Trends in Earnings Volatility, Earnings Quality and Idiosyncratic Return. Volatility: Managerial Opportunism or Economic Activity

Transcription

1 Trends in Earnings Volailiy, Earnings Qualiy and Idiosyncraic Reurn Volailiy: Managerial Opporunism or Economic Aciviy Absrac This paper examines he causes for he increasing earnings volailiy and he deerioraing earnings qualiy over he pas hree decades and he roles of hese rends in explaining rising idiosyncraic reurn volailiy. We find ha he rend in earnings volailiy is primarily caused by increases in cash flow volailiy. The key driver for he deerioraing earnings qualiy embodied in he Dechow and Dichev (2002) accruals qualiy measure is he less negaive conemporaneous correlaion beween cash flow and accruals. Collecively, he rends in earnings volailiy and is componens and he rend in earnings qualiy help explain he documened increasing idiosyncraic sock reurn volailiy. Our resuls highligh he roles of boh operaing volailiy (economic forces) and financial reporing qualiy (managerial opporunism) in explaining earnings volailiy and idiosyncraic reurn volailiy, wih economic forces dominaing managerial opporunism. This Draf: March 7, 2008

2 Trends in Earnings Volailiy, Earnings Qualiy and Idiosyncraic Reurn Volailiy: Managerial Opporunism or Economic Aciviy I. Inroducion There is evidence in he lieraure suggesing ha earnings volailiy has increased over ime (Wei and Zhang, 2006). Differen inerpreaions have been provided regarding he implicaions of he upward rend in earnings volailiy. Wei and Zhang (2006) find ha increasing earnings volailiy and decreasing corporae earnings explain he upward rend of idiosyncraic sock reurn volailiy documened in Campbell, Leau, Malkiel and Xu (200), a now well-known empirical regulariy in he sock marke. Ohers ypically use scaled earnings volailiy as a measure of income smoohing, where he scalar can be cash flow volailiy (Francis, LaFond, Olsson, and Schipper, 2004) or sales variabiliy (Imhoff, 98 and Albrech and Richardson, 990). Holding cash flow volailiy or sales volailiy consan, increasing earnings volailiy indicaes decreasing income smoohing. Ineresingly, in a similar vein, Rajgopal and Venkaachalam (2006) find a posiive associaion beween deerioraing financial reporing qualiy and increasing idiosyncraic reurn volailiy. One of heir primary measures for financial reporing qualiy is earnings qualiy based on Dechow and Dichev (2002, hereafer DD ). The DD earnings qualiy measure, or accruals qualiy as hey called i, is he sandard deviaion of residuals in a regression of change in working capial on muli-period cash flows and capures he degree of mapping beween accruals and cash flows. Since accruals are creaed when here is a difference in iming beween acual cash flow and repored earnings, firms wih lower earnings qualiy would book his shif in iming in a less imely manner. As DD

3 pu i, Firms wih low accrual qualiy have more accruals ha are unrelaed o cash flow realizaions, and so have more noise [ ] in heir earnings. DD find a srong associaion beween earnings volailiy and accrual qualiy. Indeed, DD repor ha he correlaion beween he sandard deviaion of earnings and heir accrual qualiy measure is as high as 0.82 for heir sample period of Our paper invesigaes hree quesions: () Do he above-cied earnings volailiy and earnings qualiy rends conain some common elemens? (2) If here is only parial commonaliy in hese rends, wha oher facors also conribue o hem? And (3) how do earnings volailiy and earnings qualiy rends ogeher explain he reurn volailiy rend? DD s evidence on he associaion beween earnings volailiy and accrual qualiy is only cross secional ha is, uncondiionally and across firms. Earnings volailiy is negaively correlaed wih accrual qualiy hrough a decomposiion of earnings volailiy ino cash flow volailiy and accrual volailiy. This cross secional relaion, however, does no necessarily ranslae ino a ime-series one: increasing earnings volailiy may no imply decreasing earnings qualiy. The reason is ha if he increased variabiliy in earnings is picked up by addiional flucuaions in accruals, hen he mapping beween cash flows and accruals will improve, which by DD s definiion, meaning ha earnings qualiy will improve. In oher words, he change in earnings volailiy could be aenuaed by a more exensive use of accruals. A decomposiion of earnings volailiy illusraes he above argumen: Var( E) = Var( CF + ACC) = Var( CF) + Var( ACC) + 2, Var( CF) Var( ACC), ρ CF ACC where Var( ) is he variance of a variable, E is earnings, CF is cash flow, ACC is accruals, and ρ is he correlaion beween CF and ACC. A change of earnings variance could 2

4 be caused by a change in any of he hree componens: cash flow volailiy, accrual volailiy, or ρ. Earnings qualiy measures he degree of correlaion beween cash flow and accruals, or capures ρ in he above decomposiion. This however, does no preclude he oher wo variables from conribuing o he rend in earnings volailiy. Based on his decomposiion, we show ha he rend in earnings volailiy is srongly associaed wih changes in cash flow volailiy in a large sample of quarerly firm daa over he period, 978 o We illusrae ha he rend in earnings volailiy is similar in magniude o ha of cash flow volailiy bu much sronger han he rend in accrual volailiy. Furhermore, boh he marginal conribuion and overall conribuion of he cash flow volailiy rend o he earnings volailiy rend dominae hose of accrual volailiy and hose of correlaion. Our resuls are robus o conrols of firm characerisics such as size and marke-o-book and leverage. We find ha he increase in earnings volailiy is relaed o increasing volailiy in firm operaing aciviies. Wih accrual accouning, higher operaing volailiy forces he firm o repor more volaile earnings, as i is infeasible o smooh cash flow flucuaions and shif hem o accruals. A he same ime, firms respond o increasing operaing volailiy operaing wih larger effors in accrual recogniion, which resuls in a growing accrual volailiy. However, he effor devoed o income smoohing hrough accrual managemen is no sufficien o offse he volailiy in cash flows. As a resul, he correlaion beween accruals and cash flow becomes less negaive. Therefore, we argue ha managemen opporunism and operaing volailiy boh lead o increasing earnings volailiy. However, emporally, operaing volailiy dominaes managerial opporunism in conribuing o he rend in earnings volailiy. 3

5 We furher show ha he rend in earnings qualiy repored by DD relies almos exclusively on he correlaion beween accruals and cash flow bu no on cash flow volailiy. We illusrae his by simulaing he cash flow and change in working capial series based on he parameers calibraed from empirical daa. In regressions of simulaed daa ha vary along hree dimensions, namely, increasing cash flow volailiy, relaively sable volailiy in working capial accruals and decreasing correlaion beween accruals and cash flow, he DD-measure shows a decreasing rend only when he correlaion beween accruals and cash flow decreases. We conclude ha decreasing earnings qualiy as measured by DD is neiher caused by increasing earnings volailiy nor does i serve as a leading facor for increasing earnings volailiy. Based on our resuls, we believe ha a correc way o inerpre he DD measure is o focus on is original meaning he correlaion beween cash flow and accruals. This finding is consisen wih Wysocki (2005), where he auhor also sresses ha he DD measure is driven by he negaive correlaion beween accruals and cash flow. Finally, we show ha boh earnings volailiy rend and earnings qualiy rend coexis in explaining he rend in idiosyncraic reurn volailiy. This suggess ha boh financial reporing qualiy and he economic environmen affec idiosyncraic volailiy. In sum, his paper conribues o he lieraure in a number of ways. Firs, we idenify he drivers for he rend in earnings volailiy. While a few papers have separaely idenified he rend eiher in earnings or cash flow volailiies, his paper brings hese srands of lieraure ogeher. Our main finding is ha cash flow volailiy is he primary bu no he only driver for earnings volailiy. We also documen he role of managemen opporunism in conribuing o he earnings volailiy rend. Second, we find ha earnings qualiy, financial reporing volailiy, and cash flow volailiy coexis in 4

6 explaining idiosyncraic sock reurn volailiy. Our resuls highligh he roles of boh operaing volailiy (economic forces) and financial reporing qualiy (managerial opporunism) in explaining earnings volailiy and idiosyncraic reurn volailiy, wih economic forces dominaing managerial opporunism. This sudy feaures several novelies in he research mehodology. Firs, we use quarerly daa o increase he frequency for volailiy esimaion. Prior sudies ypically rely on annual daa for volailiy esimaion, which consrains he number of observaions across he ime-series. Using quarerly daa grealy enlarges he ime-series observaions and is suiable for our rend sudies in volailiy. In order o documen rends, we mach fiscal quarers wih calendar ime. Second, o address he seasonaliy problem in he quarerly daa, we employ he X procedure of he US Census Bureau. The X deseasonalizaion is widely used in economics bu has no ye gained much aenion in accouning and finance. Finally, we use simulaion echniques o examine he drivers conribuing o he rend in earnings qualiy. The res of he paper is organized as follows. Secion II provides a brief review of he lieraure and research design. Secion III describes our sample and deseasonalizaion echnique. Secion IV examines he earnings volailiy rend and is deerminans. Secion V discusses he earnings qualiy rend and he role of he conemporaneous correlaion beween cash flow and accruals in deciding he earnings qualiy rend. Secion VI invesigaes he causes for he idiosyncraic reurn volailiy rend, Secion VII presens some robusness resuls and Secion VIII concludes. II. A Review of he Lieraure and a Descripion of he Research Design 5

7 Campbell e al. (200) find ha he level of firm-specific sock reurn volailiy has been increasing over he period from 962 o 997 in he US. Their findings are confirmed by Morck, Yeung and Yu (2000), who show ha he raio of idiosyncraic reurn risk o sysemaic risk has surged over ime. These findings indicae ha invesmen porfolios may no always have been fully diversified and ha he risk-reward relaionship ends o be weakened. In fac, Ang e al. (2006) show ha idiosyncraic volailiy is negaively correlaed wih fuure sock reurns, a puzzling resul given he radiional wisdom of a posiive risk-reurn radeoff. Two recen papers endeavor o uncover he causes for he rend in reurn volailiy. Wei and Zhang (2006) link he rising reurn volailiy o firm fundamenals. They documen a posiive associaion beween reurn volailiy and earnings volailiy, arguing ha his is consisen wih he noion ha sock reurn volailiy is naurally driven by uncerainy abou fuure profis. Moving a sep furher, Irvine and Poniff (2005) aribue he Wei and Zhang (2006) resuls o rising fundamenal cash flow shocks from produc marke compeiion. A recen paper by Rajgopal and Venkaachalam (2006) provides evidence ha he increase in reurn volailiy is relaed o he deerioraion in earnings qualiy, which is affeced by firm s accouning choices. One primary earnings qualiy measure employed in Rajgopal and Venkaachalam (2006) is he Dechow and Dichev (2002) accrual qualiy measure (he DD measure). While, Dechow and Dichev (2002) show ha he DD measure is highly correlaed wih earnings volailiy, Wysocki (2005) argues ha he DD measure is primarily driven by he conemporaneous correlaion beween cash flow and accruals. Given he relaion beween earnings qualiy and earnings volailiy, i is necessary o differeniae he conribuion of 6

8 earnings volailiy and earnings qualiy o reurn volailiy so as o disinguish beween he findings of Wei and Zhang (2006) from Rajgopal and Venkaachalam (2006). In order o make he above disincion, we decompose earnings as he sum of cash flow and accruals: E = CF + ACC, where E is earnings, CF is cash flow, and ACC is accruals. The decomposiion of earnings ino pre-managed economic earnings and he remaining noneconomic earnings is moivaed in par by he previous lieraure ha documens a pervasive use of income smoohing echniques in financial reporing (Subramanyam, 996, and Graham e al. 2005). Following he prior lieraure (e.g. Francis e al. 2004, and Irvine and Poniff, 2005), we use operaing cash flow as a proxy for pre-managed economic earnings because i is less prone o managemen manipulaion. Taking he variance of boh sides, we have ρ CF ACC Var( E) = Var( CF) + Var( ACC) + 2, Var( CF) Var( ACC). Therefore, here are hree componens o earnings volailiy: earnings volailiy, cash flow volailiy and he conemporaneous correlaion beween cash flows and accruals. Following he lieraure, we use sandard deviaion o proxy for volailiy. Denoe Var ( E), Var( CF), Var( ACC) as EV, CFV and ACCV respecively. If as suggesed in Brown (200), managers have been able o mee earnings arges more frequenly in more recen years during 984 o 999, earnings volailiy will no increase over ime unless he volailiy of economic earnings has increased o an exen ha canno be offse by he recogniion of income-smoohing accruals. We inerpre CFV as operaing volailiy or economic volailiy from firms economic aciviies. We inerpre ACCV as firms financial reporing volailiy, as higher 7

9 variabiliy in accruals implies more shifs in iming beween earnings and acual cash flows. The inerpreaion of ρ is a subjec of debae. An empirical regulariy is ha ρ CF,ACC is negaive (see, e.g., Dechow, 994). The negaive correlaion ends o arise from he smoohing of income. The debae focuses on he moive of income smoohing ha causes ρ. Dechow (994) and Dechow, Kohari and Was (998) argue ha he correlaion reflecs he smoohing of emporary cash flow componens and is indicaive of improved earnings qualiy. On he oher hand, he correlaion could be due o managerial opporunism in managing accruals and is indicaive of deerioraed earnings qualiy (see, e.g., Myers and Skinner (2002) and Leuz, Nanda and Wysocki (2003)). Regardless of he moive, here is consensus ha ρ is a measure for earnings managemen. We inerpre ρ as he degree of earnings managemen firms ha manage earnings end o have accruals negaively correlaed wih cash flow. Togeher, ACCV and ρ are reaed as accouning (or financial reporing volailiy) componens of earnings volailiy. We herefore decompose earnings volailiy ino economic volailiy and financial reporing volailiy and invesigae he conribuion of each componen. A larger ACCV or a less negaive ρ indicaes higher financial reporing volailiy. However he earnings volailiy decomposiion is no sufficien o differeniae earnings volailiy from earnings qualiy. Since he DD earnings qualiy is he sandard deviaion of he residuals from a regression of accruals on pas, conemporaneous and fuure cash flows, we simulae accruals and cash flow series o gauge he link beween accrual volailiy, cash flow volailiy and earnings qualiy. Wysocki (2005) argues ha he DD earnings qualiy measure is primarily driven by he negaive correlaion beween 8

10 cash flow and accruals and reflecs opporunisic earnings smoohing by he managemen. We arrive a he similar conclusion we find ha he rend in he DD measure is predominanly driven by he rend in ρ bu no he rends in oher componens of earnings volailiy. Thus, despie he close correlaion beween earnings volailiy and he DD measure, he rend in he DD measure is relaed o he rend in only one componen in earnings volailiy, bu no o he general earnings volailiy rend. Esablishing ha he DD measure capures he financial reporing uncerainy bu no he economic volailiy of firms enables us o sudy he simulaneous impacs of earnings qualiy and earnings volailiy on reurn volailiy. Our cenral hypohesis in he paper is herefore ha he rend in reurn volailiy is collecively caused by he rends in earnings volailiy (and some of is componens) and in earnings qualiy. The res of he paper evolves from his cenral hypohesis. III. Sample Descripion III.A. Sample Selecion and Deseasonalizaion Our sample covers nearly hree decades of daa relaing o all NYSE/NASDAQ/AMEX lised firms during he ime period beween January 978 and December All our accouning variables are from he Compusa quarerly ape, and he reurn variables are from CRSP. The fiscal quarerly accouning variables are convered o calendar ime and mached o monhly reurn observaions o creae a panel daa se a a monhly frequency. Consisen wih prior sudies, we remove financial firms (SIC ) and uiliy firms (SIC ) since hese firms are regulaed. Following he lieraure (e.g., Wei and Zhang, 2006, and Francis e al., 2005), we use he sandard deviaion of a variable esimaed over a rolling window o proxy for is 9

11 volailiy. Specifically, we define earnings volailiy (EV) as he sandard deviaion of he raio of earnings before exraordinary iems o asses over he pas 2 quarers. The esimaion window of 2 quarers is also used in Wei and Zhang (2006). Cash flow volailiy (CFV), accrual volailiy (ACCV) and sales volailiy (SALESV) are defined analogously. These volailiy inpu measures are available for a large number of firms in Compusa from 978 onwards. Since cash flow reporing is required by he SEC only afer 988 (en years afer he sar of our sample period), we do no use his repored measure bu calculae cash flow as he sum of earnings, depreciaion, and change in working capial, where working capial is defined as per Richardson e al. (2005). Accruals are defined as earnings minus cash flow. We provide he descripion of hese and oher variables in Appendix. We use quarerly daa o increase he frequency of observaions. However, quarerly operaing variables such as earnings and cash flow display srong seasonaliy. We illusrae seasonaliy in he raw daa using earnings as an example. To deermine seasonaliy, we run he regression of curren quarer earnings on earnings of he previous four quarers for each firm. In our final sample of 6,074 firms, 2,346 firms, or 38.6% display significanly posiive auocorrelaion in quarerly earnings a he 5% significance level. In comparison, 70.2% of firms in he Compusa populaion (23,583 firms) have a significan fourh lag auocorrelaion in quarerly earnings. The seasonaliy in earnings also holds rue for he raio of earnings o asses (ROA). Given he presence of seasonaliy, esimaing sandard deviaion using non-seasonalized series would produce esimaes ha are biased upward. Our resuls are robus o esimaion windows of 6 and 20 quarers. 0

12 We adop he X procedure developed by he U.S. Bureau of Census o deseasonalize he following four operaing series for every firm: earnings, cash flow, change in working capial, and sales. Developed in 953, he X procedure has been used exensively in economics as a deseasonalizaion ool. In accouning, Broche, Nam and Ronen (2007) also use he X procedure o deseasonalize quarerly cash and accruals. 2 Afer he deseasonalizaion, he number of firms in our sample wih significanly posiive fourh-lag auocorrelaion reduces o,545, or o 25.3% of he sample. Table I provides a deailed descripion of our sample selecion. The implemenaion of he X procedure requires a leas 2 consecuive observaions and ha observaions are non-missing over ime. To mee hese requiremens, we resric he sample o firms ha have consecuive observaions in earnings, cash flow, change in working capial, and sales over heir lifeime on Compusa and furher remove firms ha fail o mee he 2 consecuive observaions hreshold. In order o carry ou ime-series analysis, we expand he quarerly observaions o monhly observaions. Our final sample consiss of 27,93 firm-monh observaions. [Inser Table I abou here.] III.B. Summary Saisics and Correlaion In Panel A, Table II, we repor he summary saisics for EV, CFV, ACCV, and ρ CF,ACC he correlaion beween cash flow and accruals for all firm-monhs during he sample period. All volailiy measures have been winsorized a he 0.5 and 99.5% perceniles. In Panel B, we repor he correlaion among EV, CFV and ACCV. The mean 2 We use he buil-in X procedure of he SAS sofware. Broche, Nam and Ronen (2007) provide a brief inroducion o he X procedure.

13 of he cash flow volailiy and he mean of accruals volailiy are greaer han ha of he earnings volailiy. The correlaions beween CFV and EV, and beween CFV and ACCV are high, whereas he correlaion beween EV and ACCV is much smaller. These correlaions sugges ha EV is more closely relaed o CFV han o ACCV. [Inser Table II abou here.] IV. Trend in Earnings Volailiy and Is Deerminans In his secion, we show ha earnings volailiy and is componens have been increasing over he pas hree decades. We hen invesigae he deerminans for earnings volailiy and is rend. IV.A. Graphical Analysis of he Trend Panel (a) of Figure plos he simple cross secional mean of EV, CFV, and ACCV over ime, denoed as EV, CFV, and ACCV respecively. In he remainder of he paper, we use he bar noaion for mean. We observe a seady increase in hese series before 2002, wih a shor period of decrease in he lae 980s and followed by a decline afer This paern is comparable o he findings of Brand e al. (2005) for idiosyncraic reurn volailiy and o he findings of Wei and Zhang (2006) for earnings volailiy. Among he hree series, CFV is always larger han EV, bu heir rends mach each oher almos perfecly. The upward rend in ACCV is only modes. As a resul, alhough ACCV sared ou almos as large as CFV a he beginning of he sample period, i becomes smaller han EV in he early 2000s. The difference in he rends beween CFV and 2

14 ACCV suggess ha he correlaion beween cash flow and accruals is rending down. This is indeed he case, as shown in Panel (b), Figure, where he ime series of he simple cross secional mean of ρ is ploed. We observe ha he rend of ρ is almos similar o he rend of CFV and EV : The negaive correlaion decreases unil 2002, wih shor periods of reversal in he lae 980s and afer Overall, ρ decreases from over 0.90 in he early 980s o 0.76 in [Inser Figure abou here.] IV.B. Regression Analysis of he Trend Figure shows he average rend. To formally esablish he rend for he enire sample, we run pooled regressions of earnings volailiy and is componens on a ime rend variable. We run wo specificaions of regressions, one for he pooled sample, and he oher by indusry. The regression equaions are: Variable, i, = α + β + ε i Variable = α + β + ε () i, j, j j i, j,, where i indexes he firm, j indexes he indusry, indexes he ime, and ε is he residual. Time, runs from 978:0, which is recorded as, o 2006:2, recorded as 348. We selec he five indusries defined by Kenneh French, namely consumer producs and services, manufacuring, high-ech, healh care, and ohers. 3 We es he rend of he following five variables: EV, CFV, ACCV, ρ, and CWCV, where CWCV is he volailiy in change in working capial, defined analogously as CFV. By definiion, he difference beween 3 The indusry definiion is available a Kenneh French s websie: hp://mba.uck.darmouh.edu/pages/faculy/ken.french/daa_library.hml. 3

15 change in working capial and accruals is depreciaion. As such, change in working capial can be reaed as operaing accruals. Table III repors he resuls from regressions (). [Inser Table III abou here.] Panel A of Table III repors he full sample resuls. We observe ha all hese series are significanly upward rending. Since he scalar used o calculae EV, CFV, ACCV and CWCV is he same (asses), if we assume ha over he long erm, hese series differ only in heir mean, heir rends are comparable o each oher. From Panel A, we observe ha he 4 rend in EV is he seepes, and i is similar o he CFV rend a around 0. The rends in ACCV and CWCV are similar and are much smaller han hose of EV and CFV. Noice ha he rend in ACCV is higher han he rend in CWCV, confirming he radiional wisdom ha operaing accruals are less volaile han oal accruals. The rend in ρ 4 is large in magniude wih an esimae of A he beginning of our sample period (978:0), ρ sared a The esimaed rend in ρ ACC predics ha CF, ρ CF,ACC a he end of he sample period (2006:2) would be , which is roughly he same as he realized value of These rend analyses indicae ha over he pas hree decades, earnings and cash flow became more volaile, and cash flow and accruals became less correlaed. The above paerns exend o he five indusries, as shown in Panel B: For each indusry, EV and CFV increase a a rae higher han ACCV and CWCV, and he correlaion beween accruals and cash flow becomes less negaive. The only excepion is he manufacuring secor ha has downward rending ACCV and CWCV. Noe ha here are 4

16 significan indusry differences. In paricular, high-ech and healh care indusries experience much higher rending in hese variables han he oher indusries. We nex es wheher he rend is saisically differen across EV, CFV and ACCV. To accomplish his, we run mulivariae ess. We es he following hree null hypoheses: () EV, CFV and ACCV have he same rend; (2) he rend in EV is equal o he rend in CFV; and (3) he rend in CFV is equal o he rend in ACCV. For he firs es, we run a wo-sided es and repor he associaed F saisics. For hypoheses (2) and (3), we run one-sided ess and repor he associaed -saisics. The es saisics are repored in he boom of Panel A, Table III. All hree null hypoheses are srongly rejeced a he % significance level. The one-sided es resuls sugges ha he rend in EV is greaer han ha of CFV, which is in urn greaer han ha of ACCV. Resuls of he ess by indusry are similar and are no repored. IV.C. Deerminans for Earnings Volailiy We now explore he deerminans of earnings volailiy. Earlier we decomposed EV ino CFV, ACCV and ρ. Insead of relying on he mahemaical equivalence ha earnings variance equal o he variance of he sum of cash flow and accruals, we are ineresed in he size of conribuion of CFV, ACCV and ρ o EV. To fully undersand heir conribuions, we need o conrol for oher elemens ha affec earnings volailiy. We argue ha aside from CFV, ACCV and ρ, oher facors ha impac earnings variabiliy include sales variabiliy, growh opporuniies, degree of leverage, and macro economic condiions. The choice of hese variables are based on Wei and 5

17 Zhang (2006), where he auhors use age, size, book-o-marke and leverage as conrol variables o explain changing reurn volailiy. The hypoheical sign of each of hese conrol variables on earnings volailiy is discussed below. Sales variabiliy can be hough as anoher proxy for flucuaions in operaing aciviies and is posiively associaed wih earnings volailiy. We use age, marke capializaion, and marke-o-book o proxy for growh opporuniies. Pasor and Veronesi (2002) argue ha younger firms end o have more growh poenials and herefore higher fuure uncerainy in earnings. By he same argumen, smaller firms end o have higher growh opporuniies and hus higher earnings volailiy. Marke o book is anoher proxy for growh opporuniies frequenly used in he lieraure (e.g., Gaver and Gaver, 993). Higher marke-o-book raio implies more growh opporuniies and herefore higher earnings volailiy. Leverage increases he equiy bea and herefore earnings volailiy. We use he real GDP growh rae o capure he macroeconomic condiion. French, Schwer and Sambaugh (987) show ha sock reurn volailiy increases in bad markes. We expec earnings volailiy responds asymmerically o GDP growh in a similar manner. Finally, we use indusry dummy variables o capure he poenial indusry differences in volailiy rends documened in Table III. The full model for earnings volailiy is herefore: EV i, i, 2 i, 3 i, + η LEVERAGE 5 = α + β CFV i, + β ACCV + η RGDPG β ρ 5 j= 2 γ D j i, j, + ε + η SALESV i,, i, + η AGE 2 i, + η SIZE 3 i, (2) + η MB 4 i, where age (AGE) is measured as he logarihm of he number of monhs a firm has been in CRSP, size (SIZE) as he logarihm of he marke value of equiy a he beginning of he monh, marke o book equiy (MB) as he beginning of he monh marke equiy o he 6

18 end of he monh book equiy, leverage (LEVERAGE) as he long-erm deb o book value of asses, RGDPG is he annual real GDP growh rae, and D i j,, is a dummy variable ha equals if firm i is in indusry j a monh and zero oherwise. Table IV presens he resuls for he following four specificaions of he above equaion: () using only he main variables (CFV, ACCV and ρ ) as he independen variables, (2) using boh he main and conrol variables (SALESV, AGE, SIZE, MB, LEVERAGE, RGDPG), (3) adding indusry dummies o specificaion (2), and (4) re-esimaing specificaion (3) using he sandard cross secional regression approach firs advocaed by Fama and MacBeh (973). Noe ha in specificaion (4), RGDPG is no longer suiable as an independen variable since i has no monhly variaion across firms. [Inser Table IV abou here.] The mos elling resul across all of he specificaions from Table IV is ha EV is posiively associaed wih CFV and ρ, and negaively associaed wih ACCV. Recall ha our primary inerpreaion for CFV is operaing volailiy, for ACCV is reporing volailiy, and for ρ is earnings managemen. The coefficien on CFV is posiive and close o one, which means ha a one-uni increase in operaing volailiy is almos fully refleced in repored earnings volailiy. Afer conrolling for operaing volailiy and he correlaion beween cash flow and accruals, repored earnings volailiy is negaively relaed o ACCV, indicaing ha earnings volailiy is dampened by financial reporing volailiy. A higher degree of earnings managemen means a more negaive correlaion beween cash flows and accruals and leads o less volaile earnings. This relaionship is verified by he posiive coefficien on ρ. 7

19 Among he conrol variables, he coefficiens for SALESV, LEVERAGE and MB are posiive and he coefficien for RGDPG is negaive, as expeced. We noe ha he signs for SIZE and AGE are ofen posiive in hese specificaions. A furher examinaion reveals ha he signs for SIZE and AGE are indeed negaive if he main variables are excluded. The resuls presened in Table IV sugges ha he size and age effecs are reversed afer conrolling for he earnings volailiy componens. Table IV shows ha CFV, ACCV and ρ all conribue significanly o earnings volailiy. However, from he regression resuls we canno ell which componen is he mos significan conribuor. In paricular, here are wo quesions. (), If ACCV dampens EV while CFV increases EV, would heir combined marginal effecs cancel each oher? And (2), how do we rank he overall conribuion of he hree componens? To answer quesion (), we run an F-es of he null hypohesis ha ˆ β ˆ = β 2. To answer quesion (2), we compare he values of ˆβ CFV, ˆβ ACCV 2, and ˆ β 3 ρ in onesided ess, where CFV is he sample mean of CFV as shown in Table II, and analogously for ACCV and ρ. ˆβ CFV can be hough of as he overall conribuion of cash flow volailiy o earnings volailiy, wih similar inerpreaions for ˆβ ACCV and 2 ˆ β ρ. 3 CF,ACC We es he above hypoheses wihin specificaion (3). Using oher specificaions would yield he same conclusions. The F-saisics for ˆ β ˆ = β 2 is 7,07, srongly rejec he null hypohesis ha he marginal effec of ACCV cancels ou he marginal effec of CFV. In he horse race of he overall effec, noe ha since 8

20 ˆ β CFV > ˆ β 3 ρ, ACC > ˆ β 2 CF ACCV, we need only o do wo one-sided ess ha ˆ β CFV = ˆ β 3 ρcf, ACC and ˆ β 3 ρ = ˆ β 2 ACCV. The wo one-sided ess have Z-saisics of 65.2 and 33.7 respecively, again srongly rejecing he nulls. 4 We conclude ha in ranking he conribuion o earnings volailiy, CFV makes he mos significan conribuion, followed by ρ and ACCV in ha order. In sum, our regression resuls indicae ha earnings volailiy is primarily driven by firms operaing volailiy. IV.D. Associaion beween he EV Trend and he CFV, ACCV and ρ Trends We have examined he overall conribuion of each componen o EV and our findings indicae ha CFV is he mos significan source of conribuion in he full sample. We now examine wheher he rend in EV is mos significanly affeced by he rend in CFV. We use wo mehods o esablish he associaion beween he EV rend and he rends in is componens. In he firs mehod, we examine he ime-series of he responsiveness of each EV componen o he formaion of EV. In he second mehod, we adop a ime-series es as in Wei and Zhang (2006). We illusrae boh mehods in deail below. We firs examine he ime-series rend of he coefficiens in he cross secional regression specificaion of Equaion (2) (Specificaion, 4). We focus on he monhly ime series of ˆβ, ˆβ 2 and ˆ β 3, namely he loadings of CFV, ACCV and ρ on EV 4 The Z-saisics are from he Wald ess for he corresponding wo-sided ess, which have a chi square disribuion. Since boh he null hypoheses have a degree of freedom of one, he square roo of he chi saisics follows a sandard normal disribuion. The one-sided ess can herefore be based on he square roo of he chi saisics, or he Z saisics. 9

21 respecively. These ime series can be inerpreed as he responsiveness of he EV componens o he formaion of EV afer conrolling for he oher variables. Figure 2 plos hese monhly series, as well as heir fied rend agains he ime variable,. We observe he hree series all show an increasing rend. Since ˆβ and ˆ β 3 are increasing in magniude and ˆβ 2 is decreasing in magniude, hese rends imply ha over ime, EV is more responsive o changes in CFV and ρ bu less sensiive o changes in ACCV. Furhermore, compared o he loadings of ACCV and ρ, he loading of CFV increases he mos over ime: he rend esimae for he CFV loading is subsanially larger han eiher he loading of ACCV or ρ. In unabulaed regressions, we es he difference of he rends among hese loadings. The es saisics reveals ha he rend in he loading of CFV is saisically larger han he rend in he loadings of ACCV and ρ CF,ACC. hese resuls sugges ha he rend in CFV is he mos imporan facor for he rend in EV. [Inser Figure 2 abou here.] One drawback of he above mehod is ha he differences in values among CFV, EV and ρ may make he direc comparison of heir loadings difficul, since he scale on which hese loadings are based is differen. We herefore augmen he firs mehod by he ime-series es in Wei and Zhang (2006). In he ime series es, we es he following regression: EV + η LEVERAGE 5 = α + β + β CFV o 6 + β ACCV + η RGDPG + ε, 2 + β 3 ρ + η SALESV + η AGE 2 + η SIZE 3 (3) + η MB 4 20

22 where variables. If EV is he cross secional average of EV a monh, and analogously for oher EV shows a rend, i should be picked up by (as shown in Regression ()). Furhermore, if here exiss some oher rending variable ha explains he EV rend, he loading of will be aenuaed by his rending variable. If a variable has no rend, hen i should no have any explanaory power on esimae of. EV and herefore no effec on he coefficien Table V presens he resuls. In Regression, we presen he raw ime rend of EV. Consisen wih Table III, EV is increasing over ime. The esimaes on are grealy reduced when we conrolled from oher rending variables. In Regression 2, we conrol for he EV componens. We observe ha he coefficien esimae on is reduced from in Regression o , a reducion of abou 85%. A he same ime, he adjused R 2 improves o almos 00%. A furher conrol of oher variables renders he esimae on insignifican, as shown in Regression 2. These resuls indicae ha he rend in EV can be explained by he conemporaneous rends in is componens: CFV, ACCV, and ρ. Consisen wih Table IV, CFV conribues posiively and ACCV conribues negaively o EV. The rankings of he marginal conribuion and overall conribuion are similar o hose of Table IV. Again, hese resuls sugges ha he CFV rend is he mos imporan conribuor o he EV rend. [Inser Table V abou here.] 2

23 V. The Trend in Earnings Qualiy and Earnings Volailiy This secion describes he rend in earnings qualiy and invesigaes is causes. Our primary focus is on he Dechow and Dichev (2002) earnings qualiy measure since i is closely relaed o earnings volailiy. V.A. Trend in he Dechow and Dichev (2002) Earnings Qualiy Measure Based on quarerly daa, he Dechow and Dichev (2002) measure of accrual qualiy is he sandard deviaion of residual from he following regression: i, q = α + βcfi, a + β 2CFi, q + β 3CFi, q+ + ε i q CWC, (3) where q indexes quarer and all variables are deflaed by asses and CWC is change in working capial. he essence of Equaion (3) is o measure he degree of maching beween operaing accruals and lagged, conemporaneous, and lead cash flows. The greaer is he residual noise, he poorer is he maching and hence he earnings qualiy. Using oal accruals (ACC) insead of CWC and annual daa in he above equaion, Rajgopal and Venkaachalam (2006) show ha earnings qualiy is deerioraing over ime. To mach Rajgopal and Venkaachalam (2006) s findings, we firs examine he earnings qualiy rend in our quarerly daa. To show he rend of earnings qualiy, we esimae he sandard deviaion of ε i, q for each firm using a rolling window of he pas welve quarers. We run he above regression in hree ways. In he firs mehod, we follow Dechow and Dichev (2002) o run a regression for each firm based on our full sample daa. This is Dechow and Dichev s (2002) primary earnings qualiy measure. We call his earnings qualiy measure he DD-Individual measure. In he second mehod, we follow 22

24 Rajgopal and Venkaachalam (2006) o run a pooled regression for he full sample. We call his earnings qualiy measure he DD-Pooled measure. In boh mehods, we rely on fuure informaion o calculae he rolling residual sandard deviaion, since he coefficien esimaes (and herefore he derived residuals) are based on full sample daa. We accoun for his look-ahead bias in he hird mehod, where we esimae a pooled regression of Equaion (3) using only daa from a rolling window of pas 2 quarers. We call his earnings qualiy measure he DD-Rolling measure. To mach he previous decomposiion of earnings o cash flow and accruals, we also replace CWC in Equaion (3) wih ACC and repea he ess. As in previous secions, in order o show rends, he quarerly earnings qualiy measures are hen expanded ino monhly series. We confirm he earnings qualiy rend documened in Rajgopal and Venkaachalam (2006). Figure 3 plos he DD measures. Panel (a) shows he rends in he DD measures when CWC is used as he dependen variable and Panel (b) shows he rend in he DD measures when ACC is used as he dependen variable. We observe he upward rends in all of he DD measures. Furhermore, hese earnings qualiy rends are similar o boh CFV and EV rends. In unabulaed regressions, we also regress hese earnings qualiy measures on ime. There is srong upward rend in he DD measures. In a nushell, he evidence suggess he deerioraion of earning qualiy over ime, as measured by he DD earnings qualiy measure. Since using ACC or CWC as he dependen variable and he hree earnings qualiy measures yields idenical earnings qualiy rends, o be consisen wih previous secions, in he remainder of he paper we focus on he ACC-based DD-Pooled measure only. [Inser Figure 3 abou here.] 23

25 V.B. Causes for he Deerioraing Earnings Qualiy DD shows ha EV is highly correlaed wih earnings qualiy a 0.82 for heir sample period of using annual daa. In our sample of quarerly daa during , he correlaion beween EV and DD-Individual is By way of comparison, we find ha he correlaion beween CFV and DD-Individual is much higher a This raises he following quesion: Wha causes he rend in earnings qualiy? More specifically, is he rend in CFV responsible for he rend in earnings qualiy? Answering hese quesions enables us o differeniae earnings qualiy from earnings volailiy and herefore differeniae Wei and Zhang (2006) and Rajgopal and Venkaachalam (2006). We address he above quesions hrough simulaions. We simulae he rends in ACCV, CFV, and ρ using he esimaed rend parameers shown in Table II. We generae a simulaed sample of panel daa wih 800 firms and 348 monhs, which is comparable o our regression sample of 27,93 firm-monhs. In he simulaion, ACC and CF follow a bivariae normal disribuion wih correlaion of ρ and sandard deviaions of ACCV and CFV respecively. For simpliciy, he bivariae normal disribuion is independen across ime. Using he simulaed daa, we run regression for Equaion (3) using he monhly frequency daa. To invesigae he causes for he earnings qualiy rend, we add he rends in CFV, ACCV, and ρ in a sepwise manner. We sudy he following four cases: () he benchmark case no rend in CFV, ACCV, and ρ ; (2) only an increasing rend in CFV; (3) increasing rends in boh CFV and ACCV, and (4) increasing rends in CFV and ACCV, and a decreasing rend in ρ. Noe ha case (4) is observed empirically and ha he rends in CFV, ACCV, and ρ are disinc from each oher. 24

26 Panel A of Table VI presens he regression resuls of he above four cases. The conemporaneous CF is always significan in explaining ACC, which is no surprising given he high correlaion beween he wo variables. We noice ha when he above volailiy rends are added sequenially, he mean DD measure (as measured by DD- Pooled) increases from Case () o Case (4). Mos noably, here is a significan jump from Case (3) o Case (4), i.e., when he rend of ρ is added. Compared wih Case (), he mean DD measure in Case (4) doubles. [Inser Table VI abou here.] We show he simulaed earnings qualiy rend of he four cases in Figure 4. We observe ha alhough uncondiionally, he mean DD measure increases from Case () o Case (4), he increasing rend happens only in Case (4), i.e., only when he rend of ρ is added in he simulaion. Economically, his makes sense. Since he DD measure is he goodness of fi beween accruals and cash flows, wha maers is he correlaion beween hese wo variables. If he correlaion decreases, he goodness of fi will decrease, resuling in a noisier residual erm and hence a larger DD measure. [Inser Figure 4 abou here.] In order o es for he saisical significance of he rend shown in Figure 4, we break he full simulaed daa o five sub-periods and run regression of Equaion (3) for each subperiod. Panel B of Table VI shows he mean of he DD measure for each subperiod and ess of he difference beween he subperiod DD measures across he five subperiods. In erms of magniude for he DD measure, Case (4) is clearly disinc from he oher cases: In each subperiod in Case (4), he DD measure is no only greaer han he 25

27 oher cases, bu also he upward rend is much greaer. The (one-sided) difference ess confirm ha he upward rend in Case (4) is sronger han he oher cases, in ha he - saisics are much larger. Noe ha in Cases () and (2), here is no saisical rend in he DD measure. In Case (3), here is a saisical rend bu he magniude of upward rending is very small. In summary, we make he following conclusions from Table VI and Figure 4. Firs, earnings qualiy is significanly differen from cash flow volailiy. Increases in cash flow volailiy alone do no cause changes in earnings qualiy. Second, changes in accrual volailiy only conribue marginally o changes in earnings qualiy. Third, a major conribuor o he rend in earnings qualiy is he rend in correlaion beween accruals and cash flow. Therefore, o correcly inerpre he rend of he DD measure, one should keep in mind he role of ρ.as emphasized by Wysocki (2005). VI. Why did Individual Socks Become More Volaile? This secion links our previous findings o sock reurn volailiy. Idiosyncraic sock reurn volailiy has been increasing over he pas four decades (e.g., Campbell, Malkiel, Leau and Xu, 200, and Morck, Yeung and Yu, 2000). Wei and Zhang (2006) aribue he rend in idiosyncraic reurn volailiy o increasing earnings volailiy. In a relaed paper, Rajgopal and Venkaachalam (2006) relae he rend in reurn volailiy o deerioraing earnings qualiy. The link beween he wo argumens however has no been explored. Given he high correlaion beween earnings qualiy and earnings volailiy documened in Dechow and Dichev (2002), one may wonder wheher hese wo argumens are indeed referring o he same facor. In he previous secions we 26

28 demonsrae ha he main deerminans for EV is CFV, ACCV, and ρ, and he main driver of he deerioraing earnings qualiy is decreasing ρ. In his secion we show ha boh earnings volailiy and earnings qualiy conribue o idiosyncraic reurn volailiy. VI.A. Trend in Idiosyncraic Sock Reurn Volailiy We use wo definiions of reurn volailiy: adj V i, Re ( raw V Re i, ) is defined as he mean of he pas hree years monhly sandard deviaion of excess (raw) sock reurns. Following Campbell e al. (200) and Wei and Zhang (2006), he monhly sandard deviaion is measured as he sandard deviaion of daily sock reurns wihin he monh imes he square roo of rading days wihin he monh. We use he hree-year mean o mach our horizon for he definiion of earnings volailiy and is componens. Figure 5 shows he ime series of cross secional mean of idiosyncraic reurn volailiy in our sample firms. We observe ha he rend of idiosyncraic reurn volailiy is similar o hose documened in hese sudies. Furhermore, he reurn volailiy rend is by and large similar o EV, CFV and DD: i shows seady increase in he 980s, a shor period of reversal in he lae 980s and early 990s, and seady decline afer The 987 marke crash inroduced a large spike in monhly reurn volailiy bu was evened ou in he hree-year moving average shown in he figure. Observe ha he rends in and adj V Re raw V Re are almos idenical. In he remainder of he paper we focus only on adj V Re. [Inser Figure 5 abou here.] 27

29 VI.B. Deerminans for Idiosyncraic Reurn Volailiy We now examine he deerminans of idiosyncraic reurn volailiy. We follow Wei and Zhang (2006) in choosing he following conrol variables: E (earnings divided by asses), Re (conemporaneous sock reurn), MB (marke o book equiy), AGE, SIZE, and LEVERAGE. We run he following regression wih he addiion of DD-Pooled (DD, he earnings qualiy measure) and earnings volailiy or is componens o he above regression: V Re adj i, + β DD = α + η E i, 2 + β EV i, i, + η Re 2 + ε i,, i, + η AGE 3 i, + η SIZE 4 i, + η MB 5 i, + η LEVERAGE 6 i, (4) In he above equaion, when a variable is only available on a quarerly basis (for example, DD), he - subscrip refers o he previous quarer insead of previous monh. Wei and Zhang (2006, Table 3) run heir primary regression by five subperiods. They show ha reurn volailiy is posiively associaed wih reurn volailiy lagged by one period, Re, and negaively associaed wih E across all subperiods. The conemporaneous posiive relaionship beween Re and adj V Re poins o a risk-reurn radeoff. Their subperiod resuls on oher conrol variables are no uniform, bu generally suppor a negaive associaion beween reurn volailiy and AGE, SIZE, LEVERAGE, and a posiive associaion beween reurn volailiy and BM. Table VII presens he regression resuls. In our firs specificaion, we reproduce he Wei and Zhang (2006) resuls using our sample and pooled OLS regression. In he second specificaion, we add boh DD and EV o he regression, and find ha boh earnings volailiy and earnings qualiy are significanly posiively relaed o reurn volailiy. Finally, in he hird specificaion, we spli EV ino is hree componens, namely 28

30 CFV, ACCV, and ρ and find ha all of he componens are significan in explaining reurn volailiy. We also conrol for indusry effecs in specificaion (4), and use he sandard cross secional regression echnique in specificaion (5). [Inser Table VII abou here.] As before, we inerpre CFV as firms operaing volailiy. The posiive loading on CFV confirms he poin made by Irvine and Poniff (2005), ha he explanaory power of EV on reurn volailiy is mainly driven by operaing volailiies. We inerpre DD, ACCV, and ρ as firms financial reporing choices. DD refers o earnings qualiy, ACCV o income smoohing, and ρ o earnings managemen. Resuls in Table VII indicae ha firms financial reporing choice also affec heir (idiosyncraic) reurn volailiy. Reurn volailiy is enhanced by poorer earnings qualiy, and dampened by heavier income smoohing and earnings managemen. If marke paricipans do no know he rue value of economic earnings and rely on repored earnings and cash flows o infer economic earnings, hen sock reurn volailiy is relaed o hese financial reporing choices. A furher examinaion of he resuls reveals several poins ha are noeworhy. Firs, he explanaory power of EV is subsumed by earnings qualiy. In specificaion 2, when DD is added, we find ha he loading on EV is much smaller (while everyhing else remains almos he same). Second, when EV is decomposed ino CFV, ACCV, and ρ, we find ha he loading of DD is smaller. This is because of he previously discussed relaionship beween DD and ρ, which reduces he explanaory power of DD. Noe ha alhough he rend in ρ is responsible for he rend in DD, his is no equivalen 29

31 o an uncondiionally 00% correlaion beween ρ and DD. In fac, he correlaion beween DD and ρ is only 24% in our sample. Therefore, in a regression using he pooled sample such as specificaion 3, we see he effecs of DD and ρ side by side. Finally, we rank he conribuion of operaing volailiy and financial reporing choice elemens o idiosyncraic reurn volailiy using he mehod previously described in Secion IV.C. Our ranking is in he order of: DD, CFV, ρ and ACCV. VI.C. Associaion of he Trends As before, we associae he idiosyncraic reurn volailiy rend wih boh he DD rend and earnings volailiy rend in he following regression: V Re adj + β DD = α + β + η E β EV + ε, + η Re 2 + η AGE 3 + η SIZE 4 + η BM 5 + η LEVERAGE 6 (5) where adj V Re is he cross secional simple average of VRe adj a monh, and similarly for oher variables. Table VIII presens he resuls. [Inser Table VIII abou here.] The resuls in Table VIII indicae ha he rend in idiosyncraic reurn volailiy is indeed explained by he rend in DD and EV. From Regression, where he only explanaory variable is, idiosyncraic reurn volailiy is increasing. When DD and EV are added o he regression, mos of he rend is explained away, as shown in Regression 2. Resuls wih EV decomposing o CFV, ACCV and ρ are weaker bu sill makes he rend less srong (Regressions 3 and 5). Taken as a whole, we find ha he idiosyncraic reurn volailiy rend is aribuable o boh EV and DD rends. 30

32 VII. Robusness Tess VII.A. Decomposiion of Earnings ino Non-discreionary Earnings and Discreionary Accruals Our previous resuls are based on a decomposiion of earnings ino cash flow and accruals on he basis ha cash flow reflecs economic operaions of he firm. Our cash flow measure is earnings adjused for accruals. This migh raise he quesion of he degree o which cash flow reflecs pre-managed earnings. Anoher popular measure for pre-managed earnings is non-discreionary earnings (NDE), ha is, earnings adjused for discreionary accruals (DA). We use he Jones model of Dechow e al. (995) o calculae DA following he mehod described in Kohari e al. (2005). In paricular, DA is he residual from he following regression: ACC i = β + β PPE + ε, 0 + β + β 2Δ( Sales) i, Assei, 3 i, i where PPE is propery, plans and equipmens, and ACC, Sales and PPE are all scaled by lagged asses. We use quarerly daa and run a rolling regression of he pas four quarers by indusries defined by he firs wo digis of SIC code o derive DA. NDE is defined as he difference beween earnings and DA. Afer we calculae DA and NDE, we similarly define NDEV (DAV) as he sandard deviaion NDE (DA) over he pas 2 quarers, and ρ as he conemporaneous correlaion beween NDE and DA. NDE, DA We noe ha previous resuls peraining o he decomposiion of earnings volailiy o CFV and ACCV holds. Specifically, (), he rends he rends in NDEV, DAV, and ρ NDE, DA are almos idenical wih he rends in CFV, ACCV and ρ, (2) he deerminans of EV is sill primarily driven by NDEV, and (3) DD and NDEV, DAV and 3