costs of equity and earnings attributes€¦ · costs of equity and earnings attributes 969 equity...

THE ACCOUNTING REVIEWVol. 79. No. 42004pp. 967-1010

Costs of Equity and Earnings AttributesJennifer FrancisDuke University

Ryan LaFondUniversity of Wisconsin

Per M. OlssonDuke University

Katherine SchipperFinancial Accounting Standards Board

ABSTRACT: We examine the relation between the cost of equity capital and sevenattributes of earnings: accrual quality, persistence, predictability, smoothness, valuerelevance, timeliness, and conservatism. We characterize the first four attributes asaccounting-based because they are typically measured using accounting informationonly. We characterize the last three attributes as market-based because proxies forthese constructs are typically based on relations between market data and accountingdata. Based on theoretical models predicting a positive association between informa-tion quality and cost of equity, we test for and find that firms with the least favorablevalues of each attribute, considered individually, generally experience larger costs ofequity than firms with the most favorable values. The largest cost of equity effects areobserved for the accounting-based attributes, in particular, accrual quality. These find-ings are robust to controls for innate determinants of the earnings attributes (firm size,cash flow and sales volatility, incidence of loss, operating cycle, intangibles use/inten-sity, and capital intensity), as vi/ell as to alternative proxies for the cost of equity capital.

WI. INTRODUCTION

e investigate the association between attributes of accounting earnings and in-vestors* resource allocation decisions, using the cost of equity capital as a .sum-mary indicator of those decisions. Specifically, we analyze the extent to which

We are grateful to Alon Brav for access to estimates of ex anle costs of equily capital. We appreciate commentsfrom Terry Shevlin. two anonymous reviewers. Moshe Bareket, Christine Botosan, Alon Brav. DJ Nanda. TerryWarlield, and workshop participants at Cornell University, Duke University, New York University. University ofWisconsin, Ihe Securities and Exchange Commission, the 2(M)3 London Business School Accounting Sympo.sium.the 2003 Southeast Accounting Research Conference, the 2004 PAC 10 Doctoral Con.sortium, the 2004 PARSConference, and ihc 2004 European Accounting Association Meetings. This research was supported by the EuquaSchool of Business, Duke University, and the University of Wisconsin. The views expressed in this paper arc thoseof the authors and do nol represent positions of the Einancial Accounting Standards Board. Positions of the FinancialAccounting Standards Board are arrived at only afler extensive due process and deliberation.

Editor's note: This paper was accepted by Terry Shevlin, Senior Editor.Sithmitled June 2003Accepted April 2004

967

968 Francis, LaFond. Olsson. and Schipper

favorable values of seven earnings characteristics are associated with benefits in the formof a lower cost of equity capital. The properties we consider are attributes used in previousaccounting research and in descriptions of accounting practice to characterize desirablefeatures of earnings: accrual quality, persistence, predictability, smoothness, value relevance,timeliness, and conservatism. Our intent is to determine whether each attribute matters toinvestors, and to calibrate which attributes matter most, as evidenced by the strength of theassociation between the attribute and estimates of the cost of equity.'

Our research is predicated on a relation between the cost of equity and properties offirm-specific information and on the presumption that earnings is a premier source of suchinformation. Analytical research {described in Section II) by Easley and O'Hara (2004),O'Hara (2003), and Leuz and Verrecchia (2004) demonstrates that firm-specific informationrisk is priced (that is, it affects the cost of equity), and this risk cannot be diversified away.The presumption that earnings is a premier source of firm-specific information is supportedby empirical research (Biddlc et al. 1995; Francis et al. 2003; Liu et al. 2002) which showsthat investors rely on earnings more than any other summary measure of performance (i.e.,dividends, cash flows, or variants of earnings, such as EBITDA). Survey results also indicatethat managers view earnings as the key metric evaluated by investors and analysts (Grahamet al. 2003).

Our investigations are motivated by descriptions, in accounting research and elsewhere,of desirable attributes of financial reporting systems in general and earnings in particular;we provide examples of such descriptions in Section II. We assume that earnings attributesare desirable to the extent they reduce information risk and thereby result in a discerniblecapital market advantage. We calibrate this advantage using as our benchmark the equitycost of capital. Taking earnings as a premier source of firm-specific information, and underthe view that properties of firm-specific information risk affect the cost of equity, we in-vestigate the association between earnings attributes and costs of equity along three di-mensions. First, we test the prediction that more unfavorable values of the earnings attri-butes are associated with higher costs of equity capital. Second, we compare the relativemagnitudes of the cost of equity effects across attributes to determine which earnings at-tributes matter most to investors. Third, we test for conditional effects of earnings attributes,identify whether earnings attributes have distinct cost of equity effects (as opposed to asubset of the attributes subsuming the others), and assess whether earnings variability, aninstrument for accrual quality and smoothness, subsumes these attributes. The second andthird tests are motivated by the fact that many descriptions of earnings attributes focus ona single attribute in isolation, without regard to whether that attribute would be desirableor undesirable in the presence of other earnings attributes.

Because we recognize that the earnings attributes we consider are jointly determinedby management's (discretionary) reporting and implementation decisions and by intrinsic(innate) features of firms' business models and operating environments, we examine sepa-rately the cost of equity effects of the total (innate plus discretionary) amount of eachattribute as well as the effects of the discretionary component. Controlling for innate de-terminants of earnings attributes in cost of capital regressions, we interpret the coefficientestimates on the earnings attributes as capturing the pricing of the discretionary (incremen-tal) component of each attribute. The innate factors we consider are size, cash flow varia-bility, sales variability, length of operating cycle, incidence of losses, intangibles intensity,and capital intensity. We emphasize that our interpretation of this coefficient as the cost of

' We focus on the use of finanfial information by equily investors and ignore its contracting (e.g.. with creditors)and stewardship uses.

The Accounting Review. October 2004

Costs of Equity and Earnings Attributes 969

equity effect of the discretionary portion of that attribute relies on both the .sufficiency ofthe set of innate factors and the prediction that discretionary components of earnings attri-butes are priced in equihbrium.

We characterize the seven earnings attributes as either "accounting-based" or "market-based" to capture differences in underlying assumptions about the function of earnings,which are, in turn, reflected in the way the attributes are measured. We refer to accrualquality, persistence, predictability, and smoothness as "accounting-based." These attributestake cash or earnings itself as the reference construct and consequently are measured usingaccounting information only. We refer to value relevance, timeliness, and conservatism as"market-based." These attributes take returns or prices as the reference construct; conse-quently, measures of these attributes are based on the estimated relation between accountingearnings and market prices or returns. We believe these differences in reference constructsare related to implicit assumptions about the intended function of accounting. Specifically,accounting-based earnings attributes derive from an implicit assumption that the functionof earnings is the effective allocation of cash flows to reporting periods via the accrualsprocess, while market-based attributes derive from an implicit assumption that the functionof earnings is to reflect economic income as represented by stock returns.

Our measures of earnings attributes follow prior research. In terms of the accounting-based attributes, prior studies measure accrual quality using either the mapping of currentaccruals into cash flows or some measure of abnormal accruals (e.g., Dechow and Dichev2002; Francis et al. 2005); measures of earnings persistence typically rely on the estimatedslope coefficient in a regression of current earnings on lagged earnings (e.g.. Lev 1983);measures of earnings predictability focus on the prediction errors from a time-series earn-ings model (e.g., Lipe 1990); and smoothness measures are based on the volatility ofearnings relative to some benchmark, such as cash flows (e.g., Leuz et al. 2003; Thomasand Zhang 2002). In terms of the market-based attributes, many prior studies measure valuerelevance as the R~ from a regression of annual returns on annual earnings (e.g.. Collins etal. 1997; Francis and Schipper 1999). Reverse regressions of earnings on variables capturingpositive stock returns and negative stock returns provide measures of timeliness (the reverseregression R', e.g., Bushman et al. 2004; Ball et al. 2000) and conservatism (the ratio ofthe reverse regression coefficient on negative returns to the coefficient on positive returns,e.g., Basu (1997); Pope and Walker (1999).

From a research design perspective, our study is related to Gebhardt et al. (2001) whoexamine whether implied cost of equity estimates derived from a residual income modelare associated with beta, leverage, size, book-to-market, price momentum, growth, andearnings variability. They calibrate their implied cost of equity estimates by showing as-sociations with variables known or expected to be related to expected returns. Our focus isthe opposite; we take the cost of equity estimates as given, and use them to calibrate theearnings attributes. Our calibration is guided by prior research, which generally considersearnings attributes one at a time. Because this research describes the attributes we consideras desirable, we expect that an unfavorable outcome for each attribute, considered alone,is associated with a higher cost of capital. Our results generally confirm this prediction.

Using annual cross-sectional regressions of cost of equity proxies on beta, size, andbook-to-market ratio over 1975-2001, we find that when the attributes are added one at atime, they are (generally) significantly positively associated with the cost of equity capital(at the 0.01 level). The positive associations mean that firms with the least favorable valuesof each attribute have higher costs of equity than firms with the most favorable values,controlling for known risk factors. The magnitudes of these associations are reduced whencontrols for innate determinants are included in the regressions, consistent with a portion

The Accountini> Review. October 2004

970 Francis, LaFond, Olsson, and Schipper

of each attribute being explained by these determinants. In economic terms, the largesteffect (controlling for risk proxies and innate determinants) is a 261 basis point (bp) dif-ferential cost of equity capital between the best and worst accrual quality deciles. Theeffects associated with accounting-based attributes generally exceed those associated withmarket-based attributes; for example, after controlling for risk factors and innate determi-nants, value relevance has the largest cost of equity effect among the market-based attributes{an 81 bp difference in the cost of equity between the best and worst value relevancedeciles).

Prior research does not provide a basis for predicting either differences in the relativemagnitudes of the cost of equity effects of earnings attributes or the conditional cost ofequity relations among those attributes. As we discuss in Section VII. we believe evidenceon relative and conditional effects has practical value for investors, researchers, standardsetters, and firm managers. In conditional tests that consider all seven earnings attributesjointly and control for risk factors and innate determinants, four attributes are stronglypositively associated with the cost of equity: accrual quality (260 bp spread), smoothness(207 bp spread), persistence (110 bp spread), and value relevance (77 bp spread). Weconclude that accounting-based attributes have more pronounced cost of equity effects thando market-based attributes, and that among accounting-based attributes, accrual quality hasthe largest effects.

These results are generally robust to a battery of sensitivity tests, including alternativeeconometric specifications and alternative measures of some earnings attributes, in addition,we show that the inclusion of earnings variability reduces the cost of equity effects asso-ciated with accrual quality and smoothness, as one would expect given that earnings vari-ability is an instrument for these attributes. We note that, by construction, our findings aredependent on the reliability of the cost of equity estimates used as the dependent variablein our tests. As we discuss in Section III. we believe the Value Line estimates used in ourmain tests are preferred to other estimates. We also show (in Sections V and VI) thatanalyses using implied cost of equity estimates derived from PEG ratios and analyses usingex post cost of equity estimates based on realized future returns produce similar results.

The rest of the paper proceeds as follows. Section II frames our research questions inthe context of the relevant literature. Section III describes the sample and data. Section IVdiscusses the innate determinants of the earnings attributes. Sections V and VI report theempirical results, and Section VII summarizes and concludes.

II. CAPITAL MARKET CONSEQUENCES OE EIRM-SPECIEIC INFORMATIONAND DESCRIPTION OE EARNINGS ATTRIBUTES AS SOURCES OE

EIRM-SPECIEIC INEORMATIONIn this section we summarize research that demonstrates that firm-specific information

factors affect the cost of capital in equilibrium. We describe the earnings attributes weconsider and we discuss earnings variability as an instrument for two of these attributes.

Research on Capital Market Consequences of Eirm-Specific InformationSupport for the view that properties of firm-specific information affect expected returns

is provided by, for example, incomplete information models (e.g., Merton 1987), liquidityeffect models (e.g., Amihud and Mendelson 1986; Diamond and Verrecchia 1991; Brennanet al. 1998), and asymmetric information models (e.g., Admati 1985).- Labeling the effect

^ See O'Hara (2003) for an overview of the literature that predicts that asset returns are, in part, a function offirm-specific information.



of firm-specific information properties a "cost of capital" effect in the traditional sense,however, requires an equilibrium model in which information properties are rationallypriced. That is, it requires a model with rational agents, in which information effects cannotbe diversified away.

Two distinct approaches to developing such a model are taken by Easley and O'Hara(2004) and O'Hara (2003), and by Leuz and Verrecchia (2004). Easley and O'Hara (2004)show that in a multi-asset, multi-period setting with informed and uninformed investors,the information risk faced by the uninformed investors is not diversifiable and will thereforebe priced. In particular, required returns are affected by information risk, captured both bythe extent of private information and by the precision of both public and private information,with more private information and less precise information (both public and private) leadingto higher required returns. Taking a different approach, Leuz and Verrecchia (2004) considerthe role of performance reports (e.g., earnings) in aligning firms and investors with respectto capital investments. Poor-quality reporting impairs the coordination between firms andtheir investors with respect to the firm's capital investment decisions and thereby createsinformation risk. Anticipating this, investors demand a higher risk premium; that is. theycharge a higher cost of capital. Leuz and Verrecchia (2004) show that even in an economywith many firms and a systematic component to the pay-off from investment, a portion ofthis risk is nondiversifiable.

Empirical tests of the predicted positive relation between information risk and the costof capital use different characterizations of information risk. For example: Easley et al.(2002) focus on the information asymmetry between informed and uninformed traders,which they operationalize using probability of informed trading (PIN) scores; Francis et al.(2005) focus on accrual quality (measured as the strength of mapping of current accrualsinto cash flows); and Botosan (1997) focuses on disclosure scores based on the quantity ofannual report information. In an international context, Bhattacharya et al. (2003) examinethe association between country-level measures of the average cost of equity and earningsopacity, defined as a composite measure of earnings aggressiveness, loss avoidance, andsmoothness. Each of these studies predicts and finds a relation between the informationfactor(s) they consider and the cost of capital.'

Our research, which is premised on the existence of this relation, examines whetherseven earnings attributes that are viewed as distinct by many in accounting research and inpractice have empirically distinguishable effects on firm-specific risk premia. We interpretthe strength and magnitude of the association between a given earnings attribute and thecost of equity as a quantitative measure of the desirability of that attribute from the per-spective of investors' capital allocation decisions. In particular, we view each earningsattribute as proxying either for the uncertainty in earnings as an informative signal aboutthe pay-off structure that is of interest to investors (as captured by the accounting-basedattributes) or for investors' perception of that uncertainty (as captured by the market-basedattributes). Our study builds on Francis et at. (2005), Barth and Landsman (2003), Cohen(2003), and Barone (2003), who each examine the cost of capital effects of one financialreporting attribute—Francis et al. (2005) examine accrual quality, Barth and Landsman

(2003) examine value relevance, Cohen (2003) examines a measure that is similar to theDechow and Dichev (2002) accrual quality measure, and Barone (2003) examines the Lev

Bhattacharya et ai.'s (2003) results, based on a sample of 34 countries over 1986-1998. are sensitive to thecost of equity proxy. Earnings aggressiveness is positively associated with their dividend-based cost of equityestimates, while loss avoidance is positively associated with the international pricing model-based cost of equityestimates; in neither case does a country's earnings smoothness have a measurable eti 'cct on its cost of equity.

The Accounting Review, October 2004


and Thiagarajan (1993) fundamental score and a second (self-constructed) score based onfinancial statement ratios—but do not address whether the attribute they examine is theonly earnings property that matters to investors, or if it is the one that matters most.

Earnings AttributesIn this subsection we first describe our primary measure of each earnings attribute that

we consider and explain how prior research has characterized each attribute as desirable.We then discuss how the attributes are linked to information risk.

Accounting-Based and Market-Based Earnings AttributesAccrual quality. Several approaches to assessing earnings quality take the view that

earnings that map more closely into cash are more desirable (e.g.. Penman 2001, 611;Harris et al. 2000). Dechow and Dichev (2002) propose and test a measure of earningsquality that captures the mapping of current accruals into last-period, current-period, andnext-period cash flows, and Francis et al. (2005) demonstrate that this measure (which theyterm accrual quality) is associated with measures of the cost of debt and equity capital. Weuse the Dechow-Dichev measure to capture the quality of accruals.

Persistence. This captures earnings sustainability; persistent earnings are viewed asdesirable because they are recurring (e.g.. Penman and Zhang 2002; Revsine et al. 2002,245; Richardson 2003). Analysts sometimes focus on sustainable or recurring earnings (see,for example. AlCPA 1994. Chapter 6). We measure earnings persistence as the slope co-efficient from a regression of current earnings on lagged earnings.

Predictability. Following Lipe (1990), we define this construct as the ability of earningsto predict itself. Predictability is an element of relevance in the FASB's Conceptual Frame-work, and is therefore a desirable earnings attribute from the perspective of standard setters.Predictability is also valued by analysts (see, for example, the AIMR's [1993] descriptionof the distinction between financial reporting and financial analysis) and is an essentialcomponent of valuation (see, for example. Lee 11999] for a discussion).

Smootbness. Discussions of the benefits of smooth earnings include Ronen and Sadan(1981). Chaney and Lewis (1995). and Demski (1998). Arguments that smoothness is adesirable earnings attribute derive from the view that managers use their private informationabout future income to smooth out transitory fluctuations and thereby achieve a more rep-resentative, hence more useful, reported earnings number. We follow Leuz et al. (2003) inusing cash flows as the reference construct for unsmoothed earnings, and measure smooth-ness as the ratio of income variability to cash flow variability.

Value relevance. This construct is often measured as the ability of earnings to explainvariation in returns, where greater explanatory power is viewed as desirable. One streamof this research interprets value relevance as a direct measure of decision usefulness (e.g.,Joos and Lang 1994; Collins et al. 1997; Francis and Schipper 1999; Lev and Zarowin1999). This interpretation rests on the view that value relevance measures capture combinedrelevance and reliability, two key concepts in the FASB's Conceptual Framework (for anextended discussion, see Barth et al. 2001). Our measure of value relevance is the explan-atory power of earnings level and change for returns.

Timeliness and Conservatism. These two attributes derive from the view that account-ing earnings is intended to measure economic income, defined as changes in market valueof equity (see, for example. Ball et al. 2000). The reference construct for both measures isstoek returns; timeliness is the explanatory power of a reverse regression of earnings onreturns and conservatism is the ratio of the slope coefficients on negative returns to the



slope coefficients on positive returns in a reverse regression of earnings on returns. Con-servatism therefore differs from timeliness in that it reflects the differential ability of ac-counting earnings to reflect economic losses (measured as negative stock returns) versuseconomic gains (measured as positive stock returns). Combined timeliness and conservatismare sometimes described as "transparency," a desirable attribute of accounting earnings(see. for example. Ball et al. 2000). Watts (2003) presents several arguments supportingthe view that conservatism is a desirable attribute of earnings; broadly speaking, thesearguments derive from the asymmetric costs of overpayments versus underpayments to firmstakeholders and the role of conservative reporting in constraining such payments.

Earnings Attributes as Proxies for Information RiskInformation risk derives from imprecision (i.e., dispersion) in estimates of the pay-off

structure to investors based on available information (Easley and O'Hara 2004) or fromimpaired coordination between managers and investors with respect to capital investmentdecisions {Leuz and Verrecchia 2004). Both perspectives point to within-firm free cashflows to common equity (FCF^^) as fundamental in the assessment of information risk.That is. FCF^ ' determines investor pay-offs in the form of dividends and capital gains, souncertainty about FCF* - is one source of information risk as characterized by Easley andO'Hara (2004). Similarly, FCF*- is operating cash flows less capital investments (adjustedfor financing) and it is uncertainty about the investment element of FCF^^ that gives riseto information risk in Leuz and Verrecchia's (2004) characterization. Based on these char-acterizations (which develop information risk as a priced factor), we predict that an unfa-vorable outcome for each earnings attribute, considered alone, will be associated with ahigher cost of equity capital to the extent that attribute captures one or more aspects ofuncertainty about future FCF ' .

Prior research does not provide a basis for predicting either the relative magnitudes ofthe cost of equity effects across earnings attributes or their conditional associations (thatis, when all attributes are considered together, does a subset of attributes subsume theothers'?). We provide qualitative arguments that suggest differences in these relations. Forexample, if investors perceive that information quality increases as earnings is more effec-tive in capturing information that is already in prices/returns, then market-based attributesshould have larger cost of equity effects than accounting-based attributes. Conversely, ifinvestors believe that information quality increases as earnings is more effective in allo-cating cash receipts and disbursements across reporting periods, then accounting-based at-tributes should have larger effects.

Among the accounting-based earnings attributes we consider, we view accrual qualityas having the most direct link to information risk. Accrual quality captures variation in themapping of earnings into operating cash flows, a key element of the pay-off structure thatis of interest to investors. We also view persistence as having a direct link to informationrisk, because greater persistence is associated with a more sustainable earnings stream—ii'earnings are persistent, then investors need not be concerned about the extent to which theinnovation in this period's earnings will continue, and this reduces one source of uncertainty.

We view the link between information risk and both predictability and smoothness asless direct than the links between information risk and either persistence or accrual quality.While highly predictable earnings and smoothed earnings do much to eliminate uncertaintyahoul earnings, the link to information risk requires that these attributes not be achievedat the expense of impairing earnings' informativeness about the pay-off structure that is ofinterest to investors. For example, managers might make reporting choices opportunistically



in order to report either or both extremely smooth and predictable earnings, but if thosechoices reduce the ability of earnings to convey information about future FCF * . then theresult will not be a reduction in information risk."* That is, if cash flows are variable, thenreporting extremely smooth or predictable earnings would not assist investors in understand-ing cash flows. In addition, those reporting choices would impair accruals quality, becausethat earnings attribute captures the mapping of accruals into cash flows.

Turning now to the link between market-based earnings attributes and information risk,to the extent value relevance captures information reliability and the latter increases earningsprecision as an indicator of FCF' ' \ greater value relevance will reduce information risk.Value relevance can also be linked directly to information risk: if returns are taken as thesummary indicator of all public information on the pay-off structure that is of interest toinvestors, then an earnings number that also summarizes the same information will beassociated with less information risk. Arguments for a link between information risk andtimeliness parallel those for a link between information risk and value relevance; however,because the measures of timeliness differ from those of value relevance, there is no partic-ular reason to expect either to dominate empirically. We view the link between informationrisk and conservatism as tenuous, because the latter implies, by definition, a bias in infor-mation, while information risk pertains to uncertainty about the true values of pay-offs.That is, conservatism is related to the precision of bad news, while information risk isrelated to the precision of both good and bad news.

We do not specify formal hypotheses about either the cost of equity effects of earningsattributes considered one at a time or their conditional effects when all attributes are con-sidered together. Rather, we aim to shed light on the empirical issue of which attributes aremost strongly associated with capital market benefits in the form of a lower cost of equity.However, we believe that our qualitative discussion of the relations between informationrisk and earnings attributes suggests that, among the accounting-based attributes, accrualquality and persistence will have the most powerful effects and, among the market-basedattributes, value relevance and/or timeliness will have the most powerful effects. This ex-pectation rests on the relatively more direct mappings between these earnings attributes anduncertainty about the structure of future free cash flows.

Earnings VariabilityEarnings variability (the standard deviation of earnings) is an instrument for accrual

quality and smoothness.*^ Dechow and Dichev (2002. Table 4) show that for their sampleand time period, earnings variability is the strongest instrument for their accrual qualitymeasure. Earnings variability is also an instrument for smoothness, since the two measuresdiffer only in the presence or absence of standardization by cash flow variability. Becauseearnings variability is expected to reflect some combination of the two attributes for whichit is an instrument, we do not include it as a separate attribute. Rather, we test the sensitivityof our results to the inclusion of earnings variability; results are reported in Section VI.

'' While persistence is potentially subject to the same earnings management concerns, we believe il is more difficultto use reporting choices (such as accruals manipulation) to make an innovation appear to stay in the earningsseries when, in fact, it does not. Stated differently, predictability and smoothness relate primarily to the variabilityof the earnings series, whereas persistence relates both to the level of earnings and to the variability of theinnovation series.

' Prior research suggests earnings variability may also be an instrument for conservatism. Specifically. Givolyand Hayn (2000) argue that one manifestation of conservatism—the more immediate recognition of badnews—leads to increased variability of earnings. Because we find very small correlations between our measuresof earnings variability and conservatism (-0.02 Pearson. 0.02 Spearman), we do not include conservatism amongthe attributes for which earnings variability may be an instrument.



III. SAMPLE, DATA, AND VARIABLE MEASUREMENTOur sample covers the 27 years, / = 1975-2001. We calculate the earnings attributes

over rolling Hrm-specitic ten-year windows; a Hrm is included in the year / sample if dataare available in years t-9 to t. To mitigate concerns that differences in sample compositiondrive comparisons acros.s attributes, we further require that data on all seven attributes areavailable for each firm-year. Table 1 shows that the number of firms meeting these require-ments (the "Full Sample," representing an average of 52.9 percent of the CRSP marketcapitalization) ranges from 678 to 1.997 per year, for an average of 1.471 firms per yearand a total of 3,917 distinct firms.

Our cross-sectional tests also require data on the cost of equity capital. Our main testsuse Brav et al. (2004) ex ante measures, which are derived from price target and dividenddata provided in Value Line (VL) reports. These ex anle measures use the VL analyst'sfour-year-out target price (7'^) and his forecast of next period dividends [DIV) and dividendgrowth (,t,'). Assuming that interim dividends are reinvested at the firm cost of equity capital(CofC). Brav et al. (2004) arrive at the following expression for the ex ante expected return:

, TP ^^^L CofC -(1 + CofQ^ = - + - ^

where P = stock priee nine days prior to the date of the VL report. The value of CofCthat satisfies the equality is the estimate of the ex ante cost of equity capital. This CofCestimate is qualitatively the same as the VL measure used by Botosan and Plumiee (2002)and evaluated by Botosan and Flumlee (2005).

We evaluate the sensitivity of our results to CofC estimates derived from two PEGapproaches (Easton 2004). We choose VL CofC estimates as our primary measure andPBG-based CofC estimates as our secondary measure for three reasons. Eirst, in a com-parison of the construct validity of four proxies for the expected cost of equity (the VLCofC estimate, a Gordon growth model estimate, a residual income estimate based onOhlson and Juttner-Nauroth [2000], and an estimate based on Ea.stoiVs [2004] PEG ap-proach), Botosan and Plumiee (2005) conclude that the VL-based CofC estimate and thePEG-based estimate outpertbrm the other approaches. In separate tests, Guay et al. (2003)examine the significance of the slope coefficients from regressing future realized returns onimplied costs of equity derived from two residual income models, one implementation ofthe Ohlson and Juttner-Nauroth (2004) model, and a Gordon growth model. The highest t-statistics reported by Guay et al. (2003, Tables 3 and 6) are 0.69 for one-year-ahead returns,2.47 for two-year-ahead returns, and 0.98 for three-year-ahead returns.^ In contrast, whenwe repeat Guay et al.'s (2003) tests for our VL CofC estimates, we find t-statistics of 17.92for one-year-ahead returns, 19.53 for two-year-ahead returns, and 21.98 for three-year-aheadreturns. On the whole, we interpret these results as indicating that VL CofC estimates arestrongly preferred to estimates based on residual income models and weakly preferred toPEG-based CofC estimates.

Second, in our setting, a residual-income-based approach to estimating the cost ofequity (such as used by Gebhardt et al. [2002]), or a PEG approach (such as used by Easton

Easton and Monahan (2(X)4) show no significani positive correlations between realized returns and implied costsof equity from PEG and residual-income-based models that use analysis' earnings forecasts.


976 Francis, LaFond, Olsson. and Schipper

TABLE 1Descriptive Information on the Full Sample, the Value Line Sample, and the Traded Sample

Year

1975

1976

1977

1978

1979

1980

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

Mean

Distitict

Full Sample

# Firms % Market Cap.

678

809923

995

,047

.088

.107

.113

,461

,708

,677

1,636

1.544

1,520

1.523

1,490

1.514

1.577

1,614

1,735

1,828

1,862

1,965

1,997

1,904

1,809

1.602

1,471

3.917

51.9

52.7

51.9

50.2

51.3

50.2

53.6

52.4

58.5

63.3

61.0

59.8

59.9

59.2

56.8

57.4

54.4

51.5

47.7

49.0

47.2

48.3

48.5

49.3

46.8

47.9

46.3

52.9

Value

# Firms

524

604675710739772778769848

1,022

990926897873834781775779775814801782812801770111706

7901,865

Lme Sample

% Market Cap.

49.5

50.1

49.4

47.5

47.8

46.8

50.2

49.3

53.9

58.6

56.4

53.5

52.6

51.9

48.7

48.1

46.5

45.0

41.1

42.1

40.1

40.2

42.2

42.7

40.0

41.7

43.1

47.4

Traded

# Firms %

4.760

4.988

5,138

5,082

5.018

5,153

5,498

5,654

6,168

6,477

6.581

6.983

7.329

1,311

7.194

7,081

7.106

7,397

1,111

8,370

8,681

9,237

9,453

9,349

8,923

7.996

7.092

6,958

18.865

Sample

Market Cap.

99.0

99.8

99.9

99.9

99.9

99.7

99.7

99.7

99.5

99.7

99.1

99.6

99.6

99.7

99,7

99.8

99.6

99.5

98.8

99.5

99.3

99.1

99.4

99.4

97.7

95.8

95.4

99.2

Sample descriptions and variahle definitions:The Full Sample contains firms with data on all eight earnings attributes in a given year /, I = 1975-2QO1, TheValue Line {VL) Sample contains the subset of firms in the Full Sample with data on c.v ante costs of capitalfrom Value Line, The Traded Sample consists of all securities with at least 24 monthly CRSP returns over197.^-2001, '7f Market Cap is the ratio of the market value of the indicated sample to the market value of thetotal CRSP population with share price and shares traded data available in December of the indicated year.Mean and Distinct refer to the average number of firms per year, and the total number of distinct firms,respectively, in the indicated sample.



[2(304]) restricts the sample in important ways, because these earnings-based approachesrequire either or both positive earnings and increasing earnings.^ These restrictions likelybias the samples toward more stable and less risky firms than methods that do not imposethese restrictions (such as approaches ba.sed on price targets and dividend forecasts). Thisbias is problematic in our setting because some of the earnings attributes we consider (e.g.,persistence) may be correlated with stability and risk.

Third, within the set of CofC estimates that use target prices as inputs, prior researchshows that VL data exhibit the least optimism. Brav et al. (2004) report a mean VL CofCestimate of 21 percent compared to a mean First Gall CofC estimate of 45 percent, andthey show that the lower expectations of VL analysts are not driven by time frame or firmscovered, but rather to ]ower optimism for VL ana]ysts versus the array of se]]-side ana]ystsreporting to First CaU.^

We have ex ante cost of equity measures, CofC, for each quarter, for VL firms over1975-2001. Over this period. Value Line followed about 1,700 firms: this sample is reducedby the requirement that a firm has data on CofC and the seven earnings attributes. Table 1shows that the number of firms meeting these requirements (the "VL Sample") rangesbetween 524 and 1,022 firms per year, with an average of 790 firms per year, a total sampleof 21,334 firm-year observations and 1,865 distinct firms. The VL samp]e covers, on av-erage, 47.4 percent of the CRSP market capitalization.

Because VL's 13-week forecast cycle does not conform to either calendar or fiscalquarters, we assign the VL observations on CofC to each calendar quarter using the dateof the VL report containing the forecast information, e.g., a February VL report is assignedto the first calendar quarter. To avoid inflating significance levels by using overlappingobservations, we average the quarterly CofC estimates for year t to generate one CofCestimate for each firm-year. (We find similar results if we use the CofC estimate for thelast quarter of the year.)

Table 2 shows that the yeaHy mean values of CofC range between 12.41 percent (in1997) and 33.19 percent (in 1975); the pooled sample average (median) is 20.8 percent(20.2 percent), with a standard deviation of 7.76 percent. While these CofC estimates mayappear high, we note the following. First, for our sample firms and time period, the averagerealized annual return is 18.7 percent; assuming rational expectations, this implies averageoptimism in the VL CofC estimates of (only) about 2.1 percent. Second, inspection of thedata in Table 2 shows that the largest CofC estimates are observed in 1975-1982 (realizedreturns are also high in this period; the average equal-weighted market return was 2.58percent per month, or about 31 percent annualized). In unreported tests, we find qualitativelysimilar results if we exclude this subperiod from our analyses. Third, values of CofC for

For example, implementations of the Ohlson and Jiittner-Nauroth (2004) model (as in Botosan and Plumiee2005) require positive and increasing earnings forecasts (in order to impulc positive growth rates). Hor tirmswhere analysts' long-term growth forecasts are not availahic, Gebhardt et al. (2001) also require that a firmhave positive and increasing earnings forecasts for year f+1 and i + 2. Easton's (2004) PHG and MPEG ap-proaches also require that earnings forecasts arc both positive and increasing.Further evidence of optimism in First Call target prices is reported by Bradshaw (2002), who documents a meanratio of target price to observed price of 1.36. The lower optimism of VL is consistent with VL not performinga number of investment banking functions, which prior research finds leads to optimism in sell-side analysts'forecasts. In particular, as Francis and Philbrick (1993) discuss, VL performs no underwriting, consulting, orbrokerage functions.



TABLE 2Descriptive Statistics on Ex Ante Cost of Capital Estimates (in percent)'

Year # Firms Mean Std. Dev. 10% 2S% Median 75% 90%

197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001

Mean

524604675710739772778769848

1.022990926897873834781775779775814801782812801770111706

790

33.1929.4829.0829.1831.9732.4629.0531.2019.9722.6119.8815.1914.1817.9816.4719.8118.2816.8814.3815.2014.7813.8712.4113.3815.9919.0116.63

20.83

12.1210.098.617.619.0110.957.248.295.996.829.866.735.786,726.558.077.967.576.646.105.856.375.747.358.289.238.00

7.76

18.5617.1718.2420.0320.4818.0019.9721.3412.2414.0310.747.697.5610.799.8810.649.387.826.138.047.876.586.135.076.338.158.07

11.74

24.6722.4923.4624.1726.2025.5724.1425,8316.1618.0914.8411.0510.6513.9012.5314.0012,6611.509.6210.8110.679.318.528,2110.1212.5211.18

15.66

32.2528.3628.2929.0632.2333.0428.6830.9719,9222.4419.3514.5813.5916.9215.6718.7917.1216.2114.1314.6214.2512.8811.4712.1715,4118.3315.28

20,22

40,8635.9334.5133.4837,8738,7533.5636.2523.9126,9123.2618,4116,7521.2419.3924.0122.4221,4618.3518.6117.8317.2015.0017.1520.5624,3720.50

25.13

50,0242.7140.4338.6043.5646.1238.4141,5527.8131.2927.9622.4421.1426.1924.1730.7227.8726.3522.6622.6622.1122.3920.2322.8926.4331.0426.46

30.53

' We report summary stati.'itics on the yearly distribution of the ex ante cost of capital estimates. CofC. for theValue Line (VL) sample; see Table I for a description of this sample. CofC estimates are the values of CofC

' + cofcf - (1 + gri

that solve the following equality: (ITP

CofCf = — +

DIVCofC - g

where TP = VL 4-

year out largel price; DIV = VL forecast of next period dividends, and g = VL forecast of growth rate ofdividends, and P = stock price nine days prior to the date of the VL report.

our sample are similar to VL CofC estitnates used in other studies.** Fourth, as we show in

For a sample of 122 VL finns in 1991, Boio.san (1997) reports a mean (median) cost of equity estimate of 20.1percent (19.0 percent); tor our VL sample of 775 firms in 1991. the mean (median) CofC estimate is 18.3percent (17.1 percent), Botosan and Plumlee (2002) report mean (median) cost of capital estimates of 16.5percent 115.6 percent) for a sample of 668 firms over 1986-1996; for our VL sample, the mean (median) CofCestimate for this same penod is 16.1 pereent (15.3 percent). Finally, for a sample of 17,930 firm-year ob.ser-vations over 1979-1993. Botosan and Plumlee (2(X)5) report a mean (median) cost of equity of 20.5 percent(18.5 percent); our mean (median) estimates tor this period are 21.4 pereent (18.8 percent).



Section VI, values of the VL CofC estimates for our sample are statistically indistinguish-able from CofC estimates based on PEG ratios. Finally, even if the CofC estimates arebiased upward (for example, due to analyst optimism about long-term target prices), thisbias will not affect our inferences unless it is correlated with the earnings attributes. Somesupport for the view that bias in target prices is unrelated to earnings attributes is reportedby Brav et ai. (2004), who tind no evidence that this bias is correlated with variables knownor expected to be related to expected returns (such as beta, size, and book-to-market).

Our analyses require measures of the seven earnings attributes that we study and ofearnings variability, an instrument for two of the attributes. Our construction of these mea-sures is based on prior research. In defining the earnings attributes, we adopt the conventionthat larger values of the attribute indicate less favorable outcomes. We measure the sevenattributes and earnings variability on a firm- and year-specific basis, using the relevantaccounting and/or market information for rolling ten-year windows, ^-9,.. . , t. (We obtainnearly identical results if we lag the estimation period by one year, i.e., t-\0,..., l-\.) Theuse of the firm as its own benchmark mitigates concerns that differences among firms in agiven industry give rise to noisy measures of the constructs, as would be the case if wemeasured the attributes by reference to industry norms. However, the firm-specific approachrequires a time-series of observations about each firm, while an industry approach requiresonly a sufficient size cross-section of firms in a given industry at a point in time. The ten-year time-series requirement biases our sample toward surviving firms, which are likely tobe larger and more successful than firms that do not meet this data requirement. On balance,we believe that the benefits afforded by the use of the firm as its own benchtnark offsetthe sample bias toward larger, more successful firms.'"

Our measure of accrual quality is based on Dechow and Dichev's (2002) model relatingcurrent accruals to lagged, current, and future cash flows from operations:

where:

TCAj, = firm / s total current accruals in year /, = {ACAj, - ACLj, - ACashj,

AssetSj, = firm/s average total assets in year t and t-\; andCFOj, = cash flow from operations in year l, is calculated as net income before extraor-

dinary items (NIBE, Compustat #18) less total accruals (TA), where:= \CAj_, - ACL,, - ACashj, + ASTDEBT,, - DEPN,,;

jj = firm y's change in current assets (Compustat #4) between yeart-1 and year /;

In Table 10, we report results of tests based on factor-mimicking portfolios that include firms with as few as24 CRSP monthly returns. This sample (18.865 firms) Ineludes smaller and younger finns that lack, lime seriesdata on the earnings aitribules.We use the indirect (balance sheet) approach to estimate accruals rather than the direct (statement of cash flows)approach. Although the former suffers from measurement error in accruals, especially for firms with merger andacquisition activity or discontinued operations (Hribar and Collins 2002), it allows us to calculate accruals fora larger sample of firms and over a longer period than is possible with the direct approach. In particular, thedirect approach requires dala from the statement of cash Hows; these data are not available prior to 1988, theyear in which SFAS No. 95 was effeciive, A ten-year daia requirement would, therefore, restrict our sample tofirms with ihe necessary data in ihe sub-period 1999-2001,


980 Francis. LaFond, Olsson, and Schipper

j = firm/s change in current liabilities (Compustat #5) between yeart- 1 and year r;

\Cashj, = firm/s change in cash {Compustat #1) between year l—{ andyear r;

= firm / s change in debt in current liabilities (Compustat #34)between year l-\ and year /; and

j = tirm/s depreciation and amortization expense (Compustat #14)in year t.

For each firm-year, we estimate Equation (1) using rolling ten-year windows. Theseestimations yield ten firm- and year-specific residuals, v ,, / = f-9,..., t, which form thebasis for tbe accrual quality metric, AccrualQualityj, = a(Vj,), equal to the standard devi-ation of firm/s estimated residuals. Large (small) values of AccrualQuality correspond topoor (good) accrual quality.

Following previous research (e.g.. Lev 1983; AH and Zarowin 1992) we measure earn-ings persistence as tbe slope coefficient estimate, cfc, , from an autoregressive model oforder one (ARl) for annual split-adjusted earnings per share (Xj,, measured as firm/s netincome before extraordinary items in year t divided by tbe weighted average number ofoutstanding shares during year /):

For each firm-year, we estimate Equation (2) using maximum likelibood estimation androlling ten-year windows. This procedure yields firm- and year-specific estimates of 4),^,which capture the persistence of earnings.'- Values of 4), close to 1 imply highly persistentearnings, while values of tfj,, close to 0 imply bigbly transitory earnings. In order to con-form this variable to our ordering of attributes, we use the negative of tbe ARI parameter.Persistence = -(|), ., so tbat larger (smaller) values of Persistence correspond to less (more)persistent earnings. Our measure of earnings predictability is also derived from tbe firm-and year-specific ARl models. Based on Lipe (1990), we use the square root of the errorvariance from Equation (2), Predictability = VCT^ . ) . Large (small) values of Predictabilityimply less (more) predictable earnings.

We define smoothness as tbe ratio of firm/s standard deviation of net income beforeextraordinary items divided by beginning total assets, to its standard deviation of casb flowsfrom operations divided by beginning total assets. Smoothness^, = (r{NlBEj,)/(j(CFOj,).^^Standard deviations are calculated over rolling ten-year windows. Larger values of Smooth-ness indicate less earnings smoothness.

' We use an ARl model {with drift) of annual earnings, rather than a higher order speciHcalion suggested byFinger (1994) and Baginski et al. (1999), because we wish to estimate firm-specific persistence measures for abroad sample of firms over rolling ten-year windows. Using higher-order specifications increases the number ofparameters to be estimated and. therefore, increases the length of the time-series needed for ihe estimation; inturn, this restricts the sample to firms with the necessary data. For example. Finger (1994) estimates AR modelsof orders 2, 4, and 8 for a sample of 50 firms with at least 40 yearly observations over the period 1935-1987;Baginski et al. (1999) estimate ARIMA (2.1,0) models (among others) for 162 firms with a complete scries ofannual data for 1967-1990 (24 years).

'-' Our measure of smoothness is similar to those used by Leuz et al. (2003) and Hunt et al. (2(X)0). Leuz et al.(2003) examine the ratio of the standard deviation of operating income scaled by assets, to the standard deviationof cash flows from opierations scaled hy assets; Hunt et al. (2000) examine the ratio of the standard deviationof nondiscretionary net income (equal to operating cash flows plus nondiscretionary accruals) to the standarddeviation of cash flows from operations.


Coxts of Equity and Earnings Attributes 981

Following Francis and Schipper (1999), Collins et al. (1997). and Bushman et al.(2004). our measure of value relevance is based on the explained variability from thefollowing regression of returns on the level and change in earnings:

^_ ,, Kj^_ j ^ + t,^, (3)

where:

RETj^ - firm/s 15-month return ending three months after the end of fiscal year t;EARNj, = firm/s income before extraordinary items in year t (NIBE), scaled by market

value at the end of year /— 1; andAEARN-^ = change in firm/s NIBE in year /, scaled by market value at the end of year

r - 1 .

We estimate Equation (3) for each firm over rolling ten-year windows. To conform thisvariable to our ordering scheme, we take the negative of the adjusted R' from Equation(3), Relevance = -R]j,^^ii,y Large (small) values of Relevance imply less (more) valuerelevant earnings.

Our measures of timeliness and conservatism are derived from reverse regressions,which use earnings as the dependent variable and returns measures as independent variables:

^_ ,j . , j , , , , ^ , ^, + ^ ,, (4)

where NEG-, = 1 '\i RETj, < 0 and 0 otherwise; all other variables are as previously defined.

Similar to our other attributes, Equation (4) is estimated on a firm- and year-specificbasis, using rolling ten-year windows. Following Ball et al. (2000) and Bushman et al.(2004), our measure of timeliness is based on the explanatory power of Equation (4): sim-ilar to Relevance, we use the negative of the adjusted R' from Equation (4), Timeliness= -Klu^i^y Following Basu (1997), Pope and Walker (1999). and Givoly and Hayn (2000).our measure of conservatism is the negative of the ratio of the coefficient on bad news tothe coefficient on good news. Conservatism = - ( 3 , ^ + (3TJ)/P|^. Larger values of Time-liness and Conservatism imply less timely and less conservative earnings, respectively.

Information about the pooled sample distribution of the earnings attribute measures forthe Full Sample is reported in Table 3, Panel A; for purposes of this table, we winsorizevalues of each attribute to the 99 percent and I percent values (subsequent tests use theranks of variables to avoid outlier concerns). Results (not reported) are similar for the VLSample. The AccrualQuality measure has mean (median) value of 0.026 (0.019); as abenchmark, Dechow and Dichev (2002) report mean (median) values of 0.028 (0.020) fortheir sample of 1,752 firms over 1987-1999. Persistence, which captures (the negative of)the extent to which an earnings innovation remains in the series, has a mean (median) valueof -0.482 (-0.520); as a benchmark, the average implied ARl parameter reported byBaginski et al. (1999) for 162 firms with a complete series of data over 1967-1990 is 0.54.Predictability has a mean (median) value of 0.876 (0.536) and a standard deviation of1,054, indicating both dispersion and skewness. Finally, Smoothness, which captures thevariability of income relative to the variability of cash flows, has mean (median) value of0.640 (0.578). In comparison, Leuz et al. (2003) report a mean smoothness measureof 0.765 (for all U.S. firm-year observations, 1990-1999) and Hunt et al. (2000) reportdescriptive data implying a mean ratio of income volatility to cash volatility of 0.51.



'cuea

Ui•a

"I0X1

c

oU

? 2

lO o r-- 00 m M oU-7 Tt TJ- -H O\ fi-l [--.p p oo — p —• 00

d d — — d d r-I I

rn m QN TJ- r^ ^ - - ^m r^ oo in (N 1^ dO fN ON oo rN <N —

ON O

p in

d d o d

^ o• ^ p

d —I I

00^ ^0O 00 rn rn q ^ ^d d d d d d rj

I 1 1 1

O - ^ ' t fN — r—• ^ 00 rO >.D O ooON — (N r - 00 O

r^ O """ r-l O-l

d — d d d

2 .n

n

I

d d d d d d d

0=II 1

^1

O O 00 m O O ÔN vO o O — " n —^£1 O in i/~i — o p(N — p (N "n "n p

d d d d d d d

ON—- > n t x • ^ • - - - t ^—- • ^ ~ - . — —- QONCD ( N I O O O O N D O O C S COCD O— Ci pO\ —O — O r-JCi OCD ppc^ oci po pci pcD po pd V o o 9 V 9 V ^ V ^ V" -

00

I

( N ^ IT) 00(NO w~. O rnOOCCl \DC3 tÔpt:: po po

V ° V

p O —d \J d

OON ON O O

OOO

O OC OO CO O

1^ —- CO - - (N "-. OO O mo r O O1 O fN o r— O O"d- O tN O ( N O O

00 -~- O •— \D '^^ O ^ O TJ- O

^ O r s> ~" SO O O 'o O So V c> V o V o v <=• V •=• V

Tt -~- o ;::; 1- ;:^ o

d o - - o 00 o tN

o V

a- — cON CD C00 O Or i CD o

— CD rÔ

V V O y O v ' ^ V ' ^ V

01

^ —<. — '•'^ OCO O ON CD —in o NO Oi j O (N O

d v" d V o v



•c

2cc3o

see

I COS u

•£ P

o u

60-r

5 E

lf

• J . 1-

5 o --o i ET^ O

•- -5 '^ E

_g ^ H 5

Ti P P

'u 7:3 --^

§1

.P ejj c : ^ oj1 ^ S := .^ p

F —" F «

;r 3 '"*" 2 -c !5 -—.2 P

0

' " -

ck• a1 ;J-

S3

W

.5' pd j

Eg^tkT3

t^

adi

•6

"o

>

T3

fd

1U^^

•£

i=.lUP0

0p

rati

0

•£

0

>

:0


984 Francis, LaFond. Olsson, and Schipper

T\iming to the three market-based earnings attribute measures. Relevance (the negativeof the adjusted R' in a returns-earnings regression) has a mean (median) value ot" -0.423(-0.416); this compares to a mean (median) value of -0.466 (-0.465) for Timeliness (thenegative of the adjusted R- in a reverse regression of earnings on returns). Our samplevalues of Relevance and Timeliness are roughly similar to those reported by Bushman elal. (2004), for a sample of 784 firms in the Fortune 1000, namely, a mean R- from firm-specific regressions of Equation (3) of 0.37. and a mean R- value of 0.33 for firm-specificregressions of Equation (4). Finally. Consetralism. the negative of the ratit) of the coeffi-cients on negative returns to the coefficient on positive returns in a reverse regression ofearnings on returns, has a mean (median) value of -0.547 (-1.00) and a standard deviationof 27.46.'"' The mean value for our sample is smaller in magnitude than the values reportedby Basu (1997) (who reports 4.66) and by Givoly and Hayn (2000) (who report a range of1.7 to 25.8); our measure of Conservatism is derived from firm-specific regressions, whiletheirs are based on cross-seclional regressions.

Pearson (above diagonal) and Spearman (below diagonal) pairwise correlations amongthe seven attributes are reported in Table 3, Panel B for the Full Sample (the VL samplecorrelations [not reported) are similar). All four accounting-based attributes exhibit positivecorrelations exceeding 0.20 (p-values < .0001). Similarly, all three market-based attributemeasures are positively correlated (p-values < .(KK)1), although only the correlation betweenRelevance and Timeliness is large in economic lerms. 0.67 (both Pearson and Spearman).''^The correlations between the accounting-based attributes and Relevance and Timeliness aregenerally positive (p-values < .0001) and small (i.e.. 0.12 or le.ss). The correlations betweenthe accounting attributes and Consenatism are al.so small and in some cases the Pearsonand Spearman correlations are of opposite signs. Overall, this evidence suggests relativelylittle overlap between the accounting-based and the market-based attributes.

Further evidence on the degree of overlap among the earnings attributes is provided inthe far right column of Panel B. where we report auxiliary R' statistics. The auxiliary R^for attribute k is the R' obtained from regressing attribute k on the other six attributes; thisstatistic is analogous to a multivariate correlation between each attribute and all otherattributes (the square root of the auxiliary R' is the Pearson multivariate correlation). Whilethese data show auxiliary R's as high as 0.51 for some attributes, in no case is the squareroot of this measure substantially larger than the largest pairwise correlation. This resultindicates that the degree of overlap across attributes is reasonably approximated by (nomore than) the largest pairwise correlation for each attribute.

We draw three conclusions from the results in Table 3. First, values of the earningsattributes measured for our sample are roughly similar to values reported in prior studies,except for conservatism. Second, there is relatively little overlap between the accounting-ba.sed and the market-based attributes. This finding is consistent with the view that the twotypes of earnings attributes are premised on divergent views about the purpose of account-ing. Third, within each of these sets of attributes, the correlations across the differentmeasures are generally reliably positive, but, with the possible exception o( Relevance andTimeliness, not so strong as to indicate that any attribute subsumes another. We concludethat the seven earnings attributes are distinct constructs.

'** The large dispersion in Consen'at'ism is driven by observations with small values of the denominator (P, ^) of(his variable. Consvrvausm = • (13, , + p ,^ ) /p , , , Our lesls arc not aiTccled by extreme values of Cimsen'ulism(or extreme values of ulher attribute proxies) because our tests use tbe decile ranks of the variables.

' Relevaiiic and Timeliness are expected to be correlated, given that the regressions have similar variables. Theexplanatory power of Equations (3) and (4) is not, however, identical since each equation contains independentvariables which the other does not. Bushman et al. (2004) make a similar point.



IV. INNATE DETERMINANTS OF EARNINGS ATTRIBUTESThe earnings attributes we consider are jointly delermined by intrinsic (innate) factors,

such as firms' business models and operating environments, and by management's (discre-tionary) reporting and implementation decisions. The discretionary component of each at-tribute is composed of intentional reporting choices, forecasting errors, and implementationerrors; we do not attempt to separate these effects. In addition to documenting the cost ofequity effects of each attribute in total, we wish to isolate the pricing effect of the discre-tionary component of each attribute. To do this, we include in our tests controls for theinnate determinants of the earnings attributes. For this purpose, we augment the five innatefactors considered by Dechow and Dichev (2002) a.s explaining accrual quality (firm size,cash flow variability, sales variability, length of operating cycle, and incidence of negativeearnings realizations) with three variables shown by prior research to influence one or moreof the other earnings attributes: absence of intangibles, intangibles intensity, and capitalintensity."' Briefly. Dechow and Dichev (2002) posit and find that accrual quality is in-versely associated with firm size, and positively related to cash flow variability, sales var-iability, operating cycle, and incidence of losses. We include measures of intangibles usebased on prior .studies* findings that expenditures on R&D and/or advertising lead to im-portant differences in some of the earnings attributes. In particular. Francis and Schipper(1999) find that value relevance differs between high-tech firms (with mean R&D spendingof 9.2 percent of assets) and low-tech firms (with mean R&D spending of O.K percent ofassets): Penman and Zhang (2(X)2) argue that the expensing of R&D and advertising is animportant determinant of conservatism; and Baginski et a!. (1999) show thai intangiblesintensity is positively related to earnings persistence. Consistent with the view that capital-intensive lirms have greater volatility in earnings due to higher operating leverage, Baginskiet al. (1999), as well as Lev (1983), find that more capital-intensive fimis have less persis-tent earnings than less capital-intensive firms.

The innate factors are measured as follows. Firm size is proxied by the log of totalassets iAssets); we obtain similar results using sales revenues. Cash flow variability,(j(CFO), is calculated as the standard deviation of the firm's rolling ten-year cash flowsfrom operations, scaled by total assets. Sales variability, irisales), is the standard deviationof the firm's rolling ten-year sales revenues, scaled by total assets. Operating cycle,OperCyvle, is the log of the sum of the firm's days accounts receivable and days inventory.Incidence of negative earnings realizations. Nef^Eurn. is measured as the firm's proportionof losses over the prior ten years. Intangibles intensity. Int-Intensity. equals the sum of thefirm's reported R&D and advertising expense as a proportion of its sales revenues; missingvalues of R&D and advertising expense are set to zero. The absence of reported intangiblesis captured by an indicator variable, lnl_Dumnn\ equal to I for firms with Int-lnlensity= 0. and 0 otherwise. Finally, capital intensity, Cap^ntensity, is proxied by the ratio ofthe net book value of property, plant, and equipment to total assets.

Table 4. Panel A contains descriptive infonnation on the eight innate determinants. Forthe first five variables, the sample mean values are 5.57 for Assets. 0.074 for <T{CFO), 0.218for (TiSales), 4.74 for OperCycle (or 114 days), and 14 percent for NegEani. In comparison,Dechow and Dichev (2002) report mean values for their sample (1987-1999) of 5.50 for

'" We describe Ihese factors as innate Ui tuplurc the nolion Ihal Ihcy are prcdeiermined wiih resped lo each period'sreporting and impieineruation decisiiins. We do not characterize Ihem as exogenous because they can be changed,albeil slowly and al perhaps cunsiderable cost, by managemcnl. Por example, management can change ihebusiness model (e.g.. by increasing receivables tumovcrj or change ihe operating environment (e.g., by exitinga line of business or a geographic region). Our identificatioii of these factors as innate is merely intended tocapture the fact that these variables are difficult to change in the short nin.



SO

OU

9iJS

H•a

3 1ft a!

0*Q

c'ESi oQ«->C9BB

m 00 — in o O00 m r j o o ONtN — ^ -^ Tt- o

• ^ - ^ oo ON Q<^ ^ r— in oON O rM —I r-4

mq

d d

o rq —d d

oo qo d

C rn O—' q —Tt d d

— oq qd d

d

o"nd

a '2 — o o o o — o o

tf?

.^

ap

n

114* a.

1

Dil

— (N

d d

xnc

qd

qd

d—-II I I

d o vc N nqcj\d~-"

oo o ON(N [^ 00

S ' ^ O Ov O ' ^^ O ON O ' ^

o "n o ^ o dII II II

ONinqd

qd

d— u-i oo

r- — p-005- oors] "~0c •—-ON— ci OIN O~~- ^ " n

NO o in o

00moo , —

0 0

(N — ^ O O " ^ — Ci^ I I

r<-. r^,

or-, rq o qd N d

(N

— Ov

O IN!

— CN

q Ov qd < d

q O q q r-1d Ci -^

Of ^ 0 0 o~- O—• O(NII - " I I I I

0

0

Ov

( N0 "•-1 (N

— o*O '*'~j O^

q "n tN O\ r-;d —i d •*5-' 00

o -- —o 'n o 0 0 — ^ —(N O O O> O q

d

£0

c



SQ

co

n-ye

ar

C—

_1J

CO

n-ye

ar

—

s

ctl

UX iM>

ts r

ece:

c

cou

:ars

;

c—o

• c

X

u00

c

adve

rt!

a'^

e: a

ndL

;rw

i

x;ooy 3

« IE 'r- •=

I I I §

X = •£

•o -o

••- :0 :fl •— t 3 C

_o ^ - _o ^.5 „


988 Francis, LaFond. Olsson, and Schipper

Assets, 0.060 for a{CFO\ 0.215 for a{Sales), 141 days for OperCyde, and 10 percent forNegEarn. A comparison of median values of these variables produce.s similar results (notreported). Turning to the variables capturing intangibles intensity and capital intensity, wenote that about 38 percent of our sample firm-years report zero expenditures on R&D andadvertising. For all sample firms (including those with Int-Inlenslty = 0). the mean valueof Int-Intensity is 6.4 percent. Finally, the mean value of Cap^Intensity is 37 percent.

To determine the importance of the innate determinants and common factors in ex-plaining earnings attributes, we estimate annual regressions of each attribute on the eightinnate determinants. Table 4, Panel B reports the mean of the 27 coefficient estimates,where t-statistics are based on the standard error of the 27 yearly estimates. The resultsshow that the innate determinants explain the most cross-sectional variation in Accrual-Quality (55 percent); this compares to 15 percent, 24 percent, and 33 percent for Persist-ence, Predictability, and Smoothness, respectively. The abiUty of the innate factors to ex-plain the market-based earnings attributes is much lower: 5 percent or less. We view theseresults as indicating that innate determinants explain a potentially significant amount of thevariation in the accounting-based attributes and little or no variation in the market-basedattributes. In tests that include the innate determinants as control variables, we interpret theestimated coefficient on the earnings attribute as a measure of the cost of capital effects ofthe discretionary portion of that attribute.

The discretionary portion will have a discernible cost of capital effect (just as theearnings attribute as a whole has such an effect) to the extent that its components (whichinclude forecasting and implementation errors as well as possibly biased accruals in re-sponse to perverse incentives) give rise to nondiversifiable information risk. A natural ques-tion to ask is whether the portion of the discretionary component attributable to deliberatemanagerial choices can have, in equilibrium, a cost of capital effect? In particular, Ittnerand Larcker (2001) argue that in the presence of consistently optimal managerial choices,there will be no relation between firm performance and the observed choice variables,controlling for exogenous determinants of those choices.'^ However, we believe a relationbetween discretionary components of earnings attributes and the cost of equity will persistin equilibrium, because the cost of equity is not a performance measure; it is a measure ofrequired return (theoretically, a returns-based performance measure is abnormal return, i.e.,total return less required return). In particular, we expect managers will take some actionsthat increase risk (information uncertainty as well as other risks) provided that the risk-adjusted pay-off of those actions is positive. In such situations, we would observe a highercost of capital even though the managerial choices that gave rise to it are optimal. Stateddifferently, in both the Easley and O'Hara (2004) model and the Leuz and Verrecchia (2004)model, all that matters is uncertainty about the pay-off structure; the source of that uncer-tainty is not relevant. This reasotiing is similar to other equilibrium asset-pricing models,such as the CAPM where only beta matters for the cost of capital. In a CAPM world theremay be determinants of beta (such as a firm's business model) that are affected by mana-gerial choices; however, regardless of the source of these effects, beta remains the primitiveelement of the cost of capital.

V. CROSS-SECTIONAL TESTSIn this section, we report the results of our main tests and the results of sensitivity tests.

Our analyses use the annual VL-based ex ante cost of equity estimates, control variables.

" Consistent with this view, Cohen (2Of)3) argues that reporting quality is endogenous, and that once all exogenousfactors are controlled for, there should be no pricing effect of reporting quality.

The Accounun^ Review. October 2004


and earnings attributes. Our control variables are the innate determinants previously dis-cussed and risk proxies known to infiuence the cost of capital: beta, firm size, and book-to-market ratio (e.g.. Fama and French 1993). We measure firm/s beta (Beta) in year tusing the CAPM, estimated using monthly returns data over rolling ten-year windows; werequire a minimum of 24 monthly returns. Firm size (Size) is measured as the log of firm/ s market value at the end of fiscal year t-\. The book-to-market ratio (BM) equals thelog of firm / s book value of equity (Compustat #60) divided by its market value of equity,both measured at the end of fiscal year t-\. Similar to the treatment of the earningsattributes, we assign values of Beta, Size, and BM to each year with a three-month lag toensure the information is known to the market.

To compare coefficient estimates across attributes, we rank each attribute each year,and form deciles. Firms in the top decile (decile I) have the smallest values of the attribute,while firms in the bottom decile (decile 10) have the largest values of the attribute. Giventhe definitions of our attribute measures, this ordering places firms with the best (worst)outcome for the attribute in the top (bottom) deciles. Using the decile rank of each attribute,rather than its raw value, alleviates the effects of extreme observations and allows us tointerpret the resulting coefficient estimates as the incremental risk premium associated withadjacent deciles. To the extent that the attribute captures elements of information risk pricedby investors, we expect a positive coefficient on the decile rank, indicating that investorsattach higher risk assessments to stocks with less favorable (i.e.. larger) values of eachattribute.

Main Tests and ResultsWe begin by examining the relations between the implied cost of equity and earnings

attributes for each year t, controlling for known risk proxies:'"

CofCj, = \u, + K.,Beta., + \._,Size^_, + K,BM^, + \^, Attribute), + ^j_, (5)

where:

Attribute''^, = the decile rank of firm / s value of the kth earnings attribute in year t,

, lAccrualQuality, Persistence. Predictability, Smoothness,\ Relevance, Timeliness, Conservatism

To mitigate concerns about cross-sectional dependencies in the data, we estimate Equa-tion (5) for each of the 27 years in our sample. We report the mean of the yearly coefficientestimates, and assess statistical significance using the time-series standard errors of theseestimates (Fama and MacBeth 1973). To assess the sensitivity of our results to this pro-cedure, we also estimate pooled regressions where we assess statistical inference usingNewey and West (1987) standard errors, which control for unspecified heteroscedasticityand autocorrelation effects. Resuhs of these pooled regressions (not reported) yield similarinferences as the annual regressions.

Table 5, Panel A reports the results of estimating a base model version of Equation (5)that includes only the risk proxies (i.e., it excludes the earnings attributes and the innate

In order to avoid understating the cross sectional variation in risk factors and innate determinant.'* (and, therefore,poientially ovcrsiiiting results ft>r the earnings attributes), our regressions use the raw values of the risk factorsand the innate determinants. Results are not qualitatively affected if we use the decile ranks of these variables.


990 Francis, LaFond, Olsson. and Schipper

O ^

DCB

CQ

•ncat

<

a&C

In

ooth

1

nlA

( ^ m fN (NCM r* r ' lTf r^ -rt1

O' O v^ C

\O fO — -^o — o o

1

S i

Ci,

E:0

OI

.^ — O\CO ~ -

O O

ON (N \D — r-r*") V^ ^ ^ C J ™^u-i c d d o

I

m o

03 •< <

Accounting Review. October 2004


pa

£ —

V. ^

oo•oc

<

CQ

oo O

U

ao

CQ

U

c

00 O— oo o —d — d fN

00 (N CMp - rs<M O O

_ f

o —

O oo rnrj \D d O -H — — m —

r-. — —— d d

— 00 O CO

^ rn Or ^ f<^ ( ^CT^ fVJ fNJ

-^ d di

r i p

C O O O — < N O O

b b

=Q -s 5 T:— 2

•- K * 2 wc ^ 5 g-S

g.i

IIal

l li

ate;

'r E

atio

;

uib

ute.

uu

able

s

3 ra o 2

C *

U

n (/ 4..- ?; C "

S<i nr-as

II_

BU>.

.£:

ibu

teg

ci ic

1

the

e;

uc«a.u

tim

27 e

s

«cc01)

ipni

eg

O

E,• c

esc

pie

d

U o o 3



factors). This regression provides a validation of our CofC estimates; we expect theseestimates to be positively related to Beta (firms with higher betas have higher costs ofequity), negatively related to Size {smaller firms have larger costs of equity), and positivelyrelated to BM (firms with larger book-to-market ratios have higher costs of equity). Themean coefficients on these variables are consistent with all three relations. Specifically, themean coefficient on beta of 5.48 (t-statistic = 14.49) indicates an average implied marketrisk premium of 5.48 percent for our sample, similar in magnitude to historical market riskpremia used in prior studies. The negative mean coefficient on Size indicates that, for oursample, large stocks have smaller expected returns (\-,, = -0.602, t-statistic = -4.22).Finally, the mean positive coefficient on the book-to-market ratio of {\i, = 1.357, t-statistic- 3.48) indicates that firms with larger book-to-market ratios have higher implied costs ofequity than firms with smaller book-to-market ratios.

Panel B of Table 5 reports results of estimating Equation (5) adding, individually, theaccounting-based earnings attributes. Based on analytical models that show that informationquality is priced, we expect that the coefficient estimate on each attribute is positive. X> 0, indicating higher costs of equity for firms with worse outcomes of the attribute.Turning first to accrual quality, the results show a mean estimate of \^ = 0.412 (t-statislic= 11.60). This finding suggests that firms with the worst accrual quality experience a 371(0.412 times nine decile differences) basis point (bp) higher cost of equity capital relativeto firms with the best accrual quality. The next largest effect is found for Persistence, wherethe mean estimate of X^ = 0.216 (t-statistic - 5.85) suggests a spread of 194 bp betweenfirms with the least and most persistent earnings. The third largest cost of capital effect isassociated with Smoothness, \^ = 0.125 (t-statistic = 4.32). The cost of equity effect ofPredictability is indistinguishable from zero ( 4 = 0.046, t-statistic = 0.79).

Panel C of Table 5 reports results adding the innate determinants to Equation (5). Asexpected, many of the coefficient estimates on the innate variables are highly significant,indicating that CofC estimates are partially explained by innate determinants of earningsattributes. Including these determinants decreases the estimated coefficients (and their re-lated t-statistics) for AccrualQuality and Persistence; specifically, in Panel C AccrualQualityhas a coefficient estimate of 0.290 (t-statistic = 8.63), compared to 0.421 in Panel B andthe coefficient on Persistence is 0.113 (t-statistic = 2.52) versus 0.216 in Panel B. Includingthe innate determinants increases the coefficient estimate and related t-statistic for Smooth-ness to 0.223 (t-statistic = 9.22) from its value of 0.125 in Panel B. Finally, including theinnate determinants shifts the statistically insignificant coefficient on Predictability to anegative 0.184 (t-statistic = -4.39).

We interpret these results as indicating that cost of equity effects of management'sreporting choices—particularly those that affect accrual quality and earnings persist-ence—are generally smaller than the overall (innate plus discretionary) effects of theseattributes, but are generally still beneficial. Exceptions concern the results for Smoothnessand Predictability, where we are cautious about drawing inferences about the effects ofmanagement actions. In the case of Smoothness, the inclusion of the standard deviation ofcash flows as an innate factor alters the interpretation of the coefficient on Smoothness: inparticular, it becomes a measure of the cost of equity effect of. essentially, earnings vari-ability (we elaborate on this issue in sensitivity tests described in Section V). In the caseof Predictability, the result, taken at its face value, implies that after controlling for innatefactors that affect earnings predictability, managements' discretionary actions to increaseearnings predictability have adverse cost of equity effects. Consistent with the discussionin Section II, and notwithstanding analysts' characterizations of predictability as a desirableearnings attribute, we interpret this result as suggesting that predictability is not in and of



itself desirable, in the sense of reducing information risk; in fact, predictability that existsover and above that implied by innate determinants contributes to information risk.'''

We perform the same analyses for the market-based factors; results are shown in Table6. The base model results are identical to those reported in Panel A of Table 5 and are notreported. Panel A shows the results of adding each of the market-based attributes to thebase model. We find the largest coefficient estimate relating CofC to Relevance, X^ = 0.127(t-statisUc = 2.88), followed by Timeliness, k^ = 0.083 (t-statistic = 1.81). and Conser-vatism, ^4 = 0.029 (t-statistic = 1.22). The magnitudes of these effects suggest that firmswith the least value-relevant (least timely) earnings pay a premium of about 114 bp (75bp) over firms with the most value-relevant (most timely) earnings. Adding the innatedeterminants to the regression (Panel B) reduces all the coefficient estimates and t-statistics;the coefficient on Relevance drops to 0.090 (t-statistic = 2.10) and the other two estimatedcoefficients lack significance at conventional levels. We interpret these results as suggestingthat management's discretionary actions to affect Relevance. Timeliness, and Conservatismhave relatively modest effects (or no effects at all) compared to the effects of innate factors,and that the cost of capital effects of market-based earnings attributes are generally smallerthan those of accounting-based attributes.

To discern whether some of the earnings attributes we consider subsume the others interms of cost of equity effects, we assess the incremental contribution of each attribute, inthe presence of the others, to explaining implied costs of equity;

+ Xj^^Sizej,^

K + s,,,. (6)

Table 7 reports the results of estimating Equation (6) for the accounting-based earningsattributes (Model 1), the markel-based earnings attributes (Model 2), and all seven earningsattributes (Model 3). Panel A (Panel B) shows results excluding (including) the innatefactors. Turning first to Panel A, the results of Model I show that Accrual Quality has thelarge.st pricing effects: X' "™"' '""'''> = 0.441 (t-statistic ^ 17.79), or a 397 bp cost of capitalspread between the worst and best accrual quality firms. Controlling for other accounting-based attributes. Model I also shows the mean coefficient estimate on Persistence is^p^rsLsMu-,- ^ 0.168 (t-statistic = 3,11). In the presence of the other accounting attributes,Smoothness is insignificantly related to CofC and Predictability takes on a negative coef-ficient, \P^''d<c«'>''>''y = -0.199 (t-statistic = -2.80). That is, once we control fox Accrual-Quality and Persistence, an unfavorable outcome on Predictability or Smoothness no longerhas a reliably positive association with the cost of equity. Controlling for innate determi-nants of the attributes, the results in Panel B show that the discretionary cost of equityeffect ofAccrualQuallty is 262 bp (\f-"""iQ""'i'> = 0.291, t-statistic = 8.66). The discre-tionary cost of equity effects for Persistence and for Smoothness are smaller, 120 and 203bp, respectively (both significant at the O.OOl level) and those for Predictability are reliablynegative. We interpret these results as broadly supporting the inferences we drew fromconsidering the accounting-based attributes one at a time, in Table 5.

The results of Model 2 reveal that only Relevance is a distinctly priced market-basedearnings attribute, regardless of whether we condition on innate factors. Specifically, the

As discussed in Sections V and VI, this result is sensitive to alternative proxies for the cost of equity and analternative measure of Predictability, For these alternatives, we Hnd a weak positive association or no associationbetvi-een the cost of equity and Predictability.



TABLE 6Results of Annual Cross-Sectional Regressions of

Ex Ante Costs of Equity on Risk Proxies and Market-Based Earnings Attributes"

Panel A: Base Model, Plus Market-Based Attributes

Decile Rank Value of Market-Based Attribute

Independent Variable

BetaSizeBMAttributeAdj. R-

Panel B: Base Model,Attributes


BetaSizeBMAttributeInnate controls:

AssetsuiCFO)iT(sales)OperCycleNegEarnInt-IntensityInt-DummyCap-Intensity

Adj. R-

RelevanceCoef.

5.482-0.615

1.3410.1270.174

t-stat.

14.89-4.44

3.492.88

TimelinessCoef.

5.474-0.607

1.3190.0830.175

t-stat.

14.56-4.39

3.461.81

Plus Market-Based Attributes and Controls for

ConservatismCoef.

5.481-0.600

1.3610.0290.169

t-stat.

14.55-4.20

3.491.22

Innate Determinants of

Decile Rank Value of Market-Based AttributeRelevance

Coef.

4.121-1.926

0.2670.090

1.61015.9070.5610.5362.026

-1.15612.416

-0.1800.218

t-stat.

12.53-8.06

0.832.10

8.147.680.791.881.97

-1.333.15

-1.13

TimelinessCoef.

4.096-1.921

0.2390.055

1.60815.6850.6380.5572.076

-1.11012.829

-0.1880.219

t-stat.

12.34-8.10

0.741.18

8.137.580.901.971.99

-1.283.14

-1.18

ConservatismCoef.

4.094-1.869

0.3250.014

1.55615.3290.6300.5822408

-1.00212.919

-0.1970.214

t-stat.

12.32-7.68

0.980.58

7.926.850.842.012.32

-1.153.11

-1.25

* Each year, t = 1975-2001, we eslimate ihe cross-sectional relation between VL CofC estimates, and knownrisk proxies {Beta. Size, and BM). innate factors, and the decile ranks of the earnings attributes consideredseparately. We report the mean of the annual coefficient estimates; t-statistics are based on the standard errorsof the time-series of 27 estimates. Panel A shows results for the market-based attributes considered individuallyand excluding innate factors, and Panel B shows results for the market-based attributes considered individuallyand including innate factors.

Sample description and variable definitions: See Table 1 for a description of the Value Line (VL) sample, Table3 for definitions of the attributes, and Table 4 for definitions of ihe innate determinants.

mean estimate of x^^'''^-"'"'^ = 0.134 {Panel A, t-statistic ^ 3.40) suggests a 121 bp spreadbetween the best and worst value-relevance firms: this effect is reduced to 0.097 (Panel B,t-statistic = 2.73) when we condition on innate factors. Neither Conservatism nor Timelinessis reliably associated with the cost of equity in the presence of Relevance.

Results including the innate factors and all earnings attributes are reported in the farright columns of Table 7 (Model 3). These results are broadly similar to those reported for



TABLE 7Results of Cross-Sectional Regressions of Ex Ante Costs of Equity on

Risk Proxies and Both Aecounting-Based and Market-Based Earnings Attributes'

Panel A: Base Model, Plus Earnings Attributes

Model 1 Model 2 Model 3


BetaSize

BMAccrualQualityPersistencePredictahilitvSmoothnessRelevanceTimelinessConservatismAdj. R^

Coef.

4.289-0.157

1.7330.441

0.168-0.199

0.010———

0.198

Tests of Incremental Explanatory

Model 3 versus Model 2Model 3 versus Model 1

(-stat.

12.92-0.93

3.76

17.793.11

-2.800.41——

—

Power*"

Coef.

5.439-0.612

1.307———

—

0.134

-0.0130.0220.177

Inc. R

0.0280.007

t-stat.

14.82-4.42

3.40————

3.40-0.31

1.00

Coef.

4.270-0.169

1.706

0.4420.156

-0.1870.0040.084

-0.0100.0030.205

t-stat.

9.873.71

t-stat.

12.88-1.06

3.74

18.303.17

-2.720.172.39

-0.240.16

p-value

<.OOO1<.OOO1

Panel B: Base Model, Plus Earnings Attributes and Controls for Innate Determinants

Model 1 Model 2 Model 3


BetaSizeBMAccrualQualityPersistencePredictabilitySmoothnessRelevanceTimelinessConservatismInnate controls:

Assetsa(CFO)(T( sates)OperCycleNegEamInt-IntensityInt-DummyCap_ Intensity

Adj. R^

Coef.

3.458-2.232

0.2880.2910.133

-0.4300.226———

2.35832.330-0.614

0.4281.585

-1.5588.652

-0.0900.236

t-stat.

10.99-8.08

0.798.662.70

-8.3210.42———

10.2412.21

-0.861.581.58

-2.002.37

-0.56

Coef.

4.072- I . 9 I I

0.246

————

0.097-0.013

0.011

1.59316.0420.5930.5452.010

-1.16712.637

-0.1500.221

t-stat.

12.34-8.03

0.76————

2.73-0.29

0.46

8.117.91

0.851.921.93

-1.343.18

-0.94

Coef.

3.444-2.278

0.2330.2890.122

-0.4220.2300.086

-0.0030.007

2.39524.285-0.663

0.4041.199

-1.6968.577

-0.0480.242

t-stat.

10.95-8.50

0.678.552.72

-8.5510.602.58

-0.060.35

10.3114.53

-0.991.511.20

-2.162.43

-0.29

(continued on next page)



TABLE 7 (continued)

Tests of Incremental Explanatory Power'' Inc. R t-stat. p-value

Model 3 versus Model 2 0.021 7.70 <.0O01Model 3 versus Model t 0 006 3.79 0.001

^ Each year. ( = 1975-2001, we estimate the cross-sectional relalion between VL CofC estimates, and knownrisk factors {Beta, Size, and BM). innnte faclors, and the decile ranks of the earnings altrihules consideredjoinliy. We report the mean of the yearly coofticient eslimates; t-statistics are based on the standard errors ofthe time-series of 27 estimates. Model I shows results for the set of accounting-based attributes: Model 2shows results for the set of market-based atlributes; and Model 3 shows resulls for all attributes. Panel Areports results excluding the innale factors and Panel B reports results including them as independent variables.

'' Each year, we assess the incremental explanatory power of the accounting-based attributes (Model 3 versusModel 2) and of the market-based attributes (Model 3 versus Model 1|. We report the mean incrementalexplanatory power calculated across the 27 yearly estimates, along with t-statistics of whelher that meandifference is reliably different from zero.

Sample description and variable definitions: Sec Table 1 for a description of the Value Line (VL) sample, Table3 for definitions of the attrihutes. and Table 4 for definitions of (he innate factors.

Models 1 and 2. In particular, conditioning on innate factors (Panel B) we tind the largesteffects for accrual quality (x^"™""^""'"^ = 0.289, t-statistic = 8.55), followed by smoothness(Xf """''"••" = 0.230, t-statistic = 10.60), persistence {K^"""'""- = 0.122, t-statistic = 2.72)and relevance (\^'-''-'"""'^ = 0.086, t-statistic = 2.58). Timeliness and Conservatism are ln-signiticantly associated with CofC, and Predictability retains its negative conditional relationwith the cost of equity.

Tests of the incremental explanatory power of the accounting-based earnings attributesand of the market-based earnings attributes are reported in the bottom rows of each panelof Table 7. These results show that both sets of attributes add signiticantly to each otherand to risk proxies {Panel A), as well as to each other, risk proxies, and innate factors(Panel B) in explaining cross-sectional vaiiation in the cost of equity. In particular, theaccounting-based attributes add an average of 2-3 percentage points in explanatory power(the average is calculated across the 27 yearly regressions), reliably different from zero atbetter than the 0.001 level. The incremental explanatory power provided by the market-based attributes is smaller, averaging 0.7 percentage points (t-statistic = 3.72). In unreportedtests comparing the difference in incremental R-s provided by accounting-based attributesversus market-based attributes (that is, the mean of 2-3 percent versus the mean of 0.7percent), we tind that accounting-based attributes provide signiticantly more explanatorypower than do market-based attributes (at the 0.001 level).

Summary and Sensitivity Tests

We draw several inferences from the results of the cross-sectional tests reported inTables 5-7. First, when considered individually, all but two attributes are associated withthe cost of equity in the predicted way; the exceptions are Predictability and Conserx'atismwhere we find no reliable associations. Second, comparisons of incremental explanatorypower show that accounting-based earnings attributes explain more of the variation in exante estimates of the cost of equity than do market-based earnings attributes. Third, amongthe accounting-based earnings attributes, accrual quality is the most priced earnings at-tribute, followed by persistence and smoothness. Fourth, among the market-based earnings


Co.sts of Equity and Earnings Attributes 997

attributes, only value relevance is consistently associated with ex ante cost of equity esti-mates. Fifth, when we control for innate determinants, the cost of equity effects are gen-erally smaller, as we would expect. We interpret the resulting coefficient estimates con-trolling for innate factors as capturing the cost of equity effects of management's reportingchoices and estimates.

We perform numerous sen.sitivity checks on the results in Tables 5-7. First, we examinethe sensitivity of the findings to the choice of the benchmark model of expected returns.Results {not reported in tables) based on the CAPM generally show larger cost of equityeffects for the earnings attributes; however, the pattern of the effects is preserved.

Second, we use an alternative method to identify the discretionary component of eachearnings attribute. Specifically, we calculate the prediction errors from annual regressionsof each earnings attribute on the eight innate factors, Disc^ttrihute^. We analyze the pricingeffects of the discretionary component by substituting Disc-Attribute^ for Attribute^. Results(not reported in tables) show that, with the exception of accrual quality, the cost-of-equityeffects are similar to those reported in the tables. In the case of accrual quality, we find asmaller pricing effect for Disv^ttribute^'"'""'^""'''^ (120 bp [t-statistic = 3.95] versus 261bp in Table 5).

Third, we evaluate the robustness of our findings to using the level of total accruals asthe measure of accrual quality. Firms with larger values of absolute total accruals (wheretotal accruals, TA, is defined as in Equation (I)) are expected to have larger cost of equitypremiums than firms with smaller values of absolute TA. Results (not reported) using thetotal accruals measure are similar to those based on AccrualQuality.

Fourth, we evaluate an alternative measure of earnings predictability based on analysts'forecast errors taken from the Zacks database for 1980-2000 (the period over which Zacksdata are available); specifically, for each firm, we calculate the average absolute forecasterror of analysts" annual earnings forecasts. Forecast error is mea.sured as the analyst'sforecast of EPS less reported EPS, scaled by share price ten days before the forecast date.We do not regard this measure of earnings predictability as a measure of an accounting-based attribute because it introduces behavioral influences in at least two forms. First, themeasure contains the effects of analysts' self-selection biases, cognitive biases, and theirincentives for optimism; second, it contains the effects of management "guidance" to an-alysts (see, for example. Bartov et al. 2002). Both influences are expected to affect forecasterrors, but do not necessarily affect the quality of information contained in the earningsnumber itself. (However, management manipulations of earnings in order to affect analystforecast erTors could very well affect the quality of the information in earnings.) If theanalyst-based predictability measure is an instrument for the accounting-based measure,then we expect that larger absolute forecast errors are associated with larger costs of equitythan smaller forecast errors. Results (not reported) using this analyst-based proxy for earn-ings predictability show positive cost of equity effects (t-statistics of 2.22 unconditional oninnate factors and 1.70 conditional on innate factors).

Fifth, we evaluate an alternative measure of conservatism based on Penman and Zhang's(2002) C-Score measure, which equals the sum of the LIFO reserve, estimated R&D assets,and estimated advertising assets, scaled by net operating assets. Following their procedures,we calculate C-scores for all sample firms with a positive value of at least one of thereserve items (14.620 firm-years and 1,448 distinct firms), and we repeat the tests in Tables6 and 7 for this measure. Results (not reported) are consistent with the previously docu-mented results insofar as we find that the coefficient e.stimates on C-score are generally not



reliably different from zero. Penman and Zhang (2002) also calculate a Q-score that cap-tures the effect of conservatism on the quality of earnings; results using Q-scores (notreported) shovf smaller effects than C-score.

Our final sensitivity check considers earnings variability as an instrument for accrualquality and smoothness. Following Dechow and Dichev (2002), we measure earnings var-iability as the rolling ten-year standard deviation of tirm/s net income before extraordinaryitems, scaled by beginning of year total assets, EarnVarj, = (riNlBEj,). Larger (smaller)values of EarnVar indicate more (less) volatile earnings. For the Full Sample, the meanand median values of EarnVar are 0.048 and 0.033, respectively, with a standard deviationof 0.049 (not reported). As would be expected given our characterization of EarnVar as aninstrument for AccrualQuality and Smoothness, it is highly correlated with both: the Pearson(Spearman) correlations between EarnVar and AccrualQuality and Smoothness are 0.77(0.84) and 0.57 (0.66), respectively. Finally, EarnVar is strongly associated with the innatedeterminants; together, they explain 67 percent of the variation in EarnVar (recall, fromTable 4, that the highest explanatory power of the innate determinants for an earningsattribute is 0.55, for AccrualQuality).

Table 8 reports results of re-estimating Equation (6) to include the decile-ranked valuesof EarnVar. -" Beginning from the left, the first and second sets of results show the cost ofequity effect of earnings variability unconditionally and conditional on innate determinants.Specifically, we find a 382 bp unconditional difference (t-statistic = 10.87) between themost and least variable earnings deciles and a 329 bp conditional diiference (t-statistic= 10.42). The far right columns show the results of adding other earnings attributes (bothunconditionally and conditioned on innate determinants); we believe these results shouldbe interpreted carefully because of significant collinearity among EarnVar, Smoothness, and(j(CFO). The cost of equity effect of earnings variability increases to 472 bp (unconditionalresult, t-statistic = 12.64) and 477 bp (conditional result, t-statistic = 7.56), and there is apronounced effect on the results for both AccrualQuality and Smoothness. The cost of equityeffect of AccrualQuality, conditional on other earnings attributes and innate determinants,decreases to 135 bp (t-statistic = 4.40) from its value of 260 bp in Panel B, Table 7; thecost of equity effect of Smoothness, conditional on other earnings attributes and innatedeterminants, becomes negative or insignificant in the presence of EarnVar. These resultsindicate that EarnVar acts as a powerful instrument for both AccrualQuality and Smoothnessand therefore captures aspects of the cost of equity effects of these attributes.

VI. ALTERNATIVE MEASURES OF THE COST OE EQUITYIn this section we examine whether our results are robust to two alternative proxies for

the cost of equity capital. The first examines CofC estimates derived using earnings-basedapproaches and the second uses asset-pricing regressions based on realized returns.

Earnings-Based Cost of Equity EstimatesWe consider the sensitivity of our VL CofC results to two PEG-based approaches to

estimating the cost of equity. The first is the modified PEG ratio approach (MPEG) (Easton2004, Equation (11)), which proxies for the cost of equity as the positive root of r^

PJ - (EPSj - EPS^)/P^i = 0, where r is the implied cost of equity, DPS is the

-" We use decile ranks in itrdcr lo compare ihe cost of equity effects of EarnVar with the other attrihutes (whichare also measured in decile ranks). The use of decile ranks also breaks the perlect collinearity that wouid existatiiong u(CFO). EarnVar. and Smooihiess if these variables were measured using raw values: we note, however,that the correlation among these three variables remains high even using decile ranks.



Ifl"oJ SJ a?"

.2

Ofc'SL.

oc

30 .2

OS c

o

scu—o

C9BQ

—

—

o

CQ

"Ocn

iss:

• ^

at* 43

V)Ue

1ca

t «

"ocoai«

ec1—1

9i

5 a

Co

— r-t o O — — « -. '"^

a 00

fN d

m i N O O O O O O O O OI I I

oI

00 O O

o o —' \O fN' t (N ^I — I

fN V-,

o —o o

d

(N

d

— r- o O o — r iI I ^

— I

r-i r j —^ r~- rno rn *od d — O

o

d

3

12 ^ i >

c:k j

Jthn

es•a

nce

line

ss

a; K

e co

n

b b

2 S

II

5 e03T3 t ;

. O

^ a.t o •-

a?:

<U i ;I

- x:

. ^ ^

K "

= 1

•a o

fn =

2iic ^ «j

•3 IC

•c o

,M50

OJ

C 3 !5

u c: c;?3w -

t ^ * - ' OJ

o S o•— tj "a

iJ E y^ ^ «c « ^

- - -a iJ

•S <=

u u1) TD-a k]

« dj o

S ' gE.° o

— 5i o S

flj ,— C |_

^ <u Vto T3 > ,

Accounting Review, October 2004

1000 Francis. LaFond, Olsson. and Schipper

expected dividend per share, EPS, and EPS2 are the analyst consensus earnings per shareforecasts, and P,, is the observed share price on the forecast date. The second is the PEGratio approach (Easton 2004, Equation (12); Botosan and Plumlee 2005, Equation (4)),which proxies for cost of equity by r = \/(EPS2 - EPSt)/Pi,. For both approaches, we useend of fiscal year consensus analyst earnings forecasts from Zacks, and we set the expecteddividend per share equal to last year's value. Following Easton (2004), we also require thatEPS2 ^ EPSj > 0. In total, these requirements produce a sample of 10,944 Hrm-yearobservations from 1980 to 2(H)0.

We begin with a comparison of the levels of the MPEG CofC estimates with the VLCofC estimates; results for PEG CofC are similar and are not discussed. To control forfirm-specific factors affecting costs of equity, we focus this comparison on firms that haveboth VE CofC estimates and MPEG CofC estimates (n ^ 8,294 firm-year observations).For this sample, the average MPEG cost of capital of 17.1 percent (calculated using thejoint estimation procedure prescribed by Easton 2004) is statistically indistinguishable fromthe 17.4 percent mean VE CofC estimate.

Next, we repeat the cross-sectional tests for the sample of firms for which we haveMPEG CofC estimates (n = 10,944 firm-years). The main results, summarized in Table 9,show roughly the same pattern of coefficient estimates on the earnings attributes as the VLCofC estimates. In general, these patterns are robust to the inclusion or exclusion of theinnate determinants; the exceptions are that the positive coefficients on Predictability andConservatism disappear when the innate determinants are included. We continue to findthat the accounting-based attributes, generally, have substantially higher coefficients thanthe market-based attributes, and within the set of accounting-based attributes, Accrual-Quality has the largest coefficient estimate (0.542, t-statistic = 10.63, unconditional oninnate factors). When we condition on innate factors, the coefficient estimate onAccrualQuality decreases to 0.249 (t-statistic = 3.57), somewhat smaller than the coefficienton Smoothness (0.329, t-statistic = 6.03). We note that all coefficient estimates are largerusing the MPEG CofC estimates than using the VE CofC estimates. While the larger co-efficients for MPEG CofC could mean that the VE CofC estimates underestimate the effectsof the attributes, it is also possible that the MPEG CofC estimates overstate these effects.

Asset-Pricing RegressionsOur second sensitivity test uses cost of equity proxies based on portfolios of realized

returns, a traditional proxy for expected returns in the empirical asset-pricing literature.This proxy assumes rational expectations, i.e., that ex ante unknown information (infor-mation surprises) will cancel in aggregate, so that it is appropriate to use future realizedreturns as proxies for expected returns. In these research designs, it is crucial that inferencesare based on portfolios to allow information surprises to cancel out. The realized returnsapproach provides two advantages. First, it is not based on estimates of the cost of equitycapital, so it is not subject to the same sorts of concerns about measurement error. Second,it allows us to expand the sample considerably (we include all listed firms with at least 24monthly returns over our 27-year sample period). Thus, selection bias due, for example, tothe type of firms followed by analysts is not present. The use of realized returns also hastwo disadvantages: realized returns are noisy, necessitating large sample sizes, and therational expectations assumption may not hold (that is, investors' expectations of futurereturns may differ from return realizations).

Concerning the first issue (sample size), because we use all listed firms with at least24 monthly returns, we believe our portfolios are of sufficient size to mitigate concerns



4

I

Ofcnj

P-

t— ON

ro d1

r^

0 0or j

u

13

3

c

a;-a

a c5

— 00— 00

• f i

o C

fiS

— O

Tt — n-1 sC>Ai r - 00 cO ; — O ; ' t

d o r ; c

00

oo m— p— d00 <n O rn<~n O "A ^d — o< d

If£ <B S

3 CO

= 5 = ?-' Ir-i d 00 00 "S

I

c

o

IVI

03

"3e

O Oin 0011 d

O 00

ti S M

00 o

d -— I

— r- 00 r-O - ^ r- 00^ ^ pC —

o mI I

o\ n o oCv (N — Tl-rn r-; in -^d — rN d

r-i m fN a o

o 00 m(N Tf (N

r t O —• 00O r-) t^ Or-1 m r- O

• — • v^ c ^

p a -s "

S

O r>i O ri-j r-i

m ON — o 00sq •rr ^ p in

T1" O ~" ON•rT ri 00 OA00 p — rnd rJ r-i d

I I

r J O r ^ O o o O T f —•^ I I I

On^; in 00 — oq (N i n

— in r in r i o o ^-

U

f^ f"^ 00 ^ Ô f ^ ON " " .<n [^ r-1 pd — ri dI I

O \ c o o r [ c of N O C O O O O O O —<7N ô "n r-- r j i n o^ p

d r-i m — rr r-i ov O —

ind

r-lo 3 NC m (T-jON _ __

O —I

U

oc Or~~ 00

d —

ON

d

— i n oo i n

r-lO^t^l^•fJ•mfO'no o o r - i — - r * m O ' ô ^ t ^ i n ' ^ N O o o O '

£•"3

ij 3 • -1 - £ ^ r"- g S g rj kj rtH o 2 o iS '"''S3 - =" :S ^ ' -a

I 3 < -. I - ^

S-.2I ^ S" i"

b bI- J JC ^ ^

03

00 u d CQ



with the noise in realized returns. As for the second issue (the rational expectations as-sumption), our research design cannot address its validity. We note, however, that the maincriticism against using realized returns to proxy for expected returns is that realized returnsmay be biased over the period of study (Elton 1999; Fama and French 2002). For example,if there is systematic good news during the sample period, realized returns will be upwardlybiased proxies for expected returns. In general, such bias will not affect analyses of cross-sectional variation in expected returns; only if the bias is correlated with the investigationvariables will inferences be affected. On this point, we are not aware of any study thatlinks bias in realized future returns with the earnings attributes we examine. Further, wenote that the cross-sectional determinants of ex ante cost of equity proxies (such as VLCofC estimates) are the same as the cross-sectional determinants of realized returns (beta,size, book-to-market; see, for example, Botosan and Plumlee 2005; Easton and Monahan2004; Fama and French 1993), suggesting that cross-sectional properties are not meaning-fully affected by bias (if it exists) in realized returns.

Our asset-pricing tests use realized monthly returns, factor-mimicking portfolios, andtime-series tests to assess the significance of each earnings attribute in explaining variationin returns. To conform the yearly measures of earnings attributes to the monthly CRSPreturns, we assign the value of each yearly attribute to the months comprising that fiscalyear. To ensure that these data are available to investors, we lag the assignments by threemonths, to allow for the 90-day lO-K filing period. (Because Equation (1), used to measureaccrual quality, requires information about cash flows in year f+1, we lag the measure ofaccrual quality by one year to ensure that this information is known at time /.)

We begin by calculating a factor mimicking portfolio for each of the k earnings attri-butes. Factor';^, equal to the difference between the equally weighted monthly returns offirms in the bottom three attribute deciles (deciles 8, 9, and 10) and the equally weightedmonthly returns of firms in the top three deciles (deciles 1, 2, and 3).^' This procedureyields a series of m = 324 monthly Factor^, returns for each of the it = 7 attribute factors.We use a similar procedure to calculate factor-mimicking portfolios for the eight innatedeterminants: Factor^,, I = [Assets, (T{CFO), (T(sates), OperCycle, NegEam, Int^ntensity,IntJDummy\ Cap-Intensity ].

Correlation tests (not reported in tables) show that the innate mimicking factors arehighly correlated with each other. We attempt to reduce this collinearity by reducing thedimensionality of the set of innate factors. Specifically, we perform a factor analysis toidentify the N common factors that explain common variation in the eight innate factors.The factor analysis identifies two factors, which we label as CFl and CF2, in total, thesefactors explain 85 percent of the cumulative variation in the innate determinants. For eachcommon factor, we calculate a factor-mimicking portfolio, Factor^J^ and Factor';,!'-. Weuse these two common factor-mimicking portfolios, rather than the eight innate factor-mimicking portfolios, in subsequent tests that condition on innate determinants of earningsattributes.

We assess the importance of each attribute factor in explaining asset prices by aug-menting the standard three-factor pricing regression with Factor'^,;.

' Our construction of the attribute factors follows Carhart's (1997) construction of a price/returns momentumfactor. Our results are not sensitive to whether we equally weight or value weight securities in the attributemimicking portfolios. Results are also not sensitive to the cutoffs used to torn the factor; specifically, we drawsimilar inferences if we form the factor using the difference between the lop and bottom I, 2. or 4 deciles.



= «, + ^(/?M*.. - ^.J + -

where:

Rj„, = firm/s retum in month m;/?p,,, = the risk free rate in month m;

^Mkim = ihe market return in month m;SMB,,, = small-minus-big factor in month m;HML,,, = high-minus-low book-to-market factor in month m; and

= attribute k factor lor month m, as well as /: = CF/ and CF2.

Variants of Equation (7) are estimated oti a finn-specific basis for all listed firms withat least 24 monthly CRSP returns. Because our tests require data only on the attributefactors and the innate factor-mimicking portfolios (which are based on monthly stock re-turns), atid not on the underlying data supporting these factors (the eamitigs attributes andinnate determinants themselves), these tests are not restricted to firms in the Full Sample.Stated differently, we use data on the earnings attributes and common factors for firms inthe Full Sample to create Factor^^, Factor^^f^', and Factor';,^, which can then be correlatedwith the returns of any firm with returns data. The only requirement we impose is that thetiim have at least 24 monthly retutns observations to estimate Equation (7); in total J= 18,865 distinct firms meet this requirement (the "Traded Sample"), with these firmshaving a mean (median) number of observations of 110 (93) months. Table 1 shows thatthe Traded Sample comprises, on average, 99.2 percent of the CRSP market capitalization.

To benchmark our results, we estimate Equation (7) excluding the attribute factors andthe innate factors. Panel A, Table 10 reports the mean coefficient estimates and adjustedR s for the resulting three-factor pricing regressions, calculated across the J firm-specificregressions; t-statistics are based on the standard errors from the J coefficient estimates.Consistent with Fama and French (1993), we find that the market risk premium (R^ki- Rf.), the size factor (SMB), and the book-to-market factor {HML) have significant positiveloadings (t-statistics range betweeti 22 and 156). The mean firm-specific adjusted R- forthe three-factor regression is 14.3 percent.

Panel B of Table 10 shows the results of regressions that include Factor'',,, as an addi-tional independent variable; these tests allow us to assess the degree to which each attributefactor overlaps with and adds to known risk factors in explaining returns. Our emphasishere is on the sign and statistical significance of the loadings on the attribute factors, noton the point estimates of those loadings." Results for the accounting-based attributes showthat, in all cases. Factor^,,, enters with a significant positive loading. In particular, the meanloading on Factor^''''"''^""'"' is / = 0.523 (t-statistic = 47.85); / = 0.674 for

Unlike the comparisons of poini eslimales in Section V. it is not straight!onvard to compare the magnitudes ofthe factor loadings because the factors themselves differ across attributes. While the mean values of the factorsfor accrual quality and earnings persistence are noticeably larger than other attribute taetors (20-26 bp per monthfor accrual quality and persistence versus less than 5 bp per tnonth for the others, not reported), we are hesitantto draw inferences from these means because, like known risk factors, the attribute factors display considerableover-time variation. As Fama and t-Vench (1997) discuss in the context of R^^^, ~ R,., SMB and HML. it isdifficult to draw inferences about the magnitude of cost of capital effects when the standard error around themean factor value is large. As a benchmark, Fama and French (1997. Table I) report standard errors for R^i^,- Rf, SMB and HML of 23 bp. 15 bp, and 13 bp per month, respectively. Our attribute factors display similarvariation, ranging from 5 bp to 26 bp per month (not reported).



O(N —O O

Off

c

a.

O Of ^•Z

p

4J OS

eIS

a>"S

0Q

n4!

c

I

>£) O

^ o dO CN —

— 00

El,

pO^

t t m C QOOS — m o "^00 OS fN fN—

d d d d d

OSOS

f

d d

06 -^ — r i

d

OS00

OO

IT) —

d d

sD Tj

as n-i

O m

OS



I.so

aa

c'2.

O 03t! soca L.

C tjO CQC by

c ac

o .Su ^

on

3

COCO

cn0.

u

o CTs >/~.

— rn O

V, x~~ rr-, r-XO O CO —

00 00 O f^— r-i O —— — r ; Of

O — r-d d d

CO (N

p C

o o o o

o —

00 Tt fvj —

d d d d

Oo_d d c d

r-t r- ^ ^r- m CO ONoo ' i ; rJ rd d d d

^D (N —d d d

^D (N —d d d

< o ^\D fN —I

d d d

c o o

od ON

CO

d

— -t r- o— ON ( N 00r i I m —

•, o — r-/ r-i CO o1 O in fN

rn (N —

d d> d

0-

p S O ¥ -S '^

o ra i2 x: 5 ..

• I I O -a ^

S o ?

^fc-p g ^ g-g i t i := C i j g E « « A : ~ - a

2 ^ "y •£ S-1 o r t£

3 ca IH*

o o S-P y

•G "5 E «= .2 >. cjc 3 £ 00 "3 ;;: .E* - "O Q ' ^ ^o •> ^ II C "S .H

^ y o " °" J^ u 2 ^C O u . " "D.M h O E

£ - y

>; 3 i f 5

•^ "^ 1 -2 ^ 2' 'Bb >- •— ? "E ,— E

E 3 m ^ 2 'C -E

oj E ". 1 ' 'g "I 'I£ c "c.'^ 8 ^ E E

Accounting Review, October 2004


(t-statistic = 41.68), / = 0.515 for factor'''"''"-'"'-""' (t-statistic = 29.88), and / = 0.594 forFactor-'''"'""'""''"' {t-.statistic = 42.62). Turning to the results for the market-based attributes,each factor has a significant positive coefficient when added individually to the three-factorpricing regression (t-statistics range from 9.26 to 14.38).

Panel C of Table 10 shows that including the innate common factors reduces the load-ings on the attribute factors. With the exception of Factor^''"''""'''"-''' the loadings on theaccounting-based attribute factors remain reliably positive, though their magnitudes arebetween 13 percent and 57 percent of their values in Panel B. In contrast, none of theloadings on the market-based attribute factors has a positive sign when the innate commonfactors are also included in the regression.

Summary of Results Using Alternative Measures of the Cost of EquityThe results in Tables 9 and 10 are broadly consistent with the results reported in Section

V. Holding constant the cross-sectional estimation procedure, MPEG-based proxies for thecost of equity capital yield similar results to those we document in Tables 5 and 6. Thepattern of results is similar, with accounting-based attributes showing larger cost of equityeffects than market-based attributes, and among the accounting-based attributes, accrualquality has the largest effect. Results based on time-series asset pricing regressions showthat, unconditional on innate determinants, all earnings attributes have significant positiveloadings, consistent with firms with poor values of these attributes having higher costs ofequity. When the innate factors are included, we continue to find positive loadings on mostof the accounting-based attributes (except for a zero loading on persistence), but we findno evidence of risk premia associated with poor values of the market-based attributes.

VII. SUMMARY AND CONCLUSIONSDrawing on prior research, we identify seven earnings attributes—accrual quality, per-

sistence, predictability, smoothness, value relevance, timeliness, and conservati.sm—thathave been posited, individually or sometimes in pairs, to be desirable or advantageous.However, the benefits of the attributes are typically specific to the setting considered, so itis not possible to assess whether the earnings attributes are statistically and economicallydistinct, and whether one or a few of the attributes dominate the others. We calibrate theseearnings attributes against the cost of equity capital, a summary indicator of investor re-source allocation decisions, to learn which attributes are viewed by investors as conferringthe greatest capital market advantage, as measured by a decreased cost of equity capital.

We generally find a .statistically reliable association between each earnings attributeconsidered individually and our proxy for the cost of equity; the exceptions (or least con-sistent relations) are found for predictability and conservatism. Conditional associations,which include all seven earnings attributes plus known risk factors and factors believed todetermine the innate portions of the earnings attributes, show that the accounting-basedearnings attributes explain more of the cross-sectional variation in cost of equity estimatesthan do the market-based attributes (value relevance, timeliness, and conservatism). Con-sidered individually, accrual quality, earnings persistence, and smoothness have strong ef-fects on the cost of equity, as does the market-based attribute value relevance. On the whole,the weight of the evidence suggests that, among the seven attributes we consider, accrualquality is the dominant attribute in terms of cost of equity effects. The findings are robustto estimation procedure (cross-sectional versus time-series tests). Because the validity ofour results depends on the reliability of the cost of equity capital estimates we use, we



consider several proxies for the cost of equity (estimates based on price targets and fore-casted dividends; estimates based on PEG ratios; and estimates based on realized returns).Results are robust across these alternative proxies.

Evidence on the individual and conditional cost-of-equity effects of earnings attributeshas practical value for investors, researchers, standard setters, and managers. With respectto investors, our results suggest that attributes of earnings (especially accrual quality) factormeaningfully into discount rates; such information could be used to improve both simpletrading strategies {as discussed by Gebhardt et al. 2001) and the identification of comparablefirms. With respect to researchers (who frequently use earnings attributes to compare earn-ings numbers or, more generally, to compare the reporting systems that produce earningsnumbers), our results suggest that a focus on accounting-based attributes (rather than onmarket-based attributes) would allow for more sharply delineated comparisons in settingswhere the consideration of earnings numbers or reporting systems is linked to investors'resource allocation decisions. With respect to accounting standard setters (who use decisionusefulness as the benchmark to calibrate their decisions), our results suggest that they neednot be as concerned as previous research might suggest about the apparent declining valuerelevance of earnings (Erancis and Schipper 1999; Collins et al. 1997) or about the increas-ing conservatism of earnings (Givoly and Hayn 2000) since neither of these two attributesfeatures as prominently into the cost of equity capital as do accounting-based attributes.Finally, with respect to firm management, our results showing a link between the cost ofequity and earnings properties (both total and discretionary) supports recent survey evidencedocumenting that top management believes that the quantity and quality of informationinfluence their cost of capital (Graham et al. 2003).

Our results suggest the possibility that changes in certain attributes of earnings, suchas accrual quality, persistence or smoothness, affect the equity cost of capital. We expectthat discretionary components of earnings attributes arc likely to be quicker to change thaninnate components; the degree to which either component will change over time is a func-tion of both management's incentives (e.g., to increase or decrease accrual quality) andopportunities for change (e.g., by adding or dropping a line of business). Testing for costof equity effects of changes in earnings attributes would require an over-time analysis. Ouruse of firm-specific measures of the attributes makes it difficult to perform over-time testsbecause we require relatively long periods to measure the attributes for a given firm. Anavenue for future research would be to examine whether over-time changes in non-firm-specific measures of earnings attributes (for example, Barth and Landsman's [2003] cross-sectionally calculated, industry-adjusted measure of value relevance) are reliably associatedwith over-time changes in the cost of equity.- '

REFERENCESAdmati, A. 1985. A noisy rational expectations equilibrium for multi-asset securities markets. Econ-

ometrica 53: 629-658.AH. A., and P. Zarowin. 1992. The role of earnings levels in annual earnings-returns studies. Journal

of Accounting Research 30: 286-296.

^ FranL-is et al. (200.^) provide some evidence on this issue for their measures of accrual qualiiy. They find reliablypositive associations between changes in their cost of capital proxies and changes in accrual quality over 1970-2001.



American Institute of Certified Public Accountants (AlCPA). 1994. Comprehensive Report of theSpecial Committee on Financial Reporting. New York, NY: AICPA.

Amihud. Y., and H. Mendelson. 1986. Asset pricing and the bid-ask spread. Journal of FinancialEconomics 17: 223-249.

Association for investment Management and Research (AIMR). 1993. Financial Reportinf^ in the1990s and Beyond. Charlottesville, VA: AIMR.

Baginski, S.. K. Lorek, G. L. Willinger, and B. Branson. 1999. The relationship between economiccharacteristics and alternative annual earnings persistence measures. The Accounting Review 74:105-120.

Ball. R., S. P. Kothari, and A. Robin. 2000. The effect of international institutional factors on prop-erties of accounting earnings. Journal of Accounting and Economics 29: 1-51.

Barone, G. 2003. Perceptions of earnings quality and their association with the cost of equity capital.Working paper. The University of Texas at Austin.

Barth, M., W. Beaver, and W. Landsman. 2001. The relevance of the value relevance literature forfinancial accounting standard setting: Another view. Journal of Accounting and Economics 31:77-104.

, and W. Landsman. 2003. Cost of capital and the quality of financial statement information.Working paper. Stanford University and The University of North Carolina at Chapel Hill.

Bartov. E.. D. Givoly, and C. Hayn. 2002. The rewards to meeting or beating earnings expectations.Journal of Accounting and Economics 33: 173-204.

Basu, S. 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal ofAccounting and Economics 24: 3-37.

Bhattacharya, U., H. Daouk. and M. Welker. 2(X)3. The world pricing of earnings opacity. The Ac-counting Review 78: 641-673.

Biddle, G., G. Seow, and A. Siegel. 1995. Relative versus incremental information content. Contem-porary Accounting Research 12: 1-23.

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

, and M. Plumlee. 2002. A re-examination of disclosure level and the expected cost of equitycapital. Journal of Accounting Research 40:

and . 2005. Assessing the construct validity of alternative proxies for expected costof equity capital. The Accounting Review 80 (1): (forthcoming).

Bradshaw, M. 2002. The use of target prices to justify sell-side analysts' stock recommendations.Accounting Horizons 16: 27-41.

Brav, A., R. Lehavy. and R. Michaely. 2004. Expected returns and asset pricing. Working paper. DukeUniversity, University of Michigan, and Cornell University.

Brennan, M., T Chordia, and A. Subrahmanyam. 1998. Alternative factor specifications, securitycharacteristics and the cross-section of expected stock returns. Journal of Financial Economics49: 345-373.

Bushman, R., Q. Chen, E. Engei, and A. Smith. 2004. Financial accounting information, organizationalcomplexity and corporate governance systems. Journal of Accounting and Economics 37 (2):139-201.

Carhart, M. 1997. On persistence in mutual fund perfonnance. Journal of Finance 52: 57-82.Chaney, P., and C. Lewis. 1995. Earnings management and firm valuation under asymmetric infor-

mation. Journal of Corporate Finance: Contracting. Governance and Organization 1: 319-345.Cohen, D. 2003. Quality of financial reporting choice: Determinants and economic consequences.

Working paper. Northwestern University.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 1. Dichev. 2002. The quality of accruals and earnings: The role of accrual estimation

errors. The Accounting Review 11 (Supplement): 35-59.Demski. J. 1998. Performance measure manipulation. Contemporary Accounting Research 15: 2 6 1 -

285.



Diamond, D.. and R. Verrecchia. 1991. Disclosure, liquidity, and the cost of capital. Journal of Finance46: 1325-1359.

Easley, D., S. Hvidkjaer, and M. O'Hara, 2(X)2. Is information risk a determinant of asset returns?Journal of Finance 57: 2185-2221.

, and M. O'Hara. 2(K)3. Information and the cost of capital. Journal of Finance 59 (4): 1553-1583.

Easton, P., and S. Monahan. 2004. The empirical relation between realized returns and accountingbased measures of expected returns. Working paper. University of Notre Dame and INSEAD.

. 2004. PE Ratios, PEG ratios, and e.stimating the implied expected rate of return on equitycapital. The Accounting Review 79: 73-95.

Elton. E. 1999. Expected return, realized return, and asset pricing tests. Journal of Finance 54: 1199-1220.

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

. and K. French. 1993. Common risk factors in the returns on stocks and bonds. Journal ofFinancial Economics 33: 3-56.

, and '. 1997. Industry costs of equity. Journal of Financial Economics 43: 153-193.and . 2002. The equity premium. Journal of Finance 57: 637-659.

Finger, C. 1994. The ability of earnings to predict future earnings and cash flow. Journal of AccountingResearch 32: 210-223.

Francis, J., and D. Philbrick. 1993. Analysts' decisions as products of a multi-task environment.Journal of Accounting Research 31; 216-230.

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

and L. Vincent. 2003. The relative and incremental explanatory power of earningsand alternative (to earnings) performance measures for returns. Contemporary Accounting Re-search 20: 121-164.-, R. LaFond, P. Olsson, and K. Schipper. 2005. The market pricing of accruals quality. Journal

of Accounting and Economics 39 (2): (forthcoming).

Gcbhardt. W., C. Lee. and B. Swaminathan. 2(X)I. Toward an implied cost of capital. Journal ofAccounting Research 39: 135-176.

Givoly, D.. and C. Hayn. 2000. The changing time-series properties of earnings, cash fiows andaccruals: Has financial reporting become more conservative? Journal of Accounting and Eco-nomics 29: 287-320.

Graham, J., C. Harvey, and S. Rajgopal. 2003. Financial reporting policies: Evidence from the field.Working paper. Duke University and University of Washington.

Guay, W.. S. P. Kothari, and S. Shu. 2003. Properties of the implied cost of capital using analysts'forecasts. Working paper. University of Pennsylvania, Massachusetts Institute of Technology,and Boston College.

Harri.s, T., E. Huh, and P. Fairfield. 20(K). Gauging profitability on the road to valuation. In StrategyReport. Global Valuation and Accounting. New York, NY: Morgan Stanley Dean Witter.

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

Hunt, A., S. Moyer, and T. Shevlin. 2000. Earnings volatility, earnings management and equity value.Working paper. University of Washington.

Ittner, C . and D. Larcker. 2001. Assessing empirical research in managerial accounting: A value-based management perspective. Journal of Accounting and Economics 32: 349-410.

Joos, P., and M. Lang. 1994. The effects of accounting diversity: Evidence from the European Union.Journal of Accounting Research (Supplement) 32: 141-175.

Lee, C. 1999. Accounting-based valuation: Impact on business practices and re.search. AccountingHorizons 13: 413-425.

Leuz, C . D. Nanda. and P. Wysocki. 2003. Earnings management and investor protection: An inter-national comparison. Journal of Financial Economics 69: 505-527.


1010 Francis, iMFond. Olsson. and Schipper

, and R. Verrecchia. 2004. Finns' capital allocation choices, information quality, and the costof capital. Working paper. University of Pennsylvania.

Lev, B. 1983. Some economic determinants of the time-series properties of earnings. Journal ofAccounting and Economics 5: 31-38.

, and R. Thiagarajan. 1993. Fundamental information analysis. Journal of Accounting Research31: 190-215.-. and P. Zarowin. 1999. The boundaries of financial reporting and how to extend them. Journalof Accounting Research 37: 353-385.

Lipe. R. 1990- The relation between stock returns and accounting earnings given alternative infor-mation. The Accounting Review 65: 49-71.

Liu, J., D. Nissim, and J. Thomas. 2002. Equity valuation using multiples. Journal of AccountingResearch 40: 135-172.

Merton, R. 1987. A simple model of capital market equilibtium with incomplete information. Journalof Finance 42: 483-510.

Newey, W., and K. West. 1987. A simple, positive semi-definite, heteroskedasticity and autocorrelationconsistent covariance matiix. Economeirica 55: 703-708

O'Hara, M. 2003. Presidential address: Liquidity and price discovery. Journal of Finance 58: 1335-1354.

Ohlson, J., and B. Jtittner-Nauroth. 2004. Expected EPS and EPS growth as determinants of value.Working paper, New York University.

Penman, S. 2001. Financial Statement Analvsis and Security Valuation. New York. NY: McGraw-Hill/Irwin.

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

Pope, P., and M. Walker. 1999. International differences in the timeliness, conservatism and classifi-cation of earnings. Journal of Accounting Research 37 (Supplement): 53-99.

Revsine, L., D. Collins, and B. Johnson. 2002. Financial Reporting and Analysis. Second edition.Upper Saddle River. NJ: Prentice Hall.

Richardson, S. 2{M)3. Earnings surprises and short selling. Accounting Horizons 17 (Supplement): 4 9 -61.

Ronen, J.. and S. Sadan. 1981. Smoothing Income Numbers: Objectives. Means and implications.Reading. MA: Addison Wesley.

Thomas. J., and H. Zhang. 2002. Value-relevance properties of smoothed earnings. Working paper,Columbia University and University of Illinois at Chicago.

Watts, R. 2(X)3. Conservatism in accounting. Part 1: Explanations and implications. AccountingHorizons 17: 207-221.


costs of equity and earnings attributes€¦ · costs of equity and earnings attributes 969 equity...

Documents