Sami Keskek of Texas A&M University will present. Does market learning explain the disappearance of the accrual anomaly?

Transcription

1 Disinguished Lecure Series School of Accounancy W. P. Carey School of Business Arizona Sae Universiy Sami Keskek of Texas A&M Universiy will presen Does marke learning explain he disappearance of he accrual anomaly? on February 25, 2 :3pm in BA 365

2 Does marke learning explain he disappearance of he accrual anomaly? Sami Keskek Texas A&M Universiy Absrac: In his sudy, I invesigae wheher marke learning explains he absence of he accrual anomaly in recen years by examining hree condiions associaed wih he presence of he anomaly in prior research: (i) a differenial relaion beween fuure earnings and cash flows versus accruals, (ii) incorrec weighing of cash flows and accruals by invesors when predicing earnings, and (iii) associaion of earnings forecas errors wih reurns. All of hese condiions are widely documened in he anomaly period. In he no-anomaly period, I coninue o find a differenial relaion of cash flows and accruals wih fuure earnings. I find, however, ha invesors appear o correcly weigh accruals and cash flows in heir earnings predicions implici in beginning-of-year securiy prices, consisen wih learning. I also invesigae wheher improvemens in analys forecass conribue o invesor learning and he absence of he anomaly. The associaion beween analys opimism and accruals is weaker in he no-anomaly period, bu is sill saisically significan. Furhermore, I find ha he anomaly ended simulaneously for firms followed by analyss and for non-followed firms, suggesing ha improvemens in analys forecass alone canno accoun for improved marke efficiency wih respec o accruals. I find ha he anomaly was similar for firms held by insiuional invesors and for firms wih no insiuional holdings before he discovery of he anomaly while he anomaly ended sooner for held firms han for non-held firms afer he discovery of he anomaly, consisen wih he conjecure ha arbirage by insiuional invesors may reduce he anomaly. Overall, he findings are consisen wih marke learning and sugges ha improvemen in invesors inerpreaion of accruals afer he discovery of he anomaly explains he absence of he anomaly. This improvemen in invesor learning is no due o changes in analyss forecasing behavior, however. Keywords: Marke efficiency, accrual anomaly, invesor sophisicaion, analyss forecas bias Daa Availabiliy: Daa are available from public sources. Conac informaion: Sami Keskek, Deparmen of Accouning, Mays Business School, Texas A&M Universiy, College Saion, TX ; skeskek@mays.amu.edu. Acknowledgemens: I would like o hank my commiee chair, Dr. Senyo Tse, for coninuous suppor and guidance. I also wan o hank my oher commiee members, Dr. Anwer Ahmed, Dr. Chris Wolfe, and Dr. Sco Lee for helpful commens and valuable discussions. In addiion, I received useful commens from Dr. Lynn Rees. Preliminary and incomplee.

3 I. Inroducion: Recen evidence suggess ha invesors accrual-relaed mispricing disappeared in 22 and has been undeecable since hen (e.g., Richardson e al. 2). In his sudy, I invesigae wheher improvemens in invesors inerpreaion of accruals explain he absence of mispricing in recen years. I build on Richardson e al. s (2) conjecure ha adapive marke efficiency (e.g., Grossman and Sigliz 98), or marke learning, migh explain he end of he anomaly. Numerous sudies sugges ha analyss are sophisicaed informaion processors and improve marke efficiency by providing value-relevan informaion o invesors (Ramnah 28). Oher sudies sugges, however, ha analyss also overweigh accruals when hey issue forecass, resuling in forecas errors ha are predicably associaed wih accruals. I herefore also examine wheher reducions in analyss accrual-relaed bias conribue o he end of he anomaly. Efficien use of accouning informaion is of fundamenal imporance o invesors and firms as well as regulaors. I is useful o documen wheher marke learning can resore efficiency when he marke is discovered o be inefficien wih respec o paricular informaion. Since analyss are imporan informaion inermediaries and heir forecass exer considerable influence on securiy prices, heir possible conribuion o marke learning is of ineres o invesors and regulaors. Sloan (996) finds ha a rading sraegy ha akes a long posiion in low-accrual firms and a shor posiion in high-accrual firms earns economically large and saisically significan hedge reurns, a phenomenon referred o as he accrual anomaly. While a number of sudies argue ha he excess reurns o he rading sraegy can be explained by risk-relaed facors (Khan 28; Wu e al. 29), he evidence suggess ha he excess reurns reflec marke inefficiency and are driven by invesors failure o anicipae he lower persisence of accruals Noe ha Richardson e al. (2) do no es his predicion.

4 relaive o cash flows (e.g., Collins and Hribar 2; Richardson e al. 2; Xie 2; Beneish and Vagus 22; Hirshleifer e al. 24; Hirshleifer e al. 26; Richardson e al. 2). More generally, he anomaly is associaed wih he following hree condiions: (i) a differenial relaion beween fuure earnings and cash flows versus accruals, (ii) incorrec weighing of accruals by invesors, leading o earnings forecas errors ha are predicably associaed wih accruals, and (iii) associaion of earnings forecas errors wih reurns. The recen evidence shows ha he anomaly disappeared in 22 (e.g., Richardson e al. 2). The absence of he anomaly provides a unique opporuniy o undersand he causes of he anomaly. Mos imporanly, he absence of he anomaly implies ha a leas one of he condiions associaed wih he anomaly changed. Since changes in hese condiions lead o disinc inferences abou rends in he qualiy and value-relevance of accouning informaion or invesor sophisicaion in inerpreing accouning informaion, i is of fundamenal imporance o undersand he conribuion of each facor o he absence of he anomaly. Richardson e al. (2) conjecure ha he end of he anomaly could be driven by marke learning, suggesing ha invesors learn abou he relaively low associaion of accruals wih fuure earnings over ime and adjus accordingly. Since he absence of he anomaly coincides wih he enacmen of he Sarbanes Oxley Ac of 22 (hereafer, SOX), i seems plausible ha he accouning scandals ha led o he enacmen of SOX and he regulaory changes in SOX may have led invesors o pay more aenion o he implicaions of cash flows and accruals for fuure earnings. Oher sudies sugges ha he relaion beween accruals and fuure earnings changed afer he enacmen of he Sarbanes Oxley Ac, and his change, raher han invesor learning, may explain he absence of he anomaly (e.g., Bhojraj e al. 29). For example, Cohen e al. (28) find a subsanial decrease in he accruals-based earnings managemen 2

5 following SOX. The coincidence of he absence of he anomaly wih he enacmen of SOX also raises he possibiliy ha he absence of he anomaly could be due o a loss of invesor confidence in he financial informaion provided by firms. If invesors rely less on informaion in earnings, he associaion of earnings forecas errors wih reurns may be oo low o generae excess reurns. Since analyss play an imporan role as informaion inermediaries in he capial marke, improvemens in analyss forecas properies, or analys learning, could conribue o he absence of he anomaly. Consisen wih he view ha analyss are sophisicaed informaion inermediaries, numerous sudies sugges ha analyss undersand he implicaions of financial informaion beer han invesors and improve marke efficiency by providing value-relevan informaion o invesors (e.g., Brennan e al. 993; Walher 997; Bhaacharya 2; Elgers e al. 2; Elgers e al. 23; Gleason and Lee 23). However, prior sudies disagree abou wheher analyss reduce or srenghen he accrual anomaly. Bradshaw e al. (2) documen overopimism in analyss earnings forecass for high accruals firms and conclude ha analyss fail o aler invesors o he lower associaion of accruals han cash flows wih fuure earnings. 2 They also conjecure ha analyss accrual-relaed opimism could be driven by heir incenives raher han heir inabiliy o inerpre accruals. In conras, Elgers e al. (23) argue ha invesors could infer he lower persisence of accruals relaive o cash flows from analys forecass bu fail o do so. Boh sudies base heir inferences abou analyss conribuion o marke efficiency wih respec o accruals on samples of firms followed by analyss. However, he accrual sraegy does no require ha firms have analyss. Unlike prior sudies, I provide a direc es of analyss role in he accrual anomaly by comparing he magniude and significance 2 Bradshaw e al. (2) do no rule ou he possibiliy ha analyss parially see hrough he subsequen accrual reversals. 3

6 of he accrual anomaly for followed and non-followed firms over he anomaly and no-anomaly periods. Evidence ha he anomaly declines more rapidly and ends sooner for followed firms han for non-followed firms would suppor he widely-held noion ha analyss conribue o marke efficiency. I examine causes of he end of he anomaly using daa over I begin by documening annual excess reurns o he accrual-relaed rading sraegy of buying firms in he lowes accruals decile and selling firms in he highes decile. The findings confirm prior findings ha he accrual anomaly disappeared in 22. I also confirm prior evidence ha he accrual anomaly does no exis among loss firms (Dopuch e al. 2) and herefore focus on gain firms in subsequen analyses. Among gain firms, he mean excess reurn o he accrual-relaed sraegy is 8.9 percen during he anomaly period whereas i is.6 percen and saisically insignifican during he no-anomaly period. I nex invesigae wheher a change in he differenial persisence of accruals and cash flows accouns for he end of he anomaly. If accruals and cash flows have similar relaion wih fuure earnings during he no-anomaly period hen invesors naïve fixaion on earnings would no generae forecas errors ha are predicably relaed o accruals. I examine his by regressing fuure earnings on curren year s accruals and cash flows and find ha here is sill a subsanial difference in he relaion of cash flows versus accruals o fuure earnings. This implies ha he accrual anomaly would sill exis during he no-anomaly period if invesors naively assign he same weigh o cash flows and accruals when hey predic earnings. I nex employ he Mishkin model o esimae he weighs invesors aach o he cash flow and accruals earnings componens o predic nex-period earnings in he anomaly and noanomaly periods. I infer he weighs from he relaion beween he earnings componens and reurns in he one-year period beginning four monhs afer earnings are announced. As in prior 4

7 sudies, I find ha invesors overweighed accruals and underweighed cash flows during he anomaly period. During he no-anomaly period, he implici weighs ha invesors assign o accruals and cash flows are no significanly differen from he ime series associaion of accruals and cash flows wih fuure earnings. Thus, he Mishkin model resuls sugges ha invesors learned abou he differenial persisence of accruals and cash flows and incorporaed he correc weighs in valuing securiies in he no-anomaly period. To provide furher evidence on his issue, I esimae earnings forecas errors from a model based solely on prior-year earnings, he naïve model. By consrucion, his model ignores he lower persisence of accruals han cash flows. Forecas errors from he naïve model are significanly posiively associaed wih accruals during boh he anomaly and he no-anomaly periods. Nex, I pariion he forecas error from he naïve model ino a porion relaed o accruals and a porion ha is unrelaed o accruals. I find ha size-adjused reurns are significanly associaed wih accrual-relaed earnings forecas errors in he anomaly period bu no in he no-anomaly period. This finding suggess ha securiy pricing is consisen wih he naïve model in he anomaly period bu wih a model ha properly weighs cash flows and accruals in he no-anomaly period, consisen wih learning. I find ha invesors respond similarly o he porion of he earnings forecas error ha is unrelaed o accruals in he anomaly and no-anomaly periods. Thus, he end of he anomaly canno be due o a decline in he value-relevance of earnings for seing securiy prices. I nex invesigae wheher he correcion in invesors response o cash flows and accruals can be aribued o improvemens in analys forecass. This inference would be suppored if he anomaly and analyss accrual-relaed opimism end a he same ime and if he accrual anomaly coninues for non-followed firms. I find a subsanial decrease in analyss accrual relaed overopimism ha coincides wih he end of he accrual anomaly, bu analyss forecass are sill 5

8 opimisically biased for firms wih high accruals. There is no accrual anomaly among followed firms during he no-anomaly period. This implies ha invesors are no influenced by he remaining accrual-relaed overopimism in analyss forecass during he no-anomaly period. Consisen wih his, I find ha invesors response o analyss forecas error predicable by a given level of accruals is significan during he anomaly period while i is insignifican during he no-anomaly period. Furhermore, I find ha followed and non-followed firms had similar levels of he accrual anomaly in he anomaly period and ha he anomaly disappeared a he same ime for boh samples. These findings indicae ha he end of he anomaly is no due o improvemens in analys forecass, and sugges ha analyss forecass did no enhance marke inefficiency wih respec o accruals during he anomaly period. To provide evidence on how invesors correc accrual-relaed mispricing, I compare excess reurns o he accrual sraegy for firms held by insiuional invesors and firms wih no insiuional invesor holdings based on he conjecure ha arbirage by insiuional invesors may reduce he anomaly. I focus on insiuional invesors because hey may be beer posiioned han individual invesors o rade on knowledge abou he anomaly. Consisen wih his conjecure, I find ha reurns o he rading sraegy is similar for firms moderaely-held by insiuional invesors, for firms highly-held by insiuional invesors, and firms wih no insiuional invesor holdings in he pre-996 period, before he accrual anomaly was discovered. In he 996 o 2 period, when he anomaly became widely known and prior sudies find coninued mispricing, he excess reurns o he accrual sraegy is significanly lower for moderaely-held firms han non-held firms and is insignifican for high-held firms. I repea he analyses for muual funds and obain similar resuls. Furhermore, I find a subsanial increase in he percenage of firms held by insiuional invesors and/or muual funds, he 6

9 number of insiuions and funds holding socks a firm, and percenage of oal shares held by insiuions and funds in he no-anomaly period. These findings sugges ha insiuional invesors and muual funds played a significan role in he decline of accrual-relaed mispricing. This sudy conribues o he lieraure in several ways. Firs, I exend he lieraure invesigaing he absence of he anomaly. Green e al. (29) find an increase in boh asses managed by hedge funds and rading volume in he exreme accruals deciles during he noanomaly period, and argue ha accrual-relaed rading sraegies by large hedge funds advised by academic accounans resuled in a decline in he anomaly. Mohanram (29) dispues Green e al. s (29) conclusion by showing ha he increase in rading urnover is no unique o firms wih exreme accruals and is driven by small rades raher han by he large rades ypically made by hedge funds. He argues ha increases in he number of analyss cash flow forecass explain he recen absence of he anomaly. The findings in his sudy sugges, however, ha he absence of mispricing is no specific o followed firms, inconsisen wih he argumen ha analyss cash flow forecass explain he end of he anomaly. I find he decline in he anomaly sared earlier for firms held by insiuional invesors han for firms wih no insiuional invesor holdings. I also find increases in boh he percenage of firms held by insiuional invesors and he percenage of hose firms shares held by insiuional invesors. Thus, he findings in his sudy are more consisen wih Green e al. (29) and sugges ha an increase in rading o exploi he accrual anomaly by arbirageurs is likely o explain he absence of he anomaly. Second, numerous sudies sugges ha analyss are sophisicaed informaion processors and enhance marke efficiency by providing invesors wih value-relevan informaion (see Ramnah e al. 28). Mos imporanly, Elgers e al. (23) argue ha 6 percen of accrual anomaly would be eliminaed if invesors naively fixaed on analyss forecass. My resuls sugges, however, ha 7

10 analyss forecass remain inefficien wih respec o accruals while invesors appear o have fully learned abou he relaion of cash flows and accruals o fuure earnings. This finding is consisen wih he conjecure ha analyss accrual-relaed bias may arise from heir incenives o collude wih managemen raher han from heir inabiliy o process financial informaion (e.g., Bradshaw e al. 2). Finally, prior sudies view he biases in analyss forecas as evidence of marke wide inefficiency based on he assumpion ha analyss forecass are a reasonable proxy for marke expecaions (e.g., Bradshaw e al. 2; Bradshaw e al. 26). Richardson e al. (2) furher argue ha his should be a sandard diagnosic es. I find, however, ha alhough analyss forecass coninue o be inefficien wih respec o accruals in he no-anomaly period, his inefficiency is no refleced in securiy prices. The resuls indicae ha invesors are no influenced by he remaining accrual-relaed overopimism in analyss forecass during he noanomaly period, suggesing ha marke expecaions may significanly diverge from analyss consensus forecass. Thus, his sudy has also implicaions for sudies using analyss earnings forecass as a proxy for marke expecaions. II. Relaed lieraure and hypohesis developmen a. Behavioral versus risk-based explanaions for excess reurns o he accrual sraegy Sloan (996) finds ha he accruals componen of earnings is less persisen han he cash flows componen. He finds ha invesors price securiies as if hey naïvely predic nex-period earnings using aggregae earnings, ignoring he differenial relaion beween fuure earnings and cash flows versus accruals. As a resul, a sraegy ha akes a long posiion in he lowes-accrual firms and a shor posiion in he highes-accrual firms generaes economically large and 8

11 saisically significan excess reurns. Following Sloan (996), a large body of lieraure invesigaes he robusness of he accrual anomaly and searches for he ways o refine i. A number of sudies argue ha he anomaly is illusory and ha he excess reurns o he rading sraegy can be explained by risk-relaed facors (e.g., Khan 28 and Wu e al. 29). Mos sudies conclude, however, ha he anomaly exiss and is driven by invesors failure o undersand he lower persisence of accruals han cash flows (e.g., Collins and Hribar 2; Richardson e al. 2; Xie 2; Beneish and Vagus 22; Richardson e al. 2). Oher sudies sugges ha he anomaly is real and will endure because of significan economic barriers o arbirage (Lev and Nissim 26; Mashruwala e al. 26). b. The end of he anomaly and condiions associaed wih he presence of he anomaly Recen sudies find ha he anomaly disappeared in 22 and has been undeecable since hen (e.g., Richardson e al. 2). The absence of he anomaly in recen years implies ha a leas one of he following hree condiions associaed wih he presence of he anomaly changed: (i) a differenial relaion beween fuure earnings and cash flows versus accruals, (ii) incorrec weighing of accruals by invesors, leading o earnings forecas errors ha are predicably associaed wih accruals, and (iii) associaion of earnings forecas errors wih reurns. Among hese alernaives, Richardson e al. (2) conjecure ha adapive marke efficiency (e.g., Grossman and Sigliz 98; Lo 24), or marke learning, explains he absence of he anomaly. This explanaion suggess ha invesors learn abou he lower persisence of accruals relaive o cash flows over ime and adjus accordingly. I presupposes ha accruals coninue o have lower persisence han cash flows in he no-anomaly period, bu ha invesors correcly weigh cash 9

12 flows and accruals when predicing earnings in he no-anomaly period. An alernaive scenario is ha he persisence of accruals and cash flows is similar during he no-anomaly period, in which case earnings forecass ha ignore he cash flow and accruals componens of earnings would yield forecas errors ha are no associaed wih accruals. In his scenario, reurns would no be associaed wih accruals even if invesors naively rely on aggregae earnings when seing securiy prices. The absence of he anomaly could also be driven by a decline in value-relevance of he informaion in earnings raher han marke learning or a change in differenial persisence of accruals and cash flows. Tha is, he associaion of earnings forecas errors wih reurns may be oo small o generae excess reurns even if he differenial relaion beween fuure earnings and cash flows versus accruals persiss in he no-anomaly period and invesors fail o anicipae he lower persisence of accruals han cash flows. c. Change in differenial persisence of accruals and cash flows for he end of he anomaly A number of sudies find lower absolue discreionary accruals following he enacmen of SOX, and conclude ha firms engaged in less accrual-based earnings managemen once SOX became effecive (e.g., Cohen e al. 28). This evidence is paricularly imporan because he absence of he anomaly coincides wih he enacmen of SOX. Bhojraj e al. (29) argue ha an increase in he qualiy of accruals due o lower earnings managemen following he enacmen of SOX in 22 and FAS 46 in 23 resuled in a decrease in accrual-relaed mispricing among resrucuring firms. Thus, I firs es wheher a change in he differenial persisence of accruals and cash flows raher han marke learning accouns for he end of he anomaly. If accruals and cash flows have similar relaion wih fuure earnings during he no-anomaly period hen

13 invesors naïve fixaion on earnings would no generae forecas errors ha are predicably relaed o accruals. This leads o he firs hypohesis: H: Accruals and cash flows have similar persisence in he no-anomaly period. d. Marke learning explanaion for he end of he anomaly If he persisence of accruals and cash flow componens of earnings coninues o differ during he no-anomaly period hen he absence of he anomaly could be due o eiher marke learning or a decrease in value-relevance of earnings such ha he associaion of earnings forecas errors wih reurns is oo weak o generae excess reurns. The absence of he anomaly in recen years would be consisen wih marke learning explanaion if invesors learn o correcly weigh accruals and cash flows when incorporaing he informaion in curren year earnings ino heir forecass of fuure earnings. This leads o he second hypohesis: H2: Invesors fully anicipae he lower persisence of accruals relaive o cash flows when forming expecaions of fuure earnings during he no-anomaly period. e. Decline in value-relevance of earnings as an explanaion for he end of he anomaly Since he absence of he anomaly coincides wih he enacmen of SOX, finding evidence consisen wih marke learning would imply ha regulaory changes in SOX and associaed accouning scandals alered invesors o he differenial relaion beween fuure earnings and cash flows versus accruals and hereby conribued o marke learning. However, he coincidence of

14 he absence of he anomaly wih SOX also raises he possibiliy ha invesors los confidence in he financial informaion provided by firms due o he accouning scandals, resuling in he absence of he anomaly. If invesors rely less on informaion in earnings, he associaion of earnings forecas errors wih reurns may be oo small o generae excess reurns even if invesors fail o anicipae differenial persisence of accruals and cash flows. This leads o he hird hypohesis: H3: There is no associaion of earnings forecas errors wih reurns during he noanomaly period. f. Analyss conribuion o he end of he anomaly Analys forecass are an imporan componen of he informaion se ha is refleced in securiy prices. Numerous sudies sugges ha analyss are sophisicaed informaion processors who are more likely han invesors o undersand he implicaions of financial informaion for fuure earnings (e.g., Bradshaw e al. 2; Ali e al. 23; Elgers e al. 23; Chen and Jiang 25; Ramnah e al. 28). Consisen wih his, a large body of lieraure concludes ha analyss improve marke efficiency by providing value-relevan informaion o invesors (e.g., Brennan e al. 993; Walher 997; Bhaacharya 2; Elgers e al. 2; Elgers e al. 23; Gleason and Lee 23; Ramnah e al. 28). However, prior sudies reach conflicing conclusions wih respec o analyss conribuion o he accrual anomaly. Some sudies find a significan posiive associaion beween opimism in analyss earnings forecass and accruals (Ahmed e al. 2, Bradshaw e al. 2). Bradshaw e al. (2) conclude ha even sophisicaed informaion processors, i.e., analyss, do no undersand he lower associaion of 2

15 accruals han cash flows wih fuure earnings and hus heir forecass do no aler invesors abou differenial persisence of accruals and cash flows. They furher conjecure ha analyss accrualrelaed opimism could be driven by heir incenives o collude wih managers raher han heir misundersanding of he differenial relaion beween fuure earnings and accruals versus cash flows. In conras, Elgers e al. (23) argue ha analyss warn invesors abou fuure earnings problems associaed wih high accruals. In paricular, hey find ha boh analyss and invesors fail o fully anicipae he lower persisence of accruals han cash flows and hus overweigh he informaion in accruals in heir earnings predicions. They find, however, ha he overweighing by analyss is less han one hird of he overweighing by invesors. They conclude ha reurns from he accrual anomaly would be reduced by over 6 percen if invesors naively relied on analyss earnings forecass. Kang and Yoo (27) dispue he Elgers e al. (23) finding, arguing i is induced by omied variables. Following Liu and Thomas (2), hey conrol for he revision in analyss one year ahead forecass and find no significan differences in overweighing of he informaion in accruals by analys and invesors. They conclude ha analys forecass do no help invesors undersand he lower persisence of accruals, a conclusion similar o ha reached by Bradshaw e al. (2). The absence of he anomaly in recen years offers a unique opporuniy o undersand analyss conribuion o he accrual anomaly. Moreover, an imporan feaure of prior sudies is ha hey base heir inferences solely on samples of firms ha are followed by analyss. This research design obscures analyss conribuion o he accrual anomaly. Unlike prior sudies, I provide a direc es of analyss role in increasing or reducing he accrual anomaly by comparing he magniude and significance of he accrual anomaly for followed and non-followed firms over he anomaly and no-anomaly periods. Since analyss are sophisicaed informaion processors, 3

16 hey are more likely han naïve invesors o learn abou he lower persisence of accruals relaive o cash flows. Thus, a reducion in analyss accrual-relaed opimism, or analys learning, could explain he absence of he anomaly. Analys learning as an explanaion for he absence of he anomaly would be suppored if he anomaly and analyss accrual-relaed opimism end a he same ime and if he accrual anomaly coninues for non-followed firms. This leads o he fourh hypohesis: H4a: Analyss accrual-relaed opimism disappears during he no-anomaly period. H4b: The absence of he accrual anomaly is specific o followed firms during he noanomaly period. III. Sample selecion and descripive saisics a. Sample selecion I obain financial saemen daa from he CRSP/COMPUSTAT merged annual daabase, sock reurns daa from he CRSP monhly sock reurns files, and analys forecas daa from he IBES deail file. As in prior sudies, I exclude financial firms (SIC code beween 6 and 6999). I follow Bradshaw e al. (2) in consrucing financial variables. In paricular, I use Saemen of Financial Accouning Sandards No. 95 (SFAS 95) daa o measure accruals and cash flows. SFAS daa became available in fiscal year 988, and I herefore begin he sample period in fiscal year 988 and end i in fiscal year 28. In addiion, idenifying firms followed by analyss is of significan imporance for he purpose of his sudy. IBES coverage sars in he lae 97s and daa are only available for large firms before he lae 98s. Therefore, beginning 4

17 he sample in he lae 98s decreases he possibiliy of erroneously classifying followed firms as having no analys coverage. Following Bradshaw e al. (2), I use wo alernaive measures for accruals. The firs is based on working capial accruals and he second on oal ne accruals. Bradshaw e al. (2) find ha working capial accruals beer capure accruals ha lead o earnings reversals ha are unanicipaed by invesors and repor resuls for his measure in heir main ess. Following heir approach, I only repor he resuls for working capial accruals for conciseness, bu find similar resuls in all my analysis using he oal ne accruals measure. 3 I measure working capial accruals as follows: WCAcc= Increase in Accouns Receivable (Compusa iem RECCH) + Increase in invenory (Compusa iem INVCH) + Decrease in Accouns Payable and Accrued Liabiliies (Compusa APALCH) + Decrease in Accrued Income Taxes (Compusa iem TXACH) + Increase (Decrease) in Asses (Liabiliies)-Oher (Compusa iem AOLOCH) I use earnings before ineres, axes, depreciaion, and amorizaion (Compusa iem OIBDP) as he earnings measure and obain he corresponding cash flows measure, WCCF, by subracing WCAcc from Compusa iem OIBDP. As in prior research, I deflae all variables by average oal asses (Compusa iem AT). Table repors he sample selecion procedure. The oal number of firm-years read from he CRSP/Compusa merged file over 988 and 28 period is 3,423. The sample wih non-missing earnings (OIBDP), working capial accruals (WCAcc), and corresponding cash flows (WCCF) daa consiss of 92,988 firm-years. 3 Bradshaw e al. (2) also find ha he resuls are very similar for oal ne accruals and working capial accruals measures. Following Bradshaw e al. (2), I measure oal ne accruals as follows: TAcc= Income before exraordinary iems (Compusa iem IBC)- Ne Cash Flows from Operaing Aciviies (Compusa iem OANCF). 5

18 Sock reurns are obained from CRSP monhly reurns file and are inclusive of dividends. I obain compounded buy-and-hold reurns over welve monhs beginning four monhs afer he end of he fiscal year, and compue size adjused reurns by deducing a firm s size-mached porfolio buy-and-hold reurn from is raw buy-and-hold reurn. As in Kraf e al. (26), I obain size porfolios from CRSP calculaions of size deciles of NYSE and AMEX firms. Kraf e al. (26) idenify several sample selecion biases ha affec excess reurns o he accrual relaed sraegy, and I follow heir procedures o miigae hese problems. One sample selecion bias hey idenify is selecing socks based on he curren lising exchange insead of he acual exchange for he period during which sock reurns are measured. Since I use socks lised in all exchanges, he resuls do no suffer from his selecion bias (Kraf e al. 26). 4 Following heir procedures, I se a firm s reurn o zero for any monh in which i is missing (WRDS code.b ). If a firm is no assigned o a size decile by CRSP, I manually deermine is size porfolio by using is marke capializaion as of he beginning of he year. In addiion, if a firm is delised during he reurn accumulaion period, I use he delising reurn in he monh in which he firm deliss and assume ha he firm s reurn is equal o he reurn of is size-mached porfolio for he res of he year. The delising reurn is se o -% if a firm s delising reurn is missing and he delising is forced by he exchange or Securiies and Exchange Commission (SEC) or is due o liquidaion. Finally, I resric he sample o firm-year observaions wih fiscal-year-end sock prices greaer han one dollar. 5 The final number of firm years having financial informaion and sock reurns daa is 78,45. 4 Anoher sample selecion bias idenified by Kraf e al. (26) is he exclusion of firms no reporing earnings or accruals in year + from he rading sraegy sample. I also do no have his sample selecion issue. 5 I exclude hese firm-years o miigae he concern abou he real underlying cause of he inefficiency driven by low priced socks (Kohari 2; Kraf e al. 26). However, I obain similar resuls when I include hese firm-years in he sample or when I use $5 as he cu-off price. 6

19 My ess require idenifying firms followed by analyss and hose wih no analys following and measures of analyss forecas error for followed firms. I obain analys forecas daa from he IBES deail file and adjus analys forecass for sock splis and sock dividends using he IBES adjusmen facors. I find ha 3,348 firm-years have no analys forecass from he announcemen of year earnings o he porfolio formaion dae, and classify hese firm years as non-followed firms. 6 For firms followed by analyss, I manually compue he mean consensus forecass using forecass over he 6-day period ending prior o he porfolio formaion dae. 7 There are 4,7 firm-years for which I can obain he mean consensus forecas. Finally, I obain insiuional invesor holdings from he Thomson Financial Specrum 3F Insiuional Holdings Daabase. b. Descripive saisics I begin by calculaing yearly reurns o he accrual-relaed rading sraegy of buying firms in he lowes-accrual decile and selling firms in he highes-accrual decile and plo he reurns in Figure. This evidence confirms prior findings ha he accrual anomaly ended in 22. I herefore calculae descripive saisics separaely for he anomaly period (988-2) and he no-anomaly period (22-28) and es for differences across he wo samples. I repor he resuls in Table 2, Panel A. Firms are on average larger and more likely o be followed by analyss during he no-anomaly period. I also find ha boh he percenage of firms held by insiuional invesors and he percenage of hose firms shares held by insiuional invesors 6 Of he firms having no analys forecass before he porfolio formaion dae, 7,793 firm-years have analys forecass during he year afer he porfolio formaion dae. Eliminaing hem from no analys following group does no affec he resuls. 7 Following prior sudies, I prefer no o use IBES consensus forecas and resric he forecas age o 6 days o miigae he effecs of sale forecass. In addiion, analyss consensus forecass may beer reflec he invesors expecaions if he forecas horizon is kep shorer bu i resuls in a decrease in sample of firm-years having consensus forecas. 7

20 increased in he no-anomaly period. Firms are also on average less profiable in he no-anomaly period. In he anomaly period, mean asse-deflaed earnings are.63, and 2 percen of he firms repor losses. In he no-anomaly period, asse-deflaed earnings are.48 and 22. percen of he firms repor losses. The differences are saisically significan. The disribuions of accruals and cash flow componens of earnings sugges ha he decrease in firms profiabiliy is driven by a decrease in mean accruals. The mean asse-deflaed working capial accruals are.2 and.8 in he anomaly and no-anomaly periods, respecively, and he difference is saisically significan. The corresponding numbers for mean cash flows are.42 and.4 in he anomaly and no-anomaly periods, respecively, and he difference is insignifican. Unabulaed resuls show ha he mean accruals for he firms in he lowes accruals decile are -.25 and -.9 in he anomaly and no-anomaly periods, respecively, and he difference is saisically insignifican. In he highes accruals decile, he mean accruals in he anomaly period are.86 and are significanly larger han he mean accruals in he no-anomaly period,.37. This suggess ha he decrease in mean accruals in he no-anomaly period is due o a decrease in mean accruals of firms in he highes accruals decile. This is consisen wih prior sudies ha find a significan decrease in discreionary accruals following he enacmen of SOX (e.g., Cohen e al. 28). Thus, he findings sugges ha an increase in he qualiy of accruals due o a decrease in accruals-based earnings managemen may conribue o he absence of he anomaly. Table 2, Panel A also provides descripive saisics for analyss mean consensus forecas error, AnalysFERROR. The mean (median) consensus forecas errors are -.24 (-.3) and -.3 (-.4) in he anomaly and no-anomaly periods respecively and unabulaed resuls show ha he difference in means is saisically significan. This is consisen wih prior findings suggesing ha opimism in analyss consensus forecass early in he year decreased in recen 8

21 years (Hovakimian and Seanyasiri 29). In Table 2, Panel B, I repor he correlaions for earnings, accruals, and cash flows separaely for anomaly and no-anomaly periods. Consisen wih prior research, accruals are posiively correlaed wih earnings and negaively correlaed wih cash flows. The correlaion coefficiens are similar in he anomaly and no-anomaly periods. IV. Research Mehods, Specificaion of Empirical Tess, and Resuls a. Accrual anomaly over ime In his secion, I firs documen he excess reurn o he accrual relaed sraegy of buying firms in he lowes and selling firms in he highes accruals deciles. Figure confirms prior evidence ha he accrual anomaly disappeared in 22. Therefore, I separaely examine he anomaly and no-anomaly periods. Hayn (995) shows ha securiy prices are insensiive o losses, and argues ha his is because of equiy invesors liquidaion opion. This suggess ha one of he condiions required for he presence of he accrual anomaly, namely he associaion beween earnings and reurns, is weak or missing for loss firms. Consisen wih his, Dopuch e al. (2) find ha here is no anomaly among loss firms. Therefore, I repor he excess reurn o he accrual sraegy for he full sample as well as separaely for gain firms and loss firms. Table 3, Panel A repors he excess reurn o he accrual sraegy in he anomaly period. For he full sample, he excess reurn o accrual sraegy is abou.7 percen per year. Consisen wih Kraf e al. (26), I find ha he accrual anomaly is mainly driven by negaive excess reurn o he firms in he highes accruals decile. The excess reurn o he firms in he lowes accruals decile is small and saisically insignifican whereas i is larger and significanly posiive for he firms in he second decile of accruals. Thus, he resuls for he full sample sugges an invered U-shaped relaion beween accruals and fuure sock reurns as in Kraf e al. 9

22 (26). An invered U-shaped relaion is inconsisen wih Sloan (996) s earnings fixaion hypohesis ha he accrual anomaly is driven by invesors failure o anicipae lower persisence of accruals wih respec o cash flows. I find ha he invered U-shaped relaion is induced by loss firms, and ha he mean excess reurn o he accrual sraegy increases o 8.9 percen per year when I exclude loss firms from he sample. This change is almos enirely due o he lowes accrual decile, which has insignifican excess reurns when loss firms are in he sample bu has saisically significan mean excess reurns of 8.8 percen per year when loss firms are excluded from he sample. 8 Unlike he resuls for he full sample, he fuure sock reurns decline monoonically from lowes- o highes-accrual porfolios. Because loss firms disor reurns o he accrual sraegy, I focus on gain firms in subsequen analyses. Table 3, Panel B repors he excess reurns o he accrual sraegy over he no-anomaly period. Among gain firms, he average excess reurn o he accrual sraegy is.6 percen over Figure shows he excess reurn o he accrual sraegy for gain firms over ime. The resuls confirm prior findings ha he accrual sraegy does no earn significan excess reurns afer 2. b. Persisence of accruals in he anomaly and no-anomaly periods In his secion, I es wheher a change in he differenial relaion beween fuure earnings and cash flows versus accruals explains he absence of he anomaly in recen years. I examine his by regressing nex year s realized earnings on he accruals and cash flows componens of his year s earnings: 9 8 These findings sugges ha having loss firms in he accrual sraegy is likely o influence he inferences wih respec o causes of he accrual anomaly in prior sudies. For example, moivaed by he evidence ha he anomaly is mainly driven by he negaive excess reurn o firms wih high accruals, Kohari e al. (25) argue ha agency heory of overvalued equiy raher han invesors failure o anicipae lower persisence of accruals explains he anomaly. 9 Requiring firms o have nex year s earnings reduces he sample from 43,378 o 4,28 and from 6,85 o 5,254 in he anomaly and no-anomaly periods, respecively. 2

23 EARN α WCCF + u. () + = + γ WCAcc + γ 2 I esimae he model on a year-by-year basis for and repor Fama MacBeh coefficiens and corresponding -saisics separaely for he anomaly and no-anomaly periods in Table 4. I find ha he coefficien on accruals increases from.635 in he anomaly period o.7 in he no-anomaly period and he increase is saisically significan. There is also a significan increase in he persisence of cash flows in he no-anomaly period. I find ha he coefficien on accruals is significanly smaller han he coefficien on cash flows in boh he anomaly and no-anomaly periods. Furhermore, he gap beween he persisence of cash flows and accruals is.78 and.42 in he anomaly and no-anomaly periods respecively, and he change in he gap is no saisically significan. Overall, alhough here is an increase in he persisence of accruals in he no-anomaly, he differenial relaion beween fuure earnings and cash flows versus accruals persiss in ha period. This suggess ha a decrease in he differenial persisence of accruals and cash flows canno explain he absence of he anomaly in he noanomaly period. c. Invesors pricing of accruals and cash flows in he anomaly and no-anomaly periods In his secion, I use he Mishkin model o measure changes in invesors pricing of accruals and cash flows componens of earning in he no-anomaly period. This procedure allows me o infer he weighs ha invesors assign o accruals and cash flows in predicing nex-period earnings and he consisency of hose weighs wih heir empirical relaion wih earnings in equaion. Since accruals coninue o be less persisen han cash flows in he no-anomaly period, finding ha invesors correcly weigh accruals and cash flows would suppor he marke 2

24 learning explanaion for he absence of he anomaly. To examine his, I esimae he following model: Forecasing Model: EARN + = + γ WCAcc + γ 2WCCF + u+ γ, and (2a) Pricing Model: ARET β β EARN γ WCAcc γ WCCF ν. (2b) * * + = + ( + 2 ) + + I esimae he model on a year-by-year basis for Table 5 repors Fama MacBeh coefficiens and corresponding -saisics separaely for he anomaly and no-anomaly periods. Marke efficiency wih respec o he accruals componen of earnings requires ha γ = γ *. As in prior sudies, he resuls for he anomaly period show ha hanγ while * γ is significanly larger * γ 2 is significanly smaller hanγ 2. This implies ha invesors overweigh he informaion in accruals and underweigh he informaion in cash flows when forming expecaions of fuure earnings. * In he no-anomaly period, I find ha ha γ is no significanly differen hanγ and ha * γ 2 is no significanly differen hanγ 2, suggesing ha invesors correcly weigh he informaion in accruals and cash flows when seing securiy prices. The resuls so far show ha he differenial associaion of fuure earnings wih accruals versus cash flows persiss in he noanomaly period. However, invesors correcly price he componens of earnings by lowering he weigh ha hey assign o accruals and increasing he weigh ha hey assign o cash flows. The findings are consisen wih marke learning explanaion for he absence of he anomaly, The resuls and inferences are similar when I esimae he model over separae pooled samples in he anomaly and no-anomaly periods. 22

25 suggesing ha invesors learned abou he differenial persisence of accruals and cash flows and evenually incorporaed he correc weighs in valuing securiies. Furhermore, I find ha he reurn response coefficien, β, is saisically similar in he anomaly and no-anomaly periods. Thus, I fail rejec he hypohesis ha here is no associaion of earnings forecas errors wih reurns during he no-anomaly period. This suggess ha a loss of invesor confidence in he financial informaion provided by firms is no likely o accoun for he absence of he anomaly. d. Invesors response o earnings forecas errors ha are predicable versus unpredicable from accruals I provide furher evidence on marke learning by sudying invesors response o earnings forecass errors predicable by a given level of accruals in he anomaly and no-anomaly periods. Sloan (996) argues ha invesors naively rely on aggregae earnings and ignore he lower persisence of accruals han cash flows when forming expecaions of fuure earnings. Following Sloan (996), I esimae earnings forecas errors from a model based solely on prior-year earnings, he naïve model: EARN + = + γ EARN + u+ γ. (3) The resuls, repored in Table 6, Panel A, sugges ha he relaion beween fuure earnings and curren earnings is similar in he anomaly and no-anomaly periods. The coefficien on curren earnings is.84 in he anomaly period and.844 in he no-anomaly period. The residuals from Model (3), u +, represen he earnings forecas errors from he naïve model, NaiveFerror. Because he naive model ignores he lower persisence of accruals han cash flows, forecas errors from his model will be posiively associaed wih accruals. I regress earnings forecas errors, NaiveFerror, on curren year accruals o obain he componen of earnings 23

26 forecas errors predicable by a given level of accruals. In paricular, I esimae he following model: NaiveFerror + = + αwcacc + e+ α. (4) The resuls are repored in Table 6, Panel B, and sugges ha earnings forecas errors are significanly posiively associaed wih curren-year accruals in he anomaly and no-anomaly periods, consisen wih he findings ha accruals are less persisen han cash flows in boh periods. This implies ha he accrual anomaly would sill exis during he no-anomaly period if invesors naively assigned he same weigh o cash flows and accruals when hey predic earnings. I nex examine invesors response o he componens of NaïveFerror ha are predicable versus unpredicable using accruals. The prediced par of NaïveFerror in Model (4), Pred_NaiveFerror, is he fied value using accruals. The residuals from Model (4) are he unpredicable componen of earnings forecas errors, Unpred_NaiveFerror. Marke learning would sugges ha invesors sop responding o predicable earnings forecas errors while hey respond similarly o unpredicable errors in he anomaly and no-anomaly periods. To examine his, I esimae he following model: ARET + = δ + δpred _ NaiveFerror + + δ 2Unpred _ NaiveFerror + + ε +. (5) The resuls, repored in Table 6, Panel C, sugges ha invesors responded srongly o earnings forecas errors predicable by a given level of accruals during he anomaly period while heir response is insignifican during he no-anomaly period. In paricular, he coefficien on Pred_NaiveFerror in he anomaly period is 4.3 and is saisically significan. The 24

27 corresponding coefficien in he no-anomaly period is.98 and is insignifican. Thus, I find ha invesors response o he informaion in earnings is consisen wih he naïve model in he anomaly period bu is consisen wih a model ha properly weighs cash flows and accruals in he no-anomaly period. The resuls furher suppor he marke learning explanaion for he end of he anomaly. Furhermore, I find ha invesors respond similarly o he unpredicable componen of earnings forecas errors, Unpred_NaiveFerror, in he anomaly and no-anomaly periods. This indicaes ha reurns respond o earnings forecas errors in he anomaly and noanomaly periods, and is consisen wih he resuls from he Mishkin model. The findings sugges ha he end of he anomaly is no due o a decline in he value-relevance of earnings for seing securiy prices. e. Analyss conribuion o he end of he anomaly In his secion, I examine wheher a reducion in accrual-relaed bias in analyss earnings forecass, or analys learning, explains he absence of he anomaly in recen years. If analys forecass are he cause of improved marke efficiency hen analyss accrual-relaed opimism should disappear a he same ime as he anomaly. Furhermore, he anomaly should only end for followed firms. To examine his, I firs regress analyss mean consensus forecas error on accruals. Tha is, I esimae he following model: AFerror α α WCAcc. (6) + = + + e+ I esimae he model on a year-by-year basis for Table 7, Panel A repors Fama MacBeh coefficiens and he corresponding -saisics separaely for he anomaly and noanomaly periods. The mean coefficiens on accruals are -.34 and -.82 in he anomaly and 25

28 no-anomaly periods respecively, and boh coefficiens are saisically significan, indicaing ha analyss earnings forecass are opimisically biased for firms wih high accruals in boh periods. The difference in he coefficiens is saisically significan, suggesing a significan decrease in analyss accrual-relaed overopimism in he no-anomaly period. The findings sugges ha he anomaly would sill exis if invesors naively fixaed on analyss consensus earnings forecass. I examine his quesion by regressing reurns on he componen of analyss forecas errors ha is predicable based on accruals and he unpredicable porion. I esimae he following model: ARET + = + δpred _ AFerror + + δ 2Unpred _ AFerror + + ε + δ, (7) where Pred_AFerror is he componen of AFerror prediced by accruals in Model (7), equal o he fied values from Model (7), and Unpred_AFerror is he componen of AFerror ha is no predicable by accruals, equal o he residuals from Model (7). The resuls, repored in Table 7, Panel B, sugges ha invesors srongly responded o analyss earnings forecas errors predicable by a given level of accruals during he anomaly period while heir response is insignifican during he no-anomaly period. In paricular, he coefficien on Pred_AFerror is and saisically significan in he anomaly period while i is.7 and insignifican in he no-anomaly period. Thus, I find ha invesors are no influenced by he remaining accrual-relaed overopimism in analyss forecass during he no-anomaly period. These resuls sugges ha marke expecaions may significanly diverge from analyss consensus forecass. The evidence ha analyss coninue o issue opimisically biased forecass for firms wih high accruals while invesors appear o fully learn abou he lower persisence of accruals han cash flows is consisen wih he conjecure ha analyss accrual-relaed bias could 26