Does Information Risk Really Matter? An Analysis of the Determinants and Economic Consequences of Financial Reporting Quality

Transcription

1 Does Informaion Risk Really Maer? An Analysis of he Deerminans and Economic Consequences of Financial Reporing Qualiy Daniel A. Cohen* January 2006 ABSTRACT I invesigae he deerminans and economic consequences associaed wih financial reporing qualiy. I find evidence of a posiive associaion beween invesors demands for firm-specific informaion and financial reporing qualiy. In addiion, he evidence suggess ha higher proprieary coss (proxied by capial inensiy, produc marke compeiion, and growh opporuniies) are associaed wih a lower qualiy of financial informaion. Conrolling for he firm-specific characerisics deermining financial reporing qualiy, I find evidence of a negaive associaion beween firms oal risk and financial reporing qualiy. Decomposing oal risk ino a sysemaic componen and an idiosyncraic one, he resuls imply ha firms providing financial informaion of higher qualiy do no necessarily enjoy a lower cos of equiy capial. However, a significan negaive relaion is documened beween reporing qualiy and idiosyncraic risk. This suggess ha he qualiy of accouning informaion canno be characerized as an addiional sysemaic priced risk facor, bu raher as an idiosyncraic one, once he firmspecific characerisics deermining informaion qualiy are conrolled for. These resuls demonsrae he imporance of explicily conrolling for he deerminans of financial reporing qualiy when invesigaing he associaed economic consequences and quesion recen empirical evidence on he associaion beween reporing qualiy and he cos of equiy capial. *Sern School of Business, New York Universiy, New York NY 10012; dcohen@sern.nyu.edu. This paper is based on my disseraion compleed a Norhwesern Universiy. I hank members of my disseraion commiee: Rober Korajczyk, Rober Magee, and especially Thomas Lys (Chair) and Beverly Walher for heir valuable commens, guidance and discussions. Workshop paricipans a he 2004 AAA meeing, especially Paul Hribar (discussan), he 2003 London Business School Accouning Symposium, especially Wayne Landsman (discussan), Baruch College, Universiy of California, Berkeley, Carnegie Mellon Universiy, Emory Universiy, Harvard Universiy, Universiy of Illinois, Urbana-Champaign, New York Universiy, Rice Universiy, Universiy of Houson, Universiy of Minnesoa, Universiy of Pennsylvania, Universiy of Souhern California, Universiy of Texas, Dallas, Universiy of Torono, Universiy of Uah, and Yale Universiy, Jan Baron, Eli Barov, Sudipa Basu, Philip Berger, Mark Bradshaw, Donal Byard, Jeff Callen, John Core, Gus DeFranco, Aiyesha Dey, Tom Fields, Frank Gigler, Rebecca Hann, Paul Healy, Mingyi Hung, Nisan Langberg, David Larcker, Chrisian Leuz, Xiaohui Liu, Jim Myers, Linda Myers, Greg Miller, Jim Ohlson, Gordon Richardson, Joshua Ronen, Sephen Ryan, Bhara Sarah, K.R. Subramanyam, Shyam Sunder, Pervin Shroff, Greg Waymire, Joe Weinrop, Marin Wu, Ram Venkaaraman, and David Ziebar all provided useful suggesions and commens on earlier versions. I am graeful o he Bioechnology Research Cener a he Kellogg School of Managemen and he Zell Cener for Risk Research for financial suppor. I hank Amira Proweller for ediorial assisance. Analyss forecas informaion was provided by I/B/E/S Inc. All errors are my own.

2 1. Inroducion I invesigae he deerminans and economic consequences of cross-secional variaion concerning he qualiy of financial reporing. Wheher disclosure policies and financial reporing affec a firm s cos of equiy capial is one of he mos ineresing and imporan quesions in he accouning and finance lieraure oday. To dae, here is a growing body of evidence ha informaion qualiy and disclosure policy lowers he equiy cos of capial (e.g., Boosan [1997], Easley and O Hara [2004]). However, his evidence is roubling for wo reasons. Firs, classical asse pricing heory shows ha diversifiable risks are no priced, and here are a pauciy of compelling argumens for why informaion risk is diversifiable (e.g. Hughes, Liu and Liu [2005]). Second, mos exising empirical sudies ake disclosure policy and financial reporing qualiy o be exogenous, alhough i is generally agreed ha firms opimize heir disclosure policy. Furhermore, he specific characerisics of firms ha provide a cerain qualiy of financial informaion may also affec he consequences of financial reporing. The purpose of his sudy is o exend prior lieraure by carefully idenifying he deerminans of firms financial reporing qualiy and invesigaing wheher here is evidence ha informaion risk affecs he cos of equiy capial once he firm-specific characerisics of his informaion risk are conrolled for. Economic heory suggess ha, ceeris paribus, increasing he qualiy of financial informaion reduces informaion asymmeries and hence lowers he cos of capial (e.g., Glosen and Milgrom [1985], Amihud and Mendelson [1986], Diamond and Verrecchia [1991], and Easley and O Hara [2004]). A firm can reduce informaion asymmeries beween iself and marke paricipans and beween informed and uninformed invesors by providing informaion ha helps invesors in heir decision making process. Consisen wih hese heoreical models, empirical sudies using indirec measures of disclosure documen ha a firm s disclosure qualiy/level is posiively associaed wih capial markes valuaion benefis (Welker [1995], Healy e al. [1999]), and, in paricular, is inversely relaed o is equiy cos of capial (e.g., Boosan [1997], Boosan and Plumlee [2002]). Using more direc measures of accouning informaion qualiy, recen empirical work focuses on he associaion beween earnings qualiy and he cos of capial (e.g., Barone [2002], Barh and Landsman [2003], Bhaacharya, Daouk and Welker [2003], and Francis e al. [2004, 2005]). 1

3 A significan shorcoming of numerous empirical disclosure sudies is he failure o address he endogenous naure of disclosure qualiy. If researchers do no conrol for he deerminans of disclosure policy, heir inferences regarding he economic consequences of disclosure qualiy may be spurious (Core [2001], Fields e al. [2001]). For example, recen evidence by Hou and Robinson [2005] suggess ha firms in indusries ha are more concenraed have lower cos of equiy capial. Theory (e.g., Verrecchia [2001]) and empirical evidence (e.g., Bamber and Cheon [1998], Harris [1998]) sugges ha he level of produc marke compeiion affecs disclosure policies. Taken ogeher wih he recen evidence in Hou and Robinson [2005], his suggess ha he level of indusry compeiion is an imporan correlaed omied variable in any sudy invesigaing he associaion beween disclosure/informaion qualiy and he cos of capial. Such consideraions, among oher firm-specific characerisics ha deermine he qualiy of financial informaion, for example, demands for capial, liigaion coss, and incenive coss, make i difficul o inerpre he associaion beween informaion risk and capial markes valuaion benefis documened in prior research. The research design used in his sudy specifically addresses hese concerns by idenifying he facors ha deermine he variaion in financial reporing qualiy and he associaed economic consequences. In line wih recen sudies on consequences of financial reporing qualiy (e.g., Francis e. al [2004, 2005]), I measure reporing qualiy as he mapping beween earnings and operaing cash flows (Dechow and Dichev [2002]). These measures represen abnormal deviaions beween accouning earnings and cash flows. Such deviaions resul in larger forecas errors on fuure operaing cash flows which impair his mapping, resuling in lower qualiy and increasing informaion risk. I find evidence ha higher invesors demands for firm-specific informaion are associaed wih higher qualiy of financial reporing. In addiion, he resuls sugges ha higher proprieary coss consrain he qualiy of financial informaion. My findings also indicae ha firms wih high-qualiy financial reporing policies have lower uncerainy, lower idiosyncraic risk, and lower esimaion risk. However, I do no find evidence ha firms providing high-qualiy financial informaion necessarily enjoy a lower cos of equiy capial. The resuls documened imply ha alhough informaion qualiy is associaed wih oal risk, i is he idiosyncraic diversifiable componen which drives his relaion, raher han he sysemaic undiversifiable componen. 2

4 I firs replicae prior findings and provide evidence consisen wih he resuls documened in concurren sudies [e.g. Francis e al. [2005]). I documen ha failure o conrol for firm characerisics ha lead firms o have a cerain qualiy of accouning informaion and enjoy a lower equiy cos of capial may wrongly aribue he cos of capial benefi o informaion qualiy raher han o he underlying characerisics. In oher words, he evidence suggess ha he link found in previous research beween a firm s qualiy of accouning informaion and is equiy cos of capial resuls from a failure o consider he underlying facors deermining he qualiy of financial reporing. My analysis implies ha he informaion risk associaed wih he qualiy of financial reporing does no necessarily consiue an addiional sysemaic non-diversifiable risk facor, bu raher is an idiosyncraic one. This finding suggess ha capial markes paricipans are no likely o price he documened uncerainy as oher risk facors, such as bea, size and book-o-marke raios. This finding is consisen wih recen heoreical work by Hughes, Liu, and Liu [2005]. Hughes e al. [2005] principal resul demonsraes ha in economies wih a large number of asses, idiosyncraic risk arising from he qualiy of informaion is fully diversifiable and does no affec risk premiums. The major conribuion of his sudy is ha i accouns for he underlying firmspecific characerisics relaed o he qualiy of heir financial reporing when invesigaing he associaed economic consequence. I show ha he failure o do so affecs he inferences made and conclusions drawn by previous sudies. In addiion, his sudy s findings have imporan implicaions for research on he consequences of firms disclosure policies. The evidence I presen suggess ha he variaion in financial reporing qualiy depends no only on he benefis firms expec o derive from disclosure, bu also on oher firm-specific aribues. Fuure work on deerminans and consequences of financial reporing policies should hus consider no only he capial marke benefis associaed wih financial reporing policies, bu oher firm-specific characerisics, such as produc marke characerisics, which consrain he qualiy of repored earnings. The remainder of he paper is organized as follows. Secion 2 provides a lieraure review and presens he heoreical background on he deerminans and consequences associaed wih financial reporing policies. Secion 3 describes he research design and addresses mehodological issues. Secion 4 presens he sample selecion crieria and discusses he empirical resuls. Secion 5 concludes. 3

5 2. Relaion o Prior Research Theoreical research invesigaing he link beween disclosure and a firm s cos of capial suggess a negaive associaion beween he wo. The firs sream of research arguing for his associaion concludes ha firms wih increased levels of disclosure reduce he cos of capial arising from informaion asymmeries, eiher beween a firm and is sockholders, or beween poenial raders in he firm s shares. Examples of heoreical work in his area are Copeland and Galai [1983], Glosen and Milgrom [1985], and Diamond and Verrecchia [1991]. Diamond and Verrecchia [1991] sugges ha higher disclosure reduces he amoun of informaion revealed by a large rade in a firm s securiies, hereby reducing he negaive price impac associaed wih such large rades. In his scenario, invesors would have relaively large posiions in a paricular firm s securiies. There would be a higher demand for he firm s securiies, which would increase he price of he firm s sock, hereby reducing he equiy cos of capial. Empirical work by Amihud and Mendelson [1986] suggess ha firms whose socks have a higher bid-ask spread have a higher cos of equiy capial because invesors demand compensaion for he added ransacion coss. The auhors conend ha firms ha provide more public informaion can reduce he adverse selecion componen of he bidask spread, and hus reduce heir cos of equiy capial. A second sream of research focuses on he link beween esimaion risk and he cos of capial (e.g., Barry and Brown [1985]). This research suggess ha a firm can reduce he esimaion risk associaed wih is payoff disribuion by providing more disclosures. If invesors price he esimaion risk, providing more informaion will reduce he firm s cos of capial. In recen work, Easley and O Hara [2004] demonsrae a link beween informaion srucure (privae versus public informaion) and he cos of capial. Developing a raional expecaions asse pricing model, hey argue ha more privae informaion increases he risk faced by uninformed invesors since beer informed invesors can shif heir porfolio weighs o adjus for new informaion. Easley and O Hara [2004] imply ha firms can affec heir cos of capial hrough he precision and quaniy of he informaion hey provide invesors. Building on he above heory, Francis e al. [2004, 2005] seek o provide evidence consisen wih he pricing effecs of informaion qualiy and claim ha accrual qualiy is a sysemaic priced risk facor. The 4

6 evidence documened in Francis e al. [2004, 2005] suggess ha informaion seems o affec he cos of capial. These resuls are puzzling since he heoreical underpinning behind his finding relies on Easley and O Hara s [2004] model which is sill a predicion of how informaion asymmery affecs he cos of capial. Hughes e al. [2005] quesion he heoreical underpinnings of he specificaions used in recen empirical sudies and show ha in large economies, idiosyncraic risk as well as he asymmeric informaion risk associaed wih idiosyncraic facors is fully diversifiable and should no affec he cos of capial in a sysemaic manner. Given he wo heoreical models developed in Easley and O Hara [2004] and Hughes e al. [2005], an empirical quesion sill remains as o wheher informaion qualiy can be characerized as proxying for idiosyncraic componens of asses payoffs which do no affec risk premiums or as a sysemaic undiversifiable facor which affecs he risk premium. I is his main empirical quesion ha I seek o address in his sudy. One of he major limiaions of empirical sudies on corporae disclosures is he difficuly in measuring he qualiy of disclosure policies (Healy and Palepu [2001]). 1 Using he Associaion of Invesmen Managemen and Research (hereafer, AIMR) rankings as an indirec measure of disclosure qualiy, numerous empirical sudies examine he associaion beween hese measures and firm characerisics and capial markes valuaion proxies (e.g., Lang and Lundholm [1993], Healy, Huon, and Palepu [1999], and Welker [1995]). Lang and Lundholm [1993] find a posiive associaion beween he AIMR scores and firm size, firm performance, and securiy issues, and a negaive associaion beween he AIMR scores and he correlaion beween earnings and reurns. Welker [1995] finds ha firms wih higher AIMR scores have lower informaion asymmery, as proxied by bid-ask spreads. Healy, Huon, and Palepu [1999] find ha firms wih susained improvemens in analyss raings of disclosure qualiy (AIMR scores) show an increase in sock liquidiy, analys following, insiuional ownership, and sock performance. Previous empirical research has documened mixed and limied evidence ha disclosure reduces he cos of capial. Boosan [1997] finds ha manufacuring firms wih a low securiy analys following have a negaive associaion beween a self-consruced 1 The difficuly in measuring disclosure qualiy has led some researchers o focus on managemen forecass (e.g., Coller and Yohn [1997], Pownall, Wasley, and Waymire [1993]). Ohers examine disclosure qualiy raings, for example, Lang and Lundholm [1996], Healy, Huon, and Palepu [1999], and Welker [1995]. 5

7 index of disclosure level and he cos of equiy capial. In a relaed empirical sudy Sengupa [1998] provides evidence ha high disclosure raings (AIMR) are inversely associaed wih he cos of deb. Boosan and Plumlee [2002] find a negaive relaion beween he annual repor disclosure level (as measured by he AIMR raings) and he cos of capial. Ye hey also documen ha he cos of capial is posiively associaed wih he levels of disclosures in quarerly repors (his finding is conrary o he predicion of he heory). As menioned before, recen empirical sudies (e.g., Barone [2002] and Francis e al. [2004, 2005], among ohers) documen ha earnings qualiy is associaed wih he cos of capial, suggesing ha lower qualiy of accruals (and hus earnings) ranslaes ino a higher cos of capial. A separae bu complemenary branch of analyical research examines he coss, especially he proprieary ones, associaed wih disclosures. Models such as Dye [1985b, 1986], Verrecchia [1983, 1990], Darrough and Soughon [1990], Wagenhofer [1990], and Hayes and Lundholm [1996] argue ha, all hings being equal, he probabiliy of disclosure decreases as he associaed proprieary coss increase. Mos of hese proprieary coss borne by firms arise from ineracion wih oher paries - he coss of compeiive disadvanage from disclosing informaion o heir compeiors and regulaors, of bargaining disadvanages wih boh suppliers and consumers, and of liigaion ha migh follow informaive disclosure are hree such examples. 2 As Fields e al. [2001] sugges, he empirical evidence presened in sudies like Boosan [1997] and Sengupa [1998] provides ineresing insighs, bu hese sudies suffer from noeworhy limiaions. Mos imporanly, hese sudies do no consider he relaed coss of higher disclosure qualiy and wheher hese coss affec he disclosure decision. Firms measure he valuaion benefis of providing higher qualiy earnings agains he associaed coss. If he coss ouweigh he marke valuaion benefis, he firm will choose o provide a lower qualiy of repored earnings, which will be less informaive in predicing fuure performance. This srongly moivaes my examinaion of he coss and benefis, among oher firm-specific deerminans, associaed wih disclosure policies. Alhough Francis e al. [2004, 2005] disinguish beween wha hey refer o as innae earnings qualiy and discreionary earnings qualiy, hey do no address wheher 2 Oher coss relaed o disclosure are he coss of developing and presening financial informaion. These coss, which are non-proprieary, are of second-order effec. This sudy focuses only on hird pary relaed coss, which are assumed o be proprieary. 6

8 proxies for proprieary coss, in paricular, curren produc marke compeiion, capial inensiy, growh opporuniies, and liigaion coss, consrain he qualiy of repored earnings. Given he empirical evidence in Hou and Robinson [2005], no conrolling for he level of indusry compeiion in examining he relaion beween informaion qualiy and he cos of capial creaes an omied correlaed variables problem which inroduces bias in he analysis. 3 Recognizing he endogenous naure of financial reporing qualiy policies, I firs model he facors deermining hese policies. In paricular, I es wheher invesors demands for firm-specific informaion, demands for capial, liigaion coss, and proprieary cos facors affec he variaion in financial reporing qualiy. As for he economic consequences of financial reporing qualiy, I es wheher higher qualiy reporing is associaed wih various proxies for firm risk, boh sysemaic and idiosyncraic. 3. Research Design Firms financial reporing qualiy policies are likely o be endogenous. If facors influencing cross-secional variaion in he reporing policies also influence he associaion beween he economic consequences and he qualiy of he repored accouning informaion, failing o conrol for hese facors may lead o erroneous inferences (Maddala [1983]). Specifically, an OLS regression of empirical measures for he equiy cos of capial on firm characerisics and a reporing qualiy measure would resul in inconsisen and biased coefficiens. To address his issue, I use a wo-sage esimaion mehod (Wooldridge [2002], chaper 5). I firs discuss my empirical measures of reporing qualiy and hen address he facors deermining financial reporing qualiy and is economic consequences. 3.1 Measuremen of Financial Reporing Qualiy In order o measure financial reporing qualiy I use wo mehods. Across hese mehods, he focus is on he associaion beween accruals and cash flows. A larger deviaion beween accruals and cash flows is inerpreed as lower qualiy of accouning informaion, reflecing higher informaion risk. 3 Hou and Robinson [2005] show ha firms in more concenraed indusries have a lower cos of capial. Based on heir findings, hey infer ha firms in highly concenraed indusries face a lower disress risk due o he less compeiive environmen in which hey are compeing in. Taken ogeher wih he empirical evidence in Harris [1998] ha firms disclose less in more concenraed indusries, one should conrol for he level of indusry compeiion when invesigaing he relaion beween disclosure/informaion qualiy and he cos of capial. 7

9 The firs mehod relies on he model presened in Barh e al. [2001]. Consider he following seing: (i) Firms have an objecive o maximize heir expeced value, (ii) A he end of period (-1), firms release an audied annual earnings repor. Aggregae earnings for period (-1) and is componens (cash flow from operaions and accruals) are used by various paries (e.g., capial markes paricipans, cusomers, suppliers, and curren and poenial compeiors) o predic firms fuure cash flows a ime, (+1), ec. Analyical models (e.g., Admai and Pleiderer [2000], Baiman and Verrecchia [1996], and Easley and O Hara [2004]) ake he disclosed public informaion s precision as he measure of is qualiy. The precision is inerpreed as achieving a given level of predicabiliy of expeced fuure cash flows under he flexibiliy and discreion permied by GAAP. Thus, he higher he precision, he higher he qualiy of repored earnings, and he more accuraely fuure cash flows may be prediced. To measure he level of precision empirically, I focus on he residuals obained from a regression of fuure operaing cash flows on previous period earnings componens. The firs measure of reporing qualiy is based on he residuals obained from esimaing he model specified in equaion (1) using ordinary leas squares: CFO + 1 = α0 + β CFO + β ΔAR + β ΔINV + β ΔAP + β DEPR + β OTHER + ε + 1 (1) Where: CFOi, Cash flow from operaions for firm i a year (Compusa annual daa iem #308) minus he accrual porion of exraordinary iems and disconinued operaions per he saemen of cash flows (Compusa annual daa iem #124); Δ AR Change in accouns receivable accoun per he saemen of cash flows 4 (Compusa annual daa iem #302); Δ INV Change in invenory accoun per he saemen of cash flow (Compusa annual daa iem #303); Δ AP Change in accouns payable and accrued liabiliies accoun per he saemen of cash flows (Compusa annual daa iem #304); DEPRi, Depreciaion and Amorizaion Expense (Compusa annual daa iem #125); OTHERi, Ne of all oher accruals, calculaed as EARN (CFO+ Δ AR+ Δ INV- Δ AP-DEPR), where EARN is income before exraordinary iems and disconinued operaions (Compusa annual daa iem #18); If Compusa annual daa iems #302, #303, or #304 are missing, Δ AR, Δ INV, and Δ AP are calculaed using daa from he balance shee, i.e., accouns receivable (Compusa annual daa iem #2), invenory (Compusa annual daa iem #3), and accouns payable (Compusa annual daa iem #70 plus accrued expenses (Compusa annual daa iem #153)). 8

10 All variables are deflaed by average oal asses. 5 In order o obain he financial reporing qualiy meric, I esimae equaion (1) for each fiscal year for each wo-digi SIC indusry code. I focus on he residuals obained from esimaing equaion (1). These provide a firm-specific residual for each fiscal year. The firs empirical measure of reporing qualiy is he absolue value of hese residuals: FQ1 = e. 6 These residuals reflec he magniude of fuure operaing cash flows i,+ 1 unrelaed o curren disaggregaed earnings. In he empirical analysis ha follows, I inerpre lower absolue value as represening a higher qualiy of financial reporing, which corresponds o a higher level of cash flow predicabiliy. The second empirical measure FQ2 = σ (e ) is he sandard deviaion of firm i s residuals calculaed over years i -4 hrough. A larger sandard deviaion of residuals indicaes a lower qualiy of repored earnings. I should be noed ha a firm which has frequenly large residuals will have a low sandard deviaion implying lile uncerainy abou he qualiy of earnings. In such a case, one should no expec for his o ranslae ino any priced uncerainy. To be consisen wih curren research, he second mehod of measuring reporing qualiy is based on he Dechow and Dichev [2002] mehod as implemened in Francis e al. [2005]. Under his approach, reporing qualiy is measured by he exen o which working capial accruals map ino cash flow realizaions. I esimae he following regression for each year for each wo-digi SIC code: WC = γ + γ CFO 1 + γ CFO + γ CFO 1 + γ ΔREV + γ 5 PPE + ε (2) Δ Where: Δ WC =The change in working capial for firm i a year which is compued as he change in accouns receivable (Compusa annual daa iem #302) plus he change in invenory (Compusa annual daa iem #303) less he change in accouns payable (Compusa annual daa iem #304) less he change in axes payable (Compusa annual daa iem #305) plus he change in oher ne asses (Compusa annual daa iem #307); Δ REV =The change in sales revenues (Compusa annual daa iem #12) beween year -1 and year ; PPE = Gross value of propery, plan and equipmen (Compusa annual daa iem #7) in year ; 4 5 This specificaion is used because i has he highes predicive abiliy compared o models ha include muliple lags of cash flows from operaions and accruals componens (see Barh e al. [2001]). 6 An alernaive firm-year specific measure of reporing qualiy is he squared residual for ha year. The correlaion beween his measure and he absolue value of residuals used in he sudy is (p-value< ) which suggess ha hese measures are highly correlaed. The enor of he resuls is very similar under his alernaive specificaion. 9

11 The model esimaed in equaion (2) is based on McNichols [2002] who suggess ha adding changes in sales revenues and PPE (propery plan and equipmen) significanly increases he model s explanaory power and reduces measuremen error. Using his modified Dechow and Dichev model leads o a beer specified model and sequence of residuals. The absolue value of he residuals obained from esimaing equaion (2) form he hird measure of reporing qualiy: FQ3, while FQ4 is he sandard deviaion of he residuals obained from esimaing equaion (2) calculaed over years -4 hrough. 3.2 Deerminans of Financial Reporing Qualiy The choice firms make abou he qualiy of he financial informaion hey repor in heir public financial saemens reflecs an analysis ha weighs he expeced benefis agains he associaed coss of disclosing high-qualiy informaion. Given he expeced benefis of providing informaion of higher qualiy, one would expec firms o choose o provide he highes qualiy of financial informaion possible, absen any coss of disclosing such informaion. Thus, one would expec o observe a corner soluion where he maximum reporing qualiy is chosen. In realiy, his does no occur, implying ha here are coss associaed wih disclosure, such as direc coss (non-proprieary), liigaion coss, and proprieary coss. Given such coss, firms would selec an inerior soluion o financial reporing qualiy. Therefore, when invesigaing he facors ha deermine a firm s reporing sraegy decision, his rade-off has o be considered. In addiion o hese specific facors affecing firms financial reporing qualiy decisions, I rely on cross-secional deerminans of firms disclosure policies used by prior lieraure (e.g., Lang and Lundholm [1993, 1996], Bushee e al. [2003], Buler e al. [2003], and Baron and Waymire [2004]). The above sudies, among ohers, have provided evidence ha disclosure decisions are associaed wih financing needs, he firm s informaion environmen, incenive coss, firm performance, liigaion coss, and ownership dispersion. Building on hese idenified deerminans, I presen below he empirical model and ouline he measures I use for firm-specific explanaory variables deermining he variaion in financial reporing qualiy. A discussion of he conrol variables used in he analysis follows. In he firs sage of he analysis, I esimae he following model based on he variables discussed below: 10

12 FQ + 1 = φ 0 + φ1owner + φ 2GROWTH + φ 3CAPITAL + φ 4 A_ HERF + φ 5 ISSUE + φ 6 LIT + φ LEVERAGE i, + φ MARGIN + φ OC + φ N SEG + φ SIZE + φ AGE + ξ _ (3) Where: FQ +1 Financial reporing qualiy measure, as described in he previous secion; OWNERi, Naural log of he number of shareholders of firm i in year (in housands; Compusa annual daa iem #100) minus naural log of he mean number of shareholders (in housands) in he firm s size decile; GROWTHi, Curren year s growh in sales, calculaed as ne sales for year (Compusa annual daa iem #12) less ne sales of year -1, scaled by ne sales for year -1; CAPITALi, Ne plan, propery and equipmen (Compusa annual daa iem #8) divided by oal asses (Compusa annual daa iem #6); 7 A_ HERFi, The weighed (by segmen sales) average Herfindahl-Hirshman Index for he indusries in which firm i repors business segmen sales. The Herfindahl-Hirschman Index is calculaed as he sum of squares of marke n 2 shares in he indusry. HERF = i= 1[si S], where s i is he firm s sales and S is he sum of sales for all firms in he indusry (defined by he wo-digi SIC code), and n is he number of firms in he indusry; ISSUE i Dummy variable equal o one if he company issued deb or equiy during he curren fiscal year or he nex wo fiscal years, and zero oherwise; LIT i Dummy variable equal o one if he firm is in a high-liigaion indusry, zero oherwise; 8 LEVERAGE Long-erm deb (Compusa annual daa iem #9) plus deb in curren liabiliies (Compusa annual daa iem #34) divided by firm value (Compusa annual daa iem #199 imes Compusa annual daa iem #25); MARGINi, Gross margin percenage, calculaed as he year ne sales (Compusa annual daa iem #12) less cos of goods sold for he year (Compusa annual daa iem #41), scaled by ne sales; OC, i Operaing cycle for firm i a ime, measured in days as ( AR + AR 1) / 2 (INV + INV 1) / 2, where AR is he firm s accouns (Sales/ 360) + (COGS/360) receivable, INV is he firm s invenory, and COGS is he firm s cos of goods sold; N_SEG Number of wo-digi SIC code indusries ha he firm is engaged in year ; SIZEi, Naural logarihm of marke capializaion a he end of he fiscal year (year ), calculaed as he closing price a fiscal year-end imes he number of shares ousanding a fiscal year-end (Compusa annual daa iem #199 imes Compusa annual daa iem #25); 7 When he reporing qualiy measure is FQ3 and FQ4, I do no include he variables CAPITAL and MARGIN as a deerminan o avoid any mechanical associaion given ha PPE and a variaion of MARGIN is used as an explanaory variable in equaion (2). 8 Following Kasznik and Lev [1995] I define high-liigaion indusry as: high-echnology firms (SIC codes , , , , ). Using he definiions in Francis e al. [1994] produces idenical resuls. 11

13 AGE The firm s age, naural logarihm of number of monhs he company has been lised on CRSP; I esimae he model in equaion (3) boh across firms and ime, resuling in a pooled cross-secional ime-series specificaion. 9 Firms have incenives o respond o invesors demands for firm-specific informaion since reducing informaion asymmeries beween he firm and is invesors can lower heir cos of capial (Healy and Palepu [2001] and Verrecchia [2001]). Exernal demands for firm-specific financial informaion are expeced o vary wih he level of ownership concenraion. Higher poenial informaion asymmery, especially among invesors, and demands for firm-specific informaion is expeced for firms wih a highly dispersed invesor base. Therefore, ousiders demands for financial informaion from hese firms is expeced o be higher han for firms wih high levels of ownership holdings. To capure he effec of ownership dispersion on he qualiy of financial reporing, I use he variable OWNER, which is he log of he number of shareholders of he firm adjused by he log of he mean number of shareholders in he firm s size decile. The use of his exac variable is moivaed by Bushee e al. [2003] who used his explanaory variable as a deerminan of a firm specific disclosure decision (holding an open versus closed conference call). To proxy for he proprieary coss associaed wih he reporing decision, I use measures of a firm s capial inensiy, growh opporuniies, and characerisics of is produc marke. If a produc marke s barriers o enry are relaively high, he associaed coss of disclosure should be relaively low. High capial inensiy is generally inerpreed as a major barrier of enry (Pioroski [2003], Hou and Robinson [2005]). Therefore, capial inensiy is hough o be posiively associaed wih he qualiy of financial informaion. 10 High enry coss o a marke, as refleced by high capial requiremens, creae siuaions in which a large fracion of he capial coss are already sunk for incumben firms, bu are decision relevan o poenial enrans. To capure he feaure of capial inensiy as a barrier o enry, I use he variable CAPITAL, which 9 As an alernaive specificaion, I have esimaed also a firm-specific ime series specificaion for firms wih more han 15 years of daa. Under his mehod, he sandard deviaion of he residuals obained from he firm-specific regressions were used as a measure of reporing qualiy. While his alernaive esimaion procedure decreased subsanially he number of firms in he sample (1,476 firms), he resuls based on his alernaive specificaion are no significanly differen from he repored resuls. 10 In examining he effecs of proprieary coss on segmen disclosures, Pioroski [2003] uses capial inensiy as a proxy for barriers o enry and shows ha i is posiively associaed wih he fineness of segmen reporing. 12

14 comprises ne propery, plan and equipmen scaled by oal asses. 11 Capial inensiy proxies also for financing needs (Leuz and Verrecchia [2000]), hus, consisen wih he lieraure o dae, i is expeced ha more capial inense firms which have higher financing needs will provide higher qualiy of financial informaion. Anoher measure of proprieary coss relaes o he firm s growh opporuniies. The more innovaive a marke is and he more heavily i relies on inangible knowledge, he more a firm should inves o reain is unique saus and preserve fuure opporuniies. Given ha hese fuure opporuniies are posiively associaed wih proprieary coss, I use GROWTH, which I define as he curren year s percenage change in sales, as a proxy for fuure opporuniies ha he firm needs o proec. I expec ha i will be negaively associaed wih he level of fuure cash flow predicabiliy. 12 The lieraure idenifies exising compeiion in a firm s produc marke as being associaed wih proprieary coss. Compeiion hus influences a firm s disclosure decisions. In order o accoun for produc marke compeiion, I measure he concenraion rae of each indusry using he Herfindahl-Hirschman Index. I calculae he Herfindahl-Hirschman Index as HERF n = i= 1[si 2 S], where si is he firm s sales, S is he sum of sales for all firms in he indusry (defined by he wo-digi SIC code), S s i is he marke share of firm and n is he number of firms in he indusry. 13 Since he concenraion raio is an indusry measure, I use a weighed average degree of concenraion (compeiion) he firm faces in order o capure he effec of compeiion of firms decision o provide a cerain level of informaion qualiy. The weighed average Herfindahl-Hirshman Index is he sum of he indusry concenraion raios for all he indusries he firm is engaged in (i.e., specific segmen daa is disclosed in he annual repor), each weighed by he raio of specific segmen sales o oal sales As a sensiiviy analysis, I repeaed he analysis by using capial expendiures, a flow variable. The repored resuls are no affeced by his change. 12 As an alernaive measure, I used R&D expenses and adverising expenses. However, for many firm-year observaions, Compusa daa iem # 46 (R&D Expenses) is missing which decreased he sample by 70%. For he reduced sample, I find ha his variable is more srongly and significanly associaed wih he qualiy of financial informaion in he prediced direcion. In addiion, none of he resuls of he second sage analysis are differen using his reduced sample. 13 One can argue ha he Herfindahl-Hirshman Index is a good proxy of pre-enry compeiion in a paricular indusry. For example, in concenraed indusries where barriers o enry are low, one can expec compeiion o increase wihin a shor period of ime. 14 As an alernaive explanaory variable, I have used he firm s marke share based on oal sales (oal asses). The firm s marke share is defined as he firm s oal sales revenues (oal asses) divided by he 13

15 Given differen analyical models suggesions, I do no have a specific predicion abou he associaion beween he qualiy of financial informaion and he level of compeiion, as capured by he concenraion raio. On he one hand, higher concenraion raios proxy for monopoly rens, bu on he oher hand, his may be one reason why he indusry is highly concenraed. Thus he concenraion raio is likely o be a proxy for high enry coss (high barriers o enry). However, if he concenraion raio is a good proxy for pos-enry compeiion (consisen wih Verrecchia [1983]), hen one would expec a posiive relaion beween financial reporing qualiy and he level of indusry concenraion. Such an expecaion implies a negaive relaion beween he level of compeiion wihin an indusry and financial reporing qualiy. To be consisen wih he lieraure ha suggess ha performance is an imporan deerminan of disclosures (e.g., Lang and Lundholm [1993]), I include he variable MARGIN, defined as sales revenue ne of cos of goods sold, scaled by ne sales. As Lang and Lundholm [1993] discuss, heoreical and empirical sudies provide mixed evidence on he relaion beween disclosure policies and firm performance bu do no offer a specific predicion on he relaion beween he wo. I is agreed, however, ha he wo are relaed. On one hand, more profiable firms wih higher gross margins arac fuure compeiion and face higher hreas of poenial enrans. Thus he higher he firm s profiabiliy, capured by is gross margin, he more he cos of providing higher qualiy of financial informaion is expeced o be. On he oher hand, a posiive relaion beween disclosure and firm performance is implied by an adverse selecion argumen. Liigaion coss have been suggesed by prior sudies (e.g., Francis e al. [1994], Skinner [1994], and Lev and Kasznik [1995]) as a deermining facor of financial reporing sraegies. I define he dummy variable LIT o ake he value of 1 if he firm operaes in a high-liigaion risk indusry. If lower precision of accouning informaion is cosly o firms, I expec ha when liigaion coss are higher, he qualiy of informaion is higher. The presence of agency coss gives rise o demand for monioring, and he informaion a firm s financial saemens provide may be used o miigae agency coss (Jensen and Meckling [1976]). Highly leveraged firms have higher agency coss and hus sum of oal sales revenues (oal asses) of all sample firms in he same wo-digi SIC code. The resuls repored using his alernaive variable are no affeced and are similar. 14

16 a greaer demand for monioring. 15 Therefore, I predic reporing qualiy o vary wih a firm s capial srucure (Was [1977], Smih and Warner [1979], Lefwich e al. [1981], Buler e al. [2003], and Baron and Waymire [2004]). In a recen sudy, Baron and Waymire [2004] provide evidence ha managers incenives o supply high qualiy financial saemens increase wih he level of shareholder-debholder agency conflics as proxied by he amoun of leverage in he firm s capial srucure. They show a significan posiive associaion beween firms leverage and he qualiy of public accouning informaion and inerpre his finding as consisen wih deb conracing influencing financial reporing. If he financial informaion provided in he firm s annual repor is complemenary o he monioring informaion deb providers use, I expec more leveraged firms o provide financial informaion of higher qualiy. However, if deb providers use subsiue informaion channels o acquire monioring informaion, his will decrease he likelihood ha he previous predicion holds rue. Following Lefwich e al. [1981], I use he variable LEVERAGE, which is he firm s oal deb o firm value, o capure his deerminan on financial reporing qualiy. 16 Moivaed by he empirical evidence in prior research ha securiy issuance is associaed wih disclosure policies (e.g., Lang and Lundholm [1993]), I include he variable ISSUE as an addiional deerminan of financial reporing qualiy. Conrol Variables: Differences across firms could influence he fuure performance and predicabiliy of heir fuure cash flows. Dechow, Kohari and Was [1998] show ha abiliy of earnings o predic fuure cash flows depends on he firm s operaing cash cycle. Dechow and Dichev [2002] claim ha longer operaing cycles induce more uncerainy, making accruals noisier and less helpful in predicing fuure cash flows. To conrol for he uncerainy associaed wih he operaing environmen of he firm, I include in equaion (3) a proxy for he lengh of he operaing cycle (OC), where OC = + (measured in days). The operaing cycle variable capures ( + 1 AR AR 1) / 2 (INV INV ) / 2 + (Sales/ 360) (COGS/ 360) variaion in fuure cash flow predicabiliy ha is likely no predeermined. To furher 15 There is no consensus in he corporae finance lieraure wheher firms ha are highly leveraged have higher agency coss (Jensen [1986]). I can be argued ha deb holders provide addiional monioring and incenives ha lower agency coss. Consisen wih he sudies cied above, I expec a posiive associaion beween LEVERAGE and financial reporing qualiy. 16 Capial srucure is an endogenous variable for he firm as well. To address his concern, he specificaion of equaion (3) uses lagged values of he explanaory variables in conras o a conemporaneous specificaion. 15

17 address how he complexiy of he firm s operaing environmen affecs variaion in informaion qualiy, I also include he number of line of businesses ha a firm engages in (N_SEG). Consisen wih previous empirical sudies, I conrol for he firm s informaional environmen, by including he firm s size. The variable SIZE is defined as he naural logarihm of marke capializaion a he end of he fiscal year. In addiion, I include he variable AGE, conjecuring ha younger firms have a lower qualiy of accouning informaion. 3.3 Economic Consequences of Financial Reporing Qualiy I use four proxies for capial markes consequences of financial reporing qualiy: 1) he firm s cos of equiy capial (using boh implied cos of capial esimaes and he Fama-French [1993] hree facor model, 2) he firm s sandard deviaion of sock reurns, 3) he firm s idiosyncraic risk, esimaed as he residual variance from a regression of sock reurns on he marke s reurn (using he radiional marke model), and 4) he dispersion in analyss earnings forecass (a proxy for uncerainy and esimaion risk). The main hypohesis ha I es is wheher providing financial informaion of higher qualiy is associaed wih capial markes valuaion benefis. To do so, I esimae he following pooled cross-secional ime-series model: = α 0 + β ' X i + β FQ + ε i (4) R 1 2 ', In equaion (4) he dependen variable is one of he four proxies for capial markes consequences, X i represens a vecor of conrol variables, and FQ is an empirical measure of reporing qualiy. As discussed previously, FQ is likely o be correlaed wih he error ermε, which creaes an endogeneiy problem. This endogeneiy is generally due o omied correlaed variables. As shown in he previous secion, financial reporing qualiy may be deermined by facors ha are no capured by he conrol vecor X. This endogeneiy problem can be solved hrough an insrumenal variables approach. Following his approach, he researcher mus idenify a vecor of observable variables ha do no appear in equaion (4) and are no correlaed wih he error ermε, bu are correlaed in par wih he variable FQ. The variables idenified in secion 3.2 mee his requiremen. 16

18 I esimae a wo-sage procedure (see Wooldridge [2002], chaper 5), in he firs sage of which I esimae a financial reporing qualiy model. Using he fied values from he firs sage model as an insrumenal variable for he qualiy measure (IV), I esimae in he second sage an OLS regression of capial markes/valuaion benefis proxies on firm characerisics and his insrumenal variable. 17 Following Maddala [1983, p ], I do no include in he firs sage model [equaion (3)] a variable ha proxies for any of he capial markes benefis ha I use in he second OLS regression. Including he dependen variable of he second sage equaion in he firs sage model will lead o a logically inconsisen specificaion, unless he second sage dependen variable coefficien is resriced o be equal o zero. 18 This mehodology is similar o he approach underaken by Baron and Waymire [2004] who adop an insrumenal variables approach o conrol for he endogenous naure of financial reporing decisions when esing he capial marke s reacion o he sock crash of The following specific models are esimaed across firms and ime: R e i, = ρ + ρ BETA 0 + ρ B _ M 6 1 i, i, + ρ SIZE + ρ INDS 7 2 i, i, + ρ LEVERAGE + ρ FQ 8 3 i, + ϑ i, i, + ρ DISP 4 i, + ρ LTG 5 i, + (4a) STDRET i, χ 0 + χ 1 SIZE i, + χ 2 B _ M + χ 3 FQ + ζ i, = i, i, (4b) IDIOS i, λ 0 + λ 1 SIZE i, + λ 2 B _ M + λ 3 FQ + π i, = i, i, (4c) DISP i, ϖ 0 + ϖ 1 SIZE i, + ϖ 2 SURP i, + ϖ 3 ANALYST i, + ϖ 4 FQ + ψ i, Where: = i, (4d) R ei, Cos of equiy capial for firm year ; BETAi, The marke bea for firm i in year, esimaed using a rolling window of five years of monhly reurns where he CRSP value-weighed marke reurn is used as he marke reurn; SIZEi, Naural logarihm of marke capializaion for firm a he end of he fiscal year, calculaed as he closing price a fiscal year-end imes he number of shares ousanding a fiscal year-end (Compusa annual daa iem #199 imes Compusa annual daa iem #25); 17 In order o derive correc inferences, he sandard errors are adjused o address he correlaion beween he error erm of he firs sage choice model and he error erm in he second sage equaion. 18 As an alernaive esimaion procedure, I have esimaed he probabiliy ha he qualiy of financial informaion for a given firm is above he indusry median using a Probi model. Using he fied probabiliies of his firs sage esimaion as an insrumenal variable, I repeaed he second sage analysis in which he associaion beween his insrumen and he capial markes benefis variables was invesigaed. The resuls using his alernaive procedure provide similar resuls o hose repored in he curren version of he paper. These resuls are available from he auhor upon reques. 17

19 B_M LEVERAGE, LTG STDRET i IDIOS i Book-o-Marke raio, where marke value of equiy is calculaed as he closing price a fiscal year-end imes he number of shares ousanding a fiscal year-end (Compusa annual daa iem #199 imes Compusa annual daa iem #25), divided by he book value of common equiy (Compusa annual daa iem #60); i Long-erm deb (Compusa annual daa iem #9) plus deb in curren liabiliies (Compusa annual daa iem #34) divided by firm value (Compusa annual daa iem #199 imes Compusa annual daa iem #25); Expeced long erm growh in earnings, defined as he percenage change in he mean wo-year ahead earnings forecas (obained from IBES) from he curren earnings realizaion (Compusa annual daa iem #58); Firm i s sandard deviaion of daily holding period reurn averaged over he 12 monhs saring as of June subsequen o fiscal year ; The residual variance from a regression of firm-specific reurns on he reurns of he CRSP value-weighed marke index over he 12 monhs saring as of June subsequen o fiscal year ; DISPi, Sandard deviaion of analyss forecass of year earnings per share (IBES) for firm measured in June following fiscal year -1, scaled by beginning of period price; INDS Average indusry implied equiy cos of capial; SURPi, Absolue value of he difference beween curren year s earnings per share (Compusa annual daa iem #58) and he previous year earnings per share, scaled by average oal asses of he firm; ANALYSTi, Number of analyss issuing forecass a he same ime DISP was calculaed (IBES); ζ, υ, π ψ Error erms assumed o have zero mean and consan variance. i,, Equaion (4a) is specified similarly o he cross-secional regression of he implied risk premium in Gebhard e al. [2001, Table 7] and includes similar conrols. Using his paricular specificaion makes he analysis from an asse pricing perspecive more relevan since we can learn which of he firm characerisics are more imporan and significan deerminans of he implied equiy cos of capial. Equaions (4a-4d) are esimaed using boh pooled cross-secional ime-series regressions and Fama-MacBeh regressions. I repor he resuls of he Fama-MacBeh regressions, for which I run cross-secional regressions annually and hen average he annual coefficiens across he sample years from I use he sandard deviaion of he coefficiens across he seveneen sample years o compue he - saisics: β = ; where β is he average coefficien based on he yearly regressions, σ β n σβ is he ime-series sandard deviaion corresponding o each coefficien, and n is he 18

20 number of years. Since he auocorrelaion beween he coefficiens in he annual regressions can affec he rue sandard errors and hus inflae he -saisics, I correc he -saisics in a manner consisen wih Bernard [1995]. Nex, I discuss each one of he economic consequences models as specified in equaions (4a-4d). 3.4 Financial Reporing Qualiy and Risk Cos of Equiy Capial Following Barone [2002] and Francis e al. [2004, 2005], lower qualiy financial reporing leads o greaer uncerainy and ulimaely o higher informaion risk. If his risk canno be diversified away, i will resul in a higher cos of equiy capial. Following his raionale, I es he associaion beween he empirical measures of reporing qualiy oulined in secion 3.2 and an implied equiy cos of capial esimaed using he models presened in Claus and Thomas [2001], Gebhard e al. [2001], Ohlson and Juener-Nauroh [2005] as implemened in Gode and Mohanram [2003], and Eason [2004]. Previous sudies have used differen measures for he equiy cos of capial. Since he cos of capial is no an observed phenomenon, one needs o esimae i. Some of he esimaion mehods used in he lieraure are based on he Fama-French hreefacor model (Fama and French [1993]). As Fama and French [1997] poin ou, equiy cos of capial esimaes based on he hree-facor model are imprecise, boh a he firm level and he indusry level. Anoher alernaive is o proxy for he equiy cos of capial by using realized sock reurns. This approach has problems oo, given ha he correlaions beween expeced reurns and ex pos realized reurns are weak (Elon [1999]). These approaches have led researchers o infer ex ane equiy cos of capial raes using an implied approach. Following his approach, assuming a valuaion model, one esimaes he implici equiy cos of capial using he curren sock price and observable proxies for marke expecaions of fuure cash flows or earnings (usually analyss earnings forecass). The specific models differ in erms of he assumpions made regarding growh raes, erminal values, and forecas horizons. I summarize he specific models used in he Appendix. While here is a debae abou he meris of various ex ane measures (e.g., Gode and Mohanram [2003], Guay e al. [2003]), I ake no posiion, bu use a variey of acceped mehods o calculae he implied equiy cos of capial o ensure ha he 19