Earnings Quality, Insider Trading, and Cost of Capital. David Aboody John Hughes

Transcription

1 Earnings Qualiy, Insider Trading, and Cos of Capial by David Aboody John Hughes and Jing Liu Mailing Address: 110 Weswood Plaza, Suie D403 Anderson School of Managemen Universiy of California Los Angeles Los Angeles, CA Aboody and Hughes are from he Anderson Graduae School of Managemen a UCLA. Liu is from UCLA and he Cheung Kong Graduae School of Business. We appreciae he research assisance of Mingshan Zhang and he helpful commens of an anonymous referee, Maureen McNichols, Rober Bushman, Paul Healy, Jun Pan, Krishna Palepu, Shyam Sunder and workshop paricipans a he Cheung Kong Graduae School of Business, Harvard Business School, Hong Kong Universiy of Science and Technology, London Business School, Sanford Summer Camp, Universiy of Iowa and he 2004 China Inernaional Finance Conference.

2 Earnings Qualiy, Insider Trading, and Cos of Capial Absrac Previous research argues ha earnings qualiy, measured as he unsigned abnormal accruals, proxies for informaion asymmeries ha affec cos of capial. We examine his argumen direcly in wo sages. In he firs sage, we esimae he firm s exposure o an earnings qualiy facor in he conex of a Fama-French hree facor model augmened by he reurn on a facor-mimicking porfolio ha is long in low earnings qualiy firms and shor in high earnings qualiy firm. In he second sage, we examine wheher he earnings qualiy facor is priced and wheher insider rading is more profiable for firms wih higher exposure o ha facor. Generally speaking, we find evidence consisen wih pricing of he earnings qualiy facor and insiders rading more profiably in firms wih higher exposure o ha facor. 1

3 1. Inroducion In his paper, we examine wo closely relaed issues regarding he cos of capial effecs of asymmeric informaion: i) wheher he sysemaic componen of asymmeric informaion is priced, and ii) wheher privaely informed raders earn greaer profis when rading socks wih higher exposure o an asymmeric informaion risk facor. There is a growing empirical lieraure on boh he cos of capial effecs of asymmeric informaion (e.g., Boosan [1997], Boosan and Plumlee [2002], Boosan, Plumlee, and Xie [2004], Healy, Huon, and Palepu [1999], Francis, LaFond, Olsson, and Schipper [2002], and Easley, Hvidkjaer, and O Hara [2002]) and he associaion of insider rading wih asymmeric informaion (e.g., Rozeff and Zaman [1998], Aboody and Lev [2000], Frankel and Li [2004], Pioroski and Roulsone [2004], and Ke, Huddar, and Peroni [2003]). 1 However, apar from a sudy by Bhaacharya and Daouk [2002] ha invesigaes a correlaion beween enforced insider rading laws and cos of capial across counries, i remains o be deermined if he prospec of privaely informed rading is wha drives a cos of capial effec of asymmeric informaion. 2 I is plausible ha he risk posed by privaely informed rading may be fully diversifiable by uninformed raders, implying no cos of capial effec (e.g., Hughes, Liu, and Liu [2005]). This paper differs from prior lieraure on insider rading (e.g., Frankel and Li [2004], Aboody and Lev [2000]) in ha we examine he relaionship beween insider 1 Oher sudies by Seyhun [1998] and Lakonishok and Lee [2001] documen profiabiliy of insider rading wihou linking such profiabiliy o specific measures of asymmeric informaion. 2 Enforced insider rading laws characerize an insiuional environmen in which one would predic a higher asymmeric informaion risk premium if insider rading is he underlying cause. As we describe below, earnings qualiy plays a similar role o enforced insider rading laws by characerizing an informaional environmen in which one would make he same predicion. 2

4 rading and he sysemaic (priced) componen of informaion risk, while he exising lieraure examine he relaion beween insider rading and firm specific characerisics. 3 Theoreical suppor for he exploiaion of privae informaion by informed raders as he explanaion for a cos of capial effec of asymmeric informaion comes from Amihud and Mendelson [1986], Admai [1985], Dow and Goron [1995], and Easley and O Hara [2004]. Oher heories ha do no rely on privaely informed rading are based on incomplee informaion (e.g., Meron [1986] and Shapiro [2002]) and esimaion risk (e.g., Barry and Brown [1985], Clarkson and Thompson [1990], and Clarkson, Guedes, and Thompson [1996]). Each of he above heories exploi he idea ha he common knowledge assumpion abou he mean-variance marix of asse payoffs adoped in neoclassical asse pricing heories may no hold; allowing for he prospec ha invesors will be differenially informed abou he asse payoffs leading o a cos of capial effec if diversificaion is incomplee. A disinguishing characerisic of heories ha rely on privaely informaion is he exploiaion of ha informaion by informed raders. While we canno rule ou he possibiliy of muliple explanaions for a cos of capial effec of earnings qualiy, a finding ha insiders profi from rading on he priced componen of earnings qualiy is consisen wih an explanaion based on asymmeric informaion. Our measure for idenifying firms for which privaely informed rading is likely o be more pronounced, and, hence, pose a greaer asymmeric informaion risk o uninformed raders, is earnings qualiy defined as unsigned abnormal accouning 3 This aspec is suble bu imporan. For example, size is a firm characerisic (idiosyncraic), herefore canno be a risk facor. To consruc a risk facor relaed wih size, one has o use facor mimicking porfolios a la Fama and French [1993]. 3

5 accruals. 4 We chose earnings qualiy as our measure of informaion asymmery for several reasons: he findings of Francis e al [2002] sugges ha earnings qualiy is priced; relaive o he cash flow componen of earnings, accouning accruals are more prone o managemen discreion and manipulaion implying less privae informaion may be pre-emped by earnings announcemens; earnings qualiy is a ubiquious consruc ha applies o all publicly raded firms; and unlike he probabiliy of informed rading measure used by Easley, Hvidkjaer, and O Hara [2002], an indirec measure of he firm s informaion asymmery derived from rade daa, earnings qualiy is a direc measure of he firm s informaion environmen derived from fundamenal accouning daa conained in is financial saemens. While we believe ha i is reasonable o use unsigned abnormal accruals o proxy for informaion asymmery, we are aware ha abnormal accruals can be highly correlaed wih growh (Zhang [2005]). Hence, here could exis alernaive inerpreaions of our empirical resuls. 5 To be consisen wih he idea of diversificaion in neoclassical asse pricing heory, following Francis, LaFond, Olsson, and Schipper [2002] and Easley, Hvidkjaer, and O Hara [2002], we esimae cos of capial using a facor model based on arbirage pricing heory (APT) from ex pos regressions. As in hose sudies, we use hree facors recommended by Fama and French (1993) augmened by a fourh facor based on earnings qualiy o capure asymmeric informaion risk. If asymmeric informaion risk 4 Specifically, we employ four measures relaed o he unsigned value of a firm s discreionary accruals as defined in Francis, e al [2002]. Two proxies are based on he modified Jones model (Dechow, Sloan, and Sweeney [1995]) and wo are based on he Dechow and Dichev [2002] model. Oher proxies appearing in he lieraure include he Associaion for Invesmen and Research s assessmen of corporae disclosure pracices (Boosan [1997] and Boosan and Plumlee [2002]), dispersion in analyss earnings forecass (Boosan, Plumlee, and Xie, [2004]), and analyss coverage (Healy, Huon, and Palepu [1999]). 5 For example, if growh proxies for some priced risk facor, hen he pricing resul we documen could be due o risk facors no relaed o informaion asymmeries. Furher relaing he cos of capial effec o informed rading profis helps o disinguish beween hese compeing explanaions, since i is difficul o esablish a link beween non-informaion based risk facors wih informed rading profis. 4

6 as measured by an earnings qualiy facor-mimicking porfolio is priced, hen his risk should command a posiive risk premium and greaer exposure o his risk as esimaed from facor loadings should imply greaer profis o privaely informed rading. Lasly, we employ rades by corporae insiders (officers, direcors, and principal sockholders) subsequenly publicly disclosed o he Securiies and Exchange Commission (SEC) as our measure of privaely informed rading. While insiders are represenaive of informed raders, we envision ha a similar informaion advanage exends o insiuions and oher professional raders, implying a sufficien impac on marke order flow o consiue a grea enough poenial risk o induce deecable risk premiums. Since hese rade daa are he measure of privae informaion we use in assessing insider rading profiabiliy, he naural window for purposes of esimaing such profiabiliy is from he rade dae o he SEC filing dae. 6 Accordingly, insider s rading profis are measured by abnormal reurns from he dae of rade o one day afer filing repors of hose rades o he SEC afer conrolling for all risk facors including he asymmeric informaion risk facor. Since boh he frequency of insider rades and he profiabiliy o each rade conribue o he oal profis earned by he insiders, we examine boh aspecs and find boh are posiively correlaed wih firm s exposure o he earnings qualiy (asymmeric informaion) risk facor. On he buy side, based on resuls for a hedge porfolio ha is long in he highes quinile of asymmeric informaion risk facor loading firms and shor in he lowes quinile, he difference in abnormal reurns o insiders ranges from 1.339% 6 Noe ha sudies ha explore insider rading profiabiliy over longer windows (e.g., Pioroski and Roulsone [2004], and Ke, Huddar, and Peroni [2003]) condiion heir ess on he assumpion ha informaion revealed ex pos was known o insiders ex ane. In conras, we condiion on jus he porion of insiders privae informaion conained in rades repored o he SEC. Marke efficiency in he semi-srong form dicaes ha one canno predic abnormal reurns o his informaion beyond he filing dae. 5

7 o 3.344% for an average holding period of 27 days. Sell side resuls have he correc signs, bu are smaller in magniude and weaker in saisical significance, a finding consisen wih he populariy of sock-based compensaion and he prospec ha insider sell ransacions are more likely han buy ransacions o be moivaed by diversificaion (e.g., Ofek and Yermack [2000]) and consumpion. The resuls for he pricing effec of earnings qualiy are relaively weak bu consisen. We find ha while he quinile porfolios wih highes exposure o he asymmeric informaion facor have saisically significan posiive inerceps (i.e., Jensen s alphas) afer conrolling for he Fama and French (1993) risk facors, he quinile porfolios wih he lowes exposure have alphas ha are insignificanly differen from zero. Resuls for he hedge porfolio have correc signs, wih esimaes ranging from 0.992% per monh o 1.178% per monh, bu weak saisical significance levels, poenially due o noise in he sysem from he low exposure firms, or he diversificaion effec (e.g., Hughes, Liu and Liu [2005]). The combinaion of hese findings is supporive of he conclusions ha asymmeric informaion risk, as measured by he firm s exposure o an earnings qualiy facor-mimicking porfolio, has measurable pricing effecs, and his risk is posiively associaed wih expeced insider rading profis. The remainder of his paper is organized as follows: In Secion 2, we describe how we measure earnings qualiy. In Secion 3, we describe he sample daa employed and provide descripive saisics. Our empirical analyses of he cos of capial effecs of asymmeric informaion risk and is associaion wih insider rading are conained in 6

8 Secion 4. We conclude in Secion 5 wih a summary of our resuls and a discussion of how hey relae o exan heory. 2. Earnings Qualiy We employ wo approaches for deriving measures of earnings qualiy ha are also used by Francis e al. [2002]. Following Dechow, Sloan, and Sweeney [1995], one is based on esimaes of abnormal accruals and he oher, following Dechow and Dichev [2002], is based on he exen o which working capial accruals map ino cash flow realizaions. Boh approaches rely on accouning fundamenals o separae accruals ino non-discreionary (normal) and discreionary (abnormal) componens. Earnings qualiy is defined as he absolue value of he abnormal componen. The larger he absolue value, he lower is earnings qualiy. Specifically, oal accruals (TA j, ), oal curren accruals (TCA j, ), and cash flow from operaions (CFO j, ) for firm j and year are calculaed as shown below: TA j, = ( CA j, CL j, CASH j, + STDEBT j, DEPN j, ) CA j, = ( CA j, CL j, CASH j, + STDEBT j, ) CFO j, = NIBE j, TA j, where: CA j, = firm j s change in curren asses (Compusa #4) in year, CL j, = firm j s change in curren liabiliies (Compusa #5) in year, CASH j, = firm j s change in cash (Compusa #1) in year, STDEBT j, = firm j s change in shor-erm deb (Compusa #34) in year, DEPN j, = firm j s depreciaion and amorizaion expense (Compusa #14) in year, and 7

9 NIBE j, = firm j s ne income before exraordinary iems (Compusa #18) in year. To esimae abnormal accruals (AA j, ) for firm j in year, we perform he following cross-secional regression for each of Fama and French s [1997] 48 indusry groups conaining a leas 20 firms in each year: TA 1 REV PPE Asse Asse Asse Asse j, j, j, = k1, + k2, + k3, +ε j, (1) j, 1 j, 1 j, 1 j, 1 where: REV j, = firm j s change in revenues (Compusa #12) in year, PPE j, = firm j s gross value of propery, plan and equipmen (Compusa #7) in year, and we have deflaed by firm j s oal asses in year -1 (Asses j, -1, Compusa #6). We hen use he indusry-year-specific parameer esimaes from (1) o esimae firm-specific normal accruals ( NA j, ) for firm j in year as a percen of lagged oal asses; i.e., where: ( Α ) REV R PPE NA k k k Asse Asse Asse 1 j, j, j, j, = 1, + 2, + 3, j, 1 j, 1 j, 1 AR j, = firm j s change in accouns receivable (Compusa #2) in year. In urn, abnormal accruals ( AA j, ) for firm j in year are AA j, TA Asse j, = NAj,. j, 1 The absolue value of abnormal accruals ( AA j, ) is he firs earnings qualiy measure (EQ1), wih larger values indicaing lower earnings qualiy. Similar o oal accruals, we esimae abnormal curren accruals ( ACA j, ) using he following variaion of (1): 8

10 TCA 1 REV Asse Asse Asse j, j, =γ 1,τ +γ 2,τ + vj,. (2) j, 1 j, 1 j, 1 We use he parameer esimaes from equaion (2) o calculae each firm s normal curren accruals as a percen of lagged asses, NCA j, ( REV, ΑR, ) 1 j j 1, 2, j, 1 j, 1 =γ +γ, Asse Asse and hen calculae he abnormal componen for firm j in year as shown below; ACA TCA =. Asse j, j, NCAj, j, 1 Our second earnings qualiy measure (EQ2) is he absolue value of he abnormal curren accruals ( ACA j, indicae poorer earnings qualiy. ) calculaed as shown above. Similar o EQ1, larger values of EQ2 The wo remaining measures of earnings qualiy proceed from esimaes of oal curren accruals based on cash flows from operaions, again employing Fama and French s [1997] 48 indusries for each year wih a leas 20 firms: TCA CFO CFO CFO Aveasse Aveasse Aveasse Aveasse j, j, 1 j, j, + 1 = θ0, + θ1,τ + θ2,τ + θ3,τ + vj, (3) j, j, j, j, where: Aveasse j, = firm j s average oal asses over years and -1. Our hird earnings qualiy measure (EQ3) is he absolue value of he firm s residual ( v j, ) from equaion (3) and our fourh earnings qualiy measure (EQ4) is he ime-series sandard deviaion of hese firm-specific residuals (σ( v j,,)) calculaed using a minimum of five residual observaions. Consisen wih he consrucion of he oher 9

11 measures, larger absolue residuals and larger sandard deviaions of residuals are inerpreed as lower earnings qualiy. 3. Sample and Descripive Saisics We gaher accouning and cash flow daa from COMPUSTAT and price daa from CRSP. The insider rading daa are obained from he CDA/Invesne par of Thompson/Firs Call. The daabase conains all buy and sell ransacions made by insiders and repored o he SEC from January, 1985, hrough November, We focus on open marke buy and sell ransacions ha were conduced by direcors and officers of he firm, as officers are a priori presumed o have insider informaion. We esimae each of he four earnings qualiy measures annually over years , yielding 989,530 firm-monh observaions for EQ1 and EQ2, 910,477 observaions for EQ3, and 614,981 observaions for EQ4. Table 1 conains descripive saisics of he variables used in our analysis. INSERT TABLE 1 ABOUT HERE I is eviden from Panel A of Table 1 ha sample disribuions of EQ1 and EQ2 are highly skewed, wih means being subsanially higher han medians. This suggess he presence of ouliers in he posiive exremes, caused parially by deflaion using oal asses when oal asses for some firms are close o zero. 7 Consisen wih Francis e al. [2002], o reduce he influence of ouliers in he calculaion of correlaions, we rimmed he sample a 1% and 99% in Table 1 Panels B and C. In all oher pars of our sudy, oulier reamens are no necessary since all analyses are done a a porfolio level. Inspecing insider buying and selling, we find ha insiders sell more frequenly han hey 7 In secion 4.4, we discuss robusness checks on his issue. 10

12 buy, and he raio of he sell o buy ransacions is close o 2 o 1. This is reasonable, because insiders, ofen compensaed in socks, have incenives o sell heir company shares in order o diversify heir personal wealh (e.g., Ofek and Yermack [2000]) and consume. Moving o he correlaion marix, we find he firs wo earnings qualiy measures based on he modified Jones model (Dechow, Sloan, and Sweeney [1995]) are highly correlaed, wih a Spearman rank order correlaion coefficien a However, he wo measures based on Dechow and Dichev s [2002] model are no as highly correlaed, nor are hey highly correlaed wih EQ1 and EQ2. This suggess ha EQ1 and EQ2 are poenially capuring similar informaion consrucs abou he firm, bu ha EQ3 and EQ4 are capuring somewha disinc consrucs. As a benchmark, similar o Francis e al. [2002], we esimae a four-facor assepricing model by adding an earnings qualiy facor o he Fama-French [1993] hreefacor model. In paricular, we perform ime series regressions for he following model: ( ) R R = α + β R R + δ SMB + σ HML + φ EQ + ε (4) j, f, j j m, f, j j j j, where R j, is firm j s sock reurn; R f, reasury bill rae; R m, weighed index; SMB and facor reurns, respecively; and is he risk free rae, measured as one-monh is he marke porfolio reurn, measured using CRSP value HML are he Fama-French [1993] size and marke-o-book EQ is he hedge porfolio reurn going long in he exreme quinile of low earnings qualiy firms and going shor in he exreme quinile of high earnings qualiy firms. 8 In order o make sure ha observaions peraining o all 8 We obained he Fama-French [1993] facor-mimicking porfolio reurns from Ken French s websie: hp://mba.uck.darmouh.edu/pages/faculy/ken.french/daa_library.hml. 11

13 variables on he righ-hand-side of (4) are publicly available ex ane, we consruc he EQ facor mimicking porfolio using he realizaions of earning qualiy measures in he preceding fiscal year. We run ime series regressions on each firm using daa from January, 1985, o November, The mean and median facor loadings are repored in Panel C. The resuls are broadly consisen wih hose repored in Francis e al. [2002]. The consisenly posiive facor loading on he earnings qualiy facor-mimicking porfolio indicaes ha, like Fama-French [1993] facors, asymmeric informaion risk seems o be priced by he marke. However, since posiive loadings do no in hemselves imply a non-zero risk premium, we move o he nex secion o invesigae wheher he asymmeric informaion risk as capured by our earnings qualiy proxies is priced. 4. Empirical Analyses 4.1 Earnings Qualiy, Insider Trading, and Abnormal Reurns In his secion, we esablish ha he earnings qualiy measures are effecive proxies for informaion asymmery ha serves as a source of rading advanage exploied by corporae insiders. Because he purpose of his secion is o verify wheher he EQ measures proxy for informaion used by corporae insiders, we calculae he EQ measures conemporaneously based on annual financial saemens, and hese measures are only parially observable o he invesing public from previously released quarerly financial repors. Corporae insiders are hypohesized o know more abou hese measures han he general public. In secion 4.2 and 4.3, where we invesigae he sysemaic componen of 12

14 informaion risk, we only use publicly available informaion and consruc he facor mimicking porfolios based on EQ measures of he preceding fiscal years. To begin, in each year, we pariion firms ino quinile porfolios based on conemporaneous esimaes of earnings qualiy, and consruc a hedge porfolio ha is long in he lowes quinile of earnings qualiy firms and shor in he highes quinile. Similar o prior sudies on insider rading (e.g., Aboody and Lev [2000], Pioroski and Roulsone [2004], and Ke, Huddar, and Peroni [2003]), we use he earnings qualiy measures as proxies for informaion available o insiders, bu no o ouside invesors. To he exen ha hese measures are effecive proxies for informaion asymmery, we should be able o deec differenial insider rading profis across he quinile porfolios, i.e., insider rading profis should be negaively correlaed wih earnings qualiy measures. We examine insider rading by furher pariioning each quinile porfolio based on earnings qualiy ino hose firms in which monhly insider rades are ne buys and ne sells and esimaing abnormal reurns using he Fama-French [1993] hree facor model applied o excess porfolio reurns: 9 ( ) R R = α + β R R + δ SMB + σ HML + ε. (5) p, f, p p m, f, p p p, To measure insider profis, in each monh, we compue a firm-specific mean reurn from he ransacion daes of insiders rades o one day afer he filing dae of hose ransacions wih he SEC. These firm-specific ransacion-o-reporing reurns are averages over all he individual insider rades ha occurred during he monh. We ake 9 We classify a firm as a ne buyer if he number of shares bough by all insiders during he monh exceeds he number of shares sold. The number of ransacions, number of shares, and oal value of ransacions are highly correlaed. Replicaion of he ess wih dollar value of ransacions yielded very similar resuls o hose derived from he number of shares. 13

15 he average of rades because in many firm-monhs insiders have muliple rades. In a firm-monh where insiders rade a opposie direcions, we sill ake he average ransacion-o-reporing reurns, and he average reurn is classified as a buy (sell) if he ne ransacion is a buy (sell) in number of shares. Firm-monhs wih no rades are excluded from he analysis. 10 We group insider rading according o he ransacion daes raher han filing daes, e.g., if here are wo rades occurred in he same calendar monh, and one rade is filed wihin ha monh and he oher rade is filed in he subsequen monh, we assign boh rades o he former monh. This grouping causes some ineviable calendar ime mismaching beween componens of porfolio reurns, as well as mismaching beween porfolio reurns and Fama-French facor reurns. We believe poenial biases from such mismaching are minimal since he mean inerval beween ransacions and filing daes is only 27 days. Such mismaching may be viewed as a source of noise, and, herefore, likely o reduce he power of ess, implying a bias agains finding significan resuls. We run ime series regressions for hese porfolios based on 227 monhs of daa, from 1/85 o 11/03, o esimae he inercep and facor loadings. Because insider buying and selling are measured monhly, in effec our porfolios are rebalanced every monh. In addiion o he monhly rebalancing due o insider rading, he porfolios are also rebalanced by he end of each fiscal year o reflec changes in earnings qualiy. Inasmuch as he esimaed inerceps for he five-quinile porfolios are generally (wih minor excepion) monoone for all four earnings qualiy measures, o save space, we only repor he resuls for he exreme porfolios and he hedge porfolio in Table 2. INSERT TABLE 2 ABOUT HERE 10 We examine he no-rade sample as a benchmark. Resuls are discussed in secion

16 The facor loadings on he hree Fama-French [1993] risk facors are qualiaively similar o he benchmark case repored in Panel C of Table 1. However, hey differ in magniudes. Some differences are expeced since Table 2 is based on porfolio regressions while Panel C of Table 1 is based on firm-by-firm regressions. The mos noable and sysemaic differences in porfolio loadings occur for he hedge porfolios. These loadings on he hree Fama-French [1993] facors are, for he mos par, closer o zero as we would expec given he long and shor rading sraegy of he hedge porfolios. Our primary ineres is wih Jensen s alphas (inerceps), α p, as a measure of excess reurns for each combinaion of insider ransacion ype, earnings qualiy porfolio, and earnings qualiy measure, afer conrolling for he Fama-French [1993] risk facors. As expeced, he esimaed inerceps from ime series regressions of he difference in reurn beween low earnings qualiy firms and hose wih high earnings qualiy on he hree sysemaic facors are all posiive when insiders bough shares. For he hedge porfolio, he esimaed inercep for EQ1 is 1.342% (-saisic of 2.16), for EQ2 is 2.408% (-saisic of 3.00), for EQ3 is 1.282% (-saisic of 1.88), and for EQ4 is 0.912% (-saisic of 1.04). 11 While he resuls for EQ1, EQ2 are saisically significan a convenional levels, he resul for EQ3 is only marginally significan, and he resul for EQ4 is insignifican, suggesing he firs hree measures are beer proxies for informaion advanages experienced by insiders. Anoher plausible explanaion for he insignifican 11 The -saisics are convenional measures based on he assumpion of zero serial dependence. In all ess here and laer in Table 4, we also calculae sandard errors using he Newey-Wes procedure o adjus for poenial serial dependence. The resuls are quie close due o negligible serial dependence in sock reurns a he monhly frequency. 15

17 resuls for EQ4 is he maerial reducion in sample size ha occurs because he esimaion of EQ4 requires five years of daa. 12 Moving o insider sell ransacions, we observe much weaker resuls. For example, regardless of he earnings qualiy measure used, here are no discernable negaive abnormal reurns for he high or low earnings qualiy porfolios. Resuls for hedge porfolios are slighly sronger; while he abnormal reurns for he hedge porfolios ha are long in low qualiy firms and shor in high qualiy firms are negaive for all EQ measures, none of he esimaes are saisically significan. The esimaed inerceps for he hedge porfolios are 0.316% for EQ1 (-saisic of 0.99), 0.180% for EQ2 (saisic of 0.55), for EQ3 (-saisic of -0.31) and for EQ4 (-saisic of ). To summarize resuls in Table 2, we find evidence (primarily from buy ransacions) ha insiders rade more profiably in low earnings qualiy firms han high earnings qualiy firms. In addiion o heir saisical significance, he gains o insider buys are economically significan. The annualized gains o insiders in buy ransacions are 16.10% for EQ1, 28.89% for EQ2, 15.38% for EQ3, and 10.94% for EQ4. The weaker resuls for sell ransacions are no unexpeced. Because corporae insiders usually have disproporional financial capial and human capial invesed in heir firms, hey have srong incenives o sell heir company socks and diversify. Moreover, o he exen heir compensaion is in he form of sock hey have incenive o sell in order o consume. The selling driven by diversificaion and consumpion moives plausibly overshadows he selling driven by privae informaion, resuling in weaker resuls documened in Table Noe ha he number of observaions per regression is considerably reduced from hose indicaed for our full original sample by pariioning on earnings qualiy and direcion of insider rades. 16

18 Having esablished ha earnings qualiy measures are effecive proxies for informaion asymmeries exploied by corporae insiders, in Secion 4.2 we examine wheher informaion asymmeries have pricing implicaions. The fac ha insiders exploi heir informaion advanages in a muli-asse seing does no guaranee any pricing implicaions because he price effecs could be fully diversifiable (e.g., Hughes, Liu and Liu [2004]). In order o assess pricing implicaions, we focus on he sysemaic componen of asymmeric informaion risk and invesigae wheher he firm s exposure o he sysemaic componen of asymmeric informaion risk, as measured by he loadings on he earnings qualiy facor mimicking porfolio esimaed using he four-facor model inroduced in Secion 3, is priced, i.e., refleced in expeced reurns. Finding a priced componen o asymmeric informaion risk, in Secion 4.3 we invesigae wheher insiders exploi informaion advanages refleced in ha componen by earning abnormal reurns afer conrolling for he Fama-French [1993] risk facors as well as for he newly consruced asymmeric informaion risk facor. We hypohesize ha insiders abnormal rading gains are posiively correlaed wih he firm s exposure o he sysemaic componen of asymmeric informaion risk. 4.2 Premium on Asymmeric Informaion Risk In order o measure he firm s exposure o he sysemaic componen of asymmeric informaion risk, we again add an asymmeric informaion risk facor o he Fama-French [1993] hree-facor model mimicked by a porfolio ha is long in he quinile of socks wih he lowes earnings qualiy and shor in he quinile wih he highes qualiy as in equaion (4). To obain an esimae for each firm-monh, we use he pas 36 monhs of daa o esimae facor loadings hrough monhly ime series 17

19 regressions. As previously noed, he esimaed regression coefficien or loading on he earnings qualiy facor-mimicking porfolio, φ, measures he firm s exposure o he sysemaic componen of asymmeric informaion risk. This measure is differen from measures based on earnings qualiy per se in ha he non-priced idiosyncraic componen is filered ou. j To verify ha he sysemaic componen of asymmeric informaion risk is priced, we form quinile porfolios according o a firm s exposure o ha componen (facor loading) in each monh, and run he Fama-French [1993] hree-facor model o esimae Jensen s alphas. Because facor loadings are esimaed every monh, he porfolios are in effec rebalanced every monh. We repor resuls for measures of exposure based on each of he four earnings qualiy proxies: he quinile of firms wih he highes φ j ( Hφ ), he quinile of firms wih he lowes φ j ( Lφ ), and a hedge porfolio where we sell firms in he Lφ quinile and buy hose in he Hφ quinile. Our resuls are conained in Table 3. INSERT TABLE 3 ABOUT HERE The resuls in Table 3 are broadly consisen wih he noion ha he sysemaic componen of informaion risk commands a posiive risk premium. While he low exposure firms have alphas no significanly differen from zero, he high exposure firms all have saisically significan posiive alphas. As a resul, he hedge porfolio ha is long in Hφ firms and shor in Lφ firms has posiive alphas across he four EQ measures, implying a posiive risk premium on he EQ facor in each case. The esimaes are all economically significan, wih values of 1.031% for EQ1, 1.016% for EQ2, 0.992% for EQ3 and 1.178% for EQ4. The saisical significance for he hedge porfolio is less impressive all -saisic esimaes for he hedge porfolio are below wo. Combining he 18

20 srong resuls for he Hφ firms and he weaker bu consisen resuls for he hedge porfolios, we conclude ha i is likely he noise in he sysem, primarily from he Lφ firms, 13 ha reduced he significance level for he hedge porfolio resuls. This finding also illusraes he imporance of examining he premium on he informaion risk facor explicily. Francis e al (2002) found ha firms have highly significan exposures o he informaion risk facors and hus concluded ha he risk facors are priced. Our resuls show ha he evidence is in fac weaker han one migh have surmised from facor loading esimaes alone. Of course, his difference in conclusion is only a a quaniaive level. In he nex secion, we show ha he alpha spread beween Hφ and Lφ firms is posiively correlaed wih insider rading profis, a finding ha furher suppors he noion ha he sysemaic componen of asymmeric informaion risk facor is priced. 4.3 Sysemaic Asymmeric Informaion Risk, Insider Trading and Abnormal Reurns As indicaed earlier, if privaely informed rading lies behind he pricing of asymmeric informaion risk, hen insiders should earn abnormal rading profis afer conrolling for all risk facors, including he sysemaic componen of asymmeric informaion risk. Furhermore, we expec insiders rading profis o be posiively correlaed wih firms loadings on he earnings qualiy facor-mimicking porfolio. 14 To demonsrae hese poins, we now urn o Table Esimaes of alpha have larger sandard errors for Lφ firms han for Hφ firms. 14 We noe ha here is considerable noise conained in insider rading daa. In heory, a horough es would require consideraion of addiional phenomena such as oher moives for insider rading and insiders privae porfolio holdings of he firm s sock and oher asses, ec. Unforunaely, daa on hese addiional issues are generally no available. 19

21 INSERT TABLE 4 ABOUT HERE Table 4 is similar o Table 2 excep ha now jus he sysemaic (priced) componen of he firm s asymmeric informaion risk is he focus of our invesigaion. We esimae he firm s exposure o he sysemaic componen of asymmeric informaion risk in he same way as in Table 3. For each monh, we firs sor firms ino buy and sell porfolios by checking wheher he insiders are ne buyers or ne sellers. Then, wihin each buy or sell porfolio, we furher pariion firms ino five quiniles according o heir exposure o he asymmeric informaion risk facor as measured by loadings on ha facor. Finally, we run he four facor model given by equaion (4) for each porfolio generaing regression inerceps (Jensen s alphas) ha now represen abnormal rading profis o insiders afer conrolling for all four risk facors. We repor resuls for he boom and op quiniles, as well as for a hedge porfolio ha is long in he op quinile and shor in he boom quinile. We find srong evidence insiders profi by exploiing he sysemaic porion of heir privae informaion for buy ransacions. Judging by he hedge porfolio resuls, all esimaes of abnormal reurns (Jensen s alphas) are highly saisically significan wih he excepion of resul based on EQ4. Their significance is similarly srong in he economic sense. For example, insiders make 3.344%, 1.827%, 1.676% and 1.339% monhly abnormal profis (afer conrolling for known risk facors) for hedge porfolio reurns based on EQ1, EQ2, EQ3, and EQ4 measures of earnings qualiy, respecively. Comparing his se of resuls wih hose repored in Table 2, we noe ha boh he abnormal reurns and he significance levels are generally higher in Table 4. The fac ha he resuls are sronger in Table 4 suggess ha insider rading is associaed wih he 20

22 sysemaic componen of he asymmeric informaion risk, and parsing ou he sysemaic componen via a facor model is imporan empirically. Resuls are weaker for insider sell ransacions. Alhough all four EQ measures produce EQ facor loadings wih he correc signs and significance a convenional levels, none of he Jensen s alphas are significan. These resuls sugges ha insiders buy ransacions are more likely o have been based on privae informaion han heir sell ransacions. The above resuls are based on a sample where here are insider rades wihin he firm-monhs. However, for he majoriy of he firm-monhs in our daase, here are no insider rades. To benchmark our analysis, we repea he same analysis as in Table 4 on he sample wih no insider rades. We expec he Jensen s alphas o be close o zero for his sub-sample since he risk facors we conrol for include he EQ facor. The resuls confirm our expecaion; he poin esimaes for Jensen s alphas are very close o zero, wih esimaes of 0.016%, 0.012%, 0.006% and 0.011% for EQ1, EQ2, EQ3 and EQ4 respecively. Anoher robusness issue follows from he skewness of earnings qualiy measures repored in Table 1. In as much as he earnings managemen lieraure generally concludes ha all abnormal accruals measures are noisy, i is reasonable o quesion wheher our resuls are an arifac of ouliers. To alleviae concerns, we consruc new samples by excluding firms wih oal asses below $50 million or rimming he sample a he op 5% and boom 5% of EQ, and repea he analysis in Table 3 and 4. We find qualiaively similar resuls in all hese sensiiviy ess. In sum, he resuls from our ess of an associaion beween he firm s exposure o he sysemaic componen of asymmeric informaion risk as measured by loadings on an 21

23 earnings qualiy facor-mimicking porfolio and gains o insider rading provide supporing evidence ha asymmeric informaion risk premiums are a consequence of insider rading. The weakness of he findings on he sell side of insider rading is consisen wih he greaer likelihood of mixed moives for insiders o sell. 4.4 Trading frequencies Esablishing a relaionship beween abnormal reurns o insiders from he dae hey rade o he dae hose rades are made public and a firm s exposure o asymmeric informaion risk is of firs-order imporance in he sense ha failing o find such a relaionship precludes furher inquiry on he prospec ha insider rading is driving he risk premium on our earnings qualiy facor. Having found evidence of a relaionship, i remains o be shown wheher he higher abnormal reurns associaed wih high exposure o he earnings qualiy facor are accompanied by higher frequency of insider rades, implying ha insiders are, indeed, exploiing heir informaion advanage o realize greaer profis. Accordingly, we exend our analysis o incorporae relaive rading frequencies. Specifically, we examine he connecion beween insider rading frequencies and firms exposure o he earnings qualiy facor by comparing numbers of monhly insider rades for porfolios sored on firms exposure o he EQ facor. Table 5 provides srong evidence ha insiders rade more frequenly for high exposure firms ( Hφ ) han low exposure firms ( Lφ ): he differences are highly significan for all four EQ measures as well as for boh buy and sell ransacions. Because of he concern ha saisical significance migh be oversaed due o cross-secional correlaion, we also apply an alernaive mehod similar o Fama and Macbeh [1973] o measure he sample saisics 22

24 and associaed significance levels: insead of measuring he global pooled saisics, we firs measure he means and medians for each monh, hen rea hese numbers as a ime series and measure he mean of he mean series and he median of he median series. Significance levels are hen measured based on hese wo ime series of daa. We employ boh a parameric -es of differences in means, and a non-parameric Wilcoxon ranksum es. The resuls are largely consisen wih he pooled resuls, bu weaker in significance levels as one would expec from such a es design. 5. Concluding Remarks The research quesions addressed in his paper are i) wheher he sysemaic componen of earnings qualiy (a proxy for informaion asymmery) is priced, and ii) wheher privaely informed raders earn greaer profis when rading socks wih higher exposure o an earnings qualiy risk facor. Our resuls are consisen wih an affirmaive answer o boh quesions. Measures of earnings qualiy employed in he paper include wo based on he modified Jones model (Dechow, Sloan, and Sweeney [1995]) and wo based on he Dechow and Dichev [2002] model. In he conex of a muli-facor model of asse pricing, we use hese proxies o form earnings qualiy facor-mimicking porfolios ha, in urn, allow us o filer ou he non-priced componen of earnings qualiy. Exposure o he priced componen of earnings qualiy is measured by loadings on he earnings qualiy facor in ime series regressions of a Fama-French hree facor model augmened by he earnings qualiy facor. We find ha asymmeric informaion risk premiums are significanly posiive for firms wih high facor loadings, insignificanly differen from zero for firms wih low facor loadings, and posiive bu no significan for he hedge 23

25 porfolio ha is long in he high exposure firms and shor in he low exposure firms. These resuls sugges ha sysemaic componen of earnings qualiy is priced, bu he evidence is relaively weak, a resul no available from porfolio loadings evidence as repored in prior lieraure (Francis e al [2002]). Pariioning sample firms by he direcion of insider (officers, direcors, and major sockholders) rades subsequenly repored o he SEC, as well as by loadings on he earnings qualiy facor, we find srong evidence ha insiders rade more profiably in firms wih high facor loadings. The combinaion of he findings is supporive of he conclusions ha asymmeric informaion risk as proxied by earnings qualiy is priced, and insider rading is an imporan elemen in esablishing his effec. Given he srong resuls on insider rading, he relaive weak resul for he pricing of earnings qualiy is inriguing. One plausible explanaion is ha he laer ess have low power due o large amoun of noise in he sysem. Anoher equally plausible explanaion is ha diversificaion eliminaes some pricing effecs of asymmeric informaion risk. Hughes, Liu and Liu [2005] resuls sugges ha he cross-secional effec of asymmeric informaion on cos of capial may be fully diversified away in a pure exchange economy wih a large number of asses. I seems inuiive ha, in large economies, informaion mus direcly affec asse payoffs hrough producion and invesmen decisions in order for asymmeric informaion o ener as a separae facor in he deerminaion of cos of capial. An explici model o incorporae such ideas is no available in he lieraure and warrans fuure research. Reversing he perspecive, our resuls also sugges ha here is some ruh o he noion, popular boh in he accouning lieraure and on Wall Sree, ha earnings 24

26 numbers have differen qualiies. The fac ha unsigned abnormal accruals as proxies for earnings qualiy can generae our resuls implies ha accruals are an imporan characerisic of a firms informaion environmen, in paricular, an indicaion of asymmeric informaion risk emanaing from privaely informed rading. From his perspecive, our resuls sugges ha invesors canno fully unravel accruals and ha he exercise of discreion by insiders serves o reduce he effeciveness of public earnings announcemens as a device for resolving asymmeric informaion risk and miigaing gains o insiders whose rades conribue o he creaion of ha risk. 25

27 References: Aboody, D., and B. Lev. Informaion Asymmery, R&D, and Insider Gains. The Journal of Finance 55 (2000): Admai, A. A noisy raional expecaions equilibrium for muli-asse securiies markes. Economerica 53 (1985): Amihud, Y., and H. Mendelson, Asse pricing and he bid-ask spread. Journal of Financial Economics 17 (1986): Barry, C., and S. Brown. Differenial informaion and securiy marke equilibrium. Journal of Financial and Quaniaive Analysis 20 (1985): Boosan, C. Disclosure level and he cos of equiy capial. The Accouning Review 72 (1997): Boosan, C., and M. Plumlee. A re-examinaion of disclosure level and he expeced cos of equiy capial. Journal of Accouning Research 40 (2002): Boosan, C.; M. Plumlee; and Y. Xie. The Role of Privae Informaion Precision in Deermining Cos of Equiy Capial. Review of Accouning Sudies (2004): forhcoming. Clarkson, P., and R. Thompson. The empirical esimaes of bea when invesors face esimaion risk. Journal of Finance 45 (1990): Clarkson, P.; J. Guedes; and R. Thompson. On he diversificaion, observabiliy, and measuremen of esimaion risk. Journal of Financial and Quaniaive Analysis 31 (1996): Dechow, P.; R. Sloan; and A. Sweeney. Deecing earnings managemen. The Accouning Review 70 (1995): Dechow, P., and I. Dichev. The qualiy of accruals and earnings: The role of accrual esimaion errors. Accouning Review 77 (2002): Easley, D.; S. Hvidkjaer; and M. O Hara. Is informaion risk a deerminan of asse reurns? Journal of Finance 57 (2002): Easley, D., and M. O Hara. Informaion and he cos of capial. Journal of Finance (2004): forhcoming. Fama, E.F., and K.R. French. Common risk facors in he reurns of socks and bonds. Journal of Financial Economics 33 (1993):

28 Fama, E., and K. French. Indusry coss of equiy. Journal of Financial Economics 43 (1997): Francis, J.; R. LaFond; P. Olsson; and K. Schipper. The marke pricing of earnings qualiy. Working paper, Duke Universiy and Universiy of Wisconsin, Frankel, R., and X. Li. The characerisics of a firm s informaion environmen. forhcoming, Journal of Accouning and Economics. Healy, P.; A. Huon; and K. Palepu. Sock performance and inermediaion changes surrounding susained increases in disclosure. Conemporary Accouning Research 16 (1999): Hughes, John, Jing Liu and Jun Liu. Informaion, diversificaion and cos of capial. Working paper, UCLA, Ke, B.; S. Huddar; and K. Peroni. Wha insiders know abou fuure earnings and how hey use i: evidence from insider rades. Journal of Accouning and Economics 35 (2003): Lakonishok, J., and I. Lee. Are insider rades informaive? Review of financial sudies 14 (2001): Meron, R. A simple model of capial marke equilibrium wih incomplee informaion. Journal of Finance 42 (1986): Ofek, E., and D. Yermack. Taking sock: equiy-based compensaion and he evoluion of managerial ownership. Journal of Finance 55 (2000): Pioroski, J., and D. Roulsone. Do insider rades reflec boh conrarian beliefs and superior knowledge abou fuure cash flow realizaions? forhcoming, Journal of Accouning and Economics. Ross, S. The arbirage pricing heory of capial asse pricing. Journal of Economic Theory 13 (1976): Rozeff, M., and M. Zaman. Overreacion and insider rading: evidence from growh and value porfolios. Journal of Finance 53 (1998): Seyhun, H. Invesmen inelligence from insider rading. Cambridge, MA: MIT press, Shapiro, A. The Invesor Recogniion Hypohesis in a Dynamic General Equilibrium: Theory and Evidence. Review of Financial Sudies 15 (2002):

29 Sloan, R. Do sock prices fully reflec informaion in accruals and cash flows abou fuure earnings? The Accouning Review 71 (1996): Spiegel, M., and A. Subrahmanyam. Informed Speculaion and Hedging in a Non- Compeiive Securiies Marke. Review of Financial Sudies 5 (1992): Zhang, F. X. Wha causes he accrual anomaly growh or earnings persisence? Working paper, Universiy of Chicago,

30 Table 1 Summary saisics The sample consiss of 989,530 observaions from 1985 o EQ1 and EQ2 are earnings qualiy measures based on he Modified Jones model (Dechow, Sloan and Sweeney, 1995). EQ1 i is he absolue value of abnormal oal accruals and EQ2 i is he absolue value of he abnormal curren accruals. EQ3 and EQ4 are earnings qualiy measures based on Dechow and Dichev (2002) model. EQ3 i is based on cross-secional regressions and EQ4 i is based on he ime-series regressions. MV i is firm i s marke capializaion in millions a he end of calendar monh ; RET i is monhly calendar reurn for firm i. A firm is classified as BUY if insiders purchases exceed insiders sales, and SELL if insiders sales exceed insiders purchases. Panel A: Univariae disribuion N Mean Median STD Q1 Q3 EQ1 i 989, EQ2 i 989, EQ3 i EQ4 i 614, MV i 989, RET i 989, BUY i 51, SELL i 85, Panel B: Correlaion marix, Pearson above diagonal and Spearman below EQ1 EQ2 EQ3 EQ4 EQ EQ EQ EQ Panel C: The mean and median coefficien esimaes across 9,966 firms using he following four-facor model: R R = α + β R R + δ SMB + σ HML + φ EQ + ε ( ) j, f, j j m, f, j j j j, where R p, is porfolio sock reurn,; R f, is risk free rae, measured as one-monh reasury bill rae; Rm, is marke porfolio reurn, measured using CRSP value weighed index; SMB and HML respecively he Fama-French (1992) size and marke-o-book facor reurns; EQ is he hedge reurn going long in he low earnings qualiy firms and going shor in he high earnings qualiy firms. The reurn window is monhly, and facor loadings are esimaed using a ime series regression based on 227 monhs of daa, from 1/85 o 11/03. Rm RF SMB HML EQ Mean Median Mean Median Mean Median Mean Median EQ EQ EQ EQ

31 Table 2 Insider exploiaion of asymmeric informaion as characerized by earnings qualiy measures Time series regression resuls are obained using he Fama-French hree-facor model: Rp, Rf, = α p+ βp( Rm, Rf, ) + δ psmb + σ phml + ε j, where R p, is porfolio sock reurn, where he reurn inerval is beween ransacion dae and one day afer SEC filing dae days on average; R f, is risk free rae, measured as one-monh reasury bill rae; Rm, is marke porfolio reurn, measured using CRSP value weighed index; SMB and HML respecively he Fama-French (1992) size and marke-o-book facor reurns. The reurn window is monhly, and facor loadings are esimaed using a ime series regression based on 227 monhs of daa, from 1/85 o 11/03. -saisics are under he coefficien esimaes and in ialics. We repor resuls for insider buying and selling for hree porfolios based on four measures of earnings qualiy: he quinile of firms wih highes earnings qualiy (HQ), he quinile of firms wih lowes earnings qualiy (LQ), and a hedge porfolio where we buy LQ quinile and sell HQ quinile. Insiders are classified as Buy if hey have ne purchases during he monh, and Sell if hey have ne sales during he monh. Only monhs where a leas 15 firms are presen are included. See secion 3 for he definiions of he four earnings qualiy measures. α R m, R f, SMB HML Adjused R 2 EQ1 BUY HQ LQ LQ-HQ EQ1 SELL HQ LQ LQ-HQ EQ2 BUY HQ LQ LQ-HQ EQ2 SELL HQ LQ LQ-HQ