Assessing Earnings and Accruals Quality: U.S. and International Evidence

Transcription

1 ssessing Earnings and ccruals Qualiy: U.S. and Inernaional Evidence Peer D. Wysocki * MIT Sloan School of Managemen E Memorial Drive ambridge M 0 US wysockip@mi.edu This version: January 009 bsrac This paper examines he properies and validiy of accouning qualiy proxies derived from he Dechow and Dichev DD 00 accruals model. I analyically demonsrae ha he underlying DD model has limied abiliy o disinguish beween manipulaed and high qualiy accruals. My empirical analyses show ha prior validaions of he DD model have significan limiaions and he model canno empirically disinguish beween discreionary and non-discreionary accruals. Overall he DD model does no appear o reliably capure high qualiy accruals and in some seings will even reverse rank firms earnings qualiy. Implicaions for curren and fuure research are discussed. Keywords: ccouning qualiy ccruals ash flow Earnings qualiy Inernaional * ommens and suggesions from seminar paricipans a he Universiy of nwerp Universiy of rizona Universiy of Miami MIT Universiy of orh arolina Ohio Sae Universiy Universiy of Rocheser Washingon Universiy and Yale Universiy are graefully acknowledged. I would also like o hank Richard Frankel Luzi Hail Peer Joos SP Kohari hrisian Leuz Shyam Sunder Sephen Taylor and Sco Whisenan for helpful suggesions.

2 . Inroducion This paper examines he empirical validiy of accouning qualiy measures derived from he Dechow and Dichev DD 00 accruals model. ccouning informaion qualiy is of cenral imporance o invesors sandards seers and oher sakeholders. While accouning qualiy is muli-dimensional see for example Schipper and Vincen 00 researchers have recenly focused on he DD discreionary accruals model as a cach-all proxy for firms accruals qualiy earnings qualiy accouning qualiy and overall informaion qualiy. umerous accouning sudies use qualiy measures derived from he DD model o es various economic hypoheses including he associaion beween informaion qualiy and cos of capial see for example Francis LaFond Olsson and Schipper 005 and Ecker Francis Kim Olsson and Schipper 006 and he associaion beween informaion qualiy and invesmen efficiency see for example Biddle and Hillary 006 and Raman Shivakumar and Tamayo 008 While he DD model has been quickly and widely adoped here is a pauciy of evidence on he analyical and empirical validiy of he model. I is unclear wheher he model reliably capures accouning qualiy of U.S. and inernaional firms. In his sudy I presen analyical and empirical ess of he DD regression model ha sugges ha: i he model has limied abiliy o disinguish beween manipulaed versus high qualiy ii he model ofen displays empirical properies indisinguishable from a random decomposiion of working capial accruals iii prior empirical validaions of he model have serious limiaions iv accouning qualiy measures Examples of papers ha use he DD model o capure accouning qualiy include boody Hughes and Liu 005 shbaugh and LaFond 00 Bharah Sunder and Sunder 008 Bhaacharya Desai and Venkaaraman 007 Bradshaw and Miller 005 heng Peng and Thomas 005 Doyle Ge and McVay 007 Ecker Francis Kim Olsson and Schipper 006 Ecker Francis Olsson and Schipper 005 Francis LaFond Olsson and Schipper 00 and 005 Francis anda and Olsson 008 Francis Olsson and Schipper 005 Myers Myers and Omer 00 Pae 005 Peng 005 Raman Shivakumar and Tamayo 008 and Thornon and Webser 00. Hribar and ichols 007 and Liu and Wysocki 007 examine firms operaing volailiy as a possible correlaed facor ha corrups accruals qualiy models.

3 derived from he DD model and is exensions show weak and ofen conradicory associaions wih oher measures of accouning qualiy for U.S. and inernaional firms. My analyical and empirical invesigaion of he DD model reveals fundamenal limiaions in he model. These findings are relevan because he model is widely used and becoming a de faco empirical benchmark for accouning qualiy. umerous sudies direcly use he model wihou limied prior validaion and hen draw arguably quesionable inferences abou difference in or he imporance of accouning qualiy for U.S. and inernaional firms. The srucure of he paper is as follows. In secion I dissec and analyically evaluae he DD model. My analyses sugges ha he DD model canno disinguish beween high qualiy accruals ha are associaed wih operaing cash flows and opporunisic discreionary accruals ha are also associaed wih operaing cash flows. I also demonsrae ha a srong negaive conemporaneous correlaion beween accruals and cash flow from operaions is consisen wih boh opporunisic discreionary earnings managemen as well as high qualiy accruals. Secion describes he U.S. and inernaional samples used o empirically es and benchmark he DD accruals model. In secion I use he U.S. daa sample o re-evaluae previous claims ha correlaions wih firms innae characerisics confirm ha he DD model capures accouning qualiy. I show ha firms innae characerisics have he same correlaions wih boh he absolue variabiliy of residuals from he DD and he absolue variabiliy of explained DD accruals. These findings conradic he claims of Dechow and Dichev 00 and Francis LaFond Olsson and Schipper 005 ha heir innae ess confirm ha he DD model capures high qualiy accruals. My evidence suggess ha he DD model randomly decomposes accruals raher han separaing high and low qualiy accruals

4 and he absolue variabiliy of accruals residuals from he DD model is a flawed measure of accruals qualiy. In secion 5 I invesigae he negaive conemporaneous relaion beween curren accruals and cash flows and examine is implicaions for earnings qualiy. Dechow 99 is among he firs o sysemaically documen a srong negaive correlaion beween firms accruals and cash flows. This accruals propery has generally been viewed as a means by which accruals increase he informaiveness of a firm s earnings see for example Dechow Kohari and Was 998. Using samples of U.S. and inernaional firms I find ha a firm s overall earnings qualiy declines when here is a sronger negaive correlaion beween is curren accruals and cash flows. This evidence suggess ha a srong negaive correlaion beween curren accruals and cash flows may disor earnings abiliy o capure rue firm performance see for example Skinner and Myers 007 and Leuz anda and Wysocki 00. Moreover I find ha in boh U.S. and inernaional samples he negaive conemporaneous associaion beween accruals and cash flows dominaes he DD model and may undermine is abiliy o capure accouning qualiy. Finally I benchmark he properies of he DD model relaive o previously-validaed accouning qualiy measures. I find ha he DD model exhibis weak and inconsisen relaions wih fundamenal earnings aribues for a broad sample of U.S. firms. The DD model also shows couner-inuiive associaions wih previously-validaed measures of earnings qualiy and counry-level insiuional facors relaed o high qualiy accouning numbers. The inabiliy of he DD model o correcly capure earnings/accruals qualiy for U.S. and inernaional firms appears o be driven by he fac ha he model srongly reflecs he conemporaneous negaive associaion beween accruals and cash flows. See also Hribar and ichols 007 and Liu and Wysocki 007.

5 To address he limiaions of he DD accruals model I sugges a simple modificaion o he original DD model o capure he incremenal associaion beween curren accruals and pas and fuure cash flows over and above he associaion beween curren accruals and curren cash flows. This modificaion removes he confounding effec of he srong negaive correlaion beween curren accruals and curren cash flows. The modified accruals model displays robus and inuiive associaions wih fundamenal earnings aribues associaed wih high earnings qualiy for U.S. firms. The modified model also displays correc and robus associaions wih counry-level earnings qualiy aribues and insiuional facors for 0 counries.. dissecion of he DD model In his secion I re-examine he heoreical underpinnings of he DD empirical regression model of working capial accruals. I idenify imporan limiaions of he empirical model and demonsrae ha i canno disinguish beween high qualiy accruals measuremen error and opporunisic discreionary accruals.. The baseline DD model Dechow and Dichev s 00 heoreical model of working capial accruals is simple and inuiive. They define he deerminans of working capial accruals in period as: = ε + ε - where capures working capial accruals in period. The cash flows erms j k are ne cash flows received paid in period j bu recognized in income in period k. The error erms ε + and 5

6 ε - are adjusmens for esimaion errors and heir correcions. If all cash flow componens and + are observable hen one can esimae he following regression model: = + * - + * * + + ε Dechow and Dichev 00 focus on he abiliy of his model o deec esimaion errors in accruals. They argue ha if here are no esimaion errors and no measuremen errors hen he esimaed coefficiens of model should be = = and =-. Moreover he regression R should be 00% and residual variance should be zero. I analyically confirm hese prediced properies in secion of ppendix. his poin i is imporan o highligh ha equaion can be viewed as a model of non-discreionary accruals driven by exogenously-classified cash flow ransacions. If he classificaions of he cash flow ransacions are exogenous hen he unexplained porion of working capial accruals in period should be relaed o eiher esimaion errors or discreionary and possibly opporunisic accrual choices or boh. In oher words he DD model should hypoheically capure he deerminans of non-discreionary accruals D as: D = f n analyical assessmen of he empirical DD model s Dechow and Dichev 00 noe i is infeasible o empirically esimae he heoreical version of heir model i.e. equaion. This limiaion arises because firms public financial 6

7 repors do no conain enough informaion o deermine he cash flows direcly relaed o working capial accruals i.e and +. Therefore one mus esimae regression model using subsiue cash flow measures ha are available. In general he empirical regression model will ake he form: = + * - + * + * + + ε where k is a cash flow measure for period k. Dechow and Dichev sugges ha suiable cash flow measures are cash flows from operaions. Therefore he empirical version of heir model is: = + *O - + *O + *O + + ε a where O k is cash flow from operaions in period k. s noed by Dechow and Dichev 00 he operaing cash flow erms in equaion a capure he consrucs in heoreical model i.e. equaion wih measuremen error. The measuremen error componens are highlighed in bold below: O - = O = O + = a- a- a- For example he measuremen error in curren cash flow from operaions O is he cash flow received and recognized in earnings in period ha does no generae working 7

8 capial accruals. The presence of hese measuremen errors will bias he regression coefficiens oward zero and reduce he regression R relaive o a regression model ha includes he correc cash flows direcly linked o working capial accrual ransacions. Dechow and Dichev use simulaions o demonsrae hese measuremen error effecs. I use analyical echniques as an alernae way o quanify he impac of he measuremen error on he regression resuls see secion of ppendix. The effecs of measuremen error on he empirical DD model are summarized in he able on he nex page see Empirical model: o discreionary accruals. The obvious problem wih he empirical DD model is ha he measuremen error problem reduces he abiliy of he model o correcly caegorize firms accouning qualiy. Furhermore i is possible ha he measuremen error may be correlaed wih oher firm characerisics such as firm risk which can lead o incorrec inferences abou he relaion beween accouning qualiy and a firm characerisic such as firm risk see for example Francis LaFond Olsson and Schipper 005. I now highligh a previously unrecognized and poenially more damaging problem wih he empirical DD model. This problem occurs when managers discreionary accrual choices are correlaed wih or even deermined by he measuremen error componens of firms operaing cash flows. In secions and 5 of ppendix I analyze a very likely scenarios where a manager uses opporunisic discreionary accruals D o manage or smooh earnings o offse shocks o curren period cash flows. Dechow and Dichev 00 assume ha curren period cash ransacions are merely measuremen error in he empirical regression. However a manager ha uses opporunisic discreionary accruals o offse shocks o curren cash flows will mechanically creae a negaive associaion beween and boh discreionary accruals D and oal 8

9 working capial accruals. Therefore if he DD model is used o measure a firm s accruals qualiy hen he following regression is esimaed: = D + D = + *O - + *O + *O + + ε 5 The analyical analyses in secions 5 and 6 of ppendix show ha he empirical DD model will assign a higher accouning qualiy raing o firms ha sysemaically engage in opporunisic earnings managemen and smoohing aciviies compared o firms ha do no engage in hese aciviies. summary f he analyical evaluaions are as follows: Regression propery Summary of nalyical Resuls from ppendix Theoreically- Empirical model: Empirical model: correc model o discreionary Earnings Empirical model: Earnings smoohing accruals managemen oefficien / / / oefficien - -/ - -5/6 oefficien / / / Regression R 00% 50% 77.8% 77.8% VarResiduals 0 ve 0 /ve 0 /ve 0 VarEarnings vx 0 vx 0 vx 0 5/6vx 0 These resuls show ha he DD model is quie effecive in explaining oal working capial accruals i.e. non-discreionary + discreionary accruals bu his is of lile use if one is rying o deec earnings managemen or assess accruals qualiy. Finally he simulaions in ppendix show ha a srong negaive correlaion beween conemporaneous accruals and cash flow from operaions may indicae sysemaic earnings managemen and low qualiy earnings see for example Leuz anda and Wysocki 00. In he following secions I furher analyze his problem by examining he empirical properies of he DD model. 9

10 . Sample Selecion and Descripive Saisics To complemen he analyical analyses in secion and ppendix I also benchmark he properies of he DD accruals qualiy model using wo samples. The firs sample described in secion. is comprised of 985 firm-year observaions for 70 U.S. firms beween 987 and 006. This sample is similar o he sample used by Dechow and Dichev 00. The second sample described in secion. is comprised 7908 firm-year observaions from 0 counries beween 99 and U.S. Sample Table summarizes he main sample used in he empirical analyses. The sample is obained from he ompusa indusrial and research files beween 987 and 006. Similar o Dechow and Dichev 00 I use cash flow from operaions from he Saemen of ash Flows repored under Financial ccouning Sandard o. 95 SFS o. 95 FSB 987. The sample is resriced o firms wih complee daa on asses earnings cash flow from operaions changes in accouns receivable and changes in invenory. Similar o Dechow and Dichev 00 I runcae he mos exreme % of cash flow from operaions earnings and changes in working capial and require a leas one year of pas and fuure cash flows and earnings associaed wih curren period working capial accruals. lso similar o Dechow and Dichev 00 I require a leas eigh years of daa o esimae firm-specific regressions. These resricions reduce he sample o 985 firm-year observaions for 70 firms. My final sample of firm-year observaions is larger han Dechow and Dichev 00 because he sample period exends o 006 bu he number of firm observaions is slighly smaller han Dechow and Dichev because of he addiional sock reurn daa requiremens for my daa sample. 0

11 ash flow from operaions O for U.S. firms is ompusa iem 08. Working capial accruals in year Wcc are compued as R + Inv - P - TP + Oher sses where R is accouns receivable Inv is invenory P is accouns payable and TP is axes payable in year. Specifically Wcc is calculaed from ompusa iems as W = -iem 0 + iem 0 + iem0 + iem 05+ iem 07. I calculae earnings afer shor-erm accruals bu before long-erm accruals as Earn = O + Wcc. ll variables are scaled by lagged oal asses. Panel of Table shows ha cash flows accruals and earnings numbers are similar o hose repored in Dechow and Dichev 00. The Pearson correlaions repored in Panel B are also consisen wih Dechow and Dichev. Specifically here is a srong posiive conemporaneous correlaion beween Earn and O 0.78 a posiive conemporaneous correlaion beween Earn and Wcc 0.0 and a negaive conemporaneous correlaion beween O and Wcc -0.. I also find an economically insignifican uncondiional correlaion beween curren accruals Wcc and pas and fuure cash flows from operaions O - and O +. This finding is also consisen wih he resuls in Dechow and Dichev.. Inernaional Sample Table summarizes he inernaional sample. The sample is obained from he Worldscope daabase which conains daa from annual repors of publicly-raded companies around he world. To be included in he sample a counry mus have a leas 50 firm-year observaions. firm-year observaion is included if here are a leas four consecuive years of income saemen and balance shee daa o calculae pas presen and fuure cash flow from operaions and working capial accruals. These four consecuive years of financial daa mus include oal sales operaing earnings changes in accouns receivable and changes in invenory. Similar o he U.S. sample I runcae he mos exreme % of firm-year observaions of scaled

12 cash flow from operaions earnings and changes in working capial wihin each counry. ll observaions are scaled by lagged sales raher han asses because of he large variaion in asse inensiy and asse recogniion rules and choices wihin and across firms in his inernaional sample. These differences in repored asses can poenially lead o spurious resuls if one uses variables scaled by oal asses see for example Wysocki 00. These daa requiremens leave a sample of 7908 firm-year observaions across 0 counries. Operaing income Earn is used o measure earnings o avoid capuring sysemaic differences in oher non-operaing income across firms and counries. I calculae working capial accruals Wcc and cash flow from operaions O using he balance shee mehod because mos inernaional firms do no consisenly repor cash flow saemens. Following Dechow e al. 995 I calculae changes in non-cash working capial as Wcc = Δ -Δash - ΔL - ΔSTD - ΔTP. The variables are defined as: Δ = change in oal curren asses Δash = change in cash/cash equivalens Δ = change in oal curren liabiliies ΔSTD = change in shor-erm deb included in curren liabiliies ΔTP = change in income axes payable in year. If a firm does no repor informaion on axes payable or shor-erm deb hen he change in boh variables is assumed o be zero. Similar o Leuz anda and Wysocki 00 I approximae cash flow from operaions O for he sample of inernaional firms as he difference beween operaing income Earn and working capial accruals Wcc in year. The daa in Panel of Table shows cash flow and earnings numbers ha slighly smaller in magniude han hose repored in Table for he U.S. sample. On he oher hand working capial accruals for inernaional firms are on average slighly larger han he number repored by U.S. firms. The mean Pearson correlaions across counries repored in Panel B are also similar o he numbers repored for U.S. firms. noable difference is ha he

13 conemporaneous correlaion beween curren earnings Earn and cash flow from operaions O is only 0.7 for he inernaional sample whereas i is 0.7 for he U.S. sample. On he oher hand he persisence of repored earnings correarnearn+=0.680 is acually higher for he inernaional firms han U.S. firms which have a lower persisence number of This difference appears o be explained by greaer earnings smoohing aciviies using accruals for firms in he inernaional sample. In paricular he conemporaneous correlaion beween working capial accruals Wcc and cash flow from operaions O has a large mean value of in he inernaional sample while he U.S. sample has a negaive correlaion of -0.. This is consisen wih he earnings managemen evidence in Leuz anda and Wysocki 00 ha suggess ha U.S. firms smooh earnings less han oher firms around he world.. Do empirical ess of firms innae characerisics validae he DD model? Prior research aemps o show ha DD model capure earnings qualiy by correlaing he variabiliy of he model residuals wih firms innae characerisics. For example Dechow and Dichev 00 predic and es he correlaions beween sresid he firm-specific sandard deviaion of he residuals from he DD model and specific firm characerisics. In paricular hey predic ha if sresid capures accouning qualiy hen i should be associaed wih he lengh of he firm s operaing cycle + firm size - sales volailiy + cash flow volailiy + working capial accrual volailiy + earnings volailiy + loss frequency + and he magniude of accruals +. Dechow and Dichev 00 find empirical evidence supporing hese predicions and hey herefore claim ha i confirms ha he DD model capures firms accouning qualiy. Francis LaFond Olsson and Schipper 005 go furher and claim ha a regression of sresid on hese facors capures a firm s innae accouning qualiy.

14 I is quesionable wheher correlaing hese firm characerisics wih he absolue variaion of residuals from he DD model ells us anyhing abou accruals qualiy. For example Mcichols 00 noes ha relaing he absolue variaion in accrual residuals o he absolue magniude of accruals and o he sandard deviaion of sales cash flows accruals and earnings are likely o be induced mechanically and ha variabiliy in he residuals is increasing in accruals which in urn is correlaed wih variabiliy in sales cash flows and earnings p. 6. Mcichols 00 also suggess ha he correlaion resuls would be more readily inerpreable if he relaive variance of he errors i.e. he regression R were sysemaically correlaed wih firm characerisics. In addiion o Mcichols criique I claim ha he correlaion beween sresid and various firm characerisics provides only parial and circumsanial evidence ha sresid is a valid accouning qualiy proxy. To demonsrae his I will momenarily accep he asserion ha he DD model correcly decomposes working capial accruals ino a low qualiy componen i.e. he regression residual and a high qualiy componen i.e. he model prediced value of accruals. If his decomposiion is correc hen he firm characerisics lised above should no only be correlaed wih sresid bu hey should be consisenly correlaed wih he variabiliy of accruals explained by he DD model. Specifically he variabiliy of explained accruals from he DD model should have he opposie correlaions wih innae firm characerisics compared o he prediced correlaions wih sresid. To illusrae his poin consider he simple case where all firms have he same variabiliy of oal working capial accruals. By definiion he variance of oal working capial accruals equals he variance of explained accruals from he DD model plus he variance of residuals from he DD model. I immediaely follows ha he cross-secional On he oher hand Gu 00 discusses he problems wih making comparisons of regression R s across samples.

15 correlaion beween a firm characerisic and he variance of residual accruals will be mirrored by an opposie correlaion beween he firm characerisic and he variance of explained accruals. I es hese exended cross-secional predicions using correlaion ess similar o hose presened in Dechow and Dichev 00 and Francis LaFond Olsson and Schipper 005. Panel of Table summarizes he sample descripive saisics of he eigh innae firm characerisics. The descripive saisics in my sample are very similar o hose repored in Dechow and Dichev 00. In Panel B I replicae he simple correlaions beween sresid and he series of innae firm characerisics. The firs row of Panel B shows ha correlaions beween sresid and he eigh firm characerisics are very similar o hose prediced and documened by Dechow and Dichev 00. I hen compue he correlaions beween sdexpl he sandard deviaion of explained accruals from he DD model and he same firm characerisics. Ineresingly he correlaions are almos exacly he same as hose found for sresid raher han he opposie which would be expeced if he DD model decomposed accruals ino high qualiy and low qualiy componens. These similar associaions are also observed Panel which presens muliple regression ess for boh sresid and sdexpl. These findings conradic he conenion of Dechow and Dichev 00 and Francis LaFond Olsson and Schipper 005 ha heir correlaion ess direcly confirm ha he DD model idenifies high qualiy accruals. s a specific example I analyze he logic behind he prediced correlaions beween firm size sresid and sdexpl. Firs if smaller firms are expeced o have lower accouning qualiy hen Dechow and Dichev predic and find larger absolue variabiliy of accruals residuals for hese small firms. Second if he exac same logic is applied o he explained par of accruals hen hese smaller firms should also exhibi lower absolue variabiliy in he explained accruals from he DD model. The empirical evidence does no o suppor his predicion. I should be 5

16 noed ha if you have some doubs abou he logic of he second saemen lised above hen should have exacly he same doubs abou he firs saemen lised above. In conclusion we observe exacly he same associaions beween he firm characerisics and absolue variabiliy of explained and residual accruals. This conradicory evidence forces us o conclude ha eiher: i he DD model randomly decompose accruals raher han separaing high and low qualiy accruals or ii correlaion ess beween innae firm characerisics and he absolue variabiliy of residual accruals are uninformaive abou he validiy of he DD model or iii he absolue variabiliy of accrual residuals is an inappropriae measure of accruals qualiy or iv all of he above. 5. Empirical benchmarking of he DD model In he secion I benchmark he properies of he DD model for U.S. and inernaional firms. I specifically focus on he conemporaneous correlaion beween accruals and cash flows as major facor ha affecs he usefulness of he DD accruals model. 5. nalyzing he conemporaneous associaion of accruals and cash flows The negaive conemporaneous correlaion beween cash flows and accruals is a widelyobserved propery of accouning numbers see for example Dechow 99. In his sub-secion I examine he negaive correlaion beween a firm s curren accruals and cash flows orro Wcc where Wcc is he change in non-cash working capial i.e. working capial accruals and O is cash flow from operaions in period. There is an on-going debae wheher his negaive correlaion helps improve or undermines he usefulness of earnings as a measure of firm performance. On one hand prior 6

17 research argues ha he negaive correlaion beween accruals and cash flows reflecs he smoohing of emporary componens of cash flows and herefore improves earnings qualiy. This smoohing arises from he iming and maching principles of an accrual accouning sysem where accruals record real economic ransacions see for example Dechow 99 and Dechow Kohari and Was 998. On he oher hand smoohing can also arise from managers opporunisically using accruals o misrepresen firm performance and risk. Therefore a srong negaive correlaion beween cash flows and accruals can reflec earnings managemen see for example Myers and Skinner 007 Leuz anda and Wysocki 00 and Burgsahler Hail and Leuz 006. In fac Leuz anda and Wysocki 00 argue ha if a ypical accouning sysem underreacs o economic shocks and if managers use accruals o signal performance hen his would induce a less negaive or even posiive correlaion beween cash flows and accrual. These compeing explanaions for he correlaion beween cash flows and accruals have polar opposie implicaions for earnings qualiy. Ironically prior research direcly or indirecly uses he correlaion beween cash flows and accruals as a proxy for earnings qualiy. However here is lile sysemaic evidence wheher his correlaion acually capures high or low earnings qualiy. This paper aemps o fill his void by benchmarking he correlaion beween cash flows and accruals agains hree common measures of earnings qualiy. The firs earnings qualiy benchmark is earnings persisence Persis. I is calculaed as he firm-specific Pearson correlaion beween curren and fuure earnings Earn and Earn +. Dechow and Dichev 00 also use earnings persisence as a measure of earnings qualiy o assess he validiy of heir original accruals model. The second earnings propery is he abiliy of curren repored earnings o predic fuure cash flow from operaions Predic. If invesors ulimaely value fuure cash flows hen he predicive abiliy of curren earnings for fuure cash flows is an imporan 7

18 earnings qualiy meric for invesors see for example Finger 99; and Barh ram and elson 00. The predicive measure is calculaed as he firm-specific Pearson correlaion beween curren earnings Earn and fuure cash flow from operaions O +. The hird earnings qualiy measure ER capures he associaion beween curren repored annual earnings Earn and conemporaneous annual sock reurns Re. Firms annual sock reurns are calculaed over he same fiscal period as heir earnings and are inclusive of dividends paid during he year. 5. decomposiion of he empirical DD model The empirical version of he DD model equaion a aemps o capure he exen o which working capial accruals map ino operaing cash flow realizaions. DD summarize accruals qualiy as he inverse of sresid he sandard deviaion of he residuals from model a. Higher values of sresid indicae lower earnings qualiy because less of he variaion in curren accruals is explained by pas presen and fuure cash flow from operaions. I should be noed ha an alernae way o quanify accruals qualiy from he DD model is o measure he explanaory power djused R of regression model a see for example Mcichols 00. In he DD framework a high explanaory power of model a would indicae high accruals qualiy. very desirable feaure of he DD accruals qualiy model is ha i capures he associaion beween curren accruals and pas and fuure cash flows. If curren accruals ac o recognize real realized cash ransacions from prior or subsequen periods hen one should observe a posiive associaion beween curren accruals and pas and fuure period cash flow from operaions. These wo aribues of accruals had no been previously modeled in he 8

19 lieraure Dechow and Dichev 00. However he DD model also capures he negaive conemporaneous associaion beween accruals and cash flows. The negaive conemporaneous associaion beween accruals and cash flows is problemaic for he DD model because i can simply reflec firms opporunisic earnings smoohing aciviies he anihesis of qualiy. To invesigae he poenial confounding effec arising from he conemporaneous associaion beween accruals and cash flows I dissec he DD accruals model by comparing he incremenal explanaory power of pas and fuure cash flows over and above he explanaory power of curren cash flows. This dissecion provides insighs ino he properies of he individual cash flow componens of he DD model. I firs examine he basic conemporaneous associaion beween working capial accruals and cash flows from operaions. I esimae a resriced version of he DD model as follows: = b 0 + b O + b Regression model b capures he conemporaneous associaion beween accruals and cash flows in isolaion. Prior research documens a srong negaive relaion beween conemporaneous accruals and cash flows see for example Dechow 99 and Dechow Kohari and Was 998. nice feaure of his simple regression model is ha here is one-oone link beween square of he slope coefficien and he model s coefficien of deerminaion R. sronger negaive associaion beween accruals and cash flows implies a higher model R and vice versa. To allow a comparison beween regression models a and b I examine boh he adjused R and he sandard deviaion of he residuals sresid of he regression models. These measures can be used o deermine wha fracion of he explanaory power of he DD accruals 9

20 model i.e. model a is caused by he conemporaneous negaive associaion beween accruals and cash flows i.e. capured by he resriced model b. This is imporan because he conemporaneous negaive associaion beween accruals and cash flows may capure opporunisic earnings smoohing raher han fundamenal accruals qualiy. greaer explanaory power of curren cash flows for curren accruals would increase he overall explanaory power of he DD accruals model. However his high explanaory power is unlikely o reflec fundamenal accruals qualiy and herefore inhibis he abiliy of he DD accruals qualiy model o ruly capure earnings and accruals qualiy. If curren cash flow componen of he DD model eiher adds noise or reflecs poor accruals qualiy hen i would be advanageous o remove he effec of curren cash flows from he regression model. However he simple exclusion of curren cash flows from he DD regression model is problemaic. In paricular Dechow and Dichev 00 explain and show ha curren working capial accruals have a very weak uncondiional associaion wih pas and fuure cash flows. To verify he weak uncondiional associaion I esimae a regression ha only includes pas and fuure cash flow from operaions as deerminans of curren accruals: = b 0 + b O - + b O + + c Even hough pas and fuure cash flows are vially imporan for assessing accruals qualiy i is expeced ha regression model c will have weak explanaory power for curren working capial accruals. The weak uncondiional relaion beween accruals and pas and fuure cash flows suggess he necessiy o condiion on curren cash flow realizaions o capure he oher componens ha ruly reflec accruals qualiy. In fac he full DD model aemps o do his 0

21 by including pas fuure and curren cash flows as explanaory variables see regression model a. 5.. Modified measures of accruals qualiy In his secion I presen a simple soluion o he limiaions of he DD accruals qualiy model. On one hand i is desirable o remove he confounding effecs of curren cash flows in regression model a. On he oher hand i is necessary o condiion on curren cash flows o overcome he weak uncondiional associaion beween curren accruals and pas and fuure cash flows in model c. simple soluion o his problem is o examine he incremenal explanaory power of pas and presen cash flows over and above he explanaory power of curren cash flows for curren accruals. This can be easily measured by comparing he explanaory power of regression model a wih he explanaory power of resriced regression model b. Dechow and Dichev 00 show ha curren accruals are significanly associaed wih pas and fuure cash flows in regression model a. However hey do no invesigae how much of he variaion in accruals is explained by pas and fuure cash flows incremenal o ha of curren cash flows. I ouline hree possible merics o capure he incremenal explanaory power of pas and fuure cash flows over ha of curren cash flows: ccruals Qualiy Meric = djused R[Model a] djused R[Model b] ccruals Qualiy Meric = djused R[Model a] / djused R[Model b] ccruals Qualiy Meric = sresid[model b] / sresid/[model a] where regression models a and b are esimaed for he same sample of observaions and sresid is he sandard deviaion of he regression residuals see Dechow and Dichev 00. These hree relaed merics capure accruals qualiy hrough he incremenal relaive explanaory power of pas curren and fuure accruals compared o he explanaory power of curren cash

22 flows in isolaion. In all cases a higher value of a meric is inerpreed as indicaing a sronger relaive associaion beween curren accruals and pas and fuure cash flows i.e. higher accruals qualiy. 5.. Empirical resuls In his secion I analyze he relaion beween he correlaion of cash flows and accruals and benchmark measures of earning qualiy including earnings persisence he predicive abiliy of earnings for fuure cash flows and he conemporaneous associaion beween earnings and sock reurns. The goal of his analysis is o deermine if a srong negaive conemporaneous correlaion beween cash flows and accruals is associaed wih high or low earnings qualiy. I hen apply a similar benchmarking exercise o he original DD accruals qualiy model and for hree new modified accruals qualiy merics ha remove he confounding effec of firms earnings smoohing aciviies ha poenially corrup he DD model see secion 5.. ll accruals merics are benchmarked agains widely-used and previously-validaed variables ha capure accouning qualiy or facors ha influence accouning qualiy. 5.. ssociaion beween corrcash flows accruals and earnings qualiy measures Table presens he associaion beween hree common measures of earnings qualiy for U.S. firms and he conemporaneous correlaion of cash flows and accruals. The hree firmspecific measures of earnings qualiy are Persis Predic and ER. higher value of hese measures is suggesive of higher earnings qualiy. If a sronger negaive correlaion beween cash flows and accruals i.e. corro Wcc improves earnings qualiy hen i should be negaively relaed o he hree earnings qualiy measures. Ineresingly I find ha corro Wcc is unrelaed o earnings persisence. Moreover corro Wcc has a significan

23 posiive associaion wih he predicive abiliy of curren earnings for fuure cash flows Predic and wih ER. These findings sugges ha a sronger negaive correlaion beween cash flows and accruals i.e. smoohing leads o lower earnings qualiy. The findings for his sample of 70 U.S firms are consisen wih he inernaional resuls of Leuz anda and Wysocki 00. They also presen evidence suggesing ha a sronger negaive correlaion beween cash flows and accruals is associaed wih higher levels of earnings managemen and lower earnings qualiy. 5.. ccrual qualiy merics for U.S. sample The evidence in Table suggess ha a srong negaive correlaion beween cash flows and accruals can reduce earnings qualiy. Therefore I nex urn o an evaluaion of he DD model of accruals/earnings qualiy which capures he negaive correlaion beween cash flows and accruals. Table 5 presens he esimaes of regression models based on he original DD model and resriced versions of he DD model. Panel summarizes he pooled regression resuls for he 985 firm-year observaions and Panel B summaries he average firm-specific resuls esimaed for each of he 70 firms in he U.S. sample. In Panel he firs row summarizes he pooled esimaion of he original DD accruals model model a where curren working capial accruals are regressed on pas presen and fuure cash flow from operaions. The coefficiens and saisical significance of he regression model are very similar o hose repored in Dechow and Dichev 00. The coefficiens on pas presen and fuure cash flows are and 0.0 respecively. The overall explanaory power of he model as capured by he adjused R is 0.5% and he sandard deviaion of he residuals sresid is I should be noed ha he adjused R and sresid are by consrucion

24 negaively-relaed wih each oher. In row of Panel a resriced version of he original DD model model b is esimaed where he only deerminan of working capial accruals is conemporaneous cash flow from operaions. I can be seen ha his simple model has srong explanaory power wih an adjused R of 9.7% and sresid is This resriced model suggess ha close o wo-hirds of he explanaory power of he full DD model i.e. model a is explained by he negaive conemporaneous correlaion beween accruals and cash flows. This can be problemaic if his negaive conemporaneous correlaion reflecs firms use of accruals o opporunisically smooh repored earnings. In paricular DD argue ha a high explanaory power which equaes o a low value of sresid implies high accruals qualiy. On he oher hand he high explanaory power of he DD model may be largely aribuable o firms opporunisic earnings smoohing aciviies which are he anihesis of high qualiy earnings. To remove he poenially confounding effecs of earnings smoohing from he DD model I esimae a reduced-form regression model ha only includes pas and fuure cash flows as deerminans of working capial accruals. This model is presened in row of Table 5. Similar o he findings of Dechow and Dichev 00 here is essenially no associaion beween curren accruals and pas and fuure cash flows. This confirms he need o capure he incremenal relaive explanaory power of pas and fuure cash flows over and above he explanaory power of curren cash flows. For descripive purposes I quanify he average incremenal explanaory power of pas and fuure cash flows by comparing he adjused R from he full DD accruals regression model a wih he adjused R from he resriced accruals regression model b. For he overall pooled sample of 985 observaions he adjused R increases from 9.7% o 0.5% when pas and fuure cash flows are added o he basic model of ha only includes

25 conemporaneous cash flows O as a deerminan of working capial accruals Wcc. Based on he F-es he 0.8% increase in he adjused R is saisically significan. I re-esimae he accruals models using firm-specific regressions for he sample of 70 firms are summarized in Panel B of Table 5. The regression resuls for models a b and c are summarized in rows - respecively of Panel B. The resuls parallel he pooled regressions findings wih slighly higher average explanaory power for boh models a and b. The resriced version of he DD model i.e. model b explains an average of 7.% of he variaion in firms working capial accruals compared wih 5.6% for he unresriced model i.e. model a. Given ha over a majoriy of he explanaory power of model a is explained by curren cash flows his again raises he concern ha much of he explanaory power of he full DD accruals model is aribuable o firms smoohing aciviies. If his negaive conemporaneous correlaion beween accruals and cash flows reflecs opporunisic earnings smoohing hen he DD model will be a poor measure of accruals and earnings qualiy. On he oher hand a meric ha capures he incremenal explanaory power of pas and fuure cash flows for accruals does no suffer from his problem. For he firm-specific regressions he average incremenal explanaory power of pas and fuure cash flows over and above curren cash flows is 8.%. 5.. Benchmarking accruals qualiy models for U.S. sample In he prior subsecion I demonsrae ha a sizable fracion of he explanaory power of he original DD accruals model is aribuable o he negaive conemporaneous associaion beween accruals and cash flow from operaions. If his conemporaneous componen does no capure accruals qualiy hen i can handicap he DD model in capuring accruals qualiy. In secion 5. I ouline hree simple alernae merics of accruals qualiy based on he incremenal 5

26 relaive explanaory power of pas and fuure for curren accruals. In his sub-secion I compare how he original DD model and he hree alernae merics relae o fundamenal earnings properies ha are generally associaed wih high qualiy earnings for U.S. firms. I use hree fundamenal earnings properies Persis Predic and ER as benchmarks for earnings qualiy. Table 6 presens he correlaions beween he DD accruals measures he hree modified accruals qualiy merics and he hree fundamenal earnings properies for he 656 firms in he sample. The original DD merics are summarized by -*sresid he negaive of he sandard deviaion of DD regression residuals and he adjused R from he DD regression model. The hree modified merics capure he incremenal relaive explanaory power of pas and fuure cash flows as he eiher he difference or raio of adjused R s beween regression models a and b or he raio of he sresid from model b o he sresid from model a. In all cases a higher value of a meric is inerpreed as higher accruals qualiy. I firs invesigae wheher he original DD accruals qualiy measures are associaed wih he hree fundamenal earnings aribues of Persis Predic and ER. The key findings are presened in he lower lef corner of he correlaion marix. The original DD accruals qualiy measures exhibi only a weak posiive associaion wih he fundamenal earnings aribues. Boh -*sresid and he dj R from he original DD model have a saisically significan correlaion wih earnings persisence which is consisen wih he findings documened in Dechow and Dichev 00. However Predic and ER show essenially insignifican associaions wih he original DD merics. This resul is no ha surprising given he srong associaion beween he original DD merics and he correlaion beween cash flows and accruals corrowcc. I appears ha he original DD merics are essenially capures he smoohing of cash flows wih accruals. 6

27 I nex proceed o validae wheher he hree modified accruals merics are beer able o capure firms earnings qualiy. gain I compare he modified accruals merics wih he hree fundamenal earnings qualiy aribues of Persis Predic and ER. These associaions are repored in he lower righ corner of he correlaion marix in Table 6. In all cases he new modified merics have a srong posiive and significan associaion wih each of he fundamenal earnings qualiy aribues. This suggess ha he hree incremenal relaive explanaory power merics are srongly and posiively relaed o measures of high qualiy earnings. This evidence suggess ha he measures from original DD accruals qualiy model are poor proxies for accruals and earnings qualiy for U.S. firms. Moreover he modified accruals qualiy merics based on he incremenal relaive explanaory power of pas and fuure cash flows for accruals have a very srong posiive associaion wih radiional measures of fundamenal earnings qualiy. 5.. ccrual qualiy merics for inernaional sample Table 7 presens he esimaes of accruals regressions for inernaional firms based on he original DD model and resriced versions of he DD model. Panel summarizes he pooled firm-year regression resuls and Panel B summaries he mean counry-specific resuls esimaed for each of he 0 counries. In Panel he firs row summarizes he esimaion of he original DD pooled regression model i.e. model a where curren working capial accruals are regressed on pas presen and fuure cash flow from operaions. The resuls of he regression model are similar o he U.S. resuls repored in Table 5. The overall model explanaory power as capured by he adjused R is 6.% which is slighly higher han he U.S. sample. I should be noed ha according o he DD model a higher overall explanaory power or lower sresid from model a implies higher 7

28 earnings qualiy. The higher explanaory power for he inernaional sample suggess ha ha he original DD model may no ruly capure accruals qualiy because U.S. firms are generally considered o have beer earnings qualiy han firms form oher counries in he world see for example Hung 00 Leuz anda and Wysocki 00 and Bhaacharya Daouk and Welker 00. To check for he possible confounding effec of earnings smoohing a resriced version of he original DD model is esimaed in row of Panel. This esimae of model b includes conemporaneous cash flow from operaions as he sole deerminan of working capial accruals. I can be seen ha his simple model has srong explanaory power wih an adjused R of 9.6%. The esimaes of he resriced model sugges ha over half of he explanaory power of he full DD model i.e. model a is explained by he negaive conemporaneous correlaion beween accruals and cash flows in his inernaional sample. If his conemporaneous correlaion capures poor-qualiy accruals in he DD model hen his would explain why he original DD model has higher explanaory power for inernaional firms compared o U.S. firms. These empirical resuls again sugges ha he anomalously-high explanaory power of DD model for inernaional firms may be aribuable o widespread earnings smoohing aciviies in counries ouside he U.S. To remove he poenially confounding effecs of earnings smoohing from he DD model I examine he average incremenal explanaory power of pas and fuure cash flows for accruals. This is calculaed by comparing he adjused R from he full DD accruals regression model a wih he adjused R from resriced accruals model model b. For he overall pooled sample of 7908 inernaional firm-years he adjused R increases from 9.6% o 6.%. Based on he F-es his 6.7% increase is saisically significan. To deermine if he pooled regression resuls are robus I re-esimae he accruals models using counry-specific regressions for he sample of 0 counries. These resuls are summarized in Panel B of Table 7. 8

29 The 0 counry regressions are separae pooled regressions of firm-year observaions wihin each counry. The average regression resuls for models a b and c are summarized in rows - respecively of Panel B. The resuls parallel he regressions findings presened in Panel. The resriced version of he DD model i.e. model b explains an average of 9.6% of he variaion in firms curren working capial accruals wihin each counry. On he oher hand he incremenal explanaory power of pas and fuure cash flows is an average of.% higher for all counries in he sample Benchmarking accruals qualiy models for inernaional sample In his sub-secion I examine he how he qualiy measures from he original DD model compare wih previously-validaed counry-level earnings managemen and opaciy measures and counry-level insiuional facors associaed wih earnings qualiy. I also examine he associaion of he hree modified accruals qualiy merics wih he same counry-level earnings managemen/opaciy measures and counry-level insiuional facors. The counry-level earnings managemen and opaciy merics are obained from Leuz anda and Wysocki 00 and Bhaacharya Daouk and Welker 00. The firs earnings managemen measure ggregae EM Score is he overall earnings managemen score abulaed from four earnings managemen measures from Leuz anda and Wysocki 00. This overall score capures earnings smoohing relaive o cash flows he use of accruals o offse cash flow shocks he exen of accruals use and exen of loss avoidance. Given ha he DD accruals qualiy measure may unknowingly capure earnings smoohing I also use he Leuz anda and Wysocki 00 measure of smoohing as he second earnings managemen measure. This measure is labeled EM Smoohing and is he inernaional ranking of he spearman correlaion 9

30 beween accruals and cash flows for firms wihin each counry. The hird earnings managemen measure ccruals is he average raio of absolue working capial accruals o absolue cash flow from operaions for each counry from Leuz anda and Wysocki 00. Finally he Earnings Opaciy meric is abulaed from Bhaacharya Daouk and Welker 00. The four counry-level insiuional facors relaing o earnings qualiy are also abulaed from previous inernaional sudies. Prior research has shown ha hese facors relae o overall earnings qualiy see for example Hung 00 Leuz anda and Wysocki 00 and Wysocki 00. The sudies show ha he four insiuional variables namely high qualiy accouning sandards beer disclosure rules srong invesor proecion and effecive legal enforcemen reduce firms incenives and abiliy o disor repored earnings. The firs counry-level insiuional variable is ccouning Sandards. This variable measures he inclusion or omission of 90 iems from firms annual repors and is abulaed for a large number of counries in La Pora e al 998. The second variable Filing Disclosure capures a counry s mandaed sock exchange disclosure filing requiremens abulaed by La Pora e al 00. The hird and fourh insiuional variables are Invesor Righs and Legal Enforcemen. Boh of hese variables reflec he laws and enforcemen of laws ha proec he righs of ouside invesors. Invesor Righs is he anidirecor righs index abulaed by La Pora e al 998 and Legal Enforcemen is he mean score of hree legal variables also abulaed by La Pora e al 998. Table 8 presens he correlaions beween he original DD accruals measures he hree modified accruals qualiy merics he counry-level earnings qualiy proxies and he insiuional facors for he 0 counries in he sample. gain he modified merics capure he incremenal relaive explanaory power for accruals as he difference or raio of he adjused R of regression models a and b or he raio of he sresid from model 9 o he sresid from model a. The 0