The Impact of Initiating Dividend Payments on Shareholders' Wealth Author(s): Paul Asquith and David W. Mullins, Jr.

Source: The Journal of Business, Vol. 56, No. 1 (Jan., 1983), pp. 77-96 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/2352747 Accessed: 09/09/2009 15:14
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the 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 at http://www.jstor.org/action/showPublisher?publisherCode=ucpress. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Business.

http://www.jstor.org

Paul Asquith and David W. Mullins, Jr.

Harvard University

The Impact of Initiating Dividend Payments on Shareholders' Wealth*

This study investigates the impact of dividends on stockholders' wealth by analyzing 168 firms that either pay the first dividend in their corporate history or initiate dividends after a 10-year hiatus. The empirical results exhibit larger positive excess returns than any previous study on dividends. This result does not depend on any other events (such as earnings announcements) and the excess return is positively related to the size of the initial payment. Subsequent dividend increases for the same sample of firms are also investigated. Compared with the initiation of dividends, the results suggest that subsequent increases may produce a larger positive impact on shareholders' wealth. The results also indicate that other studies may have underestimated the effect of dividend increases. The findings for both initial and subsequent dividends are consistent with the view that dividends convey unique, valuable information to investors.

I.

Introduction

The objective of this study is to add to the understandingof the effect of dividendson shareholders' wealth. The impact of a firm's dividendpolicy on its value is an unresolved issue. In their seminalwork, Millerand Modigliani(1961)demonstrate that, absent imperfections, dividend policy shouldnot affect shareholders'wealth. As Miller and Scholes (1978) subsequently demonstrate, under U.S. tax code this result may survive even if there is differentialtaxation of dividends and capital gains. Dividend irrelevanceis also supportedby the empirical work of Black and Scholes (1974). Alluding to an argumentin paper, Black and Scholes the Miller-Modigliani emphasizethe abilityof firmsto adjustdividends to appeal to tax-inducedinvestor clienteles and argue that this supply effect may account for their finding of no significant relationship between dividendyields and stock returns.

* We wish to thankEugene F. Fama, RobertR. Glauber, JohnV. Lintner,Scott P. Mason, RobertC. Merton,Merton H. Miller, Franco Modigliani,RichardRuback, an anonymous referee, and especially Stewart C. Myers for helpful commentsandsubstantivesuggestions.We wouldalso like to of thankthe participants the MITFinanceWorkshopand the membersof the CRSP Seminaron the Analysis of Security supportwas Generousfinancial Pricesfor theircontributions. provided by the Division of Research, HarvardUniversity GraduateSchool of Business Administration.
(Journal of Business, 1983, vol. 56, no. 1) ? 1983 by The University of Chicago. All rights reserved. 0021-9398/83/5601-0002$01.50 77

78

Journal of Business

Challengesto the dividend irrelevancepropositionhave focused on imperfections.Dividend income is taxed at a higher rate than capital gains, and this suggests a negativewealth impact.The existence of taxinduced clienteles is investigatedby surveys of investors' holdingsby Lewellen et al. (1978)and Blume, Crockett, and Friend (1974)and by the empirical results of Elton and Gruber(1970) concerning the exdividendbehavior of stock prices. Capitalasset pricingmodels incorporatingdifferentialpersonal taxes have been developed and/ortested by several researchers including Brennan (1970), Litzenberger and Ramaswamy(1979, 1980), Rosenbergand Marathe(1979), and Blume (1980). Empiricalresults presentedin these papers suggest that, if risk (i.e., beta) is held constant, before-tax returnsare an increasingfunction of dividend yield. These results are criticized by Miller and Scholes (1981) and Hess (1981).The formerauthorsfindno supportfor an after-taxcapitalasset pricingmodel. Hess, on the other hand, finds that before-taxexpected returnsare relatedto dividendyields. However, he concludes that the natureof this relationshipis not consistent with the tax-inducedeffects hypothesized by an after-taxcapital asset pricingmodel. Finally, a negativewealth impactmay resultfrom other costs associated with paying dividends. In additionto the cost of administering a dividendprogram,the firmmay incurtransactioncosts associatedwith issuing new equity. With a given investment policy and capital structure, an increase in dividends must be funded with new equity. A positive wealth impacthas also been suggestedby manyresearchers. A traditionalviewpoint is that investors preferreturnsin the form of dividends, possibly because of institutionalconstraints. This position is expressed by Grahamand Dodd (1951), Gordon (1959), and Gordonand Bradford(1979)and is supportedby Long's (1978)examination of the returnson the dual series common stock of one firm. A positive wealth impactmay also result from a dividendpolicy that communicatesvaluable informationto investors. Dividends may provide a vehicle for communicatingmanagement'ssuperiorinformation concerning their interpretationof the firm's recent performanceand their assessment of future performance.This view is consistent with the results of empirical studies examining firms' dividend policies (Lintner 1956;Brittain1966;Fama and Babiak 1968).Ross (1977)and Bhattacharya(1979, 1980)present asymmetricinformationmodels in and which dividendsserve as signals of the firm'scurrentperformance future prospects. In a general equilibriumanalysis, Hakansson(1982) demonstratesthat informativedividends improve efficiency when investors are heterogeneous in some respect or financial markets are incomplete. This result can hold even in the presence of deadweight costs associated with dividends. Millerand Rock (1982)show that as a result of information asymmetrybetween investors and managers,div-

Dividend Payments on Shareholders' Wealth

79

idend changes can result in market price reactions. Kalay (1980) studied dividend reductions and, consistent with his signalingmodel, he could not reject the hypothesis that dividend reductions contain information. A number of other empirical studies have examined whether dividends containinformation.Pettit (1972)found that dividendannouncements do convey valuable information. However, Watts (1973) and Gonedes (1978) came to the opposite conclusion. They contend that unexpected dividend changes communicate no informationbeyond that reflected in other contemporaneous variables (e.g., earnings). Laub (1976) and Pettit (1976) challenged Watts's findings, and Watts (1976) rebutted these challenges. However, all of these studies are based primarilyon monthly stock returns. Two studies employ daily returndata. Charest(1978)found that the announcementof a dividend increase generates an excess return of about 1%.1 Because his study makes no effort to remove the effect of contemporaneous earningsannouncements,Charestconcludes that his evidence does not necessarily reveal the presence of informationin dividendannouncements.Aharonyand Swary (1980)documenta small but significantdividend announcementeffect separatefrom the information impact of earnings announcements.Their analysis focuses on dividendannouncementdates that differfrom earningsannouncement dates by at least 11 days. For dividend increases they found a significant averageexcess returnof about 1%over the 2-dayannouncement period. Their study also supports the semistrong form of the efficient capital markethypothesis. There is no leakage of information prior to the dividend announcement, and the full impact of the announcementis concentratedin the 2-day announcementperiod. This study investigates the mannerin which dividends affect shareholders' wealth. The disagreementamong previous empiricalstudies stems from three sources. The first is inadequate identificationand control of other simultaneoussources of informationsuch as earnings announcements. Our analysis, like that of Aharony and Swary, uses daily data to allow explicit identificationand control of contemporaneous information. The second source is the difficulty of isolating and controllingfor investors' expectations. To identify unexpected dividend increases most studies either assume a naive dividendexpectations model (i.e., any change in dividendsis unexpected)or employ some dividendfore1. Charest's interest is primarilythe efficiency of the stock marketwith respect to dividendannouncements,ratherthan the precise measurement the information of impact. The resultreportedabove is basedon a briefextensionof his monthlyanalysis.No adjustment risk is incorporated his analysis of daily excess returns,and Charest for in assumedthe announcement apparently date to be mid-month insteadof determining the actual news date.

80

Journal of Business

casting model to capture investors' expectations. The firms in our study paid no dividendseither duringtheir entire corporatehistoriesor for at least 10 years. In this case the dividendforecastingmodels employed in other studies collapse into the naive expectations model. Compared with subsequent dividend changes, we believe that the naive model accuratelyreflects investors' expectationsfor initialdividends and that initial dividends are more likely to be unexpected. This view implies that hypothesized dividends effects should be most visible at initiation.If dividendinitiationis unexpected, the market reaction on announcementday should capturethe full effect. The excess returnshouldreflectinvestors' estimates of the presentvalue of factors, such as the tax burden associated with dividends and the benefits of establishing a mechanism for communicatingmanagerial information.Subsequentchanges in dividends, exploredin other studies, may be more accuratelyforecast by investors. If so the unexpected portionof a changein dividendsis no longerequal to the entirechange, and the full effects of the dividendchange are not visible in the excess returnon announcementday. The portionof the marketreactionassociated with the expected dividend change is already incorporatedin stock prices on announcementday. The marketreactionon announcement day is only for the unexpected portionof the change in dividend, not its full impact. Initiatingdividends may also induce a change in investor clienteles. Before the initial dividend, stockholders of our sample firms receive returnssolely in the form of capitalgains. If tax clienteles are relevant, these firmsshould be owned by high tax bracketinvestors. Therefore, the dividendannouncementmay induce a changein investorclienteles. The present value of any transactioncosts and other expenses associated with the expected change in clienteles as well as the presentvalue of any tax burden imposed on this new clientele will appear in the market's reaction. Thus, if initial dividends are largely unanticipated,examiningdividend initiationsallows us to purgeinvestors' expectations that may be incorporatedin subsequent dividends. The result should be a clearer view of the true impact of dividends on shareholders'wealth. A final deficiency in much of the previous empirical work is its failure to relate the wealth effect to the magnitudeof dividends. The theory suggests that the two may be related. Exploringthis relation also helps identifytruly unexpected changes in dividendsand provides additionalinsight into the sources of the wealth effect of dividendsas well as of the abilityof investors to forecast both initialand subsequent dividends. We find for our sample that initiatingcash dividends is associated with a significantpositive excess return.The averagemarketresponse to initiationis largerthan the average effect of large subsequentdivi-

Dividend Payments on Shareholders' Wealth

81

dend increases analyzed here and in other studies (Charest 1978; Aharonyand Swary 1980).However, the size of the initialdividendis also larger than subsequent changes. Adjustingfor the magnitudeof dividendchanges, subsequentdividendincreasesappearto generateas large or largeran effect on shareholders'wealth. Ourresults also suggest that previous studies may have underestimated wealth effect the of subsequentdividendincreases. The understatement stems from the failure of previous work to (1) capture accuratelyinvestors' anticipation of dividends and (2) incorporate the magnitudeof dividend increases in their analysis. Ouranalysis supportsthe view that dividends convey to investors valuableinformationin additionto that contained in contemporaneousinformationsources. Further,the benefits of this informationappearto outweigh any costs associated with paying dividends.
II. Issues

The precedingsection suggests a numberof hypotheses concerningthe effect of dividends on shareholders'wealth. Assumingrationalinvestor expectations, an efficient stock market, and no anticipationby investors of the initiationof dividendpayments,investors' estimatesof the present value of the various hypothesizedeffects should be incorporated into the stock price on the day a firm announces an initial dividend. These present values may already be incorporatedin stock prices when subsequentdividendsare announced.If the dividendsare partiallyforecast, the effect of subsequent dividend announcements should reflectonly the communicationof incrementalinformation plus any unexpected changes in the other hypothesized effects. The factors responsiblefor changes in stockholders'wealth can be groupedinto those which may have a positive wealth effect and those which may have a negative wealth effect. Factors that increase shareholders' wealth include the present values of (1) establishinga mechanism for communicatingmanagerialinformation,(2) reducinginstitutional constraints on investors, and (3) benefits associated with the traditionalists' (e.g., Gordon1959)view that investorspreferreturnsin the form of cash dividends. In addition, the initiation impact should include the value of the incrementalinformationcommunicatedby the initial dividend. Factors which decrease shareholders'wealth include the present values of (1) the additional tax burden associated with receiving dividends now and in the future and the adjustmentcosts incurredby tax-inducedchanges in clienteles and (2) any other costs (e.g., administrativecosts, transaction costs associated with issuing new equity) incurredin paying dividends now and in the future. If investors anticipatesubsequentdividend announcements,the effect of these announcements should reflect only the unexpected

82

Journal of Business

changes in the present values discussed above. Subsequentwealth effects should result primarilyfrom the communicationof unexpected incrementalinformationtransmittedthrough an established dividend program. Obviously, it is not easy to isolate the separateeffects of the disparate factors presented above. To provide insight into the effects of dividendswe shall explore four issues: (1) the average effect on shareholders' wealth of both initial and subsequent dividend announcements; (2) the relationshipbetween the wealtheffect and the magnitude of dividendsfor both initialand subsequentdividends;(3) the comparative effects of initialand subsequentdividends;and (4) investors' anticipation of initial and subsequent dividends. The first issue poses a simple question. Which set of factors dominates the announcementeffect? On balance, do our sample dividend announcementsincrease shareholders'wealth? The mannerin which the effect is incorporatedin stock prices throughtime also provides a test of the semistrongform of the efficient capital markethypothesis. The second question is whetherthe magnitudeof the wealth effect is relatedto the size of the dividendpayment. Several of the hypotheses suggest it shouldbe. For example, if dividendssignalvaluableinformation, the wealth effect shouldbe relatedto the size of the signal. Candidates for the size variable include the change in dividend yield and dividendpayout. The thirdissue involves comparinginitialand subsequentdividends. The objective is to compare the effect of establishinga dividendprogram with its use once established. This comparison should provide with insightsinto the variouspresent-valueeffects associated primarily the initiationof dividends. Finally, we shall examine the possibility of an anticipationeffect. When the firmpays no dividends, investors' informationsets includes all informationavailable or inferrablefrom a dividend series constant at zero. Once dividend payments are initiated, investors' information may be augmented by the informationcontained in a nonconstant series of dividends. As a result, investors may be more successful in forecastingthe magnitudeand timing of future dividendchanges. Unlike episodic stock repurchaseswith timingat the discretionof management, investors have expectations concerning the frequency of dividend payments. Aroundthe time a subsequentdividendannouncement is expected, the stock price should already incorporateinvestors' expectations of the impendingdividend. If dividend increases are good news and investors anticipate an increase, a positive excess return should be observed only if the actual increase exceeds the expected
increase.

Some previous empirical studies of dividends have assumed the naive expectationsmodel, that any increase in dividendsis unexpected

Dividend Payments on Shareholders' Wealth

83

(e.g., Aharonyand Swary 1980).As a result, these studies may underestimate the wealth effect of an increase in dividends. Other studies have employed dividend forecasting models to captureinvestors' expectations (e.g., Lintner 1956;Watts 1973).The models are estimated by regressingdividends in a given period againstpast dividends, earnings, and other variables. With a variable series of past dividends, these models may produce forecasts superiorto those implicit in the naive model. When confronted with a constant past series of dividends, these models generallycollapse into the naive model. Thus, for the initiationof dividends the naive model may be difficultto improve
on.2

The empiricalresults help clarify the question of the efficacy of the naive model and the validity of findingsbased on it. The relation between excess returnsand the magnitudeof dividendchanges provides an insightinto the failureof the naive model. The abilityof investors to anticipate initial and subsequent dividends provides evidence of the errorsin the findingsof earlier studies that employ the naive expectations model.
III. Data

This study analyzes a sample of 168 firms that initiate a dividend to common shareholders. The dividend is either the first dividend in a firm's corporatehistory or the resumptionof a dividendafter a hiatus of at least 10 years. The initial 10-yearscreen used was January1954December 1963, and thereforeall first dividendpaymentsin this study occur afterDecember 1963.The period studiedextends to 1980.For all firms, this initial dividend was paid at least 1 year after the firm was listed on either the NYSE or ASE. This requirementguaranteesthe availabilityof data. The set of firmsthat initiateddividendpayments came from several
sources, including Moody's Dividend Record, Standard and Poor's Dividend Record, the Center for Research in Security Prices, and the

Wall Street Journal. The dividend announcement dates and the amount of dividends paid by these companies were then collected. A dividend announcementdate is the date when news of the forthcoming dividend first appears in the Wall Street Journal. Neither the exdividend day nor the day the dividend is paid is considered to be an announcementday.
2. This statement should be limited. Investors may well expect the firm to initiate dividendssometimein the future.Theirestimateshouldbe reflectedin the preannounceeffect shouldreflectonly the unexpectedcomment stock price, and the announcement ponent, not the full effects of initiation.It may be possible to gain some insight into investors'estimatesof the likelihoodof dividendinitiation,but we have not pursuedthis resultsprovideinsightinto investors'abilityto anticpossibility.However, our empirical ipate dividendinitiation.

84

Journal of Business

Dividendannouncementdates and dividendamountswere collected not only for the initial dividend but also for the largest dividend increase that occurred during the following 12 quarters.This provides on information the dividendhistories of the samplefirmsfor the 3-year period following the initial dividend and also allows a comparisonof initialand subsequentdividends. Of the 168initialfirms, 114increased their dividend within 3 years, seven decreased their dividendand the remaining47 kept their dividendat the initiallevel. The results clearly delineatewhetherthe announcementdate is for that firstor subsequent dividends. To control for other events, all other announcementsthat occurred within ? 10 days from a dividend announcementwere collected from the Wall Street Journal Index. The majority of such an-

nouncementsare earnings statements. Finally, stock prices were collected for the month end before all dividend announcements to calculatechanges in dividendyield, and earningsper shareinformation was collected for the previousfiscal year to calculatechangesin payout ratios.

IV. Methodology This study uses daily stock returndata to compute excess stockholder returnsand to examine dividend announcementsfor each firm in the data base. The daily excess returnfor a security is estimated by
XRit = Rit
-

E(Rit),

(1)

where t = the day measured relative to an event; XRit = the excess

returnto securityi for day t; Rit = the returnon securityi duringday t;

E(Rit) = the expected rate of return on security i for day t. The term

E(Rit) is estimated by grouping annually all securities listed on the NYSE and the ASE into 10 equal control portfolios rankedaccording to their Scholes-Williamsbeta estimates. Each security is therefore assigned to one of 10 portfolios. The observed return to the control portfolio which security i is assigned to is then used as the estimate of E(Rit). The daily returns file of the Center for Research in Security Prices (CRSP)provide the observed returnsfor each securityRit. The excess returnfor each security, XRit, is then calculated as the difference between the actual returnto a security and the returnto its control portfolio. Average excess returnsfor each relative day are calculatedby
N

XRt

XRit,

(2)

where N is the numberof securities with excess returnsduringday t. Daily average cumulative excess returns, CERs, are formed by sum-

Dividend Payments on Shareholders' Wealth

85

ming average excess returnsover event time as follows:

L

CER

=
t=K

XRt,
=

(3)

where the CER is for the period from t

K days until t = L days.

In addition, a 2-day average excess return is generated for each dividendannouncementexamined. A 2-day excess returnis necessary to capturethe entire impactof a dividendannouncement.Day t = 0 is the day the news of the dividendis publishedin the Wall Street Journal. In many cases, however, the news is announcedon the previous
day, t = - 1, and reported the next day. If a dividend is announced

before the marketcloses, then the market'sresponse to the news actually predatesthe announcementday by one. If the news is announced after the marketcloses, the marketwill respond the next day and the announcementday is indeed zero. Thus in reality there is a 2-day
announcement "day," t
=

-1

and t = 0. This 2-day return is calN

culated as XR(_1,o)=

XRi( o)

(4)

where XRi(- 1,0) = XR_ I + XRjO;XR_ I = the excess return to

security i on the day priorto the publisheddividendannouncementin the Wall Street Journal; and XRjO = the excess returnto security i on the day the dividend announcementis published in the Wall Street
Journal. Finally, a t-statistic is calculated for XR(- 1,O)by XRt( 10)= XR( lO)/SXR,( IO)/V (5)

where SXR, 10) = the standard deviation of the 2-day excess returns; and N = the numberof firms in the sample.

V. Results This section examines how shareholders'wealth changes with initial and subsequent dividend announcements.Table 1 gives the average daily excess returnsand cumulativeexcess returnsfor the 20-day pethe riod surrounding initialdividendfor 160 firms.3The results for the 2-day announcementperiod are both large and significant.The 2-day
3. Therewere 168firmswhichinitiatedividendsin the sample.Of these 160combined of their announcement dividendinitiationwith the amountof the dividendand are included in tables 1 and 2. Eight firmsmade a separatedividendinitiationannouncement of followed by a second announcement the amountof the dividend.Since this was not an the customarymethodof announcing initialdividend,these eightwere kept separate. day The average2-dayexcess returnsfor the eightare + 4.7%on the firstannouncement day. The t-statisticsare 2.56 and 1.13. and + 1.8%on the second announcement

86 4444
0

Journal of Business

42

> V

~~m

V' 1,

rD 00 C O

r
-0

+ + + + + + + + ++

I ++ u C~~~~~~~

++++
+

1 +1

I I +++

I+ + ++

Dividend Payments on Shareholders' Wealth TABLE 2 Cross-sectional Distribution of Excess Returns for 160 Firms on Announcement Date of First Dividend Percentage of Firms 0 0 0 .6 3.8 2.5 8.8 16.3 13.1 14.4 13.1 18.8 3.8 3.2 1.3 .6

87

Size of Excess Return (XR) XR -.30<XR -.24 < XR -.18 < XR -.12 < XR -.06 < XR -.04 < XR -.02 < XR ?.00 < XR +.02 < XR +.04 < XR +.06 <XR + .12 < XR +.18 < XR +.24 < XR +.30 < XR -.30 -.24 -.18 -.12 -.06 -.04 -.02 .00 +.02 +.04 +.06 + +.12 +.18 +.24 +.30

Cumulative Percentage of Firms 0 0 0 .6 4.4 6.9 15.6 31.9 45.0 59.4 72.5 91.3 95.0 98.1 99.4 100.0

<

excess returnis +3.7% and the associated t-statistic is 6.59. Table 2 shows the cross-sectional distributionof 2-day announcementexcess returnsfor the firms in table 1. For almost 70%of the firmsexamined there is a positive marketreaction to the announcementof initial dividend.4 These results are several times larger than either Charest's (1978) averagedaily excess returnof 1%or Aharonyand Swary's (1980)2-day average excess returnof 1%. Most important,the results supportthe hypothesis that any negative wealth effect dividends generate (either throughchanges in tax-inducedclienteles or throughincreasedfuture financingcosts for the firm) is, on average, more than offset by the positive value investors place on being paid a dividend. It can be argued that a single dividend payment should not have much impact on tax clienteles or on the future financingneeds of the firm. The vast majorityof firms in this sample, however, did not pay only a single dividend. Of the 160firms, 153continuedthe paymentof dividends after the initial dividends for at least 3 years.5 In addition, 114 of these 153 firms increased the dividend at least once duringthe first 12 quartersfollowingthe initialdividend. If this establishmentof a dividend policy is foreseen by investors when an initial dividend is
4. For the few sample firmswith negative excess returns,the result may be due to failure of the naive expectationsmodel ratherthan to an adverse reactionto dividend initiation. 5. All seven firmsthatdid not continuepayingdividendshadan announcement that to effect. The average2-day excess returnfor this announcement date is - 7.9%and the tstatisticis - 4.30. All firmsthatreducedividendseliminated thementirelyduring first the 3 years.

88

Journal of Business

paid, the wealth effect on investors should exceed that associatedwith any single dividend. A separateand more importantissue is whether the strongpositive made available results exhibitedin table 1 are due to other information dividendannouncement. to the capitalmarketsat the same time as the Aharonyand Swary's (1980)findingof separateannouncementeffects for earningsand dividends and Pettit's (1972)findingof large dividend both suggest excess returnswhen there is positive earningsinformation that earnings announcements must be separated from dividend announcements.This raises the possibilitythat the results in tables 1 and of other thanthe announcement an 2 mightbe the result of information initial dividend. To test this possibility, any other events that occurred within + 10 days of any dividend announcementwere identified.6For 66 of our 160 initial dividend announcements, there is an earnings announcement within ? 10 days. Thirty-fiveof these earnings announcements occur within ?1 day of the dividendannouncement.In additionthere are other events, such as merger negotiations, divisional spin-offs, legal actions, and the announcementof new products, which contain the informationand which occur duringthe 21-dayperiod surrounding initial dividend announcement.Some of the firms have both earnings announcementsand nonearningsevents duringthe 21-dayperiod. For event in this period.The remainsix firmsthere was only a nonearnings become publicwithinthe 21ing 88 firmshad no importantinformation the day period surrounding announcementof an initial dividend.7 The excess returnanalysis describedin equations(1)-(5) was redone for each of these three subsamplesand is reportedin table 3. The 2-day
excess returns and associated t-statistics are XR = 4.7% and t = 5.88 for the 88 firms with no other new information, XR = 2.5% and t =

3.08 for the 66 firms with earnings announcements,and XR = 1.6%

schemesare used to eliminatethe influenceof otherevents. One, 6. Two classification mentionedabove, is to segregateany events which occur within + 10 days of the dividend announcement.The other scheme segregates any events which occur simultaneously with the dividend announcement.There are no importantdifferencesin the results for the two criteria,and only the results for the first classificationscheme are reportedhere. 7. There is the possibility that the Wall Street Journal may report a dividend announcementbut not an earningsannouncement.To test for this, we checked the Wall any that surrounded diviearningsannouncements Street Journalfor the two quarterly not dend announcement.There were only four quarterlyearningsannouncements reported. Three are for regularinitial dividendannouncements,and one is for an initial of that had separateannouncements initiationand amount(see dividendannouncement n. 3). The 2-day announcementday excess returnsfor these four firms are + 3.4%, +-7.4%,+ 0.9%and -2.1%, respectively. None of the subsequentdividendannouncements have a missing earningsannouncement.Since the lack of a reportedquarterly earningsstatementdoes not mean that an earningsreportbecame public in the period + 10 days of the dividendannouncement, these firmswere classifiedas havingno earnin ings information the period ? 10 days.

Dividend Payments on Shareholders' Wealth TABLE 3 Two-Day Average Excess Returns and t-Statistics for Initial Dividend Announcements and Subsequent Dividend Increases

89

Excess Return(%) Initialannouncements: All initialannouncements No other events + 10 days Earningsannouncements+ 10 days Otherevents + 10 days Subsequentincreases: Largestabsoluteincreases Subsequentincrease > initial dividend Subsequentincrease > 100% Largestincrease-no other events Subsequentincrease > initial dividend-no other events Subsequentincrease > 100%-no other events 3.7 4.7 2.5 1.6 1.6 1.6 1.7 1.2 .8 .7

t-statistic 6.59 5.88 3.08 1.78 3.07 1.84 1.79 2.07 .53 .41

Number of Firms 160 88 66 6 114 37 30 66 16 11

and t = 1.78for the six firmswith other news. Thus, an initialdividend announcement results in positive excess returnseven when there is no other informationreleased simultaneously.These results suggest that the market'spositive reactionto the dividendannouncement not due is to other events. To the contrary, other informationappearsto negate the impact at the dividendannouncementby reducingthe information content of the dividend announcement. Aharony and Swary (1980) examined the differentialeffect of earnings and dividend announcements and demonstratedthat dividend and earnings announcements are not perfect substitutes.Ourresults suggest that they may be partial substitutes. The results in this study also support the semistrongform of the efficientmarkethypothesis. The excess returnsin table 1 are largeand significant for the 2-day announcement day, t = - 1 and t = 0, but are smaller and insignificantfor all other days. There is no subsequent marketreaction after the announcement.There also appearsto be no leakage of informationprior to the dividendannouncement.Although consistent with Aharony and Swary's results, this absence of prior marketreactionis unusualcomparedwith results of other recent studies using daily data, such as Dodd (1980) and Asquith (1982). Dodd's and Asquith'sresults show a marketreactionto the announcement a of mergerbid from 20 days before until the actual bid. The market'sreactionto subsequentdividendannouncements also is examined.All firmsare followed for 3 years afterthe initialdividendto determinedividendpolicy. As mentionedabove, 153 of the 160 firms continued to pay dividends over this period; 114 firms increased the dollaramountof their dividendat least once, with many firmsdoing so

90

Journal of Business

several times. Excess returns and t-statistics are then calculated for several subsets of subsequentdividends.There are 37 subsequentdividend increases where the absolute increase was largerthan the initial dividend. The 2-day announcementexcess return for this subset is 1.6% and the t-statistic is 1.84. Thirty subsequent increases were greaterthan 100%of the previousperiod's dividend.These had a 2-day excess return of 1.7% and a t-statistic of 1.79. There is of course considerableoverlap between these two samples. The largestabsolute increases in dividendsduringthe 3 years for all 114firmshad an excess returnof 1.6%and a t-statistic of 3.07. Eliminatingsubsequentincreases where there are other events during the period + 10 days lowers the average excess returnsand the tstatistics for all three of the subsets. The 37 firms with subsequent dividendchanges absolutelygreaterthanthe initialchangeare reduced to 16, with an average excess returnof 0.8% and a t-statistic of 0.53. There are only 11 remainingsubsequentincreasesgreaterthan 100% of the currentdividend, and the average excess returnand t-statisticare 0.7%and 0.41, respectively. Finally, 66 firmshad subsequentdividend increases with no other event information.The largest subsequentincrease for these 66 had an average excess return of 1.2% and a tstatistic of 2.07. These results for subsequentdividendincreases are small compared to the initial increase and are comparableto those found by Charest and by Aharonyand Swary. There is a problem,however, in interpreting these results to suggest that initiatinga dividendpolicy has a positive present value over and above that observed with subsequentdividend increases. First, there is no control for the size of the dividend. Presumablyif dividendsare a signalingdevice, the size of the dividend is a measure of the magnitudeof the signal. Initial dividends may be largerthan subsequentincreases and thus may explain the largerexcess returns. Second, the expectationsmodel for initialand subsequentdividends may differ. As suggestedearlier,the naive expectationsmodel may be a more accurate reflection of investors' expectations at the time of dividend initiation. Once a dividend policy is in place, the past sequence and timing of dividend changes may provide informationthat allows investors to construct a better forecasting model. Thus, the expected dividend increase may not be zero as assumed in the naive model. If so, the full effects of a dividend increase are no longer reflected in the excess returnon announcementday. At that time the effects of the expected dividend are already incorporatedin stock
prices.

If dividend increases are expected and received as good news, the announcementday excess return reported by earlier studies understates the market reaction to an increase in dividends. Studies em-

Dividend Payments on Shareholders' Wealth

91

ploying the naive expectations model for subsequent increases measure only the average reaction to the unexpected portion of the increase, which is less than the full effect of a partiallyforecast dividend increase. Examiningthe magnitudeof initialand subsequentdividend changes illustrates the extent to which expected dividend increases are alreadyincorporated stock prices on announcement in day. This objection, that initial dividends cannot be compared directly with subsequent increases, is supportedby the results. The average change in the dividend, measured either by yield or payout ratio, is larger for initial payments than for subsequent ones. For initial dividend payments the average increase in the dividend yield was 3.02% and the average increase in the payout ratio is 17.85%.For the largest subsequent increases the average increase in the dividend yield was 1.07%and the averageincreasein the payoutratiowas 7.34%.Examining the subset of 14 firms where the subsequentincrease in dividend yield was greaterthan it was for the initialdividend,the average2-day excess returnfor subsequentincreases was 2.9% and the average excess returnfor initial dividends was 3.3%. Thus it is possible that the averageexcess returnsare largerfor initialdividendsbecause the average change in the yield or payout ratios is larger. To explore the relation between the wealth effect and the size of dividends, the market'sreactionto an initialdividend(as measuredby each firm'sexcess return)is regressedagainstthe annualizedchangein yield.8 This cross-sectional regression equationis
XRETi = a + a (A yield1) +
Ei,

(6)

where XRETi = the estimated excess return for firm i on the 2-day announcementday as calculated by equation (4); and A yield = the annualizedchange in yield for firmi as calculatedby the new dividend minusthe old dividenddividedby the firm'sprevious-month-end stock
price.

Table 4 reports the regression results for equation (6). There is a positive and significantrelationshipbetween the size of the initialdividend and the size of the firm'sexcess returnon the announcement day. This relationshipholds for the sampleof all initialdividendsandfor the
8. All of the followingregressionsare estimatedfor both changesin yield and payout ratios. Changesin payout ratios may be a better signalingmechanismthan changes in yield, but payoutratiosare unfortunately more subjectto measurement errors.Earnings per sharedata are often impreciseand differentfirmsemploy differentaccountingpractices. Furthermore,a large number of the sample firms changed certain accounting proceduresduringthe periodstudied.Becauseof this the R2for the sampleof firmsusing changesin payoutratiosis always less thanthe R2 for the same sampleusingchangesin yields. The following reported results use only changes in yield, but the results for changesin payoutratios are all consistentwith those reportedwith respect to both sign and significance.A similarrelationis independently derivedby Millerand Rock (1982) from an assumptionof asymmetricinformation (see their eq. [22]).

92 TABLE 4

Journal of Business OLS Estimates of Equation (6) for Initial Dividends and Largest Subsequent Dividend Increases ct (%) .38 (.43) .34 (.27)
-.96

Sample
All initial dividends Initial dividends-no other events Largest subsequent increases Largest subsequent decreasesno other events
NOTE.-t-statistics are in parentheses.

P
1.22 (4.63) 1.45 (4.37) 2.20 (4.43) 2.94 (4.71)

R2 .120 .182 .149 .258

N
160 88 114 66

(- 1.26) - 1.98 (-2.35)

sample where there is no other informationduring the period + 10 days. The positive slope, which captures the effects of an unexpected dividend increase, demonstratesthat increases in dividends are good news. If increases are good news and dividend initiation is largely unforecasted, dividend increases would be expected to produce only positive market reactions. The nonnegative intercept term suggests small that, on average, investors were not disappointedby arbitrarily initialdividends.This findingimpliesthat the naive expectationsmodel may be accurate for dividend initiation. The intercept term should measure the effect of dividend initiationseparatefrom the size of the differentfrom zero. dividend.The interceptis small and insignificantly Finally, the regressions also suggest that the dividend effect may be large enough to offset any investor tax differential.9 Equation (6) is used also to regress the excess returns associated with subsequent dividend increases against changes in yields. These results are reportedin table 4. The interceptfor subsequentdividend increases is negative and the coefficient is much larger. Again, larger dividends are associated with largerexcess returns. This negative intercept suggests the failureof the naive expectations model for subsequentincreases. The negative interceptmay reflectinvestors' anticipation of subsequent dividend increases already incorporatedin stock prices. The excess returnon announcementday then capturesonly the
9. For example, suppose the worst-case investor is subject to a 100%tax on the dividend,with all other investorstax exempt. The stock price would decline by the full amountof the dividendon the ex-dividendday. However, the regressionsdemonstrate day is at least as large as the dividendyield. that the excess returnon announcement day Therefore,the capitalgainon announcement is at least as largeas any capitalloss on the ex-dividendday. An investor would not lose even if the entire dividendis taxed away. Further,the yields employedin the regressionsare annualized,while manyof the dividendswere actuallyquarterlydividends.This suggests that dividendsmay produce net gains even for high tax bracketinvestors.

Dividend Payments on Shareholders' Wealth

93

unexpected component of the dividend increase, rather than its full effects. The fitted regressionline should equate zero excess returnswith the expected level of the dividend increase. If the increase is less than expected, the excess return will be negative. Only if the increase is greaterthan expected will the excess returnbe positive. Of the excess returnsfor the largest subsequent dividend increases, 42% are negative. If the model is correctly specified, the negative of the intercept term measures the returnassociated with the expected portion of the dividendincrease. If the firmannouncedno increase in dividends,this return, which is incorporatedin stock prices before the dividend announcement, would be reclaimed through a negative excess return. The intercept is significantfor the sample of subsequentdividendincreases with no other events.10 This result supports the proposition that the expectations models for initial and subsequentdividends differ. The a terms in table 4 show that the marketreacts more strongly,as measuredby excess returns,to subsequentincreases in dividendsthan to initialdividends. In both instances the a termmeasuresthe market's reactionto the magnitudeof an unexpectedchange in dividends.When there are no other events, the 2.94 a for subsequentdividendincreases is twice as largeas the 1.45 a for initialdividends,and the t-statisticfor Ho: PI 0 12 is 2.11.11 This raises the possibility that the marketreacts less favorably to initial dividends than to subsequent dividend increases. In summary,the regression results supportthe hypothesis that part of the largerexcess returnassociated with initialdividendsis due to the largerincrease in the dividendyield. This result is also consistent with signalingtheory in that the market'sreaction is significantlyrelatedto the size of the dividend change. Finally, the negative intercept for subsequentdividendincreases suggests that they are partiallyforecast. Previous empiricalmeasures of the market's reaction to dividend increases, therefore, may understate the real reaction since only the unforecastportion is captured. VI. Conclusions Our results demonstratethat, for this sample of firms, initiatingdividends increases shareholders'wealth. The same is true of subsequent
10. If all expected subsequentdividendincreasesare not equal, then using eq. (6) as an aggregate functionintroducespossibleproblemsof aggregation. This possibilityis not dealt with explicitlyhere and thus care shouldbe taken in interpreting stronglythe too applicationof eq. (6) to the subsequentdividendincreases. 11. Here againthere is an assumptionthat the two samplesare independent each of other.

94

Journal of Business

the dividendincreases. Incorporating effects of the magnitudeof diviand investors' anticipationof subsequentincreases, the wealth dends effect of subsequent dividend increases appears to be as large if not largerthan the effect of initiation.A smallerinitiationeffect may result from the negative present value of taxes and/orfinancingcosts. The effects we find are largerthan those presented in other studies and do not appearto be caused by contemporaneousannouncements such as earnings reports. These results are consistent with the view that dividends convey unique, valuable informationto investors. As Lintner (1956) and others have documented, managers'behavioralso appearsto be consistent with this view. Many authorshave dismissed this informationrole as unimportant. They suggest that equally efficacious, cheaper alternatives exist throughwhich managerscan disseminateinformation(e.g., Millerand Modigliani 1961; Pettit 1972, Black 1976; Stern 1979). As a purely empiricalmatter, our results as well as those of Aharony and Swary (1980) demonstratethat dividend announcementsconvey information over and above that contained in other announcements. Dividend policy has several attractive aspects as an information transmissionmechanism.Unlike the detailedfocus of other announcements, dividends can be used as a simple, comprehensive signal of management'sinterpretationof the firm's recent performanceand its future prospects. Unlike most announcements, dividend announcements must be backedwith hardcold cash. The firmmust eithergenerate this cash or convince the capitalmarketsto supplyit. In additionto the credibilityof cash signals, dividends are also highly visible compared with other announcements. These advantages are not unique to dividends but are shared by stock repurchases. As the study by Vermaelen (1981) demonstrates, repurchases stock repurchasesmay convey information.Furthermore, may be more attractive to investors because of the tax treatmentof and capitalgains. However, the timingof stock repurchasesis irregular at the discretionof management.An advantageof dividendsfor investors is the fixed, periodicnatureof announcements.Once dividendsare initiated, shareholdersapparentlyanticipatea periodic signal by management and managementis forced to submit to a periodic review. Finally, tax considerations appear to preclude the replacement of a dividendpolicy with repurchasesof equal magnitudeand frequency.12 Thus, dividend payments are more frequent than repurchases, and investors should benefit from the regular release of valuable information.
12. If repurchaseswere executed with equal frequency, they would be essentially equivalentto dividends.In this case the tax advantageto repurchaseswould likely be removed. Repurchaseswhich are essentiallydividendsare treatedas dividendsfor tax purposesunderU.S. tax code.

Dividend Payments on Shareholders' Wealth

95

Despite promisingtheoreticaland empiricalwork, more researchis needed to provide a satisfactory theoretical model, more explicitly specifiedempiricaltests, and guidanceto managersconcerningoptimal dividend policy. The result of this research may be a solution to the dividend puzzle. In conclusion we can say that initiatinga dividend policy does matter, probably as an informationsource, and that the marketreaction is strong and positive.

References
and Aharony,J., and Swary, I. 1980. Quarterlydividendand earningsannouncements stockholders'returns:an empiricalanalysis. Journalof Finance 35 (1): 1-12. and Asquith,P. 1982.A two-eventstudyof mergerbids, marketuncertainty, stockholder returns.WorkingPaper. Boston, Mass.: HarvardUniversity. Bhattacharya,S. 1979. Imperfectinformation,dividend policy, and "the bird in the
hand" fallacy. Bell Journal of Economics 10 (1): 259-70.

and Bhattacharya, 1980.Nondissipativesignalingstructure dividendpolicy. Quarterly S.

Journal of Economics 95 (1): 1-24. Black, F. 1976. The dividend puzzle. Journal of Portfolio Management 2 (2): 5-8.

Black, F., and Scholes, M. 1974. The effects of dividendyield and dividendpolicy on common stock prices and returns. Journal of Financial Economics 1 (1): 1-22. Blume, M. E. 1980.Stock returnsand dividendyields: some more evidence. Review of
Economics and Statistics 62 (4): 567-77.

in Blume, M. E.; Crockett,J.; and Friend, I. 1974. Stockownership the United States: and Business 54 (11): 16-40. characteristics trends. Surveyof Current Brennan,M. J. 1970. Taxes marketvaluationand corporatefinancialpolicy. National
Tax Journal 23: (4): 417-27.

DividendPolicy. Washington,D.C.: BrookingsInstitution. Brittain,J. 1966.Corporate stock returnsand marketefficiency.II. Journal Charest,G. 1978.Dividendinformation,
of Financial Economics 6 (2/3): 297-330.

wealth.Jourdiscretionand stockholder Dodd, P. 1980.Mergerproposals,management

nal of Financial Economics 8 (2): 105-38.

tax stockholder ratesandthe clientele.Review Elton, E., andGruber,M. 1970.Marginal

of Economics and Statistics 52 (1): 68-74.

analysis.Journalof the Fama,E. F., and Babiak,H. 1968.Dividendpolicy:an empirical

American Statistical Association 63 (4): 1132-61.

Gonedes, N. J. 1978.Corporatesignaling,externalaccounting,and capitalmarketequiitems. Journalof Accountlibrium: evidence of dividends,income, and extraordinary

ing Research 16 (1): 26-79.

Gordon,M. J., 1959,Dividends, earningsand stock prices. Review of Economics and

Statistics 41 (2): 99-105.

Gordon,R. H., and Bradford,D. F. 1979.Taxationand the stock marketvaluationof capital gains and dividends:theory and empiricalresults. WorkingPaper no. 456. Mass.: NationalBureauof Economic Research. Cambridge,
Graham, B., and Dodd, D. L. 1951. Security Analysis: Principles and Techniques. 3d ed.

New York: McGraw-Hill. Hakansson,N. H. 1982.To pay or not to pay dividends.Journalof Finance 37 (2):41528.

Hess, P. J. 1981.Dividendyields, and stock returns:a test for tax effects. Unpublished manuscript.Columbus:Ohio State University. Kalay, A. 1980. Signaling,informationcontent, and the reluctanceto cut dividends.
Journal of Financial and Quantitative Analysis 15 (4): 855-69.

content of dividends.Journalof Business 49 (1): Laub, P. M. 1976.On the information

73-80.

Lewellen, W. G.; Stanley, K. L.; Lease, R. C.; and Schlarbaum,G. G. 1978. Some

96

Journal of Business

direct evidence on the dividend clientele phenomenon.Journal of Finance 33 (5):

1385-99.

Lintner, J. 1956. Distributionof incomes of corporationsamong dividends, retained

earnings, and taxes. American Economic Review 46 (2): 97-113.

Litzenberger, H., and Ramaswamy,K. 1979.The effects of personaltaxes and diviR. dends on capital asset prices: theory and empiricalevidence. Journal of Financial
Economics 7 (2): 163-95.

Litzenberger,R. H., and Ramaswamy,K. 1980. Dividends, short selling restrictions, tax-induced investorclientelesandmarketequilibrium. Journalof Finance35 (2):46982. Long, J. 1978.The marketvaluationof cash dividends:a case to consider.Journalof
Financial Economics 6 (2/3): 235-64.

Miller, M. H., and Modigliani,F. 1961. Dividendpolicy, growth and the valuationof
shares. Journal of Business 34 (4): 411-33.

Miller,M. H., and Rock, K. 1982.Dividendpolicy underasymmetric information, 1. pt. Unpublishedmanuscript.Chicago:Universityof Chicago. Miller, M. H., and Scholes, M. S. 1978. Dividends and taxes. Journal of Financial
Economics 6 (4): 333-64.

Miller,M. H., and Scholes, M. S. 1981.Dividendsand taxes: some empiricalevidence. Unpublishedmanuscript.Chicago:Universityof Chicago. Pettit, R. R. 1972.Dividendannouncements,securityperformance, capitalmarket and
efficiency. Journal of Finance 27 (5): 993-1007.

Pettit, R. R. 1976.The impactof dividendsand earningsannouncements: reconciliaa

tion. Journal of Business 49 (1): 86-96.

Rosenberg,B., and Marathe,V. 1979.Tests of capitalasset pricinghypotheses.In Haim Levy (ed.), Research in Finance. Greenwich,Conn.: JAI. Ross, S. A. 1977. The determination financialstructure:the incentive signalingapof
proach. Bell Journal of Economics 8 (1): 23-40.

Stern, J. 1979.The dividendquestion. WallStreet Journal,July 16, 1981. Vermaelen,T. 1981. Common stock repurchasesand market signalling:an empirical
study. Journal of Financial Economics 9 (2): 139-83.

Watts, R. 1973.The information content of dividends.Journalof Business 46 (2): 191211. Watts,R. 1976.Commentson "The impactof dividendand earningsannouncements: a reconciliation."Journalof Business 49 (1): 97-106.