Rajiv Banker a,* Sudipta Basu a Dmitri Byzalov a Janice Y.S. Chen a

Transcription

1 Direcion of Sales Change and Asymmeric Timeliness of Earnings Rajiv Banker a,* Sudipa Basu a Dmiri Byzalov a Janice Y.S. Chen a a Fox School of Business, Temple Universiy, Aler Hall, Philadelphia, PA 19122, Unied Saes Absrac Sales decreases have a greaer impac on earnings han sales increases because of cos sickiness. We hypohesize ha his correlaed omied variable consiues a confounding effec in sandard models of condiional conservaism. Adding sales change direcion o he Basu (1997) and Khan and Was (2009) models decreases measured conservaism significanly. We also documen a moderaing effec ha likely arises because sales decreases provide informaion ha is used in impairmen esing for individual angible asses. Increases in conservaism around SOX are concenraed in he sales decrease subsample, consisen wih his argumen. When GDP increases, a Modified CScore incorporaing only he moderaing effec predicably decreases (conrary o he Khan-Was CScore ha increases). Keywords: informaion asymmery; imely loss recogniion; Sarbanes-Oxley Ac; asse wrie-downs JEL codes: D24, M41, M42, M55 * addresses: banker@emple.edu (R. D. Banker), sudipab@emple.edu (S. Basu), dbyzalov@emple.edu (D. Byzalov), janiceysc@emple.edu (J. Chen). 1

2 1. Inroducion Empirical research documens a piecewise linear relaion beween earnings and sock reurns, and ofen aribues his paern o (condiional) conservaism asymmeric recogniion of good versus bad news (e.g., Basu, 1997). Numerous sudies invesigae he effecs of various drivers of demand for conservaism (such as conracing or liigaion) in his piecewise linear relaion (e.g., Qiang, 2007; LaFond and Was, 2008; Khan and Was, 2009). In conras, we examine he effecs of operaional facors, represened by he direcion of sales changes. There are wo reasons why an operaional facor (such as sales increases or decreases) may play an imporan role in he asymmeric relaion beween earnings and sock reurns. Firs, recen research on sicky coss (e.g., Anderson e al., 2003) documens ha many coss behave asymmerically for operaional reasons. Because of his cos asymmery, operaional facors may have a confounding effec on asymmeric imeliness esimaes. Second, even afer conrolling for his confounding effec, operaional facors may conain incremenal informaion abou fuure gains and losses ha are bookable in accruals, 1 and hence may have a moderaing effec on imely loss recogniion. Similar o conracual and liigaion drivers of demand for conservaism, operaional facors vary across firms and over ime, which may lead o sysemaic variaion in asymmeric imeliness. Conservaism implies ha bad news (ypically proxied by negaive sock reurns) is recognized in earnings more quickly and more fully han good news (proxied by posiive reurns), and herefore he associaion beween earnings and sock reurns should be sronger for negaive reurns han for posiive reurns (Basu, 1997; Was, 2003a). Numerous empirical sudies find resuls consisen wih his predicion and infer he presence of conservaism (e.g., Pope and 1 Ball and Shivakumar (2006) inroduce he noion of bookable gains and losses (i.e., gains and losses ha poenially can be recognized via accouning accruals ) o moivae heir analysis of differen book-based and marke-based proxies for gains and losses, including combinaions of muliple proxies. Roychowdhury & Was (2007) similarly disinguish beween separable ne asses ha are recognized and rens (growh opions, monopoly rens) ha accounans ypically do no recognize unless hey are acquired. 2

3 Walker, 1999; Ball e al., 2000; Was, 2003b). In his paper, we examine he role of a key operaional facor he direcion of sales changes. This facor is operaional in he sense ha i reflecs changes in real economic aciviy of he firm. As such, i is fundamenally differen from he moderaing facors examined in prior sudies, which proxy for various sources of agency-driven demand for conservaism arising from conracing, liigaion, axaion and regulaion (Was, 2003a). Examples of such facors include liigaion exposure (Basu, 1997; Holhausen & Was, 2001; Qiang, 2007; Park & Wynn, 2008), deb conracing (Zhang, 2008; Wienberg-Moerman, 2008; Nikolaev, 2010) and managerial ownership (LaFond & Roychowdhury, 2008). Operaional facors may have a confounding effec on conservaism esimaes because of asymmeries in he underlying economic behavior. Recen sudies of sicky coss (e.g., Anderson e al., 2003; Weiss, 2010) documen asymmeric behavior of coss, and aribue his paern o deliberae resource commimen decisions by managers who face adjusmen coss (such as hiring and firing coss for labor). Because of adjusmen coss, managers reain some unuilized resources when sales fall. By conras, when sales rise, managers have o add sufficien resources o accommodae he increased sales. Because of his operaional asymmery, many coss are sicky hey rise more in response o sales increases han hey fall in response o sales decreases. The asymmery in coss implies an asymmeric relaion beween earnings and sales changes. Since sales changes are correlaed wih concurren sock reurns, i also implies an asymmeric relaion beween earnings and sock reurns, which is likely o disor he sandard esimaes of asymmeric imeliness. To conrol for his confounding effec, we esimae our conservaism measures separaely for he sales increases and for sales decreases subsamples. The asymmeric 3

4 imeliness esimaes wihin each subsample will reflec condiional conservaism bu no he confounding operaional asymmeries beween sales increases and decreases. In addiion o his confounding effec, operaional facors refleced in he direcion of sales changes are likely o have an imporan moderaing effec on conservaism. Condiional on sock reurns, sales increases and decreases are likely o conain incremenal informaion abou he bookable porion of fuure gains and losses. 2 Therefore, he direcion of sales changes is likely o have a sysemaic moderaing effec on recogniion of fuure losses in curren period earnings, leading o greaer asymmeric imeliness for sales decreases han for sales increases. We esimae our models using daa for publicly-raded U.S. firms for We examine he confounding and moderaing effecs of operaional facors refleced in sales change direcion using he Basu (1997) model and an exended model ha conrols for he demand for conservaism following Khan and Was (2009). 3 We find ha adding conrols for he confounding effec of he direcion of sales changes reduces he average esimaes of conservaism by 11.6%-13.6%. Thus, as expeced, he sandard measures of conservaism exhibi (moderae) upward bias because hey do no conrol for he confounding effec of operaional facors. More imporanly, we find ha, even afer conrolling for his confounding effec, he direcion of sales changes has a srikingly large moderaing effec on asymmeric imeliness. The Basu asymmeric imeliness coefficien is on average 61% higher in he sales decrease subsample han in he sales increase subsample. In annual regressions, he same paern holds for 2 As we explain in secion 2, even hough in an efficien marke sock reurns fully capure all informaion abou fuure cash flows, no all of hose fuure cash flows are bookable (relevan for curren period accruals). 3 Alhough a few researchers such as Dierich e al. (2007), Givoly e al. (2007) and Paaoukas & Thomas (2011) criique he Basu (1997) asymmeric imeliness regression on economeric grounds, he model is defended by Ryan (2006), Basu (2009) and Ball e al. (2011a, b). We believe our economic concerns would be valid in modified models as well, and analyze he asymmeric imeliness regressions mos commonly used in prior research o develop deeper insighs ino exising resuls. 4

5 41 years ou of 45 in our sample. Even afer conrolling for he Khan-Was conservaism demand proxies, he average (median) of firm-year asymmeric imeliness measures is 40% (72%) higher for sales decreases han for sales increases. These differences are highly significan boh saisically and economically. The esimaes indicae ha asymmeric imeliness is paricularly pronounced when bad news in sock reurns is compounded by unfavorable operaional facors as refleced in decreasing sales. Consisen wih hese argumens, we show ha sales decreases have greaer explanaory power for angible asse wriedowns, which likely reflec reduced cash flow forecass, han for goodwill impairmens, which likely reflec reduced growh opions. Recognizing he confounding and moderaing effecs of operaional facors refleced in he direcion of sales changes also has imporan consequences for Khan and Was (2009) firm-year esimaes of conservaism. We exend he Khan-Was (2009) model o compue a Modified CScore ha incorporaes he moderaing effec and excludes he confounding effec of operaional facors when esimaing he firm-year conservaism measure. When we compue he CScore wih and wihou conrols for boh effecs, he correlaion beween he wo measures is jus 0.34 in he full sample, and even lower (0.26) in he sales decreases subsample, indicaing a dramaic impac on firm-year conservaism esimaes. When we examine he relaion beween GDP growh and he wo conservaism measures, we find ha he ime-series variaion in our Modified CScore is consisen wih he heoreical predicions ha higher macroeconomic growh reduces he demand for conservaism, whereas he ime-series variaion in he Khan- Was CScore violaes he heoreical predicions. Analyzing he confounding and moderaing effecs of sales changes direcion provides imporan new insighs ino variaion in conservaism hrough ime. For example, conservaism 5

6 increases primarily in he sales decreases subsample during recessions as well as around SOX, which is consisen wih our argumen ha asse impairmens are condiioned on he direcion of sales changes. We also documen ha informaion asymmery (bid-ask spread) increases conservaism only in he sales increases subsample, which we aribue o he laer being siuaions in which informaion asymmery is paricularly high. The res of his paper is srucured as follows. In secion 2, we develop our empirical hypoheses. In secion 3, we describe he daa and he esimaion models. Secion 4 describes our main empirical resuls for he effecs of operaional facors and for our Modified CScore, and secion 5 presens addiional empirical examples o illusrae he new insighs provided by our approach. Secion 6 concludes. 2. Hypohesis Developmen In developing our main hypoheses, we build on prior research in wo areas: conservaism in financial accouning (e.g., Basu, 1997; Pope and Walker, 1999; Ball e al., 2000) and sicky coss in cos accouning (e.g., Anderson e al., 2003; Weiss, 2010; Chen e al., 2012). Conservaism and cos sickiness are fundamenally differen phenomena: conservaism generaes asymmery in he informaion sysem ha ranslaes (managerial repors of) real economic aciviy ino accouning daa, whereas cos sickiness denoes asymmery in real economic aciviy iself. Neverheless, operaional facors represened by he direcion of sales changes have imporan asymmeric effecs on earnings via boh phenomena, and boh ypes of effecs should be aken ino accoun o ensure accurae inferences in conservaism research. 4 In financial accouning, (condiional) conservaism is inerpreed as he higher degree of 4 Similarly, boh ypes of effecs should be aken ino accoun in sudies of sicky coss. However, he implicaions for sicky coss research are beyond he scope of his paper. 6

7 verificaion o recognize good news as gains han o recognize bad news as losses (Basu, 1997; Was, 2003a). Basu (1997) showed ha conservaism implies asymmeric imeliness of earnings wih respec o good versus bad news. His primary measure of asymmeric imeliness is based on a regression of earnings on sock reurns, wih separae slopes for posiive and negaive reurns (proxies for good and bad news, respecively). He found ha he slope coefficien and R 2 is higher for negaive reurns han for posiive reurns, i.e., earnings reflecs bad news more quickly han good news, consisen wih his heoreical predicions. This empirical model has been widely used o examine he asymmeric imeliness of earnings in various conexs, and demonsrae cross-secional and ime-series variaion in his asymmeric imeliness ha is consisen wih hypoheses regarding he demand for conservaism (see Ball e al., 2011a, for a brief review). Recen sudies of sicky coss in cos accouning (e.g., Anderson e al., 2003; Weiss, 2010) have emphasized anoher source of asymmery in coss and earnings, which arises for operaional reasons. These sudies documen ha many coss are sicky in he sense ha hey increase proporionaely more for sales increases han hey decrease for sales decreases. Anderson e al. (2003) explain ha cos sickiness is prevalen because (1) many coss arise as a resul of managers resource commimen decisions, and (2) managers have o incur adjusmen coss when changing resource levels, such as insallaion and disposal coss for equipmen or hiring and firing coss for labor. Because of adjusmen coss, managers resource commimen decisions are in general asymmeric wih respec o sales decreases versus sales increases. When sales decrease, managers opimally choose o reain some slack resources o avoid adjusmen coss associaed wih cuing resources. Consequenly, coss decrease less han proporionaely o he decrease in sales. However, when sales increase, managers have less discreion as hey have 7

8 o acquire enough addiional resources o be able o accommodae he higher sales. Therefore, on average, coss will be more sensiive o sales increases han o sales decreases, reflecing operaional asymmery in resource commimens. Alhough sicky coss research has mosly focused on asymmeries in coss, he asymmeric behavior of coss also implies an asymmeric relaion beween earnings and sales increases versus decreases, which is a mirror image of he asymmery for coss (e.g., Banker and Chen, 2006). Empirical models of condiional conservaism ypically use sock reurns as he primary proxy for fuure gains and losses. 5 However, operaional facors refleced in he direcion of sales changes may have wo imporan effecs on he piecewise linear relaion beween earnings and sock reurns. Firs, he behavior of coss and earnings is likely o be asymmeric for operaional reasons relaed o cos sickiness, which could have a confounding effec on he esimaes of he asymmery ypically aribued o condiional conservaism. Second, by providing a more deailed view of he firm s operaions, sales increases and decreases may conain incremenal informaion abou he bookable porion of fuure gains and losses refleced in sock reurns, and hence he direcion of sales changes could have a moderaing effec on asymmeric recogniion of gains and losses. Operaional facors are likely o lead o upward bias in he sandard esimaes of conservaism because of he prevalence of sicky coss. As recen sudies (e.g., Anderson e al., 2003; Banker and Chen, 2006; Weiss, 2010) have shown, mos of he major componens of operaing coss decrease significanly less for sales decreases han hey increase for sales increases. As coss ener earnings wih a negaive sign, he direcion of his asymmery is reversed for earnings, i.e., 5 One noable excepion is Ball and Shivakumar (2005), who sudy privae firms ha by definiion do no have observable sock reurns. They use earnings decreases (Basu, 1997) and negaive cash flows (or cash flow decreases) as alernaive non-marke proxies for bad news. Buijink e al. (2011) examine earnings skewness (Basu, 1995) as an addiional non-marke conservaism measure when analyzing privae firms. 8

9 earnings are more sensiive o sales decreases han o sales increases when coss are sicky. Since sales changes are posiively correlaed wih sock reurns, cos sickiness implies ha he associaion beween earnings and sock reurns should be sronger for negaive reurns han for posiive reurns even if conservaism is absen. 6 The sandard models do no conrol for his confounding effec of operaional facors. Therefore, hey may suffer from omied variable bias, as he esimaes of asymmeric imeliness in hose models capure boh he asymmeric effec of condiional conservaism and he addiional confounding effec of he operaional asymmery beween sales increases and decreases. Boh ypes of asymmery ac in he same direcion, leading o upward bias in he sandard esimaes of asymmeric imeliness. By conrolling for he direcion of sales changes (using separae subsamples for sales increases and decreases), we are able o separae he asymmeric effec of condiional conservaism from he confounding effec of operaional facors refleced in sales increases and decreases, hus avoiding he upward bias in asymmeric imeliness esimaes. 7 This reasoning leads o he following hypohesis: Hypohesis 1: Adding conrols for he confounding effec of he direcion of sales changes in he sandard models reduces he esimaes of asymmeric imeliness. 6 For example, in our daa, he correlaion beween log-change in sales and sock reurns for he full sample is 0.161, significan a he 1 percen level, which means ha he magniude of sales changes is a correlaed omied variable ha may affec he esimaes of slope in a regression of earnings on sock reurns. The correlaion beween he negaive porion of sales changes (i.e., log-change in sales imes he sales decrease dummy) and he negaive porion of sock reurns (i.e., sock reurns imes he negaive reurns dummy) is 0.186, significan a he 1 percen level, which means ha he magniude of sales decreases is a correlaed omied variable ha may affec he esimaes of he asymmery in a piecewise-linear regression of earnings on sock reurns. 7 In our main models, we conrol for he direcion bu no he magniude of sales changes. This allows us o use sandard conservaism models, esimaed separaely for he sales increase and sales decrease subsamples. The magniude of sales change is a correlaed omied variable in hese regressions, which may affec he esimaes of he average slope on sock reurns. However, because cos sickiness represen asymmery beween sales increases and decreases bu no asymmery wihin each subsample, his omied variable will no affec he esimaes of he asymmery in he relaion beween earnings and sock reurns wihin each subsample. In robusness checks, we direcly conrol for boh he direcion and he magniude of sales changes, wih similar resuls. 9

10 In addiion o his confounding effec, operaional facors also have a poenially imporan moderaing effec on asymmeric recogniion of fuure gains and losses in earnings. Alhough sock reurns in an efficien marke fully summarize informaion abou fuure cash flows, no all of hose fuure gains and losses are bookable (i.e., can be recognized in accruals, Ball and Shivakumar, 2006). For example, fuure growh opions and synergies are fully refleced in sock reurns bu no recognized as asses on he balance shee, and herefore fuure gains and losses associaed wih hem should no be capured in accruals for he curren period (Roychowdhury and Was, 2007). Therefore, by providing a more deailed view of he composiion of fuure gains and losses, operaional facors are likely o provide incremenal informaion (beyond ha conained in sock reurns) abou he bookable componen of hose gains and losses. For he same oal level of losses embedded in sock reurns, we expec he bookable porion of losses o be larger in he case of decreasing sales, for wo reasons. Firs, when sales decrease in he curren period, he losses refleced in sock reurns are more likely o be associaed wih asses already in place (as opposed o fuure growh opions), and hence hey are more likely o be bookable in curren period accruals. Second, audiors are required o examine muliple impairmen indicaors, including operaional facors such as sales or cash flows. In paricular, Saemen on Audiing Sandards (SAS) 59 (AICPA, 1989) recommends (paragraph 6) ha he audior examine muliple indicaors o evaluae wheher he eniy can coninue as a going concern and if necessary (paragraphs 10-11) evaluae he effec on financial saemens of disconinued operaions and asse impairmens. Financial accouning sandards recommend alernaive impairmen indicaors. Saemen of Financial Accouning Sandards (SFAS) 121 (FASB, 1995) requires a wo-sep impairmen es for long-lived asses. The firs sep is o es wheher he book value of he asse is less han he 10

11 sum of undiscouned fuure cash flows from he asse, which would normally be based on sales forecass less cos forecass. If he asse is deemed o be impaired based on his comparison, in he second sep he asse is wrien down o is fair value, esimaed as he sum of discouned fuure cash flows (in he absence of quoed marke prices) or based on marke prices (for asses raded in acive markes). A sales decrease in he curren period is likely o direcly reduce he esimaes of fuure cash flows in impairmen ess, which increases boh he likelihood (hrough reduced esimaes of oal undiscouned fuure cash flows) and he size of he asse impairmen (hrough reduced esimaes of oal discouned fuure cash flows or fair value). 8,9 Therefore, for he same level of losses indicaed by sock reurns, companies will be more likely o recognize hose losses in he curren period when oher impairmen indicaors (such as sales changes) also indicae bad news. Similarly, he FASB recommends use of muliple indicaors for assessing wheher goodwill is impaired. Even when fair value of a reporing uni can be deermined based on marke prices of equiy, SFAS 142 (FASB 2001, paragraph 23) cauions ha marke capializaion may no be represenaive of he fair value and herefore, sock prices need no be he sole measuremen basis of he fair value of a reporing uni. Alernaive measuremen mehods include presen value echniques (paragraph 24) and valuaion based on muliples of earnings or revenue or a similar performance measure (paragraph 25). Thus, even when marke prices of equiy provide a direc measure of fair value, SFAS 142 leaves room for addiional valuaion echniques based on operaional performance measures such as sales revenue. 8 For asses ha are raded in acive markes, he marke prices (and, hence, he fair value esimaes in sep 2) may no fully reflec he consequences of a (firm-level) sales decrease. However, sep 1 is based on esimaes of undiscouned fuure cash flows regardless of availabiliy of quoed marke prices (paragraph 6 in SFAS 121). Therefore, even when a sales decrease does no affec he fair value in sep 2, i increases he likelihood of impairmen in sep 1, hus increasing he expeced value (likelihood size) of asse impairmen. 9 Beaver and Ryan (2005) model and simulae he muli-period effecs of probabilisic wrie-downs while Basu (2005) develops he implicaions of probabilisic wrie-downs for modifying he conemporaneous Basu (1997) asymmeric imeliness regressions. 11

12 As his discussion demonsraes, for he same amoun of bad news refleced in sock reurns, he losses recognized in curren period earnings are likely o be larger in he case of decreasing sales, reinforcing he asymmery in he relaion beween earnings and sock reurns. Thus, in addiion o he confounding effec described in Hypohesis 1, he direcion of sales change is likely o have an imporan moderaing effec on asymmeric imeliness. We summarize his predicion in our second hypohesis: Hypohesis 2: The direcion of sales change has a moderaing effec on asymmeric imeliness of earnings, where asymmeric imeliness is greaer in he case of sales decreases han in he case of sales increases Differenial impac of sales change direcion on separable and firm-level asse wriedowns Our predicions in Hypohesis 2 rely on he conjecure ha he direcion of sales changes provides an addiional signal for he bookable porion of gains and losses, and ha companies use his signal o decide wheher o wrie down an asse, and if so, how much. To validae his conjecure, we direcly examine he relaion beween wrie-downs (of angible asses and goodwill) and wo signals: negaive sock reurns and decreasing sales. This supplemenary evidence is essenial because wrie-downs are by definiion linked o condiional conservaism, and unrelaed o any confounding operaional effec of sales increases and decreases on concurren coss and earnings. Condiional on sock reurns, we expec sales decreases o be associaed wih larger wrie-downs, reflecing he moderaing effec of he direcion of sales changes. Compusa separaes wrie-downs of goodwill and oher unamorized inangibles from oher 12

13 asse wriedowns, which disinguishes asses wih indefinie fuure lives from hose wih finie lives. Noably, goodwill and oher unamorized inangibles ofen incorporae unverifiable esimaes of fuure growh opions. The relaive impac of sales decreases versus negaive sock reurns is likely o differ sysemaically beween hese wo ypes of wrie-downs. In paricular, sales decreases reduce he esimaes of fuure cash flows from asses in place, which direcly increases he likelihood and size of asse wrie-downs under SFAS 121 (FASB, 1995) and SFAS 144 (FASB, 2001), whereas heir impac on he value of goodwill is far less direc. 10 By conras, sock reurns reflec all economic losses associaed wih asses in place and goodwill equally. Therefore, we expec he impac of sales decreases o be relaively more imporan for asse wrie-downs han for goodwill impairmen, and he impac of negaive reurns o be relaively more imporan for goodwill impairmen han for asse wrie-downs Modified firm-year conservaism score Khan and Was (2009) develop a new firm-year conservaism measure CScore, which is based on annual asymmeric imeliness regressions wih conrols for hree sandard proxies of demand for conservaism: firm size, marke-o-book raio and leverage (we will refer o hem as Khan-Was conrols for breviy). Afer esimaing hose regressions, hey compue he conservaism score for each firm-year (KW_CScore in our noaion) as he prediced asymmeric imeliness coefficien condiional on he values of he Khan-Was conrols for ha firm-year. They show ha he cross-secional and ime-series variaion in KW_CScore is 10 For example, paragraph 29 of SFAS 142 saes: If goodwill and anoher asse (or asse group) of a reporing uni are esed for impairmen a he same ime, he oher asse (or asse group) shall be esed for impairmen before goodwill... If he asse group was impaired, he impairmen loss would be recognized prior o goodwill being esed for impairmen. Thus, alhough sales decreases may lead o impairmen of boh angible asses and goodwill, he size of goodwill impairmen will reflec only he residual amoun afer fully recognizing he impairmen loss for angible asses. Therefore, he associaion beween sales decreases and goodwill impairmen is likely o be less direc han ha for angible asse wrie-downs. 13

14 consisen wih oher manifesaions of condiional conservaism, which indicaes ha KW_CScore successfully capures some of he sysemaic differences in conservaism across firms and over ime. 11 We argue ha he Khan-Was conservaism score can be improved by recognizing he effecs of operaional facors, and propose a new Modified CScore (Mod_CScore), which incorporaes he moderaing effec of he direcion of sales changes on asymmeric imeliness and excludes is confounding effec. A valid firm-year conservaism measure has o have wo key properies: (1) i has o successfully separae variaion in conservaism from oher confounding sources of variaion in asymmeric earnings behavior (as he laer will disor he inferences based on he firm-year conservaism measure because of a correlaed omied variable problem), and (2) i has o successfully capure a large share of variaion in (unobserved) rue firm-year conservaism. As we show nex, accouning for he effecs of operaional facors in a firm-year conservaism score is likely o subsanially improve is performance on boh dimensions. Firs, as we explain in developing Hypohesis 1, operaional facors are likely o disor he sandard esimaes of asymmeric imeliness because of he confounding effec relaed o sicky coss. As prior sudies demonsrae, he degree of cos sickiness varies sysemaically across firms and over ime (e.g., Anderson e al., 2003). This sysemaic variaion in cos sickiness is likely o bias no only he esimaes of average asymmeric imeliness in sandard models (Hypohesis 1) bu also he esimaes of cross-secional and ime-series variaion in asymmeric imeliness, as changes in he degree of cos sickiness can be misaken for changes in he degree of asymmeric imeliness. Therefore, in esimaion of firm-year conservaism measures, i is essenial o add conrols for his confounding effec of operaional facors (hese conrols do no 11 The variaion in conservaism score is driven by differences in Khan-Was conrols across firms and over ime, and differences in conservaism parameer esimaes across years. Therefore (and by design), CScore does no capure oher sources of variaion in conservaism ha are unrelaed o hese observable facors. 14

15 ener he compuaion of conservaism scores). Second, as we argue in Hypohesis 2, even afer separaing his confounding effec from he conservaism measures, operaional facors are likely o have a subsanial moderaing effec on asymmeric recogniion of gains and losses, leading o sysemaic variaion in asymmeric imeliness boh across firms and over ime. Sandard asymmeric imeliness models do no accoun for his source of variaion, which reduces heir abiliy o explain variaion in he underlying rue firm-year conservaism. Incorporaing he moderaing effec of operaional facors in esimaion of conservaism scores is herefore likely o subsanially improve heir abiliy o explain cross-secional and ime-series variaion in conservaism. Based on hese argumens, we develop a Modified CScore (Mod_CScore), which excludes he confounding effec and incorporaes he moderaing effecs of operaional facors represened by he direcion of sales changes, along wih he Khan-Was conrols. To accoun for boh effecs, we esimae he Khan-Was asymmeric imeliness regression (and hen compue he prediced firm-year scores) separaely for he sales increase subsample and for he sales decrease subsample. Since he asymmeric imeliness esimaes in his approach capure variaion in earnings asymmery wihin each subsample (bu no variaion in earnings asymmery beween sales increases and decreases), hey are no conaminaed by he confounding effec of he direcion of sales changes (and, more imporanly, hey are no conaminaed by ime-series and cross-secional variaion in his confounding effec). 12 Since he asymmeric imeliness parameers are allowed o differ beween he wo subsamples, hey direcly capure he moderaing effec of he direcion of sales changes on asymmeric imeliness, and hence, accoun for his addiional source of variaion in condiional conservaism across firms and over ime. 12 Imporanly, our approach is no equivalen o merely adding a sales decrease dummy in Khan-Was model. Insead, all parameers have o be subsample-specific, o ensure ha he esimaes are based only on he wihinsubsample variaion. 15

16 Therefore, our Modified CScore is likely o improve on he Khan-Was CScore because (1) i separaes he sysemaic variaion in conservaism from variaion in he confounding effec of he direcion of sales changes, and (2) i fully accouns for he variaion in conservaism explained by he Khan-Was conrols, 13 plus addiional ime-series and cross-secional variaion explained by he moderaing effec of he direcion of sales changes. To illusrae he firs advanage of he Modified CScore, we exploi variaion in GDP growh, which has been shown in prior sudies o affec boh sickiness (e.g., Anderson e al., 2003) and conservaism (e.g., Jenkins e al., 2009). Anderson e al. (2003) and Banker e al. (2012) show ha higher GDP growh increases cos sickiness because i makes managers more opimisic abou fuure sales. When he economy is booming, managers are relucan o cu commied resources if sales decrease (because hey view he sales decrease as emporary) and are willing o expand commied resources if sales increase (because hey expec he increased sales o persis in he fuure). This increases cos sickiness. As we show in Hypohesis 1, sandard conservaism models are likely o misake cos sickiness for conservaism, and hence hey are likely o misake increased cos sickiness for increased conservaism. Therefore, he Khan-Was CScore, which is based on sandard models, is likely o overesimae he impac of GDP growh on asymmeric imeliness (which also leads o biased inferences abou he role of he business cycle in ime-series variaion in conservaism). By conras, he Modified CScore is robus o his confounding variaion, leading o he following hypohesis: Hypohesis 3a: The impac of GDP growh on Khan-Was CScore is biased upwards relaive o he impac of GDP growh on he Modified CScore. 13 Technically, since our model ness he Khan-Was specificaion under he resricion ha all parameers are idenical beween he wo subsamples, i mus capure all of he variaion explained by he Khan-Was model (plus he addiional variaion due o he moderaing effec of sales changes). 16

17 Jenkins e al. (2009) find ha condiional conservaism increases during recessions, and aribue his finding o (1) increased liigaion risk during economic conracions, when sharp decreases in sock prices (which may lead o lawsuis) are more likely, (2) increased scruiny on he poenial downside risks from regulaors, and (3) greaer need for exernal financing. All of hese facors increase he demand for conservaism during periods of weak macroeconomic growh (measured using GDP growh rae 14 ), i.e., higher GDP growh reduces conservaism scores: Hypohesis 3b: Higher GDP growh reduces conservaism scores. 15 Imporanly, he upward (posiive) bias and he negaive main effec of GDP growh (Hypoheses 3a and 3b, respecively) are manifesaions of wo fundamenally differen effecs of GDP growh on asymmeric earnings behavior, which ac in opposie direcions. The upward bias reflecs he posiive impac of GDP growh on cos sickiness (i.e., asymmery in real resource commimens, where favorable macroeconomic condiions lead o more aggressive choice of resource levels due o managerial opimism), which increases he asymmery in earnings behavior for operaional reasons. The negaive main effec reflecs he impac of GDP growh on condiional conservaism (i.e., asymmery in he accouning sysem, where favorable macroeconomic condiions reduce he need for imely loss recogniion), which reduces he asymmery in earnings behavior due o less conservaive financial reporing choices. The 14 We use a coninuous measure of macroeconomic growh (raher han a dummy for economic conracions) because all hree of he facors menioned above are likely o vary wih he magniude (and no jus he sign) of macroeconomic growh raes. 15 Because of he upward bias in he impac of GDP growh on he Khan-Was CScore (Hypohesis 3a), i may violae Hypohesis 3b if he upward bias is sufficienly srong. 17

18 differen direcions of hese wo effecs (and he differen signs hey imply in Hypoheses 3a, 3b) are essenial empirically because, in addiion o evaluaing he performance of our Modified CScore vis-a-vis he Khan-Was CScore, hey provide a more general es of wheher our approach successfully separaes he effecs of condiional conservaism from he confounding effec of he direcion of sales changes. 16 The direcion of sales changes also plays an imporan moderaing role in he impac of GDP growh on conservaism, affecing all hree mechanisms in Jenkins e al. (2009). Firs, while weak macroeconomic growh increases liigaion risk for all firms, firms wih unfavorable operaional facors are likely o face paricularly high risk of lawsuis as hey are more likely o experience sharp drops in sock prices. Second, he increased scruiny from regulaors during downurns is likely o be focused disproporionaely on firms wih unfavorable operaional performance. Third, firms are more likely o be cash consrained when slow macroeconomic growh is compounded by unfavorable operaional facors, which increases heir need for ouside funding. All hree mechanisms imply ha he (negaive) impac of GDP growh on conservaism scores should be sronger for firms wih unfavorable operaional facors, such as decreasing sales: Hypohesis 3c: Higher GDP growh reduces conservaism scores o a greaer exen for sales decreases han for sales increases. In supplemenary ess, we direcly examine he incremenal effec of GDP growh and is ineracion wih sales decreases on wrie-downs, because wrie-downs are direcly linked o 16 If he effecs of conservaism are no successfully separaed from he confounding effec of operaional facors, hen boh esimaes would reflec a combinaion of hese wo phenomena, and herefore, he esimaes of boh effecs of GDP growh would likely have he same sign. Therefore, he prediced opposie signs of he wo effecs of GDP growh make his a powerful es of wheher we successfully disenangle he wo phenomena. 18

19 condiional conservaism. Condiional on sock reurns, higher GDP growh should have a negaive effec on wrie-downs, and his negaive effec should be sronger (i.e., more negaive) in he case of sales decreases Daa and Esimaion Models 3.1. Sample selecion and descripive saisics We use he combined Compusa/CRSP annual sample beween 1963 and Following Khan and Was (2009), for each firm-year observaion we require valid daa for curren year s earnings, he previous fiscal year-end sock price and book values of asses and equiy, and reurns daa. We also discard 1% exreme values on each ail for all regression variables. The variable definiions are presened in Table 1. We presen he descripive saisics and he correlaion marix in panels A and B of Table 2. Earnings before exraordinary iems and disconinued operaions is scaled by lagged marke value of equiy (or, equivalenly, earnings per share is scaled by lagged share price). On average, earnings before exraordinary iems and disconinued operaions is equal o 4.7% of lagged marke value (he median is 6.4%). Scaled earnings is negaively skewed (mean < median) consisen wih he presence of condiional conservaism (Basu, 1995; Ball e al., 2000). Annual sock reurns are 11.3% on average (median 6.1%). Sales decrease observaions ( =1) are 23.4% of he sample, while sock reurns are negaive ( =1) for 43.5% of he sample. The correlaion beween sock reurns ( ) and log-changes in sales (ΔlnSALES ) is 0.161, 17 The sign of his ineracion effec is easier o inerpre in erms of decreases in GDP growh. When he decrease in GDP growh (i.e., less favorable macroeconomic condiions) is compounded by sales decreases, wrie-downs should increase o a greaer exen. 18 We end he sample in 2008 because companies loss recogniion behavior and operaional behavior may have changed subsanially following he financial crisis of Our main resuls hold when we exend he sample o

20 significan a he 1 percen level, suggesing ha sales changes may be an imporan correlaed omied variable in he relaion beween earnings and sock reurns. The correlaion beween he negaive reurns dummy and he sales decrease dummy is 0.142, and he correlaion beween he negaive porion of sock reurns ( ) and he negaive porion of sales changes ( ΔlnSALES ) is 0.186, boh significan a he 1 percen level, suggesing ha sales decreases may be an imporan correlaed omied variable in he piecewise-linear relaion beween earnings and posiive versus negaive sock reurns. A he same ime, he correlaions beween hese variables are far less han 1, indicaing ha here is enough independen variaion o reliably isolae he effecs of he direcion of sales changes The esimaion models We sar from he sandard models in conservaism research, and hen exend hem o accommodae he confounding and moderaing effecs of operaional facors represened by he direcion of sales changes. Our saring poin is he basic Basu (1997) model, esimaed on he full sample: Model A0 (Basu model) X / MVi, 1 0 1i, 2i, 3i, i, (1) where X represens earnings of firm i in year, MV -1 is he marke value of he firm a he end of year -1, is sock reurns compued for he 12-monh period of fiscal year, is a dummy variable equal o 1 if sock reurns are negaive, is a dummy variable equal o 1 if sales decreased from year -1 o year, and ε is an error erm. Sock reurn,, proxies for news regarding fuure cash flows, and negaive reurns 20

21 ( =1) proxy for bad news. The Basu asymmeric imeliness coefficien, α 3, capures he difference in slopes beween posiive and negaive reurns (good and bad news, respecively). Condiional conservaism implies ha α 3 should be posiive, i.e., bad news (negaive reurns) should be refleced in earnings more quickly and more fully han good news (posiive reurns). We exend he Basu model o incorporae (1) he moderaing effec of he direcion of sales changes on asymmeric imeliness, using subsample-specific coefficiens on he conservaism erm for he sales increase and sales decrease subsamples, and (2) he confounding operaional effec of he direcion of sales changes, using subsample-specific parameers for oher variables. This model is equivalen o esimaing Model A0 separaely for each subsample. To conduc hypohesis ess across he wo subsamples, we also esimae his model as a pooled regression wih ineracion erms: Model A1 (modified Basu model) X / MV 1 ( ( 0, SU 0, 1, SU 1, 2, SU 2, 3, SU 3, ) SU ) (2) where SU =1- is a dummy variable equal o 1 if sales of firm i increased in year, and he remaining variables were defined previously. The erms α 2,SU SU and α 2, (along wih he lower-level ineracions) in Model A1 capure he confounding effec of he direcion of sales changes. In oher words, he esimaes of α 2,SU and α 2, reflec no only he exen of recogniion of good news in earnings (he convenional inerpreaion in he Basu model) bu also he asymmeric effecs of he correlaed omied variable (he magniude of sales changes). The asymmeric imeliness coefficien in Model A1 is equal o α 3,SU for sales increases and α 3, for sales decreases. Hypohesis 1 predics ha adding conrols for he confounding effec of 21

22 he direcion of sales changes should reduce he esimaes of conservaism, i.e., he esimae of average asymmeric imeliness in Model A1 (he weighed average of α 3,SU and α 3, ) should be lower han he esimae of α 3 in Model A0. Hypohesis 2 implies α 3,SU < α 3,, i.e., bad news embedded in negaive sock reurns should be recognized in earnings more quickly and more fully in he case of decreasing sales han in he case of increasing sales, reflecing he moderaing effec of he direcion of sales changes. We also examine he impac of operaional facors on he Khan and Was (2009) CScore measure, which has been used in several recen conservaism papers (e.g., Hui e al., 2012; Eredge e al., 2012). Khan and Was (2009) exend Basu-ype analysis by esimaing asymmeric imeliness condiional on observable correlaes of demand for conservaism (size, leverage and marke-o-book raio), and use he esimaes o generae a firm-year conservaism score. Since he direcion of sales changes may be sysemaically correlaed wih he Khan- Was insrumens, and boh ypes of facors lead o sysemaic variaion in asymmeric imeliness across firms and over ime, we also examine he confounding and moderaing roles of he direcion of sales changes in Khan-Was-ype analysis. We firs include he Khan-Was conrols in he Basu model: Model B0 (Khan-Was model) X / MV 1 0, 9..11, 1, KW 1 2, 12, 3..5, KW , 6..8, KW KW 1 1 (3) where KW -1 is he vecor of beginning-of-year Khan-Was conrols, which consiss of firm size SIZE -1, book-o-marke raio BM -1 and leverage LEV -1, and he res of he erms were 22

23 defined previously. 19 Following Khan and Was (2009), we esimae Model B0 using annual cross-secional regressions, and hence he parameers α 0, -α 15, are indexed by year. The prediced asymmeric imeliness for observaion (i.e., he incremenal slope on negaive sock reurns, condiional on he values of he conrol variables KW -1 ) in Model B0 is equal o: Khan-Was conservaism score KW _ CScore KW SIZE BM LEV (4) 12, , 1 12, 13, 1 14, 1 15, 1 This expression is he sandard formula for regression slope wih respec o one of he independen variables ( in his case), condiional on he values of oher independen variables, in regression models wih ineracions (i.e., by definiion, in a modified model wih addiional or differen conrol variables, he relevan CScore will consis of all regression erms ha are ineraced wih ). 20 Khan and Was go beyond his convenional inerpreaion, and frame KW_CScore as a firm-year conservaism score, which can be employed as an independen variable in analysis of he impac of condiional conservaism on oher oucomes. Following Khan and Was (2009), we expec larger firms o exhibi less conservaism (α 13 < 0) as hey have a richer informaion environmen (e.g., more analys following), which reduces informaion asymmeries, and hence, lowers he demand for conservaism (see also Giner and Rees, 2001; Basu, 2001). Firms wih higher leverage should exhibi more conservaism (α 14 > 0) as hey face a more severe agency conflic beween lenders and 19 We use beginning-of-year values of size, book-o-marke and leverage, because end-of-year values of hese variables are mechanically affeced by concurren earnings (our dependen variable), which could disor he esimaes for end-of-year variables. The resuls are robus when we use end-of-year variables. We use book-omarke raio (raher han marke-o-book) because i is monoone wih respec o posiive and negaive book value (he resuls are robus when we use marke-o-book raio and resric he sample o posiive book values). 20 In fac, Khan and Was cauion agains mechanical applicaion of heir CScore, and poin ou ha addiional or differen conrols can be used in place of KW (page 148). 23

24 shareholders, which increases he demand for conservaism (Ball and Shivakumar, 2005; Zhang, 2008). Higher book-o-marke raio should be associaed wih greaer conservaism (α 15 > 0) because (1) higher beginning book-o-marke proxies for lower uncondiional (and condiional) conservaism, and (2) he associaion beween prior uncondiional conservaism and subsequen condiional conservaism is likely o be negaive (e.g., Giner and Rees, 2001; Basu, 2001; Beaver and Ryan, 2005; Pae e al., 2005). We nex incorporae he wo effecs of he operaional facors, represened by he direcion of sales changes, by making all parameers subsample-specific. This specificaion is equivalen o esimaing Model B0 separaely for he sales increase and sales decrease subsamples, or i can be esimaed as a pooled regression wih ineracion erms. Model B1 (modified Khan-Was model): X / MV 1 ( ( 0,, SU 9..11,, SU 0,, 9..11,, 1,, SU KW KW 1 1,, 1 2,, SU 12,, SU 2,, 12,, 3..5,, SU 3..5,, KW ,, SU KW ,, KW KW 6..8,, SU ,, 1 KW KW 1 1 ) SU ) (5) where all variables were defined previously. The erms α 6..8,,SU KW -1 SU and α 6..8,, KW -1 (along wih lower-level ineracions) capure he effec of he Khan- Was variables on cos sickiness, and hus allow us o separae he variaion in he confounding effec relaed o cos sickiness from he impac of hese variables on condiional conservaism (he laer is capured by he ineracion erms KW -1 wih subsample-specific coefficiens α ,,SU and α ,, ). Similar o Model A1, he asymmeric imeliness in Model B1 varies wih he direcion of sales changes (reflecing is moderaing effec), and similar o Model B0, i varies wih he Khan- 24

25 Was correlaes of demand for conservaism. 21 Condiional on he conrol variables SU,, KW -1, i is equal o: Modified conservaism score Mod _ CScore ( ( 12,, SU 12,, SU ,, SU 13,, SU KW SIZE 1 1 ) SU 14,, SU ( BM 12,, ,, 15,, SU LEV KW 1 1 ) SU ) (6) ( 12,, 13,, SIZE 1 14,, BM 1 15,, LEV 1 ) Similar o he original conservaism score (KW_CScore ) of Khan and Was, our Modified CScore (Mod_CScore ) is based on he sandard formula for regression slope wih respec o (condiional on he values of he variables is ineraced wih). Hypohesis 1 implies ha he average and median of Mod_CScore in Model B1 should be lower han he average and median of KW_CScore in Model B0, reflecing he confounding effec of he direcion of sales change. Hypohesis 2 implies ha he average and median Mod_CScore should be lower for sales increases han for sales decreases, represening he moderaing effec of he direcion of sales changes. Hypoheses H3a-H3c focus on he effecs of GDP growh on KW_CScore and Mod_CScore; we es hem by regressing he firm-year esimaes of conservaism scores on GDP growh as described laer in subsecion 4.2. Imporanly, Model B1 ness he Khan-Was Model B0 (and, consequenly, our Modified CScore ness he Khan-Was CScore) under he resricion ha all parameers are idenical beween he wo subsamples. Therefore, our Modified CScore is able o fully reproduce he variaion capured by he Khan-Was CScore, because he wo models are equivalen if his resricion holds in he daa, along wih addiional variaion (no capured by he Khan-Was 21 Imporanly, boh sources of variaion in asymmeric imeliness in Model B1 are afer conrolling for he parallel variaion in he confounding effec of sicky coss (hese conrols do no ener he compuaion of conservaism scores). We allow he coefficiens on he Khan-Was variables in asymmeric imeliness o vary wih he direcion of sales changes because he direcion of sales change may also have a moderaing effec on he relaion beween he Khan-Was conrols and conservaism. 25

26 CScore) in case his resricion is no appropriae in he daa (which will be he case if eiher he confounding or he moderaing effec of he direcion of sales changes is presen in he daa). In supplemenary ess o validae he incremenal informaion conen of he direcion of sales changes (which is he source of is moderaing effec on conservaism), we direcly examine he relaive impac of sales decreases and negaive sock reurns on wrie-downs. The esimaion model is: Model C WDi, / MV (7) where WD is he wrie-down amoun for firm i in year, and he res of he variables were defined previously. Similar o he sandard conservaism models, he associaion beween wriedowns and sock reurns is likely o be sronger for negaive reurns han for posiive reurns, which implies α 3 < The main parameer of ineres is α 4, which capures he incremenal effec of sales decreases afer conrolling for sock reurns. 23 We predic α 4 > 0, i.e., sales decreases should be associaed wih greaer bookable losses, and hence, larger wrie-downs for he same level of sock reurns. We esimae Model C for wo dependen variables from Compusa: goodwill impairmen and (oher) asse wrie downs. As discussed in secion 2, we expec he impac of sales decreases (capured by α 4 ) o be relaively more imporan for asse wrie-downs han for goodwill impairmen, and he impac of negaive sock reurns (capured by α 3 ) o be relaively more imporan for goodwill impairmen han for asse wrie-downs. In addiional ess, we also examine he incremenal effec of GDP growh (ΔGDP ) and is 22 As wrie-downs ener earnings wih a negaive sign, he prediced sign of α 3 is reversed relaive o he sandard models for earnings. 23 Unlike in Model A1, we do no include higher-level ineracion erms of wih and *, because hey would make i harder o characerize he relaive magniude of he impac of sales decreases in a ransparen way. 26