barth, cram, nelson (2001)

THE ACCOUNTING REVtEWVol. 76. No. 1Januury 2001pp. 27-58

Accruals and the Prediction ofFuture Cash Flows

Mary E. BarthStanford UniversityDonald R Cram

California State University, FullertonKaren K. NelsonStanford University

ABSTRACT: Building on the Dechow et al. (1998) model of the accrual pro-cess, this study investigates the role of accruals in predicting future cashflows. The model shows that each accrual component reflects different infor-mation relating to future cash flows; aggregate earnings masks this informa-tion. As predicted, disaggregating accruals into major components—changein accounts receivable, change in accounts payable, change in inventory, de-preciation, amortization, and otheraccruals—significantly enhances predictiveability. Each accrual component, including depreciation and amortization, issignificant with the predicted sign in predicting future cash flows, incrementalto current cash flow. The cash flow and accrual components cf current earn-ings have substantially more predictive ability for future cash flows thanseveral lags of aggregate earnings. The inferences are robust to alternativespecifications, including controlling for operating cash oyole and industrymembership.

Key Words: Accruals, Cash flow, Earnings, Cash flow prediction.

Data Availability: The data used in this study are from the public sourcesidentified in the text.

The authors appreciate comments by Sudipta Basu, Bill Beaver, Gerald Feltham, S. P. Kothari, Stephen Ryan,two anonymous reviewers. Ken Gaver (the Associate Editor), and workshop participants at Ihe American Account-ing Association 1999 Annual Meeting, the Ninlh Annual Conference on Financial Economics and Accounting.Stanford Universily Accounting Summer Camp, University of British Columbia. California State University, Ful-lerton. Baruch College-CUNY, The George Washington University. Massachusetts Institute of Technology, andRutgers University. The support of the Massachusetts Institute of Technology Sloan School of Management, andIhe Stanford University Graduate School of Business Faculty Trust and Financial Research Initiative i.s gratefullyacknowledged.

Submitted November 1999Accepted August 2000

27

28 The Accounting Review, January 2001

I. INTRODUCTION

This study investigates the role of accruals in predicting future cash fiows. A firm'sability to generate cash flow affects the values of its securities. For this reason, theFinancial Accounting Standards Board (FASB) indicates that a primary objective of

financial reporting is to provide information to help investors, creditors, and others assessthe amount and timing of prospective cash fiows {FASB 1978. 1137-39). Moreover, theFASB asserts that information about earnings and its components is generally more pre-dictive of future cash flows than current cash flow (FASB 1978. 1144). Several prior studiestest the relative abilities of aggregate earnings and cash flow to predict future cash fiows.but do not examine how the components of earnings affect its ability to predict future cashfiows. We build on the model of Dechow et al. (1998) (hereafter DKW) to develop predic-tions about the role of accruals in predicting future cash flows. As predicted, we find thatdisaggregating earnings into cash fiow and the major components of accruals significantlyenhances earnings" predictive ability. Thus, this study extends our understanding of thetemporal relations among accruals, cash fiows, and earnings, and provides evidence con-sistent with the FASB's assertion that knowledge ofthe components of earnings is importantfor predicting future cash fiows.

Our extended analysis of the DKW model, which focuses on predicting cash flow nextperiod, reveals that the various accrual components of earnings capture different informationnot only about delayed cash flows related to past transactions, but also about expectedfuture cash fiows related to management's expected future operating and investing activity.Aggregate earnings, and thus aggregate accruals, masks this information by weighting thecomponents equally. Thus, we predict that disaggregating earnings into cash flow and thecomponents of accruals enhances earnings' predictive ability relative to aggregate earnings.A key insight from our extended analysis of the model is that current cash flow and themajor components of current period accruals reflect the same information about cash flownext period as do multiple lags of aggregate earnings.

The empirical tests focus on annual amounts and disaggregate accruals into six majorcomponents: change in accounts receivable, change in inventory, change in accounts pay-able, depreciation, amortization, and other accruals.' Based on the weights on the accrualcomponents implied by the model, we predict that each accrual component has a differentrelation with future cash flows, and that increases in accounts receivable and inventory anddecreases in accounts payable are associated with higher future cash flows. We also predictthat depreciation of fixed assets and amortization of intangible assets are associated withhigher future cash flows. We find that the relation between cash flow next year and currentcash flow and each component of accruals is significant and has a sign consistent withpredictions. Moreover, the relations differ significantly from each other, in absolute value.Thus, as predicted, we find that aggregate earnings masks information relevant for pre-dicting future cash flows. We also find that long-term accruals, i.e., depreciation and am-ortization, have significant predictive ability for future cash flows. This evidence is incon-sistent with suggestions in the financial press that depreciation and amortization do notpredict future firm performance (e.g.. MacDonald 1999a, 1999b). calling into questionanalysts' recent focus on "cash earnings," which excludes these accruals.

The model leads to the prediction that disaggregated current earnings has the samepredictive ability for next period cash flow as do current and two lags of aggregate earnings.

The term "accruals" refers lo adjustments from operating cash flow lo earnings, e.g., change in accounts receiv-able, rather than adjustments from cash to nel worth, e.g., accounts receivable, which we refer to as balancesheet accruals.

Barth, Cram, and Nelson—Accruals and the Prediction of Future Cash Flows 29

However, our empirical analysis reveals that disaggregated current eamings has significantlymore predictive ability than current and up to six years of lagged aggregate eamings. Thisevidence suggests that the modeFs simplifying assumptions about the accrual process un-derstate the predictive ability of the accrual components relative to aggregate eamings.

We investigate whether disaggregated eamings' significantly greater predictive abilityis attributable to disaggregating cash flow and aggregate accruals or to disaggregating ac-cruals. The findings reveal that disaggregating eamings into cash fiow and aggregateaccruals significantly increases predictive ability relative to aggregate eamings. but thatdisaggregating accruals into its major components further significantly increases predictiveability. Our inferences regarding the superior predictive ability of disaggregated eamingsare robust to a variety of sensitivity checks, including predicting cash flows several yearsin the future, controlling for firms' operating cash cycles and industry membership, andusing share prices, retums, and discounted future cash flows as proxies for expected futurecash flows.

The remainder of the paper is organized as follows. Section II relates this study toextant research. Section III develops the model and empirical predictions. Section IV spec-ifies the estimation equations. Section V describes the sample and presents the primaryfindings. Section VI presents results from additional analyses. Section VII summarizes andconcludes.

II. RELATION TO PRIOR RESEARCHThe role of accruals in predicting future cash flows is a fundamental question under-

lying financial reporting. Prior research directly addressing this question typically uses smallsamples limited by long time-series requirements; the results of this research are mixed.Bowen et al. (1986) does not find that aggregate earnings provides better predictions offuture cash flows than past cash flow. In contrast. Greenberg et al. (1986) concludes thataggregate eamings has more predictive ability than cash flow, and Lorek and Willinger(1996). focusing on quarterly rather than annual amounts, finds that accruals have predictiveability incremental to cash flow. Finger (1994) finds that cash flow is marginally superiorto aggregate eamings for short prediction horizons, but eamings and cash flow performequally well for longer horizons. Using a larger sample, Burgstahler et al. (1998) finds thatcash flow has more predictive ability than aggregate eamings.-

Our analysis of the model reveals that neither current aggregate eamings nor currentcash flow is an unbiased predictor of future cash flows, and that the bias in each is afunction of accruals. Thus, one possible explanation for the mixed results of prior researchis that the relative magnitudes of the biases depend on sample composition, which is man-ifest in the small sample studies. Similar to Burgstahler et al. (1998). for our large sampleof firms we find that current cash flow has more predictive ability for future cash flowsthan current aggregate eamings. However, we leave to future research clear resolution ofthe mixed findings of prior research. Our contribution to this research is showing thatlimiting the inquiry to aggregate eamings masks the ability of accrual components to predictfuture cash flows incremental to current cash flow.

DKW models cash flow and the accrual process related to accounts receivable, accountspayable, and inventory to derive the prediction that current eamings is the best predictorof future cash flows. DKW reports that firm-specific variation in cash flow forecast errorsbased on aggregate eamings is significantly lower than that based on cash flow. DKW alsoreports that in firm-specific regressions of future cash flows on aggregate current eamings

~ Several of these studies also predict cash flow more than one period ahead; we do so in Section VI.


and cash flow, both have incremental explanatory power. DKW does not explore the model'simplications for the predictive ability of eamings components, including the componentsof accruals.

We extend the analysis of the DKW model to show that eamings' superiority for pre-dicting future cash flows stems from disaggregating eamings into cash flow and the com-ponents of accruals. The differences across accrual components are not evident from DKW'sempirical analysis because DKW implicitly permits the aggregate eamings coefficient todiffer firm by firm. Our analysis of the model and empirical results also extend DKW byshowing that several past aggregate eamings have explanatory power for predicting futurecash flows, incremental to aggregate current eamings, and that disaggregated current eam-ings has significantly more predictive ability than several lags of aggregate eamings.

DKW's analysis also indicates that the predictive ability of aggregate eamings relativeto cash flow varies with firms' operating cash cycles. We show that even after partitioningon operating cash cycle, following DKW, aggregate eamings masks significant informationrelevant to predicting future cash flows. Finally, we show that long-term accruals aid inpredicting future cash flows; DKW focuses on working capital accmals.

This study also relates to research comparing the predictive abilities of eamings andcash flow using share prices as an implicit or explicit proxy for expected future cash flows.For example. Ball and Brown (1968), Beaver and Dukes (1972), and Dechow (1994) findthat returns are more highly associated with aggregate earnings than with cash flow. Severalother studies (e.g., Rayburn 1986; Wilson 1986, 1987; Bowen et al. 1987; Ali 1994; Chenget al. 1996; Pfeiffer et al. 1998) document that aggregate eamings and cash flow are incre-mentally informative for retums. Some prior research also finds that components of eam-ings, including accruals and accrual components, have different pricing multiples, as pre-dicted by differences in the components' persistence (e.g., Lipe 1986; Barth et al. 1990;Barth et al. 1992; Barth et al. I9Q9, 2000). We contribute to this research by showing thatcash flow and the major accrual components of eamings have different multiples whenpredicting future cash flows, as predicted by a model of the accmal process.

We focus on the implications of cash flow and the accrual components of eamings forfirms' expected future cash flows rather than for firms' share prices for two primary reasons.First, cash flow prediction is fundamental to assessing fimi value as reflected in share prices.Thus, cash flow is a primitive valuation construct. Our use of realized future cash flows asa proxy for expected future cash flows assumes rational expectations, as does prior ac-counting research in a variety of contexts (e.g., McNichols and Wilson 1988; Penman andSougiannis 1998; Aboody et al. 1999). Second, prior research provides evidence that shareprices fail to reflect accurately the differential persistence of accruals and cash flow (e.g.,Sloan 1996; Barth and Hutton 2000; DeFond and Park 2000; Xie 2000). This evidencecalls into question the use of share prices as a proxy for expected future cash flows whenexamining the predictive ability of accruals and cash flow. Nonetheless, to facilitate com-parison with prior research, we estimate price- and retums-based specifications with nochange in inferences regarding the predictive ability of disaggregated eamings.

Til. MODELDKW models operating cash flows and the accrual process to generate predictions for

the relative abilities of eamings and cash flow to predict future cash flows. Because ourobjective is similar to DKW, we expand the analysis of the DKW model to obtain ourpredictions.


Model AssumptionsThe model assumes earnings, EARN, equals a constant proportion of sales, S, and sales

follows a random walk:

EARN, = TTS, and S, = S, , + e,, (1)

where 0 < TT < 1 is the profit rate, t denotes time period, and e, is a mean zero randomshock.^ The model also includes three balance sheet accruals: accounts receivable, AR;accounts payable, AP; and inventory, INV. AR is modeled as a constant proportion, a, ofsales and AP is modeled as a constant proportion, p, of purchases or production, P,:'*

AR, = aS, and AP, = pP, /2j

AAR, = ae, and AAP, = pAP,,

where A denotes a one-period change, e.g., AAR, = AR, - AR,_i. Assuming 0 < a < 1allows a portion of sales to be received in cash in the next period. Similarly, assuming0 < p < 1 allows a portion of purchases to be paid in cash in the next period.

Following Bernard and Stober (1989), DKW models an inventory policy with a targetinventory level proportional to cost of sales, which is not fully achieved in the period ofthe sales shock. The firm adjusts inventory to the target level over two periods. Two modelparameters, 7, and 7^, reflect the inventory policy, where 0 < 71, 7^ < 1. 71 is the fractionof cost of goods sold, (1 - IT) S,, that is the target inventory level and 7, is the fraction ofthe current sales shock, e,, not included in inventory in the current period because it isdeferred to the next period.^ Thus, current period purchases equals current period cost ofsales, plus the initial inventory adjustment for the current sales shock, plus the laggedadjustment for the prior sales shock:

P, = (I - TT)S, + 7,(1 - 'n-)E, - 7,72(1 - 'ir)Ae, ^3^

= (1 - 17)5, + 7,(1 - TT)[(I - 72)e. + 72e.-,]-

Equation (3) indicates that if 7, = 0, i.e., there is no inventory, then purchases equalscost of goods sold. If 7^ = 0, i.e., management contemporaneously fully adjusts inventoryin response to a sales shock, then purchases equals cost of goods sold plus the total inven-tory effect of the sales shock, 7,(1 - -iT)e,. Because purchases equals cost of goods soldplus the change in inventory, equation (3) can be used to obtain expressions for the currentperiod change in inventory, AINV,, and the expected inventory change next period,EJAINV,,,]:

We restrict TT to be positive because, even though a firm can have negative eamings in a particular year, -IT is anintertemporally constanl firm-specific parameter. Thus, negative T- would assume a firm always has negativeeamings, which is not reasonable.Henceforth, references to purchases should be interpreted as purchases or production.Although the inventory assumptions might not mirror precisely the policies of real firms, they capture the notionthat nol all accruals reverse in a single period and that, as explained below, accruals reflect more informatiotithan simply the one-period delayed payments or receipts associated with past purchases or sales. In particular.accruals can reflect information related to management's expected future activity (see also Bernard and Noel1991).


AINV. = 7,(1 - Tr)[(\ - 7,)e, + 72^,-,] (4)

E,[AINV,,,] = 7i72n - '^)ep

where E,[-] is the expectations operator conditioned on time t information. With a delayedinventory adjustment, i.e., 7,, 7, > 0, E,[AINV,+ i] equals 0 only in the rare case when S,= S,_|. i.e., e, = 0.

Aggregate Eamings as a Predictor of Future Cash FlowNext period cash flow. CF,+|, equals cash inflows from sales, adjusted for uncollected

amounts reflected in the change in accounts receivable, minus outflows from purchases,adjusted for unpaid amounts reflected in the change in accounts payable. That is:

CF,., = (S.,, - AAR,,,) - (P,.. - AAP,,,). (5)

Equations (1) through (3) show that equation (5) can be expressed in terms of S,+, andthree sales shocks, E,+ ,, E,, and E,_,. In particular, DKW shows that:

CF,,, = T7S,,, - [a + (1 - 7T)7, - P(l - 'iT)]e,,, (6)

+ 7,(1 - iT)[p + 72(1 ~

Because earnings equals cash flow plus accruals, i.e., EARN,+ , = CF,,, + AAR,+, +AINV,+, - AAP,^, in the model, and EARN,,, = irS,+ ,, the E terms in equation (6) areaccruals.'^ As DKW explains, the second term in equation (6) reflects the permanent changein working capital accruals (accounts receivable plus inventory minus accounts payable)that results from the current sales shock. This term equals the change in working capitalaccruals only if the entire target inventory adjustment occurs and is paid for in periodt + 1. DKW refers to the multiple on E^, , in the second term as the firm's operating cashcycle, 8, expressed as a fraction of a year. The third and fourth terms in equation (6) reflectthe one- and two-year effects on cash flow of cash payments related to the new levels ofcost of goods sold and inventory arising from the sales shocks.

DKW ignores the second line of equation (6) to obtain current eamings as the bestpredictor of next period cash flow, and all future cash flows.^ However, the second line ofequation (6) does not equal 0 in expectation at time t. Specifically, E J A E , , , ] ^ - e . andEJAE, ] = e, - E,_,, where e, and E,_, are the time t and t - 1 realizations of the randomvariable e. which only equal 0 by chance. As we show below, including these terms revealsthat expected next period cash flow does not equal current earnings, i.e., EARN, is not anunbiased predictor of CF,+i, and, more importantly, facilitates insights into the incrementalrole of accruals in predicting future cash flows.

Equation (6) can be used to express expected next period cash flow as a function ofcurrent and two lags of earnings. In particular, from equation (6):

Because P,^, = (1 - ir) S,,, + AINV,,,. equation (6) can be expressed as CF,,, = 77 S,,i - AAR,,, -+ AAP,^.,. which is earnings minus accruals.From equation (6). ignoring the second line, E,[CF,,|| = E,(7rS,,, - 5 £,,,1 = EJ-rrS,] = EARN,, which alsosuggests that -nS, is an optimal forecast of next period cash flow. The second line of equation (6) does not affectthe predicted sign of the correlation between cash flow and eamings, or accruals, which is a primary focus ofDKW.


From equation (1), e, = IT ' (EARN, - EARN,.,) and e, , = tr ' (EARN,_, - EARN,.,).Thus, current and two lags of earnings provide information about the sales shocks relevantto expected next period cash flow." Substituting EARN variables for the e terms in equation(7) and rearranging yields equation (8);

E,[CF,^,] - (I - 7,(1 - 7r)TT-'[p + 7,(1 - p) - P7,])EARN,

+ 7i(l - TT)-;! '[P + 7^(1 - P) - 2P7,]EARN,_, (8)

Equation (8) shows that expected next period cash flow equals current earnings, ad-justed for the one- and two-year effects of inventory changes and associated payments. Forexample, if the two prior years' sales changes, i.e., e, and e,_,, are positive, then EARN,overstates expected cash flow in period t + 1 because EARN, omits the future cash floweffects of payments related to delayed inventory increases. In this case, cash flow in periodt + 1 will be less than earnings in period t because of payments related to (1) the periodt + 1 inventory increase arising from the period t sales increase, (2) the period t accountspayable for the period t inventory increase arising from the period t sales increase, and (3)the period t accounts payable for the period t inventory increase arising from the periodt — 1 sales increase.

Cash Flow and Components of Accruals as Predictors of Future Cash FlowThe analysis thus far focuses on using aggregate earnings to predict next period cash

flow. However, our objective is to understand the relation between earnings and its com-ponents, and future cash flows. We next use the model to derive this relation. Key tounderstanding this relation is the observation that equation (5) ean also be used to expressexpected next period cash flow in terms of the components of current earnings.

Specifically, using equation (5) to obtain expressions for CF, and E,[CF,^,| shows thatnext period cash flow is expected to differ from current period cash flow because the firmcollects the amount of the change in receivables, pays the amount of the change in accountspayable, and pays an amount associated with the change in expected purchases next period.Thus, CF, is not an unbiased predictor of CF,+ ,:

E,[CF,,,] = CF, + AAR, - AAP, - (1 - P)(E,[P,.,| - P,) ^^^

= CF, + AAR, - AAP, - (I - P)(EJAINV,.,1 - AINV,).

The second line of equation (9) follows from the first by noting that purchases equalscost of goods sold plus the change in inventory and recalling that expected cost of goodssold in period t + 1, E,[(l - ir)S,,,], equals (1 - T7)S,. TO state equation (9) in terms ofthe components of current earnings, recall from equations (2) and (4) that the sales shock,e,, affects current period change in receivables and expected change in inventory next period.Thus, the expected change in inventory can be stated in terms of current change inreceivables.

Because EARN, = irS,, equation (7) can be equivaJently stated in terms of uurTent and two lags of sales.

34 The Accounting Review, January 200!

Expressing E,[AINV,+ |] in terms of AAR, and collecting terms yields an expression ofexpected next period cash flow in terms of the components of current eamings:

E.[CF,,,I = CF, + (1 - (1 - P)7,72(l - '^)« ')^AR. + (1 - (i)AlNV, - AAP, (10)

Thus, under the assumptions of the model, expected cash flow can be expressed as afunction of either (1) current and two lags of aggregate eamings, as in equation (8). or (2)current eamings disaggregated into cash flow and the components of accruals, as in equation(10). In other words, the model predicts that equations (8) and (10) have equal predictiveability.

In equation (10), the first part of the multiple on AAR, i.e., 1, reflects the expectedcollection next period of the current period change in receivables. The second part of themultiple, i.e., -{I - P)*yi*Y2(' ~ iT)a"', reflects the expected payment next period of theexpected change in inventory, because the expected change in inventory next period, likethe change in accounts receivable, depends on the current period sales shock. The multipleon AINV reflects the payment deferred to next period of the current period change ininventory. The multiple on AAP reflects the expected change in cash next period associatedwith the current change in accounts payable. Equation (10) shows that accruals reflectinformation about expected future cash fiows relating to management's expected futurepurchasing activity, as well as relating to collections and payments associated with currentperiod transactions, i.e.. collecting accounts receivable and paying accounts payable. Thus,the predictive ability of accruals for future cash flows is not limited to the "mechanical"delayed cash flow effects of past transactions.^

PredictionsThe model provides the basis for the predictions we test in Section V. First, equation

(8) leads us to predict that current and two lags of aggregate earnings are significant inpredicting next period cash flow. Equation (8) includes three eamings variables because theassumed target inventory policy is a two-period phenomenon. However, asset investmentpolicies and related cash payment policies likely differ across short-term assets, such asinventory, and long-term assets, such as property, plant, and equipment. If the policy werelonger lived, as one might expect for long-term assets, then equation (8) would includeadditional lags of eamings. Because firms invest in long-term assets, we expect that eamingslags greater than two also are significant in predicting future cash flows.'"

Our predictions assume that managers choose accounting policies to portray accurately the firm's economicsituation. To the extent managers choose accounting policies opportunistically, we expect eamings lo be arelatively poor predictor of future cash flows, thereby biasing against finding evidence consistent with ourpredictions. Also, it is possible that components of cash flow could enhance the predictive ability of aggregatecash flow for future cash flows, just as components of earnings enhance the predictive ability of aggregateeamings. However, such components are available only for the relatively few lirms using the direct methodunder Statement of Financial Accounting Standards No. 95. Moreover, because accruals inherently incorporateinformation about future cash flows, whereas past cash flow does not, it would not be surprising to find thataccruals enhance the predictive ability of disaggregated cash flow.This can be seen by noting that equation (8) includes one more EARN term than the sales shock affecting cashflow next period. In lhe modeled case of a two-period investment policy, the sales shock from period t - 1affects period l + 1 cash flow. Thus. EARN,.. is included in equation (8). In the case of a six-period investmentpolicy, the sales shock from period t - 5 affects period t -t- I cash flow. Thus, EARN,_^ would be included inequation (8),

An equation analogous to equation (8) using current and lagged cash flows as predictor variables wouldinclude an infinite number of lags of cash flow. Even if the only accrual relates to accounts receivable, if theseries of changes in cash flows is invertible, i.e., a < IT/2 , then expected cash flow equals an infinite series ofpast cash flows. Relatedly, as an additional analysis in Section VI, we investigate the explanatory power ofcurrent and lagged aggregate and disaggregated eamings for cash flows up to four years into the future.

Barth. Cram, and Nelson—Accruals and the Prediction of Future Cash Flows 35

Second, equation (10) leads us to predict that current earnings, disaggregated into cashflow and the major components of accruals, has the same predictive ability for future cashflows as current and two lags of aggregate eamings. This prediction derives from the dif-ferent prediction equation weights on cash flow and the accrual components. Although theweights depend on the magnitudes of model parameters, the weights would be equal, inabsolute value, only if there were no accounts payable, i.e., p ^ 0, and either there wereno inventory, i.e., 7, = 0, or there were no delayed inventory adjustment, i.e., 72 = 0- Thisprediction is equivalent to the prediction that the components of accmals add to cash flowin predicting next period cash flow.

Third, equation (10) permits us to predict the sign of the weights on each eamingscomponent. Specifically, we predict that the weights on CF and AINV are positive andthe weight on AAP is negative. The sign of the weight on AAR depends on the magnitudesof model parameters p, 7,, 7 * and -rr in equation (10). Thus, we calculate firm-specificestimates of these parameters, as in DKW. and then calculate the expression in equation(10) to predict the sign of the weight on AAR. Because untabulated statistics reveal thatthe expression is positive for all sample firms with data sufficient for its calculation, wepredict that the weight on AAR is positive.

Although we predict the signs of the weights on the eamings components, we do notpredict their magnitudes because, although the model captures the essence of accmals. itis based on several simplifying assumptions. For example, the model assumes only threecurrent accruals with relatively simple time-series processes. Also, the model does notinclude long-term accruals; the empirical tests include them because they are componentsof eamings for sample firms. In developing expectations for the signs of the weights onlong-term accruals, specifically depreciation and amortization, we note that the model fo-cuses on predicting future operating cash flows, which do not include expenditures relatedto long-term investments. Such investments might include the purchase of property, plant,and equipment or intangible assets. Presumably, a firm makes such investments becausethey are expected to generate, over multiple future periods, higher cash flows than wouldbe generated from the firm's previously existing asset base. Depreciation and amortizationare intended to match the costs ofthe investments to their benefits. If matching is achievedand the investment earns a positive retum. then the cash inflows associated with the in-vestment will exceed its depreciation or amortization in each period, even if the rate ofreturn is lower than the firm's cost of capital. Thus, consistent with Feltham and Ohlson(1996), we expect that future operating cash flows are positively related to depreciation andto amortization, and predict that the weights on depreciation and amortization are positive.

IV, ESTIMATION EQUATIONSThe first set of tests relates to the predictive ability of current and past aggregate

earnings for future cash flows, and is based on the following equation:

k

CF,., , - 4) + S 4),-. EARN,,_, + u , , (11)"I, l-T • " | ,P

T=0

where i and t denote firm and year, and k ranges from 0 to 6. Based on equation (8), wepredict that 4>|, 4)^_|, and (|), . are significantly different from 0." For the reasons set forth

Although intuition suggests the eamings coefficients should be positive and decline with longer lags, this is notpredictable from equation (8). Thus, we do not predict the signs or relative magnitudes of the coefficientestimates.


in Section III, we estimate equation (11) using up lo six lags of EARN and expect that atleast some c}),_j, are significantly different from 0 for k > 2. We use up to six lags ofeamings because this results in the same number of explanatory variables in equation (H)as in the accmal components equation (12), ensuring that any difference in explanatorypower is not attributable solely to the number of explanatory variables.

The second set of tests disaggregates eamings into its major components:

CF,,,., = ^ + <f>,.,CF,, + CI

where DEPR is depreciation expense, AMORT is amortization expense, and OTHER is theaggregate of other accmals, i.e., OTHER = EARN - (CE + AAR + AINV - AAP- DEPR - AMORT).'"

As explained in Section III, we predict that equation (12) has the same predictive abilityas equation (11) estimated using current and two lags of EARN. We also predict that thecoefficients on the accmal components in equation (12) differ from each other and fromthat on cash flow. That is, the components of accmals add to cash flow in predicting futurecash flows. We also predict that <t>(.Fi 4>AR '1>I. <I>D' ^^^ ^ AM ^ ^ positive, and 4)^? is negative.We have no prediction for 4>o-

Tests of predictions comparing the explanatory power of equations ( I I ) and (12) arebased on a Vuong (1989) Z-statistic. Tests of predictions relating to differences in coeffi-cients in equation (12) are based on E-tests of coefficient equality. Results associated withother tests of coefficient constraints provide additional insights Into the predictive abilityof eamings components and facilitate comparing the predictive abilities of cash flow andaggregate earnings, as in prior research. In particular, constraining the coefficients on cashflow and the accmal components to be equal with signs implicit in EARN is equivalent toincluding only aggregate eamings in equation (12). Constraining the coefficients on theaccmal components to equal 0 is equivalent to including only cash flow in equation (12).Similarly, constraining the coefficient on cash flow to equal 0 is equivalent to includingonly the accmal components in equation (12).

V. DATA AND EMPIRICAL RESULTSData and Descriptive Statistics

Data are from the 1997 Compustat annual industrial and research files. The samplespans 1987-1996 because the analyses require at least one year of future cash flows anduse cash from operations reported under Statement of Financial Accounting Standards No.95 (SEAS No. 95).'* EARN is income before extraordinary items and discontinued opera-tions, and CE is net cash flow from operating activities, adjusted for the accrual portion of

The accruals disaggregation follows the nuidel in Section III. An alternative disaggregation follows the lineitems typically reported in firms' income statements, i.e.. sales, cost of goods sold (COS), selling, general, andadministrative expenses (SGA), DBPR. AMORT, and OTHER. The two alternatives are closely related in thatchange in receivables typically relates to sales, change in inventory relates lo CGS, and change in accountspayable relates to CGS and SGA. Not surprisingly, our inferences are unaffected by using this alternativedisaggregation.SFAS No. 9? requires firms to present a statemenl of cash flows for fiscal years ending after July 15. 1988(FASB 1987). Our sample begins in 1987 because some firms early-adopted SFAS No. 95. We predict cashflows for 1988-1997. depending on the length of the prediction horizon and the number of lagged predictorsin the equation.


extraordinary items and discontinued operations. The accrual components are from thestatement of cash flows, when available, and calculated from balance sheet data otherwise.Following Sloan (1996), all variables are deflated by average total assets."^

The sample excludes financial services firms (SIC codes 6000-6999) because the modelis not developed to reflect their activities. It also excludes observations with sales less than$10 million, share price less than $1, EARN or CF in the extreme upper and lower 1percent of their respective distributions, and studentized residuals greater than 3 in absolutevalue.'"^ Because we seek to compare the findings from equation (11) for k = 2 and equation(12), sample firms in the primary analyses must have data sufficient for estimating thesetwo specifications. These criteria result in a primary sample of 10,164 firm-yearobservations.

Table I presents descriptive statistics for the variables used in the estimation equations.Panel A reports distributional statistics and Panel B reports Pearson and Spearman corre-lations. Consistent with prior research, e.g.. Sloan (1996), Panel A reveals that the meansand medians of EARN and CF are positive and those of aggregate accruals, ACCRUALS= EARN - CF, are negative. The negative mean and median for ACCRUALS reflect thefact that aggregate accruals includes depreciation and amortization, but acquisition of de-preciable and amortizable assets is an investing, not operating, activity under SFAS No. 95.Also consistent with Sloan (1996), current accruals, i.e.. AAR, AINV, and AAP, are smallerin magnitude and more variable than DEPR, a long-term accrual. Finally, less than one-half of the sample firms report amortization expense, as evidenced by a median AMORTof 0.00.'"

Panel B of Table 1 reveals that, as expected, EARN is significantly positively correlatedwith CF and ACCRUALS, and CF and ACCRUALS are significantly negatively corre-lated.'^ With the exception of AMORT, the accrual components are individually signifi-cantly correlated with EARN and CF, and generally are also correlated with each other.Untabulated statistics indicate that EARN, ACCRUALS, and CF are significantly autocor-related. The persistence of CF, 0.47, is substantially greater than that of ACCRUALS, 0.26,consistent with prior research such as Sloan (1996) and Barth et al. (1999). The persistenceof EARN, 0.66, exceeds those of ACCRUALS and CF.'"

Results: Aggregate EarningsTable 2 presents regression summary statistics from estimating equation (11), which

tests the predictive ability of current and past aggregate earnings for next period cash fiow.The results reveal that current earnings, EARN,, is significant in predicting one-year-ahead

'•* Our variable definitions follow DKW, excepi that DKW calculates cash from operations from balance sheet andincome statement amounts. Also. DKW deflates by the number of shares outstanding. Seciion VI reports thatour inferences are unaffected by the source of cash from operations or the choice of deflator.

'^ Our inferences are unaffected by eliminating observations based on any of the stated criteria.' If amortization expense is missing, ihen AMORT equals lotal depreciation and amortization minus depreciation

and minus depletion: if depletion is missing and the firm is not in an industry for which depletion is common,i.e., SIC codes 1000-1499. 2900-2999. and 4600^699. then we set depletion equal to 0. Depletion is includedin OTHER.

''' Throughout, the term "significant" refers to statistical significance at less than the 0.05 level."* The statistics in Table 1 differ somewhat from those in some prior studies, e.g., Dechow (1994), Sloan (1996),

DKW. and Barth etal. (1999), but are consistent with those in others, e.g.. Burgstahler et al. (1998). For example,we report that the standard deviations of EARN. CF. and ACCRUALS are similar, whereas Dechow (1994) andDKW find that earnings is less variable than cash flow. Untabulated statistics reveal that the differences areprimarily attributable to the definition of accruals, sample selection, and deflators. Despite these differences indescriptive statistics. Section VI reports that our inferences are robust to these choices.


o p q p q p p p pd o d o d d o d d

o

* 1 -

*

ENo

*

d1

*

O1

od

*

o

* •

sd1

*5o1

5d

oo1

od1

*co

od

oo

*

d1

co1

*od1

e

3

ntit

* * * * *in D fN r*-, \O —

p (N rn p p pd d d d d dI I I I I

d dI I

#

•—1

d—— 1

d—od

*m

d

*(^d

*. — •

CD o*od

£ 5•S o

S p o o p p p p pd d d d d d d d d

I I * * * *•d- O m Or-1 rn in —

d d o d

ts o

2 a^ Eat w

sss p p q q q p

on-J<D

U

(N — Tj

d d d

d d

— O ^ m COri Tf- q q qd d d d d

o r- rN — m•rr Tf p -t qd d d d d

pd d d d d d1 1 1

* * * * * * *i/ r^ c^ c^ '—' O O Od d d d d d d d

P J U < < 1 < I < 1 Q < O


inui/iu

cra

So"

Etoao

;rat

i

2 1o-o

KU

o

1•aB

ace

tfi

CF:

3C

. . .

oin%

o

ash

rno*

O

E C c• - O c

e • - =

s § : •

Z !-•

"S o .5> U "H^ - oS s

C i - ^

u

. c xi= E ^

^ r-i ^•£ ° tt

I = ^

• ^ « II

I c ^ ^^T' r/ ^ m ^U I U I—M r^ ^^

. 3 ^C "o iJ c o

"ra 5

.^ Z " .XI a: ^•r <

o X5; H G

ou

— f^ 00 r~~ r-i c-i —3 C O O

^co —

O O O O O O C O O

00 CN -^ O Cj -^

^?0

p ( p p O pO O O O O O

•a I

I - • - <= I

^V

d

p r p O p pcJ o o o d o

o —

3 I

- i cS iSEl

p " p p Pd o d d o

O TT••C 00iO OS

•2 E

01 aP rn p ^d d d d

—1 O\rn 1/1d —

CO

E

z z z z z<

40


cash flow, CF, + |, explaining 15 percent of its variation. However, the results also revealthat, consistent with predictions, additional lags of earnings are significant in predictingnext period cash flow. Consistent with the model and equation {8), we find that current andtwo lags of earnings are significant in the prediction equation, explaining 17 percent of thevariation in next period cash flow. In fact, up to six lags of earnings are significant inpredicting next period cash flow, consistent with long-lived asset investment policies beingomitted from the model. The adjusted R-s increase monotonically from 0.15 for the spec-ification with current earnings only to 0.19 for the specification with current and six lagsof earnings. Untabulated statistics indicate that the EARN coefficients significantly differfrom one another in each specification. Taken together, these results indicate that aggregatecurrent earnings is not an unbiased predictor of future cash flows, permitting a role foraccruals in predicting future cash flows as the model predicts.

Results: Cash Flow and Components of AccrualsTable 3 presents summary statistics from estimating equation (12), which disaggregates

current earnings into cash flow and the major components of accruals. Panel A reveals that,as predicted, the six accrual components are all significant in predicting next period cashflow, with signs consistent with predictions; the t-statistics range in absolute value from10.54 to 28.58.''' The significance of the DEPR and AMORT coefficients is inconsistentwith assertions in the financial press that depreciation and amortization are not predictiveof future cash flows.

Comparing the adjusted R's associated with the specification with current and two lagsof aggregate earnings in Table 2, 0.17, and disaggregated current earnings in Panel A ofTable 3, 0.35, reveals that the disaggregated earnings specification has substantially morepredictive ability. An untabulated Vuong {1989) Z-statistic indicates that this difference issignificant. The model predicts that the predictive abilities of these two specifications arethe same. Thus, this finding suggests that the model's focus on working capital accrualsand its other simplifying assumptions understate the predictive ability of current cash flowand the components of accruals relative to current and two lags of aggregate earnings. Weinvestigate below the extent to which the superior predictive ability of disaggregated currentearnings is attributable to {1) disaggregating cash flow from aggregate accruals, or {2)disaggregating accruals into its major components.

Untabulated statistics associated with comparisons of the adjusted R^s of the disaggre-gated current earnings specification to each specification in Table 2 reveal that disaggregatedcurrent earnings has significantly more predictive ability than up to seven years of aggregateearnings. This result indicates that the higher adjusted R' in Panel A of Table 3 is notmerely attributable to a larger number of explanatory variables in the disaggregated earningsspecification.

Panel A of Table 3 also presents results from tests of coefficient restrictions comparingthe explanatory power of disaggregated earnings {Unrestricted) to cash flow {CE only),aggregate earnings (EARN only), and the accrual components {Accrual components only).As predicted, all coefficients on the accrual components significantly differ from 0. and,thus, the unrestricted specification has significantly more predictive ability than the CE-only specification. Also as predicted, the coefficients on cash flow and the six accrualcomponents significantly differ. Thus, disaggregating earnings significantly enhances

Unlabulated findings reveal that some coefficients on up to four lags of the predictor variables differ significantlyfrom 0.


TABLE 3Summary Statistics from Regressions of Future Cash Flow on

Current Cash Flow and Components of AccrualsSample of Compustat Firms 1987-1996

Panel A: Regression Summary Statistics, All Variables Deflated by Average Total Assets

Variable

InterceptCF,AAR,AINV,AAP,DEPR,AMORT,OTHER,

Adj. RnTests of coefficient restrictions:

UnrestrictedCF onlyEARN onlyAccrual components only

Prediction7

+++-++

Coefficient

0.010.590.420.35

-0.560.420.470.15

0.3510,164

Adf R'

0.350.240.15

o.n

t-statistic

7.8961.3428.1021.75

-28.5816.3911.0510.54

p-value

<0.01<0.01<0.01

Panel B: Regression Summary Statistics, All Variables Undefiated

Variable Prediction Coefficient t-statistic

CF_U, +AAR_U, +AINV_U, +AAP_U,DEPR_U, +AMORT_U, +OTHER_U, ?

Adj. RnTests of coefficient restrictions:

UnrestrictedCF onlyEARN onlyAccrual components only

0.540.670.15

-0.280.761.780.32

0.95610,164

Adj. /?-

0.9560.9450.9360.947

39.3725.82

5.01-10.02

25.4311.9821.25

p-value

<0.01<0.01<0.01

See Table t for variable definitions._U denotes variables are undefiated. p-values associated with F-tests of coefficient restrietions as follows: CF only(*AR = *i = -4>AP = ^a= 4»AM = •i'o ^ 0). EARN only (<})„• = <1> R = ^ I = -<t> |, = -4),, = -( |)^M = ^J. andAccrual components only {()>fp = 0).


predictive ability relative to aggregate earnings.^" Panel A also indicates that the adjustedR- of the CF-only specification exceeds that of the EARN-only specification; an untabulatedVuong (1989) Z-statistic indicates that the difference is significant. Finally, the accrualcomponents alone have the least predictive ability for next period cash flow.

Panel B of Table 3 presents results from estimating equation (12) using undefiatedvariables and including fixed firm and year effects. Estimating this equation ensures thatour findings are not attributable to scaling by contemporaneous average total assets. Theinferences drawn from Panel B are similar to those drawn from Panel A, although, as istypical when estimating undefiated specifications, the level of the adjusted R^s is higher.

Consistent with Panel A, the six accrual components are all significant in predictingone-year-ahead cash flow. Tn addition, untabulated findings indicate that the adjusted R^from a fixed-effects specification with current and two lags of aggregate earnings is 0.937,which is lower than 0.956 from the disaggregated current eamings specification. The resultsfrom tests of coefficient restrictions in Panel B are also generally consistent with those inPanel A in that the unrestricted specification has the most predictive ability and the pre-dictive ability of the CF-only specification exceeds that of the EARN-only specification.The only difference between the findings in Panels A and B is that in Panel B the speci-fication with accrual components only has more predictive ability than either CF only orEARN only. Because our primary inferences are Insensitive to deflation and to facilitatecomparison to prior related research that typically estimates deflated specifications, hence-forth we tabulate only findings from deflated specifications.

Results: Cash Flow and Aggregate AccrualsThe model predicts that current and two lags of aggregate eamings has the same pre-

dictive ability as disaggregated current eamings. However, the results in Tables 2 and 3indicate that the latter specification has significantly more predictive ability. We estimateequation (13) to investigate whether the higher adjusted R- for the disaggregated earningsspecification is attributable to (I) disaggregating cash flow and aggregate accruals, or (2)disaggregating the major accrual components:

k k

CF,,. , = <t + 2 4)cF.,-. CF,,,_, + 2 *A.. . ACCRUALS,,., + u,,. (13)

where aggregate accruals, ACCRUALS, equals dAR + AINV - AAP - DEPR - AMORT+ OTHER and, for the sake of parsimony, k :S 3.

Table 4 presents the findings. It reveals that for all specifications, the CF coefficientsare significantly positive and current aggregate accruals, ACCRUALS,, adds significantlyto cash flow in predicting next period cash flow. For specifications including more than twoyears of predictor variables, lagged ACCRUALS has no significant explanatory power in-cremental to cash flow and current ACCRUALS. The adjusted R-s increase monotonicallyfrom 0.27 to 0.35 as additional lags of cash flow and aggregate accruals are included.

Comparing the predictive ability of multiple lags of CF and ACCRUALS to the unre-stricted specification with only year t variables estimated on the same sample indicates that

Because we do not predict the sign of t|>(, and predict the sign of 4)^^ based on empirical estimation, we testthe coefficient restrictions excluding one or both of these coefficients. Untabulated results reveal the sameinferences as when including these coefficients. Untabulaled tests also reveal that the four current accmalscoefficients are not all equal; we cannot reject the hypothesis that the two long-term accruals coefficients areequal.


o

qd

T — — q — q q qc J d d o o d d d

oc(5 d d dV V V V

su•oc

I

ON

_ r-"O 00

J ^ s

^ * I"

1

3g

-e-OT3C

C

3

O >-O 00' _ p ^rJ 00 ir-)

r ( Otn 00 qri d —

qc o d d

qd d d o

2: od d d

t/D CO

-J -JD,

ti. u. a. u.u u u u

3 ;D

u uu u< <

ID Da: a:u uu u< < < c

q q q qd d d dV V V V

rn rn rn — qd d d d d

o q q qd d d dV V V V

>o O 00 r~ mrn rn (N — qd d d d d

d d d dV V V V

d d d d

U

•d

u- <

9-U


current cash flow and the accrual components have significantly more predictive ability thanup to four years of CF and ACCRUALS. However, the significance of the difference be-tween the two specifications declines with the inclusion of more lagged variables, anduntabulated findings reveal that the difference is insignificant in a specification includingcurrent and four lags of CF and ACCRUALS. Recall from Tables 2 and 3 that current andsix lags of aggregate earnings have significantly less predictive ability than current earningsdisaggregated into cash flow and the components of accruals. The findings in Table 4 revealthat the superior predictive ability of disaggregated current earnings relative to multiplelags of aggregate earnings derives from disaggregating cash flow from aggregate accruals,as well as from disaggregating the components of accruals.

VI. ADDITIONAL ANALYSESOperating Cash Cycle

The model in Section III is based on a single firm, whereas the estimation equationsare cross-sectional. Thus, differences in model parameters across firms could affect ourinferences. DKW shows that the predictive ability of earnings for future cash flows dependson the firm's operating cash cycle, 8. This is because, as the first line of equation (6) shows,the difference between realized CF,^, and EARN, depends on accruals, as reflected in 8,where 8 = [a + (1 - i r h , - p(l - -IT)]. TO investigate whether the incremental predictiveability of disaggregated accruals is robust to controlling for 8, we repeat our analyses afterpartitioning firms into 5-quartiles, as in DKW.''

Table 5, Panel A, presents descriptive statistics for 6. The full sample distribution iscomparable to that in DKW. The statistics reveal that partitioning firms into quartiies sub-stantially reduces the variation in 8 relative to the full sample, especially for the first threeportfolios. Table 5, Panel B, presents regression summary statistics from estimating thedisaggregated earnings specification by 8-quartile. The findings are consistent with thosein Table 3, indicating that the Table 3 results are not solely attributable to differences in 8.In particular, in each of the four 8-quartile portfolios all accrual component coefficientshave predicted signs and differ significantly from 0, with the exception of AMORT inquartile 1. Finally, in each of the four portfolios the results of tests of various coefficientrestrictions in Panel C are consistent with those in Table 3.

Industry MembershipBecause the types and mix of accruals likely vary by industry, we also estimate the

disaggregated earnings specification for 13 industries identified by Barth et al. (1998). Table6 presents the findings. Panel A reports descriptive statistics on operating cash cycle, 8, byindustry. Not surprisingly, 8 varies substantially across industries. The substantial variationin 8 within industries reveals that operating cash cycle and industry membership are notperfectly correlated. For eight of the 13 industries the interquartile range of 8 exceeds themaximum interquartile range of 0.12 for the four 8 portfolios in Table 5, Panel A.

Panel B of Table 6 presents summary statistics from equation (12) estimated by indus-try. It reveals substantial variation across industries in the predictive abilities of cash flowand the accrual components; the adjusted R^s range from 0.76 to 0.12. However, the sep-arate-industry findings are largely consistent with those in Table 3. The coefficients on CF,

As in DKW, a = (AR, + AR,_J - 2 S,, p = (AP, + AP,_,) ^ 2 S, (I - IT), IT = E, - S,, averaged acrossthe number of years for which data are available, and 7i = gi " (1 ^ i^). where gi is the estimated coefficienton S, in a regression of INV, on S, and i S , . Thus. 8 is average days accounts receivable and inventory minusaverage days accounts payable. DKW interprets S as a fraction of a year; multiplying 5 by 360 permits inter-pretation of 8 in terms of days. Estimating firm-specific inventory regressions requires at least five years of data.

46 The Accounting Review. January 2001

so

o d o den

t

codu

i

oU•oe

o

j i

Uc

9

uo

o

(A

Auu3

OK

_oI/IlA

DC

uio

(A

n

a>"u

JS

n

o

tim

ate

VI

CO

• n

aB3• nat

EL.O

Un3

o•ai«^*35uu

• ^

I

1987

-

SUI

L.

n^

aoUO

mpl

3

-1-

0=-'

1•0-

O

•© • •

4-

CKT

+

l i .

"©-4-

:

-1-

a:

+

u

+

t o

..

O

rtit

e

a3

" ^

3

"1-£

.^

r-i ci o o 00d d d d ^

Ci U-) r-J 3— — q oo o o o

o q q qd d d d

q 00d

O O O O

'^

O O v O —

p rt r-i — ^ I-'-, Tt qd d d d d d d d

2CJ Tt - - — ^c in m rn "*d d d d d

1

^ j ^ _f . O-) _d d o

f-~j - ^ —

d d d d d d d d

— T)- o fnr-i -^ xn 00

d d d d d d d d

O' • -t- + + I + +

—

c .2 Q CT-

o It T3 -2=eg

.a


3

.sOU

3

-0-+

a:a.

+

Uo

4-

& '

q q p qo d d dV V V V

q — q

d d d d d

Ci Ci Ci a>d d d dV V V V

CT\ — r ^ O^ Orr-, r ^ — fN —

d d d d d

a3

odV

odV

odV

ooV

in in (N ^ r jd d d d d

^8| <ti- < ti- '->U P-1 U <

odV

odV

odV

dV

<1IIM

C

-e-

II ^

(J -

-e- ?-

5 I

-l< 5 ^

•d "^ •>-^ to o

{J O £y II V.

•2 1!CJ _

5 -0-

ii'O O bU

^ [ J "P

-& Ei

E Z

=- -p

> " O 00

•= +

^ 'et

11

T3 o

48 The Accounting Review, January 2(X)1

s

oa.

oU1 3

SCO

II

U 2

CD C 1S<: o 3

I/I O

O5

+

;ssi

SO

UIO

rV""

k

+•s-

g

I

o c \ D r o o o ^q q — I — q o

DC

O O O O O O O O O O O O O O Si

^^ C::^ ^ ^ „ e^-^C) ~^^

O

o o o o o o o o o d d d d

niM-i<N(NOON (N

CO O —m 00 CO


IO

CQ

3+

i4

+

5

I+

4)

ddddddddddddd j "

c* * * * * * * * * * * *fNt^ — [ ^ o o O r - i Q o o m m — 00 O t ^ g

d d d d d d d d d d d d — a^ .s

\ 0 ( N o o o i n o d t ^ ^ ' ^ c j \ ' A f r i „ oo"^1 O T ^ o ^ n ' C O T ^ m o ^ l ^ ' ^ o m _ \DQV

d — ^ d d d d d d d d — ^ d o o — r n ^I I — —

* * * * * * * * *in '—' *o 'n 00 in ^n *" [ ^^ i^ CJ ^D ,. r* rj

I c^ ^n c^ ^^ c^ cT ^f ^f r*~i rn rn *c ~~ ^ i ov ^^d d d d d d d d d d d d - ^ — Tf^

I

* * * * * * * * * * * *

' d d d d d d d d d d d d - ^ ^ rn—I I I I I I I I I I I I I Y 7

* * * * * * * *_L r n q — r n i ^ ^ r n ' ^ - H — q r ^ q ^ '^•~~^

d d d d d d d d d d d d - ^ — rn —

* * * * * * * * * * * * *-J- in CN r^ m ^3 lO 'n o-i m <^ rN f* f r< *n oo

d d d d d - ^ d d d d d d - ^ — ri in

* * * * * * * * * * * * *in ^^ 00 o^ ^ T D ^ m c v^ o^ r* * r* r***.

4_ r^ ^f v^ in w in in ^o v^ sO ^f TJ" OO rn ~ ood d d d d d d d d d d d d — inrn

* * * * * * * * *

r-. p p p o p p p p p p p p p ^. ~^

d d d d d d d d d d d d d dr-

o

oa.c s > S

00

"So

•ca. C 3

3 T3

1 1 'E 11 .Si; t; IU c tij c

W H U U O. S

•c ^

N N


-©--I-

a.SO

u-©-

"Ic3

p p p p p p p p p p p p p ^

' v v v v v v v v v v v v v I"s:co

r i p p p p p p p p p p p p p p 3d'zicid'did'ddicicicidid' d d -g

e

p p p p p p p p p p p p pdddddddddddddV V V V V V V V V V V V

ddddddddddddd d d

p p p p p p p p p p p p i n

V V V V V V V V V V V V

O O O O O O O O O •—' O O '*G

V V V V V V V V V V V

ddddddddddddd d d

0= ^O t ^ t^ O^ r^ oc lO ^^t ^ r* o-l '•^ c* O^ v~it^ r^ ro f^ r'*! '^ '^ ' ^ r^ • ^ f**^i c^ ^ c'j c^

ddddddddddddd d d

Tt" O^ ^^ f" I ' r ^r*" \f\ i/-'i ^ ^ 00(N — p ^ r-~ V,

rn

a:

IIII.H

bo

tI•cQ.

X g -C G ,W h- U U I

2 q

CO<


- E

ail

11OS

alu

>II

-e"

II

N

II

N

ta__

N


AAR, AINV, AAP. DEPR, and AMORT have predicted signs in 13, 13, 10, 13, 12, and 1!of the 13 industries, respectively, a result that has less than a 1 percent probability of beingobserved by chance, based on a binomial test. The Zl and Z2 statistics also indicate thatthe coefficients on cash flow and on each of the six accrual components differ significantlyfrom 0 when the separate industry results are taken together.-"

Panel C of Table 6 presents statistics associated with tests of coefficient restrictionsand reveals that the separate-industry inferences largely are consistent with those drawnfrom the full sample. Across-industry mean and median adjusted R-s are highest for thedisaggregated earnings specification (0.39 and 0.35), followed by the specification withcash flow and aggregate accruals (0.31 and 0.27), cash flow only (0.27 and 0.23), aggregateearnings only (0.16 and 0.14), and, finally, accrual components only (0.04 and 0.03). Theseindustry averages reflect relations that largely are consistent across industries. In all but theAgriculture and Pharmaceutical industries, the adjusted R- is significantly highest for thedisaggregated earnings specification. Food and Agriculture are the only industries in whichthe adjusted R- associated with the EARN-only specification is higher than that associatedwith the CF-only specification.-^

Homogeneous Model ParametersWe also estimate equation (12) using a sample of firms with relatively homogeneous

model parameters. Specifically, we estimate the parameters determining firms' operatingcash cycle, i.e., a, p, -TT, and 7,, as described in footnote 21, for each firm-year A firm-year observation is included if the estimate of each parameter lies in the middle two quar-tiies of that parameter's distribution. This procedure substantially reduces the sample size,reflecting the effects of intersecting the middle two quartiies of the distributions of the fourmodel parameter estimates and the requirement for multiple years of data for firm-specificparameter estimation. Untabulated findings reveal that inferences from this sample are thesame as those drawn from Table 3.

Year-by-Year EstimationThe findings thus far are based on pooling observations over time and across firms.

This raises the possibility that residual correlation results in overstated test statistics. Thus,as a robustness check, we estimate equation (12) separately by year. The untabulated find-ings indicate that our inferences are consistent across years, and when based on Zl and Z2statistics that aggregate the separate-year findings. In addition, in all ten years the tests ofcoefficient restrictions indicate that the adjusted R-s of the disaggregated earnings specifi-cation are highest, followed by the specification with cash flow and aggregate accruals,cash flow only, earnings only, and finally, accrual components only.

Predicting Cash Flow More than One Year AheadThe analyses thus far focus on predicting one-year-ahead cash flow, whereas some

related prior research predicts cash flows further in the future. To investigate the predictive

Zt = ([/VT)21^|(t,Vki/(k, - 2)), where T is number of industries, j . t, is the t-statistic and k| is the degrees offreedom (Healy et al. 1987). Z2 = mean l-statistic/(standard deviation of t-statistics/v'(T - 1)) (White 1984;Bernard 1987). Z! assumes residual independence; Z2 relaxes this assumption.DKW argues that the predictive ability of earnings relative to cash flow increases as the operating cash cycleincreases. Untabulated statistics reveal that the difference in adjusted R^ between the EARN-only and CF-onlyspecifications is significantly positively correlated with & (Spearman corr. = 0.53. t = 2.14). However, consistentwith 5 not fully explaining our findings relating to accrual components, the difference in adjusted R- betweenthe unrestricted, i.e., disaggregated earnings, and CF-only specifications is not significantly correlated with h(Spearman corr. = -0.04. t = -0.12).


abilities of aggregate and disaggregated earnings for cash flows more than one year ahead,we estimate equations (11) and (12) using CF,+,. CF,.,, or CF,+, as the dependent variable.We estimate equation (11) using current and three lags of EARN. We use more than twolags of EARN because our findings in Tables 2 and 3 indicate that more than two lags ofEARN are significant in predicting future cash flows; we limit the lags to three for the sakeof parsimony. Table 7 presents the findings. Panel A reveals that current and up to threelags of EARN are significant predictors of cash flows more than one year in the future,although as the prediction horizon increases, the prediction equation's explanatory powerdecreases monotonically.

Panel B of Table 7 reveals that cash flow and the accrual components of current earn-ings all have significant predictive ability for cash flow up to four years ahead, and thesigns are consistent with those estimated when predicting one-year-ahead cash flow. Aswith Panel A, the explanatory power of the regression decreases as the prediction horizonincreases.'^ Consistent with the one year-ahead cash flow prediction findings. Panel B re-veals that the cash flow and components of accruals specification has the most predictiveability for cash flows up to four years in the future, followed by cash flow and aggregateaccruals, cash flow only, and multiple lags of aggregate earnings in Panel A. The currentEARN-only and current year accrual components-only specifications have the least predic-tive ability. All differences are significant.

Prices, Returns, and Discounted Cash Flows as Alternative Dependent VariablesAs noted in Section II, prior research concludes that aggregate current earnings has a

higher association with prices and returns than does current cash flow. Viewing prices andreturns as proxies for future cash flows, these findings conflict with ours. To investigatewhether the differences in findings are attributable to variable definitions or sample selec-tion, we estimate equation (12) using as the dependent variable: (1) market value of equity,deflated by average total assets, (2) annual returns, or (3) discounted cash flows, deflatedby average total assets.^'^

Table 8 presents the findings. The coefficients have predicted signs and are significantlydifferent from 0, with two exceptions: DEPR's coefficient is significantly negative in themarket value and returns specifications, and the coefficient on AMORT is insignificantlydifferent from 0 in the returns specification. Untabulated statistics from estimating the one-year-ahead cash flow and discounted cash flows specifications with cash flow defined asCF minus capital expenditures also reveal signiticantly negative coefficients on DEPR.These results indicate that the negative DEPR coefficients in the price-based specificationsare attributable to market prices implicitly reflecting investing cash flows, whereas CF isoperating cash flow. Possible explanations for the insignificant AMORT coefficient in thereturns specification are that amortization expense is not informative for returns and thatAMORT's year-to-year variation is too small to be detectable.

Nonetheless, consistent with our primary results, for each dependent variable the spec-ification permitting the coefficients on cash flow and the components of accruals to differhas a significantly higher adjusted R^ than the constrained coefficient specifications. Inaddition, when CF or the discounted cash flow measure is the dependent variable, cash flow

^* Untabulaled findings reveal that our inferences are unaffected by including up to four lags of the predictorvariables, although some of the coefficients on the additional lagged variables are significantly different from 0.

-^ We discount at a rale of 12 peri:enl realized cash flow. CF, for each of three future years and assume thatrealized cash fiow four years in the future is a perpetuity. The additional data required to calculate discountedcash flow reduce the sample size for the estimation using this variable.

54 The Accounting Review. January 2001

TABLE 7Summary Statistics from Regressions of Future Cash Flows onEarnings or Current Cash Flow and Components of Accruals

Sample of Compustat Firms I987-I996

Panel A: Regression Summary Statistics. Aggregate Eamings Only

,,_,+ u,,

CF CF,,, CF

Variable

InterceptEARN,EARN,_|EARN, 2EARN,_,

Adj. R'n

Coefficient

0.070.280.090.050.06

0.147.709

t-siatistic

66.1920.646.653.324.69

Coefficient

0.070.220.080.030.08

0.106.565

t-statislic

61.2314.075.161.725.68

Coefficient

0.070.220.070.010.08

0.085.481

(statistic

55.2012.473.800.765.01

Panel B: Regression Summary Statistics, Cash Flow and Accrual Components

CF,

Variable Prediction

Intercept ?CF, +AAR, +AINV, +AAP,DEPR, +AMORT, +OTHER, ?

Adj. R'n

Coefficient

0.020.520.320.38

-0.410.430.360.17

0.277,709

Tests of coefficient restrictions:

UnrestrictedPanel A .specificationCF and ACCRUALSCF onlyEARN onlyAccrual componentsonly

CF,

Adf R'

0.270.140.220.190.130.09

t-statistic

11.2444.0817.3019.54

-16.4214.227.749.19

p-value

<0.01<O.OI<0.01<0.01<O.OI

Coefficient

0.020.440.230.26

-0.290.490.180.11

0.226.565

CF,

Adj. R'

0.220.100.180.170.080.09

t-statistic

12.3633.3411.1912.04

-10.2314.28

1.725.40

p-vatue

<0.01<0.01<O.OI<0.01<O.OI

Coefficient

0.020.420.300.22

-0.330.570.520.16

0.215.481

CF,,

Adj. /e-

0.210.080.150.140.070.10

t-statistic

10.2928.0412.489.07

-10.2514.473.547.01

•4

p-value

<0.01<0.01<0.01<O.OI<O.OI

See Table 1 for variable definitions.p-values comparing the Panel A specification to the unrestricted specification in Panel B are based on Vuong s(1989) Z-statistic. All other p-values are associated with F-tests of coefficient restrictions as follows: CF andACCRUALS ((|), K = <t>i = -"t-AP = -*r) = -*AM = 4»o). CF only (<f>,.\H= *, = -it^p = <bn = *AM = *<» = 0),EARN only ^i;>^J^: = ^^^ = (t»i = -<I>AP = ~^n = " ^ A M - <J'o)> ''nd Accrual componenis only ((J),-.p = 0).

Barth, Cram, and Nelson—Accruals and the Prediction of Future Cash Flows

.i "3IS g

I

2C/3

c S pr

J- 00

Is

l i

-I-

2I

I I

I Id o

I I

r N ^ ^ C O ^— m CM —1 I —

— o o m r - J t ^

r i d ^ o d

d d d d d d d d

q q p qd d d dV V V V

O r*i in— q q

H- -I- -I- I -I- -f

CL.

uDtCr>

<

<:^^ O_ o c:i_d d d dV V V V

r** *—* o o^ ^tN r-j tN q qd d d d d

q q q qd d d dV V V V

O O t*-> t^ -rtm CM — — —o o o o o

!=> a c> ad d d diV V V V

O<-n rn — tN —d d d d d

a:

"C -a

Io

1 2C U, < U. q -^3 CJ W U < O

a; «

.p -

55

+ z_ o

UJ

0-e-

. o r e

> "J

t S

fe^

^ K.S

S n - u S -S

V CL.

• r - fl>

5 l 3

(u . 3•S ^o >

II

0.-9-


has more predictive ability than aggregate earnings. However, consistent with prior price-and returns-based research, when market value or returns is the dependent variable, aggre-gate earnings has more predictive ability than cash flow. This difference in findings likelyis attributable to the difference in the sign of the DEPR coefficient—DEPR is subtractedto arrive at EARN, not added—rather than to sample selection or variable definitions.

Other Robustness ChecksFour additional sensitivity checks do not affect our inferences: (1) estimating all spec-

ifications using undeflated variables, and using sales, number of shares outstanding, andmarket value of equity as alternative deflators; (2) defining EARN as operating earningsafter depreciation, which excludes special items, as in Sloan (1996); (3) limiting the sampleto firms disclosing accrual components on the statement of cash flows;-" and (4) limitingthe sample to firms trading on NYSE and AMEX, as in Dechow (1994) and Sloan (1996).

VII. SUMMARY AND CONCLUSIONSOne of the FASB's stated objectives is that earnings and its components, which include

accruals, provide a better indication of future cash flows than current cash flow. However,extant research does not clearly establish this fundamental accounting link. Using a modelof the accrual process, we provide insights into the role of accruals in predicting futurecash flows and provide evidence consistent with predictions based on the model.

Our predictions are based on an extended analysis of the Dechow et al. (1998) modelof the working capital accrual process. The model contains several simplifying assumptions;it includes only three current accruals with relatively simple time-series processes. We leaveto future research the development of a more comprehensive model, including explicitmodeling of long-term accruals. Whereas we limit our predictions to the signs of theweights on all earnings components, a more complete model would permit specific predic-tions about the magnitudes of the weights. It also would permit specific predictions aboutthe role of long-term term accruals in predicting future cash flows, and about the relationbetween current earnings and its components and cash flow more than one year ahead. Wealso leave to future research the modeling of industry-specific differences in accountingpolicies and practices, which could sharpen predictions and inferences.

Despite these limitations, however, the model provides insights into the role of accrualsin predicting future cash flows. In particular, the model reveals that the predictive abilityof accruals for future operating cash flows derives from management's expected futureinvestments in operating assets, in addition to delayed cash receipts or payments related topast transactions. Each major accrual reflects different information about future cash flows,resulting in different weights in prediction. In contrast, aggregate earnings implicitly placesthe same weight on each earnings component, masking information relevant to predictingfuture cash flows. A key insight from the model is that expected future cash flow can beexpressed as a function of several lags of aggregate earnings or as a function of the cashflow and major accrual components of current earnings.

Consistent with predictions, we find that disaggregating earnings into cash flow and sixmajor accrual components—change in accounts receivable, change in inventory, change inaccounts payable, depreciation, amortization, and other accruals—significantly enhancesthe predictive ability of earnings. Each accrual component's relation with future cash flowsis significant and of the predicted sign, indicating the accrual components aid in predicting

Pearson (Spearman) correlations between ihe accrual components disclosed under SFAS No. 95 and thosecalculated from balance sheet data are 0.7.'i (0.X2), 0.82 (0.86). and 0.75 (0.79) for AAR. AINV, and AAR

Barth. Cram, and Nelson^Accruais and the Prediction of Future Cash Flows 57

future cash flow beyond eurrent cash flow. In contrast to claims in the financial press, long-term accruals, specifically depreciation of long-lived tangible assets and amortization ofintangible assets, have significant predictive ability for future eash fiows. We also find thatthe explanatory power of disaggregated current eamings exceeds that of current and up tosix lags of aggregate eamings, and that of current and up to four lags of eamings disag-gregated into cash flow and aggregate accruals. Disaggregating cash flow from aggregateaccruals significantly increases predictive ability relative to aggregate eamings, but disag-gregating accruals into its major components further significantly increases predictive abil-ity. Our findings are robust to predicting cash flows several years in the future and usingshare prices, returns, or discounted cash flows as a proxy for future cash flows. Our findingsalso are robust to controlling for operating cash cycles and industry membership, suggestingthat knowledge of accounting accruals aids in predicting future cash flows over and abovethese firm characteristics.

REFERENCESAboody D.. M. E. Barth. and R. Kasznik. 1999. Revaluations of fixed assets and future firm perform-

ance: Evidence from the U.K. Journal of Accounting and Economics 26: 149-178.Ali. A. 1994. The incremental information content of eamings, working capital from operations, and

cash flows. Journal of Accounting Research 32: 61-74.Ball, R.. and P. Brown. 1968. An empirical evaluation of accounting income numbers. Journal of

Accounting Research 6: 159-178.Barth, M. E.. W. H. Beaver, and M. A. Wolfson. 1990. Components of eamings and the structure of

bank share prices. Financial Analysts Journal 46: 53-60.. , and W. R. Landsman. 1992. The market valuation implications of net periodic

pension cost components. Journal of Accounting and Economics 15: 27-62.and . 1998. Relative valuation roles of equity book value and net income as

a function of financial health. Journal of Accounting and Economics 25: 1-34.-, J. R. M. Hand, and W. R. Landsman. 1999. Accruals, cash flows, and equity values.

Review of Accounting Studies 4: 205-229.and . 2000. Accrual components, eamings forecasting, and equity

values. Working paper, Stanford University, Graduate School of Business.and A. P Hutton. 2000. Information intermediaries and the pricing of accruals. Working

paper, Stanford University.Beaver, W. H.. and R. E. Dukes. 1972. Interperiod lax allocation, eamings expectations, and the

behavior of security prices. The Accounting Review 47: 320-333.Bernard, V. L. 1987. Cross-sectional dependence and problems in inference in market-based account-

ing research. Journal of Accounting Research 25: 1-48.—, and T L. Stober. 1989. The nature and amount of information in cash flows and accruals.

The Accounting Review 64: 624-652.-. and J. Noel. 1991. Do inventory disclosures predict sales and eamings? Journal of Account-

ing. Auditing, and Finance 6: 145-181.Bowen, R. M.. D. Burgstahler, and L. A. Daley. 1986. Evidence on the relationships between eamings

and various measures of cash flow. The Accounting Review 61: 713-725.. , and . 1987. The incremental information content of accruals versus cash

flows. The Accounting Review 62: 723-747.Burgstahler. D.. J. Jiambalvo. and Y. Pyo. 1998. The infomiativeness of cash flows for future cash

flows. Working paper. University of Washington.Cheng. C . C. Liu. and T. Schaefer. 1996. Eamings permanence and the incremental information

content of cash flows from operations. Journal of Accounting Research 34: 173-181.Dechow. P. M. 1994. Accounting earnings and cash flows as measures of firm performance: The role

of accounting accruals. Journal of Accounting and Economics 18: 3-42., S. P. Kothari. and R. L. Watts. 1998. The relation between eamings and cash Hows. Journal

of Accounting and Economics 25: 133-168.


DeFond. M. L., and C. W. Park. 2000. The reversal of abnormal accruals and the market valuationof earnings surprises. Working paper. University of Southern California.

Feltham, G. A., and J. A. Ohison. 1996. Uncertainty resolution and the theory of depreciation mea-surement. Journal of Accounting Re.search 34: 209-234.

Financial Accounting Standards Board (FASB). 1978. Objectives of Financial Reporting by Busine.ssEnterprises. Statement of Financial Accounting Concepts No. I. Stamford, CT: FASB.

. 1987. Statement of Cash Flows. Statement ot Financial Accounting Standards No. 95. Stam-ford, CT: FASB.

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

Greenberg, R. R., G. L. Johnson, and K. Ramesh. 1986. Earnings versus cash flow as a predictor offuture cash flow measures. Journal of Accounting. Auditing, and Finance 1: 266-277.

Healy, P., S. Kang, and K. G. Palepu. 1987. The effect of accounting procedure changes on CEO'scash salary and bonus compensation. Journal of Accounting and Economics 9: 7-34.

Lipe, R. 1986. The information contained in the components of earnings. Journal of AccountingResearch 24: 37-68.

Lorek, K. S., and G. L. WilUnger. 1996. A multivariate time-series prediction model for cash-flowdata. The Accounting Review 71: 81-101.

MacDonald, E. 1999a. Analysts increasingly favor using cash flow over reported earnings in stockvaluations. Wall Street Journal (April 1): C2.

. 1999b. First call will add cash EPS estimates for 20 web firms, responding to street. WallStreet Journal (April 14): A4.

McNichols, M.. and G. P. Wilson 1988. Evidence of earnings management from the provision forbad debts. Journal of Accounting Research 26: !-40.

Penman, S. H., and T Sougiannis. 1998. A comparison of dividend, cash flow, and earnings ap-proaches to equity valuation. Contemporary Accounting Research 15: 343-383.

Pfeiffer, R. J., P. T. Elgers, M. H. Lo, and L. L. Rees. 1998. Additional evidence on the incrementalintormation content of cash flows and accruals: The impact of errors in measuring market ex-pectations. The Accounting Review 73: 373-385.

Rayburn, J. 1986. The association of operating cash flow and accruals with security returns. Journalof Accounting Research 24: 112-133.

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

Vuong, Q. H. 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econo-metrica 57: 307-333.

White. H. 1984. Asymptotic Theory for Econometricians. Orlando, FL: Academic Press, Inc., HarcourtBrace Jovanovich.

Wilson, G. P. 1986. The relative information content of accruals and cash flows: Combined evidenceat the earnings announcement and annual report release date. Journal of Accounting Research24: 165-200.

• . 1987. The incremental information content of the accrual and funds components of earningsafter controlling for earnings. The Accounting Review 62: 293-322.

Xie, H. 2000. Are discretionary accruals mispriced? A reexamination. Working paper, The Universityof Arizona.

barth, cram, nelson (2001)

Documents

future cash fiows

cash flow prediction

current cash flow fasb

operating cash oyole

future cashflows

future cashfiows

role of accruals

components of earnings