Are Accounting Standards Diversifiable? Evidence of the Aggregate Valuation Effects of Standards ΓF

Transcription

1 Are Accouning Sandards Diversifiable? Evidence of he Aggregae Valuaion Effecs of Sandards ΓF Bjorn N. Jorgensen, Jing Li, Gil Sadka April 3, 2009 Absrac Prior research documens ha individual sock reurns respond o earnings differenly under new accouning sandards, regulaions, or changes in enforcemen. This paper examines wheher his resul exends o he aggregae sock marke. We ake a macro perspecive and sudy he properies of aggregae earnings. Firs, we documen ha aggregae earnings and operaing income have remained relaively sable afer 1951, nowihsanding numerous changes in accouning sandards. Second, we documen ha he relaion beween aggregae sock marke reurns and aggregae earnings also appears sable over long ime periods. Third, we show ha he relaion beween aggregae earnings, GDP and indusrial producion has remained sable over ime as well. Finally, we show ha common earnings aribues (such as asymmeric imeliness, asymmeric earnings persisence, and predicabiliy of cash flows), while presen in firm-level analyses, disappear in he aggregae. Γ We would like o hank Marc Giannoni, Ami Khandelwal, Nellie Kim, Efraim Sadka, Jenny Tucker (discussan a FARS) and workshop paricipans a 2009 Financial Accouning Research Conference, Buron Workshop, Four School Conference, Harvard Universiy, Indiana Universiy, Universiy of Colorado a Boulder, Universiy of Missouri a Columbia, Universiy of Souhern California, and Washingon Universiy in S. Louis for valuable commens and suggesions. Any errors are our own. Bjorn is from Universiy of Colorado a Boulder, Jing is from Carnegie Mellon Universiy, and Gil is from Columbia Universiy. The auhors s are bjorn.jorgensen@colorado.edu, jl2491@columbia.edu, and gs2235@columbia.edu. 1 Elecronic copy available a: hp://ssrn.com/absrac=

2 1. Inroducion The accouning sandards issued by he Financial Accouning Sandards Board (FASB) are inended o resul in financial saemens ha inform a diverse group of users, 1 including equiy invesors.f F In a similar vein, he U.S. Securiies and Exchange 2 Commission (SEC) is mandaed o proec invesors.f F The informaion in financial saemens is supposed o help financial saemen users make beer decisions. Ye, a basic enan of finance heory is ha invesors should diversify. Therefore, i is worhwhile invesigaing from he perspecive of a diversified invesor wheher he informaional role of earnings is neural o changes in accouning sandards, regulaions, 3 and enforcemen o dae.f F Moreover, one can argue ha accouning informaion is aimed a providing useful informaion o sophisicaed invesors ha engage in sock picking. In his case, he marke porfolio provides an aggregaion of all such invesors and all pricing decisions. In addiion, he aggregae analysis allows invesigaion of he average effec on all socks. Cross-secional sudies, commonly used o es he implicaions of sandards, exclude by consrucion he average impac on all firms. Pu differenly, aggregae earnings measure he profiabiliy for a represenaive firm and he marke porfolio measures he reurn from invesing in ha represenaive firm. We noe ha he represenaive firm changes over ime. For example, during he 90s, a represenaive firm 1 Financial reporing should provide informaion ha is useful o presen and poenial invesors and crediors and oher users in making raional invesmen, credi, and similar decisions. The informaion should be comprehensible o hose who have a reasonable undersanding of business and economic aciviies and are willing o sudy he informaion wih reasonable diligence. FASB (1978). 2 The mission of he U.S. Securiies and Exchange Commission is o proec invesors, mainain fair, orderly, and efficien markes, and faciliae capial formaion. Hhp:// 3 Wha we mean by sandards includes no only SFAS, bu also he enforcemen by regulaors, such as he SEC and audiors. As deailed below, numerous prior sudies documen ha announcemens of changes in sandards can affec individual socks and he aggregae marke reurns (for a recen paper, see Zhang 2007). However, hese papers do no address he effec on aggregae earnings and heir informaion role. 2 Elecronic copy available a: hp://ssrn.com/absrac=

3 should have a larger invesmen in R&D. Since we use he aggregae earnings for each period, our analysis allows for he represenaive firm o change over ime consisen wih he changes in he economy. Prior lieraure conains ample evidence supporing he noion ha mandaory changes in accouning rules affec individual firms repored earnings. This lieraure furher documens ha affeced firms sock prices reac o he announcemen and adopion of new accouning sandards. As 4F noed above, mos of his lieraure uses cross-secional ess ha focus on he difference beween firms in he cross-secion and excludes he average effec on all firms. 5 In his paper, we invesigae simulaneously boh he aggregae effec of sandards o dae on all firms and wheher a diversified invesor s porfolio is neural agains hese changes in accouning sandards. Specifically, we focus on he represenaive invesor, or average invesor, who by consrucion owns he represenaive firm. To es he impac of accouning sandards on an average invesor s use of earnings, we sudy he effecs of accouning sandards on earnings and is relaion o he reurns of he marke porfolio. To our knowledge, our paper is he firs o es he implicaions of accouning sandards on aggregae earnings aribues as prior sudies focus on firm-level and cross-secional implicaions of sandards. To address our research quesion, we follow a recen lieraure ha aggregaes 6 from he firm-level of analysis o he aggregae level of analysis.f F Specifically, we documen for he aggregae level ha he relaion beween earnings (measured as 4 The effec of on an individual firm s repored earnings can arise eiher from differen accruals being repored, due o say changes in accouning esimaes, or from differen real operaing decisions. 5 By analogy, macro-economic analyses of he effec of individual household incomes have considered boh cross-secional disribuional effecs as well as he average or aggregae growh in income over ime. 6 See Kohari, Lewellen, and Warner (2006), Sadka (2007), and Hirshleifer, Hou, and Teoh (2008) among ohers. 3 Elecronic copy available a: hp://ssrn.com/absrac=

4 aggregae earnings growh) and marke reurns has no changed significanly since he Securiies Exchange Ac of We documen ha common earnings aribues such as asymmeric imeliness have no changed over ime. Specifically, we es wheher he earnings-reurns relaion has changed wih he incepion of he FASB. We find no evidence ha he aggregae earnings-reurns relaion has changed wih he incepion of he FASB. Consisen wih Kohari, Lewellen, and Warner (2006), our findings sugges ha aggregae earnings are negaively associaed wih conemporaneous aggregae sock reurns and posiively associaed wih lagged aggregae sock reurns. Using hisorical daa exraced from Wilson and Jones (2002), we find ha he aggregae earnings-reurns relaion is differen in he pre-sec period of relaive o he pos-sec period. We documen ha in he pre-sec period aggregae earnings are posiively associaed wih conemporaneous aggregae sock reurns and posiively associaed wih lagged aggregae sock reurns. The difference in he aggregae earnings-reurns relaion in he pre- and pos-sec period suggess ha while enforcemen has affeced he average invesor, accouning sandards have no. We cauion he reader ha he hisorical daa is likely less reliable han more recen daa. Also, he number of public firms included in our sample in he pre-sec era, for which we have daa, is smaller and may no represen he marke porfolio. One possible disadvanage of using he aggregae earnings-reurns relaion is ha 7 as earnings properies change, reurns migh change correspondingly.f F Therefore, alhough aggregae earnings have changed significanly, he aggregae earnings-reurns relaion could remain unchanged. To address his concern, we idenify an independen 7 Jamal and Sunder (2008) documen an increasing, bu sill relaively small, number (159) of FASB accouning sandards ha are complex relaive o oher areas of sandardizaion, such as, he Inerne Engineering Task Force. 4

5 esimae of economic aciviy ha does no change over ime GDP growh. This 8 measure is imporan as he GDP esimaes have no changed significanly since 1947.F F In conras, during his period, many new accouning sandards have been implemened, ye he relaion beween GDP growh and earnings growh has remained unchanged since Specifically, we find ha GDP growh is posiively associaed wih earnings growh. In addiion, his relaion has no changed significanly pre- and pos-incepion of he FASB. Since corporae earnings are a porion of GDP, we also include Indusrial Producion in our analysis and find similar resuls. Casual inspecion of aggregae earnings ime series reveals ha aggregae earnings were unusually low and high in 2001 and 2003, respecively. In conras, he ime series of aggregae operaing income does no exhibi hese wo recen years as ouliers. We herefore proceed o invesigae wheher any paricular income saemen line-iem beween operaing income and earnings leads o hese unusual paerns in he aggregae. We find ha asse impairmen of goodwill or, more generally, indefiniely 9 lived inangible asses, cause he unusual paerns.f F These wrie-offs inroduced wha appears o be a shor-lived aggregae shock o ne income as a resul of he implemenaion of SFAS 142. We follow Ely and Waymire (1999) and invesigae he effecs of nonrecurring iems by replicaing our es using growh in operaing income insead of earnings growh. Firs, we documen ha he associaion beween aggregae earnings and marke reurns is more pronounced when he years 2001 hrough 2003 are excluded han when 8 Noe ha 1947 is he earlies dae for which he Deparmen of Commerce produced esimaes of commodiy oupus. The esimaion of GDP and real GDP growh does change over ime. However, when he Bureau of Economic Analysis changes he esimaion i resaes he prior esimaes of GDP such ha he GDP esimaes are all based on he same esimaion procedure. 9 This was also recognized by he press a ha ime, see Sender (2002). 5

6 hese years are included. In conras, he associaion beween aggregae operaing earnings and marke reurns is consisen and independen of wheher he years 2001 hrough 2003 are included or excluded in he analysis. In fac, he relaion beween growh in aggregae operaing income and aggregae sock reurns remains consan during he enire sample period, consisen wih prices and invesors being unaffeced by SFAS 142. In oher words, invesors idenified he abnormal decline in aggregae earnings as a shor-run accouning effec. In addiion o GDP and aggregae sock reurns, we also es for changes over ime in some common earnings aribues, such as asymmeric imeliness. Basu (1997) documens, using cross-secional regressions, increased asymmeric imeliness pos- FASB. We re-apply he ess using firm-level ime-series regressions. We find evidence consisen wih an increase in firm-level asymmeric imeliness. In conras, asymmeric imeliness does no emerge for he aggregae marke boh pre- and pos Moreover, aggregae earnings are no imely. Invesors anicipae conemporaneous earnings changes during he prior year (refleced by he significan associaion beween aggregae earnings changes and prior-year aggregae sock reurns). Basu (1997) furher documens asymmeric earnings persisence for posiive and negaive earnings changes. While we find consisen evidence in firm-level ime-series ess, we fail o find consisen evidence for aggregae earnings changes. In addiion, we do no find a change in he aggregae earnings persisence and/or asymmeric persisence pos Finally, we es wheher earnings growh becomes a beer predicor of fuure cash flows over ime. While firm-level ess show ha earnings are informaive abou fuure cash flows, aggregae earnings do no predic fuure cash flows. In fac, i appears ha 6

7 aggregae earnings reflec pas cash flows beer over ime, bu hey do no predic fuure cash flows. A number of economics papers focus on he inroducion of he SEC on marke reurns, including Benson (1973), Simon (1989), and Sigler (1964). More recenly, Beardsley and O Brien (2003) find ha he sandard deviaion of sock marke reurns have changed wih he inroducion of regulaory changes in Ausralia, UK, and US. These papers, however, generally invesigae he effec of changes in accouning sandard seing and enforcemen wihou reference o or use of earnings in he analyses. In a noable excepion, Ely and Waymire (1999) compare he value relevance of earnings under various accouning sandard seers using he goodness-of-fi from cross-secional firm-level regressions. They find limied evidence in suppor of increasing value relevance around changes in accouning seers during he period. Our invesigaion exends heir analyses by sudying aggregae earnings properies over ime from 1871 o Consisen wih heir findings, we also find limied suppor of changes in aggregae earnings properies. One imporan cavea is ha his paper focuses on he informaional role of earnings in equiy markes and ha we do no address oher uses of accouning informaion, such as he sewardship role and conracing (Was and Zimmerman, 1986; Dye and Verrecchia, 1995; Holhausen and Was, 2001; Sivakumar and Waymire, 2003; Was, 2003a, 2003b; Ball, 2008; Ball, Robin and Sadka, 2008; Wienberg-Moerman, 2008). Our findings sugges ha while prior sudies documen ha sandards had a firmby-firm effec, he economy-wide effecs on he represenaive firm and on he represenaive invesor are negligible. Alernaively, i is possible ha accouning 7

8 sandards had an impac in erms of mainenance such ha wihou hese sandards he properies of aggregae earnings and heir relaion wih oher economic indicaors would have changed (Sunder, 1996). 1.1 Why Accouning Sandards Migh no Maer in he Aggregae There are hree reasons why accouning sandards migh no maer in he aggregae, while affecing firm-level earnings-reurns relaions. The firs possibiliy is ha regulaory agencies are incapable of wriing and enforcing sandards ha can influence he aggregae marke porfolio and invesors as a whole. While his argumen is possible, we doub ha his is he case. There is overwhelming evidence of significan firm-level impac. In addiion, he FASB is moving owards implemening fair value accouning. If his move persiss and expands, here should be an aggregae effec on invesors. Also, a leas in one case, SFAS 142, he FASB had an easily discernable effec, albei shor-lived, on aggregae earnings. The second possibiliy is ha sandards are generally diversifiable in he sense ha some firms gain while ohers lose, wih lile or no aggregae/average effec. For example, some firms lobby in favor of a proposed sandard while oher firms lobby agains. This paern implies ha while sandards have off-seing firm-level effec, hey need no have an overall impac on aggregae earnings. The hird possibiliy why sandards migh no maer in he aggregae is ha sandards o dae deal wih minor issues. In oher words, he sandards were no economically significan enough o have an aggregae effec. For example, he accouning lieraure has exensively sudied he implicaions of SFAS 123. However, 8

9 he overall value of opion awards (daa iem OPTION_AWARDS in Compusa Execuive Compensaion) is approximaely 0.5% of overall earnings for he firms in he sample. While i appears ha SFAS 123 had an impac in erms of conracing (employee compensaion), he overall sock opion expenses are a fairly small amoun compared o aggregae earnings. In sum, he aggregae earnings in he economy are generaed mosly from he firms regular operaing aciviy and he accouning sandards wih regard o recognizing he profis from hese aciviies have no changed significanly over ime. The remainder of he paper is organized as follows. Secion 2 analyzes he relaed lieraure and is implicaions for our sudy. Secion 3 describes he daa and is sources. Secion 4 repors our findings. Secion 5 concludes. 2. Relaed Lieraure This paper conribues o various sreams of research. Prior empirical capial markes research was mainly done a he firm-level (and mosly cross-secional). A growing empirical lieraure invesigaes wheher he firm-level findings also hold in he aggregae, economy level. For example, Bernard and Thomas (1990) invesigae he pos-earnings announcemen drif a he firm-level and porfolio-level. In conras, Kohari, Lewellen and Warner (2006) find no evidence of pos-earnings announcemen drif in he aggregae. Similarly, while Sloan (1996) documens he accrual anomaly a he firm-level, Hirshleifer, Hou, and Teoh (2008) find no evidence of he accrual anomaly in he aggregae. These resuls sugges ha marke inefficiencies, which persis a he firm level, disappear a he overall economy level. Oher papers ha ake a macroeconomic perspecive on accouning issues include Sadka (2007) on he 9

10 relaionship beween aggregae earnings and sock price volailiy and Anilowski, Feng, and Skinner (2007) on he macro-effec of aggregae managemen earnings guidance. In a similar vein, we ake as given he exan lieraure on firm-level analyses of he economic consequences of new accouning sandards. In a nice deparure from he sandard approach, Kohlbeck and Warfield (2008) simulaneously sudy muliple accouning sandards on firm-level reurns. In his paper, we underake aggregae-level analyses of he consequences of new accouning sandards. Specifically, we invesigae wheher he new accouning sandards affeced he porfolio held by a well-diversified invesor or in oher words, he aggregae porfolio held in he economy. Prior sudies on he effecs of accouning sandards es he implicaions of sandards for individual firms. These sudies are largely cross-secional and hus do no focus on he average effec, which is he impac on all invesors. Second, our paper relaes o he lieraure on how mandaory disclosure rules can change he iming of arrival of informaion. Saring wih Ross (1989), heoreical resuls in he finance lieraure predics no effec from early resoluion of uncerainy. While i is possible ha early mandaory disclosure has real effecs, Ross (1989) provides sufficien condiions under which he pricing kernel is unaffeced by he iming of disclosure. Therefore, individual sock prices could remain he same oday even wih a fully anicipaed change in he iming of a fuure disclosure. There are wo scenarios under which changes in disclosure iming can have an effec on individual sock prices. Firs, disclosure iming can have real effecs. For example, Hirshleifer (1971) shows how early disclosure can lead o a breakdown in insurance markes. Second, non-expeced uiliy may generae prices ha depend on he iming of disclosures and creae a preference for 10

11 he resoluion of uncerainy. Examples of models ha allow invesors o have nonsandard preferences ha differ from expeced uiliy include Kreps and Poreus (1978), Duffie, Schroder, Skiadas (1996), and Gran, Kajii, and Polak (2000). 3. Daa We use several sources of daa. In our iniial ess, we rely on Compusa and CRSP daa for he period 1951 hrough Our iniial sample consiss only of firms wih December fiscal year-ends o avoid misspecificaions due o differen reporing periods. We obain he reurn daa from CRSP monhly file and he annual reurn is measured by he cumulaive reurn from April of year unil March of year +1. The equal-weighed aggregae marke reurn is calculaed using all individual socks in our sample in each year. We also obain he ne income before exraordinary iems (Compusa Daa #18) and operaing income afer depreciaion (Compusa Daa #178) from Compusa indusrial annual file in each year. Second, we perform ess for a longer, hisorical ime-period covering he period from 1871 hrough 1999 using daa from Wilson and Jones (2002). This gives us annual sock reurn daa for Sandard & Poors (S&P) index as well as earnings for all firms in 10 ha index.f F In he analysis, we exclude 1933 as earnings growh exceeds 6, due o an increase in he number of firms included. Finally, we use Gross Domesic Produc (GDP) aken from Federal Reserve Economic Daa (FRED) as an alernaive non-gaap accouning measure of income 10 This index will include firms wih non-december fiscal year ends, which may resul in misspecificaions due o overlapping reporing periods for firms wih non-december fiscal year. Unlike our iniial sample, however, we did no aggregae ourselves and do no have access o he firm-level daa o only include firms wih December fiscal year end. 11

12 ha is unaffeced by changes in financial accouning rules governing privae companies, wheher hese rules are enforced by SEC, FASB, or oher sandard seing bodies. In fac, he benefi of using GDP as a measure of aggregae economic aciviy is ha is esimaion has no changed since One similariy beween aggregaed income based on financial accouning rules and GDP is ha GDP measures he value of he economy s oupu (goods and services) and earnings measure he profis generaed from he oupu when i is sold. There are, however, maerial differences in he eniy for which income is measured. For example, o avoid double couning, GDP only includes he marke value of final producs. In conras, our aggregaion of accouning income across companies will include revenue and expenses for inermediae goods and services ha would have canceled ou as inercompany ransacions if he economy had only consised of a single represenaive firm. Noneheless, his effec is likely minor because we consider accouning income ha is revenue ne of expenses. Second, GDP measures he value of he economy s oupu including invenories, while earnings only record he profis of unis ha are sold and mee he accouning rules wih regard o revenue recogniion. Thus, curren period GDP excludes marke ransacions based on exising goods and services, whereas accouning earnings can include gains or losses on sale of such goods and services. Third, curren GDP conains gross domesic invesmen o improve fuure producion, which includes he purchase of new propery plan and equipmen and changes in invenory as a par of revenue. Ne Domesic Produc deducs depreciaion from GDP bu is less generally available. 12

13 4. Summary of Analyses We firs used Compusa daa o creae wo accouning-based measures of aggregae income by aking he sum of all firms earnings and operaing incomes, respecively. Tha is, hese aggregae income measures derive from he cross-secional aggregae across all firms in he marke porfolio for each calendar year. We hen ploed he growh raes in hese wo aggregae earnings variables in Figure 1. Figure 1A shows ha aggregae earnings were unusually low in 2001 and unusually high in In conras, casual inspecion of Figure 1B reveals no paricular paerns in operaing income for hose years. We hen creaed an aggregae income saemen and found ha he reason for he unusual earnings growh is due o large wrie-offs in he aggregae (see Figure 4) ha creae he unusual aggregae earnings for years 2001 hrough We herefore run our subsequen analysis boh wih and wihou hose years. Figure 2 shows ha boh our aggregae earnings measures covary wih GDP. Figure 3 plos he aggregae earnings o sales raio and aggregae operaing income o sales raio. We observe ha operaing income o sales raio exhibis a relaively sable increasing paern over he period, while he aggregae earnings o sales raio is decreasing over he period. Table 1 offers descripive saisics for growh raes in our aggregae variables. Panel A presens he saisics using he daa sample from Compusa from 1952 o The means for earnings and operaing income are 12.3% and 11.1% respecively, and he medians for earnings and operaing income are quie similar a 12.5%. Bu he sandard deviaion for aggregae earnings is almos wice ha of operaing income, which suggess ha boom line earnings only inroduce more volailiy relaive o operaing income. The growh rae in GDP is relaively lower a abou 3.3%. The median of aggregae annual 13

14 equal-weighed reurn is abou 13% and he median of aggregae value-weighed reurn is abou 15%, consisen wih prior lieraure. Panel B presens he descripive saisics using he hisorical long ime-series daa. The average earnings growh is abou 6.6%, lower han ha in Panel A. GDP growh is similar o recen periods. 4.1 The Informaion Conen of Aggregae Earnings We firs es he ime-series relaion beween he aggregae earnings growh and conemporaneous and lagged reurn. Table 2 presens he OLS resuls of he following regressions for differen sample periods: R R u = ρ 0 + ρ1 X / X 1 + ε (1) u = 0 + ρ1 OI / OI 1 ρ + ε where R -u is he aggregae marke reurn in year -u, for u=0,1, X /X -1 he growh of aggregae earnings before exraordinary iems in year, and OI /OI -1 is he growh of aggregae operaing income afer depreciaion in year. Panel A repors he resuls for esimaing he regressions using he enire sample periods from 1951 o Panel B repors he resuls afer excluding he years from 2001 o 2003 due o he exreme earnings change repored during his periods. The aggregae earnings growh appears o be posiively correlaed wih conemporaneous reurn and no significanly correlaed wih lagged reurn. However, afer excluding 2001 o 2003, aggregae earnings growh is negaively correlaed wih conemporaneous reurn and posiively correlaed wih lagged reurn, which is consisen wih findings in prior lieraure on aggregae earnings reurn relaion (e.g., Kohari, Lewellen and Warner, 2006; and Sadka and Sadka, 2008). The coefficien of X /X -1 is - 14

15 0.48 wih -saisics of in he conemporaneous equal-weighed reurn regression and wih -saisics of in he lagged equal-weighed reurn regression. The regressions using value-weighed reurn also show similar resuls. The aggregae operaing income measure, however, is no affeced by he sample period selecion. The equal-weighed reurn regressions show ha aggregae operaing income growh is significanly negaively correlaed wih conemporaneous reurn and posiively correlaed wih lagged reurn, including or excluding 2001 o 2003 in he sample periods. Excluding 2001 o 2003 does increase he magniude (from o ) and -saisics (from o ) of he coefficien of operaing income growh in he conemporaneous regression. However, he lagged regression appears o show only a marginal decrease in he coefficien magniude and significance. The valueweighed reurn regressions show similar resuls, excep ha in he conemporaneous reurn regression he coefficien of aggregae operaing income growh, alhough negaive, is no saisically significan. Overall, operaing income regression has higher R 2 han he aggregae earnings regression. Panel C repors he resuls of aggregae earnings and reurn regressions using he long hisorical daa for sample periods before and afer SEC. Our analysis shows ha before 1932, he earnings growh is posiively relaed o boh conemporaneous and lagged reurns, while afer 1932 he earnings growh is consisen wih he findings using recen period daa. We proceed o es he relaion beween earnings (and operaing income) growh and GDP growh, which we consider o be a sable measure of economic aciviy. Table 3 presens he resuls of he following regressions: 15

16 F Excluding X / X ω + ε (2) 1 = 0 + ω1 GDP / GDP 1 OI / OI 1 = 0 + ω1 GDP / GDP 1 ω + ε Where X /X -1 is he growh of aggregae earnings before exraordinary iems in year, and OI /OI -1 is he growh of aggregae operaing income afer depreciaion in year. GDP /GDP -1 is he growh of real GDP in year. We conduc he regressions of aggregae earnings growh and aggregae operaing income growh on boh GDP growh and indusrial producion, using differen samples. The resuls show ha boh aggregae earnings growh and aggregae operaing income growh are significanly correlaed wih conemporaneous GDP growh. F 11 he sample period 2001 o 2003 significanly increases he significance of correlaion beween aggregae earnings growh and lagged GDP growh, bu does no change he magniude of coefficien much. In addiion o GDP growh, we use indusrial producion as well (e.g., Fama, 1990; and Schwer, 1990). The resuls are repored in Table 3. The resuls imply ha earnings growh posiively relaes o he growh in indusrial producion and ha his relaion is sable over ime. Consisen wih he resuls for GDP he relaion beween he growh in earnings and he growh in indusrial producion srenghens when he sample excludes he period 2001 o Panel C of Table 3 presens he resuls using he hisorical daa from Wilson and Jones (2002). The esimaion for he period shows he same resuls as in Panel A and B ha earnings growh is posiively and significanly correlaed wih conemporaneous GDP growh. Noe ha he Wilson and Jones (2002) daa uses a 11 Ball, Sadka and Sadka (2008) also find ha aggregae earnings shocks are posiively correlaed wih real GDP growh and indusrial producion. 16

17 differen ime period ( compared o for Compusa daa) and differen earnings (S&P 500 compared o he enire Compusa sample wih December fiscal year end). While boh he daa and he ime periods differ, he relaion beween aggregae earnings growh and GDP growh remains similar, furher supporing he conclusion ha accouning sandard had no discernable impac on he aggregae. The esimaion for he period , reveals similar resuls. The aggregae earnings growh is significanly and posiively correlaed wih conemporaneous GDP growh. This seems sugges ha SEC had lile impac on he relaion beween aggregae earnings and GDP measures. Bu noe ha here are fewer firms in he pre-1933 sample and differen GDP calculaion mehods. In fac, he GDP esimaes, exraced from Romer (1989), are reconsruced esimaes using a variey of assumpions. We now proceed wih examining wheher accouning sandard changes have affeced he documened earnings-reurn and earnings-gdp relaionships in he above ables. Specifically we include a dummy variable in he regressions in able 2 and able 3 o es he pre- and pos- FASB resuls. This dummy variable, D 1973, equals 1 for years afer 1973, he year Financial Accouning Sandard Board was formed, and 0 oherwise. Table 4 presens he resuls of reurn regressions on aggregae earnings growh and operaing income growh afer including he ineracion erm beween earnings growh (operaing income growh) and he dummy variable. The sample in able 4 excludes he years from 2001 o Panel A presens he resuls using equal-weighed reurns and Panel B presens he resuls wih value-weighed reurns. The coefficiens of he ineracion variable are no significan in boh he conemporaneous and lagged equalweighed reurn regressions on aggregae earnings growh (X /X -1 ). The ineracion 17

18 variable is also no saisically significan in he conemporaneous value-weighed reurn on aggregae earnings growh regression, bu moderaely significan wih -saisics of in he lagged reurn. This suggess ha accouning sandards do no significanly change he aggregae earnings reurn relaion. For he operaing income growh measure, he coefficiens of he ineracion variable are also no saisically significan in general, excep in he conemporaneous value-weighed reurn regression. The -saisic of he coefficien of he ineracion beween he operaing income growh and D 1973 is ; in sum in he pos-fasb period we do no find significan and consisen evidence on he changes in he relaion beween reurns and boh aggregae earnings and operaing income growh. Table 5 presens he resuls of regressions of earnings measures on GDP growh, afer including he ineracion beween he GDP growh variable and he dummy variable. Similar o Table 4, we find ha he relaion beween aggregae earnings (operaing income) growh and GDP growh does no change significanly before and afer FASB. The GDP growh remains significan in boh aggregae earnings and operaing income regression afer including he dummy variable. The ineracion beween GDP growh and D 1973 is no significan in he aggregae earnings growh regression. Bu he ineracion variable is marginally significanly negaive in he operaing income growh regression, wih a coefficien of and -saisics of We also perform he regression using indusrial producion and resuls are repored in Table 5 as well. The indusrial producion becomes slighly insignifican in he earnings growh regression, bu remains significan in he operaing income growh regression. The ineracions beween indusrial producion growh and he dummy 18

19 variable are no significan in boh earnings growh and operaing income growh regressions. 4.2 Asymmeric Timeliness and Conservaism Basu (1997) firs proposed running he following reverse regression of earnings on reurns: X j, / Pj, 1 = α 0, j + α1, jdr j, + β1, jr j, + β 2, jdr j, R j, + ε j, (3) Where DR j, = 1 if R j, < 0 and 0 oherwise, and oher variables are as defined previously. The posiive coefficien of β 2,j suggess ha bad news is more imely recognized in earnings han good news. Basu (1997) measures condiional conservaism based on he esimaed coefficiens in he above equaion and finds significan firm level resuls using he pooled and cross-secional regressions. Oher sudies, such as Ball, Kohari and Robin (2000), Bushman e al. (2004), Francis e al. (2004), and Ball, Robin and Sadka (2008), also use he R 2 of his regression as a measure of imeliness of earnings. Figure 5 plos he ime series of he coefficien of negaive news in Basu (1997) and he dispersion of aggregae earnings in Jorgensen, Li and Sadka (2008). Boh he asymmeric imeliness and earnings dispersion increase significanly afer 1973, suggesing ha earnings a firm level show significan changes pos FASB. We perform he Basu regression using aggregae earnings and aggregae marke reurn. Table 6 presens he resuls using he Compusa sample and he long hisorical daa sample. Overall here is no evidence of asymmeric imeliness in he aggregae daa. Panel A presens he aggregae level Basu es resuls using he equal-weighed reurn and Panel B presens he aggregae level Basu es resuls using he value-weighed reurn. 19

20 Boh he equal-weighed and value-weighed reurn regressions show ha he coefficiens of boh posiive reurn and ineracion wih negaive reurn are no significan in eiher aggregae earnings or operaing income regression. When we furher include he dummy variable D 1973 in he regression, he ineracion beween he dummy variable and negaive reurn is no significan. Furher, we also use he hisorical daa o run he Basu regression and he resuls are in Panel C and D. Panel C presens resuls using aggregae earnings growh as dependen variable, and here is no evidence of differenial asymmeric imeliness for he period afer FASB was formed relaive o he period from 1951 o Panel D presens he resuls using aggregae earnings scaled by beginning marke value and again here is no evidence of differenial asymmeric imeliness for he period before SEC relaive o afer he SEC was formed. For purpose of comparison, Table 7 repors he resuls of firm-level analyses of asymmeric imeliness pre-/pos FASB. For each firm we require a leas 20 observaions in eiher pre-fasb or pos-fasb period. This limis our sample size o 219 firms in pre- FASB period and 1079 firms in pos-fasb period. We run he firm level ime-series Basu regression and repor he descripive saisics of he regression coefficiens in Table 7. Panel A presens he resuls of aggregae earnings regression and Panel B presens he resuls of aggregae operaing income regression. The mean of coefficien β 1,j in Panel A is in pre-fasb period and in he pos-fasb period, and he wo-sample - es shows hese wo are no significanly differen from each oher wih -saisics (no repored in he able) of The mean of coefficien β 2,j shows a significan increase from in he pre-fasb period o in he pos-fasb period and he wo-sample -es also confirms ha he increase is saisically significan wih -saisics of 4.07 (no 20

21 repored in he able). Panel B shows ha he mean of coefficien β 1,j in he operaing income regression increases from in he pre-fasb period o in he pos- 12 FASB period and he increase is saisically significan.f F In conras, he mean of β 2,j decreases from o and he decrease is also saisically significan. The resuls confirm ha he asymmeric imeliness of firm level earnings increases significanly afer 1973, bu he operaing income s asymmeric imeliness does no. Indeed operaing income imeliness shows he opposie paern afer Basu documen ha earnings persisence is also asymmeric for posiive and negaive news. Similar o he asymmeric persisence regression in Basu (1997), we also perform he earnings persisence regression in he aggregae level. Specifically we perform he following regressions in able 8: ΔX ΔOI / P 1 = φ 0 + φ1δx 1 / P 2 + φ2δx 1 / P 2 * D + ε (4) / P 1 = 0 + φ1δoi 1 / P 2 + φ2δoi 1 / P 2 * φ D + ε where ΔX /P -1 and ΔOI /P -1 are he changes in aggregae earnings and aggregae operaing income for period, scaled by he beginning marke value. The dummy variable D equals 1 if he corresponding ΔX /P -1 or ΔOI /P -1 is negaive and 0 oherwise. Panel A presens he resuls of earnings persisence a he aggregae level. We find no evidence of asymmeric earnings persisence a he aggregae level. The coefficien φ2 in boh aggregae earnings and operaing income regressions is negaive a and respecively. The negaive coefficien is consisen wih Basu (1997) ha negaive earnings is more likely o reverse immediaely, however, i is no saisically significan wih -saisics of and We hen include he dummy 12 For more abou changes over ime in firm-level earnings-reurns relaion see Lev (1989) and Francis and Schipper (1999). 21

22 variable D 1973 in he regression and inerac D 1973 wih oher variables o es he pre and pos FASB effec. The resuls show ha he ineracions erm beween D 1973 and oher independen variables are in general insignifican. Indeed he ineracion beween ΔX -1 /P -2 D (and ΔOI -1 /P -2 D) and D 1973 shows a posiive coefficien of (1.038), which suggess ha pos FASB period here is even less evidence of low persisence of negaive earnings. Similar o Table 7, we also perform he ime-series regression of earnings persisence a he firm level and resuls are repored in Panel B of Table 8. We again require each firm has a leas 20 observaions o be included in he pre or pos FASB period sample. This gives us 213 firms in pre FASB period and 1027 firms in he pos FASB sample. In he persisence regression of oal earnings, he mean of coefficien φ 1, he persisence of posiive earnings, is in he pre-fasb period and 0.27 in he pos- FASB period, which suggesing ha earnings become more persisen in he pos-fasb period. The coefficien φ2 decreases from o and he decrease is saisically significan wih -saisics of 1.77 (no repored in he able). This shows ha a firm level he negaive earnings become more likely o reverse in he pos-fasb period, consisen wih increase in he conservaism. Overall, our resuls sugges ha even hough a firm level he earnings persisence becomes more asymmeric in he pos FASB period, his does no appear a he aggregae level. 4.3 Predicabiliy of cash flows In his secion, we es earnings predicabiliy of cash flows a he aggregae level and wheher earnings have become a beer predicor of cash flows in he pos-fasb period. 22

23 We use he following ses of models o es he relaion beween curren cash flows, curren earnings and fuure cash flows a he aggregae level: CFO + 1 CFO = 0 + λ1 X / X 1 + λ2cfo / CFO 1 / λ + ε (5) CFO + 1 / CFO = 0 + λ1oi / OI 1 + λ2cfo / CFO 1 λ + ε CFO CFO + 1 / TA = γ 0 + γ 1X / TA + γ 2CFO / TA + ε (6) + 1 / TA = 0 + γ 1OI / TA + γ 2CFO / γ TA + ε The firs se of models uses he growh measure of all variables. The second se of models is he same model as used in Dechow, Kohari and Was (1998) and Kim and Kross (2005) o es he relaionship beween fuure and conemporaneous cash flows and earnings. Dechow, Kohari and Was (1998) run a ime-series regression of he model for each firm, while Kim and Kross (2005) run a cross-secional regression of he model and examine he rend of explanaory power of earnings over ime. We conduc he regression analysis of boh models, wih and wihou including he ineracion erm wih he dummy variable D The resuls are presened in Table 9. Panel A in Table 9 presens he regression resuls of predicing fuure cash flow growh using Model (5). The resuls show a srong negaive auocorrelaion beween curren cash flow growh and fuure cash flow growh. Neiher earnings growh nor operaing income growh has an explanaory power in predicing fuure cash flow growh. The role of earnings (operaing income) growh does no change in he pre and pos FASB periods, as he ineracion erm beween X /X -1 (OI /OI -1 ) and D 1973 is no saisically significan wih -saisics of (0.234). The resuls also sugges ha he predicabiliy of curren cash flow growh on fuure cash flow growh does no change eiher in he pos-fasb period. 23

24 Panel B in Table 9 presens he resuls of predicing fuure cash flow using model (6). Oher firm level sudies (Kim and Kross, 2005; Dechow e al, 1998) usually find ha boh curren cash flows and earnings have explanaory power in predicing fuure cash flows. Bu he analysis in Panel B of Table 9 shows slighly differen resuls a he aggregae level: Earnings have explanaory power of fuure cash flow and curren cash flows comparaively do no have incremenal explanaory power beyond earnings; in conras, he regression using operaing income measure show ha operaing income does no have incremenal explanaory power of fuure cash flows beyond curren cash flows. Moreover, adding he ineracion erms beween he key variable and dummy variable D 1973 in he regression, we also ge similar conclusion as in Panel A ha earnings predicabiliy of fuure cash flows do no show significan change in he pos-fasb period. Indeed he coefficien of X / TA D 1973 is negaive, hough no significan, which suggess ha earnings have no become a beer predicor of fuure cash flows in he pos- FASB period. To furher es he pre and pos-fasb effec on he earnings relaion o cash flows, we also compare he accrual qualiy (Dechow and Dichev, 2002) in he pre and pos-fasb period. The accrual qualiy (Sd(v)) is measured as he sandard deviaion of residuals from he following regression model of oal accruals on he cash flows: TCA + 1 / TA = ζ 0 + ζ 1CFO 1 / TA + ζ 2CFO / TA + ζ 3CFO+ 1 / TA + ε (7) Panel C in Table 9 repors he accrual qualiy resuls. The accrual qualiy shows a significan increase in he pos-fasb period, wih Sd(v) decreases from o However furher analysis of he regression coefficiens indicae ha he 24

25 improvemen in he accrual s mapping ino cash flows comes from he pas and curren cash flows, no fuure cash flows. Finally, we also perform he ime series regression of firm-level cash flow predicabiliy and he resuls are repored in Table 10. Panel A and B repor he summary saisics of firm specific regressions of model (6) using earnings and operaing income measures, respecively. The resuls show ha a he firm level, earnings abiliy o predic fuure cash flows significanly increase in he pos-fasb period. The mean coefficien of earnings (γ 1 ) of firm-specific regressions increases from o 0.396; he unrepored wo sample -es shows he difference is saisically significan wih - saisics of The mean coefficien of operaing income increases from o 0.472, wih a -saisics of Panel C repors he summary saisics of firm-level accrual qualiy measure in he pre and pos-fasb period. The resul shows ha he accrual qualiy a firm level acually decrease, as measured by he increase of he sandard deviaion of residual from esimaion of model (7) using firm-specific ime series daa. However, he accruals abiliy o predic fuure cash flows a he firm level increases, as he coefficien of fuure cash flows increases from o wih an unrepored -saisics of Overall, he firm level resuls sugges a significan increase of earnings abiliy o predicing fuure cash flows, however, a he aggregae level, we do no find such significan change in he pos-fasb period. 4.4 Addiional Tess 25

26 We also perform several robusness ess using differen aggregae measures. We firs use aggregae earnings scaled by aggregae oal asses and aggregae earnings change scaled by aggregae oal asses respecively as he independen variable in he conemporaneous and lagged reurn regression. The resuls are consisen wih our finding ha here is no significan change in he earnings reurn relaionship in he preand pos-fasb period. In addiion, following he value relevance lieraure, we also run he regression of he equal-weighed and value-weighed reurn on he aggregae earnings and he aggregae book value, boh scaled by he aggregae beginning period marke value. Since he book value daa sars only from 1960, our daa is limied for he pre-fasb sample. However, we do no find significan change in he coefficiens of boh aggregae earnings and aggregae book value measures. 5. Conclusion This paper sudies he effecs of accouning sandards on earnings and heir aribues from he aggregae level poin of view of a well-diversified average invesor, who owns he represenaive firm. Our findings also have implicaions for sophisicaed invesors, engaging in sock picking, because he aggregae of all such invesors consiue (approximaely) he aggregae marke, i.e., he combined porfolio held by all sophisicaed invesors is approximaely he marke porfolio. In conras o firm-level sudies, we find ha he informaiveness of earnings o an average well-diversified invesor is largely unaffeced by changes in accouning sandards. Specifically, we do no find any significan changes in aggregae earnings aribues. In oher words, a 26

27 represenaive invesor who values he oal asses in he economy, appears from his perspecive o be unaffeced by he new accouning sandards ha were passed during our sample period. As a noable excepion, SFAS 142 significanly affeced aggregae earnings due o increased goodwill wrie-offs in 2001 and Furher analyses reveal ha while he sandard affeced aggregae earnings, he average invesor was largely unaffeced by he accouning change. Moreover, afer he iniial adopion period of his sandard, he relaion beween aggregae earnings and aggregae sock marke reurns appear o reurn o is pre-2001 relaion. We reierae ha our findings do no sugges ha changes in accouning sandards and pracice canno affec a diversified invesor. Firs, i is possible ha wihou he sandards he relaion beween earnings and reurns would have changed. Second, assuming ha sandards had no aggregae impac, our findings imply ha he average invesor was largely unaffeced by accouning changes o dae. Our focus is on he informaional role of aggregae earnings, or lack hereof, and our analysis does no consider oher uses of accouning informaion, such as sewardship and conracing roles. I is possible ha beer accouning pracices have conribued o overall economic growh hrough more efficien conracing and invesmens and hus indirecly affeced he diversified invesor as well. Ye, our findings sugges ha even if accouning pracices faciliaed growh, i was no because aggregae earnings provided imely informaion o invesors. 27

28 References Anilowski, C., M. Feng, and D.J. Skinner Is guidance a macro facor? The naure and informaion conen of aggregae earnings guidance. Journal of Accouning and Economics 44: Ball, R Wha is he acual economic role of financial reporing? Working Paper, Universiy of Chicago. Ball, R., and P. Brown An empirical evaluaion of accouning income numbers. Journal of Accouning Research 6 (Auumn): Ball, R., S.P. Kohari, and A. Robin The effec of insiuional facors on properies of accouning earnings. Journal of Accouning and Economics 29: Ball, E., A. Robin and G. Sadka Is financial reporing shaped by equiy markes or by deb markes? An inernaional sudy of imeliness and conservaism. Review of Accouning Sudies H13: H Ball, R., G. Sadka, and R. Sadka Aggregae earnings and asse prices. Working Paper, Columbia Universiy. Basu, S., 1997, The conservaism principle and he asymmeric imeliness of earnings. Journal of Accouning and Economics 24, Beaver, W.H., and S.G. Ryan Condiional and uncondiional conservaism: conceps and modeling. Review of Accouning Sudies 10: Benson, G. J Required disclosure and he sock marke: An evaluaion of he Securiies Exchange Ac of American Economic Review 63: Bernard, V., and J. Thomas Evidence ha sock prices do no fully reflec he implicaions of curren earnings for fuure earnings. Journal of Accouning and Economics 13: Bushman, R., Q. Chen, E. Engel, and A. Smih Financial accouning informaion, organizaional complexiy and corporae governance sysems. Journal of Accouning and Economics 37: Beardsley, C., and J. R. O'Brien The effeciveness of mandaory disclosure under Briain's Financial Service Auhoriy: An economic analysis. Working paper, Carnegie- Mellon Universiy. Dechow, P.M., and I. Dichev, 2002, The qualiy of accruals and earnings: The role of accrual esimaion errors. The Accouning Review 77 (Supplemen):

29 Dechow, P.M., S.P. Kohari, and R.L. Was, 1998, The relaion beween earnings and cash flows. Journal of Accouning and Economics 25: Duffie, D., M. Schroder, and C. Skiadas Recursive valuaion of defaulable securiies and he iming of he resoluion of uncerainy. Annals of Applied Probabiliy 6: Dye, R., and R. Verrecchia Discreion vs. uniformiy: Choices among GAAP. The Accouning Review 70: Ely, K., and G. Waymire Accouning sandard-seing organizaions and earnings relevance: Longiudinal evidence from NYSE common socks, Journal of Accouning Research 37: Fama, Eugene F., 1990, Sock reurns, expeced reurns, and real aciviy, Journal of Finance 45, Financial Accouning Sandards Board Conceps Saemen No. 1. Objecives of Financial Reporing by Business Enerprises (Issue Dae November 1978). Hhp:// Francis, J., R. LaFond, P.M. Olsson, and K. Schipper The coss of equiy and earnings aribues. The Accouning Review 79: Francis, J., and K. Schipper Have financial saemens los heir relevance? Journal of Accouning Research 37: Gran, S., A. Kajii, and B. Polak Temporal resoluion of uncerainy and recursive non-expeced uiliy models. Economerica 68: Hirshleifer, J The privae and social value of informaion and he reward o invenive aciviy. The American Economics Review 61: Hirshleifer, D., K. Hou, and S.H. Teoh Accruals and aggregae sock marke reurns. Journal of Financial Economics (forhcoming). Holhausen, R.W., and R.L. Was The relevance of he value-relevance lieraure for financial accouning sandard seing. Journal of Accouning and Economics 31: Jamal, K., and S. Sunder Monopoly and Compeiion: Sandard Seing in he Public and Privae Secor. Working paper, Yale School of Managemen. 29