Differences in the Price-Earning-Return Relationship between Internet and Traditional Firms

Transcription

1 Dfferences n he Prce-Earnng-Reurn Relaonshp beween Inerne and Tradonal Frms Jaehan Koh Ph.D. Program College of Busness Admnsraon Unversy of Texas-Pan Amercan jhkoh@upa.edu Bn Wang Asssan Professor Compuer Informaon Sysems and Quanave Mehods College of Busness Admnsraon Unversy of Texas-Pan Amercan bnwang@upa.edu Phone: (956) Fax: (956) ABSTRACT Ths paper examnes he effecs of he begnnng sock prce and he curren earnngs on he sock reurns and analyzes he dfferences n he prce-earnng-reurn (P-E-R) relaonshp beween nerne frms and radonal. Our resuls show ha he sock reurns are more affeced by he prevous msprcng han by he curren performance and ha he prevous msprcng has more effec on he nerne sock reurns han on he radonal sock reurns and sugges ha nerne socks earn hgher reurns despe of less earnngs han radonal ones because he nerne frms are under-valued n he begnnng prce compared wh he radonal ones. 1. INTRODUCTION Before nerne bubbles burs n he sprng 2000 many new valuaon measures for nerne frms (IF) had popped up o jusfy he hgh prces of nerne shares. Accordng o Trueman Wong and Zhang (2000) (TWZ) he fundamenal reason why dfferen mehods came no exs o value IF s ha no hsorcal fnancal nformaon was useful o forecas fuure performance of hese frms because he ndusry was so young and growng so fas. Now more han 7 years have passed snce nerne bubbles burs and mos bubbles seem o dsappear. In fac accordng o TWZ as of November Yahoo! had a P/E of 1382 ebay a P/E of 3351 and Amazon.com raded a a mulple o sales revenue of 22.9 ( had been unprofable snce ncepon). However as of he end of December 2007 Yahoo! has a P/E of 46.9 ebay a P/E of and Amazon.com rades a a mulple o sales revenue of 2.6 ( s profable now and s P/E s 81). Hence may be beer han ever o compare IF wh radonal frms (TF) by he same model. 640

2 In hs research we nvesgae by usng pos-nerne-bubble daa wheher any sgnfcan dfferences exs n he prce-earnng-reurn (P-E-R) relaonshp beween IF and TF. Our resuls show ha he sock reurns are more affeced by he prevous msprcng han by he curren performance and ha he prevous msprcng has more effec on he nerne sock reurns han on he radonal sock reurns and sugges ha nerne socks earn hgher reurns despe of less earnngs han radonal ones because he nerne frms are under-valued n he begnnng prce compared wh he radonal ones. Ths paper conrbues o he exsng leraure n he followng dsnc ways. Frs few researchers have suded IF by usng daa afer 2002 he year nerne bubbles seemed o have gone. Second we analyze resduals from he regressons o capure he dealed dfference n he P-E-R relaons beween wo ypes of frms and nroduce he expeced reurn lne (ERL) o draw he P-E-R relaonshp. The remander of hs paper s organzed as follows; he nex secon brefly revews leraures and he hrd secon explans he model and mehodology. The fourh provdes sample selecon crera and daa he ffh repors he resuls of he emprcal ess and he fnal secon concludes he paper. 2. LITERATURE REVIEW Mos papers examnng he sock reurns of IF usng daa before 2002 do no fnd any sgnfcan P-E-R relaonshp snce hese researches manly amed a denfyng facors oher han fnancal nformaon o explan IF valuaon. A ha me researchers could no explan he P-E-R relaonshp of IF usng radonal valuaon mehods whch rely on fnancal daa. Furhermore few researches have examned he P-E-R of IF usng daa afer 2002 he year when he Inerne bubble seemed o have dsappeared. In general academcans and praconers have found he P-E-R relaons n seekng ably o predc sock reurns; n general he rao of book value o marke value (BM) descrbes msprcng n he begnnng prce and eher earnngs yeld (= earnngs-prce rao = EP = EPS/Prce EPS = earnngs per share) or PE mulple (= Prce/EPS) explans he relaonshp beween earnngs and reurns. Jaffe Kem and Weserfeld (1989) analyze he relaon beween sock reurns and he effecs of sze and EP rao and fnd he EP effec o be sgnfcan. Fama and French (1992) argue ha wo varables sze measured by he marke value (MV) and he BM rao capure much of he average sock reurns and also conclude ha he EP s sgnfcan when s he unque explanng varable for he sock reurns bu s sgnfcance dsappears when he BM rao s also aken no accoun. Ponff and Schall (1998) fnd ha he BM rao provdes some predcve ably whch sems from he relaon beween BV and fuure earnngs. Kohar and Shanken (1997) fnd relable evdence ha he BM rao rack me-seres varaon n expeced real sock reurns for he US sock marke whle Lamon (1998) argues ha he PE has ndependen predcve power for excess reurns n addon o he dvdend-prce rao. Bae and Km (1998) usng a sample of Japanese frms show ha compared wh he radng sraegy based on EP or BM alone he radng sraegy based on he combnaon of boh EP and BM generaes subsanally hgher reurns mplyng ha BM (or EP) capure ceran aspecs of equy values ha are no capured by EP (BM). The regresson resuls furher ndcae 641

3 ha he predcve ably of EP s domnaed by ha of BM conssen wh he US resuls of Fama and French (1992) P-E-R Relaonshp R 3. MODELS AND METHODOLOGY Reurns are calculaed by he followng formula. D P P 1 (1) P 1 Where R s he reurn on sock a year D s he dvdends per share (DPS) for sock a year P and 1 P s he prce of sock a year and -1. In he mos fnance leraures dvdends and prce are consdered as he funcon of earnngs. D d E P p E Where E s he earnngs per share (EPS) for sock a year d s he dvdend payou rao (DE) = D/E p s he prce-earnng mulple (PE) = P/E. Hence D P ( d p ) E (2) Subsung (2) no (1) yelds ( d p ) E P 1 R hen P 1 E R ( d p ) 1. (3) P 1 The equaon (3) shows he P-E-R relaonshp ha sock reurns are he funcon of he begnnng prce and he curren EPS f d and p are assumed o be consan. In oher words here are wo sources of he unexpeced reurns: msprcng n he prevous year and unancpaed EPS n he curren year e.g. here would be posve (negave) unexpeced reurns f he sock s under (over)-valued n he prevous year or f he frm s over (under)- performng n he curren year. In hs paper he comparson beween IF and TF s relave no absolue; for example under-valued (performed reurned) nerne socks means over-valued (performed reurned) radonal ones and vce versa. Because hey are compared each oher one ype should be beer han he oher and vce versa unless hey are equal Models To measure he P-E-R for each frm we use he followng regresson models because PE mulple prof margn (PM = EPS/SPS SPS = sales per share) and are wdely used. 642

4 P a be v (4) E m ns e (5) 1 R M r Where P 1s he prce of sock a year -1 (begnnng year) E and E 1 s he EPS of frm a year and -1 S s he SPS of frm a year R s he reurn on sock a year M s he marke reurn a year a b m n are consan v 1 e r are he error erms. The expeced values for P-E-R are he fed values from he regressons and error erms v 1 e and r are he unexpeced poron of P-E-R for each sock. The sock wh posve (negave) error erms s regarded as havng hgher (lower) value han expeced.e. over (under)-valued performng or reurned. To examne he P-E-R relaonshp and analyze he dfference n hs relaonshp beween wo ypes of frms usng he resduals from he models (4) (5) and (6) we regress v 1 and e agans r. r v e (7) 1 Where r s he unexpeced (poron of) reurns on sock a year v 1s he wrong-valued (poron of) prce n he prevous year (-1) e s he unancpaed (poron of) EPS a year are consan. In addon o regressons we perform he ndependen -es o fnd wheher any sgnfcan mean dfference exss n he mporan varables such as P 1 E R v 1 e and r beween IF and TF. (6) 3-3. Expeced Reurn Lne We analyze he resduals v 1 e and r by he graph represenng he P-E-R relaonshp. In [Fgure 1] he expeced reurn lne (ERL) represens he frms whose sock reurns are same as he expeced reurns from he model (6) ( r = 0). The slope of he ERL depends on he relave degree of effec beween v 1 and e on r. The lower (hgher) he slope s he sronger s he effec of v 1 ( e ). We can calculae he slope of he ERL n he followng way. From he graph n [Fgure 1] he ERL can be descrbed by v 1 e (8) Then 643

5 The more negave r ( r <<0) v 1 ERL ( r =0) C B O e The more posve r A ( r >>0) v 1= Wrong-valued prce a -1 e = Unancpaed EPS a r = Unexpeced reurns a ERL = Expeced Reurn Lne Lef-upper sde of he ERL shows he negave r (.e. lower reurn han expeced) and rgh-under sde of he ERL means he posve r (.e. hgher reurn han expeced). The farher away from he orgn o he NW (SE) means he more negave (posve) r.e. he lower (hgher) reurns han expeced. A he orgn (O) v 1= 0 and e = 0 and hence r = 0. Pon A s on he ERL whch means r = 0 by cancellng ou he posve effec of under-valued prce by he negave effec of under-performed EPS. r < 0 (lower reurns han expeced) by greaer negave effec of Pon B s over he ERL whch means under-performed EPS han he posve effec of under-valued prce. Pon C s under he ERL whch means r > 0 (hgher reurns han expeced) by greaer posve effec of over-performed EPS han he negave effec of over-valued prce. [Fgure 1] Prce-Earnng-Reurn relaonshp and expeced reurn lne v 1. (9) e From he equaon (7) r v e (7) 1 r r and (10) v 1 e From (9) and (10) where means absolue value. (11) should be posve wh <0 and >0 f v 1 and r are negavely relaed and e and r are posvely relaed. 644

6 Subsung (11) no (8) yelds v 1 e ( 0) e 0( 0) Ths (12) represens he ERL. (12) 4. DATA We use he frms ncluded n he Nasdaq Inerne Index (92 frms) for he IF and he frms n he S&P 100 Index (excludng Google) for he TF. The followng frms are excluded from he sample; 1) Fnancal frms. 2) Frms whch have dfferen fscal year from calendar year. 3) Frms whch do no have full daa from 2003 o In fac we perform hs research over 4 years from 2004 o 2007 bu we need 2003 daa for he prevous year prce and EPS n he model (4) and (7). 4) Frms whch show sgnfcanly dfferen behavors from he ohers 2. The fnal number of frms n hs research s 38 for he IF 61 for he TF and hence 99 for oal frms resulng n oal 396 observaons 99 frms*4 years = 396 observaons. Daa are colleced from CompuSa and CRSP (for reurn). Sock reurns are annual reurn calculaed by compounded monhly reurns from CRSP excep 2007 reurns whch s he frs 9 monh (Jan. o Sep.) reurns from CRSP mulpled by he las quarer (Oc. o Dec.) reurns compued by he sock prce and dvdends from COMPUSTAT. We use he Russell 3000 ndex for he marke reurn of boh IF and TF because boh ypes of frms should be compared by he same sandard and he Russell 3000 ndex racks almos 99 percen of he socks ncluded n porfolos of nsuonal nvesors. 5. RESULTS Before runnng he regresson Panel B n [Table 1] shows ha TF on average have he hgher begnnng prce hgher EPS n boh prevous (-1) and curren () year hgher SPS n he curren year bu lower reurn n he curren year and hese dfferences are all sascally sgnfcan a 1% sgnfcance level excep reurns sgnfcan a 5% sgnfcance level. Why do nerne socks earn hgher reurns despe of lower EPS and SPS han TF? [Table 2] shows he regresson resuls for he models (4) (5) and (6) and [Table 3] examnes he resduals from he [Table 2]. Panel A n [Table 3] ells ha sgnfcan dfferences n he P-E-R exs beween IF and TF. The msprcng n he prevous year v 1 he unexpeced performance e and unancpaed reurns r ell ha on average IF are less valued n he prevous year by $13.16 worse performed n he curren year by $.85 n EPS and more earned by 12.7% n annual reurn han TF. Therefore Panel A n [Table 3] explans ha he resuls of Panel B n [Table 1] have he reason. Panel B n [Table 3] suggess ha msprcng n he prevous year s negavely and curren EPS s posvely relaed o he curren sock reurns. In oher words over (under)-valued socks n he prevous year have lower (hgher) reurns n he curren year and over (under)- 645

7 Panel A: Descrpve sascs Varables N Mnmum Maxmum Mean Sd. Devaon Pre-P Pre-E All E S R M IF TF Pre-P Pre-E E S R M Pre-P Pre-E E S R M Panel B: Mean comparson (from he panel A) Varables All IF TF -es for mean equaly Pre-P Pre-E E S R (.026) M (1.000) p-values n ( ) Pre-=Prevous year P=Prce E=EPS S=SPS R=Reurn and M=Marke reurn [Table 1] Descrpve sascs and mean comparson performng socks n he curren year earn more (less) reurns and hs s conssen wh wha we expec from he equaon (3). However here s dfference beween IF and TF n he P-E-R relaons. From he second and hrd columns of he panel B n [Table 3] only he msprcng n he prevous year has a sgnfcan effec on he reurns for he IF bu boh msprcng n he prevous year and he curren performance have a sgnfcan effec for he TF. Whch has more mpac on he unexpeced reurns beween prevous msprcng and unancpaed performance? From he panel B n [Table 3] s greaer han whch may mslead o conclude ha he unexpeced reurns r are more affeced by he unancpaed performance e han he msprcng n he prevous year v 1. However v 1 s much greaer 646

8 Model (4): P 1 a be 1 v Model (5): 1 E m ns e Model (6): R M r Regresson Resduals v ) ( 1 a b Regresson m n.464 (.005).053 Regresson.082 (.088) (.006) Adj R Adj R Adj R Mean.000 Mean.000 Mean.000 Mn Resduals Mn Resduals Mn Max ( e ) Max ( r ) Max S. Dev S. Dev S. Dev..452 [Table 2] Regressons Panel A: Group Sascs and -es for mean equaly Sd. IF N Mean Devaon Sd. Error Mean v e r IF =1: Inerne Frms IF =0: Tradonal Frms -es for mean equaly (.025) Panel B: Regressons: Model (7) r v 1 e All IF TF (1.000) (.647) (.001) (.012) (.025) (.086) (.619) (.001) Adj R p-values n ( ) [Table 3] Analyses of resduals han e (15 mes greaer on average from Panel A n [Table 3]) causng v 1 o be greaer han e and hence v 1 has greaer effec on r han e does. Ths resul suppors he Fama- French (1992) and Bae and Km (1998). As seen n [Fgure 2] for he IF mos frms are under-valued and he number of underperformed frms s slghly greaer han ha of over-performed ones leadng mos IF o he posve area of unexpeced reurns (under he ERL). On he oher hand for he TF overperformed frms ounumber under-performed ones by more han double and over-valued frms 647

9 Msvalued prce a -1 Msvalued prce a -1 P-E-R Relaonshp n he Inerne frms Unancpaed EPS a 0 10 P-E-R Relaonshp n he Tradonal frms Unancpaed EPS a 0 10 Slope of he ERL = / =3.4 from he frs column of Panel B n [Table 3] (See he equaon (12)) [Fgure 2] Dfference n P-E-R Relaonshp beween IF and TF are a few more han under-valued ones bu hey seem o be almos evenly dvded by he ERL because he reurns are more nfluenced by he over-valuaon han he over-performance. 6. CONCLUSION We have examned he P-E-R relaonshp of he frms and analyzed he dfferences n hs relaonshp beween he IF and he TF analyzng resduals from he regressons. From he resuls we fnd ha he sock reurns are more affeced by he prevous msprcng han by he 648

10 curren performance supporng he Fama-French (1992) and Bae and Km (1998) and ha he prevous msprcng has more effec on he nerne sock reurns han on he radonal sock reurns and conclude ha nerne socks earn hgher reurns despe of less earnngs han radonal ones because he nerne frms are under-valued n he begnnng prce compared wh he radonal ones. NOTES 1. The model (6) R M r can be derved from he CAPM (Capal Asse Prcng Model) n he followng way. The CAPM says R RF ( M RF) (1 ) RF M. If we use as an average of all hen R (1 ) RF M r makng he model (6) f (1 )RF s regarded as consan. 2. Afer collecng all daa we exclude 2 frms from he sample because hey show sgnfcanly dfferen behavors from he ohers: Cogen Communcaons Group Inc. (CCOI) and General Moors (GM). REFERENCES Bae K. Hong Jeong-Bon Km(1998) The Usefulness of Earnngs versus Book Value for Predcng Sock Reurns and Cross Corporae Ownershp n Japan Japan and he World Economy 10 pp Fama E.F. K.R. French (1992) The Cross-Secon of Expeced Sock Reurns Journal of Fnance 47 pp Jaffe J. D. B. Kem R. Weserfeld (1989) Earnngs Yelds Marke Values and Sock Reurns The Journal of Fnance 44 pp Kohar S.P. Jay A. Shanken (1997) Book-o-Marke Dvdend Yeld and Expeced Marke Reurns: A Tme- Seres Analyss Journal of Fnancal Economcs 44 pp Lamon O. (1998) Earnngs and Expeced Reurns Journal of Fnance 53 pp Ponff Jeffrey Lawrence D. Schall (1998) Book-o-Marke Raos as Predcors of Marke Reurns Journal of Fnancal Economcs 49 pp Trueman B. M.H. F. Wong and X.Zhang (2000) "The Eyeballs Have I: Searchng for The Value n Inerne Socks" Workng paper Haas School of Busness Unversy of Calforna Berkeley. 649