The Market Pricing of Earnings Quality

Jennifer Francis*
(Duke University)

Ryan LaFond
(University of Wisconsin)

Per Olsson
(Duke University)

Katherine Schipper
(Financial Accounting Standards Board)

We provide large sample evidence on whether the equity and debt markets impound information about
the quality of earnings. We examine eight proxies for earnings quality (four based on the modified Jones
approach to estimating abnormal accruals; three based on the Dechow and Dichev [2002] approach which
relates working capital accruals to cash flows; and one based on a factor analysis of the other seven).
Across all eight metrics, we find that firms with lower quality earnings have higher costs of capital as
evidenced by lower debt ratings, larger realized costs of debt, larger industry-adjusted earnings-price
ratios, larger equity betas, and positive loadings on an earnings quality factor added to one-factor and
three-factor asset pricing regressions. The documented effects are statistically significant and
economically meaningful: the results show, for example, that firms with the best earnings quality enjoy
discounts of 80-160 basis points in their costs of debt and 150-300 basis points in their costs of equity
relative to firms with the poorest earnings quality.

Draft: October 2002

* Corresponding author: Fuqua School of Business, Duke University, Durham, NC 27708. Email address,
[email protected]. This research was supported by the Fuqua School of Business, Duke University and
the University of Wisconsin. Analysts’ earnings forecasts are from Zacks Investment Research. The
views expressed in this paper are those of the authors. Official positions of the Financial Accounting
Standards Board are arrived at only after extensive due process and deliberation. We appreciate
comments from workshop participants at Duke, Northwestern, Rochester, the Southeast Accounting
Research Conference 2002, and the Stockholm Institute for Financial Research Conference, and from Jan
Barton, Sudipta Basu, George Benston, Larry Brown, Stephen Brown, Marty Butler, Qi Chen, John
Graham, John Hand, Grace Pownall, Michael Smith, Charles Wasley, Ross Watts, Greg Waymire, Joanna
Wu and Jerry Zimmerman.
 The Market Pricing of Earnings Quality

1. Introduction

This study investigates the relation between earnings quality and the costs of debt and equity

capital for a large sample of firms over the period 1988-1999. We find that earnings quality metrics are

statistically associated with ex-ante and ex-post costs of debt, industry-adjusted earnings-price ratios, and

systematic risk from the capital asset pricing model and the Fama-French [1993] three-factor model.1 In

all cases, the results show statistically significant and economically meaningful adverse consequences of

low earnings quality.

Because the earnings quality metrics we use capture the combined effects of managerial decisions

and predetermined economic factors, we do not attempt to separate the cost of capital effects associated

with management’s discretionary accounting and governance choices from effects associated with innate

(and largely uncontrollable) features of the economic environment. Rather, our aim is to document a key

capital market effect of earnings quality, regardless of the specific determinants of that quality. We

interpret our results as documenting cost of capital effects that are consistent with rational asset pricing,

specifically, that earnings quality is a nondiversifiable information risk factor that is priced by investors in

ways that are similar to the pricing of other observable risk factors such as size and book-to-market ratios.

The earnings quality (EQ) metrics we consider are based on two distinct but related measurement

approaches. The first approach is based on Jones’ [1991] separation of total accruals into its normal

component (accruals statistically associated with changes in revenues, and property, plant and equipment)

and its abnormal component (the difference between total and normal accruals). This measurement

approach assumes that accruals shift with accounting fundamentals as captured by revenues and fixed

assets, with deviations from this relation capturing abnormal accruals. The Jones model has been used in

investigations of earnings management, as manifested by the behavior of abnormal accruals at or around a

specific event or in a specific context (e.g., import relief investigations, Jones [1991]; share offerings,

1
We use conventional measures of (or proxies for) the unobservable cost of capital. For simplicity, we refer to
these measures as cost of debt and cost of equity, respectively.

1
Teoh, Welch and Wong [1998]; meeting previously-issued earnings forecasts, Kasznik [1999]). Earnings

management research tends to focus on the signed abnormal accrual because the research context

typically generates a directional prediction about earnings management. In contrast, the Jones-based

earnings quality metrics that we examine focus on the unsigned abnormal accrual, which we interpret as

an inverse indicator of earnings quality.2 The second approach we use to assess earnings quality is based

on Dechow and Dichev’s [2002] model which posits a relation between current period working capital

accruals and operating cash flows in the prior, current and future periods. In this framework, working

capital accruals reflect managerial estimates of cash flows, and the extent to which those accruals do not

map into cash flows (due to intentional and unintentional estimation errors) is an inverse measure of

earnings quality.

Our main tests examine the relation between eight EQ metrics (four based on the Jones model,

three based on the Dechow-Dichev model, and one based on a factor analysis of the other seven) and

firms’ costs of debt and equity capital. We find that firms with lower quality earnings have lower debt

ratings and higher ratios of interest expense to interest-bearing debt than firms with higher quality

earnings (all differences significant at the .001 level). Controlling for other variables known to affect debt

costs (leverage, firm size, return on assets, interest coverage, and earnings volatility), the results suggest

that firms with the best earnings quality enjoy an 80-160 basis point lower cost of debt relative to firms

with the worst earnings quality. In terms of the cost of equity, our most direct tests focus on the effects of

earnings quality on loadings from asset pricing models.3 We find that betas from a one-factor capital

asset pricing model increase monotonically across earnings quality quintiles, with a difference in betas

between the lowest and highest quintiles of 0.25-0.51 (significant at the .001 level). Assuming a 6%

market risk premium, this difference implies a 150-300 basis point higher cost of equity capital for firms

with the worst earnings quality relative to firms with the best earnings quality. Controlling for size and

2
Other studies that examine unsigned abnormal accruals include Warfield, Wild and Wild [1995], Becker, DeFond,
Jiambalvo and Subramanyan [1998], Bartov, Gul and Tsui [2000], and Klein [2002].
3
Indirect tests show that firms with lower earnings quality have significantly (at the .001 level) larger earnings-price
ratios relative to their industry peers; that is, a dollar of earnings commands a lower price multiple when the quality
of those earnings is low.

2
book-to-market factors (in addition to the market risk premium), we find that earnings quality adds

significantly to the three-factor model in explaining variation in returns. We conclude that earnings

quality not only influences the loadings on other documented risk factors, but contributes significant

incremental explanatory power over and above these factors.

As noted earlier, our results do not speak to what portion of the cost of capital effects is

attributable to firm-specific governance and accounting choices and what portion is attributable to

features of firms’ operating environments. While the Jones-type model was developed to identify

management’s intentional estimation errors, research indicates that this identification is imperfect (e.g.,

Dechow, Sloan and Sweeney [1995]).4 In addition, earnings quality measures based on the Dechow-

Dichev model reflect accruals estimation errors from all sources, including firm-specific accounting and

governance choices, managerial expertise and business fundamentals. In fact, Dechow-Dichev report

statistically reliable associations between their measure of earnings quality and firm characteristics such

as length of operating cycle and firm size. We interpret the cost of capital premiums we document for

poor earnings quality firms as arising from some combination of discretionary actions and innate features

of firms’ operating environments. We note, however, that all of our results are robust to the inclusion of a

measure of the underlying volatility of earnings.

Our results contribute to three streams of research. The first investigates the capital market

effects of financial reporting. Research has documented adverse capital market consequences (in the

form of shareholder losses) when earnings are of such low quality as to attract regulatory or legal

attention.5 However, research on severely low earnings quality firms does not establish a general relation

between earnings quality and capital market consequences. Our results show that earnings quality has

4
The EQ metrics based on the Jones, or modified Jones, model reflect deviations from industry-matched (and
industry- and performance-matched) accruals models. While these benchmark models attempt to control for
underlying features of the firm’s operating environment and changes in economic conditions, there is no assurance
that they achieve this goal.
5
Previous research has documented severe economic consequences for earnings of sufficiently low quality as to
attract SEC enforcement actions (Feroz, Park and Pastena [1991]; Dechow, Sloan and Sweeney [1996]; Beneish
[1999]), shareholder lawsuits (Kellogg [1994]; Francis, Philbrick and Schipper [1994]), or restatements (Palmrose,
Richardson and Scholz [2001]). In addition, the financial press provides ample anecdotal evidence of catastrophic
shareholder losses associated with the (arguably) lowest quality earnings, those resulting from financial fraud.

3
economically meaningful consequences for broad samples of firms, unconditional on external indicators

of extremely poor earnings quality.

The second stream of research explores a different, and explicitly anomalous, form of capital

market effects of accruals. This research shows that investors overprice both accruals and abnormal

accruals, as indicated by trading strategies showing that firms with relatively (high) low magnitudes of

signed accruals, or signed abnormal accruals, earn (negative) positive risk-adjusted returns (Sloan [1996];

Xie [2001]; Chan, Chan, Jegadeesh and Lakonishok [2001]; Richardson, Sloan, Soliman and Tuna

[2002]).6 In addition to the difference in metrics studied (we study unsigned metrics, while anomaly

research studies signed metrics), our paper differs from this research in at least two ways. First, anomaly

research takes the perspective that abnormal returns associated with observable firm attributes arise from

slow or biased investor responses to information. An alternative perspective—and one that is more

consistent with the perspective we take—is that observable firm characteristics (such as size and book-to-

market ratio) are proxies for underlying, priced risk factors. Second, our tests are not structured to shed

light on whether investors fully incorporate factors related to earnings quality in their pricing decisions.

That is, our results speak to whether investors price the factor captured by our EQ metrics, and we do not

attempt to determine whether the pricing function used by investors is fully rational. In that sense, we

document a capital market consequence of earnings quality that requires relatively weak assumptions

about investor rationality.

Finally, our paper contributes to research which investigates how the supply of information

affects the cost of capital. Easley and O’Hara [2001] develop a multi-asset rational expectations model in

which the private versus public composition of information affects required returns and thus the cost of

capital. In their model, relatively more private information increases uninformed investors’ risk of

holding the stock, because the privately informed investors are better able to shift their portfolio weights

6
Subsequent research has probed the extent to which factors such as ownership by institutions and analyst following
reduce the anomalous pricing. Desai, Rajgopal and Venkatachalam [2002] summarize this literature and, in related
tests, provide evidence that the “accruals anomaly” is related to, and subsumed by, cash-flow-to-price ratios. By
anomaly we mean a systematic pattern in average returns not explained by the one-factor capital asset pricing model
(Fama and French [1996]).

4
to take advantage of new information. Uninformed investors thus face a form of systematic information

risk, and will require higher returns (charge a higher cost of capital) as compensation. Firms can affect

the required return by reducing the extent of private information or by increasing the precision (quality) of

both public and private information. This second effect is pertinent to our analysis, in that higher quality

earnings provide a more precise signal to investors. Indeed, Easley and O’Hara explicitly note an

important role for precise accounting information in reducing the cost of capital by decreasing the

(information-based) riskiness of shares to uninformed investors. In a companion empirical paper, Easley,

Hvidkjær and O’Hara [2002] find results that are broadly consistent with the prediction that firms with

more private information and less public information have larger expected excess returns. Our analysis

complements their research by considering a second implication of Easley and O’Hara’s model, namely,

that more precise (higher quality) accounting information reduces the cost of capital.7

Empirical accounting research on the relation between information and the cost of capital has

often focused on the quantity of information or on information measures that broadly reflect

quality/quantity attributes of information communicated to investors. For example, Botosan [1997] finds

evidence of a higher cost of equity for firms with low analyst following and relatively low disclosure

scores, where the scores capture information quantity (the existence of specific pieces of information in

annual reports) not the precision or quality of the information. Research has also found a relation

between both the cost of equity (Botosan and Plumlee [2002]) and the cost of debt (Sengupta [1998]) and

analyst-based (AIMR) evaluations of aggregate disclosure efforts, where the evaluations take into account

annual and quarterly reports, proxy statements, other published information and direct communications to

7
Penman and Zhang’s [2002, section VI] trading rule tests provide indirect evidence about market pricing of
quality. Specifically, controlling for risk proxies, they show a positive association between subsequent year returns
and current year Q scores; the latter captures the combined effects of changes in investment and accounting
conservatism, and is measured as the relative ratio of LIFO reserves, uncapitalized R&D expense and uncapitalized
advertising expense to net operating assets. Although Penman and Zhang refer to Q as a measure of quality, they
note that Q does not ordinally rank firms on earnings quality; that is both large positive and large negative Q scores
indicate poor earnings quality, while Q scores of zero indicate high earnings quality. In addition, it is not clear that
Easley and O’Hara’s notion of quality (which relates to the precision of information) is reflected in Q scores (which
capture the combined effects of changes in investment and accounting conservatism).

5
analysts. Our analysis adds to this work by providing evidence on the link between the costs of debt and

equity capital and measures of the quality of earnings information.

The rest of the paper is organized as follows. In the next section we describe the earnings quality

proxies examined in the subsequent tests. Section 3 describes the sample and provides descriptive

information on the earnings quality metrics. Section 4 reports tests of factors related to the cost of capital.

Section 5 reports the results of a robustness check examining the relations between changes in earnings

quality metrics and changes in cost of capital measures. Section 6 summarizes the results and concludes.

2. Earnings Quality Metrics

2.1 Approaches to measuring earnings quality

We use two measurement approaches to obtain seven EQ metrics (our eighth EQ metric is the

common factor determined from a factor analysis of the other seven metrics). Both approaches rely on

associations between accruals and accounting fundamentals to separate an accruals measure (either total

accruals or working capital accruals) into normal components and abnormal components. Under our

framework, the larger the unsigned abnormal component of the accruals measure, the lower is earnings

quality. We do not consider measurement approaches which assess earnings quality by reference to

discontinuities in the distribution of earnings outcomes around some target (e.g., Burgstahler and Dichev

[1997]) because these approaches cannot be applied to the majority of the earnings distribution which is

not close to the target.

The first set of EQ metrics is based on abnormal accruals estimated from the Jones [1991] model

as modified by Dechow, Sloan and Sweeney [1995]. Applying the modified Jones approach to our

setting, earnings quality is related to the extent to which accruals are well captured by fitted values

obtained by regressing total accruals on revenues adjusted for receivables and PPE. We identify two

adjustments to the modified Jones approach. First, following Teoh, Welch and Wong [1998], and to

enhance comparability with the Dechow-Dichev based metrics, we estimate an EQ metric based on

abnormal current accruals. Second, based on results in McNichols [2000] and Kothari, Leone and

6
Wasley [2001], we adjust the abnormal accruals measures (and abnormal current accruals measures) for

firm performance, as proxied by return on assets. Altogether we identify four EQ metrics based on the

modified Jones approach: EQ1 = unsigned abnormal accruals estimated from the modified Jones model;

EQ2 = unsigned abnormal current accruals estimated following Teoh et al. [1998]; EQ3 = EQ1, adjusted

for performance; EQ4 = EQ2, adjusted for performance. The estimation of these four metrics is described

in section 2.2.

The second set of EQ metrics is based on Dechow and Dichev’s [2002] model in which earnings

quality is captured by the extent to which working capital accruals map into operating cash flow

realizations. This model is predicated on the idea that, regardless of management intent, earnings quality

is affected by the measurement error in accruals. Intentional estimation error arises from incentives to

manage earnings, and unintentional error arises from management lapses and environmental uncertainty;

however, the source of the error is irrelevant in this approach. Dechow and Dichev’s approach separates

accruals based on their association with cash flows by regressing working capital accruals on cash from

operations in the current period, prior period and future period. The unexplained portion of the variation

in working capital accruals is an inverse measure of earnings quality (a greater unexplained portion

implies lower quality). We define three EQ metrics based on the Dechow-Dichev approach, where EQ5

and EQ6 are based on cross-sectional regressions and EQ7 is based on firm-specific time series

regressions: EQ5 = the absolute value of the firm-specific regression residual; EQ6 = the firm-specific

time-series standard deviation of the regression residuals (we require at least five firm-specific residuals);

and EQ7 = the time-series standard deviation of the regression residuals (we require at least eight firm-

specific residuals).

In principle, applying the Dechow-Dichev model to total accruals would produce earnings quality

metrics more directly comparable to metrics derived from the modified Jones model. As a practical

matter, however, the long lags between non-current accruals and cash flow realizations preclude this

extension. It is an empirical question whether the Dechow-Dichev approach, linking earnings quality to

the working capital-cash flows relation, provides similar or different information about earnings quality

7
than the modified Jones approach linking earnings quality to the association between accruals and

accounting fundamentals.

2.2 Estimates of EQ metrics

Our approaches to measuring earnings quality rely on a measure of accruals. We calculate total

accruals using information from the balance sheet and income statement (indirect approach).8 Our

definitions follow those of Kothari et al. [2001]:

TA j ,t = (∆CA j ,t − ∆CL j ,t − ∆Cash j ,t + ∆STDEBT j ,t − DEPN j ,t )

TCA j ,t = (∆CA j ,t − ∆CL j ,t − ∆Cash j ,t + ∆STDEBT j ,t )
CFO j ,t = NIBE j ,t − TA j ,t
where TA j ,t = firm j’s total accruals in year t,
TCA j ,t = firm j’s total current accruals in year t,
CFO j ,t = firm j’s cash flow from operations in year t,
∆CA j ,t = firm j’s change in current assets (Compustat #4) between year t-1 and year t,
∆CL j ,t = firm j’s change in current liabilities (Compustat #5) between year t-1 and year t,
∆Cash j ,t = firm j’s change in cash (Compustat #1) between year t-1 and year t,
∆STDEBT j ,t = firm j’s change in debt in current liabilities (Compustat #34) between year t-1
and year t,
DEPN j ,t = firm j’s depreciation and amortization expense (Compustat #14) in year t,
NIBE j ,t = firm j’s net income before extraordinary items (Compustat #18) in year t.

To apply the modified Jones model, we estimate the following regression for each of Fama and

French’s [1997] 48 industry groups with at least 20 firms in year t.9

TA j ,t 1 ∆Rev j,t PPE j ,t

= κ1 +κ2 + κ3 + ε j ,t (1)
Asset j ,t −1 Asset j ,t −1 Asset j ,t −1 Asset j ,t −1

8
We use the indirect approach rather than the statement of cash flows (or direct method, advocated by Hribar and
Collins [2002]) because two of our EQ metrics (EQ6 and EQ7) require time-series data. The time-series
requirement means that we cannot calculate these metrics using the direct method for years prior to 1992 (for EQ6)
and prior to 1996 (for EQ7). In addition, Compustat statement of cash flow data are less complete than balance
sheet and income statement data, especially in the early post-SFAS No. 95 period. Because our research question
and research design rely on substantial variation in the EQ metrics, the indirect method is preferred because it
provides for a larger cross-section of observations. As described in note 11, requiring cash flow data reduces the
sample by nearly 90%.
9
Consistent with the prior literature and throughout this section, we winsorize the extreme values of the distribution
to the 1 and 99 percentiles.

8
where Asset j ,t −1 = firm j’s total assets (Compustat #6) at the beginning of year t,
∆Rev j ,t = firm j’s change in revenues (Compustat #12) between year t-1 and year t,
PPE j ,t = firm j’s gross value of property, plant and equipment (Compustat #7) in year t.

The industry- and year-specific parameter estimates obtained from equation (1) are used to

estimate firm-specific normal accruals (as a percent of lagged total assets),

1 (∆Rev j ,t − ∆AR j ,t ) PPE j ,t

NA j ,t = κˆ1 + κˆ 2 + κˆ3 , where ∆AR j ,t = firm j’s change in accounts
Asset j ,t −1 Asset j ,t −1 Asset j ,t −1

receivable (Compustat #2) between year t-1 and year t, and to calculate abnormal accruals in year t,

TA j ,t
AA j ,t = − NA j ,t . The absolute value of abnormal accruals is the first earnings quality proxy,
Asset j ,t −1

AA =EQ1, with larger values of AA indicating lower earnings quality.

To calculate EQ2, the abnormal current accruals version of EQ1, we estimate equation (2) for

each of the 48 Fama-French industries with a minimum of 20 firms in year t:

TCA j ,t 1 ∆Rev j ,t
= γ1 +γ2 + υ j ,t . (2)
Asset j ,t −1 Asset j ,t −1 Asset j ,t −1

We use the parameter estimates from equation (2) to calculate each firm’s normal current accruals (as a

1 (∆Rev j ,t − ∆AR j ,t )
percent of lagged assets), NCA j ,t = γˆ1 + γˆ2 , and then calculate the abnormal
Asset j ,t −1 Asset j ,t −1

TCA j ,t
component, ACA j ,t = − NCA j ,t . EQ2 is the absolute value of the abnormal current accruals
Asset j ,t −1

calculated according to equation (2), ACA =EQ2. Similar to EQ1, larger values of EQ2 indicate poorer

earnings quality.

Our performance-matching procedure first partitions the sample of firms in each of the 48 Fama-

French industries into deciles based on the firm’s prior year return on-assets, ROA, defined as net income

before extraordinary items divided by beginning of year total assets. Performance-adjusted earnings

9
quality metrics are calculated as the difference between firm j’s metric and the median metric for its

industry ROA decile, where the median calculation excludes firm j:10

AA _ PM j ,t = AA j ,t − IndAA j ,t

ACA _ PM j ,t = ACA j ,t − IndACA j ,t

where IndAA j ,t = median abnormal accruals for firm j’s industry ROA decile,
IndACA j ,t = median abnormal current accruals for firm j’s industry ROA decile.

EQ3 and EQ4 are the absolute values of these measures; that is, AA _ PM j ,t = EQ3 and

ACA _ PM j ,t = EQ4.

To calculate earnings quality metrics using Dechow and Dichev’s model, we estimate equation

(3) for each year t for each of the 48 Fama-French industry groups with at least 20 observations:

TCA j ,t = φ 0 + φ1CFO j ,t −1 + φ 2 CFO j ,t + φ 3 CFO j ,t +1 + ν j ,t (3)

These estimations yield firm- and year-specific residuals, which form the basis for the fifth and

sixth earnings quality metrics. EQ5 is the absolute value of firm j’s residual in year t, νˆ j,t . EQ6 is the

time-series standard deviation of these firm-specific residuals, σ (νˆ j ,t ) , calculated using a minimum of

five residualobservations. Consistent with the construction of the other metrics, larger absolute residuals

and larger standard deviations of residuals suggest poorer earnings quality.

Finally, we calculate a seventh earnings quality metric based on firm-specific time-series

estimations of (3). For each year t, we require a minimum of 10 years of data for each firm, yielding at

least eight firm-specific observations to estimate the firm-specific model. EQ7 is the standard deviation

of the firm-specific residuals, σ ( µˆ j ,t ) ,with larger values of σ ( µˆ j ,t ) indicating poorer quality earnings.

Note that EQ7 is both firm- and year-specific because the time-series equation is re-estimated each year.

10
Our implementation of performance matched portfolios differs from Kothari et al.’s approach which matched their
treatment firms (firms with seasoned equity offerings or SEO firms) with the non-SEO firm closest in ROA and in
the same two-digit SIC code. The portfolio adjustment is better suited to our broad sample study where there are no
well-defined treatments (e.g., equity offerings, import relief investigations).

10
 To summarize, we apply methods based on the modified Jones approach and the Dechow-Dichev

approach to separate either total accruals or current accruals into normal components (i.e., the portion

associated with accounting fundamentals) and abnormal components (i.e., accruals that are not

statistically associated with accounting fundamentals). Under the modified Jones model approach,

accounting fundamentals are revenues adjusted for receivables and PPE. Under the Dechow-Dichev

approach, lagged, current and lead cash flows from operations are the accounting fundamentals. Our

seven EQ metrics capture various aspects of the abnormal component of accruals; the weaker the

association between accruals and accounting fundamentals, the lower is earnings quality.

3. Sample and Description of Earnings Quality Proxies

We calculate EQ metrics for t=1988-1999. To be included in any of the market-based tests, and

to facilitate within-firm comparisons across the metrics, we require that each firm-year observation have

the data necessary to calculate each EQ metric. (We find qualitatively similar results if we do not require

firms to have values for all EQ metrics.) Because the most restrictive data requirement is the availability

of at least 10 years of data (for the firm-specific estimations of equation (3)), the sample is (effectively)

restricted to firms with at least 10 annual observations. This restriction likely biases our sample to

surviving firms which tend to be larger and more successful than the population. We expect this

restriction will, if anything, reduce the variation in the EQ metrics, making it more difficult to detect

effects. In total, there are 33,770 firm-year observations with data on all seven EQ metrics.11 The

number of firms each year ranges from about 2,500 observations during 1988-1990 to roughly 3,200

observations per year beginning in 1995. Table 1, Panel A reports summary statistics on the EQ measures

11
The loss in sample size from using the direct method for calculating cash flows and accruals (see note 8) is
substantial. We repeat our calculations of EQ1-EQ7 for all firms with statement of cash flow data on Compustat,
imposing the same estimation requirements as we did in calculating EQ1-EQ7 using the indirect approach. The
number of firm-year observations with all seven EQ metrics is 4,019 (versus 33,770 firm-year observations using
the indirect approach). The pairwise correlations between the direct and indirect EQ metrics range from 0.55 to 0.80
(significance levels of 0.0001), suggesting that, at least for the sample where both methods can be applied, the
earnings quality metrics are highly positively correlated.

11
for the pooled sample. Since all of the metrics capture levels or variation in components of accruals, it is

not surprising that all have a similar mean and median values (the range is about .03 to about .06).

Evidence on the overlap among the seven EQ metrics is provided by the pairwise correlations in

Panel B of Table 1. The four metrics derived from the modified Jones model (EQ1-EQ4) are highly

correlated; pairwise Pearson correlations are at least .83 (Spearman correlations are lower, but all are at

least .68). In contrast, the three metrics derived from the Dechow-Dichev approach are less highly

correlated: the largest Pearson correlation (between EQ6 and EQ7) is .86 and the other correlations are

below .50. Correlations between metrics derived from the two different approaches are also generally

lower: the highest (between EQ2 and EQ5) is .68 and most correlations are about .40 or lower. The

exception is EQ5, which is correlated with EQ1-EQ4 at .57 or more. We conclude from these

correlations that there is substantial overlap in the information contained in the seven EQ metrics, but that

none of the measures seems to fully impound the others.

Panel C reports summary information on the over-time variation in the EQ metrics, specifically,

the cross-sectional distribution of firm j’s rolling five-year standard deviation of EQ(k). (We exclude

firm-year observations with incomplete 5-year data). These data indicate considerable over-time

variability. For example, the mean (median) standard deviation of EQ1 of .0453 (.0335) is about 75%

(85%) of the mean (median) of EQ1, reported in Panel A, of .0597 (.0394). EQ2-EQ5 also show large

over time variability, while EQ6 and EQ7 are more stable.

The evidence in Panels B and C suggests that the seven EQ metrics may be noisy measures of a

single underlying (unobservable) earnings quality factor. To explore this possibility, we conduct a factor

analysis, and identify a common factor that explains over 80% of the variance in the seven EQ metrics.

As shown in Panel D, Pearson correlations between the common factor and each of the modified-Jones

based metrics (Dechow-Dichev based metrics) exceed .91 (range from .57 to .71), with all correlations

significant at the .001 level. We obtain similar results from a maximum likelihood factor analysis (also

reported in Panel D). Based on these results, we conclude there is a common factor underlying EQ1-

12
EQ7; we label it EQ8 and include it in the EQ metrics examined in subsequent tests. We retain EQ1-EQ7

because our results do not indicate that EQ8 fully summarizes the other seven metrics.

Table 2 reports summary information on selected financial variables. The sample firms are large

(median market value of equity is about $142 million and median assets are about $197 million);

profitable (median ROA is about 0.04 and median income scaled by market value is about 0.06); growing

(median log of sales growth is 0.078); and solvent (median ratio of long-term debt to assets is 0.170 and

median bond rating is investment grade). In unreported tests, we compare these sample attributes to those

of the Compustat population for the same time period. Although the sample firms are larger and more

profitable than the typical Compustat firm (the median Compustat firm has a market value of equity of

$78 million, assets of $94 million, and ROA of 1.5%), they have similar growth and solvency (the median

Compustat firm has log sales growth of 0.082 and an investment grade debt rating).

The last two rows of Table 2 report summary data on two measures of earnings volatility: the

coefficient of variation of net income before extraordinary items, CV(NI), and the standard deviation of

earnings per share (adjusted for stock splits), Volatility. We calculate CV(NI) and Volatility using 5-year

rolling windows; for this analysis, we exclude observations with incomplete 5-year data. We prefer

Volatility to CV(NI) as a measure of earnings variability because the latter is not defined over variables

that can take on negative values, such as net income (Blom [1984]). As evidenced by the summary

statistics, CV(NI) is negative for over 25% of the sample firms. Pairwise correlations between each of the

EQ metrics and Volatility (not reported) are reliably (at the .001 level) negative, ranging from –.021 to –

.055. These results suggest that our EQ metrics are not simply proxying for earnings volatility.

Notwithstanding this observation, we control for earnings volatility (where appropriate) in subsequent

tests.

We next provide evidence that the EQ metrics capture two notions of earnings quality used in

prior research: the explanatory power of annual earnings for annual returns, and the response coefficient

13
from a regression of short window abnormal returns on unexpected earnings news.12 Following Gu’s

[2002] argument that the squared residuals from the regression of annual returns on earnings are a better

measure of value relevance than the more-conventional R2 measure, we focus on the squared residuals

obtained from regressions of annual returns on the level and change in annual earnings. For our purposes,

an important feature of the squared residuals is that they provide a firm-specific measure of the extent to

which the estimated sample parameters explain variation in an individual firm’s returns. We expect that

firms with highly relevant (that is, high quality) earnings have small squared residuals, while firms with

less relevant (that is, low quality) earnings have large squared residuals. We formalize this hypothesis by

testing whether λ1 > 0 in the following regression:

εˆ 2j ,t = λ0 + λ1 EQ(k ) j ,t + ζ j ,t (4)

where εˆ 2j ,t = the square of firm j’s residual, obtained from the year t regression of 12-month returns on
the level and change in earnings.
EQ(k ) j ,t = firm j’s value of EQ metric k in year t, k=1,…,8.

Table 3, Panel A shows the results of estimating (4), separately for each EQ metric, using both

pooled regressions and Fama-MacBeth [1973] regressions;13 we also report results of rank regressions

which replace the raw value of each independent variable by its decile rank (firms with the largest EQ

values (i.e., those with the poorest quality earnings) are assigned to decile 10 while those with the

smallest values are in decile 1). The sample contains 26,398 firm-year observations with annual returns

and annual earnings data. The consistent result across both estimation procedures and both specifications

12
Barth, Beaver and Landsman [2001] argue that value relevance, as captured by earnings’ explanatory power for
returns, captures combined relevance and reliability, key concepts in the FASB’s Conceptual Framework. Also, see
Lev [1989], Collins, Maydew and Weiss [1997] and Francis and Schipper [1999]. Kothari [2001, section 4.1.1]
discusses research on earnings response coefficients, including their association with earnings persistence, which has
been used by, for example, Lev and Thiagarajan [1993] as an indicator of earnings quality.
13
The Fama-MacBeth tests calculate t-statistics based on the time-series standard errors of the estimated
coefficients; these tests mitigate concerns that cross-sectional correlations in the data lead to inflated t-statistics for
the pooled regressions.

14
is a strong positive association between squared residuals (indicating lower value relevance of earnings)

and the EQ metrics (indicating lower quality earnings).14

We also expect that earnings quality affects investors’ immediate reactions to unexpected

earnings, because low earnings quality is identified with low persistence (e.g., Lev and Thiagarajan

[1993]) and low persistence reduces earnings response coefficients (e.g., Collins and Kothari [1989],

Easton and Zmijewski [1989]). We estimate the earnings response coefficients (ERC’s) relating the two-

day market reaction to quarterly earnings announcements to the earnings news conveyed in the

announcements, and predict smaller ERC’s for firms with lower quality earnings. That is, we expect

ω 2 < 0 in the following regression:

CAR j ,q = ω 0 + ω1UE j ,q + ω 2UE j ,q * EQ(k ) j ,t −1 + ζ j ,q (5)

where CAR j , q = firm j’s cumulative market-adjusted returns on days (-1,0) where day 0 is the
announcement date of earnings for quarter q;

A j ,q − F j ,q
UE j ,q = = unexpected earnings conveyed in firm j’s quarterly earnings
Pj , −20
announcement; Aj,t = firm j’s reported earnings for quarter q; Fj,t = median analyst
forecast of earnings for quarter q, analysts’ forecasts are from Zacks Investment
Research; Pj,-20 = firm j’s share price 20 days prior to the q’th earnings announcement.

EQ(k ) j ,t −1 = the year t-1 value of firm j’s EQ metric. We use the prior year value of the EQ metric
rather than the contemporaneous value, because the latter may not be known to investors
at the date of the quarterly earnings announcement.

The coefficient estimates and t-statistics associated with equation (5) are shown in Table 3, Panel

B; we report pooled and Fama-MacBeth tests as well as raw and decile rank specifications. The sample

contains 32,149 quarterly earnings announcements with data on the EQ metrics, daily returns and

unexpected earnings. The first row in each panel shows the coefficient estimate on UE, ω1 , excluding all

interaction terms. Consistent with prior studies, the estimates of ω1 show a significant (at the .001 level)

14
The results are not affected by the choice of 12- or 15-month periods to measure returns, the use of raw or market-
adjusted returns, the inclusion or exclusion of change in earnings, the inclusion of interaction terms capturing loss
observations, or the inclusion of variables capturing firm size and market-to-book ratio. Also, we draw similar
inferences using absolute, rather than squared, residuals.

15
positive association between short term market reactions and the news in quarterly earnings

announcements. The results adding UE * EQ(k ) as an independent variable show that all the estimated

ω 2 ’s are negative, although not all are reliably different from zero. Except for two of the Jones-model

based metrics (EQ3 and EQ4) in the pooled raw regression, the pooled results show significantly (at the

.05 one-tailed level) smaller ERC’s for lower quality earnings for both the raw and rank specifications.

The results for the Fama-MacBeth regressions also show negative coefficients on the UE * EQ(k )

terms, although the statistical significance of the estimates is weaker.15

Overall, we interpret the results in Table 3 as evidence that the EQ metrics we examine capture

notions of earnings quality used in prior research. We do not include εˆ 2j ,t or ERC as EQ metrics because

both measures are based on returns data. That is, since our focus is the cost of capital effects of earnings

quality, we wish to avoid using measures of this construct that are calculated from market prices.

4. Earnings Quality and the Costs of Debt and Equity Capital

Our main tests examine the association between earnings quality and three factors related to the

cost of capital: cost of debt (section 4.1), and cost of equity, as captured by earnings-price ratios (section

4.2) and factor loadings in the one-factor and three-factor asset pricing models (section 4.3). For each of

these tests, we merge the sample described in section 3 with all observations with the market and

accounting data dictated by that test. Of the 33,770 firm-year observations with earnings quality metrics,

7,982 firm-year observations have Compustat data on debt ratings (our proxy for the ex ante cost of debt),

22,940 have data on average interest expense as a percent of interest bearing debt (our proxy for the

realized cost of debt) and 19,633 have the necessary data to calculate earnings-price ratios. The primary

sample used in the asset pricing tests focuses on the 4,124 firms with monthly returns data.

15
These results are consistent with DeFond and Park’s [2001] ERC results (which condition both on signed
abnormal accruals and the nature of unexpected earnings news) insofar as both studies find that investors price the
information in abnormal accruals. DeFond and Park find lower ERCs for good news firms that also report income
increasing abnormal accruals than for good news firms with income decreasing abnormal accruals; and they find
lower ERCs for bad news firms reporting income decreasing abnormal accruals than for bad news firms with income
increasing abnormal accruals.

16
4.1. Cost of debt

Our first test examines whether EQ metrics explain variation in the cost of debt. We explore this

association for both an ex ante cost of debt (proxied by Standard & Poor’s [S&P] debt ratings) and an ex

post, or realized, cost of debt (proxied by the ratio of interest expense in year t+1 to average interest

bearing debt in year t+1). Data on S&P Issuer Credit Ratings are taken from Compustat (data item #280).

S&P ratings range from AAA (highest quality) to D (default), which Compustat codes using a numerical

scale. We re-code the Compustat data to remove unassigned and similar codes, as shown in the

Appendix. The re-coded variable, DebtRating , ranges from 1 (AAA rating) to 20 (default).

Evidence on the relation between DebtRating and earnings quality is detailed in Panel A of

Table 4, where we report the mean debt rating for each quintile of the ranked EQ metric distributions.

These data show that the worst earnings quality firms (Q5) have mean debt scores of 10-12

(corresponding to ratings of BBB- to BB), just at or below investment grade, while the best earnings

quality firms (Q1) are consistently well above investment grade. For all EQ metrics, there is a monotonic

increase in debt rating scores over the quintiles, with a significant (at the .001 level) difference between

the mean debt rating scores for the worst and best earnings quality quintile (Q5 versus Q1). These

differentials are economically meaningful. For example, the modified Jones EQ metrics (EQ1-EQ4)

show differential ratings of BBB+ (Q1) versus BBB- (Q5). Even more striking, for EQ6-EQ8, the Q5-Q1

difference corresponds to a rating of A versus BB+/BB .

These effects are likely overstated because the tests do not control for the effects of other factors

known to affect debt ratings: financial leverage, firm size, return on assets, interest coverage, and earnings

volatility (Kaplan and Urwitz [1979]; Palepu, Healy and Bernard [2000]). If earnings quality is not

subsumed by one or more of these factors, and if debt rating agencies view firms with low quality

earnings as riskier than firms with high quality earnings, we expect a positive relation between debt rating

scores and EQ metrics, or θ 6 > 0 in the following regression:

17
DebtRating j ,t = θ 0 + θ 1 Leverage j ,t + θ 2 Size j ,t + θ 3 ROA j ,t + θ 4 IntCov j ,t + θ 5Volatility j ,t + θ 6 EQ ( k ) j ,t + ς j ,t
(6)
where Leverage j ,t = firm j’s ratio of long term debt to total assets in year t.
Size j ,t = log of firm j’s total assets in year t.
ROA j ,t = firm j’s return on assets in year t.
IntCov j ,t = firm j’s ratio of interest expense to net income in year t.
Volatility j ,t = standard deviation of firm j’s earnings per share over the rolling prior 5 year.

Panel B, Table 4 reports the results of estimating equation (6).16 The first five rows show the

coefficient estimates and t-statistics when only the control variables are included. As expected, Leverage

is significantly (at the .001 level) positively correlated with DebtRating , and Size and ROA are

significantly (at the .001 level) negatively related; for our sample, interest coverage ratios and earnings

volatility are insignificantly related to debt ratings. The remaining rows in Panel B show the results of

adding each EQ metric as an independent variable.17 In all cases, these results show that EQ(k) is

positively correlated with DebtRating , with t-statistics ranging from 9.6 to 44.5. The magnitudes of the

coefficient estimates ( θ 6 ’s) from the decile rank regressions suggest the economic importance of these

effects. These estimates, which average about 0.17 for the modified Jones EQ metrics and about 0.40 for

the Dechow-Dichev metrics, suggest a difference of 1.7 and 4 rating categories between the worst and

best EQ deciles. Given that the mean debt rating for firms in the best EQ decile is roughly A, a 1.7 (4)

category difference corresponds to a BBB+ (BBB-) rating for the worst EQ firms.18

We also examine the ability of earnings quality to explain the realized cost of debt ( CostDebt ),

calculated as the ratio of firm j’s interest expense in year t +1 (Compustat data item #15) to average

16
We report the results of simple OLS regressions. Because the dependent variable in equation (6) is discrete and
ordinal, a preferred approach is to estimate expression (6) as an ordered multinomial logit. Results of the latter
approach yield similar inferences and are not reported.
17
For brevity, we do not report the (new) coefficient estimates and t-statistics on the control variables in the
presence of the EQ metrics. In general, neither the magnitude nor the significance of the control variables is
significantly affected by the addition of EQ(k).
18
Applying these ratings differences to corporate bond spreads as of June 2002 (see www.bondsonline.com,
6/11/2002), we find differences in 1-year, 5-year and 10-years spreads of 95, 54 and 57 basis points for A versus
BBB+ bonds and 326, 222 and 158 basis points for A versus BBB- bonds.

18
interest bearing debt outstanding during year t +1 (Compustat data items #9 and #34).19 Panel C, Table 4

reports the results of estimating equation (6) replacing DebtRating with CostDebt , for the 22,940 firm-

year observations with data on all variables. Turning first to the results for control variables, Size is

negatively related to CostDebt, and Volatility displays a weak positive relation; none of the other control

variables enters significantly in the regression. When we add each EQ metric as an independent variable,

we find consistently positive coefficients, with t-statistics ranging from 7.2 to 20.4. The typical decile

rank value of θ 6 for the modified Jones based metrics (EQ1-EQ4) is .07-.08, suggesting a 70-80 basis

point difference between the costs of debt of firms in the best and worst EQ deciles. For the Dechow-

Dichev based metrics (EQ5-EQ7), the decile rank coefficient estimates are larger (0.08-0.16), and imply

an 80-160 basis point premium paid by the worst earnings quality firms.

In summary, the results in Table 4 indicate that earnings quality affects the cost of debt,

incremental to financial leverage, size, return on assets, interest coverage and earnings volatility. The

results are consistent across ex ante and ex post measures of the cost of debt, across estimation procedures

and across specifications. The realized cost of debt regressions suggest a 70-160 basis point differential

between the best and worst earnings quality firms.

4.2 Earnings-price ratios

Following Liu, Nissim and Thomas [2002], we view the multiple attached to earnings as a short-

hand valuation, which places a price on a dollar of earnings. While there is scant empirical evidence on

the relation between earnings quality and the price multipliers attached to earnings, intuition suggests that

lower quality reporting leads to greater risk which (if not diversifiable) results in lower price-earnings

ratios (Penman [2001]). Building on this intuition, we investigate the relation between the EQ metrics

and industry-adjusted earnings-price ratios. We use earnings-price (EP) ratios to address concerns with

the effects of small values of earnings in the denominator, and we industry-adjust based on Alford’s

[1992] finding that industry membership works well for selecting comparable firms.

19
Summary information reported in Table 2 shows a mean (median) cost of debt of 9.2% (8.7%), with 80% of the
sample having a cost of debt between 5.8% and 12.9%.

19
 To calculate industry-adjusted EP ratios, we first calculate the median EP ratio for all firms with

positive earnings in year t in each of the 48 Fama-French industry groups; we require a minimum of five

positive earnings firms in the industry in year t (excluding firm j). We calculate firm j’s industry-adjusted

EP ratio, IndEP j ,t , as the difference between its EP ratio and the median industry EP ratio in year t. (We

draw similar inferences using the ratio of firm j’s EP to the median industry EP.) If investors apply lower

multiples to lower quality earnings, we expect such earnings to be associated with larger EP ratios.

Evidence on the relation between industry-adjusted EP ratios and earnings quality is provided in Panel A,

Table 5, where we report the mean value of IndEP for each quintile of the ranked EQ distributions. For

all EQ metrics, these data show that the poorest earnings quality firms have the largest IndEP, and that the

difference between the mean IndEP for the worst earnings quality quintile (Q5) is significantly (at the

.001 level) larger than the mean for the best earnings quality quintile (Q1).

More formal tests of whether earnings quality explains industry-adjusted EP ratios are shown in

Panel B of Table 5, where we report the coefficient estimates and t-statistics from estimating equation (7):

IndEPj ,t = ϕ 0 + ϕ1Growth j ,t + ϕ 2 EQ(k ) j ,t + ς j ,t (7)

Based on prior research showing that growth is positively related to price-earnings ratios (e.g., Alford

[1992]), equation (7) includes the log of the firm’s average sales growth over the past five years (Growth)

as an independent variable.20 Similar to our previous tests, we first estimate (7) excluding EQ(k) and then

report results with EQ(k). As expected, results for Growth show ϕ1 < 0 (t-statistics of –3.6 and –1.7 for

the pooled and Fama-MacBeth tests, respectively). The remaining rows of Panel B show consistent

evidence (across estimation procedures and specifications) that firms with larger EQ scores have larger

earnings-price ratios: the estimates of ϕ 2 are positive for all eight EQ metrics, with t-statistics ranging

from 5.6 to 13.4. We interpret these results as indicating that as the quality of earnings decreases, so too

20
In unreported tests, we find that earnings volatility (Volatility) is positively correlated with IndEP, but has no
effect on the magnitude or significance of EQ(k).

20
does the amount investors are willing to pay for a dollar of earnings, implying a higher cost of equity

capital to firms with lower quality earnings.

4.3 Factor loadings in one-factor and three-factor asset pricing models

Our final analyses investigate the effects of earnings quality on equity costs of capital, as manifest

in the factor loadings and explanatory power of one-factor and three-factor asset pricing models.21 We

begin by examining whether lower earnings quality is associated with a larger factor loading on

systematic risk (beta) in a traditional one-factor model. If earnings quality is associated with this risk, we

expect a positive association between the EQ metrics and beta; that is, poorer earnings quality is

associated with larger betas and, therefore, higher costs of capital.

To allow for differences in firms’ fiscal year ends as well as over-time changes in earnings

quality, we use a dynamic portfolio technique to assign firms to earnings quality portfolios. Specifically,

beginning in April 1989, we form five portfolios on the first day of each calendar month m based on the

firm’s most recent EQ signals known prior to month m; firms with the smallest (largest) EQ scores are

placed in the first (fifth) portfolio.22 We calculate the average monthly excess return for each portfolio for

the period April 1989 to March 2001, yielding a time series of 144 monthly excess returns for each of the

p=5 portfolios constructed for each EQ metric. The portfolio beta is the coefficient obtained from

regressing each portfolio’s monthly excess return on the monthly excess market return:

21
This approach differs from Barone’s [2002] analysis of the relation between implied costs of equity capital and
two earnings quality metrics (Lev and Thiagarajan’s [1993] fundamental score and a second measure developed
from relations between financial statement line items), conditioning on market value of equity, book-to-market ratio,
growth, beta and leverage. He calculates implied costs of equity using a residual income formula, using as inputs
observed prices, book value of equity, analysts’ earnings and growth forecasts and an assumption about mean
reversion in return on equity. Because these estimates are, by construction, a negative function of market value of
equity and a positive function of book-to-market (where market value, and therefore, book-to-market, are
themselves functions of the cost of equity), it is not clear what incremental role earnings quality plays in explaining
variation (except potentially as a proxy for analysts’ forecasts of earnings). Consistent with these mechanical
relations, Barone reports t-statistics greater than 4 in absolute value on size, book-to-market and growth, and finds
that earnings quality is generally insignificant in the presence of these variables. When the shared variation between
earnings quality and book to market and size is attributed to earnings quality, he finds that the earnings quality
variables enter with t-statistics ranging from –1.9 to –4.6.
22
We assume each EQ signal is known by investors as of the first day of the fourth month following the firm’s fiscal
year-end. For example, for the month of April 1998 firms are ranked into portfolios based on the EQ signals
calculated using annual data for fiscal year-ends between January 1997 and December 1997. This procedure means
that firm j’s EQ signals for year t, where fiscal year t ends in month n, will influence firm j’s ranking for months n+4
through n+15.

21
 R p , m − R F , m = α p + β p ( RM , m − R F , m ) + ε p , m (8)

Panel A, Table 6 reports the betas obtained from estimating equation (8) for each portfolio and

EQ metric. These data show that for nearly all metrics, betas increase monotonically across the

portfolios, from 0.60 to 0.76 for Q1 to 0.99 to 1.06 for Q5. The differences in betas between the Q5 and

Q1 portfolios range from 0.25-0.49, with t-statistics of 4.2-5.1. Assuming a 6% market risk premium, the

far right column of Panel A shows that the Q5-Q1 difference in betas implies that firms with the highest

quality earnings enjoy a 150-300 basis point reduction in the cost of equity capital relative to firms with

the worst quality earnings.

More explicit tests of the effects of earnings quality on the cost of equity capital are conducted

using firm-specific asset-pricing regressions. We begin by estimating one-factor models for each of the

J= 4,124 firms with EQ metrics and at least 12 monthly returns.23 The row labeled “none” of Panel B,

Table 6 reports the mean values of the coefficients and adjusted R2s, and reports t-statistics of whether the

mean coefficient estimate equals zero. The results show an estimated beta of 0.81 (t-statistic of 83.94),

and an adjusted R2 of 8.2%. (The finding of an average beta less than one is expected for our sample of

relatively large, stable firms.) To this traditional CAPM, we add a variable capturing earnings quality.

Specifically, we calculate EQ(k) factor-mimicking portfolios, EQfactor (k ) , equal to the difference

between the monthly excess returns of the top two EQ(k) quintiles (Q4 and Q5) and the bottom two

EQ(k) quintiles (Q1 and Q2). This procedure (similar to that used by Fama and French [1993]) yields a

series of 144 monthly EQfactor (k ) returns for each of the eight EQ mimicking portfolios. The

remaining rows of Panel B show the results of regressions which include each EQfactor (k ) as an

additional independent variable; these tests allow us to assess the degree to which earnings quality

overlaps with and adds to the market risk premium in explaining returns. Specifically, we report the

mean of the J=4,124 loadings, β j and λ j , from firm-specific estimations of equation (9):

R j ,m − RF ,m = α j + β j ( RM ,m − RF ,m ) + λ j EQfactor (k ) m + ε j ,m (9)

23
The mean (median) sample firm has 111 (124) monthly returns.

22
 The mean loadings on the EQfactor (k ) ’s (the λ j ’s) are positive and highly statistically

significant, with t-statistics generally above 30. The estimated betas remain statistically (at the .001 level)

positive in the presence of EQfactor (k ) , but their magnitude is reduced by 15-22% relative to the point

estimate of 0.81 from the regression excluding EQfactor (k ) . While this result suggests that some of the

information in EQfactor (k ) overlaps with the market risk premium, the statistical significance of both

variables indicates that neither beta nor earnings quality subsumes the other. Evidence on the extent to

which EQfactor (k ) adds to explaining returns is provided in the rightmost column of Panel B, where we

report the incremental explanatory power of the EQfactor (k ) , equal to the difference in adjusted R2s

from estimations of (9) which include and exclude EQfactor (k ) . These results show that

EQfactor (k ) increases the average adjusted R2 from 8.2% to 11.8%, or by about 44% (the statistical

significance of this increase is identical to the significance of EQfactor (k ) ).

Based on prior research which shows that earnings volatility is positively correlated with beta

(Beaver, Kettler and Scholes [1970]), we investigate whether the positive loadings on EQfactor (k ) are

explained by volatility. Our test examines the significance of EQfactor (k ) controlling for a volatility

factor-mimicking portfolio ( Volfactor ), equal to the difference between the monthly excess returns of the

top two quintiles of firms ranked on the standard deviation of earnings per share (i.e., those with the

largest standard deviations) and the bottom two quintiles (i.e., the smallest standard deviations). Results

of estimating equation (10) are reported in Panel C, Table 6:

R j ,m − RF ,m = α j + β j ( RM , m − RF , m ) + λ j EQfactor (k ) m + ρ jVolfactor + ε j ,m (10)

Inspection of the coefficient estimates shows that Volfactor is generally not significant in

regressions which include the modified Jones based EQ metrics (EQ1-EQ4), but is significant in

regressions which include the Dechow-Dichev metrics (EQ5-EQ7) or the common factor (EQ8); in all

cases, the factor loadings are small (the mean ρ j ranges between 0.01 and 0.11, with t-statistics of

between 0.30 and 6.20). Most importantly, the inclusion of Volfactor enhances, rather than detracts from,

23
the significance of EQfactor (k ) : factor loadings on EQfactor (k ) increase, as do the associated t-

statistics and incremental R2s. Given both the low correlation between Volatility and EQ(k) and the

absence of any meaningful effects on EQ(k) from adding Volatility in this and prior tests, we conclude

that the EQ metrics are not merely proxying for earnings volatility.

Finally, we investigate the ability of EQfactor (k ) to explain returns by examining its contribution

to the three-factor asset pricing model. This analysis provides evidence on whether EQfactor (k ) proxies

for either or both the size factor (SMB) or the book-to-market factor (HML), both of which have been

shown to be incrementally relevant for asset pricing (Fama and French [1993]). We begin by estimating

the three-factor model for each of the J=4,124 firms:

R j ,m − RF ,m = a j + b j ( RM ,m − R F ,m ) + s j SMBm + h j HMLm + ε j ,m (11)

The row labeled “none” of Panel D reports the mean coefficient estimates and t-statistics. These

results show that each of the factor loadings is highly significant, with t-statistics of 83.89 (b), 44.25 (s)

and 13.55 (h). Together, the three factors explain an average of 12.9% of the total variation in the sample

firms’ excess returns. The remaining rows of Panel D report the mean coefficient estimates and t-

statistics for regressions which include EQfactor (k ) :

R j ,m − RF ,m = a j + b j ( RM ,m − R F ,m ) + s j SMBm + h j HMLm + e j EQfactor (k ) m + ε j ,m (12)

For all EQ metrics, the results show e > 0 , with t-statistics of 11.9-17.7. Inspection of the

incremental R2’s (reported in the far right column) and the changes in the estimates of b, s and h, indicates

that the significance of EQfactor (k ) comes both from additional explanatory power (the average adjusted

R2 increases from 12.9% to 14.5%, an increase of 13%) and from overlap with the other three factors. By

far, the most significant overlap of EQfactor (k ) is with SMB, where the factor loading (s) declines by

24%-45%, from 0.67 to 0.37-0.51. The significant impact of EQfactor (k ) on s is consistent with Berk’s

[1995] conclusion that size factor loadings reflect misspecification and estimation errors (of the asset

24
pricing model). In particular, if the three-factor model is misspecified due to the exclusion of

EQfactor (k ) , we would expect its inclusion to reduce the magnitude of the loading on SMB.

The results in Table 6 suggest that earnings quality plays a statistically and economically

meaningful role in determining the cost of equity capital. To mitigate concerns that these findings are

specific to the sample of firms used to calculate EQfactor (k ) , we repeat the one-factor and three-factor

tests using the 14,398 publicly traded firms with at least 12 monthly returns during the period April 1989

through March 2001.24 If firms with poor earnings quality have higher costs of capital, their excess

returns should exhibit positive loadings on EQfactor (k ) . The results of these tests, reported in Table 7,

are, if anything, stronger than those reported in Table 6. In particular, the CAPM tests (Panel A) show

significant positive loadings on EQfactor (k ) (t-statistics exceed 50), with EQfactor (k ) contributing a

mean incremental explanatory power of 4.4%, an increase of over 50% from the regression excluding

EQfactor (k ) . The three-factor results (Panel B), show that EQfactor (k ) retains statistical significance in

the presence of the other three factors, and provides average incremental explanatory power of 2.4%, an

increase of about 17% over the model excluding EQfactor (k ) .

Overall, we interpret the results in Tables 6 and 7 as providing strong evidence that earnings

quality affects market perceptions of equity risk. The result that firms with poor quality earnings have

larger costs of capital than firms with high quality earnings is consistent both with intuition and with

predictions from Easley and O’Hara [2002]. We also find that earnings quality is distinct from market,

size, book-to-market, and earnings volatility factors, and adds meaningfully to these factors in explaining

returns.

24
This analysis is facilitated by the factor mimicking portfolio design which maps firm- and year-specific earnings
quality values into month-specific excess returns (the EQfactor (k ) ’s). Because EQfactor (k ) is not firm-specific, it
can be correlated with the excess returns of any firm, irrespective of whether the firm has the necessary data to
calculate the EQ metric itself.

25
5. Changes in Earnings Quality and Changes in Costs of Capital

As a final sensitivity check, we investigate whether the change in a firm’s earnings quality is

positively correlated with the change in their costs of capital. While this test controls for firm-specific

factors that are constant over time, it has lower power (relative to our previous tests) because of the

considerable over-time variation in the EQ metrics (documented in Table 2). For each firm with EQ data

for the entire period (J=1,226 firms), we calculate the mean change in EQ(k) between 1988-1993 and

1994-1999, ∆EQ (k ) j ; we also calculate the mean change between these sub-periods in each of firm j’s

cost of capital proxies: ∆DebtRating j , ∆CostDebt j , ∆IndEPj and ∆Beta j . Table 8 reports the

coefficient estimates and t-statistics from regressing the change in each cost of capital measure on

∆EQ (k ) j . Although the results are statistically weaker than those reported in Tables 4-7, they generally

show significant (at the 0.10 level or better, with some exceptions) positive correlations between changes

in firms’ costs of capital and changes in their earnings quality metrics.

6. Conclusion

We find that investors price securities in a manner that reflects their awareness of earnings

quality: lower quality earnings are associated with lower debt ratings, higher realized costs of debt,

smaller price multiples on earnings, and larger equity betas. Moreover, earnings quality loads as a

separate factor in explaining variation in excess returns in both a one-factor model and a three-factor

model. Our results are consistent across securities (debt and common equity), estimation procedures

(pooled and Fama-MacBeth regressions), variable specification (raw and decile) and model specification

(cross-sectional levels versus over-time changes), and are robust to the inclusion of control variables

known to affect the hypothesized relations, including earnings volatility. We also provide evidence that

the earnings quality metrics we examine are related to notions of earnings/accounting quality used in prior

research (i.e., the ability of annual earnings to explain annual returns, and the response coefficients from a

26
regression of two-day returns on quarterly earnings news). Taken as a whole, we interpret these results as

indicating systematic capital market consequences associated with differences in earnings quality.

We also believe that the weight of the evidence supports the view that the capital market

consequences of differences in earnings quality arise because earnings quality is a nondiversifiable risk

factor as opposed to a variant of the previously-documented accruals anomaly. In particular, our asset

pricing results are robust to controls for size and book-to-market factors, which Fama and French [1996]

show capture value-glamour strategies (such as cash flow-to-price), which itself has been shown to

capture most or all of the accruals anomaly effect (Desai et al. [2002]). Moreover, our results are

consistent with a rational asset pricing model (e.g., Easley and O’Hara [2002]) in which nondiversifiable

information risk is a priced factor.

While our results document economically significant capital market effects of firm-specific

differences in earnings quality, our analysis does not speak to underlying determinants of those

differences. We do not separate earnings-quality-related cost of capital effects into those associated with

fundamental strategic choices (such as business models) and innate firm characteristics (such as volatility

of operating environments) and those associated with accounting judgments and governance-related

choices.25 As other studies have noted, discretionary accounting choices and governance decisions are

likely to be endogenously determined, while other factors (such as volatility of the operating

environment) are likely to be predetermined. Future research might investigate the relative strength of

endogenous managerial decisions versus predetermined factors as determinants of earnings quality.

25
Results of prior research are consistent with board characteristics, ownership structure and auditor quality
influencing earnings quality. Specifically, prior studies show larger absolute abnormal accruals for firms: where the
board of directors is insider controlled and where a smaller percent of independent directors sit on the board or audit
committee (Klein [2002); with smaller insider holdings (Warfield, Wild and Wild [1995]); and audited by non-Big 6
auditors (Becker, Defond, Jiambalvo and Subramanyan [1998]).

27
 Appendix: Debt Rating Scores

Our debt
Compustat rating score,
S&P numerical
Debt Rating code(s)26
DebtRating
AAA 2 1
AA+ 4 2
AA 5 3
AA- 6 4
A+ 7 5
A 8 6
A- 9 7
BBB+ 10 8
BBB 11 9
BBB- 12 10
BB+ 13 11
BB 14 12
BB- 15 13
B+ 16 14
B 17 15
B- 18 16
CCC+ 19 17
CCC or CC 20,23 18
C 21,24 19
D or SD 27,29,90 20

26
Compustat data item #280.

28
 Table 1
Descriptive Information on the Earnings Quality Metrics, 1988-1999

Panel A: Summary statistics on the earnings quality metricsa

Variable mean 10% 25% median 75% 90%

EQ1 0.0597 0.0067 0.0172 0.0394 0.0802 0.1406
EQ2 0.0546 0.0050 0.0134 0.0337 0.0725 0.1345
EQ3 0.0624 0.0067 0.0179 0.0423 0.0841 0.1462
EQ4 0.0515 0.0024 0.0106 0.0309 0.0696 0.1301
EQ5 0.0478 0.0046 0.0123 0.0301 0.0639 0.1164
EQ6 0.0591 0.0187 0.0302 0.0495 0.0793 0.1148
EQ7 0.0431 0.0107 0.0197 0.0342 0.0572 0.0904

Panel B: Pairwise correlations between earnings quality metricsb

EQ1 EQ2 EQ3 EQ4 EQ5 EQ6 EQ7

EQ1 0.9036 0.9025 0.8441 0.6357 0.4040 0.3562
(.0001) (.0001) (.0001) (.0001) (.0001) (.0001)
EQ2 0.7812 0.8309 0.9207 0.6767 0.4239 0.3730
(.0001) (.0001) (.0001) (.0001) (.0001) (.0001)
EQ3 0.7948 0.6844 0.8878 0.5774 0.3987 0.3551
(.0001) (.0001) (.0001) (.0001) (.0001) (.0001)
EQ4 0.6766 0.7997 0.7783 0.6223 0.4054 0.3533
(.0001) (.0001) (.0001) (.0001) (.0001) (.0001)
EQ5 0.5197 0.6038 0.4749 0.5235 0.4790 0.4336
(.0001) (.0001) (.0001) (.0001) (.0001) (.0001)
EQ6 0.4016 0.4506 0.4058 0.4060 0.4751 0.8595
(.0001) (.0001) (.0001) (.0001) (.0001) (.0001)
EQ7 0.3527 0.4011 0.3575 0.3524 0.4310 0.8515
(.0001) (.0001) (.0001) (.0001) (.0001) (.0001)

Panel C: Over-time variation in EQ metricsc

Standard deviation of EQ(k)

Variable mean 10% 25% median 75% 90%
EQ1 0.0453 0.0117 0.0192 0.0335 0.0577 0.0945
EQ2 0.0434 0.0095 0.0170 0.0315 0.0563 0.0926
EQ3 0.0473 0.0130 0.0212 0.0356 0.0599 0.0972
EQ4 0.0438 0.0102 0.0182 0.0321 0.0561 0.0929
EQ5 0.0320 0.0074 0.0130 0.0228 0.0406 0.0695
EQ6 0.0032 0.0004 0.0008 0.0016 0.0036 0.0074
EQ7 0.0032 0.0003 0.0006 0.0014 0.0034 0.0079

29
Panel D: Principal factor analysis of EQ metricsd

Principal Factor Analysis ML Principal Factor Analysis

Correlation with Correlation with
EQ metric common factor p-value common factor p-value
EQ1 0.9278 0.0001 0.9503 0.0001
EQ2 0.9436 0.0001 0.9708 0.0001
EQ3 0.9089 0.0001 0.9290 0.0001
EQ4 0.9269 0.0001 0.9580 0.0001
EQ5 0.7132 0.0001 0.6920 0.0001
EQ6 0.6203 0.0001 0.4603 0.0001
EQ7 0.5726 0.0001 0.4087 0.0001
Cumulative explained variation 0.8015 0.8878

Variable definitions: EQ1 = unsigned abnormal accruals estimated from the modified Jones model; EQ2 = unsigned
abnormal current accruals estimated following Teoh et al. [1998]; EQ3 = performance-matched EQ1; EQ4 =
performance-matched EQ2; EQ5 (EQ6) = absolute (standard deviation of) residuals from cross-sectional estimations
of Dechow-Dichev model; EQ7 = standard deviation of residuals of firm-specific time series estimations of
Dechow-Dichev model. EQ8 = the common factor identified from principal factor analysis of EQ1-EQ7.
a
Panel A reports summary information on the pooled sample distribution of the EQ metrics. The sample contains
33,770 firm-year observations over t=1988-1999 with Compustat data to calculate all EQ metrics.
b
Panel B reports Pearson (Spearman) correlations above (below) the diagonal; significance levels are in
parentheses.
c
For each firm-year, we calculate the standard deviation of the rolling 5-year EQ(k). Panel C reports summary
information on the cross-sectional distribution of these standard deviations.
d
Panel D reports the pairwise correlation between each EQ metric and the common factor identified from principal
factor analysis and separately, from maximum likelihood (ML) principal factor analysis; significance levels for the
correlations are also reported.

30
 Table 2
Summary Financial Information About the Sample Firms, 1988-1999a

Variable mean 10% 25% median 75% 90%

Market value of equity ($mils) 1808.3 9.0 29.5 142.1 785.0 3201.1
Total assets ($mils), Size 2061.6 13.6 43.2 196.9 1041.6 4468.9
Sales revenue ($mils) 1944.9 13.7 49.5 226.9 1050.6 4065.8
Return on assets, ROA 0.023 -0.075 0.007 0.041 0.076 0.117
Market to book ratio 4.64 0.66 1.06 1.67 2.77 4.84

Fiscal year stock return 0.191 -0.318 -0.082 0.145 0.382 0.698
Net income (before extra.), scaled by MV 0.202 -0.114 0.012 0.060 0.095 0.149
Change in net income, scaled by MV 0.036 -0.115 -0.024 0.007 0.038 0.144

CAR(-1,0) 0.002 -0.050 -0.021 0.001 0.024 0.057

Unexpected earnings, UE -0.009 -0.021 -0.005 0.000 0.001 0.005

Earnings-price ratio, EP 0.065 0.023 0.041 0.061 0.083 0.111

Industry-adjusted earnings-price ratios, IndEP 0.006 -0.034 -0.016 0.002 0.021 0.048
Log of average sales growth over prior 5 years, Growth 0.093 -0.024 0.027 0.078 0.143 0.224
Debt rating score (see Appendix), DebtRating 8.65 4.00 6.00 8.00 12.00 14.00
Long term debt to assets ratio, Leverage 0.213 0.000 0.036 0.170 0.310 0.440
Interest expense to long term debt ratio, CostDebt 0.092 0.058 0.071 0.087 0.106 0.129
Coefficient of variation of net income, CV(NI) 0.547 -2.562 -0.677 0.315 0.817 2.295
Standard deviation of earnings per share, Volatility 0.935 0.114 0.226 0.448 0.913 1.813

a
We report summary data on selected financial variables for the sample of 33,770 firm-years, t=1988-1999, with
Compustat data to calculate all earnings quality metrics. All variables are measured as of the end of fiscal year t.

31
 Table 3
Tests of the Association Between Earnings Quality and the Returns-Earnings Relation, 1988-1999

Panel A: Regression of squared residuals on EQ metricsa

Raw EQ Decile rank of EQ

Pooled results Fama-MacBeth results Pooled results Fama-MacBeth results
Indep. var. coef.est. t-stat coef.est. t-stat coef.est. t-stat coef.est. t-stat
EQ1 1.646 16.61 1.581 5.88 0.028 13.71 0.026 7.07
EQ2 1.671 16.60 1.629 5.81 0.028 13.93 0.027 7.04
EQ3 1.616 16.72 1.531 5.53 0.027 13.41 0.026 7.04
EQ4 1.635 15.99 1.551 5.50 0.025 12.42 0.024 6.46
EQ5 1.844 15.01 1.790 7.05 0.027 13.11 0.025 8.29
EQ6 3.899 24.57 3.543 8.47 0.047 23.18 0.043 8.87
EQ7 4.403 22.74 3.962 7.52 0.043 21.27 0.039 8.48
EQ8 0.131 21.22 0.124 6.19 0.039 19.47 0.037 7.70

Panel B: Tests of the association between ERCs and EQ metricsb

Raw EQ Decile rank of EQ

Pooled results Fama-MacBeth results Pooled results Fama-MacBeth results
Indep. var. coef.est. t-stat coef.est. t-stat coef.est. t-stat coef.est. t-stat
UE 0.0679 11.35 0.0922 7.80 0.0716 7.65 0.1645 5.02
UE*EQ1 -0.1172 -1.63 -0.2118 -0.72 -0.0054 -2.61 -0.0054 -0.93
UE*EQ2 -0.1611 -2.21 -0.1456 -0.45 -0.0088 -4.25 -0.0064 -1.23
UE*EQ3 -0.0051 -0.07 -0.2576 -0.94 -0.0033 -1.66 -0.0056 -1.07
UE*EQ4 -0.0987 -1.28 -0.3215 -1.14 -0.0052 -2.61 -0.0077 -1.45
UE*EQ5 -0.3036 -4.04 -0.1554 -0.47 -0.0074 -3.48 -0.0072 -1.19
UE*EQ6 -0.7164 -4.98 -0.5241 -1.32 -0.0074 -3.39 -0.0035 -0.87
UE*EQ7 -0.6054 -3.25 -0.7010 -1.58 -0.0064 -2.64 -0.0073 -1.93
UE*EQ8 -0.0111 -2.36 -0.0190 -1.15 -0.0103 -4.26 -0.0105 -2.19

Variable definitions: See Table 1 for definitions of the EQ metrics.

a
The sample used in the squared residual tests contains 26,398 firm-year observations over t=1988-1999. We report
the results of estimating the relation between firm j’s squared residual, εˆ 2j ,t (calculated from the year t regression of
fiscal year returns on the level and change in annual earnings in year t) and the value of firm j’s EQ(k) metric in year
t. The decile-rank tests replace the (raw) value of the EQ metric with its decile rank. We report pooled results as
well as results from Fama-MacBeth tests of the mean of the 12 annual coefficients, with the t-statistic based on the
time-series standard deviation of those coefficients.
b
The sample used in the quarterly earnings announcement tests contains 32,149 quarterly announcement dates over
t=1988-1999 for which EQ metrics, daily returns data and unexpected earnings data are available. We report the
results of estimating the regression of the 2-day abnormal return to the earnings announcement on unexpected
earnings (UE) and unexpected earnings interacted with EQ. The decile-rank tests replace the (raw) value of the EQ
metric with its decile rank. We report pooled results as well as results from Fama-MacBeth tests of the mean of the
48 quarterly coefficients, with the t-statistic based on the time-series standard deviation of those coefficients.

32
 Table 4
Tests of the Association Between Earnings Quality and Debt Ratings, 1988-1999
Panel A: Mean debt rating, by EQ quintilesa
EQ Quintile (1=High EQ score; 5=Low EQ score) Q5-Q1
EQ metric Q1 Q2 Q3 Q4 Q5 Diff. t-stat.
EQ1 7.99 8.06 8.19 8.82 10.17 2.18 16.63
EQ2 7.68 7.96 8.37 8.96 10.26 2.59 20.31
EQ3 7.66 7.86 8.55 8.93 10.23 2.58 20.11
EQ4 8.03 7.89 8.23 9.03 10.05 2.03 15.54
EQ5 7.55 7.85 8.40 9.05 10.39 2.84 22.04
EQ6 5.97 7.40 8.25 9.78 11.82 5.85 56.79
EQ7 6.02 7.18 8.20 9.91 11.91 5.90 57.17
EQ8 6.33 7.80 8.71 9.62 10.76 4.43 38.27

Panel B: Regression of debt rating on EQ, controlling for other factorsb

Raw EQ Decile rank of EQ
Pooled results Fama-MacBeth results Pooled results Fama-MacBeth results
Indep. var. coef.est. t-stat coef.est. t-stat coef.est. t-stat coef.est. t-stat
Leverage 5.07 29.51 5.26 12.86 5.07 29.51 5.26 12.86
Size -1.15 -48.39 -1.20 -45.08 -1.15 -48.39 -1.20 -45.08
ROA -11.93 -29.03 -13.83 -7.97 -11.93 -29.03 -13.83 -7.97
IntCov 0.00 -1.02 -0.01 -2.07 0.00 -1.02 -0.01 -2.07
Volatility 0.06 7.30 0.19 3.37 0.06 7.30 0.19 3.37
EQ1 9.90 14.00 8.95 12.10 0.14 12.69 0.12 10.36
EQ2 10.86 14.76 10.13 10.94 0.16 14.97 0.15 11.60
EQ3 9.74 14.48 9.14 11.13 0.17 15.27 0.16 9.65
EQ4 9.85 13.44 8.98 13.53 0.13 12.30 0.12 10.76
EQ5 12.77 13.68 11.17 12.89 0.17 15.59 0.16 12.12
EQ6 47.03 36.20 47.66 13.31 0.44 40.39 0.42 19.51
EQ7 58.73 37.39 57.62 12.27 0.47 44.50 0.45 16.24
EQ8 1.02 21.90 0.97 13.27 0.31 28.14 0.29 16.07
Panel C: Regression of realized cost of debt on EQ, controlling for other factorsc
Raw EQ Decile rank of EQ
Pooled results Fama-MacBeth results Pooled results Fama-MacBeth results
Indep. var. coef.est. t-stat coef.est. t-stat coef.est. t-stat coef.est. t-stat
Leverage -0.09 -1.81 -0.47 -3.19 -0.09 -1.81 -0.47 -3.19
Size -0.31 -29.24 -0.24 -5.85 -0.31 -29.24 -0.24 -5.85
ROA 0.00 -0.28 -2.08 -7.08 0.00 -0.28 -2.08 -7.08
IntCov 0.00 0.34 0.00 -1.77 0.00 0.34 0.00 -1.77
Volatility 0.11 13.58 0.15 4.93 0.11 13.58 0.15 4.93
EQ1 3.93 9.95 3.48 10.89 0.07 8.81 0.07 14.04
EQ2 4.36 10.76 3.82 8.23 0.08 10.35 0.08 11.08
EQ3 3.69 9.60 3.37 9.18 0.07 8.34 0.06 9.90
EQ4 4.17 10.20 3.77 7.18 0.08 9.50 0.07 7.93
EQ5 5.62 11.48 4.38 7.20 0.08 10.28 0.07 6.36
EQ6 11.14 15.08 13.41 16.81 0.13 12.88 0.17 17.69
EQ7 14.36 16.67 15.53 19.82 0.16 17.15 0.18 20.41
EQ8 0.35 13.57 0.34 12.70 0.13 14.26 0.13 17.60

33
Variable definitions: DebtRating = numerical scoring of S&P’s Credit Rating (see Appendix for the transformation
of the debt ratings into scores); CostDebt = ratio of interest expense to average interest bearing debt. Leverage =
ratio of interest bearing debt to total assets; Size = log of total assets; ROA = return on assets; IntCov = interest
coverage ratio; Volatility = standard deviation of earnings per share calculated over the past 5 years. See Table 1 for
definitions of the EQ metrics.
a
The sample used in the debt rating tests (Panels A and B) contains 7,982 firm-year observations over t=1988-1999
with Compustat data on S&P debt ratings. The sample used in the cost of debt tests (Panel C) contains 22,940
observations. Panel A shows the mean debt rating score for each EQ quintile; the scores range from 1 (highest
credit rating, AAA) to 20 (lowest credit rating, default). The columns labeled “Q5-Q1” show the difference in the
median values between the worst (Q5) and best (Q1) earnings quality quintiles, along with t-statistics of whether the
difference is zero.
b
We report the results of estimating the relation between firm j’s debt rating and the value of its EQ metric,
controlling for other factors known to affect debt ratings. The first five rows show the results excluding EQ(k) from
the regression. The remaining rows show the coefficient estimate and t-statistics on EQ(k), controlling for these
factors. The decile-rank tests replace the (raw) value of the EQ metric with its decile rank. We report pooled results
as well as results from Fama-MacBeth tests of the mean of the 12 annual coefficients, with the t-statistic based on
the time-series standard deviation of those coefficients.
c
Panel C is similar to Panel B except the dependent variable is the realized cost of debt (CostDebt) rather than debt
rating. Also, see note a for differences in samples used in Panels B and C.

34
 Table 5
Tests of the Association Between Earnings Quality and Industry-Adjusted Earnings-Price Ratios, 1988-1999

Panel A: Industry-adjusted earnings-price ratios, partitioned by EQ quintiles

EQ Quintile (1=High EQ score; 5 =Low EQ score) Q5-Q1

EQ metric Q1 Q2 Q3 Q4 Q5 Diff. t-stat.
EQ1 0.0051 0.0047 0.0054 0.0058 0.0096 0.0045 5.50
EQ2 0.0048 0.0045 0.0046 0.0072 0.0095 0.0047 5.71
EQ3 0.0049 0.0046 0.0054 0.0068 0.0090 0.0041 5.03
EQ4 0.0045 0.0046 0.0051 0.0071 0.0093 0.0047 5.70
EQ5 0.0049 0.0042 0.0047 0.0069 0.0099 0.0049 6.08
EQ6 0.0030 0.0027 0.0060 0.0074 0.0115 0.0085 10.56
EQ7 0.0034 0.0027 0.0057 0.0066 0.0122 0.0088 10.88
EQ8 0.0035 0.0039 0.0053 0.0072 0.0107 0.0072 9.04

Panel B: Regression of industry-adjusted EP ratios on Growth and EQ metrics

Raw EQ Decile rank of EQ

Pooled results Fama-MacBeth results Pooled results Fama-MacBeth results
Indep. Var. coef.est. t-stat coef.est. t-stat coef.est. t-stat coef.est. t-stat
Growth -0.0066 -3.56 -0.0059 -1.70 -0.0066 -3.56 -0.0059 -1.70
EQ1 0.0348 7.47 0.0338 5.75 0.0005 6.12 0.0005 5.22
EQ2 0.0385 8.23 0.0379 5.63 0.0006 7.25 0.0006 6.31
EQ3 0.0313 6.94 0.0310 6.51 0.0006 6.44 0.0006 6.43
EQ4 0.0395 8.32 0.0385 6.10 0.0006 7.38 0.0007 6.33
EQ5 0.0573 9.29 0.0603 5.85 0.0006 7.31 0.0007 6.04
EQ6 0.1037 13.45 0.1018 7.07 0.0011 12.91 0.0011 7.02
EQ7 0.1265 13.42 0.1208 6.35 0.0011 11.97 0.0010 6.25
EQ8 0.0031 10.60 0.0030 6.89 0.0009 10.22 0.0009 6.78

Variable definitions: Industry-adjusted EP ratio, IndEP = firm j’s earnings-price ratio less the median earnings-price
ratio of its industry. Growth = log of the average past five years sales changes. See Table 1 for definitions of the EQ
metrics.
a
The sample used in the earnings-price tests contains 19,633 firm-year observations over t=1988-1999. Panel A
shows the mean industry-adjusted earnings-price for each EQ quintile. The columns labeled “Q5-Q1” show the
difference in the mean values between the worst (Q5) and best (Q1) earnings quality quintiles, along with t-statistics
of whether the difference is distinguishable from zero.
b
We report the results of estimating the relation between firm j’s industry-adjusted earnings price ratio, IndEP , and
Growth and EQ(k). The row labeled Growth shows the results excluding EQ(k) from the regression. The remaining
rows show the coefficient estimate and t-statistics on EQ(k) controlling for Growth. The decile-rank tests replace
the (raw) value of EQ(k) with its decile rank. We report pooled results as well as results from Fama-MacBeth tests
of the mean of the 12 annual coefficients, with the t-statistic based on the time-series standard deviation of those
coefficients.

35
 Table 6
Tests of the Association Between Earnings Quality and Cost of Capital
For the Earnings Quality Sample, 1989-2001

Panel A: Beta coefficients from CAPM regressions, partitioned by EQ quintilesa

EQ Quintile (1=High EQ score; 5 =Low EQ score) Q5-Q1 EQ effect on implied

EQ metric Q1 Q2 Q3 Q4 Q5 Diff. t-stat cost of capital
EQ1 0.76 0.74 0.83 0.91 1.01 0.25 4.16 1.50%
EQ2 0.74 0.76 0.83 0.91 1.03 0.30 4.84 1.78%
EQ3 0.74 0.79 0.84 0.89 1.00 0.27 4.45 1.60%
EQ4 0.73 0.78 0.82 0.90 1.03 0.29 5.07 1.76%
EQ5 0.67 0.78 0.86 0.96 0.99 0.33 4.84 1.96%
EQ6 0.60 0.75 0.84 0.96 1.11 0.51 4.27 3.05%
EQ7 0.63 0.78 0.81 0.93 1.11 0.49 4.45 2.92%
EQ8 0.61 0.79 0.86 0.95 1.06 0.45 4.68 2.69%

Panel B: Firm-specific cost-of-capital regressions, CAPM and EQ factorb

Rm-Rf factor EQfactor(k)

EQ metric coeff. t-stat coeff. t-stat Adj.R2 Inc. R2
none 0.81 83.94 - - 0.082 -
EQ1 0.63 66.74 0.57 32.73 0.116 0.034
EQ2 0.65 69.69 0.50 29.48 0.115 0.033
EQ3 0.64 68.77 0.60 35.01 0.117 0.035
EQ4 0.66 71.03 0.55 32.79 0.116 0.035
EQ5 0.64 67.64 0.47 33.48 0.119 0.038
EQ6 0.69 74.15 0.32 33.54 0.121 0.039
EQ7 0.68 74.36 0.36 34.67 0.122 0.040
EQ8 0.67 71.73 0.40 32.00 0.119 0.037

Panel C: Firm-specific cost-of-capital regressions, CAPM, EQ and Volatility factorsc

Rm-Rf factor EQfactor(k) Volfactor

EQ metric coeff. t-stat coeff. t-stat coeff. t-stat Adj.R2 Inc. R2
EQ1 0.62 62.28 0.59 32.21 0.03 1.35 0.124 0.042
EQ2 0.63 64.28 0.52 29.78 0.04 1.90 0.123 0.041
EQ3 0.63 65.06 0.62 35.89 0.02 0.87 0.125 0.043
EQ4 0.65 66.50 0.56 32.58 0.01 0.30 0.124 0.042
EQ5 0.62 62.58 0.50 35.20 0.05 2.92 0.127 0.045
EQ6 0.65 68.45 0.34 36.34 0.11 6.20 0.128 0.046
EQ7 0.65 68.61 0.39 36.81 0.09 4.93 0.130 0.048
EQ8 0.64 66.04 0.42 32.54 0.08 4.22 0.126 0.044

36
Panel D: Firm-specific cost-of-capital regressions, three-factor model and EQ factord

Rm-Rf factor SMB factor HML factor EQfactor(k)

EQ metric coeff. t-stat coeff. t-stat coeff. t-stat coeff. t-stat Adj.R2 Inc. R2
none 0.81 83.89 0.67 44.25 0.22 13.55 - - 0.129 -
EQ1 0.76 75.72 0.46 26.39 0.29 16.82 0.32 15.23 0.143 0.015
EQ2 0.77 78.02 0.51 29.52 0.28 16.60 0.24 11.86 0.143 0.014
EQ3 0.76 77.35 0.45 26.02 0.28 17.67 0.36 17.67 0.145 0.016
EQ4 0.78 78.62 0.47 26.56 0.28 16.37 0.31 15.08 0.144 0.016
EQ5 0.76 75.89 0.43 21.64 0.27 15.59 0.28 14.64 0.146 0.018
EQ6 0.79 81.02 0.37 18.61 0.29 16.26 0.22 15.94 0.147 0.019
EQ7 0.78 80.21 0.34 15.83 0.25 14.65 0.27 16.11 0.148 0.020
EQ8 0.78 79.70 0.43 23.15 0.29 16.64 0.25 14.74 0.147 0.019

Variable definitions: See Table 1 for definitions of the EQ metrics. EQfactor (k ) = the returns to the EQ(k) factor-
mimicking portfolio. Volfactor = the returns to the Volatility factor-mimicking portfolio.

a
The sample consists of 4,124 firms with data on the EQ metrics and at least 12 monthly returns between April 1989
and March 2001. Panel A shows the beta coefficient from a one-factor model for each quintile of firms ranked on
the noted EQ metric. The columns labeled “Q5-Q1” show the difference in the median values between the worst
(Q5) and best (Q1) earnings quality quintiles, along with t-statistics of whether the difference is zero. The far right
column shows the implied effect on the cost of capital, calculated as the Q5-Q1 difference in equity betas times an
assumed market risk premium of 6%.
b
Panel B reports the average coefficient estimates, across the J=4,124 estimations, of the one-factor model (row
labeled “none”), and the one-factor model that includes EQfactor (k ) . The far right column shows the incremental
R2 associated with EQfactor (k ) .

c
Panel C reports the average coefficient estimates, across the J=4,124 estimations, of the one-factor model
augmented by Volfactor and EQfactor (k ) . The far right column shows the incremental R2 associated with
EQfactor (k ) .

d
Panel D reports the average coefficient estimates, across the J=4,124 estimations, of the three-factor model (row
labeled “none”), and the three-factor model that includes EQfactor (k ) . The far right column shows the incremental
R2 associated with EQfactor (k ) .

37
 Table 7
Tests of the Association Between Earnings Quality and Cost of Capital
For All Traded Firms, 1989-2001a

Panel A: Firm-specific cost-of-capital regressions, CAPM and EQ factorb

RM-RF factor EQ factor (k)

EQ metric coeff. t-stat coeff. t-stat Adj.R2 Inc. R2
none 0.88 114.03 - - 0.083 -
EQ1 0.61 82.34 0.67 56.52 0.125 0.042
EQ2 0.63 85.04 0.61 52.16 0.125 0.042
EQ3 0.63 59.17 0.71 59.17 0.126 0.044
EQ4 0.63 85.97 0.67 58.54 0.126 0.043
EQ5 0.63 84.81 0.57 57.42 0.128 0.045
EQ6 0.67 91.59 0.37 56.05 0.129 0.046
EQ7 0.67 91.02 0.44 58.77 0.130 0.047
EQ8 0.64 87.99 0.46 53.19 0.128 0.045

Panel B: Firm-specific cost-of-capital regressions, three-factor model and EQ factorc

RM-RF factor SMB factor HML factor EQfactor(k)

EQ metric coeff. t-stat coeff. t-stat coeff. t-stat coeff. t-stat Adj.R2 Inc. R2
none 0.81 100.87 0.72 67.02 0.15 11.75 - - 0.140 -
EQ1 0.73 85.71 0.46 34.24 0.28 20.10 0.43 27.45 0.162 0.022
EQ2 0.74 87.36 0.48 38.61 0.28 20.55 0.38 26.17 0.162 0.022
EQ3 0.74 88.84 0.43 34.13 0.27 19.45 0.50 32.85 0.164 0.023
EQ4 0.75 88.58 0.45 35.13 0.30 20.92 0.46 31.18 0.163 0.023
EQ5 0.73 86.24 0.37 27.80 0.27 19.61 0.43 32.06 0.165 0.025
EQ6 0.77 89.98 0.32 21.56 0.28 19.71 0.30 29.01 0.166 0.026
EQ7 0.76 89.88 0.29 19.09 0.25 18.39 0.36 32.02 0.166 0.026
EQ8 0.75 88.56 0.38 28.21 0.30 20.33 0.35 28.93 0.165 0.025

a
The sample consists of J=14,398 publicly traded firms with monthly returns data on CRSP between April 1989 and
March 2001.
b
Panel B reports the average coefficient estimates, across the J=14,398 estimations, of the one-factor model (row
labeled “none”), and the one-factor model that includes EQfactor (k ) . The far right column shows the incremental
R2 associated with EQfactor (k ) .

c
Panel C reports the average coefficient estimates, across the J=14,398 estimations, of the three-factor model (row
labeled “none”), and the three-factor model that includes EQfactor (k ) . The far right column shows the incremental
R2 associated with EQfactor (k ) .

38
 Table 8
Changes in Costs of Capital and Changes in Earnings Quality, 1988-1999a

Change in cost of capital estimate

∆DebtRating (n=458) ∆DebtCost (n=1070) ∆IndEP (n=1005) ∆Beta (n=992)
Indep. var. coef.est. t-stat coef.est. t-stat coef.est. t-stat coef.est. t-stat
∆EQ1 0.0450 1.06 0.1282 2.16 0.0011 1.99 0.0306 2.64
∆EQ2 0.1330 1.97 0.1770 2.97 0.0006 1.13 0.0235 2.01
∆EQ3 0.1096 1.63 0.1579 2.67 0.0011 1.98 0.0319 2.80
∆EQ4 0.1244 1.85 0.1064 1.81 0.0006 1.11 0.0247 2.15
∆EQ5 0.1230 1.83 0.1042 1.79 0.0004 0.78 0.0271 2.39
∆EQ6 0.2426 2.54 0.2520 1.83 0.0018 1.38 -0.0047 -0.17
∆EQ7 0.4420 4.83 0.4073 3.22 0.0023 1.94 0.0211 0.84
∆EQ8 0.0951 1.92 0.2074 2.86 0.0014 2.02 0.0334 2.36

Variable definitions: ∆EQ(k ) j = the change in the mean value of EQ(k) between 1988-1993 and 1994-1999;
∆DebtRating j , ∆CostDebt j , ∆IndEPj and ∆Beta j are the changes in firm j’s mean debt rating, realized cost of
debt, industry-adjusted earnings-price ratio and equity beta between the same two sub-periods.
a
The sample consists of 1,226 firms with data on all EQ metrics for the entire period; for each regression, the
sample size is reduced to firms with data on the noted cost of capital measure. We report the coefficient estimates
and t-statistics from regressing the change in each cost of capital measure ( ∆DebtRating j , ∆CostDebt j , ∆IndEPj
and ∆Beta j ) on ∆EQ(k ) j .

39
 References

Alford, A. 1992. The effect of the set of comparable firms on the accuracy of the price-earnings valuation method.
Journal of Accounting Research 30, 1: 94-108.

Barone, Gerhard. Perceptions of earnings quality and their association with the cost of equity capital. Working
paper, University of Wisconsin, 2002.

Barth, M., W. Beaver and W. Landsman. 2001. The relevance of the value relevance literature for financial
accounting standard setting: another view. Journal of Accounting and Economics 31 (1-3): 77-104.

Bartov, E., F. Gul and J. Tsui. 2000. Discretionary-accruals models and audit qualifications. Journal of Accounting
and Economics 30: 421-452.

Beaver, W., P. Kettler and M. Scholes. 1970. The association between market determined and accounting
determined risk measures. The Accounting Review, 654-682.

Becker, C., M. DeFond, J. Jiambalvo and K.R. Subramanyan. 1998. The effect of audit quality on earnings
management. Contemporary Accounting Research 15: 1-24.

Beneish, M.D. 1999. Incentives and penalties related to earnings overstatements that violate GAAP. The
Accounting Review 4: 425-457.

Berk, J. 1995. A critique of size-related anomalies. Review of Financial Studies 8, 2: 275-286.

Blom, G. 1984. Probability theory and applications. 2cd edition. Lund (Sweden).

Botosan, C. 1997. Disclosure level and the cost of equity capital. The Accounting Review 72: 323-349.

Botosan, C. and M. Plumlee. 2002. A re-examination of disclosure level and the expected cost of equity capital.
Journal of Accounting Research 40, 1: 21-40.

Burgstahler, D and I. Dichev. 1997. Earnings management to avoid earnings decreases and losses. Journal of
Accounting and Economics 24 (1): 99-126.

Chan, K., L. Chan, N. Jegadeesh and J. Lakonishok. Earnings quality and stock returns: the evidence from accruals.
National Taiwan University and University of Illinois working paper, 2001.

Collins, D. and S.P. Kothari. 1989. An analysis of inter-temporal and cross-sectional determinants of earnings
response coefficients. Journal of Accounting and Economics 11: 143-181.

Collins, D., E. Maydew and I. Weiss. 1997. Changes in the value-relevance of earnings and book values over the
past forty years. Journal of Accounting and Economics 24: 39-67.

Dechow, P. and I. Dichev. 2002. The quality of accruals and earnings: The role of accrual estimation errors.
Forthcoming, The Accounting Review.

Dechow, P., R. Sloan and A. Sweeney. 1995. Detecting earnings management. The Accounting Review 70 (2):
193-225.

Dechow, P., R. Sloan and A. Sweeney. 1996. Causes and consequences of earnings manipulation: an analysis of
firms subject to enforcement actions by the SEC. Contemporary Accounting Research, 13 (1): 1-36.

DeFond, M. and C. Park. 2001. The reversal of abnormal accruals and the market valuation of earnings surprises.
The Accounting Review 76, 3: 375-404.

40
Desai, H., S. Rajgopal and M. Venkatachalam. Value-glamour and accruals mispricing: one anomaly or two?
Southern Methodist University and Duke University working paper, 2002.

Easley, D., S. Hvidkjaer and M. O’Hara. 2002. Is information risk a determinant of asset returns? Journal of
Finance 57:5 (October): 2185-2221.

Easley, D. and M. O’Hara. 2001. Information and the cost of capital. Cornell University working paper.

Easton, P. and M. Zmijewski. 1989. Cross-sectional variation in the stock market response to accounting earnings
announcements. Journal of Accounting and Economics 11: 117-141.

Fama, E. and K. French. 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial
Economics 33: 3-56.

Fama, E and K. French. 1996. Multifactor explanations of asset pricing anomalies. Journal of Finance 51: 55-84.

Fama, E., and K. French. 1997. Industry costs of equity. Journal of Financial Economics 43: 153-193.

Fama, E. and J. MacBeth. 1973. Risk, return and equilibrium: Empirical tests. Journal of Political Economy 81:
607-636.

Feroz, E., K. Park and V. Pastena. 1991. The financial and market effects of the SEC’s accounting and auditing
enforcement actions. Journal of Accounting Research (Supplement): 107-148.

Francis, J., D. Philbrick and K. Schipper. 1994. Shareholder litigation and corporate disclosures. Journal of
Accounting Research 32, 2: 137-164.

Francis, J. and K. Schipper. 1999. Have financial statements lost their relevance? Journal of Accounting Research
37: 319-352.

Gu, Z. Cross-sectional incomparability of R2s and additional evidence on value relevance changes over time.
Carnegie Mellon University working paper, 2002.

Hribar, P. and D. Collins. 2002. Errors in estimating accruals: i134.mplications for empirical research. Journal of
Accounting Research 40, 1: 105-134.

Jones, J. 1991. Earnings management during import relief investigations. Journal of Accounting
Research 29, 193-228.

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

Kasznik, R. 1999. On the association between voluntary disclosure and earnings management. Journal of
Accounting Research 37 (Autumn): 57-81.

Kellogg, R. 1984. Accounting activities, securities prices, and class action lawsuits. Journal of Accounting and
Economics 6: 185-204.

Klein, A. 2002. Audit committee, board of director characteristics, and earnings management. Journal of Accounting
and Economics 33: 375-400.

Kothari, S.P. 2001. Capital markets research in accounting. Journal of Accounting and Economics 31 (1-3): 105-
231.

Kothari, S.P., A. Leone and C. Wasley. Performance matched discretionary accrual measures. Massachusetts
Institute of Technology working paper, 2001.

41
Lev, B. 1989. On the usefulness of earnings and earnings research: Lessons and directions from two decades of
empirical research. Journal of Accounting Research 27 (Supplement): 153-192.

Lev, B. and R. Thiagarajan. 1993. Fundamental information analysis. Journal of Accounting Research 31(2): 190-
215.

Liu, J., D. Nissim and J. Thomas. 2002. Equity valuation using multiples. Journal of Accounting Research 40(1):
135-172.

McNichols, M. 2000. Research design issues in earnings management studies. Journal of Accounting and Public
Policy. 19: 313-345.

Palepu, K., P. Healy, and V. Bernard. 2000. Business Analysis and Valuation. 2nd edition. Southwestern Publishing

Palmrose, Z., V. Richardson and S. Scholz. 2001. Determinants of the market reaction to restatement
announcements. University of Southern California and University of Kansas working paper.

Penman, S. 2001. Financial statement analysis & security valuation. 1st edition. McGraw-Hill Irwin, New York.
NY.

Penman, S. and X. Zhang. 2002. Accounting conservatism, the quality of earnings, and stock returns. The
Accounting Review 77, 2: 237-264.

Richardson, S., R. Sloan, M. Soliman and I. Tuna. 2002. Information in accruals about earnings persistence and
future stock returns. University of Michigan working paper.

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

Sloan, R. 1996. Do stock prices fully reflect information in accruals and cash flows about future earnings? The
Accounting Review 71, 289-315.

Teoh, S.H., I. Welch, and T.J. Wong. 1998. Earnings management and the long run underperformance of seasoned
public offerings. Journal of Financial Economics 50: 63-99.

Warfield, T., J. Wild and K. Wild. 1995. Managerial ownership, accounting choices, and informativeness of
earnings. Journal of Accounting and Economics 20: 61-91.

Xie, H. 2001. The mispricing of abnormal accruals. The Accounting Review 76, 3: 357-373.

42