Stockholder Wealth Implications of the Firm s Choice Between Dividends. and Stock Repurchases

Transcription

1 Stockholder Wealth Implcatons of the Frm s Choce Between Dvdends and Stock Repurchases by Noel R.Reynolds for The Unversty of the West Indes, St. Augustne Campus Inaugural Internatonal Conference on Busness, Bankng & Fnance THIS IS A WORKING DRAFT PLEASE DO NOT QUOTE Draft: Aprl 2004

2 An Abstract: Corporate dsbursements have a sgnfcant mpact on the stock markets and, hence, stockholders wealth. Exstng emprcal studes have revealed sgnfcant stock prce reactons to announcements of unexpected corporate cash dstrbutons (that s, dvdends and stock repurchases). Notwthstandng the hgher observed wealth effect assocated wth stock repurchase announcements, emprcal evdence shows an unexplaned preference by frms for usng cash dvdends. Drawng on data gathered from frms tradng on the US stock markets (NYSE, AMEX, and NASDAQ) between 1984 and 1995, ths research emprcally addresses the queston arsng from the above observaton: snce stock repurchases create a greater value change n stockholders wealth, are managers maxmzng stockholders wealth when they use other forms of cash dstrbutons? Applyng a lmted dependent varable methodology known as self-selectvty, I examne the determnants of the choce between ncreasng dvdends and utlzng an open market stock repurchase and the mpact of that choce on the stockholders wealth poston. The expectaton s that even n the presence of asymmetrc nformaton, agency costs, and dfferng expected stock prce reactons to the varous mechansms of cash dsbursements, frms, on average, choose the cash dstrbuton method that maxmzes the expected gan assocated wth the dstrbuton. The fndngs ndcate that open market repurchasng frms make optmal dsbursement choces that s reflected n the reacton of the stock market to the dsbursement announcement. However, smlar results were nconclusve wth regard to frms choosng to utlze dvdends as ther cash payout mechansm. SECTION 1: INTRODUCTION Corporatons n the Unted States utlze varous mechansms to dstrbute cash to ther stockholders. Frms currently use fve prncpal methods of corporate cash dstrbutons: regular cash dvdends, specally desgnated dvdends, open-market stock repurchases, ntrafrm repurchase tender offers, and targeted or negotated share repurchases. These forms of cash payout have been the focus of numerous studes n the fnancal lterature over the past years. Early theoretcal work on cash dstrbutons, for the most part, dd not dfferentate between the dfferent types of dsbursements. 1 For example, the agency cost motvaton (to allevate agency problems assocated wth montorng and rsk averson of managers) of Easterbrook (1984), the cash flow sgnalng argument (to nform the market of an ncrease n the frm's earnngs) of Mller and 1 Except for the tax-clentele theores, early research nto ths queston treated stock repurchases and dvdends as equvalent mechansms for cash payout to stockholders. 2

3 Rock (1985), and the free cash flow theory (to reduce agency costs assocated wth excess free cash flow) of Jensen (1986) apply equally to both dvdends and stock repurchases. More recent models have consdered the choce between dfferent payout methods and have suggested possble explanatons for the form of cash dstrbuton chosen by frms. Ofer and Thakor (1987) and Persons (1994) suggest sgnalng models where the level of asymmetrc nformaton (extent of undervaluaton) determnes the payout choce. Barclay and Smth (1988) propose an alternatve asymmetrc nformaton model that concentrates on cost-mnmzaton as the determnng factor n the frm s choce of the form of the payouts to shareholders. Bagnol, Gordon, and Lpman (1989), Dens (1990), and Bagwell (1991) dentfy takeover defense as an alternatve motvaton for repurchases. Hausch and Seward (1993) model the choce as one between a determnstc (dvdends) and a stochastc (share repurchases) dsbursement and conclude that t depends on the form of the frm s producton functon (analogous to absolute rsk averson for a utlty functon). Jagannathan, Stephens, and Wesbach (2000) and Guay and Harford (2000) hypothesze that the fnancal flexblty nherent n stock repurchases contrbutes to the choce of payout method used by frms and ndcate that the permanence of the frm s cash flows are mportant n ths regard, whle Fenn and Lang (2001) examne the extent to whch management stock optons nfluence the choce. Corporate dsbursements also have a sgnfcant mpact on the stock markets and, hence, stockholders wealth. Exstng emprcal studes have revealed sgnfcant stock prce reactons to announcements of unexpected corporate cash dstrbutons (that s, dvdends and stock repurchases) [for example, Brckley (1983); Rchardson, Sefck and Thompson (1986); Smth (1987); Healy and Palepu (1988); Bajaj and Vjh (1990); and Stephenson (1994)]. On average, the market s reacton to stock repurchase announcements has been sgnfcantly hgher than the reacton to dvdend announcements. The average cumulatve (3-day) abnormal return on stock repurchase announcements has been documented to be between 5 percent and 9 percent. The correspondng excess returns for unexpected dvdend announcements (that s, ntatons, ncreases, and specally desgnated dvdends) have been observed to be between 2 percent and 3 percent. 2 2 The reacton to open-market repurchases s sgnfcantly lower than that to tender-offers -- 4 percent compared to between 7 percent to 15 percent [see Masuls (1980a), Vermaelen (1981), and Stephenson (1994)]. Ths observaton also apples to dvdends, wth the reacton to specal dvdends averagng 1 percent, dvdend ncreases 1 percent to 2 percent, and dvdend ntatons 3 percent to 4 percent [see Brckley (1983), Dens, Dens, and Sarn (1994), and Reynolds (1994)]. 3

4 Notwthstandng the hgher observed wealth effect assocated wth stock repurchase announcements, emprcal evdence shows an unexplaned preference by frms for usng cash dvdends (at least up to the md 1980s). 3 For example, durng the perod 1983 to 1986, 81 percent of all frms on the New York Stock Exchange (NYSE) used cash dvdends whle only 14 percent of the frms made stock repurchases. 4 Durng ths perod, these cash dstrbutons averaged $94 bllon per year -- representng approxmately 6 percent of the market value of the total equty base of all the lsted frms. Ofer and Thakor (1987), pertanng to the perod pror to the md 1980s, conclude that whle the majorty of US frms pay dvdends, only a relatvely small percentage utlzes stock repurchases. Interestngly, n the subsequent perod (post 1980s) there has been a marked declne n the ncdence of frms utlzng dvdends. Concurrently, the usage of stock repurchases has ncreased dramatcally. Fama and French (2001) reveal that between 1978 and 1999 the proporton of frms payng cash dvdends fell from 66.5 percent to20.8 percent. Jagannathan, Stephens, and Wesbach (2000) report qute the reverse for stock repurchases. They ndcate that between 1985 and 1996, the number of open market stock repurchase programs announced by U.S. ndustral frms ncreased from 115 to 755 (a 650 percent ncrease) whle the value of these transactons ncreased from $15.4 bllon to $113 bllon (a 750 percent ncrease). However, they observe that whle the ncdence of dvdend payments has decreased the value of these dstrbutons contnue to rse over the same perod, movng from $67.6 bllon to $141 bllon (a 109 percent ncrease). Overall, repurchases have not replaced dvdends as the prmary cash dsbursement mechansm as dvdends contnue to be sgnfcantly hgher n value than repurchases (more than double the total value of actual share repurchases). Fama and French (2001), among others, surmse that snce dvdends are usually taxed at a hgher rate than captal gans (realzed though a stock repurchase), the common presumpton s that dvdends are less valuable than captal gans. Emprcal evdence tends to support ths asserton as the stock prce reacton and, hence, wealth mpact of a stock repurchase announcement on average exceeds the wealth mpact assocated wth a dvdend decson (see endnote 4). Gven the emprcal 3 4 Ofer and Thakor (1987), Barclay and Smth (1988), and Hausch and Seward (1993), provde only partal explanatons for ths phenomenon. Extracted from Table I, Barclay and Smth (1988, p.62). The remanng 5 percent of frms s dvded approxmately equally between the frms that utlzed specal dvdends and those that nether pad dvdends nor repurchased stock over the perod covered by ther study. 4

5 evdence suggestng that hgher wealth gans to shareholders would result from the use of repurchases nstead of dvdends, the fact that frms contnue to ncrease the sze of dvdends annually and not ncrease repurchases at an even faster rate ndcate that dvdends reman an engma. The present research attempts to provde some answers n ths regard. The problem at hand, then, s to provde nsghts as to why some frms choose to pay cash dvdends whle others choose to repurchase ther stock, and, consequently, whether the choce made s n the best nterest of the frm s stockholders (that s, s t a wealth maxmzng decson?). My results ndcate that open market repurchasng frms generally are makng ther payout decsons n order to maxmze the returns to ther stockholders (through the resultng expected stock market prce reacton to the dsbursement announcement). However, smlar results were nconclusve wth regard to frms choosng to utlze dvdends as ther form of cash dstrbuton. The focus of ths study, then, s to fll the gap n the exstng lterature by emprcally examnng the stockholder wealth maxmzaton mpact of management s dsbursement choces, thereby supplementng and extendng current research n ths area. In ths research, I examne the determnants of the choce between dvdends and stock repurchases and the mpact of that choce on the stockholders wealth poston. The expectaton s that even n the presence of asymmetrc nformaton, agency costs, and dfferng expected stock prce reactons to the varous mechansms of cash dsbursements, frms, on average, choose the cash dstrbuton method that maxmzes the expected gan assocated wth the dstrbuton. Hence, managers, on average, make stockholder wealth maxmzng dsbursement choces (that s, the dsbursement choce s made wth reference to the expected excess returns generated on announcement of the decson). My results fnd support for ths proposton only wth regard to frms utlzng open market stock repurchases, notwthstandng the nfluence of other factors on the decson. A caveat s n order: whle I draw on a broad cross-secton of theoretcal underpnnngs (for example, asymmetrc nformaton/sgnallng, agency costs, free cash flow, cash flow permanence, and fnancal flexblty) n developng my emprcal model, the tests utlzed n addressng my prmary hypothess are not desgned to dfferentate between the effects of these varous theores. Further, t s not my ntent to test all the possble proxes for the varables dentfed and utlzed n the model. I am prmarly concerned wth the relatonshp between the frm s cash dsbursement choce (dvdend versus stock repurchase) and the mpact of that choce on share prces. Therefore, I employ varables from two strands of the cash dsbursement lterature (not necessarly ndependent) 5

6 that have been used n prevous studes to explan the dsbursement choce and the magntude of the assocated wealth effects. The remander of ths research paper s organzed as follows. In the next secton I further develop my hypothess and defne the factors to be used n my model, as well as dscuss the methodology to be appled n the research, ndcatng my data sources. Secton 3 then provdes the results of the study and presents an nterpretaton of these results n the present research context. A summary of the research concludes the study n Secton 4. SECTION 2: RESEARCH DESIGN 2.1 Motvaton and Development of Hypothess The mpact of a frm s choce of dstrbuton method s non-trval. In fact, the choce facng managers s one that has mplcatons for the value of ther frm. For example, Fama and French (2001) ndcate that due to tax mplcatons, frms that pay dvdends are at a compettve dsadvantage snce they have a hgher cost of equty than frms that use stock repurchases. In addton, Persons (1994) ponts out that whle the admnstratve expenses assocated wth payng a dvdend are nconsequental, repurchases usually nvolve substantal transactons costs. Hence, wth taxes, transactons costs, and asymmetrc nformaton, the frm s choce of a payout method s an mportant decson wth valuaton mplcatons. Wth the plethora of theoretcal and emprcal fnancal research relatng frm characterstcs to the valuaton mpact (excess stock market returns) of cash dstrbuton announcements (for example, Dann (1981), Vermaelen (1984), Mller and Rock (1985), John and Wllams (1985), Jensen (1986), Bajaj and Vjh (1990), Dens, Dens, and Sarn (1994), Chhachh and Davdson (1997), Guay and Harford (2000), among others), t would seem an easy task to assess the opportunty cost of a partcular dsbursement choce (that s, the dfference between the expected wealth mpact of the method used and that of an alternatve method). However, f ndeed frm s make the dsbursement choce on the bass of value maxmzaton, the ssue s much more complcated snce the dsbursement choce would then be endogenzed; that s, the wealth effect assocated wth a partcular choce of cash dstrbuton would be drectly related to the choce model tself. In ths case, correct specfcaton of the expected wealth mpact of an alternatve cash payout method would requre jont modelng of the dsbursement choce 6

7 equaton and the wealth effects models for both alternatve payout methods. Ths ssue has not prevously been addressed n the fnance lterature and serves as the major contrbuton of the present research. My prmary hypothess, then, may be stated as: H o : H a : Managers do not dscrmnate n ther choce of a payout method. Managers dscrmnate between dvdends and repurchases by maxmzng the expected abnormal returns followng the dsbursement announcement. The valdty of the hypothess s examned n two stages. Frst, relevant varables are extracted from the lterature as t relates to motvatons for cash dsbursements and these are utlzed n jontly estmatng (usng a self-selectvty modelng approach) the manager s dsbursement decson and the resultng stock market excess returns around the announcement of the decson. In step two I examne what the expected excess returns would have been had the alternate choce been made by the manager and then conclude whether the choce was a stockholder wealth maxmzng one. 2.2 Methodology Development and Determnaton of Test Statstcs The present study examnes a frm s choce between dvdends and stock repurchases, specfcally, the choce between a dvdend ncreases and an open market stock repurchases. An ncrease n the cash dvdend generally nvolves a commtment by the frm to mantan an ncreased payout over the foreseeable future 5. Open market repurchase programs nvolve frms merely announcng ther ntenton to buy back shares over an extended perod of tme at the prevalng market prce. It nvolves an ongong commtment by the frm to make cash dsbursements to ts stockholders (albet not all stockholders). The natural comparson between open market repurchases and dvdend ncreases (frequent, perodc payout) s supported by recent emprcal and theoretcal work (for example, see Chhachh and Davdson (1997), Stephens and Wesbach (1998), Guay and Harford (2000), and Fenn and Lang (2001)). In modelng the above choce I apply a lmted dependent varables estmaton technque known as self-selectvty. Self-selectvty mples that we do not observe frms randomly choosng to 7

8 dstrbute cash to ther stockholders n the form of dvdends or stock repurchases. Rather, the observed choce of dsbursement method s the result of a delberate and specfc decson made by the frm. Accordng to Maddala (1991): The self-selecton model s based on the dea that ndvduals choose one of two groups on the bass of expected benefts from belongng to the two groups. Sometmes the benefts can be captured by the stock prce As such, the observed cross-sectonal nformatonal effect s condtonal on the choces made. Therefore, we would not expect the same average effect to be observed for frms randomly choosng to engage n the event. The process generatng observed abnormal stock market returns s thus modeled as a swtchng regresson model wth endogenous swtchng, 6 requrng the researcher to smultaneously estmate: () the uncondtonal cross-sectonal announcement perod cumulatve abnormal return experenced for each event type, () the decson process adopted by the frm n choosng between the dfferent methods of dsbursement, and () the mpact of the frm s choce of dsbursement type on the observed announcement perod cumulatve abnormal returns. Our foundatonal premse s that a frm, havng decded to dstrbute cash to ts stockholders, wll make a dvdend payment only f the net gan from ths opton s greater than the gan arsng from a stock repurchase. That s, dvdends wll be used as the cash dsbursement choce f ( VD V ) ( VR V0 ) > CD CR 0 (1) where V D and VR are the values of the frm after makng the dvdend payment or stock repurchase, respectvely, V 0 s the value of the frm before makng the cash dstrbuton, and C D and CR are the respectve costs assocated wth the dvdend payment or the stock repurchase. If we standardze all varables n terms of the value of the frm before the dsbursement, V 0, then the frm wll utlze dvdends f * I R R c > 0 (2) = D R where I * s the net gan from payng dvdends rather than repurchasng stock, R D s the return from makng a dvdend payment, R R s the return from makng a stock repurchase, and c s the 5 Emprcal fndngs ndcate that frms are unlkely to ncrease dvdends unless they perceve that the ncreased dvdend can be mantaned. See for example Mller and Rock (1985), Ofer and Segel (1987), and Dens, Dens, and Sarn (1994). 6 See Maddala (1983), pg. 223 et seq. for a more detaled treatment of the model. 8

9 dfference n cost of makng a dvdend payment relatve to a stock repurchase expressed as a fracton of the value of the frm. I *, the decson varable, s a latent unobservable varable. The frm wll make a dvdend payment where the net gan, I *, s greater than zero and utlze a stock repurchase where t s less than zero. Although the decson varable s not observed, we do however observe the frm s choce, and ths s modeled by the bnary selecton ndex (dummy varable) defned as: I 1 f I * 0 (for dvdends) = I = 0 f I 0 * < (for stock repurchases) Smlarly, for each frm makng a cash dstrbuton, the cumulatve abnormal return around the announcement of the dsbursement choce s observed ex-post. We can thus specfy the excess returns equatons for frms makng dvdend payments and stock repurchases as follows: * CAR D X D β + µ ff I 0 (3) = D D * CAR R X R β + µ ff I < 0 (4) = R R Equaton (3) represents the cumulatve abnormal return to be expected by a frm on announcement of a dvdend dsbursement whle equaton (4) represents a smlar effect for the frm choosng a stock repurchase. The X terms represent the exogenous factors expected to nfluence the wealth effect assocated wth the dsbursement. These are outlned and dscussed n a later secton. β D and β R are vectors of coeffcents that may dffer dependng on whch dsbursement choce s used, whle µ D and µ R are the error terms n the respectve regresson equatons. We can substtute equatons (3) and (4) nto equaton (2) to yeld a reduced form selecton ndex, namely: X ( β β ) + ( µ ) I = µ D R or D R I = γ µ (5) Z The coeffcents n equaton (5) are not drectly observable, however, due to the selfselectvty bas resultng from the dsbursement choce beng endogenously determned. That s, the selecton bas arses because the choce of a dsbursement method and the abnormal returns on announcement of the decson are jontly determned by a common set of unobservable factors. The result s that the error terms n equaton (3) and (4) wll be correlated wth the error term n equaton (5) and wll have non-zero expectatons. Accordng to Shehata (1991): 9

10 Recent developments n econometrcs suggest that, n the presence of self-selecton bas, usng OLS n the usual fashon to estmate regresson models could result n neffcent and nconsstent estmates. Gven the observatons I, I use the probt maxmum lkelhood to estmate the parameter γ. However, γ s estmable only up to a scale factor and I thus set Var (µ )=1 [Maddala (1991) ndcates that the assumpton of Var(u )=1 s because I * s observed only as a dchotomous ndcator]. Fnally, I assume that µ D, µ R, and µ I have a trvarate normal dstrbuton wth mean vector zero and covarance matrx: = σ σ σ σ σ σ σ DU σ 1 2 D DR DU 2 DR R RU RU (6) Snce σ DR s not estmable by maxmum lkelhood (by desgn I treat repurchases and dvdends as separate observatons and never group these for the same frm), I can set t equal to zero and transform the Σ matrx n (6) to obtan: 2 σ D 0 2 = 0 σ R σ DU σ RU σ DU σ RU 1 (6a) The lkelhood functon for the model s then gven by: L 2 2 ( β, β, σ, σ, σ, σ ) D R Zγ D R Dµ I ( CAR X β, µ ) dµ f ( CAR X β, µ ) = g D D Rµ D Zγ R R R dµ ( 1 I ) (7) where g(.) and f(.) are the bvarate normal densty functons of (µ D, µ ) and (µ R, µ ) respectvely. Although maxmzaton of the lkelhood functon n equaton (7) s possble, t can be qute cumbersome. Lee (1978) outlned a smpler two-stage structural probt estmaton method that nvolves frst estmatng γ from the reduced form bnary choce equaton (5) by probt maxmum lkelhood (ML) and then usng ths estmate to transform and solve equatons (3) and (4) by ordnary least squares (OLS). The predcted beneft dfferental, ( CAR D CAR R ), s then 10

11 ntroduced n the dsbursement choce equaton (5) to obtan the structural form probt equaton that allows for consstent estmaton, agan by applyng maxmum lkelhood procedures. The detaled two-stage structural probt estmaton procedure s as follows: Frst, obtan the expected values of µ D and µ R condtonal on the frm s choce of beng n the sample (another way to thnk of ths s that we are consderng the expectaton of the abnormal returns condtonal on the dstrbuton beng observed, whether a dvdend or a repurchase), whch s defned as 7 : E ( γ Z µ ) µ D = σ Dµ ( γ Z ) ( γ Z ) φ (8) Φ and E ( γ Z < µ ) µ R = σ Rµ φ( γ Z ) ( 1 Φ( γ Z )) (9) In equatons (8) and (9) the frst term on the rght-hand sde of the equatons measures the relatonshp (covarance) between the manager s decson (choce of dsbursement method) and the outcome of the decson (resultng abnormal return), whch ndcates whether managers are actng on shareholders behalf. It essentally s the lnear regresson coeffcent that results from regressng the error terms n the decson model (equaton (5)) aganst the error terms n the abnormal returns models (equatons (3) and (4)). The second term, referred to as the Inverse of the Mlls Rato (or the non-selecton hazard), s an expectaton of the value of the error term n the decson model condtonal on the frm usng ether a dvdend or a stock repurchase respectvely (snce the condtonal dstrbutons of these error terms are normal). The Inverse of the Mlls Rato s the rato of the probablty to the cumulatve densty functons evaluated at the pont at whch the dstrbuton s separated. As the probablty of beng n the selecton sample (n ths case, dstrbutng through dvdends) ncreases, the cumulatve densty functon approaches one and the probablty densty functon approaches zero, so the Inverse of the Mlls Rato approaches zero. Hence, a postve (negatve) coeffcent on ths varable n the 7. Ths s a necessary correcton for the condtonal expectaton gven that we have non-random selecton, that s, certan unts from the underlyng populaton do not appear n a random sample due to ther ndvdual dsbursement choce. 11

12 dvdend (repurchase) abnormal returns equaton ndcates that sample selecton s mportant and that ndeed managers are makng decsons wth regard to the welfare of the frm s stockholders. Ths result mples that the error terms n the abnormal return regresson equatons wll have non-zero expectatons (and, hence, the self-selectvty bas). Gven these two expectatons, defne W rewrte equatons (3) and (4) as: CAR D ( γ Z ) ( γ Z ) φ = and W Φ R φ( γ Z ) ( Φ( γ Z )) = 1, and then β X + σ µ W + ε for I = 1 (3a) D = D D D D D CAR R β X + σ µ W + ε for I = 0 (4a) = R R R R R where the new error terms, ε D and ε R have zero condtonal means. Equatons (3a) and (4a) provde an nsght nto the self-selectvty ssue. Instead of lnear equatons we have two non-lnear equatons after the non-zero means have been adjusted. Equaton (3a) shows that the expected CAR for a frm that announces a dvdend conssts of two separate components. The frst term, β D X D to announce a dvdend payment. The second term, σ, s the expected stock market effect for a random frm that elects D µ WD, s the adjustment for self-selectvty that may be nherent n the sample. The covarance term, σ D, s of partcular mportance. It ndcates µ that a randomly selected frm, were t to choose to pay a dvdend, would not experence a smlar stock prce effect to that experenced by frms that actually pad dvdends. Smlar reasonng would apply to the terms n equaton (4a). Usng our estmate of γ from the probt maxmum lkelhood estmaton of equaton (5) we obtan estmates for W D and W R n equatons (3a) and (4a) respectvely. We then proceed to solve these equatons by ordnary least squares regresson, whch wll provde consstent estmates for β D, β R, σ Dµ, and σ Rµ. A test for the presence of self-selectvty bas s then performed by examnng the statstcal sgnfcance of the coeffcent on the W terms n the revsed abnormal returns equatons (3a) and (4a). Two potental problems arse wth ths estmaton procedure, however. Frst, the resduals ε D and ε R n equatons (3a) and (4a) are heteroscedastc. The second potental problem wth the two-stage structurtal probt approach was dentfed by Lee, Maddala, and Trost (1980) who show 12

13 that the true varances n equatons (3a) and (4a) wll be underestmated snce the selectvty varables are themselves estmates, that s, they are generated regressors. However, the computer package used n estmatng these equatons n the present research, LIMDEP, provdes a full nformaton maxmum lkelhood estmator (FIML) that jontly estmates all the parameters n the model and corrects for these dffcultes. Ths methodologcal approach s thus utlzed n the present study nstead of the two-stage structural probt approach outlned above. Havng estmated the two abnormal return regresson equatons, Maddala (1991) suggests that our next step s to examne whether there are, n fact, any sgnfcant changes n the estmates of the effect of the explanatory varables. Ths s done by comparng the coeffcents on the varables n the regresson equatons estmated wth and wthout correcton for the self-selectvty bas. Ths wll ndcate whether gnorng the non-random selecton process has ndeed produced msleadng results. We next proceed to estmate what the predcted abnormal return would have been had the frm used the alternate dsbursement choce, by applyng the relevant varables nto the estmated CAR models. Ths, n effect, s the man purpose of the analyss. In ths procedure the selectvty terms are not needed and, hence, are omtted. The purpose of estmatng the selectvty equatons (3a) and (4a) was to obtan estmates of β D and β R that are free of the selectvty bas and hence any further analyses uses these parameter estmates. If managers are makng ther dsbursement decsons n the best nterests of the frm s stockholders then we would expect that the dfference between the excess returns resultng from the frm s dsbursement choce and the predcted excess returns from choosng the alternate payout method would be postve and statstcally sgnfcant. Ths s tested by examnng the dfference between the mean abnormal returns for frms that made a partcular dsbursement choce and the mean predcted abnormal return for those frms had they chosen the alternate method. The fnal step s to estmate a beneft dfferental (BENEFIT), calculated as the dfference between the predcted abnormal returns for all frms f they choose to use dvdend payments and the predcted abnormal returns had they nstead chosen a stock repurchase (that s, CAR ˆ CAˆ R ). Ths addtonal explanatory varable s then ncluded n the dsbursement choce equaton (5), producng a structural form selecton ndex, whch s re-estmated by the probt maxmum lkelhood method. A statstcally sgnfcant coeffcent on the beneft dfferental varable ndcates D R 13

14 that managers make ther choce of a dsbursement method on the bass of the dfferental n the expected abnormal returns (net-beneft). To estmate equatons (3a) and (4a) we need to provde unbased estmates of the uncondtonal CAR experenced by frms around the announcement of the relevant dsbursements. Ths s done usng standard event-study procedures employng estmated market-model parameters. For ths purpose I use returns for each frm (from the CRSP data base) over 190 tradng days (approxmately nne calendar months) from day -210 to day -21, relatve to the announcement day, to estmate a market model of the form: R t = α + β R mt + ε t (10) R t s the return on frm s stock on day t, R mt s the return on the CRSP value-weghted ndex 8 on day t, ε t s the error term n the model (assumed to be normally dstrbuted wth a common mean but unequal or nonhomogenous varance -- that s, heteroscedastc), and α and β are the parameters that wll be estmated n the OLS regresson. The estmaton perod s chosen so as to be close enough to the event perod to approxmate the true beta durng the announcement nterval whle beng far enough to be uncontamnated by the event. Usng the returns generated from the estmated model, the abnormal return for frm s stock on day t (AR t ) s calculated as the devaton of the predcted (estmated) return for day t from the actual return on day t. That s: AR t t t ( ˆ α + ˆ β R ) mt = ε = R (11) The abnormal returns for each frm are then summed for days -1 to +1 to arrve at the three day cumulatve abnormal return around the announcement: CAR + 1 = AR t = 1 t (12) We nclude n the announcement nterval day -1 because a leakage of nformaton may cause a substantal prce reacton on ths day whle day +1 s ncluded to account for announcements that are made after the stock market has closed for tradng. The CAR values are then used n the OLS (or WLS) estmatons of equatons (3a) and (4a). The sgnfcance of the coeffcents on β D and β R, as well as the coeffcents on the self-selectvty varable, can then be examned by usng standard t-test statstcs. 14

15 2.2.1 Tests for Uncondtonal Wealth Effects We are prmarly nterested n examnng the dsbursement choces of managers and ts mpact on stockholders wealth as measured by the abnormal returns observed around the dsbursement announcement. However, Ross (1989) shows that ncreases n the rate of flow of dosyncratc nformaton manfest themselves n ncreases n stock prce volatlty. In lght of ths, changes n the varance of the stock returns dstrbuton may be mstakenly dentfed as wealth effects. Snce the cumulatve abnormal returns observed around the dsbursement announcement s the dependent varable n the self-selectvty models (specfcally the wealth effects regresson equatons), t s prudent to frst nvestgate the statstcal sgnfcance of the abnormal returns assocated wth dvdend and repurchase announcements, condtonal on any varance changes (snce the varance changes would overstate the true wealth effect). In lght of ths, I examne the effect of stock repurchase and dvdend announcements on frm volatlty by usng the uncondtonal mean (wealth) effects test of sgnfcance suggested by Sanders and Robns (1991), that ncorporates the effect of an event nduced change n the varance of the stock returns dstrbuton. To provde an uncondtonal test of the mean CAR around the event announcement, Robns and Sanders (1993) (RS) suggested a multple-day event perod analog to the t-statstc developed by Collns and Dent (1984)(CD) to test sngle-day average abnormal returns measures. The CD statstc s shown to be asymptotcally the best lnear unbased estmator of the average abnormal return and ncorporates n ts formulaton any seral correlaton between the market returns over the estmaton perod. The RS analog s calculated as follows: t CAR = I = 1 ACAR 2 2 [( CAR ACAR) σcar ] ( ) I 2 I 1 = 1( 1 σcar ) (13) ACAR, the average cumulatve abnormal return, s calculated usng the formula: ACAR = I = 1 2 ( CAR σcar ) I = ( σcar ) (14) 8 CRSP provdes a sngle composte ndex ncorporatng all frms on NYSE, AMEX, and NASDAQ. 15

16 and: σ K K 2 = σ + + T 2 2 CAR + 1 ( t = 1( rmt rm) ) 10 τ = 150 ( rmτ ) r m 2 2 (15) where: 2 σ K T r mt r mτ r m resdual varance from estmaton of the market model for frm 3; the number of days accumulated n the calculaton of CAR number of returns used to estmate the market model for frm return to the market portfolo on event-day t return to the market portfolo on estmaton day τ mean return to the market portfolo over the estmaton perod Ths procedure, n effect, employs an estmated generalzed least squares methodology to calculate the cumulatve abnormal returns (CAR ). Under the null hypothess that the average cumulatve abnormal returns equals zero, t CAR follows a Student-t dstrbuton wth I-1 degrees of freedom. For comparatve purposes, I also calculate the smple average cumulatve abnormal return (AVGCAR) and a Z-test based upon the average standardzed cumulatve abnormal return (ASCAR), as these are frequently reported n the event study lterature. These are: AVG CAR I CAR = = 1 I (16) and Z CAR = I I = 1 [ CAR σ CAR ] I = I ( ASCAR) (17) Although the Z-test adjusts for and ncorporates any seral correlaton n the predcton errors (abnormal returns), t nevertheless gnores any event nduced changes n the resdual varance of the abnormal returns dstrbuton. Further, Dens and Kadlec (1994) observed that non-synchronous tradng -- the tendency for prces recorded at the end of a day to represent the outcome of a transacton occurrng earler n that day -- causes seral cross-correlatons n securty returns, leadng to based estmates of systematc rsk when usng smple ordnary least squares regresson to estmate the market model. In addton, they fnd sgnfcant decreases n tradng actvty followng share repurchases. Gven that I have requred frms n my sample to have no mssng returns durng the announcement perod and no 16

17 more than 15 days mssng returns durng the estmaton perods, ths s not expected to be a cause for concern n ths study Sample Selecton and Descrpton The data used n the study covers the perod and conssts of the followng subsamples: 1. The sample of frms wth dvdend ncreases are selected by randomly searchng the Center for Research n Securty Prces (CRSP) Daly Returns Master Fle for frms wth ncreases n consecutve regular quarterly dvdends per share over the perod covered by the study. In addton, no other type of dstrbuton must be made by the frm durng the perod between the two quarterly dvdends. Ths comprses all frms lsted on ether the New York Stock Exchange (NYSE), the Amercan Stock Exchange (AMEX), or the North Amercan Securtes Dealers Automated Quotaton (NASDAQ) System. The market reacton theores presented earler predct a prce reacton only to announcements of unexpected dvdend ncreases. In an attempt to capture ths, I requre that the ncrease must be at least 10 percent n order for the announcement to be ncluded n the sample. Ths lower bound of 10 percent ensures that only economcally sgnfcant dvdend changes are ncluded n the sample 10. In addton, to mnmze the effect of outlers, I mpose an upper bound of 700 percent on the sze of the dvdend ncrease. Further, to dentfy and quantfy unexpected dvdend sgnals, the dvdend ncrease must be the frst n any seres of consecutve regular quarterly ncreases of Notwthstandng, results of all the above statstcal tests are reported after re-estmatng the market model (more specfcally the systematc rsk component, β) usng the methodology proposed by Scholes and Wllams (1977). As reported n Fowler and Rorke (1983), the re-estmated beta s gven by the followng, shown to be a consstent estmator: where: 1 0 β plm $ = ( β + β + β ) ( 1+ 2ρ ) β = the parameter estmate obtaned from the smple regresson of R t aganst R mt-1 β = the parameter estmate obtaned from the synchronous smple regresson +1 β = the parameter estmate obtaned from the smple regresson of R t aganst R mt+1 ρ = the frst order seral correlaton coeffcent for the market ndex, R 1 m Elmnatng small dvdend changes would also mnmze problems arsng from msspecfcaton n the model of expected dvdends snce large dvdend changes are lkely to be categorzed as dvdend surprses regardless of the expectaton model employed. 1 17

18 smlar magntude. To quantfy the dvdend changes I apply the nave expectatons model, whch states: ^ t, = t, 1 D That s, the best estmate at tme (t-1) of dvdends n tme (t) s the dvdends pad at tme (t-1). Usng ths model, unexpected dvdends s thus represented by the actual amount of the dvdend ncrease. D The use of the naïve model s supported by the emprcal observaton that frms generally do not change ther dollar dvdends frequently and hence follow a farly stable, predctable dvdend payment polcy. Damodaran (2001), usng data from Compustat, reports that between 1989 and 1998, n most years the number of frms that do not change ther dollar dvdends far exceeds the number that do 11. Lntner (1956) n hs classc study on how managers make dvdend decsons, found that they stablze dvdends wth gradual, sustanable ncreases whenever possble, establsh an approprate target payout rato, and avod dvdend cuts, f at all possble. Fama and Babak (1968) reevaluated Lntner s model and concluded that t contnues to perform well relatve to alternatve specfcatons usng both economywde earnngs and dvdend data as well as data for ndvdual frms. 2. I dentfy the ntal sample of open market repurchase programs and repurchase tender offer announcements from the followng sources: The appendx to Comment and Jarrell (1991) coverng announcements from 1984 to A general search of the repurchases database of the Securtes Data Company (SDC). A general search of the WSJI of the LEXIS/NEXIS reference database. A general search of the CRSP master fle. Ths sample s reduced by excluson of repurchase offers avalable only to odd-lot holders, those offers by closed-end nvestment companes, and offers whose ntenton was to take the frm prvate. The ntal samples are reduced by applyng the followng screens to the data: 11. See Damodaran (2001) Fgure 21.6, page

19 1. Snce the model mples a mutually exclusve choce between dvdends and repurchases, I exclude from the sample frms that concurrently announce both a dvdend and a stock repurchase Frms must have returns data avalable on CRSP for at least 250 tradng days (one calendar year) before and 150 tradng days (seven calendar months) after the date of the dsbursement announcement. In addton, there can be no more than 15 days mssng returns durng the estmaton perod from 210 to 21 days pror to the event date, and no mssng returns over the 3-day event perod. 3. Frms must have the relevant accountng data avalable on the COMPUSTAT database for calculaton of the varous measures used n the decson models (these are detaled n a later secton). 4. I elmnate from the sample fnancal frms (SIC codes ), utltes (SIC codes ), and regulated telephone companes (SIC code 4813) 13. Event dates for the varous announcements are taken from the relevant sources (that s, CRSP, WSJI, SDC database, or Comment & Jarrell s Appendx). The fnal sample conssts of 2,423 dvdend ncreases and 1,931 open market repurchases. Table 1 shows the dstrbuton of announcements across the sample perod, broken down wth respect to dsbursement type and year. It appears that the observatons are farly evenly spread across the sample perod. The notable excepton to ths s the number of open market repurchase announcements n 1987 and Ths can be accounted for by the documented ncrease n repurchase authorzatons after the stock market crashes n 1987 and 1990, supposedly n response to the belef that stocks were hghly undervalued at these tmes. Overall, the sample of dsbursement announcements does not dsplay any major problems of clusterng n any sngle year. 2.4 Identfcaton of Explanatory Varables A number of factors emerge from the fnance lterature as potental dscrmnators of dsbursement type. Based on the overwhelmng support for nformaton sgnalng by fnancal researchers, proxes for sgnalng should be useful n emprcally dfferentatng between managers' There were 59 frms that announced both a dvdend ncrease and an open market repurchase program smultaneously. Fnancal frms are consstently omtted from smlar studes prmarly because ther repurchases are not consstently reported (Fenn and Lang, 2001), (Fama and French, 2001). Heavly regulated frms (utltes and 19

20 choces of the form of ther cash dstrbutons. I use two proxes to measure managers sgnalng of prvate nformaton and the level of nformaton asymmetry. These are () the change n annual earnngs per share between the year pror to and the year subsequent to the dsbursement, scaled by the frm s stock prce 5 days before the announcement date (DEPS), and () the resdual volatlty n daly stock returns n the year precedng the event announcement, (RVOL), measured as the standard devaton n the market-adjusted daly stock returns. Table 1. Dstrbuton of Sample Announcements by Type and Year Year Dvdend Increases Open Market Repurchases TOTAL TOTAL DEPS s used to proxy for sgnalng snce the theory posts that mproved operatng performance s ncluded n the content of the sgnal. Derkns(1991) and Krshnaswam and Subramanam (1999) suggest that nformaton asymmetry (hgh when managers have a relatvely large amount of value-relevant, frm-specfc nformaton that s not shared by the market) can be captured by the market-adjusted standard devaton of the daly stock prce abnormal returns (R t R mt ). Hence we use RVOL as a proxy for the level of nformaton asymmetry. Ofer and Thakor telephone companes) are omtted because ther payout polces may be sgnfcantly affected by ther regulated status (Fenn and Lang, 2001). 20

21 (1987) suggest that greater nformaton asymmetry should be characterstc of the stock repurchasng frms relatve to frms that use dvdend payments. Hence, I expect comparatvely larger values for these varables to be assocated wth the use of repurchases, whle smaller values should be assocated wth dvdends. The sgnalng hypothess also posts that repurchasng frms are undervalued and, n ths regard, we would expect the market s valuaton of frms utlzng repurchases to be lower, ceters parbus, than for those dstrbutng cash through dvdends. Tobn s Q, (TOBINQ) a measure of the frm s nvestment opportunty set, s used as a proxy for classfyng frms as ether growth frms / value-maxmzers (Q>1) or overnvestors (Q<1). I adopt Chung and Prutt s (1994) equaton 2 to proxy for Tobn s Q: q = (MVE + PS +DEBT) / TA where MVE s the market value of the frm s common stock, PS s the lqudatng value of the frm s outstandng preferred stock, DEBT s the value of the frm s short-term labltes net of ts shortterm assets, plus the book value of the frm s long-term debt, and TA s the book value of the total assets of the frm. They show that ths approxmaton to Q explans at least 96.6 percent of the varablty n the more theoretcally correct model of Tobn s Q. Lang and Ltzenberger (1989) fnd that frms wth Q<1 have, on average, greater stock prce reactons to dvdend changes than do frms wth Q>1. Dens, Dens, and Sarn (1994) also fnd evdence that Tobn s Q and dvdend yeld are negatvely correlated. Snce Q s used as a measure of growth opportuntes we expect that hgher ratos should be assocated wth hgher-valued frms and lower ratos assocated wth lower valued frms. Because the sgnalng/undervaluaton hypothess suggests that repurchases are used manly by frms that are undervalued, I expect frms choosng dvdends to be those wth hgher ratos for Tobn s Q. Closely lnked to ths s the use of a proxy measure for the level of free cash flow exstng wthn the frm at the tme of the dsbursement decson, (FCF). As prevously used by Maquera and Meggnson (1994), ths s calculated as the after-tax undstrbuted cash flow of the company (cash flow from operatons net of debt payments, preferred dvdends and common dvdends) dvded by the market value of ts equty. Free cash flow theory posts that corporate dsbursements are used to reduce free cash flows and thereby lower the assocated mtgatng agency costs. Takng ths nto consderaton, as well as the emprcal observaton that the mones dstrbuted by companes durng stock repurchases usually represent a larger fracton of ther outstandng equty as compared wth 21

22 dvdends, 14 we can expect hgher levels of free cash flow to be assocated wth greater utlzaton of stock repurchases. However, as dscussed earler, only a small percentage of repurchases should be undertaken for the specfc purpose of reducng agency costs -- snce emprcal observatons suggest that repurchases usually nvolve external fnancng. In ths regard, t s not certan, ex-ante, how well the level of free cash flow wll perform as a dscrmnatory varable. A potentally useful factor n the model, as suggested by Bagnol, Gordon, and Lpman (1989) and Bagwell (1991), s a measure for corporate control, specfcally defense aganst hostle takeovers (TKOVER). Ths s ntroduced as a dummy varable representng the presence of such actvtes facng the frm wthn one year precedng the dsbursement announcement 15. In the present framework only stock repurchases has been suggested as a possble mechansm for such control. I would thus expect a varable measurng the presence of takeover actvty (and possbly the exstence of agency problems) to be related to the form of dsbursement used by the frm. One testable predcton of captal structure hypothess s that repurchasng frms should have less leverage than non-repurchasng frms. In the decson model, I use the frm s debt/equty rato (LTDEQ) -- measured as long-term debt dvded by the book value of equty -- as a measure of the frm s fnancal leverage. Fenn and Lang (2001), n studyng the relatonshp between open market repurchases and dvdend payment, fnd that repurchases are postvely related to proxes for free cash flow and negatvely related to proxes for margnal fnancng costs. Frm sze has been emprcally related to both market return and dsbursement characterstcs, and s a plausble proxy (nverse) for margnal fnancng costs. Hence, I nclude a factor for sze, (SIZE), calculated as the natural log of the market value of the frm s equty 5 days pror to the announcement date. However, snce Fama and French (2001) conclude that smaller frms are less lkely to pay dvdends, the ex-ante relatonshp of frm sze to dsbursement choce s not certan. Fenn and Lang (2001) also conclude that the presence and level of management stock optons nduces a preference for open market repurchases compared to dvdend payments. Gven ths, I nclude a proxy for management stock optons (MNSTK) n the dsbursement decson equaton. The proxy I use s adopted from ther paper and s calculated as the number of common shares reserved for converson for stock optons, convertble securtes, and warrants, dvded by the total number of shares the frm has outstandng. 14 See Ofer and Thakor (1987) for a theoretcal justfcaton of ths observaton. 15 Data on hostle takeover target announcements are taken from the Securtes Data Company database. 22

23 Dvdend yeld s also expected to be an mportant varable n the frm s choce between dvdends and repurchases. Ths can be consdered as a proxy for a frm s tax-clentele. The varable DIVYLD represents the average dvdend yeld of the frm for the three years leadng up to the dsbursement announcement. Based on the clentele argument, frms wth hgh dvdend yelds pror to the dsbursement wll be more lkely to contnue usng dvdends as a means to dstrbute cash to shareholders. Addtonally, f stock repurchases and dvdends are partal sgnalng substtutes, then we would expect the stock market s prce reacton to a repurchase announcement to be negatvely related to the frm s pror dvdend yeld. The fnancal flexblty hypotheses of Guay and Harford (2000) and Jagannathan, Stephens, and Wesbach (2000) ndcate that measures of earnngs volatlty, cash flow permanence, and pror stock performance are mportant n dscrmnatng between dvdends and repurchases. In smlar fashon, I use EARVOL the standard devaton n the rato of operatng ncome to total assets of the frm over the fve years leadng up to the announcement to measure earnngs volatlty and AVGRET the average daly stock return n the year precedng the announcement to estmate pror stock performance. I apply two varants of ther measures of cash flow permanence: RELPERM measures the relatve proporton of permanent cash flows and s calculated as the average of the rato of operatng to total ncome (operatng plus non-operatng ncome) over the three years pror to the announcement and CFPERM measures the dfference n the average rato of cash flow from operatons to total assets n the three years before and after the announcement. Fnally, n lne wth the conclusons of Grullon, Mchaely, and Swamnathan (2002) that the abnormal returns around dvdend announcements are related to the declne n systematc rsk, I nclude DBETA n the abnormal returns equatons to proxy for the change n systematc rsk (measured as the dfference n the CRSP market-model beta of the frm, estmated for 150 tradng days before and after the announcement). Descrptve statstcs for each of the factors mentoned above are provded n Table 2 for the 4354 frms n the fnal sample (separated accordng to the dsbursement method used). SIZE, AVGRET, CFPERM, and DBETA appear to be approxmately normally dstrbuted. However, all the other varables dsplay defntely skewed dstrbutons, wth the means generally beng larger than the correspondng medan (except for FCF and RELPERM that have medans hgher than ther means). The average sze of frms n the sample was (equvalent to $382 mllon), whle the 23