When Micro-news Interacts with Macro-News: Market Reaction to Earnings. Announcements on Macroeconomic News Announcement Days

Transcription

1 When Micro-news Ineracs wih Macro-News: Marke Reacion o Earnings Announcemens on Macroeconomic News Announcemen Days Absrac We es and provide evidence o suppor he hypohesis ha he complemenary informaion in macroeconomic news helps invesors inerpre earnings news and leads o more efficien sock valuaion. Despie he fac ha invesors allocae relaively less aenion o earnings announcemens on days wih imporan macroeconomic news announcemens, marke underreacion o earnings announcemens wih concurren macroeconomic news announcemens is significanly weaker. The pos-earnings-announcemen drif is reduced by up o 50% over he shor horizon due o he effec of macroeconomic news. In addiion, we show ha he effec is sronger for firms wih greaer informaion uncerainy. Neverheless, here is no evidence ha managemens ime earnings announcemens based on pre-scheduled macroeconomic news announcemens. Keywords: Earnings announcemens; macroeconomic news announcemens; invesor underreacion; limied invesor aenion; caegory-learning behavior. JEL Classificaion: G12, G14 2

2 I. Inroducion The lieraure documens srong evidence of invesor underreacion o corporae evens. The mos well-known phenomenon is he pos-earnings-announcemen drif (PEAD) in sock reurns (Ball and Brown, 1968; Foser, Olsen, and Shevlin, 1984; and Bernard and Thomas, 1989; 1990). Tha is, firms reporing posiive (negaive) unexpeced earnings, on average, experience posiive (negaive) abnormal reurns following he earnings announcemen. 1 The lieraure has proposed a number of poenial explanaions, from boh raional and behavioral perspecives, for invesor underreacion. Several recen sudies aribue PEAD o limied invesor aenion and oher forms of invesor cogniive consrain (e.g., Hirshleifer and Teoh, 2005; DellaVigna and Polle, 2009; Hirshleifer, Lim, and Teoh, 2009). For example, Hirshleifer, Lim, and Teoh (2009) find ha he immediae invesor reacion o a firm s earnings surprise is much weaker and drif o earnings surprises is much sronger when a larger number of relaed companies also announce earnings on he same day. This is because invesors have limied power o process large amouns of informaion a he same ime. In his paper, we examine he effec of concurren macroeconomic news announcemens on invesor reacion o earnings news. Our sudy is moivaed by he following heories and argumens. Firs, all news does no receive equal aenion from invesors and, in paricular, macroeconomic news may receive more aenion from invesors han firm-specific news as a resul of he aenion consrain. Based on psychological evidence ha aenion is a limied cogniive resource, Peng (2005) and Peng and Xiong (2006) show ha invesors have o be conscious in allocaing heir limied aenion capaciy given he vas amoun of informaion available in he markeplace. More specifically, Peng and Xiong (2006) show ha limied 1 Several sudies also documen invesor underreacion o oher corporae evens, such as share repurchase announcemens (Ikenberry, Lakonishok, and Vermaelen, 1995), dividend iniiaions and omissions (Michaely, Thaler, and Womack, 1995), and sock spli announcemens (Ikenberry and Ramnah, 2002). 1

3 invesor aenion leads o caegory-learning behavior, ha is, invesors end o process informaion abou macroeconomic fundamenals before processing firm-specific informaion. 2 Second, many invesors, such as insiuional invesors, employ a op-down approach in heir porfolio managemen. 3 A op-down invesor firs makes decisions on asse allocaions wih he desired risk reurn rade-off based on economic oulook and macroeconomic fundamenals. Thus, i is likely ha invesors pay more aenion o macroeconomic news. We posi and empirically es wo compeing hypoheses wih regard o he effec of macroeconomic news announcemens on invesor reacion o earnings news. On he one hand, by paying more aenion o macroeconomic news, invesors are disraced from earnings announcemens due o he limied informaion processing abiliy. As such, he disracion will likely aggravae invesor underreacion o earnings news. On he oher hand, he complemenary informaion in macroeconomic news may acually help invesors inerpre earnings news and lead o weaker invesor misreacion. This is because macroeconomic news no only conains imporan informaion abou he sae of he economy bu also helps disenangle he sysemaic componen of earnings news and, as such, leads o more efficien sock valuaion. 4 The argumen is formally presened under he model of Vuoleenaho (2002) in Secion III.B and suggess ha he complemenary informaion in macroeconomic news abou updaed discoun raes and 2 Aenion is a consrain no only for unsophisicaed individual invesors bu also for professional invesors. For example, Sims (2003) sudies he implicaions of informaion-processing consrains in a general dynamic conrol problem. Kacperczyk, Nieuwerburgh, and Veldkamp (2009) examine how muual fund managers allocae heir limied aenion beween aggregae marke-level informaion and firm-specific informaion during differen phases of he business cycle. Given he fac ha many specialiss handle muliple securiies on he NYSE, Corwin and Coughenour (2008) and Chakrabary and Moulon (2009) documen he effec of he limied aenion of specialiss on marke making. 3 As described in Bodie, Kane, and Marcus (Essenials of Invesmens, 9h ediion, 2013, McGraw- Hill/Irwin), a op down acive invesmen sraegy involves hree main seps: asse allocaion, securiy selecion, and implemenaion. 4 Exising sudies documen evidence ha macroeconomic news leads o more raional pricing of individual socks. Savor and Wilson (2014) show ha sock reurn paerns are much easier o reconcile wih sandard asse pricing heories on macroeconomic news announcemen days. 2

4 earnings announcemens abou firm cash flows may help invesors value socks more efficienly and reduce misreacion o earnings surprises. In his paper, we es he above hypoheses based on invesor reacion o earnings announcemens. Specifically, we compare he marke reacion o earnings announcemens wih concurren imporan macroeconomic news announcemens wih he marke reacion o hose wihou. The daa used in our sudy includes he CRSP daabase for sock reurns and Compusa for earnings announcemens and oher firm characerisics. The informaion on pre-scheduled macroeconomic news announcemens is obained from Bloomberg. The sock sample in our empirical analysis includes all common socks raded on he NYSE, AMEX, or NASDAQ. Our sample period is from January 2001 o December The main resuls of our sudy show ha while here is no significan differences in immediae marke reacions o earnings announcemens wih concurren imporan macroeconomic news announcemens and hose wihou, drif following earnings announcemens wih concurren imporan macroeconomic news announcemens is significanly weaker. For insance, for earnings announcemens wih no concurren macroeconomic news announcemens, he average reurn differenials beween he op and boom SUE (sandardized unexpeced earnings) deciles are 1.378%, 3.004% and 3.971% over one-week, one-monh, and one-quarer horizons following earnings announcemens, respecively. For earnings announcemens wih concurren macroeconomic news announcemens, hese corresponding numbers are significanly lower, a 0.632%, 1.941%, 3.022%, respecively. Tha is, macroeconomic news helps reduce invesor underreacion o earnings surprises by up o 50% over he shor horizon. We confirm ha he resuls are robus when we conrol for oher firm characerisics, such as size, he book-o-marke raio, liquidiy, idiosyncraic volailiy, and heir ineracions 3

5 wih SUE, as well as lagged sock reurns over differen horizons. Noe ha DellaVigna and Polle (2009) find less immediae responses and more drif for earnings announcemens on Fridays, when invesor inaenion is more likely, han on oher weekdays. As a robusness check, we also include a day of week dummy o conrol for poenial weekday effec in he regressions. Moreover, we divide macroeconomic news ino differen caegories and show ha he effec on invesor reacion is pervasive across differen ypes of macroeconomic news. Furhermore, we compue earnings surprises based on analys forecass insead of hisorical earnings and confirm he robusness of our main findings. The main implicaion of caegory-learning behavior is ha limied aenion capaciy leads invesors o pay more aenion o macroeconomic news and allocae relaively less aenion o firm-level news. Neverheless, i is also possible ha, as macroeconomic news announcemens draw invesor aenion o he overall marke, invesors may pay more aenion o firm-level news as well. Therefore, he weaker marke underreacion documened in our sudy could be aribued o increased invesor aenion o earnings announcemens raher han informaion conen in macroeconomic news. We empirically es he implicaions of caegorylearning behavior. Firs, we show ha rading volume and price variaion, proxies of invesor aenion, are higher on days wih macroeconomic news announcemens han on oher days. Second, using he same proxies, we show ha invesors allocae relaively less aenion o earnings announcemens on days wih macroeconomic news announcemens han on oher days. These findings are consisen wih implicaions of caegory-learning behavior and sugges ha i is more likely he informaion conen of macroeconomic news conribuing o weaker invesor underreacion o earnings surprises on days wih macroeconomic news announcemens. 4

6 In addiion, exising lieraure documens ha as invesors misreac o news, informaion uncerainy could furher exacerbae such misreacion. In paricular, invesor underreacion o corporae evens is sronger for firms wih greaer informaion uncerainy (Jiang, Lee, and Zhang, 2005; Zhang, 2006; Francis, Lafond, Olsson, and Schipper, 2007). If informaion in macroeconomic news helps reduce informaion uncerainy, we expec he effec of macroeconomic news o be sronger for firms wih greaer informaion uncerainy. Using idiosyncraic volailiy and analys coverage as proxies for informaion uncerainy, we show ha he effec of macroeconomic news on invesor reacion is indeed sronger for firms of higher idiosyncraic volailiy or lower analys coverage. For insance, for he sample of firms wih high idiosyncraic volailiy, macroeconomic news announcemens have a significan effec on invesor reacion o earnings surprises and reduce he drif following earnings announcemens by over 65% over he shor horizon. On he oher hand, for he sample of firms wih low idiosyncraic volailiy, macroeconomic news has an overall weaker effec on invesor reacion o earnings surprises. Finally, given he findings ha invesors pay relaively less aenion o earnings announcemens on days wih imporan macroeconomic news announcemens, i is naural o speculae ha managemens may have incenives iming earnings announcemens based on prescheduled macroeconomic news announcemens. Exising lieraure documens evidence ha managemens end o schedule earnings announcemen wih negaive surprises during days or hours wih less invesor aenion (Michaely, Rubin, and Vedrashko, 2013; and dehaan, Shevlin, and Thornock, 2015). Comparing variaions in reporing lags and surprises for earnings announcemens on days wih imporan macroeconomic news announcemens wih hose on days wihou, we find no evidence ha managemens ime earnings announcemens on 5

7 macroeconomic news announcemen days. There is no evidence eiher ha firms hide bad news by announcing negaive earnings on days wih macroeconomic news announcemens or highligh good news by announcing posiive earnings on days wihou macroeconomic news announcemens. Our sudy conribues o he following srands of he lieraure. Firs, our sudy conribues o he lieraure on invesor learning behavior. We provide empirical evidence supporing he predicions of Peng and Xiong (2006) on caegory-learning behavior for invesors wih limied informaion processing power. Second, our sudy conribues o he lieraure on he effec of macroeconomic news on sock valuaion. Savor and Wilson (2014) show ha sock prices behave more raionally on days wih imporan macroeconomic news announcemens. Our sudy provides one seing ha illusraes how macroeconomic news helps invesors inerpre earnings news and leads o more efficien sock prices. Third, our sudy sheds new ligh on wha drives invesor underreacion. Exising lieraure provides evidence ha more informaion of he same ype disracs invesor aenion and exacerbaes misreacion o earnings surprises. Our sudy shows ha informaion in macroeconomic news is complemenary o firm-level news and acually helps reduce invesor misreacion. Finally, our sudy provides furher evidence on he effec of informaion uncerainy on invesor reacion o news. We show ha resoluion in informaion uncerainy helps reduce invesor misreacion. The res of he paper is srucured as follows. Secion II describes he daa and mehodology employed in our analysis. Secion III presens he main empirical resuls wih various robusness checks. Secion IV performs furher analysis and Secion V concludes. 6

8 II. Daa The main daa used in our empirical analysis includes he CRSP daabase for sock reurns, Compusa for earnings announcemens and informaion on oher firm characerisics, he IBES daabase for analys forecass, and Bloomberg for macroeconomic news announcemens. CRSP sock reurns are adjused for delisings o avoid survivorship bias, following Shumway (1997). The sock sample includes common socks raded on he NYSE, AMEX, or NASDAQ in he CRSP daabase. Bloomberg provides he daes and imes for almos all pre-scheduled macroeconomic news announcemens. Due o daa availabiliy on macroeconomic news announcemens, our sample covers he period from January 2001 o December A. Unexpeced Earnings and Firm Characerisics We follow he lieraure (Foser, 1977; Foser, Olsen, and Shevlin, 1984) and consruc he measure of sandardized unexpeced earnings (SUE) as follows: E ( Qi, ) Qi, 4 i ( Qi, 1 Qi, 5 ) i (1) SUE i, Qi, E( Qi, ) (2) [ Q E( Q )] i, i, where Q i, denoes he quarerly earnings of firm i in quarer. The parameers i and i are esimaed using he mos recen 20 quarers of daa. Firm characerisics in he empirical analysis include size (SIZE), he book-o-marke raio (BM), momenum (MOM), he Amihud (2002) illiquid measure (ILLIQ), and idiosyncraic volailiy (IVOL). All variables are consruced following convenion in he lieraure (e.g., Fama and French, 2008), as described below: SIZE: he naural log of marke capializaion a he end of June of a year. 7

9 BM: he naural log of he book-o-marke raio. The book value of equiy is sockholders equiy plus balance-shee deferred axes and invesmen ax credi (TXDITC, from Compusa), if available, minus preferred sock liquidaing value (PSTKL), if available, or redempion value (PSTKRV), if available, or carrying value (PSTK). Depending on availabiliy, sockholders equiy is he Compusa variable SEQ, or CEQ+PSTK, or AT-LT, in ha order. All Compusa iems are measured for he fiscal year ending in calendar year - 1. The marke value of equiy is sock price imes shares ousanding a he end of December of year - 1, from he CRSP. We exclude firms wih negaive book value of equiy. MOM: 11-monh buy-and-hold reurn from July of year - 1 o May of year. ILLIQ: he Amihud (2002) illiquidiy measure is calculaed as he raio of he absolue daily sock reurn divided by he daily dollar rading volume and averaged over a given period. Since rading volume is defined differenly for NASDAQ socks and NYSE/AMEX socks, he rading volumes of NASDAQ socks are adjused by a facor of 0.7 (Boehmer, 2005). IVOL: he sandard deviaion of he residuals in he Fama French (1993) hree-facor model esimaed from daily reurns over a given period. Table I repors he cross-secional saisics of SUE and firm characerisics for seleced years in our sample period. During our sample period, he US sock marke experienced he collapse of he Inerne bubble, is recovery, he financial crisis, and he pos-crisis period. The saisics in he able clearly reflec he effecs of hese evens. The average of earnings surprises (SUE) is lower in 2001 and 2009 han in 2005 and 2013, as is average firm SIZE. The negaive log BM raio indicaes ha he book value is, on average, below marke value and he raio, on 8

10 average, is closer o one in 2001 and 2009 han in 2005 and The median MOM is negaive in 2001 and 2009 bu posiive in 2005 and Illiquidiy (ILLIQ) peaks in 2009 during he financial crisis period and remains high even aferward, in Furhermore, average volailiy (IVOL) is higher in 2001 and 2009 han in 2005 and B. Macroeconomic News Announcemens The lis of macroeconomic announcemens used in our analysis includes iniial jobless claims, changes in nonfarm payrolls, he Federal Open Marke Commiee (FOMC) rae decision, gross domesic produc (GDP) growh, he consumer confidence index, he ISM Manufacuring Index, he Consumer Price Index, he Universiy of Michigan Consumer Senimen Index, durable goods orders, new home sales, housing sars, he unemploymen rae, and reail sales. These announcemens are considered imporan because hey have significan impacs on financial markes based on he average Bloomberg relevance index during our sample period and a number of exising sudies (Flannery and Proopapadakis, 2002; Gerlach, 2007; Beber and Brand, 2009; Brenner, Pasquariello, and Subrahmanyam, 2009; Lee, 2012; Savor and Wilson, 2013, 2014, 2015). Table II repors he lis of macroeconomic news announcemens. N denoes he oal number of announcemens during he period from January 2001 o December Day and ime denoe he weekday or day of he monh and he ime (ET) of announcemen, respecively. Mos announcemens occur a eiher 8:30 a.m. or 10:00 a.m. excep he FOMC rae decision a eiher 12:30 p.m. or 14:15 p.m. The able also repors he number of announcemens wih no surprises, ha is, he acual announcemen is he same as he marke consensus. Oher han 9

11 FOMC rae decisions, of which abou 95% are consisen wih marke expecaions, mos news iems have a significan porion of announcemens wih surprises. III. Main Empirical Analysis A. Marke Reacions o Earnings Announcemens The lieraure documens ha invesors underreac o earnings informaion (e.g., Ball and Brown, 1968; Foser, Olsen, and Shevlin, 1984; Bernard and Thomas, 1989, 1990). These sudies show ha firms reporing posiive unexpeced earnings, on average, ouperform hose reporing negaive unexpeced earnings afer he earnings announcemen. In his secion, we examine sock reurns following earnings announcemens in our sample period. Following he lieraure, each quarer socks are assigned o deciles based on he SUE breakpoins of he previous quarer. The SUE are esimaed following he procedure described in Secion II.A. Table III repors he average SUE, average cumulaive abnormal reurns (CARs in percenage erm) for each decile porfolio, as well as spreads beween he op and boom deciles (D10-D1) over he wo-day announcemen window and differen horizons following earnings announcemens. The wo-day announcemen window covers he day of and he day afer he earnings announcemen. As documened in Michaely, Rubin, and Vedrashko (2013), abou 45% of earnings announcemens are made afer 4:00 p.m. or afer marke close. For hese announcemens, immediae marke reacions are refleced in reurns over he nex rading day. The five-, 10-, 21-, and 62-day horizons correspond o one-week, wo-week, one-monh, and one-quarer pos-earnings-announcemen periods. The decile D1 includes firms wih he lowes SUE rank and D10 includes firms wih he highes SUE rank. Abnormal daily sock reurn is calculaed as he difference beween daily sock reurns and he average daily reurn of he 10

12 corresponding size decile porfolio formed a he beginning of each calendar year. The able also repors he -saisics of he reurn spreads based on Newey-Wes (1987) sandard errors ha are adjused for boh heeroskedasiciy and serial correlaion in reurns. The resuls in Table III show ha here is a significan immediae marke reacion o earnings announcemens. The abnormal reurns of socks in he op SUE decile (D10) are significanly higher han hose in he boom SUE decile (D1) wih a spread of 4.312% during he wo-day announcemen window. More imporanly, he reurn spreads beween he op and boom deciles are posiive and highly significan over all horizons following earnings announcemens. The spreads are 0.714%, 1.160%, 2.092%, and 3.183% over one-week, woweek, one-monh, and one-quarer horizons, respecively. These spreads are clear evidence of marke underreacion o earnings surprises during our sample period. B. The Effec of Macroeconomic News Announcemens The main research quesion of our sudy is wheher macroeconomic news aggravaes or reduces PEAD. As noed in he inroducion, if macroeconomic news announcemens disrac invesor aenion away from earnings announcemens due o limied informaion processing power, his could aggravae invesor underreacion o earnings surprises. The main predicion under his hypohesis is ha here is a sronger correlaion beween earnings announcemen reurns and pos-earnings-announcemen sock reurns for earnings announcemens wih concurren macroeconomic news announcemens: where EA MAC r E r r E r I I E r E r, r E r E I (3) 1, 1 EA I denoes informaion conained in he earnings announcemen, EA MAC I denoes informaion conained in he macroeconomic news announcemen, and 0. As such, we 11

13 expec a sronger drif following earnings announcemens wih concurren macroeconomic news announcemens. On he oher hand, macroeconomic news may help invesors inerpre informaion conained in earnings news. The argumen can be formally presened in he model of Vuoleenaho (2002) who decomposes individual sock reurns ino a cash-flow componen and an expeced-reurn componen, namely, r r E 1 Ncf, N r,, (4) where Ncf denoes cash-flow news and Nr denoes expeced-reurn news. Under he framework of Campbell and Shiller (1988) and Campbell (1991) for aggregae sock reurns, Vuoleenaho (2000) derives an earnings-based model and decomposes he log book-o-marke raio ( ) as 1 1 j 0 j j j 0 j r e f, (5) j where r denoes he excess log sock reurn, e denoes he reurn on equiy which is defined as he log of one plus earnings (X) and he book equiy (B-1) raio, f denoes he log of one plus he ineres rae, (<1) is he discoun coefficien, and is a consan plus he approximaion error. Taking changes in expecaions from - 1 o ( E), he sock reurn can be decomposed ino a cash-flow componen and an expeced-reurn componen: r E 1 where E j j j r E e j f j E r j j 0, (6) j 0 1. The firs wo erms capure he effec of cash-flow news ( cf N, ) and he las erm capures he effec of expeced-reurn news ( N, ). Vuoleenaho (2002) shows ha expeced-reurn news is predominanly driven by he sysemaic macroeconomic componen. More imporanly, we noe ha under he model in Eq. (4), sock reurns are driven no only by shocks o expeced cash flows and discoun raes, bu also by he ineracion of hese wo r 12

14 componens. As shown by Savor and Wilson (2015), earnings announcemens conain informaion abou boh he prospec of issuing firms and expeced aggregae cash flows. The informaion in macroeconomic news helps invesors exrac hese wo componens and reac more raionally o earnings surprises. Similarly, Li, Richardson, and Tuna (2014) show ha informaion in macroeconomic news is also relevan in assessing fuure corporae earnings. Chordia and Shivakumar (2005) and Basu, Markov, and Shivakumar (2010) provide direc evidence ha PEAD is relaed o invesors underesimaing he impac of expeced inflaion on fuure earnings changes. Thus, he complemenary informaion in macroeconomic news abou updaed discoun raes and earnings announcemens abou firm cash flows as well as heir ineracions may help invesors value socks more efficienly and reduce underreacion o earnings surprises. Tha is, EA MAC r E r r E r I I E r E r, r E r E I (7) 1, 1 As such, we should expec a weaker drif following earnings announcemens wih concurren macroeconomic news announcemens. To es above hypoheses, each quarer we classify earnings announcemens ino wo subsamples: hose wih concurren macroeconomic news announcemens and hose wihou. An earnings announcemen is classified as having a concurren macroeconomic news announcemen if here is a leas one imporan macroeconomic news announcemen on he day of or he day afer he earnings announcemen. This classificaion is consisen wih he definiion of earnings announcemen window. Tha is, marke reacions o earnings announced afer marke close occur during he nex rading day. The lis of imporan macroeconomic news announcemens can be found in Secion II.B. During our sample period, roughly 40% of he days have macroeconomic news announcemens and earnings announcemens on abou 55% of he days are classified as EA 13

15 having concurren macroeconomic news announcemens. Based on he classificaion, socks in each decile of Table III are hen divided ino wo subsamples accordingly. Since he deciles are formed based on he SUE breakpoins of he previous quarer, his is equivalen o forming SUE decile porfolios separaely wihin each subsample of earnings announcemens. Table IV repors earnings surprises (SUE) and cumulaive abnormal reurns (CARs in percenage erm) of he op and boom SUE deciles and he reurn differenials beween he op and boom deciles as well as heir -saisics for earnings announcemens wih concurren macroeconomic news announcemens and hose wihou. The -saisics for he differences are based on Newey-Wes sandard errors. A he boom of he able, we also repor he differences in SUE spreads and drif beween he wo earnings announcemen subsamples. The resuls show ha differences in SUE spreads beween hese wo earnings announcemen subsamples are saisically insignifican, suggesing ha any differences in subsequen drif are likely driven by effecs addiional o earnings surprises. In our furher analysis, we find no evidence ha firms wih posiive or negaive earnings surprises ime earnings announcemens based on pre-scheduled macroeconomic news announcemen daes. Moreover, differences in immediae marke reacions beween hese wo earnings announcemen subsamples during he wo-day earnings announcemen window are saisically insignifican, suggesing ha any differences in subsequen drif are likely due o he compounding effec of macroeconomic news on invesor reacions o earnings surprises raher han invesor reacions o he macroeconomic news announcemens. Table IV shows ha for boh earnings announcemen subsamples, he spreads beween he op and boom deciles (D10 D1) are posiive and highly significan over all horizons following earnings announcemens. Tha is, here is a significan drif following earnings 14

16 announcemens, regardless of wheher here are concurren macroeconomic news announcemens. However, he drif following earnings announcemens wih concurren macroeconomic news announcemens is significanly weaker. The differences in drif beween wo earnings announcemen subsamples (-saisics in absolue value) are % (2.60), % (2.73), % (2.67), and % (1.44) over one-week, wo-week, one-monh, and one-quarer horizons, respecively. These differences are highly significan excep over he onequarer horizon. In paricular, macroeconomic news reduces he drif by up o 50% over he shor horizon. Furhermore, he resuls repored a he boom of he able show ha he reducion of drif for earnings announcemens wih concurren macroeconomic news is driven by boh posiive earnings surprises (D10) and negaive earnings surprises (D1). Tha is, macroeconomic news has a significan effec on invesor reacion o boh posiive and negaive earnings surprises. These resuls are consisen wih he conjecure ha macroeconomic news helps invesors inerpre earnings news, leading o more efficien pricing of individual socks. C. Mulivariae Tess: Conrolling for Oher Firm Characerisics In his secion, we perform mulivariae ess on he effec of macroeconomic news on invesor reacions by conrolling for oher firm characerisics. Specifically, we perform he following even-based Fama-MacBeh (1973) regression of cumulaive abnormal reurns (CARs) over differen horizons following earnings announcemens on SUE and is ineracion wih a macroeconomic news announcemen dummy as well as oher conrol variables: CAR i,[ 1, h] a SUE 5 0 LRET i, d [-11m,-6m] 1 MAC SUE i, d MAC RLAG Oher Conrols 6 2 LRET 3 [-1m,] i, LRET 4 [-5m,-1m] (8) 15

17 where CARi, [ 1, h] denoes cumulaive abnormal reurns over he horizon [ + 1, + h] for firm i wih an earnings announcemen on day and d MAC is a dummy variable ha is se equal o 1 if he earnings announcemen has concurren macroeconomic news announcemens and 0 oherwise. The main difference beween an even-based Fama-MacBeh regression and a convenional Fama-MacBeh regression is ha, in his seing, sock reurns and lagged variables are defined on even daes insead of calendar daes. LRET denoes lagged cumulaive sock reurns over various horizons. For example, LRET[ 5m, 1m] is he lagged cumulaive sock reurn over he pas five monhs. I is imporan o conrol for lagged reurns in our analysis since Aboody, Lehavy, and Trueman (2010) show ha socks wih he highes prior 12-monh reurns experience significanly negaive marke-adjused reurns immediaely following earnings announcemens. Following he lieraure (e.g., Grinbla and Moskowiz, 2004), we include pas reurns over differen horizons as conrol variables. We also include reporing lag (RLAG) in he regressions. Exising lieraure documens ha firms end o announce good news earlier han bad news. Oher conrol variables include SIZE, he book-o-marke raio (BM), he Amihud (2002) illiquidiy raio (ILLIQ), idiosyncraic volailiy (IVOL), and heir ineracions wih SUE in he regression. All firm characerisics are lagged by a leas one quarer. For deails on he definiions of hese variables, please refer o Secion II.A. As noed earlier, DellaVigna and Polle (2009) compare invesors responses o Friday earnings announcemens wih responses o announcemens on oher weekdays and find ha here is more drif for earnings announced on Fridays. Their argumen is ha invesors likely pay less aenion o announcemens on Fridays han announcemens on oher weekdays. We include a day of week dummy o conrol for poenial weekday effec in he regressions. We also replicae he analysis by excluding earnings 16

18 announced on Fridays. The resuls confirm ha he empirical findings in Table IV are robus o he Friday effec documened in DellaVigna and Polle (2009). Each quarer we perform he cross-secional regressions in Eq. (8). Since BM is included as a conrol variable in Eq. (8), we exclude financial firms in he regressions. Table V repors he average coefficien esimaes of he cross-secional regressions wih -saisics based on Newey- Wes sandard errors. The resuls show ha, consisen wih Aboody, Lehavy, and Trueman (2010), sock reurns following earnings announcemens generally have a negaive relaion wih lagged reurns. As expeced, marke reacions are negaively relaed o reporing lag. Consisen wih he soring resuls in Table IV, here is a significan immediae marke reacion o earnings surprises and macroeconomic news announcemens do no significanly affec he reacion in eiher direcion. More imporanly, he resuls show ha, for sock reurns over all horizons following earnings announcemens, he coefficien esimaes of he ineracion erm beween SUE and he macroeconomic news dummy are negaive and highly significan over shor horizons. The resuls confirm ha he empirical findings in Table IV are robus o conrolling for oher firm characerisics. D. The Effecs of Differen Types of Macroeconomic News The lieraure documens ha no all macroeconomic news has he same effec on sock reurns because he informaion conen varies among differen ypes of news. For insance, Flannery and Proopapadakis (2002) examine he effec of macroeconomic news on sock marke reurns and marke reurn volailiy. They find ha six macroeconomic news announcemens (hree nominal and hree real) are pricing facors. Bernanke and Kuner (2005) show ha he effec of unanicipaed moneary policy acions on sock prices is mainly driven by changes of 17

19 expeced excess reurns. Boyd, Hu, and Jagannahan (2005) show ha unemploymen news conains informaion on all hree primiive pricing facors: he fuure ineres rae, he expeced growh rae of corporae earnings and dividends, and he equiy risk premium. They also documen ha he effec of he changes of unemploymen raes on sock prices depends on he economic cycle. In his secion, we sudy he effec of differen ypes of macroeconomic news on PEAD. We classify he macroeconomic news iems in our lis ino hree caegories: (1) news relaed o ineres rae expecaions or expeced discoun raes, including he FOMC rae decision, he Consumer Price Index, iniial jobless claims, changes in nonfarm payrolls, and he unemploymen rae; (2) news relaed o real aciviies, including GDP growh, he ISM Manufacuring Index, durable goods orders, he consumer confidence index, he Universiy of Michigan Consumer Senimen Index, and reail sales; and (3) news relaed o he housing marke, including new home sales and housing sars. We include news on labor marke condiions in he same caegory as he FOMC rae decision because employmen is one aspec of he dual mandae (he oher is inflaion) of he Fed moneary policy and conains informaion abou he fuure ineres rae (Boyd, Hu, and Jagannahan, 2005). We confirm ha he resuls are robus when we furher classify news on labor marke condiions in he second caegory. Flannery and Proopapadakis (2002) classify housing marke news as real economic aciviies in heir analysis bu noe ha he marke paricularly waches he unemploymen and housing repors. We place housing news in a separae caegory since he housing marke is one of he mos observed news announcemens during he financial crisis in our sample period. We perform he even-based Fama-MacBeh regressions of cumulaive abnormal reurns (CARs) following earnings announcemens on SUE and is ineracion wih dummies of hree differen ypes of macroeconomic news announcemens as well as oher conrol variables: 18

20 CAR i,[ 1, h] a SUE 5 0 LRET i, [-1m,] d Oher Conrols 1 6 DR i, SUE LRET i, d [-5m,-1m] 2 RA SUE 7 i, LRET d 3 HM [-11m,-6m] SUE 8 i, d 4 RLAG MAC (9) where d DR = 1, d RA = 1, and d HM = 1 if he earnings announcemen has a concurren macroeconomic news announcemen in he firs, second, or hird caegory, respecively. All oher variables are he same as in Eq. (8). Each quarer, we perform he cross-secional regressions as specified in Eq. (9). Table VI repors he average coefficien esimaes of he cross-secional regressions wih -saisics based on Newey-Wes sandard errors. The coefficien esimaes of he conrol variables and heir saisical significance are similar o hose in Table V. The resuls in Table VI show ha, he coefficien esimaes of he ineracion erms beween SUE and he dummy variables for differen ypes of macroeconomic news are negaive over all horizons. This is evidence ha macroeconomic news in all caegories has a significan effec on invesor reacions o earnings surprises. Judging by he significance level, as measured by he -saisics of he coefficien esimaes, news relaed o he discoun rae has he sronges effec on invesor reacion over shor horizons, followed by news relaed o real aciviies, which is furher followed by news relaed o he housing marke. Tha is, among differen ypes of news, macroeconomic news relaed o he discoun rae seems o have an immediae effec on invesor reacions o earnings surprises. This finding corroboraes hose in Chordia and Shivakumar (2005) and Basu, Markov, and Shivakumar (2010) ha PEAD is relaed o invesor underesimaion of he impac of expeced inflaion on fuure earnings changes. Ye, our resuls show ha he macroeconomic news effec on PEAD is pervasive and goes beyond inflaion-relaed news. 19

21 E. Earnings Surprises Based on Analys Forecass In our main empirical analysis, we compue earnings surprises (SUE) based on he seasonal random walk model in Eqs. (1) and (2) from hisorical earnings. The lieraure has also examined PEAD using earnings surprises based on analys forecass (e.g., Mendenhall, 2004; Livna and Mendenhall, 2006; Francis, Lafond, Olsson, and Schipper, 2007). In his secion, we follow Livna and Mendenhall (2006) and compue SUE as follows: SUE i, ~ ( X X i, i, (10) P i, ) where Xi, is primary earnings per share before exraordinary iems for firm i in quarer and Pi, is he price per share for firm i a he end of quarer from Compusa. Boh Xi, and Pi, are unadjused for sock splis and ~ is he median of forecass repored o IBES in he 90 days X i, prior o he earnings announcemen. As in Secion III.A, each quarer we divide earnings announcemens ino deciles based on he ranks of SUE and examine pos-earningsannouncemen sock reurns. Similar o Secion III.B, we hen classify earnings announcemens ino wo subsamples: hose wih concurren macroeconomic news announcemens and hose wihou. Again, we are ineresed in he differences in sock reurn drif beween he wo subsamples of earnings announcemens. The resuls repored in Table VII show ha, for boh subsamples of earnings announcemens, here is significan PEAD. Consisen wih Livna and Mendenhall (2006), we find ha PEAD based on earnings surprises compued from analys forecass is sronger han ha based on earnings surprises compued from he seasonal random walk model. The drif in boh subsamples of earnings announcemens, as shown in Table VII, is larger han heir respecive cases in Table IV. In addiion, we noe ha he resuls in Table VII are based on he sample of 20

22 socks wih analys forecass of nex quarer s earnings. As shown in our furher analysis in nex secion, he effec of macroeconomic news is relaively weaker for firms wih high analys coverage han for firms wih low analys coverage. Neverheless, differences beween drif following earnings announcemens wih macroeconomic news announcemens and drif following earnings announcemens wihou macroeconomic news announcemens are negaive and saisically significan over mos horizons. The resuls confirm ha our main findings are robus when earnings surprises are compued using analys forecass. IV. Furher Analysis Our empirical resuls sugges ha macroeconomic news announcemen helps reduce invesor misreacion o earnings surprises. In his secion, we perform addiional analyses o undersand exacly wha mechanism drives he effec of macroeconomic news announcemens. Firs, we es he implicaions of caegory-learning behavior on he allocaion of invesor aenion. Second, we examine wheher he informaion conen of macroeconomic news has a sronger effec on firms wih greaer informaion uncerainy. Finally, we invesigae he possibiliy ha managemens may acically ime earnings announcemens based on prescheduled imporan macroeconomic news announcemens. A. Invesor Aenion o Macroeconomic News and Earnings Announcemens The caegory-learning behavior of Peng and Xiong (2006) predics ha invesors generally pay more aenion o macroeconomic news announcemens and, as a resul, allocae relaively less aenion o firm-level news. This predicion has direc implicaions on he inerpreaion of our findings. If invesors pay relaively less aenion o earnings announcemens 21

23 on days wih macroeconomic news announcemens, hen i is more likely ha he informaion conen in he macroeconomic news helps reduce misreacion o earnings surprises. On he oher hand, since he announcemen of macroeconomic news draws invesor aenion o he overall marke, i is possible ha invesors pay more aenion o earnings announcemens on hose days as well. The increased invesor aenion o earnings announcemens on days wih macroeconomic news announcemens could help reduce invesor underreacion o earnings surprises. In his secion, we empirically es he implicaions of caegory-learning behavior. Firs, we examine wheher invesors pay more aenion o he overall marke on days wih macroeconomic news announcemens han on days wihou. We use wo variables o measure invesor aenion, namely excess marke rading volume and absolue marke reurns. As poined ou in he survey by Bamber, Barron, and Sevens (2011), boh rading volume and absolue price changes have been used in he lieraure o measure marke responses o informaional evens. Following he lieraure, we define excess rading volume for day as ETV ln( TV / ATV[ 21, 3] ) where TV is he dollar rading volume on day and ATV [ 21, 3] denoes he average dollar rading volume over [ - 21, - 3] or he pas monh. We calculae he daily marke rading volume as he oal dollar volume of all CRSP common socks. We calculae absolue marke daily reurns using he CRSP value-weighed and equal-weighed indexes. Table VIII repors he average daily marke excess rading volume and average absolue daily marke reurns during days wih and wihou macroeconomic news announcemens. For macroeconomic news announcemen days, he resuls for each news iem are also repored. In addiion, he able repors he differences in excess rading volume and absolue reurns beween days wih and wihou macroeconomic news announcemens, as well as heir -saisics. 22

24 The resuls in Table VIII show ha he difference in excess marke rading volume beween days wih and wihou macroeconomic news announcemens is and highly significan. For mos individual news iems, he average excess marke daily rading volume on announcemen days is also significanly higher han on days wih no macroeconomic news announcemens. The resuls in Table VIII also show ha average absolue daily reurns for boh CRSP value-weighed and equal-weighed indexes are higher on days wih macroeconomic news announcemens han on days wihou, alhough he differences are saisically insignifican. These findings show ha invesors pay more aenion o he overall marke on days wih macroeconomic news announcemens han on days wihou. In paricular, he significanly higher rading volume on days wih macroeconomic news announcemens suggess ha invesors updae heir valuaion of socks and adjus heir porfolios following he arrival of macroeconomic news. Second, we examine wheher invesors allocae relaively less aenion o earnings announcemens on days wih macroeconomic news announcemens han on days wihou. Again, we use wo variables o measure invesor aenion, namely, excess rading volume and absolue sock reurns. Since we examine average rading aciviies across individual socks, we use urnover insead of he dollar rading volume. Consisen wih he lieraure (e.g., Bamber, 1987; Ajinkya and Jain, 1989), we compue average urnover over he hree-day announcemen window [ - 1, + 1]. Specifically, for a sock wih an earnings announcemen on day, we compue excess urnover as ETO ln( TO / ATO[ 21, 3] ) where TO is he average daily urnover over he earnings announcemen window [ - 1, + 1] and ATO [ 21, 3] denoes he average daily urnover over [ - 21, - 3] or he pas monh. Noe ha, as shown in Table VIII, he excess rading volume on days wih macroeconomic news announcemens is higher han on 23

25 days wihou. Since our focus is he allocaion of aenion o firms wih earnings announcemens relaive o oher firms, we adjus he macroeconomic news announcemen effec when comparing rading aciviies on days wih macroeconomic news announcemens o hose on days wihou. Specifically, we firs calculae he average daily urnover of all socks on days wih and wihou macroeconomic news announcemens separaely during each quarer. We hen use he raio of hese wo averages as he adjusmen facor for he effec of macroeconomic news announcemens. As a second measure of invesor aenion, we calculae he absolue daily reurns ( RET ) as he average over he earnings announcemen window [ - 1, +1] for a sock wih an earnings announcemen on day. Table IX repors he average excess urnover and average absolue daily reurns of socks in each SUE decile on days wih and wihou macroeconomic news announcemens. The able also repors he averages of hese variables for all socks. The differences in average excess urnover and absolue daily reurns beween days wih and wihou macroeconomic news announcemens are repored in he righ panel. The resuls show ha he average absolue daily reurns and excess urnover over he earnings announcemen window are significanly lower on days wih macroeconomic news announcemens han on days wihou. The paern also holds for almos all SUE deciles. The findings in Tables VIII and IX show ha invesors pay more aenion o he overall marke on days wih macroeconomic news announcemens and allocae relaively less aenion o firms wih earnings announcemens on days wih macroeconomic news announcemens. These findings are consisen wih he caegory-learning behavior and sugges ha i is more likely he informaion conen in macroeconomic news raher han increased invesor aenion o earnings 24

26 announcemens ha helps reduce misreacion o earnings surprises on macroeconomic news announcemen days. B. Informaion Uncerainy and he Effec of Macroeconomic News The lieraure documens ha invesors exhibi sronger behavioral biases when here is higher informaion uncerainy. Using several differen proxies for informaion uncerainy, Jiang, Lee, and Zhang (2005) show ha earnings momenum effecs are much sronger among firms wih high informaion uncerainy. Zhang (2006) ess he impac of informaion uncerainy on invesor underreacion o news and finds ha underreacion is sronger for firms wih greaer informaion uncerainy. Similarly, Francis, Lafond, Olsson, and Schipper (2007) find evidence ha invesors have more mued iniial reacions o unexpeced earnings signals of greaer informaion uncerainy. Kumar (2009) provides furher evidence ha individual invesors make larger invesmen misakes and exhibi sronger behavioral biases when socks are more difficul o value. If macroeconomic news helps inerpre earnings news hrough he resoluion of informaion uncerainy, i is reasonable o expec ha he effec is sronger for firms wih greaer informaion uncerainy. To es he hypohesis, we use wo proxies for informaion uncerainy in our analysis, namely idiosyncraic volailiy (IVOL) of sock reurns and analys coverage (COV). Boh measures have been used in he lieraure as proxies for informaion uncerainy. Each quarer, we divide socks ino hree subsamples based on IVOL or COV. We focus on he subsamples of socks above he 60h percenile (op 40%) or below he 40h percenile (boom 40%) of IVOL or COV. For boh sock subsamples, we replicae he analysis in Table IV. 25

27 Table X repors he resuls for he op 40% high IVOL firms (Panel A) and he boom 40% low IVOL firms (Panel B). Boh panels repor he differences beween he drif following earnings announcemens wih concurren macroeconomic news and he drif following hose wihou. For robusness, we require a leas 20 socks in boh he long and shor porfolios each quarer. Since PEAD is sronger for he subsample of high IVOL socks han for he subsample of low IVOL socks, for comparison purposes we also repor relaive reducions in he drif due o he effec of macroeconomic news. The resuls in Table X show ha, while here is a reducion in PEAD due o he macroeconomic news effec for boh he op 40% high IVOL and boom 40% low IVOL firms, he effec is more pronounced for he op 40% high IVOL firms based on he magniude, saisical significance and he relaive reducion. The resuls in Panel A show ha for he op 40% high IVOL firms, he differences in drif beween earnings announcemens wih concurren macroeconomic news announcemens and hose wihou are significanly negaive over all horizons. The relaive reducion in he drif due o he effec of macroeconomic news is more han 65% over shor horizons. The resuls in Panel B show ha, for he boom 40% low IVOL firms, he differences in drif beween earnings announcemens wih concurren macroeconomic news announcemens and hose wihou are only significan over he very shor horizon and insignifican over longer horizons. The relaive reducion in drif due o he effec of macroeconomic news is also much lower. Table XI repors he resuls based on analys coverage (COV). The resuls are consisen wih hose repored in Table X. For he sample of firms in he op 40% of analys coverage, macroeconomic news announcemens have a significan effec on invesor reacion o earnings surprises only over he very shor horizon and insignifican over longer horizons. On he oher hand, for he sample of firms in he boom 40% of analys coverage, macroeconomic news has a 26

28 significan effec on invesor reacion o earnings surprises over all horizons. The relaive reducion in drif is more han 60% over he shor horizon. The evidence in Tables X and XI suppors our conjecure ha he effec of macroeconomic news on PEAD is sronger for firms of higher idiosyncraic volailiy or lower analys coverage. The evidence furher suppors he conjecure ha he weaker PEAD on macroeconomic news announcemen days is likely due o he informaion conen in macroeconomic news. The updaed informaion on macroeconomic fundamenals helps resolve informaion uncerainy associaed wih earnings surprises. C. Do Managemens Time Earnings Announcemens Based on Pre-scheduled Macroeconomic News Announcemens? Given he fac ha almos all imporan macroeconomic news announcemens are prescheduled, i is naural o speculae ha managemens may have incenives o ime earnings announcemens based on macroeconomic news announcemens. Exising lieraure has documened evidence ha managemens end o schedule earnings announcemen wih negaive surprises during days or hours wih less invesor aenion. For insance, dehaan, Shevlin, and Thornock (2015) find evidence consisen wih managers reporing bad news afer marke hours, on Fridays, on busy days, and wih less advance noice, and wih earnings receiving less aenion in hese seings. Moreover, heir findings suppor he conjecure ha managers hide bad earnings news by announcing during periods of low marke aenion, and conversely, managers highligh good earnings news by announcing earnings during periods of high marke aenion. To examine wheher managemens ime earnings announcemens on macroeconomic news announcemen days, we perform wo ses of empirical ess. Firs, for each earnings 27

29 announcemen, we calculae he relaive change of reporing lag as he difference beween curren quarer s reporing lag and ha of four quarers ago divided by curren quarer s reporing lag. In each SUE decile, we divide earnings announcemens as hose wih concurren macroeconomic news and hose wihou. For each group, we repor he average absolue value of he relaive change of reporing lag. If managemen inenionally ime earnings announcemens on earnings announcemen days o avoid invesor aenion, we should see more variaion in change of reporing lag for earnings announcemens wih concurren macroeconomic news. Second, in each SUE decile, again we divide earnings announcemens as hose wih concurren macroeconomic news and hose wihou. For each group, we repor he average SUE. If firms wih negaive earnings surprises inenionally ime earnings announcemens on macroeconomic news announcemen days o hide bad news, we should expec lower average of SUE in D1 for earnings announcemens wih concurren macroeconomic news announcemens. If firms wih posiive earnings surprises inenionally ime earnings announcemens on non-macroeconomic news announcemen days o highligh good news, we should expec higher average of SUE in D10 for earnings announcemens wihou concurren macroeconomic news announcemens. Table XII repors average absolue relaive change in reporing lag ( RL ) and average SUE for he whole sock sample and socks in each SUE decile. The resuls are repored separaely for earnings announcemens wih concurren macroeconomic news announcemens and hose wihou. The able also repors he differences in RL and SUE beween earnings announcemens wih concurren macroeconomic news announcemens and hose wihou and heir Newey-Wes -saisics. The resuls show ha he variaion in reporing lag on average is acually lower for earnings announcemens on days wih macroeconomic news announcemens. This is evidence agains he conjecure ha managemens inenionally schedule earnings 28