Dividends and Taxes: Some Empirical Evidence Merton H ...ecsocman.hse.ru/data/887/126/1231/miller_scholes_-_dividends_1982.pdf · prior permission, you may not download an entire

Dividends and Taxes Some Empirical Evidence

Merton H Miller Myron S Scholes

The Journal of Political Economy Vol 90 No 6 (Dec 1982) pp 1118-1141

Stable URL

httplinksjstororgsicisici=0022-38082819821229903A63C11183ADATSEE3E20CO3B2-4

The Journal of Political Economy is currently published by The University of Chicago Press

Your use of the JSTOR archive indicates your acceptance of JSTORs Terms and Conditions of Use available athttpwwwjstororgabouttermshtml JSTORs Terms and Conditions of Use provides in part that unless you have obtainedprior permission you may not download an entire issue of a journal or multiple copies of articles and you may use content inthe JSTOR archive only for your personal non-commercial use

Please contact the publisher regarding any further use of this work Publisher contact information may be obtained athttpwwwjstororgjournalsucpresshtml

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academicjournals and scholarly literature from around the world The Archive is supported by libraries scholarly societies publishersand foundations It is an initiative of JSTOR a not-for-profit organization with a mission to help the scholarly community takeadvantage of advances in technology For more information regarding JSTOR please contact supportjstororg

httpwwwjstororgSun Oct 21 082844 2007


Merton H Miller and Myron S Scholes C t i l l ( l t (ij ( ~ I ( ( I ~ o

Ihis plpel r e e s ~ m i n e s some recent tests o f whe the r ho lde r o f slii~es uith higlier d ividend yields receive higher risk-acijusted rates of Ietul n to compensa te f o r tlie I iea ier taxes on d i i d e n d payments tliln o11 long- term capit~l gains O u r particular concern is ~ v i t h tests using r1iolt-run measures of d iv idend yield-that is measures tha t seek to d e d u c e the differenti~l tax b u r d e n on d i idends (gtel- long- tel-xn capital gain5 frorn diffelences in rates of r e t u ~ m o n sli ~res tha t do a n d shares that d o not pay I cash dividend dul-ing tlie r e t u ~ n intel ~l e shov that such meisures ire i nappropr i a t e f o r that pu1p)e Any yield- elated effects associ~ted ~vi t l i such measures 11ll1st I] ire frorn s o u r c e othel tlian tlie long- term tax di f ferent i ~l For thc s l i o ~ t - ~ u n measures considered he re t he yield-related effects lbunt l i n some tests a r e traced to biases o n e o f a fairly subtle k ind i n t rod~cced11v dividend a n n o u n c e m e n t effects

I Introduction

The publication of Black and Scholess (1974) study came after nearly tu0 decades of intense academic controversy over the effects of dividends on stock prices Using the large Center for Research in Security Prices (CKSP) data base and employing new econometl-ic methods

Ineirliel- velsion of this papel- iras prcsenteti at the Conference on Tzisation ~ tthe Unirrsitv of Southel-n C~lifor-nix Lms Angeles Jlnuary 22 1981 Ve ickno~vledge ith thiiiks the comments o f our discusslnt Marc Keinganuni ~ n d of the conference pll-ticiprnts espcciallv Richarti Roll Helpful comments o n earliel- dl-afts -el-e 1-eceived tl-om Fisther Black Geol-ge Constantiniclcs Eugene Farna Robert Harnada Patrick He Rol~crt Holthauscn Jon Ingelsoll hfichael Jensen Richard Leftwich James Loric Kenneth Reid Inti IVilliam Sch~vel-t Our thanks ~lso to the American National Hank and to the Center fol- Kesearch in Security Prices for rupport of thir research to the letel-ce fi)r I nurnlx~ of valuable suggestions and especially to Donalti Keim ti)r help in designing earl-ving out and interpl-eting the empil-ical tests

that avoided many of the difficulties that had hampered earlier testing efforts Black and Scholes found no significant I-elation betlveen stock returns and dividend yield or- dividend payout Their resu1t~ thus lent support to neither of tlvo contending hypotheses about dividend effects-the conventional view that the market prefers to obtain the income from stock as dividends and the contrary vie~v widely held among academics that the market demands higher- re- tur-ns on dividend-paying shares to compensate for tax penalties on dividend income

Although acadernic I-eseal-chel-s continued to speculate about lvhy so seemingly important a yield-related effect as the tax penalty on dividends should have left so small a trace in the data further empir-i- cal I-esear-ch seerned unpromising barring new sources of data (perhaps from foreign countries) 01- new and more polver-ful methods of data analysis Within the last few years studies claiming just such improvements in methodology or data have appeared (eg Long 1978 Litzenbel-gel- and Ramaslvamy 1979 1980 198 1 Rosenber-g and Marathe 1979 Stone and Bal-tter 1979 Banz 1980 Blurne 1980 GOI-don and Bradford 1980 Hess 1980 1982 Morgan 1980n 1980h) with some but by no means all reporting significant yield-related tax effects

LVe will not here review these many studies in detail 0u1- purpose rather is to warn against accepting as tax effects the yield-related effects I-epol-ted in those studies using shor-t-run definitions of dividend yield-that is in tests seeking to deduce the differential tax burden on dividends over long-term capital gains from differences in rates of r-eturn on shares that do and shares that do not pay a cash dividend during the r-eturn inter-val Tests employing such shol-t-run definitions of dividend yield ar-e inappropriate for that purpose Ex- dividend day returns that reflected the long-term tax differential ~vould irnply substantial profit opportunities in short-term trading around ex-dividend days particular-ly for brokers and dealers in securities The cum-ex pr-ice differentials that maintain market equilibr-ium and keep such profit opportunities fr-om arising obliter- ate the traces of the long-run tax differential that the tests ~vith short-run yield definitions seek to measure

llthough tests lvith shor-t-I-un measures of expected dividend yield cannot reliably calibrate the effects of differences in the tax treatment of dividends and long-term capital gains they can and frequently do turn up lvhat appear- to be significant yield-related differences in rates of return Kesear-chers reporting such differences ho~vevel- must recognize that they have a problem not a solution T h e challenge is to account for them For- the particular shol-t-run measures considered here the yield-related effects found in some tests are traced to biases

one of a fairly subtle kind introduced by dividend announcement effects

T h e plan of the paper is as follolvs Section I1 descr-ibes ou r method of testing for yield-I-elated tax effects-essentially tests of an after-tax capital-asset pricing model (CAIPM) using the familial- Fama-MacBeth (1973) rnethod of tirne-series pooling of cr-oss-sectional coefficients (but applied to individual stocks I-ather- than portfolios) T h e prob- lems of defining the dividend-yield variable are then briefly explored along with some discussion of the properties of estimates based on short-I-un measures of yield Section I11 presents the estimates of yield-related tax effects for- several alter-native shol-t-I-un yield mea-sures T h e estimates most of ~vhich seem to imply substantial tax effects are sensitive to the choice of dividend variable largely lve argue because the shor-t-run measur-es are distorted to different degrees by dividend information effects Purging these measures of dividend yield of infol-mation effects gives estimates of yield-related tax effects that are both statistically and economically insignificant T h e next two sections show that this failure to find significant tax effects -ith oul- purged yield rneasur-es cannot plausibly be attributed to the inadequacies in our controls for risk (Sec IV) nor to our neglect of nonlinear clientele effects (Sec V) Nor should this failure be regarded as surprising Ye sholv in oul- concluding Section VI lvhy tests relying on shor-t-run responses to dividend payrnents cannot be expected to PI-ovide reliable estimates of the effects of a tax penalty on dividends over long-term capital gains Whether such a tax effect if it exists can be detected lvith yield rneasur-es reflecting long-run dividend policies rathel- than shol-t-run payrnents remains an open ques- tlon

11 The Dividend Coefficient and the Dividend Variable

Ye seek to estimate the dividend coefficient u in the I-egl-ession equation

where R i is the I-ate of return on share i dur-ing period t Rn is the riskless I-ate of interest dur-ing period t hit is the estimated beta or- systematic risk coefficient for stock i for period t and (litis an estimate of the dividend yield of stock i in period t We estimate equation ( 1 ) by the now familiar- three-step pooled cr-oss-section and time-series approach of Farna and MacBeth (1973) and Farna (1976) Fir-st the risk coefficient beta is estimated from a market model regression of the form

over- the 60 months pr-evious to the test month t For- month t the risk coefficient L i t and an estimate of the dividend yield for each company are then treated as independent ariables in a cross-sectional multiple I-egr-ession of the form

This step is repeated month by month with the estimated risk and di~idend-yield ~ar-iables updated each tirne In the final step the coefficient (I is estimated as the sarnple mean (Iof the monthly cr-oss-section regression coefficients tiT h e standard el-ror- of the estimate is computed as u f l wher-e u is the standard deviation of the time series of ri and t is the number- of months in the sample

Unlike Farna and MacBeth (1973) Black and Scholes (1974) or Farna (1976) we apply the three-step rnethod using individual company data instead of pol-tfolios or- gr-ouped data We do this to sim- plify cornparison of various dividend measures and not from any belief that tests with ungr-ouped data ar-e necessar-ily mor-e efficient than tests ~vith gr-ouped data

Readers ar-e ~varned that the distr-ibutional assumptions justifying the three-step procedur-e are not always lvell appr-oximated T h e coefficients ci for- example are not generally distributed inde-pendently and identically over- the sarnple period and their distribu- tions ar-e sometimes asymmetr-ic and fat-tailed Nor- does the sequence of coefficients ci which can be interpr-eted as the returns or1 a zero beta portfolio allvays have a zero beta LVhen the sequence of coefficients A varies systematically with the market Black and Scholes (1974) have suggested a fourth step in tvhich the sequence of coefficients il is regressed on the market excess rate of return At appropriate points in the discussion we ~vill take this fourth step as veil In the first part of the analysis holvever 12e stipulate the statisti- cal assumptions as the lawyers might put it and concentrate on the dividend var-iable

T h e dividend coefficient Z if positive and lvithin a reasonable range (positive but less than 6)is often interpr-eted as an implicit tax bl-acket or tax differ-ential This interpretation can be Justified I-igorously in a CAPM framework by assurning investors nlaxirnize expected after-tax returns subject to the standard constraints Br-en- nan (1970) formulated such an after-tax CAPM relating expected befhre-tax returns to I-isk Pi 2nd dividend yield of the following form

1 I 2 2 ]OYIISL O F P O I I T I ( A I PCONOLfl

where T the coefficient of the dividend term is a weighted-average marginal tax differential of dividends over capital gains The Bren- nan model (and 1itzenbergel- and Ramaslvarnys [I9791 genel-alization of it) irnplies that the dividend coefficient has the same value for all shares This prediction was tested by Hess (1980) who finds it not descriptive of the data Despite this evidence of rnisspecification of afte1-tax models such as (4) and its variants the conventional reference to Z3 as an irnplicit tax bracket o r a tax differential will be maintained at least until Ive present alternative interpretations

The appropriate measure of dividend yield in tests for- tax o r other yield effects is by no means clear T h e underlying valuation rnodels call for t measure of the markets expectation of future dividend yield But over hat horizon is that expectation to be measur-ed Is it a one-step-ahead forecast O r is it a forecast of the aver-age dividend yield that might be r-ealized from holding the share over a considel-a- Ide pel-iod of time And if the latter- how long

Rernemher- also that the tax differential bet~veen dividends and capitill gains is itself a function of the expected length of the holding period Does the market accrue the tax differential adjusting returns Inore or less evenly over time 01-is the adjustment for tax effects concent~ated mainly at times when dividends ar-e paid and tax liabilities incurl-ed If the former a long-run measure of expected dividend yield is the appr-opr-iate dividend variable Black and Scholes (1974) for example took as their yield measure the r-ealized dividend yield of po~tfolios selected by ranking securities by the sum of dividends per share paid during the previous year- divided by the price per share at the end of the year Their ranking variable is thus a way albeit a simple one to approximate the average annual dividend yield expected by someone ~ v h o bought one of their portfolios at the start of the year and planned to hold it for a year 01-more

Blick and Scholess failure to find significant yield-I-elated tax effects ~vith their long-I-un variable led researchers to try the short-term ippl-oach by focusing on returns in and around the actual ex-dividend dates T h e appropriate expected yield for- the test equations

Ihc ueighth ilc the risk tolerince of each investol- ~elitive to the total risk toler- intr If intiividu~lc uc ~surned to h~ve utilitl functions with constant 1-elative risk ielion rhcn tllc weights beco~ne propo~tionl to market holdings Xiodel of equilil~~ium that d o not imply the same tax differential fol- all v~lu~tion

lii~-ehhaxe 1-ecently been dexeloped by (onstantinides (1980) and b) Li t~enberge~ and Kilnixam ( 1980)

is then not sorne long-run aver-age but only the dividend yield if any expected by the rnarket during the nest return interval

If the return interval is a month as here and in most of the cited studies of yield effect^^ the expected dividend yield for about t~vo- thirds of the firms in the sample will be zero because the vast bulk of dividend-paying firms folloi a quarterly payment cycle Each monthly slope coefficient a in the second-pass regression will corn- bine t~vo sources of variation in monthly dividend yields The first is the cross-sectional variation in dividend yields among those firms expected to go ex dividend during the month If a tax effect exists the conditional mean returns of the high-yield ex-dividend firms ~vill be higher than for the lower-yield firms X scatter plot for these ex-dividend firms would then be upward sloping T h e second source of variation in yields is that between the ex-dividend firms as a group and the non-ex firms T h e returns of the non-ex firms ~vill lie along a vertical line through the origin (or- more precisely through the negative of the riskless rate since most tests are run ~vith the variables in premium form) T h e location of the mean return of the non-ex firms (about t~vo-thirds of the sample each month) ~vill thus have substantial Iveight in determining the intercept of each monthly regression (and hence of course also indirectly the slope coefficients a)

Including both groups I-ather than only those expected to go ex dividend increases the effective range of the critical yield variable and thereby presumably also the efficiency of the estimates of any yield- related tax effects Llhether these hopes for greater efficiency can in fact be realized by the short-run approach or ~vhether they are th~varted by other disadvantages will be our principal concern in the sections to follo~v

111 Estimates of Yield-related Tax Effects under Alternative Short-Run Definitions of Yield

Even if Ie accept the logic of the short-run approach the markets expectations of the cash dividends to be paid in month t must still be specified In obvious first approximation is the actual dividend payment in month t-that is assume the actual dividend payment

Some dividend tudie use dtil data though in a test fill-mat sotnewhit differerlt fro111 the three- or four-pass procedut-e usetl here (see eg Elton and GI-uber 1970 Black and Scholes 1973 Kalay 1977) Orle (Blurne 1980) uses a cluartrrl t-eturri i n t e n d (and a long-run diidrntl rneasut-e) precisely to meat- o e r the curn-ex diffrr- entials that rnotiiate the hol-t-term yield approach I tartel- plot o f tnorlthlv excess returrl tgainst dividend jielda for a cornbined

salnple of ex-divitlerltl and non-ex-tlividet~tl firlns is ho-n in fig 1 Rcadcrs at-r a~-nedhoeet- that the ptctut-e is intended at this poirlt orllv to pol-tray the two source5 of variation analvsis of t tu t scatter must he defetred

equaled the expected dividend payment Such an assumption will not be uni-ersally true of course for dividend surpl-ises d o occur but a m o n ~ h is a short forecasting interval aftel- all and as Hess (1980) points out the perfect-foresight estimate can at least be consitle~-ed one bound on the set of pel-missible approximations Certainly if no dividend effect turned u p ~vith this approximation there ~vould be little point in going further

Thr Ppfr(t-Forp~ght D ~ j n ~ t l o n Dzuzdpnd Yzeldof

Ihe estimated tax effect coefficients obtained under this definition of dividend yield are sho~vn in the first rol of table 1 For the pel-iod 1940-78-~vhich for reasons noted later is the longest benchmark period over lhich the effects of the different dividend variables can be compared-the coefficient aturns out to be 317 with a t-ratio of no less than 102 Thus at first sight the case for the shol-t-run approach seems amply vindicated T h e coefficients are not only in the plausible range for a tax effect but appear estimated ~vith great precision

These appearances may be deceptive however because fol-ces at ~vor-kin the data impart an upward bias to the dividend coefficient Bet~veen 30 and 40 percent of the firms going ex dividend in month t

Ieel-1-eird 1 l K) 11~o11~li1 tliitlcnd ieltl

1)ivitlelrtl ieltl of 12 rnotlth ago ( n o 1-tg~rtlto - ( t i i t l rnd t n o l ~ t h )

Positi c l K di i t ie t~t i )ieltlu r t to d iv i t l e~~t l ielcls ot 3 6 tnotltlis igo

1K tliitlentl itltl et to I i t poiti c (ck-di ident l ~nonr l i ) othet-uisr 0

gt I F - t - t l t ~ r p i r c ~ t h c w I ~ ~ I I I I I I I I 111 u w lt I 15

R - K= 0 -ltgtA + i 1- R) + ( 1 1

lt i ~a i -I lt ~ I I lt I l ~t i A I ~ ~ h c Krw+rlth1 1 S C I L I ~ I t ~ p r ~ C I C lt i ~ ~ d c n d f r ~ ~ (cntcr f o r F ~ I L ~ I(KSPI ~ t ~ u n t h l

also declared during the same month t Considel- t~vo such stocks and suppose that at the end of the PI-evious month the markets expected dividend Ivas $100 for each On the basis of this expectation assume the price of each has been set at $50 per share Suppose one firm increases its dividend to $150 a share and the other cuts its dividend to $050 a share The first firm ill now be I-ecol-ded as having a dividend yield of 3 percent and the second of only 1 percent But the high-yield firm will also have a highel- realized rate of return for the month if as is often the case the market interprets the unexpected increase in dividends as signaling improved earnings PI-ospects for the company By the same token the company ~vith the disappointing cut in dividends may find its return for the month dragged do~vn by a fall in the end-of-month price that reflects the markets do~vnward I-evision of the companys future earnings Errors in dividend antici- pations and realized rates of return ~vill thus tend to be positively con-elated

T h r Lztzmhrgrr-Ramaculam3 Variable

1-itzenberger and Ramas~vamy (1979) sought to eliminate this information bias by exploiting an additional piece of information on the CRSP tape to wit the declaration date On the basis of the repol-ted dividend declai-ation date they defined a revised dividend variable as follovs ( I ) If a firm declared prior to month t and vent ex dividend in month t (about 60 percent of those going ex dividend in t ) then the expected dividend yield rL Ivas computed using the actual dividend paid in I D divided by the price at the end of month t - 1 (2) I f the firm 1151th declared and vent ex dividend in the same month t then d i t vas computed using the last previous I-egulal- dividend (looking back u p to one year) If the back~val-d search yields no such 1-egular dividend or if the dividend Ivas an extra dividend then (Iitis set equal to zero (about 10 percent of those declaring and paying in t ) Ye call this definition of the expected dividend yield the level-revised (1-R) dividend variable

T h e results obtained ~ith this dividend variable are shoivn in line 2 of table 1 Note the large response to this seemingly minor technical adjustment T h e coefficient drops substantially from 317 to 179 or by nearly 45 percent though it remains highly sigrlificant and still within the plausible range for a tax effect

Out- e5tirnate is somewhat lolver thari that repot-tetl h Lirzenberger ant1 Ralna- a11i (1979) lvhose sample peritxl is 5olnelvhat different (1936-57 I-arher than 1040-78) te start f t-o~n 1040 because the GRSP tape lists few cleclaration dates PI-ior to ttiit eat Hence cort-ections cannot be made (either hv the LR method or others to he considered below) fol- most of the cases in which the declaration date and ex date are in

Exploiting the declaration date in this fashion reduces the up~vard bias in Nj but does not eliminate it One group of firms still remains for ~vhich the 1R variable is the actual dividend yield even though the firms have declared and paid in the same month These are the firms that might have paid a dividend in month t on their regular quarterly schedule but whose directors at some time during the month have votccl to omit dividends T h e CRSP tape will report correctly no dividends paid during the month and no date on ~vhich a dividend Ivas declared But absence of a dividend declaration during the month is information not available to the market at the beginning of the month And as the old story goes there may be an important clue in knolving that a dog did not bark

-10 see ho~ overlooking this clue can bias regressions employing the LR dividend variable consider again the two similar firms for ~vhich the markets expected divitlencl was $100 per share Suppose that for each firm separately the markets consensus probability Ivas 5 that the firm ~vould declare a dividend of $200 and 5 that the firm would ornit the dividencl Suppose further that the price of the shares post announcement will double if the dividend announced is $2 and will halve if the dividend is omitted If the initial price for each firm Ivere say $10 the realized rate of return ~oulcl be 120 percent ~vhen a $200 dividencl is announced and -50 percent hen the dividend is set to zero ~hich implies an ex ante (and of course also average ex post) rate of return of 35 percent The ex ante expected dividend ield for both shares is 10 percent Regressions of realized returns on expected dividend yields for such ex ante matched pairs ~oulcl show no relation bet~veen the t~vo variables

Suppose however that in those regressions the 1R variable had been used instead of the markets expected dividend yield And suppose for concreteness that the most recent regular dividend ~vithin the last l h m o n t h s had been $200 for each firm A firm noI an- nouncing a clividenci of $200 in month t would be recorded as having a dividend yield of 20 percent and a realized return of 120 percent A firm passing its dividencl would have a realized return of -50 percent but a dividend yield of zero under the LK definition Thus the regressions ~ith the LK variable would show hat appears to be a positive association between returns and dividend yields even though returns are actually independent of correctly measured ex ante clivi- dend yields

the me rnonth Tests that Include the PI-e-1940 years ill to that extent still be subject to the i~lfi)rrnatiorl-effect bias deicribed above

rhe 1iis introduced bv le1o-dividend firms under the LR definition can persist for u p to three idditionil ctuartel-gt after the first regular di-idend is passed Those firms thit I-esurne I-egulal- dividends one tvo 01-three quarters later will have higher I-eturns

Inforrtrutzon E f f r c t~ or Tax Effects

That valuable information may be contained merely in the knowledge that a dividend Ivas o r was not paid during a month is strongly suggested by some of the other tests reported in table 1 Note for example that when we use not the firms actual dividend yield but its dividend yield 12 months ago if any no significant coefficient emerges Yet the cross-sectional variation anlong the firnis is just as large and the yields show substantial stability over time But when Ive use ~vhat will presumably be an even more out-of-date predictor namely the dividend yield of 3 years ago Ive get a highly significant positive coefficient provided Ie restrict the nonzero entries to those firms for ~hich the level-revised yield was nonzero in t Even rnore striking is the fact that a significant coefficient is produced ~vithout reference to any specific dollar value for the dividends but merely a dummy variable set equal to one if the IR dividend variable Ivas nonzero

Although these tests are suggestive they cannot by themselves tell Ivhether the ex-dividend month shows u p because of information effects or tax effects But there is a way to find out

Yzeld V u ~ ~ u b l e ~ Utzng Only Dzvzdends Declared zn Advance

Table 2 presents estimates of Z3 for several different yield measures in each of ~vhich the nonzero elements of the vector include dividends paid in month t but only if declared in a month prior to t For these firms at least un~anted information effects of the kind found in the previous definitions are absent Yet the logic underlying the short-run approach is maintained because these firms did go ex dividend in month t7

In the first panel of part A the entry in the yield vector for any firm that both declared and paid its dividend in month t is set to zero This approach seeks to offset the d o ~ v n ~ a ~ - d pull on the intercept by the dogs that did not bark with the upward pull from those declaring and paying nonzero dividends during the same month Announce-

and higher- LK dividend vields than those othetwise coriipatable firrna that disap- pointed the tnarket bv not resuming their regular tiiviciendi during the 12-month inter-a1 following the cut

Table 2 pr-esents estimates both for the entire sample per-iod 1940-78 and tot four- subper-iods (I-eat caution should be taken in rnterpretrng the subperiod results lvhrch ire presented only to grve some idea of the substantial -ariabrlitv in the divrdend coefficients over- time The patterns of subperiod coefficients appear unr-elated to changes in the tax law or to the downward drift in the weighted-average rnarginal tax bracket (see n 1 ) that would presuniablv have accompanied the steady-increase in the pr-opal-tion of stocks held bv tax-exempt institutions over the sample peritxl

k l 1 h l l k 1 j 1 4 1 Fkkk(l (OkPbl( I F X I F O R ViRIOLS DIFIXIIIO O F IxlP( -151)

DI I I ) P X I ) Y I P I1) Il)lr I P ~ FOK EFPL(I X F O K I I ~ I I O N I S

Definition ( la) Divirlentl Yielti Set to Zero if DivlendA

Not Declared 111-ior to Ex Rfonth ( O f r = (i)

Definition ( 1b) Nonzero Uividencl Pa)ers Sot Xntlouncecl il Xdy~nce

Excluded Z ~ I - o Dividend Pa)ers Kept (tl = (17)

rnent effects should net to zero if all announcements are included T h e disadvantage of the approach is that setting a nonzero dividend to zero creates an el-rors-in-va1iables problenl that attenuates the estimate of (1 But the degree of attenuation from this source is known-the bias tolvard zero can be shown to be approxirnately the p~mpol-tionof firms whose yields are changed (about 40 percent over the sample period)-and can be taken into account before conclusions are drawn

Ascan be seen from the table however interpretations are unlikely to be much affected by adjustrnents for attenuation T h e dividend coefficient for the entire sample period 1940-78 is close to zero ( - O268) and statistically insignificant

T h e tests in the second panel differ frorn those just mentioned by eliminating firrns both declaring and paying dividends in month t

- - - Year (I I 11 a$

Definition 1 lc) Onl) F~I-rnwith divide tic^ Decl~redin

-Idvatice Included ( t l = 11 )

rather than shifting then1 to the zero dividend group T h e attenuation bias of ( l a ) is thereby reduced in ( lb) but at the cost of rernoving the offset to the net negative information effects in the zero dividend group As expected the estirnated Z3for the overall period 0368 is indeed higher than in the first panel but still far too small to be considered a significant tax effect

T h e third panel of table 2 goes a step furthe1 and drops frorn the sample all firms except those that both paid dividends in t and announced thern in advance This definition avoids the biases in the first two sets of tests but sacrifices any contribution to the precision of

estimating the tax effect that might come from including both ex-dividend and non-ex-dividend shares in the same sample Iglance at this third panel shows that when the known biases are

removed in this fashion no relation remains between excess returns and dividend yields for the ex-dividend firms T h e point estimate of the coefficient U3 falls to an economically negligible and statistically insignificant value of 0135 Thus if yield-related effects d o exist they clearly cannot be attributed to differences in the dividend yields of the firms actually paying dividends in the month

Part B of table 2 presents a test that combines advantages of some of the previous approaches but in a more efficient way Rather than thl-ow auay observations (as with [Ib] and [Ic]) o r change their values (as 12ith [ la]) the test uses a slope dummy to estimate the dividend coefficient separately for those ex-dividend firms that did and that did not declare in advance T h e dummy variable is set to unity if the dividend is declared in advance and zero otherwise Hence the sum of the two coefficients measures the effect of dividend yields on returns net of direct announcement effects For the overall sample period the coefficient is negative and not statistically significant though somewhat larger in absolute value than in part A ( l a ) T h e last column of ])art B is an adjustment (see the discussion in Black and Scholes 119741) to allow for the correlation between the monthly dividend coefficients and the market return coefficients ( I This coi-rection furthe1 reduces the size and significance of the dividend coefficient

Summing up to this point we find the relation between returns and dividend yields to be sensitive to the definition of dividend yield T h e

LVe helieve that the dumrn) variable approach using all the data is a safer and mol-c ~-eliat)lelvav of utiliring the non-ex firms that1 tl-ying to remove information effects from that group I thl-oing some away T h e dange1- is that the non-ex firnms retained tna be a n unrepresentatiw sarnple of ~ero-dividend firms and ma bias the regression (oefht ient isan example of how biases miy inadve1-tently creep in J-hen some but riot 111 zel-o-dividend firms ire retained consider the seemingly harmless step (proposed in 1ivenl)erger and Rarnaswarny (1981 p 91) of including as zero in the expected yield vector fo1- rnorlth t onl those firms that went ex in month t - 1 and for which therefore the 1na1-ket could reasonably presume n o dividend would be paid in t T h e tl-ouhle is holvever that those firms currently or permanently following a no-dividend policy vould be excluded from the ze1-o-dividend group These excluded no-dividend firrns tend to be ni~ll firms arid given the so-called small-firm effect (see Banr 1981 K e i ~ i ~ ~ n u m1981) they Lvill also tend to have above-average excess returns T h e 7ero- dividend firms going ex in t - 1 and not excluded will thus tend to have lower than average excess I-eturns and will thereby impart an upward twist presumably unrelated to dividend tax effects to the regression coefficient

I n this test the coefficients ci and u2 are coristrairied to be the same for all firms ~vhich is ce1-tainly reasonable it1 this context $Ye have also rut1 the tests with separate slope and inte1-cept durnrnies to free those coefficients as well Ihe results fol- the dividend coefficienta are essentiall the same

differences in estimated yield effects appear to reflect differences in the degree to which the various short-run measures of expected dividend yield introduce unrvanted information effects After correcting those measures for their information effects 12e find no significant remaining relation between returns and expected dividend yields-certainly nothing that could be considered a yield-related tax effect of the classic kind Before we seek to explain this finding prudence suggests a check for other deficiencies in the testing proce- dure that may be obscuring a tax effect

IV Possible Biases from Inadequate Risk Corrections

The tests of table 2 have been designed to remove information effects but other biases affecting the dividend coefficient may still remain1deg Dividend yields for example may be proxying for risk If dividend yields tend to be lo~zer for high-risk f i ~ m s as seems likely and if risk is measured 12ith error as seems even more likely the dividend coefficient in a cross-sectional multiple regression of the kind in tables 1 and 2 will be given a negative twist (at least over those sample periods in which the market return is positive)

A tuist of the opposite kind occurs horvever to the extent that firms adjust their dividends slo~zly to changes in their earnings prospects In the months follorving the announcement of good nelvs a firms price pel share will tend to be higher and its true beta tend to be lower (because of leverage and option effects if nothing else) than in the months before the announcement If at the same time the firm maintains its dividend 01increases it only gradually its dividend yield 12ill be smaller than before For these good nervs firms low dividend yields will appeal to be associated rvith abno~rnally low returns in the months after the announcement because the measured beta based mainly on preannouncernent months rvill be too high rvhile for firms maintaining their dividends in the face of bad news high dividend yields will be associated lvith high abnormal returns

Correcting for biases of these kinds is more difficult than for the information effects of the previous section Since these biases arise in part out of nonstationarities and misspecifications in the underlying

I klan) ofthe isues cotlidered in thi section are tl-eated in KI-eater detail in Hess ( 1980)

Bias due to C~I-relation in beta cannot of dividend )ields with 1neaiureInerlt e~-I-01-s he el~rninated rnerel) by taking palns to assr11-e that the measures of expected dividenti vielcia in the tet use only pat data I n fact extr~polative measure of expected dividends (of the hind proposed say in hlorgan [1980n 198061 or Litzenhe~-ger and Kamis~~my[1981 pp 7-81) a ~ - elikely to itlcreaie the bias Extrapolative measu1-e t)y their nature ill give the appearance of dividenci 5tat)ilizationVof the kind ciescriheci even whet1 the firm has actually adapted its dividend to the change in its circu~~~starlces

R DLFINII l o 1 l d ) WITH l ( p r - l ) - Iuu~I )$

pl-ocesses describing returns Tve cannot rely on ordinary errors-in- variables approaches (such as those proposed eg in 1itzenbel-ger and Rarnasarny [1979]) But Tve can at least check the sensitivity of the dividend coefficient Z to variations in the risk measures

Part A of table 3 sholvs the effects of a risk measure even worse than the stlndard first-pass 5-year b i t used in tables 1 and 2 to wit none at all (For simplicity the I-esults are presented only for the dummy variable test corresponding to part B of table 2 since that test is the most efficient) Note that for the overall sample period the value of the dividend coefficient is little changed

Finding a better risk proxy is not as simple as finding a Tvorse one But the search for risk proxies need not be restricted to single- variable measures such as b i The discussion of the biases above suggests that the current end-of-previous-month price p i - may

-

Dividends Set to Zero

(l~entcle ctuil LC el-revised If S o t Declared (~oup Diidend Piid Yield ariahle in hdvince

contain additional and more recent information about the firms risk and prospects than the 5-year h i t Part B of table 3 shows the effect of iddirlg that variable-actually its reciprocal lpit-l-to make its scal- ing comparable to that of the dividend-yield variable Note that it does indeed contribute significantly (more so in fact than hi itself) But the key dividend coefficient remains as small and insignificant as before

V The Results for Subgroups

Significant tax effects i11 the aggregate sample may perhaps have been obscured by nonlinear clientele effects of the kind discussed in Lit- zenberger and Karnas~varny (1980)As a check ive present tests similar to theirs in ~vhich (I is estimated separately under various short-run

1 he vat-iahle l ip can alao provide a stt-iking illustration of o u r warning (aee nn 8 1 1 ) that short-run measurea of expected dividend yield even vhen put-ged ot alitlouliCelilerlt effecta 1111 still lead to biased estimates of yield-rclated tax effecta That ai-iahle aftet- all can be conaidered a measure of expected dividend yield in rxhich the fot-ecasted dividend of every firm next period is taken to be $1 Such a fot-ecast is crude but is at leajt based entirely on past infot-mation and hence is ft-ee from innouncement effects o t the kind considered earlier For all its crudeneaa as a forecast however l p - haa a positive and significant coefficient hen uaed aa the meaaut-e of expected dividend yield in tests for tax effecta Since the numerator is a constant the ignihcant positie ~ a l u e of the coefficient can come only from the information about the fit-ms future prospects additional to that in bi conveyed by the current pt-ice in the denonlinator All) other measure of dividend kield with p i - in the denominatot- auch is those in Litzenberger and Ratndswamy (1981) or in Auet-hach (1981 esp pp 16-18) lvill ot cout-se be subject to the same ~1p~vard bias

definitions of dividend yield for subgroups of firms ranked by past average dividend yield In the tests shown in table 4 the ranking is bv the Black-Scholes yield-r anking variable (previous year dividends divided by end-of-year price) updated ~nonthly Group I contains the 20 percent of firms ranked loivest by this measure group 11 the next 20 percent and so o n to group V the highest

The estilr~ates ofz for the subgroups in table 4 appear to be no less sensitive to the defi~lition of dividend yield than the overall satnple estiliiates considered earlier But note that for these tests correcting for information bias (last column) by the method in A(1a) of table 2 now does not reduce all the dividend coefficients t o insignificance The coefficient for the loivest yield group remains positive and significant even when the nonzero components of the yield variable include only dividends declared in advance

A closer look at the underlying data however raises doubts about the reliability of the estimate for the loiv-yield group T h e cumulants fractiles and other relevant sample statistics of the 468 monthly cross-sectional regression coefficients G for the lo~vest dividend-yield group are given in table 3 Note that the observations cluster very tightly around the modal value of zero A number of extreme outliers are present hoivever at both ends of the scale The highest estimated tax bracket is the 1333 percent reached in October 1968 and the lowest is the -1041 percent in January 197 1 Not only are the positive outliers larger but they are more numerous T h e right ske~v- ness in the distribution is apparent in the fractiles reported in table 3 Sote in particular hoiv far the median is below the mean-098 as co111pired ivith 373

Ample justification exists for throwing out these extrerne observations in computing an average tax effect for the period as a ~vhole But the case becomes more compelling given the underlying observations o n yields and returns that produced the extrerne coefficients Figure 1 for example sho~vs the scatter plot of the relation bet~veen returns and yields for the lo~vest yield group in the month April 1967-another month ~vith estimated ri gt 15 Table 6 shoivs the nunlerical values for the first 40 observations ranked by size of dividend yield Note first that only 16 of the 178 firms in the group had nonzero dividend yields-zero yield being entered not at zero but at - 0042 after subtracting the estimated riskless rate of interest Thus in the scatter plot of figure 1 91 percent of the observations lie along the vertical straight line through the -0042 point on the dividend-yield axis Of the 16 firrns ivith nonzero dividends t~vo happened to have extremely large increases in price during the month-one ivith a return of over 60 percent and one ~vith a return of 228 percent These two points alone account for the sign size and apparent

Dividend Yield (minus riskless rate)

11( 1 -Extrs returns o11 diitiend iel(i fot- lowest divitientl-iellti group i l l Apt-il 1967

tui-11 out to be largely responsible for the seemingly significant value of for the loest yield group T h e sensitivity o f the sample mean to the extreme values ofri is shotvn in table 7 When rve chop off the 33 cases or- 7 percent tith cross-sectio~~al regression coefficients greater than 3 in absolute value the sarnple mean falls from 373 tvith a t-value of 28 to 098 ~ith a t-value of only 1O If e chop a further 32 rvhose coefficients fall outside a range of plus or tninus 4 the estimated implicit tax bracket falls to 036 ~vith a t-value of 044

In principle we might go on to examine the observations in the other four subgroups But hy bother Subgroup tests of the kind in table 4 and some other recent studies are too inefficient and u~lreliableeven apart from their vul~~erability to outliers to throw

l3 ctudlly of C O L I ~ S C we did examine the other groups and filund as might be expected that the outlier prollem was less severe because nlol-e dividend entries were nonLero in tho5e higher-yield groups in the typical month T h e coefficients however $ere still jensitive to trimming

--

0I)e1 itiori Ketur-n Dilidend Yield Milill Rikless R ~ t e

an)- light on the issues in dispute T h e ~vithin-group estimates of tax brackets reflect only ~vhat little variation in short-run dividend yields remains after the substantial bet~veen-group variation in average yields has beer1 removed I n fact if the preclassificatio11 ~vere cotn-pletely successful the within-group variation ~vould be mostly noise4

rile uhstantial negative coefficient that emerges for the highest yield portfolio in table 4 aftel- announcement effects are eliminated remains a puzzle Cum-ex tl-ading b

--

JOL UK I O F POI I-II(IF ( O S O X f Y

- --

hle~nof SD of Stantiat ti ltilueof Nurnl~e~- hlonthly hlonthlv Errol ofof

Inclutied Ohsel- ations ~lues Values hlean 1-Ratio

An insti-unlent so unreliable permits no firm conclusiorls about the existence o r absence of tax clienteles But we can say at least that the subgroup tests presented provide no grounds for suspecting that important yield-related tax effects in our aggregate sar-nple are obscured by fitting a straight line to a nonlinear relation

VI Why Tax Effects Should Not Be Expected in Tests Using Short-Run Measures of Expected Dividend Yield

Ye recognize as did Black and Scholes (1974) that our f a1 1ui-e to detect yield-related tax effects leaves something of a puzzle Why does i tax effect that seems so plausible to so many on a pi-iori grounds appear to leave no detectable track in the data In the case of the short-run dividend r-neasures at least Tve believe the puzzle 111ay have it relatively simple solution Recall that the presumed tax effect ~vi th that yield variable is essentially the average difference in rates of retui-11on those shares that go ex dividend in a given month and those that d o not Some or- all of this differelice is supposedly the equalizing differential necessary to r-nake those o~vning the shares at the start of the month indifferelit bet-een cont i~ iu i~ ig to hold the shares (and paying full tax 011 the forthcoming dividend) atid selling the shares

corporate il~vestors might conceivabl be responsible But a mot-e likel) candidate is the pl-o- bit discussed earlier Fol-tunately no great urgency attaches to resolling the 11u~~le I e show in the next section tests of the kind in tiible 4 could not shed any I-cli~hlelight on tax-related clienteles even if their econometric deficiencies Lvere re- p~ircd

I highl tionlineal in fhct U-shaped I-elation between I-eturns and yields is repol-tcd 11) Hlunle (1980)although in tests using a long-run rneasure of dividend yield r-ttllel th111 t l ~ e sllol~t-lun ~neaiures considered here Later work by Keirn (1982) hocel- uggests that Klurnes yield effects are confounded ith the so-called small-firm effect After one controls for sire of firm the nonlinear yield effects largely vlt~nislgt

curn dividend (and payi~ig tax 011 the ir-nplicit divide~id at the lo~ig- term capital gains tax rate) As Elton and Gruber- (1970) have sho~vn if the capital gains tax rate is less than the rate on dividends the price fall from the last cum-dividend day to the first ex-dividend day rill be less than the dividend paid T h e (risk-acl-lusted) rate of return on the ex-divide~id day (and by extension on the months co~ltaining the ex-dividend davs) ~vill thus be higher- than on no11-ex days a ~ i d ~ n o ~ l t h s

This plausible a ~ i d oft-i~ivoked expla~iat io~i fhtalhas ho~vever a flaiv If the price falloff on the ex-dividend day Yere really less than the dividend (after correcti~ig for risk) then shor-t-term traders buy- ing cum-divide~id shares and selling them ex dividend vould earn above-1lorrna1 profits Remember that for such in-and-out traders (and also for tax-exennpt institutions) the tax rate on dividends and 011

capital gains-in this tr-ansactio~i actually capital losses-is precisely the sa~ne

Short-terrn traders can be expected to dor-ni~iate the short-ter-111 equilibrium and hence to eliminate the presumed cor-npensati~lg curn-ex differe~itial In principle every i~ivestor taxable o r not can exploit that profit opportunity from short-term trading ~vher-eas the potential sellers are only those taxable investor-s ~ v h o happen to ovn the shares (and are ~ i o t locked i11 by holding period I-estr-ictio~is o r by potential taxes on past unrealized gains) Some individual taxpayers terripted to exploit the profit opportunities in shor-t-term cum-ex tr-ading may find therriselves constr-ained by the capital loss limitations and the rvash-sale rules But these provisio~ls d o not apply to brokers or dealers in securities For such brokers and dealers organized as corporations moreover- and for cor-porate traders generally (including casualty insur-a~ice companies) the profit pote~itial in short-ter-~n curn-ex trading is further enlarged by the exclusio~i of 85 per-cent of divide~ids received from any taxable US corpor-atio~i fro111 taxable cor-porate profits

TI-ansactions costs rei~lforce the dominance of short-terrri buyers in setti~lg the equilibrium cum-ex differential T h e round-tr-ip costs faced by pote~itial sellers among the taxable i~ivestors a re substantially larger than those i~ icur red by the brokers and dealers s t a~ id i~ ig ready to capitalize o11 deviations fro111 the short-run tr-ading ecjuilibriurn

T h e presence of tr-ansactio~is costs even the comparatively small inside ~na rke t costs of the 111-okers and dealers may cell keep the ex-divide~id pr-ice from al~vays falling by the full amount of the divi-

1 nurnber of notably Kala) (1977) have pointed out this flaw in the i~-ite~s Eltot1-(1 uhel- I-casoning In fhct the point appears already to have reaclled the textbook Iccl-iitnes thc discussion in Copeland and LVeston (1979 p 353)

denct And since transactio~is costs are roughly propor-tiorla1 to price the obser-ved price falloff may well be smaller relative to the dividend yield for lolv- than for- high-yield shares exactly as the tax-clientele model seems t o predict But such yield-related effects are 11ot tax effects or at least not tax effects reflecti~ig the presurried tax penaltv on dividends over lo~lg-term capital gains that the shooti~ig is all aboutI7

Solving the puzzle for- the shor-t-run measures of dividend yield still leaves unanslvered why yield-related tax effects have ~ i o t been con- vinci~igly delrro~istrated i11 tests using long-run measures like those of Black and Scholes (1974) Per-haps the explanatio~i is that yield- related tax effects d o not accrue ~t lonth by r r~o~ l th as assumed im- plicitly in many standard after--tax rnodels of asset pr-icing They car1 sho~vup i11 pri~lciple o1i1y i11 ex r-nonths and there they are elir-ninated

short-r-un tax trading Or- perhaps month-by-1~011th accruals of tax-related differences i11 retur-11s could arise but are eli~ninated by supply acjustnients as described i11 Black and Scholes (1974) or bv tax shelteri~lgas i11 hliller and Scholes (1978) C)r perhaps the tax effect does accr-ue rno~ith by rrio~ith but the data ar-e so 11oisy that the tax effect has escaped detectio~i by existing instr-ur-nents 6thich of these explanations is cor-r-ect remains to be seen But lve hope that the search can pr-oceed rrior-e effectively 11orv that -e krlolv at least -here not to look

References

4uethitll rlln I Stocklloltler T a x Rates a11d Fir111 Attrih~~tes orking Paper no 817 Crc~~~ ) l idge Xlaslr Nat Bur E c o ~ i Res 198 1

Binr Rolf Fulthel Evidellce 011 the Existence of the Iiclcl alicl Size kffetts C~iput)lirllcd m a ~ l ~ c ~ c ~ - i p t Chicago U ~ i i Chicago 1C180

lhe Rclatiollship hettcen Return allcl Xlarket Value of (ommoll StoclIFirlcirlricil Eco~r9 (larch 1981) 3- 18

B1acL Fischc1 all([ Scholes X l ~ - o ~ l ReturnsS The Behavior of Sccurit arourld Ex-Divitiend I)as Lll~puhlished inanuscript (hicago Univ (hicigo 1973

The Effetr5 of I)iviclt~lti Yiclcl and I)iitiellti Polic 011 ( o~nrnon Stock Prices and Keturlls Fir~ccrccictl ficor 1 (hiah 1)74) 1-22

Bl~clllc Xla~sllall E Stock Returns and Dividend Yieltls Sorile hlorc Evi- c1cnce Kc11 ficorl cod Sf(rtic 62(Novclllher 1980) 567-77

Brellnarl l l ichacl J 1-axes Slalket Valuation anti Corporactl Financial Pol- ic (it T(ix1 23 (I)eccnll~er 1970) 417-27

(onta~iti~iiclcsGeorge hi Capital hiarket Ecjuilihtiuln ~v i th Pelso~ial Tax orling Paper no 33 (llicago Cniv Chicago Cklltcr Kes Securit Ptice 1980

Iests rrjccti~~g the tax interpretation o f ex-tiividctid clay retul-ns are PI-escnted in 1 0 I-ccctlt stutiies one ~ I IC S data by Hess (1982) atid ollc o n Canadiati data b) laCo~~il~oCalld Verniaelen (1981)

( oplancl I ho~nasE a11tl estor~ I F1cd Filccl~tcicll 7-ro1gt crttcl Co~porcctr Poiir j K e a t i i ~ ~ g h1lss ldciiso~i-eslc 15)79

t l to~~EtlinJ ant1 (rul)e~ h la r t i~ l I 11argi1lal Stockliol(icr I-ilx K a t e ~ n d thc (lir~~tcle Eftecr Kt i Erott (it(( Static 52 (Fet3ruar~ 1970) 68-74

Fallla tugenc F Foztrtclotio~e(fFitecrrtc~rS e ~ cIork Basic 1956 Fl~llli I-r~genc F I I I ~1IacBeth Jnr~les 1) Risk Keru~-11 and lcluilih~iu~~l

Elllpirical Ie5ts JPb 8 1 110 3 (111i]uric 1lt173) 605-36 (ortlon Roger H and BI-atifortl Ilavici F 1-axar io~~and tlie Stocl h 1 a l k ~ ~

Valu~tionof (apital Gains a n d Diitlcntls Theor ant1 Enipirical Results I Plihiir ficore 14 (Octoher 1980) 105)-36

Hc55 P~trick J L)ividclitl Yields lncl Stock Kcturris A lest for lax Fffcc~s P11D tlissr~ririo~~ 1)80Llniv (hicago

Ihe Ex-1)ividelitl L)a Behavior of Stock Returns F~11tlier Eviderlce o n IIs Etkcts J Firtce~ecr 37 (1Ia 1982) 445-36

Ki11 ~ ~ l e r 1-hr Behavior of the Stock Prices o n tlie Ex-l)ividc~id l)n-a Rcexalllinatioli of rhe Clientcle Effcct Ph1) tlissertation C n i Ruches- [el 1977

Keim Dol~ald B t fur the^ Evitierict 011 S i ~ eEffects and Yield E f f c t s 1-he lr~lplicitio~liof Stock Return Seasorialit Cnpublisheti ~ ~ ~ a n u s c r i p t Cliic ago C l ~ i v (hiclgo 1982

1akonisliok J o s e p h and Vermae1cn Theo Tax Keform a ~ l t l Ex-I)iitierid I) Behavior Vorkirig Paper n o 7)0 V a ~ ~ c o u v e s Criiv British (alum-bia Facult Con~niercc allti Bus Atl~nin 198 1

12irre~il)ergerRobert H a n d Ramasivam~Krislina Ilic Effcct of Pc~so~ial l axes and 1)ividends o n Capital isset P~ices 1-heor a~lci E~lipirical Ei- tlenceJ Fiticetrriccl Ecor~ 7 (Juric 1)7)) l(i3-3

Diviticnds Short Selling Restriction Iax-i~ltluceti Ir~vestor (lien- tries and Iiarker Ecpilil)~iurn F i ~ c a t t c ~ 1980) 469-82 35 ( h i a ~

Thc Effects of I)iicierid or1 Cornrno~l Stock Pr ices I ax Effects 01

1nfor111irioll Effect Unpublishetl ~nar~usc~ipt Stanfortl Calif Starifortl U I I ~ - 1981el Io1k Colunrhia Llniv 1)ecember

Long J o h l ~ B ] I The h1arkct Valuation of (ash Divide~ictsA Casc to (o~lsitlcrJ Fitecrtctic~l fi(ort 6 (Jur1ciScpre1nbc1- 15158) 2 3 - 6 4

1lillcr 1lerton H 11it1 Sclioles h11ro11 S Divitlcricis a r~ t l 1-ixes JFitcclrecial F(orl 6 ( I )eccr~~her1978) 333-134

)lorgall Ieuan ( Divitlends a r~ t l Stock Price Behavior i l l (~nlda L11rput)- lislietl ~nar~uscr ip t Kir~gston Olit (2ueeris Cliiv School Bus h l a ~ 1980 ( ( 1 )

l)iitlcntls and (lpital Asset Prices C~lpublisheci ~ ~ i a r ~ u c ~ i p t Kingston (hit Quecns Criiv Scliool Bus J u l ~ 1)80 ( b )

Rei~iganuni hiarc R Iisspecificatio~~ of Capital Asset Pricing Empirical 411omalie Bascd on Earliir~gs1icltis arid 1Iarkct ValucsJ Fitcc~tcciccl Ero~c I)(hiarcli 1981) 1)-46

Koscnhvrg B a u and hiarathe V Iests of Capital Asset Pricir~g H ~ p o t h e - sis Rv Fit tc~tic ~1 (lj79) 113-3

Stone Br1nel1 K and Barttvr B ~ i t t T h e Effect of Dividcl~ti Iicltl o n Stock R e t u r n Er~ipi~ical Eidcrice or1 the Re1cvarlce of Divider~tis Voski~lg Papcr 110 E-76-8 ltlarita (a Georgia Ir~st Tech 1979


Merton H Miller and Myron S Scholes C t i l l ( l t (ij ( ~ I ( ( I ~ o

Ihis plpel r e e s ~ m i n e s some recent tests o f whe the r ho lde r o f slii~es uith higlier d ividend yields receive higher risk-acijusted rates of Ietul n to compensa te f o r tlie I iea ier taxes on d i i d e n d payments tliln o11 long- term capit~l gains O u r particular concern is ~ v i t h tests using r1iolt-run measures of d iv idend yield-that is measures tha t seek to d e d u c e the differenti~l tax b u r d e n on d i idends (gtel- long- tel-xn capital gain5 frorn diffelences in rates of r e t u ~ m o n sli ~res tha t do a n d shares that d o not pay I cash dividend dul-ing tlie r e t u ~ n intel ~l e shov that such meisures ire i nappropr i a t e f o r that pu1p)e Any yield- elated effects associ~ted ~vi t l i such measures 11ll1st I] ire frorn s o u r c e othel tlian tlie long- term tax di f ferent i ~l For thc s l i o ~ t - ~ u n measures considered he re t he yield-related effects lbunt l i n some tests a r e traced to biases o n e o f a fairly subtle k ind i n t rod~cced11v dividend a n n o u n c e m e n t effects

I Introduction

The publication of Black and Scholess (1974) study came after nearly tu0 decades of intense academic controversy over the effects of dividends on stock prices Using the large Center for Research in Security Prices (CKSP) data base and employing new econometl-ic methods

Ineirliel- velsion of this papel- iras prcsenteti at the Conference on Tzisation ~ tthe Unirrsitv of Southel-n C~lifor-nix Lms Angeles Jlnuary 22 1981 Ve ickno~vledge ith thiiiks the comments o f our discusslnt Marc Keinganuni ~ n d of the conference pll-ticiprnts espcciallv Richarti Roll Helpful comments o n earliel- dl-afts -el-e 1-eceived tl-om Fisther Black Geol-ge Constantiniclcs Eugene Farna Robert Harnada Patrick He Rol~crt Holthauscn Jon Ingelsoll hfichael Jensen Richard Leftwich James Loric Kenneth Reid Inti IVilliam Sch~vel-t Our thanks ~lso to the American National Hank and to the Center fol- Kesearch in Security Prices for rupport of thir research to the letel-ce fi)r I nurnlx~ of valuable suggestions and especially to Donalti Keim ti)r help in designing earl-ving out and interpl-eting the empil-ical tests






































R - K= 0 -ltgtA + i 1- R) + ( 1 1





























- - - Year (I I 11 a$






















-
















--




--


- --


















References




































































R - K= 0 -ltgtA + i 1- R) + ( 1 1





























- - - Year (I I 11 a$






















-
















--




--


- --


















References
































































R - K= 0 -ltgtA + i 1- R) + ( 1 1





























- - - Year (I I 11 a$






















-
















--




--


- --


















References



























































R - K= 0 -ltgtA + i 1- R) + ( 1 1





























- - - Year (I I 11 a$






















-
















--




--


- --


















References






















































R - K= 0 -ltgtA + i 1- R) + ( 1 1





























- - - Year (I I 11 a$






















-
















--




--


- --


















References















































R - K= 0 -ltgtA + i 1- R) + ( 1 1





























- - - Year (I I 11 a$






















-
















--




--


- --


















References








































R - K= 0 -ltgtA + i 1- R) + ( 1 1





























- - - Year (I I 11 a$






















-
















--




--


- --


















References


























































- - - Year (I I 11 a$






















-
















--




--


- --


















References





















































- - - Year (I I 11 a$






















-
















--




--


- --


















References
















































- - - Year (I I 11 a$






















-
















--




--


- --


















References








































- - - Year (I I 11 a$






















-
















--




--


- --


















References































- - - Year (I I 11 a$






















-
















--




--


- --


















References
















































-
















--




--


- --


















References










































-
















--




--


- --


















References



































-
















--




--


- --


















References































-
















--




--


- --


















References








































--




--


- --


















References




































--




--


- --


















References































--




--


- --


















References































--


- --


















References









































References


































References




















































Documents

Dividends and Taxes: Some Empirical Evidence Merton H ...ecsocman.hse.ru/data/887/126/1231/miller_scholes_-_dividends_1982.pdf · prior permission, you may not download an entire