years does net
income-to-equity, a measure that re#ects the e!ect of leverage on the return to
shareholders. However, the results pertaining to this measure are less clear due
to the increasing presence of "rms with negative book values of equity. In
summary, all four measures show a signi"cant decline over time similar to that
exhibited by ROA. Since the main "ndings of the paper hold for all of these
measures and for both the full and constant samples, for the sake of brevity we
discuss primarily the results based on ROA.
Very few "rms had negative book values in the early years; by the 1991}1998 subperiod, over
5% of the "rms had negative book values.
See footnote 14 for a description of the signi"cance tests by subperiod. All comparisons are
signi"cant at the 1% signi"cance level except for the di!erence in the mean income-to-sales between
the "rst two subperiods and the di!erence in the mean of the tested variables between the 1971}1975
and 1976}1980 subperiods. When signi"cance tests are performed on the di!erences between
decades, all comparisons are signi"cant at the 1% level.
300 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
results, the constant sample of "rms is partitioned each year into "ve "rm-size
portfolios based on total assets. The results (not tabulated here) show that while
the decline in pro"tability and the increase in losses are somewhat more
pronounced for smaller "rms, the patterns observed in the earnings data hold
for all "rm-size portfolios. Similar results (not reported) are obtained for the full
sample.
To determine whether these trends are limited to only a few industries, we
divide the constant sample into 272 industry groups based on their 4-digit SIC
codes. The results of this analysis (not shown) indicate that the decline in
pro"tability of the accounting return measure occurred in almost all of the
industries. Between the early years and the later years, the ROA dropped in over
two-thirds (66.2%) of the 272 industries examined and the cross-sectional
average of the within-industry percentage of losses increased from 5.1% in
1950}1960 to 14.8% in 1991}1998. Similar results are obtained when, instead of
breaking down the sample by SIC codes, we partition the "rms into two groups:
those in &high technology' and those in &low technology' industries.
The fact that the patterns of reduced pro"tability, increased incidence of
losses and the drop in reported pro"tability are observed across "rm size groups
and industries suggests that these phenomena are widespread, a!ecting the
majority of U.S. "rms.
This classi"cation, used by Francis and Schipper (1999), is based on the likelihood that "rms in
an industry will have substantial unrecorded intangible assets.
Data for CFO in 1987}1998 are obtained from Compustat item 308. For earlier years, we derive
CFO from funds from operations (FFO), Compustat item 110, as follows: CFO " FFO ! (*Cur-
rent Assets #*Debt in Current Liabilities ! *Current Liabilities !*Cash). Because it is based
on balance sheet changes, our derivation of the CFO for the earlier years is subject to measurement
error (Collins and Hriber, 1999). However, this measurement error is related to the incidence of
mergers and acquisitions as well as discontinued operations. As explained later in the paper, our
conclusions do not change when net income before discontinued operations is considered or when
"rms engaging in mergers and acquisitions are excluded from the analysis. Therefore, measurement
errors are unlikely to a!ect our results.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 301
The de#ator of the cash #ow measure, total assets, depends to some extent on accounting
procedures. However, since it also serves as the de#ator for ROA, the earnings measure, any
di!erence between the cash #ow ratio and the earnings measure is unlikely to be a!ected by
accounting practices. To further ascertain that the de#ator does not a!ect the conclusions, we
conducted the analysis using total sales as the de#ator, with essentially the same results.
Extending the period to earlier years reduces the sample size considerably due to missing data.
Depreciation accruals are excluded because their reversal is not captured by cash #ows from
operations. The related cash out#ows (arising from the acquisition of long-lived assets) are reported
as cash #ows from investing activities.
302 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
Table 3
Cash #ow from operations-to-total assets, CFOA (constant sample of 896 "rms)
Freq. of CFOA Freq. of CFOA
negative negative
Year cases (%) Mean Median Subperiod cases (%) Mean Median
Fig. 1. Cumulative accruals by type, 1965}1998 (constant sample of 896 "rms). Total Accruals
(before depreciation): (Net Income#Depreciation)! Cash Flow from Operations. Operating
Accruals: *Accounts Receivable#*Inventories#*Prepaid Expenses!*Accounts Payable!
*Taxes Payable. Nonoperating Accruals: Total Accruals (before depreciation) minus Operating
Accruals.
the operating accruals was relatively slow in the early years, accelerating from
the late 1970s onward.
Extracting depreciation, amortization and operating accruals components
from total accruals results in accruals consisting primarily of such items as loss
and bad debt provisions (or their reversal), restructuring charges, the e!ect of
changes in estimates, gains or losses on the sale of assets, asset write-downs, the
accrual and capitalization of expenses, and the deferral of revenues and their
subsequent recognition.
To distinguish these accruals from the operating accruals, we refer to these
remaining accruals as &nonoperating' accruals. Note that although some of these
accruals are dictated by GAAP, the timing or amount of most of them are
subject to management discretion.
The accumulation of these nonoperating accruals over time, our "rst measure
of conservatism, is illustrated by line 3 of Fig. 1. Note that nonoperating
accruals accumulate fairly steadily over the sample period to a sizable negative
amount. The net negative accumulation is most pronounced in the more recent
years. The total accumulation of these accruals is quite substantial, representing
2.8% of the cumulative sales of the sample "rms for that period and 32.2% of
their total assets as of the end of 1998. The accumulation of negative accruals in
the second half of the period is common to most "rms and industries. The
average annual accumulation of accruals per "rm in the second half of the
period is signi"cantly lower (i.e., more negative) than in the early half of
the period. Further, the negative accumulation of nonoperating accruals
in the second subperiod occurs in 264 of the 272 4-digit industries. The ratio
between this accumulation in the second subperiod and that in the "rst sub-
period is 6.1 for the constant sample, with an inter-quartile range across the 272
industries of between 2.9 and 7.7.
The "nding of a prevalent and signi"cant accumulation of negative
nonoperating accruals is consistent with an increase in reporting conservatism
over the last several decades.
Restructuring. One economic development over the last two decades is the
restructuring or re-engineering activity undertaken by many companies. Such
real activities are often associated with charges to income that are not accom-
panied by a contemporaneous cash out#ow. These charges, while resulting in
a decline in immediate pro"tability and an accumulation of negative nonoperat-
ing accruals, do not necessarily imply a greater reporting conservatism.
To examine the validity of this explanation, we recomputed the nonoperating
accruals (as de"ned in Section 4.2.2) excluding two components that are most
likely to capture the e!ect of restructuring activities } special items and gains or
losses from discontinued operations. The results, presented in line 4 of Fig. 1,
show that while these components do indeed dampen reported earnings, the
extensive accumulation of negative accruals persists even after their exclusion.
The restructuring explanation is also inconsistent with the "nding (presented
earlier) that the drop in pro"tability is prevalent across almost all industries,
a!ecting not only those in which restructuring was common (most notably the
manufacturing sector), but also industries (such as those belonging to the
high-technology group) that were less a!ected by this trend.
Mergers and acquisitions. Another possible explanation for the drop in corporate
pro"tability is the recurring waves of mergers and acquisitions (M&A) during
the "ve decades spanned by our sample. Most of the acquisitions in the last
decade were accounted for by the purchase method (Davis, 1990). The purchase
method tends to dampen pro"tability (and increase negative accruals) due to the
added amount of depreciation and amortization arising from the revaluation of
the acquired tangible assets and the establishment of goodwill. Further, since
most acquisitions are highly leveraged, subsequent pro"ts are also adversely
a!ected by the increased "nancial charges.
To test for the plausibility of the M&A explanation for the decline in
pro"tability, we "rst examine the trend in earnings before depreciation, amorti-
zation and interest. The results (not shown) reveal a pattern of decline similar to
that observed for the other measures of pro"tability such as the ROA or
EBIT/Assets (as shown in Table 2). Second, we form each year a sub-sample of
all "rms that did not engage in an acquisition in any of the preceding "ve years
(identi"ed by a footnote code to Compustat item 12). The "ndings indicate that
the decline in ROA over the years for this sub-sample is statistically indistin-
guishable from that of the full sample. Finally, we reproduced the accrual
accumulation results of Fig. 1 excluding "rms that engaged in M&A activity
during the period. The results (not shown) are again similar to those reported for
the entire sample.
Growth and Inyation. A potential technical explanation for the observed accu-
mulation of negative accruals is the growth of the sample "rms. Growth,
whether real or nominal (due to in#ation), results in an increase in the magni-
tude of earnings and cash #ows and, consequently, of accruals. While we control
for growth by de#ating the accruals by total assets, this de#ator may not fully
capture the scale of operations that gives rise to the accruals. While the constant
sample "rms indeed exhibit impressive growth throughout the examined
period, the growth explanation is not supported by further analyses for several
reasons. First, the average geometric annual growth rate in sales is even higher
in the early part of the period (1950}1980) than in the most recent one
(1981}1998), 13.5% vs. 6.9%. Second, the results remain intact when we use sales
or the change in sales rather than total assets to de#ate the accumulation.
The results for accrual accumulation using alternative de#ators are presented
in Table 4. The table shows the mean (and median) annual cumulative accruals
in the "rst and second subperiods. Note that the accumulation of both total and
nonoperating accruals is signi"cantly more negative in the later subperiod even
The average (geometric) annual growth rate in nominal sales of the constant sample "rms
during the 49-year period from 1950 to 1998 is 9.6%.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 307
Table 4
Accumulation of de#ated accruals, by subperiod (constant sample of 896 "rms)
Total Accruals (before depreciation): (Net Income # Depreciation) !Cash Flows from Opera-
tions. Nonoperating Accruals: Total Accruals (before depreciation) ! Operating Accruals,
where Operating Accruals equal *Accounts Receivable#*Inventories # *Prepaid Expenses
} *Accounts Payable ! *Taxes Payable.
t-values for the di!erence in means in the two subperiods are shown in parentheses.
Observations where sales declined were excluded from the analysis.
after controlling for growth through several alternative de#ators. The table also
shows that, in contrast to the pattern of accruals, de#ated operating cash #ows
remain fairly constant over the entire period.
Further, the accumulation of negative accruals could not be due to in#ation.
Note that there is no discernible accumulation of negative accruals in the 1960s
or the 1970s. Yet, the mean annual in#ation rate in these years (4.9%) was even
higher than in the later period from 1981}1998 (4.1%) where the major accumu-
lation is observed.
In summary, the above tests are inconsistent with several alternative explana-
tions for the decline in the reported pro"tability and the increase in the
accumulation of negative accruals over the last two decades, making it more
likely that increased reporting conservatism is the explanation for these
phenomena.
Table 5 shows the average annual values of each of the four measures based
on the earnings}return association obtained from regression (1) over "ve-year
Table 5
The di!erential earnings}return association in good and bad news periods: 308
Results by subperiod for Regression (1): (EPS /P "a #a DR # b R #b R* DR ) (constant sample of 896 "rms)
GR GR\ GR GR GR GR
Total bias
in earnings
(de#ated by
Subperiod N a a b b Adj R (b #b )/b R /R lagged price)
Overall period 23,612 0.077 0.002 0.048 0.133 7.54 3.77 3.92 0.011
(58.51) (1.01) (17.77) (15.05)
1950}1955 544 0.095 0.005 0.026 0.019 7.56 1.73 0.71 0.002
(25.13) (0.60) (4.88) (1.85)
1956}1960 612 0.065 0.012 0.034 0.057 8.01 2.68 1.61 0.010
(23.87) (2.14) (6.45) (2.24)
1961}1965 1,080 0.080 0.006 0.023 0.074 13.63 4.22 1.95 0.011
(38.93) (1.78) (5.18) (3.65)
1966}1970 1,978 0.063 0.005 0.025 0.085 12.58 4.28 2.22 0.013
(30.70) (1.86) (8.73) (5.86)
1971}1975 2,930 0.092 0.006 0.052 0.199 10.83 4.83 2.89 0.012
(20.20) (1.85) (9.43) (6.28)
1976}1980 3,616 0.142 0.005 0.043 0.292 10.95 7.79 2.73 0.025
(43.08) (1.81) (6.63) (8.90)
1981}1985 3,658 0.082 0.015 0.020 0.346 9.64 18.30 8.10 0.023
(22.84) (2.14) (3.75) (7.19)
1986}1990 3,705 0.068 0.005 0.025 0.449 12.09 18.96 18.25 0.026
(19.36) (1.79) (2.83) (12.02)
1991}1998 5,481 0.047 0.005 0.021 0.521 9.41 25.81 22.17 0.031
(16.73) (1.96) (3.82) (12.87)
EPS is the earnings per share of "rm i in "scal year t, P is the price per share at the beginning of the "scal year, R is the return of "rm i from nine
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
GR GR\ GR
months before "scal year-end t to three months after "scal year-end t, DR is a dummy variable that is equal to 1 if R is negative and 0 otherwise. `Total
GR GR
biasa is measured as (1/k!b ) R Pr (good)!(b #b !1/k) R Pr (bad), R (R ) where R (R ) is the mean return over a good (bad) news
period, de"ned as a period with a positive (negative) return and Pr (bad) (Pr (good)) is the relative frequency of good (bad) news periods. The parameter
k was estimated by the reciprocal of the intercept of regression (1).
The most extreme 1% EPS/P values are truncated each year.
t-values are shown in parentheses.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 309
intervals. The results indicate that "nancial reporting is, in general, conservative
in that it defers recognition of good news and accelerates the recognition of bad
news. More importantly, there is a considerable and signi"cant increase over
time in the degree of conservatism as captured by these measures.
The "rst measure, b , the incremental response to bad news relative to good
news, is positive and signi"cant in the subperiods spanning 1950}1998, sugges-
ting a di!erential stock price response in periods of bad and good news. Further,
there is a marked increase in b over time. The coe$cient's value increases
steadily from the range of 0.02 to 0.08 in the early years through 1975 to 0.449 in
the subperiod 1986}1990 and to 0.521 in the last subperiod, 1991}1998. The
hypothesis that the value of the annual b is the same in the later years as in the
earlier years is rejected at the 1% signi"cance level. These results suggest that,
consistent with accounting conservatism, earnings re#ect bad news more
promptly than good news. Further, consistent with an increased level of conser-
vatism, the di!erential response time to bad news relative to good news has
become more pronounced in recent years.
The second measure, (b #b )/b , which assesses the sensitivity of earnings
to bad news relative to their sensitivity to good news, shows a similar pattern.
Speci"cally, the ratio is always above one, indicating the greater propensity to
recognize bad news in a more timely manner than good news. Further, the &over-
recognition' of bad news is more pronounced in later years than in early years.
The ratio averages 1.73 in the earliest subperiod and increases almost monotoni-
cally to 4.83 in the years 1971}1975 and to over 25 in the more recent years.
In a similar vein, the third measure based on regression (1), the ratio of R
in bad news periods to R in good news periods, increases considerably over
the 49-year period, from 0.71 in the earliest subperiod to over 20 in the
1990s.
The &total bias' measure, which expresses the net downward e!ect on earnings
due to the over-recognition of bad news and the under-recognition of good news
as a fraction of the true earnings, is shown in the last column of Table 5. The
total (downward) bias in the earnings-to-price ratio is on average 0.011. Given
that the average earning-to-price ratio across "rms and years in our sample is
0.09, an average bias of 0.011 suggests an average understatement of about 12%
in earnings (0.011/0.09). The bias, however, increases over time. In the earliest
subperiod the bias is only 0.002 (or about 2% of earnings) but it increases fairly
steadily to 0.031 (or about 30% of earnings) in the latest subperiod. These bias
results are in line with the "ndings regarding the accumulation of negative
accruals. They also con"rm previous studies' "ndings on accounting conserva-
tism (Ball et al., 1999, Basu, 1997; Holthausen and Watts, 2000; Pope and
Walker, 1999; Stober, 1996).
In summary, the four measures based on regression (1), which capture various
aspects of the timeliness with which bad news is reported relative to good news,
consistently point to an increase in reporting conservatism over time.
310 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
4.5.1. Results
Table 6 provides results on the variability of the earnings distribution, our
next measure of conservatism. Earnings have become increasingly dispersed,
with the standard deviation of ROA rising from an average of about 0.060 in the
years prior to 1971 to an average of 0.517 (0.287) in the 1990s for the full
(constant) sample.
A similar result (not shown) is obtained when the time series of earnings of
individual "rms are considered. The mean ratio of the standard deviation of
ROA during the later subperiod (1981}1998) to the standard deviation of ROA
during the early subperiod (1951}1980) for the constant sample of "rms is 3.07,
signi"cantly (at the 1% level) above 1.0. Further, about 75% of the "rms
As in the earlier tests, the test of increased variability is based on the comparison of each of the
nine subperiods with the subperiod that preceded it. The F-statistics are signi"cant across all paired
comparisons for the full and constant samples at the 1% signi"cance level, except for the comparison
between the subperiods 1961}1965 and 1966}1970 for the constant sample.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 311
Fig. 2. Time-series and cross-sectional skewness measures of earnings and cash #ows (constant
sample of 896 "rms). Skewness is de"ned as y"[E(x!k)/p where k and p are the mean and
standard deviation of the x distribution. All variables are de#ated by Total Assets. Values shown for
each year are the "ve-year moving average of the skewness measure, centered on that year. The
cross-sectional value shown for each year is the average value of the skewness measure computed
across the sample "rms. The time-series value shown for each year is the average sample value of the
skewness measure computed for each "rm. The skewness measure for each "rm in any given year is
based on the time series consisting of eleven annual observations centered on that year.
Table 6
Standard deviation of net income-to-total assets (ROA) and cash #ows from operations-to-total
assets (CFOA), by subperiod
The most extreme (0.5%) of the cases at either end of the distribution each year were truncated.
Variance(ROA)"Variance(CFO/Assets)#Variance(Accruals/Assets)
The tests based on the standard deviation of the level variable were replicated using the
standard deviation of the changes with essentially similar "ndings.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 313
Fig. 3. Decomposition of the variance of net income into cash #ow and accrual components
(constant sample of 896 "rms). All variables are de#ated by Total Assets. Values shown for each year
are the "ve-year moving average of the variable, centered on that year.
cash #ows. Note that in more recent years, the variance of the earnings measure
exceeds that of the cash #ows measure.
A closer look at the contributors to the increased earnings variability is
possible by representing the variance of total accruals as the variance of the sum
of nonoperating accruals and all other accruals (see Section 4.2.2 for the
de"nition and discussion of nonoperating accruals). An examination of
the behavior of these components over time (not shown here) reveals that the
variance of the nonoperating accruals is the greatest contributor to the increase
in earnings variability in recent decades. The results from further analysis (not
reported) indicate that the greater earnings and accrual variability is due
primarily to the rising incidence of one-time large nonoperating accruals (as
documented by Elliot and Shaw (1988), Collins et al. (1997) and Givoly and
Hayn (1994)) and the evidence noted in footnote 24).
The results are essentially the same when this analysis is conducted using income from
continuing operations (before discontinued operations, extraordinary items and the cumulative
e!ect of accounting changes) instead of net income. In this case, total accruals (and nonoperating
accruals) exclude these &below-the-line' items.
314 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
Fig. 4. Aggregate market-to-book ratios (constant sample of 896 "rms). The aggregate market-to-
book ratio is the aggregate market value of the constant sample "rms divided by their aggregate
book value at year end. The aggregate adjusted market-to-book ratio re#ects the adding-back of the
nonoperating accruals in the denominator.
The use of the sample aggregate M/B ratio rather than the simple average ratio across
individual companies has the advantage of being independent of the cross-sectional variance in this
ratio (which a!ects the mean ratio). Nonetheless, results based on the simple mean (equally
weighted) M/B ratio show a similar pattern.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 315
U-shaped curve, with the ratio declining from a level of about 2.0 in the years
through the early 1970s, falling to slightly above 1.0 during the remaining 1970s,
rising again to 2.0 by the end of the 1980s, and increasing to over 3.0 in the
1990s. These "ndings, which are consistent with those of Basu (1997) and
Stober (1996), support the hypothesized increase in conservatism.
As a further indication of the extent to which the increase in the M/B ratio in
the second subperiod is related to reporting conservatism, we re-estimated the
ratio using as the denominator an estimate of the book values that would have
been reported had there been no accumulation of negative nonoperating ac-
cruals. This adjusted book value was computed as the reported book value plus
the accumulated nonoperating accruals as of the end of the period. The resulting
aggregate adjusted M/B ratio, appearing as the dotted line in Fig. 4, shows
relatively little change in its level from the mid-1970s to 1998.
The observed variations in the M/B ratio over time might, however, re#ect
changes in the market expectations of growth rather than a change in the degree
of reporting conservatism. To control for growth expectations, we ranked all
"rm-years by expected growth rates and measured the median M/B ratio in each
portfolio for each of the two subperiods 1968}1980 and 1981}1998. Two
alternative proxies were used for the expected growth. One was the annual
geometric growth rate in sales of the "rm over the succeeding "ve years (years
t#1 to t#5) and the other was the actual annual geometric growth rate in the
preceding "ve years (years t!5 to t).
Table 7 shows that, as expected, the M/B ratio is higher in the second
subperiod relative to the "rst subperiod (a median ratio of 1.67 vs. 1.30). Further,
the ratio within each subperiod increases with expected growth. More relevant
to our examination, the M/B ratio for any growth portfolio is greater in the
second subperiod than in the "rst subperiod. Except for the highest growth
portfolio, all di!erences are statistically signi"cant at the 1% level.
We replicated these results using the adjusted book value in the computation
of the market-to-book ratio. The "ndings, reported in the last two columns of
Table 7, reveal that after controlling for growth, the median adjusted M/B ratio
in the early subperiod is not discernibly di!erent from the median adjusted M/B
Basu (1997) "nds a drop in the M/B ratio in the mid-years and an increase in this ratio in the
later years, consistent with changes in the audit liability environment.
Underlying the forward-looking approach is the assumption that investors project future
growth without bias. The backward-looking estimate assumes that investors project future growth
based on past growth rates.
The results in Table 7 are based on growth expectations estimated each year from the preceding
"ve-year growth rate. Use of subsequent years to estimate the expected growth rate led to essentially
the same conclusions.
316
Table 7
Market-to-book ratios (constant sample of 896 "rms)
Median growth rate (%) Median market-to-book ratio Median adjusted market-to-book ratio
First subperiod Second subperiod First subperiod Second subperiod First subperiod Second subperiod
Growth portfolio 1968}1980 1981}1998 1968}1980 1981}1998 1968}1980 1981}1998
HThe means of the two subperiods are signi"cantly di!erent at the 1% con"dence level.
HHThe means of the two subperiods are signi"cantly di!erent at the 5% con"dence level.
Estimated as the geometric mean annual growth rate in sales over the preceding "ve-year period.
The adjusted book value is computed as the reported book value less the accumulated nonoperating accruals through the end of the period. Since these
accruals are generally negative, the adjusted book values are typically larger than the reported ones.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 317
in the second subperiod (1.24 vs. 1.26). Thus, the higher M/B ratios observed in
recent years may be driven by depressed book values.
The adjusted M/B test may overstate the contribution of increased reporting
conservatism to the increase in the M/B ratio over time. Note that the &true'
adjusted book value at the end of the period is likely to be lower than our
estimate (which is the reported book value plus the accumulated negative
nonoperating accruals at that time). This overstatement is due to the potential
substitution between certain nonoperating accruals and future operating ex-
penses. For example, a write-o! of a depreciable asset in the current period leads
to lower depreciation charges in subsequent periods unless the written o! asset
is replaced. Likewise, the establishment of a provision for future costs (another
negative discretionary accrual) in one period results in lower recorded expenses
in subsequent periods. It is di$cult to assess the extent of the upward bias in our
adjusted book values, yet it may be small. New depreciable assets that tend to at
least maintain the previous level of depreciation are likely to replace most of the
assets written o!. Furthermore, because many of the provisions for future
expenses are &settled' only over a long period (e.g., post-retirement bene"ts), this
bias may not be very pronounced.
To further assess the role of accruals in the observed increase in the M/B ratio,
we examine whether the extent of the increase in each "rm's M/B ratio between
the two subperiods is associated with the acceleration in the accumulation of the
"rm's (negative) nonoperating accruals in the second subperiod. The results (not
shown) indicate that the "rms with the greatest accumulation of negative
nonoperating accruals in the second subperiod relative to the "rst subperiod
(portfolio 1) are also those with the highest M/B ratios in the second subperiod
and the greatest increase in the M/B ratio between the two subperiods.
inconsistent with the data, it is possible that other factors contribute to the
results. The absence of an acceptable de"nition of conservatism and proven
measures to gauge its level make it di$cult to make more de"nitive statements.
Nonetheless, more conservative accounting would suggest that the high earn-
ings and book value multiples in today's stock market and the prevalence of
losses do not necessarily indicate overpricing.
It has often been suggested that the increase in the M/B ratios over time is due
to the failure of conventional accounting to properly measure major value
drivers, such as investment in intangibles like R&D, human resources or
information technology (Amir and Lev, 1996). The e!ect of these o!-balance-
sheet intangibles on the M/B ratios is, however, quite distinct from the phenom-
enon of conservatism as measured in this paper. While the immediate expensing
of the investment in intangibles re#ects conservatism, it does not generate any
accruals. Therefore, neither the level of these investments nor their increased
prominence in recent years can explain the signi"cant accumulation of negative
accruals. Thus, the manifestations of increased conservatism documented by this
study are in addition to the conservative e!ect on the "nancial statements
stemming from the omission from the books of "rms' increasingly prominent
investment in intangibles.
References
Amir, E., Lev, B., 1996. Value relevance of non-"nancial information: The wireless communications
industry. Journal of Accounting and Economics 22, 3}30.
Ball, R., Kothari, S.P., Robin, A., 1999. The e!ect of international institutional factors on properties
of accounting earnings. Working Paper, University of Rochester.
Barth, M.E., Beaver, W.H., Landsman, W.R., 1998. Relative valuation roles of equity book value and
net income as a function of "nancial health. Journal of Accounting and Economics 25, 1}34.
Basu, S., 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal of
Accounting and Economics 24, 3}37.
Beaver, W.H., Lambert, R., Morse, D., 1980. The information content of security Prices. Journal of
Accounting and Economics 2, 3}28.
Berton, L., 1992. Legal-liability awards are frightening smaller CPA "rms away from audits. The
Wall Street Journal, March 3, B1, col 3.
Carslaw, C.A., Mills, J.R., Developing ratios for e!ective cash #ow statement analysis. Journal of
Accountancy, November 1991, pp. 63}69.
Cole, K., Helwege, J., Laster, D., 1996. Stock market valuation indicators: is this time di!erent?
Financial Analysts Journal 52, 56}64.
Collins, D.W., Hribar, P., 1999. Errors in estimating accruals: implications for empirical research.
Working paper, University of Iowa.
Collins, D.W., Kothari, S.P., 1989. An analysis of intertemporal and cross-sectional determinants of
earnings response coe$cients. Journal of Accounting and Economics 11, 143}181.
Collins, D.W., Maydew, E.L., Weiss, I.S., 1997. Changes in the value relevance of earnings and book
values over the past forty years. Journal of Accounting and Economics 24, 39}67.
Davidson, S., Stickney, C.P., Weil, R.L., 1985. Intermediate Accounting: Concepts Methods and
Uses, 4th Edition. Dryden Press, Forthworth.
D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320 319
Davis, M.L., 1990. Di!erential market reaction to pooling and purchase methods. The Accounting
Review 65, 696}709.
Dreman, D., 1998. It's tulip time on Wall Street. Forbes 161 (10) (May 18) 325.
Easton, P.D., Harris, T.S., 1991. Earnings as an expectation variable for returns. Journal of
Accounting Research 29, 1}86.
Elliott, J.A., Hanna, J.D., 1996. Repeated accounting write-o!s and the information content of
earnings. Journal of Accounting Research 34 (Suppl.), 135}155.
Elliott, J.A., Shaw, W.H., 1988. Write-o!s as accounting procedures to manage perceptions. Journal
of Accounting Research 26 (Suppl.), 91}119.
Feltham, G.E., Ohlson, J.A., 1995. Valuation and clean surplus accounting for operating and
"nancial activities. Contemporary Accounting Research II (2) Spring, 689}731.
Francis, J., Hanna, J.D., Vincent, L., 1996. Causes and e!ects of discretionary asset write-o!s.
Journal of Accounting Research 34 (Suppl.), 117}134.
Francis, J., Schipper, K., 1999. Have "nancial statements lost their relevance? Journal of Accounting
Research 37 (2), 319}352.
French, K., Poterba, J.M., 1991. Were Japanese stock prices too high? Journal of Financial
Economics 29, 337}363.
Givoly, D., Hayn, C., 1994. Special items } information content and earnings manipulations.
Working paper, University of California at Irvine.
Gray, S.J., 1980. International accounting di!erences from a security analysis perspective. Journal of
Accounting Research 18, 64}76.
Harris, T.S., Lang, M., Moller, H.S., 1994. The value relevance of German accounting measures: an
empirical analysis. Journal of Accounting Research 32, 187}209.
Hayn, C., 1995. The Information content of losses. Journal of Accounting and Economics 20, 125}154.
Healy, P.M., Palepu, K.G., Ruback, R.S., 1992. Does corporate performance improve after mergers?
Journal of Financial Economics 31, 135}175.
Holthausen, R.W., Watts, R.L., 2000. The relevance of the value relevance literature for "nancial
accounting standard setting. Working paper, University of Rochester.
Hughes, P., Sankar, M., 1998. The impact of litigation risk on discretionary disclosure. Working
paper, University of California, Los Angeles.
Jan, C., Ou, J., 1994. The role of negative earnings in the valuation of equity stocks. Working paper,
New York University.
Joos, P., Lang, M., 1994. The e!ects of accounting diversity: evidence from the European Union.
Journal of Accounting Research 32 (Suppl.), 141}168.
Kasznik, R., Lev, B., 1995. To warn or not to warn. The Accounting Review 10, 113}134.
Kothari, S.P., Lys, T., Smith, C.S., Watts, R.L., 1988. Auditor liability and information disclosure.
Journal of Accounting Auditing and Finance 4, 307}339.
Krishnan, J., Krishnan, J., 1997. Litigation risk and auditor resignations. The Accounting Review 72,
539}560.
Leftwich, R., 1995. The agenda of the Financial Accounting Standards Board. Working paper,
University of Chicago.
Lev, B., Zarowin, P., 1998. The boundaries of "nancial reporting and how to extend them. Journal of
Accounting Research 37 (2), 353}385.
McNichols, M., 1988. A comparison of the skewness of stock return distributions at earnings
announcement date. Journal of Accounting and Economics 10, 239}273.
Minow, N.N., 1984. Accountants' liability and the litigation explosion. Journal of Accountancy 158,
70}86.
Ohlson, J.A., 1995. Earnings, book values, and dividends in equity valuation. Contemporary
Accounting Research II (2), Spring, 661}687.
Ohlson, J.A., Zhang, X.J., 1998. Accrual accounting and equity valuation. Journal of Accounting
Research 36 (Suppl.), 85}111.
320 D. Givoly, C. Hayn / Journal of Accounting and Economics 29 (2000) 287}320
Pope, P., Walker, M., 1999. International di!erences in timeliness, conservatism and classi"cation of
earnings. Journal of Accounting Research 37 (Suppl.), 53}99.
Rayburn, J., 1986. The association of cash #ows and accruals with security returns. Journal of
Accounting Research 24, 112}133.
Skinner, D.J., 1994. Why "rms voluntarily disclose bad news. Journal of Accounting Research 32,
38}60.
Skinner, D.J., 1997. Earnings disclosures and stockholder lawsuits. Journal of Accounting and
Economics 23, 249}282.
Smith, J.M. Jr., Skousen, F.K., 1987. Intermediate Accounting, 9th Edition. South-Western Publish-
ing, Cincinnati, Ohio.
Speidel, S.L., Bavishi, V.B., 1992. GAAP arbitrage: valuation opportunities in international ac-
counting standards. Financial Analysts Journal. November/December, 58}66.
Stice, J., 1991. Using "nancial and market information to identify pre-engagement factors associated
with lawsuits against auditors. The Accounting Review 66, 516}533.
Stickney, C.P., Weil, R.L., 1994. Financial Accounting: An Introduction to Concepts, Methods, and
Uses, 7th Edition. The Dryden Press } Harcourt Brace College Publishers, Forth Worth, TX.
Stober T.L., 1996. Do prices behave as if accounting book balues are conservative? Cross-sectional
tests of the Feltham}Ohlson [1995] valuation model. Working paper, University of Notre
Dame.
Walters, D., 1992. Liability crisis threatens auditors; lawsuits multiply against accounting "rms as
regulators dig into fraud, bankruptcy cases. Christian Science Monitor 85 (22), Dec. 28, 9 (1).
Watts, R., Zimmerman, J., 1986. Positive Accounting Theory. Prentice-Hall, Englewood Cli!s, NJ.
Wolk, H.I., Francis, J.R., Tearny, M.G., 1989. Accounting Theory: A Conceptual and Institutional
Approach. PWS-KENT Publishing Company.
Wyatt, J., 1996. The incredible, death-defying stock market: love it and fear it. Fortune 134 (12)
(December 23), 110}118.