Journal of Financial Econchcs 21(1988) 149-l 75. North-Holland

Received December 1986, final version received March 1988

Firms that initiate dividend payments have positive earnings changes bath before and after the dividend policy change, ,while those omitting dividend payments have negative earhgs changes. subqwt eamiqp #dlallges aIe padi* ldat4!d to the diM -t retum, imd stock price reactions at subsequent eamings -ts are smaller than usual, sug%esting thattheseeamingschang~areparhlljyiticipatedatthedividend annouucement. The results iadicate that investors il@rpret announcements of dividend initiations and omissions as maxqers’ forecasts of ffihbre eanhlgs cbges.

This paper examks whether dividend p&q changes convey informatkn about Wure exningq a to+ i- wideigr discussed in the finance literature. Miller and Modig!&i (1961) demonstrate that, under conditions of perfect capital markets and zero taxes, dividends do not affect the value of the firm. They point out that dividends may have information content if managers have better information than investors abotit the km’s future earnings and use that infoiwtion to set current dividemkl Dividend changes can thus be thought of as management forecasts of future earnings changes substantiat&by cash. This dividend information hypothesis has been formal&d by P,hattachaqa

*The paper benefited from the comments of seminar partici!xzNs at the Utiversiiiy of CaL fornia, Berkeley, Camcgie--Mellon University, Unixrsity of Chbgo, Ehr~ard Universitv, Mas- sach~tts institute of Technology, ?iorthwe;terr! Utivzrsity, Utiversity of Oregon, University of Pennsylvania, University of Rochester, and Stanford University. We ar: particularly grateful to Paul .bquitb, Ihi, Kc?lan, Richard Leftwich, Stew Myers, Pat O’Brien, Rick Ruback, Ross Watts, an anonymous referee, and Michael Jensen and John Long (the editors) for their suggestions and to Paul Asquith and David Muliins for making their dividend initiation sampk available for the paper. This research has benefited from fundhg by the Division of Research, Harvard BU&ZPES E;chool.

‘Lmtner (1956) provides empirical evidence that managers consider past as wel;l as future earnings in setting current dividends.

150 P M. Ht-a/y and K.G. Palepu, Dividend policy and earnings informtion

(1979,1980), John and Williams (1985), Miller and Rock (1985), and Ofer and Thakor (1987).

vFzwatts (1973), in an early empirical study of the information content of die&ends, examines the r&tion between unexpected dividend changes and futu-; earnings, and abnormal stock returns for firms that announce unex- pected dividend changes. He finds that unexpected dividend changes provide little Gaformation about future earnings and there are no abnormal returns in months surrounding the dividend announcements.2 As Watts points out, his study has two limitations. First, the use of monthly, rather than daily, stock pries datz. makes it difficult to distinguish between the effect of dividend and other contemporaneous information releases. Second, the potential noise In the dividend expectation model reduces the power of the tests. A recent study by Asquith and.MuUins (1983) reexamines the stock price

reaction to dividend announcements, using daily stock price data to control for other contemporaneous information announcements. To mitigate problems of specifying dividend expectations, they analyze dividend initiatiotis, which they believe are least likely to be anticipated. Their results show sign&ant positive abnormal returns at dividend initiation announcements. other studies, including Aharoney and Gus- LwVcuy (i%O), Brickley (1983), JCalay and Lowenstein (1985), ad Diehan and Oppenheimer (1984), also document abnormal returns at the announcement of unanticipated dividend increases and de- creases,

The above studies, however, do not reexamine the relation between dividend policy changes and subsequent earnings, the second issue analyzed by Watts. This paper provides fresh evidence on this issue. 0ur tests differ from those of Watts in two ways: (1) we use dividend announcement returns, rather than unexpected dividends, as a measure of ditidcnd information, and (2) we focus on dividend initiations and omissions, the two dividend policy changes that bZW4Z been documented in the literature as having the largest average an- nouncement retums.3

Our sample comprises 131 Grms that pay dividends for the first time or resume payments after a hiatus of at least ten years and 972 firms that omit dividend payments for the first time or after gqing continuoudy for at least ten years. The tests exmine three issues. First, they document changes in a firm’s earnings performance for five years before and five years after a dividend initiation or omission. Next, they examine whether subsebTzcnt earn- ings changes are related to the infmmation r&2~d at the dividend annoumce_ ment, as measured by the _.=_cI aouncement return. Finally, they analyze the market raction to earnings announcements after the dividend p&y change

2Gonedes (19Xj reports stilar fM&s. ‘Since the sample comprises extreme dividend changes, our findings may not be relevant to all

dividend policy changes.

P.M. Healy and K.G. Pakprr, Dividend policy and earnings information 151

to assess whether the market anticipates these earmngs from the dividend announcement.

The results indicate that firms that initiate and omit dividends have signifi- cant increases and decreases in their annual earnings for at least one yea before and the year of the Gvidend policy change. These fMiugs are mn- sistent with those reported by Liitner j1956), Fama and Babiak (P96$), and Watts (1973), and suggest that dividend initiations md omissions can, in part, be predicted by ch s in past and current s, However, there is a signifbnt market reaction to the announcement of these dividend policy changes, indicating that they cannot be perfectly predict4 arid that they convey new information.

Tests of earnings performance after dividend policy changes show that firms that initiate and omit dividends have sign&ant increases and decreases, respectively* in ~zrnings for at least one year afterward. The earnings changes are positively related to the two-day abnormal stock price reaction at the dividend initiation or omission announcement. Finally, the magnitude of the siock price reactions to earnings announcements following the dividend ini- tiation or omission is significantly less than normal, indicating that these earnings changes are anticipated by the market at the date of the dividend announcement,

Consistent with the dividend information hypothesis, tie above findings _ _ indicate that the information conveyed by dividend initiations and omissions is related to earnings changes following ‘the announcement of these dividend policy changes. The results are also consistent with Lintner’s observation that managers consider past, current, and future earnings when making dividend policy decisions. Investors therefore inte_qret dividend initiations and omis-

sions as changes in managements’ earnings forecasts. Although the tests in this paper do not address whether investors value dividends per se, that is, whether higher payouts are associated with higher stock prices, the results do suggest -- that stock price responses to dividend initiations and omissions coot be attributed solely to investors’ preferences for cash dividends.

In the next section, we describe the data used in our empirical analysis. The third maim de&ks the an@kd tests and results, and the paper comcludes Ain a summary and discussioit G &e results,

2. Data

Our dividend initiation sample comprises the fins used by %xpaith and Mullins (1983) in fieir study of the e&z; of initiating dividend announce- ments on shareholders’ wealth. Asquith and ullins define an initiating dividend as tihe first dividen

152 .P. M. Heuly and ,K G. Palepu, Divide&d policy and earnings information

&i&d after a hiatus of at least ten years. Their initial ten-year screen was January 1954 to December 1963. AlI L-st dividend payments in the sample therefore occur after i963, and the period studied extends to i980. For all sample ~~~IPIS, the initial dividend was paid at least one. year after the finn was listed on either the New York C+m’* V L.ren Exchange (NYSE) or the American Stwk Ex&uige (AhfKj.

Asquith and M&ins sample of 16$ fhms is selected from several sources: &&y’s Dividend Record, Stand&d and Poor’s Dividend Record, the Center

for Research in Security Prices (CRSP), and the Wall Street Journal. The dividend announcement date. denned as the date when news of the foh*com- ing dividend fist appeared m the VW! Street Journal, tie stock price two days ~&GE the dividend announcement, and stock returns %r the day before and the day of the dividend announcement are collected by Asquith and Mullins for each of the sample firms.

For the above 168 ti we collect the following additional data: (1) the six &a&year earnings announcement dates prior to the dividend initiation an nouncement and the five subsequent annual earnings announcement dates (from the Wall Street Joumal Index), and (2) annual earnings per share before extraordinary items and discontinued operations reported’ at the above 11 earnings announcement dates (from the 1984 Compustat Annual Industrial and Research tapes). To emure that sufhcient observations per Grm are available for cross-sectional tests, &ms are included in the final sample if at least 8 of the 11 ezzings mouncement dates and earnings &+a zre available. This restriction leads to a potential PX ;- SW stival bias. The effect of this bias on the resqrttc- _r3 is discussed later in t.he paper.

Our usable sample comprises 131 Grms. Of the 37 Grms eliminated from Asqtith and MuUins’ sample, 14 are not listed on Comptlrtat ties, 10 do not have the required number of WGzLI Street Journal ear&gs announcement dates, 3 were new lisiinm on the NYSE or AMEX and have insuflicient eami&s data before the*dividend announ~~cz,, and 10 have insufficient earnings data after the dividend initiation annoluncement (8 of these Grms were acqtied, 1 was hoived h an exchange transaction, and 1 was delisted).

The fiscd-year earnings announced immediately prior to the dividend titiation announcement are defined as earnings for year -1; the annual earnings announced immediately after the di~<dend amretncernent are defined as f:afnhgs for year 0.” The five annual earnings announced prior to the year - 1. s.‘lsnnouncement are defined as earnings for years _- 6 to = 2. Similarly, the four iannual earnings announced subsequent to the year c) announcement are d&nGd as earnings for years P to 4.

ol’ 4%x fkms iqort annual earnings concurrently with the dividend initiation announcement. Each these earnings announcements is assigned to year - 1, thereby ensuring that all the earnings

assigned to years 0 to 4 are annnuaced &et the dividend aoauncement. &&ding these firms l:om the analysis does not alter our conclusions.

PM. Heaiy and KG. Palepu, Di:*idendpolicy and earnings information 153

Since dividend I ~nGzun~men&s OCCUR * Aroughout the fiscal year, year 0 earnings defined above include quarterly earnings that are announced before, as well as after, the dividend announcement. This introduces a potential bias to our findings regarding earnings performance following the dividend ini- tiation. To address this concern, we perform additional tests using quarterly earnings data.

The first quarterly earnings announcement date prior to the dividend initiation date is collected from the W&l1 Street Journal In&x. Quarterly earnings per share before extraordinary items and discontinued operations re$orted on this date and earnings reported at the three prior quarterly announcements are summ& and d&ned as earnings for year -1. The four quarterIy earnings reported following the dividend date are s to con- struct earnings for year 0. E!amings for year -2 are also constructed from quarterly data using this procedure. Defined this way5 earnings for years -2 and -1 are announced strictly &fore the dividend initiation announcement, and earnings for year 0 are announced after the dividend date. We use Standar$ and Poor’s Compustat Quarterly 3 rdmtrial tapes as the souxz of quarterly earnings data? Quarterly data are q.vailable on t&se tapes for 129 al our 131 test firms.

The 131 dividend initiation &ms are in 38 two-digit SIC industries. There is no evidence of industry clustering within the sample. Table 1 presents the number of sample firms initiating dividends by year. The most frequent years of dividend initiation are 1976 (32 firms) and 1977 (25 fkms). The dividend initiation year for the remaining 74 firms ranges from 1970 to 1979.

2.2. Dividend omission sample

Tk dividend omission sample is selected by searching the 1984 Standard and Poor’s Compustat Annual Indwtrial and Research tapes and the CRSP tapes for NOSE and AMEX firms that omitted dividends during the period 1969-1980. Firms from this list are in&de6 in the sarrple if they onA dividends for the first time in their history or after paving divklends continua ously for at least ten years. Out of the 240 firms identied from this search, the Wall Street Journal Index does not list the dividend omission announcement date for 30 firms, which reduces the sample to 210 companies.

The following data are collected for each of the 210 firms: (1) the dividend omission announcement date, the date when news of the forthcoming dividend oxnission first appeared in the Wall Street JoumaC hkxz (2) the stock price t=o days before the divide:&5 omission announcement date froicin the

%kmdard and &or’s does not construct a Research tape for quarterly data. To Akt bfomation for companie, that wexb deli: ?, we searched old copies of the Compurrat Quarter& hdwtriac tape.

154 F.M. Healy and K.G. ?aiepu, Diuidendpoky and earnings information

Table 1

w-.,&b ,W”Li_=i t nf samPIe firms initiating and omitting dividends by year in the period 1969-1980.“b

Companies Companies initiating dividends omitting divideads

Year Number Percent Number Percent

1969 0 0.0 11 6.4 1970 2 1.5 50 29.1 1971 3 2.3 33 19.2 1972 10 7.6 16 9.3 1973 15 11.5 7 4.1 1974 13 9.9 13 7.6 1975 13 9.9 16 9.3 1976 32 24.4 4 2.3 1?7? 25 19.1 2 1.1 1978 15 11.5 7 4.1 1979 3 2.3 10 5.8 1980 0 0.0 3 1.7

TOtd 131 100.0 172 100.0

aDividend initiations are Grst dividend payments in a km’s history or dividend resumptions after a hiatus of at least ten years Divrdend omissions are fkst-time eliminations by Grms that paid dividends continuously koughout their history or for at least ten years.

bTo be ineluded in the sample, Grms must meet the following requirements: (1) be listed on the NYSE or AMEX, (2) have dividend initiation/omission annotuwement dates available in the WfzU Street Jow& In&x, (3) have stock price data available for two days before the dividend initiation/omission anno uncement on the CRW Daily Mzster Tape or &wu&wd and Boor’s Da@ Stuck price #kc&, (4) have stock returns available on the C’RSP data f&s for the day before and the day of the dividend initiation/omission announcement, (5) have annual earnings per share before extraordinary items and dissntinued operations available from the 1984 Compustat for 8 of the 11 years surrounding the dividend initiation,/omission announcement date, and (6) have earnings an.nounceme=nt &tes available in the Wall Sfreer J-1 Index for these same years.

Daily Masts Tape or Stan& r~i and Poor’s Daily Stock Price Record, (3) stock returns for &e day before and the day of the dividend omission announcement from the CRSP Da@ Return file&and (41 annual earnings announcement dates and reported earnings per share before extraordinary items and discon- tinued operations for years -f; *A 4 rick, *ho wnw&tres b-jescpzbed above for v ou ‘, UUIIIej wu ye”---

- tics dkidend initiation sampkB The final san~ple consists of 172 firms. Of the 38 fums excluded for data

. awlability reasons, 17 have. missing earnings announcement dates in the WJak’ Stteet Joutnal Index, 10 have insufficient data prior to the dividend omission, as they were nc* listings on the NYSE or AMEX, and 11 have insufficient data after the divkknd announcement (4 are acquired, 6 are dehsted and 1 I firm’s exchange liatinrj is suspended).

@kfteen firms report annual earnings concurrently with *h: dividend omission announcement. Each of these earnings announcements is assigned to year - 1, thereby ensuring that aI1 the etiqs assigned to years 0 to 4 are announced after the dividend announcement. Again, excludiug these observations does not alter our couc!usions.

P.M. Healy and KG. Palepu, Diuidend&cy and earnings information 1.s

As noted for dividend initiations, the method of aligning annual earnings in relation to the dividend announcement d.ate leads to a potential bias in our tests of post_smIssion earning E. To correct this problem, the procedure de- scribed in section 3.1 is adopted to construct earnings variables for years -2, - 1, and 0 using quarterly earnings per share before extraordinary items and di,~ntinu& operations. Quarterly data are available for X40 of the 172 test firms.’

The 172 dividend omission are Li 42 digit SIC industries. There is no evidence of industry clustering within the sample. Table 1 presents the number of sample firms omitting dividends by year. The most frequent years of dividend omission are 1970 (50 Grms) and 1971 (33 &ms). The dividend omission years ibr the remaining 89 firms range from 1969 to 1980.

A matched sample of comparison firms is collected to provide an earnings benchmark for evalua+ung test fiams’ earnings performance. Each wmparison firm is randomly selected from the same industry as its test firm match. For test firms currently listed on the Compustat Research tape, the comparison Grms are selected from the Research tqx to mM wv+orstip bias in the comparison samples. For the remaining test firms the comparison Grms are selected from the regular C&Z~UH~~ tape. Indus*pi matches are based on the test &m’s SIC Cole at the date of the dividend initiation or omission announcement. &ch &m in the comparison sample is required to: (1) :x listed on either the NOSE or the AMEX, (2) have stock price data on the CRSP Master Tape or in Standard and Poor’s Daily Stock Price Record two days prior to the dividend initiation or omission by its match test ti -1, and (3j meet the same earnings data requirements as the initiation and omission test firms.

Comparison firms are available for 13(3 dividend initiation test Erms and 171 dividend omission test firms. Where possible, SIC matches are based on four-digit codes. There are 122 four-digit matches for the initiation sample and

l n#u mnt La&! -rLI k%?aA 159 such matches for the omission sample. The rem-g lIIQccUwrr) cuU uUbacJ on two-digit industry codes. One fkm each in the dividend initiation and eksion samples could not be matched! For tests that require quartcr~y c,‘-_tings per share, data are available for 118 initiating m~parison %S ami

101 omission comparison firms,

7Quarterly data are available for relatively fewer dividend omission firms than for the initiating sample. As can be seen in table 1, the omissions are clustered in the early 197Os, when quarterly data available on Computstat are less complete, whereas the initiations tend to occur in the mid497Os.

‘The dividend initiation firm that is unmatched is in the amusement and recreation services, except motion picture, industry; the dividend omission firm that is unmatched ia in the apparel and accessog stores industry.

156 P. M. He& md K. 6. Palepu, Dividend policy and earnings inform%m

Table 2

Abnormal returns for 131 dividend initia+&g ad i72 dividend omitting firms for selected hdding pmiods surro~mcbg the Was:’ Street .bumu~ announcement date, PD (t statistics in parentheses).a

Days in holding period’

PD-6OtoPD-21

FD-2OtoPD-11

PD-lOtoPD-2

PD-1 toPD

PD+ltoPD+!!!

PD+ 11 toPD+20

Dividend initiating fhS

Dividend omitting firms

3.5% (4.8)

1.1 (2.7)

4.0 (i&!-g]

3.9 (15.4)

1.4

.-a m.4. (3.6)

0.6 (1.4)=

- 7.0% (-4.0)

- 2.7 (- 3.1)

- 7.0 ( - 8.0)

- 9.5 (-24.8)

-1.2 (- 1.4)C

-0.5 ( - 0.5)’

“Abnormal returns are m~~et-s!@sf~-J returns using CRSB equal-we&&d market returns. ‘!3e sample firms initiate,!omit dividends in the period 1%9 to 1980.

“The holding periods include the b-g and ending date. Xll returns other than these are significant at the 5% level or better.

3. Testsandresults

The results of folur +fic*- r-Cb are reported in this section. First the tnarket reaction to the announcen~nt of dividend initiations and omissions is de- scribed. Second, earnings changes in the five years before and five years after these dividend events are examined. Next, the relation between the market reaction to the dividend announcesnents and subsequent earnings changes is analyzed. Finally, we test whether the market reaction to the subsequent earn~gs announcements is less than the normal reaction.

2.1. Market reaction tr? dividend initiations and mimi~ns

Abnormal returns are estimated for dividend initiation and omission firms for the period 60 days before to 20 days after the announeernent. Abnormal returns are defined as market-adjusted returns, that is, the diBerence between firms’ returns and returns on the CWSP equal-weighted market portfolio.

Mean abnormal returns for various holding sreti.& surrounding the di- vidend announcements are rcpor~d in table -2. b The mean announcement

‘Abnormal returns are also estimated as risk-adjusted returns from a market model. The results reported in table 2 and the other tests reported later in the paper are not sensitive to the definition of abnormzJ returns,

JW. Healy and K.G. Palepu, Dividend policy and earnings information 151

return (days -1 enrd Q) for ?he initiation firms is 3.9% and is statistically siguificant at the i% PC&. For the dividend omission thms, the mean t.~o-day announce:ment return is -M%, s&r&ant at the 1% level. There is also evidence that initiating and omitting ohms have significant returns in the preannou83cement period. studies?”

These results are similar to Endings of earlier

The above findings indicate that investors partially anticipate dividend initiations and omissions from other information available before ihe an- nouncement of the dividend p dky change. However, these events are not fully _

anticipated: the actual announcement of the policy changes conveys info_rz+ tion to the market.

3 v.2. Earning9 changes surrounding dividend initiations and omi.wir n

Ball and Brown (1968), Ball and Watts (1973, and Watts and Leftwich (1977) suggest that annual earnings follow a random walk; the average eadgs changes for a random sample of firms are therefore expected to be zero. However, Liitner (1956) implies that dividend initiations (omissions) are preceded by a numbe: of years of earnings increases (decreases). Further, if dividend policy chsnges convey inform ation on future earnings, dividend initiations (omissionsj are expected to be followed by earnings increases (decreases).

To examine whether firms that initiate or omit dividends show systematic earnings patterns, we calculate earnings changes for five years before (years - 5 to -l), the year of (year 0), and the four years after (vears 1 to 4) the dividend policy change. To aggregate results across firms, wedexpress earnings changes in these years as a percentsge of the stock price two days before the dividend announcement, Pj. The standardized change in earnings for firm J in year t, dEjg, is therefore defined as

0 I

where E. are ear&p per share before extraor&nm item-s ~3~~2 &;i;~~~&~u~~ Jr > operations for firm i in vear a. ii’

Stadardized earnings changes are computed for yezs -5 to 4 for the dividend initiation and omission firms and for the same fiscal years for the comparison matches. Hn add&ion, we ~o~mpute &qdustryadjuusted standardized

l’ksquith and MJL. ~1 inc (1983) report a two-day announcement return of 3.7% for the full initiation saqple gf ‘168 firms. Dielman and Oppenheimer (1984) report a two-day announcement return of 3.5% for a sample of 39 firms that resume cash dividends and - 8.1% for a sample of 53 firms that o:mitted cash dividends.

158 P.M. Healy and K.G. Palepu, Dividend policy and eamings information

earnings changes for the test firms, defind as the Werence in standardized earning changes for the initiation and omission Grms and comparison firms.

Mean and median earnings changes as a percentage of equity price are reported in panel A of table 3 for the dividend initiating firms for years - 5 to 4.lr Panel B reports industry-adjusted numbers for these firms. Mean stan- dardized earnings changes for the initiation firms are insignificant in years - L to - 2. In year - 1 the mean is 4.3% and is significant at the 1% level. In year 0, the yes of the dividend announcement, there is a further 5.5% increase in standardized eamings. As the results in panel B indicate, the earnings in- creases of the initiating firms are matched by similar earnings increases for the industry comparison firms in year -I. I=Iowever, the earnings increase for initiatiq &rms in year 0 cannot be attributed to industry factors. These fin&p indicate that dividend initiating &ms are in growth industries but have superior earrkgs performance in the year they Grst pay dividends.

FoIIowing the dividend announcemat there are significant mean stan- dardized earnings changes in years 1 and 2 of 2.2% and 3.5%, respectively. The year 2 increase is significantly larger than that for the industry comparison firms.‘* Mean earnings changes are ins&&ant in years 3 and 4. Median eamings change results are generally consistent with the above findings.

Thus, firms that initiate dividends experience earnings growth starting at least a year before the dividend announcement. The earnings growth continues in the year of the dividend announcement and for two subsequent years. These findings are consistent with the hypothesis that managers mnsider past and current performance as weti as expectations of future earnings in the dividend initiation decision.

One limitation of ah? m &mve rexlts is that they are based on earnings data that are reported annually, whereas changes in dividend policy are reported throughout the year. The m an nuinber or trading days betxxn the ti- rxxuxement of the dividend initiation and the 5rst subsequent annual earnings announcement date is 171. This indicates that dividend initiation announces ments occur after approximately one quarter of the fiscal year. Therefore, part of the year 0 earnings in table 3 may have been repotied in qtarteriy earnings announcements that precede the dividend announcement.

‘?he number of firms with available earnings change data differs across years - 5 to 4, she we m@e that firms have earnings data in only eight of the eleven years surrounding the dAdend announcement.

‘%ihe mean earniugs change for year I, reported in table 3, includes an obstz-&on that has an earnings decline that is 86% of price. This mrnpzuzy is Valmac Industries, which in t&L year acquiied Rite Care Corporation and recorded a large loss on this new business. If this observation is excluded, the sample mean in year 1 is significant at the 1% level in both panels A and B.

P.M. Heaty md KS. Patepu, Dividend

Table 3

surmrmq stat.istics on chaqys in eamings per share as 8 percent ot’ equity price for 131 fims initiating dividends in years surrounding the dividend mt. Sample is

197~1979?

-5 130 0.94% -4 130 1.49 -3 129 -ml -2 128 129 -1 131 4.27

0 . 130 5.58 E 130 2.20 2 130 3.50 3 128 0.35 4 126 - 3.04

om 0.01 0.07 0.01 0.82 0.12

LQ2 1.74 0.76 l.oJ 280 ‘02

3.59 282 1.95 1.51

0.a 0‘01 0.91 0.02 0.01 0.m om o.or 0.14 0.88

-5 -4 -3 -2 -1

0 i 2 3 4

hnel B: I&~mtdm-np clkulwd 127 236 0.10 021 127 0.53 0.73 0.75 126 4.27 0.02 1.14 124 - 0.35 0.87 0.66 123 1.82 0.24 1.18 124 3.84 0.01 2.@6 126 2.56 0.18 1.18 127 5.80 0.02 2&f)

119 - 0.01 0.98 - 0.19 115 - 2.16 0.41 0.63

0.36 O-0$1 0.M 0.36 0.15 O&i 0.08 0.W 0.89 0.37

Thanges in 4ixmiqs per share before extraordkkarv items and disco&iuzd operztioz~ for euh sample firm are standard&d by the &m’s SW,-+ P-L l A& AZ SW days before the ditidend initiation

TIztief&d as the first SscaI year whose earnings are armounced aftei the dividend anno-t.

‘student t and Wilcoxon statistics test the hypotheses that the mean and median earnings chaqes an5 difkezct from zero. The probability levels repoti are for two-tailed tests of si@cance.

dIndustry-adjusted earning changes are defined as the difkence iu standardized earn@s chaqgs for the initiation Grms and for matched comparison kms that arc -do& &cted from the same four+it SIC industry. Industry matches are avaiWle for 130 of the imtiation firms.

-

To exanCne whether our results are sensitive to the use of fiscal-year e&gs data, we replicate the above tests using earnings for years -2 to 0 constructed from quarterly data. Earnings chzmszs are c cuWed for yews 0 and -1 and are standardized by the stock price two davs before the dividend announcement. Mean and median values-of raw e&gs changes for the initiation ti are reported in panel fiPW@C -*mu vu are reported in panel B. Year 0 c able 3, are annomiced strictly a

160 P.M. Healy and KG hkpu, Dividend policy and earnings information

Table 4

summary statistics on alwaked quarterly eansngs change s a percent of equity $5x fur 129 fhsiniti dividends in years surrounding the dividend announcements. Sample period is

1970-1919.8

Year in relation PO dividend iIithti0~

NUmbeS offums Mean

student t probabilityc Median

WilCoMIb plDbabilityc

-1 0

Pme!A:Rmueamiin~~g~

129 5.44% 0.01 3.73% 0.01 129 4.51 0.01 3.36 0.01

-1 0

Panel B: bldus~mtedeamiip cihangesd

118 285% 0.05 1.X% 0.01 118 123 0.39 1.72 0.03

‘TWO ei the original sample of !kms initiating dividends are excluded from these tests since qwterly eamhgs are unav&labk Famings changes are estimated using quarterly eanhgs that areannouaoedinthedgbt~~priortothediti~ mtandthefourqm

the km% stock pIice two days before the dividesd menL

bYearOi§dcfincdasthefirstquartaw~eamimysare~ after~divide!nd announoemcntandthethreesubsbqwntq~

~~ttandWilccMonstatistics~tthchypothesesthatthemeaaaadmedian~ bnges are clifkent from zero. The probability levels reported are for swohkd tests of . . w Indz+djusted earnhgs changes are defined as the diffkrencc h stu&rdhd earnings changes f& the initiation Rrms and for m&bed comparison Grms that are randomly selected from thes f+xs=digit SC industry:Industry matches se available for 118 of the initiation firms.

Mean standar are 5.4% and 4.5% in years - 1 and 0, respectively, both highly In contrast to resuk for fiscal-yeear earnin%s, the eamings change in the year before the dividend announcement is

change in the event year. This ~difkrence imp&s that the fiscal understate earnings change before the dividend initiation and

anges fokving the event. The mean industry-adjusted standar- dized earnings change is s&n&ant for year -1 but not for year 0. However,

appears to be attributable to a small number of extreme observations in the comparison sample, since the median industry-adjusted change for this year is statistically sign&ant. These findings indicate that the signScant annual earnings changes for year 0 in table 3 cannot be attributed to quarterly earnings annou~& before the dividend announcement.

el A of table 5 resents summary statistkn for the dividend omission DOWArrrr NT&.! 3 LllY s andard;

in years - 5 to 4.

P.M. Healy and K. G. Palepu, Dividend policy and earnings information 161

Table 5

Summy statistics on changes in amings per share as a percent of ecpity price for 172 firms omitting dividends in years sur~unding the dividend annoucement. Sample period is 1%9-148r).a

Year in relation to dividend omissioIP

Munber of firms Mean

student t probabiity’

Wikoxon Median

-5 -4 -3 -2 -1

0 1 2 3 4

151 0.58 171 0.38 172 - 0.57 t7 2 - 1.21 172 - 7.73 172 - 13.50 l?O 6.29 169 9.37 164 2.03 151 0.22

Panel A: Raw eamings chgs

0.32 1.29 0.57 0.95 0.49 1.37 0.05 - O.&I! 0.01 - 5.10 0.01 - 9.82 0.01 4.18 0.01 5.56 0.24 3.25 0.93 1.65

-5 -4 -3 -2 -1

0 1 2 3 4

P4nel B: h&twry~usted&?l~ changes*

138 1.43 0.19 0.65 165 ‘r.ig) 0.35 0.22 l?P - I.03 0‘39 1.28 171 09.25 481 - 0.32 170 - 5.62 0.01 - 2.82 169 - 11.35 @Iv - 7.19 164 6.13 0.01 4.58 167 5.89 0.01 3.07 152 2.44 0.24 2.72 151 - 0.32 3.92 0.70

0.27 O.?l 0.17 0.47 0.01 0.01 0.01 0.01 0.01 0.17

“changes in ~UI@S pet hue before extraordinary items and discontinued operations for each sample firm are standardked by the firm’s stock price two dqs befo,c the dividend omission annmmt.

bYeat 0 is defined 8s the fkst &al year whose earnings are announced after the dividend _L

cstudent l and Nilcoxon statistics test the hypothess that the mean zud median earniq WCS are cWkrent from zero. Ihe probability levels reported are for tw*taikd icsts of significance.

dInaustry-adjusti gay* chaqes are deiined. as the Merence in stas&u&ed earnings changes for the omission firms and for matched comparison firms that are randomly sekcted from the same f--digit SIC industry. Industry matches are available for 171 of the omission firms.

The test firms’ mean standardized earnings changes ale insignihant in years -5 to -3. The means for yews -2 and ---I are -1.2% and -?.? respectively9 and are statirtM!y sigzihtit. The ‘largest decrease in me earnings (- 13.5%) occurs in year 0, t,L;e year of the dividend Q results in panel B indicate the &gs decline in yew -2 can be ntttibuted to industry-related on. However, in years - 1 and 0 the omin- son Grms have eahL -**gs &&es even after adjusting for mance. Thus dividend omissions, like divides e

162 P.M. Hea& and KG. Palepu, Dividend poky and earrings infomsiion

Table 6

Gmmuy statistics on anmalized quarterly earnings changes as a percent of equity price for 140 firms omitting dividends in years surrounding the dividend announcements. Sample period is

1969-1980.’

Year in relation to dividend omissior?

Number of firms Mean

Student t probability’ Median

wilcoxon probability'

-1 0

140 140

Panel A: Raw earnings changes

- 10.32% 0.01 - 9.97 0.01

- 6.76% 0.01 - 1.97 0.01

Panel B: Indtrrtryadjiurted earnings change#

-1 101 - 9.04% 0.01 - 7.00% Q;01 0 101 -. 6.64 0.01 - 5.88 0.01

“Thirty-two of the original sample of firms omitting dividends are excluded from there tests since quarterly earnings are unavz&ble. Earnings changes are estimated “using quart&y earnings that are announEId in the eight quartgr; p&r to the dividend announcement and the four quarters subsequent. Changes in ea_rgs per share before extraordinary items and discontinued operations for each firm are standard&d by the firm’s stock price two days before the dividend announcement.

bYear 0 is defined as the first quarter whose earnings are announced after the dividend announcement and the three subsequent quivers.

‘Student t and Wilcoxon statistics test the hypotheses that the mean and median earnings changes are different from zero. The probability levels reportt are for two-tailed tests of Significarnce.

dIndustry-adjusted earnings changes are defined as the difference in standardized earnings changes for the omission firms and for matched compatison firms that are randomly selected from the same four4li@t SX industry. Industry matches are available for 101 of the omission firms.

Following :?e dividend otission announcement, omitting firms experience two years of s&nificant posi+-5 fi vti earnings. The ezurnings increases cannot be attributed to industry factors, as they persist in panel B. These results differ from the earnings patterns following dividend initiations. The earnings in- creases prior to a dividend initiation persist for several years afterward, whereas the declines in earn~gs before dividend omissions reverse in subse- quent years.

The above findings show that dividend omissions are preceded by declines _ in earnings; :.hese declines do not persist beyond ,vear 0, Year 0 earnings include quarr.zAy eatigs announced both before and after the dividend omission. (The median number of trading days between the announcement of the divkiend omission and the Crst subsequent annual earnings announcement date is 1.36 days, about two fiscal quarters.) Hence, from the above results, we cannot infer that there are earning; declines following dividend omissions. To explore this issue further, we analyze quarterly earnings data around the

end omission date.

PM. Healy and KG. Palepu, Dividend p!icy and earnings inj’orma lion 103

Again, we redefine annual earnings so that they are strictly aligned with the dividend announcements. Mean and median changes in raw earningS (as a percentage of equity price) in years - 1 and 0 are reported in panel A of table 6 and industry-adjusted numbers are reported in panel B. The mean etigs chmgcs are - 10.3% and - 10.0% in years - 1 and 0, respectively, both significant. The mean industry-adjusted chmgcs zz -9.0% k ye;pr - 1. and -&$$!@I in ye= 0. S;ia au ===f- ycsr 0 earnings changes here are announced after the dividend omission date, these results indicate that there are significant earnings declines for up to iour quarters after the dividend omission.

In summary, dividend initiating firms have positive earnings changes for at least one year bef,,, nr~ and in the ye= of tb_e dividep_d annodr&cement; the dividend omitting firm show negative earnings changes for up to two years hefnre and the ygm of &c &y&gyJ event= m_~jse YIIVII -I w-m1 psQer2s p&s_i even after we

cmtmi lm ihe performance of the firms’ industries. They are consistent with the proposition that these dividend decisions ar% preceded by systematic earnings patterns.

Evidence on the postannouncev+ p Aw LI etigs patterns is mixed. For ini- tiations, we find earnings growth for two years after the dividend announce- ment. The increased level of earnings for these firms appears to be permanent. In contrast, the dividend omission firms have earnings declines for only one year after the divide& announcement. Further, the earnings decline experi- enced by these firms before and shortly after the omission announcement appears to be temporary. This finding is somewhat puzzling, given the large negative market reaction to the omission announcement. 0ne possible ex- planation for the postannouncement earnings performance of the omission firms is the suirvival bias in our sample, an issue we discuss later.

3.3. Relation betweeti dividend infomatkm and earnings changes

we next test -Aa+ ?. iLCgher t,he postannouncement earnings changes documented in the previous section are related to the market reaction to the announcement of the dividend initiation Or omission. If dividend policy changes are based on managers’ expectations of future earnings, there will be a positive relation between dividend znnounpement retl~rns and subsequent earnings changes. A simple regression framework is used to test this prediction for fisr;al year earnings changes in years 0 to 4 and annualized quarterly earnings changes in year 0. Tests using fiscaLyear earnings are described below; modifications to these tests for annualized quarterly earnings are discussed iz the results section.

In examining the relation between the earnings changes and the market reaction to the dividend announcement, our tests control fWS fir ~pxf~g~IQ&)fl on

future earnines from sorirces other than the dividend announcement. First, tables 3 and <in icate that the ea

164 P.M. Healy and K.G. Palepu, Dividend policy and earnings infomatioj;i

omission &ms deviates from a random walk. Therefore, prior earnings changes may be used to forecast subsequent earnings changes. The standardized ch~ge in earnings in year t - 1 is included as an independent variable in the regression model for year t to control for this earnings pattern.

Tzhe second so-urce of information on future earnings we control for is information released to the market between the earnings announcement for ‘year -1 and the dividend initiation or omission announcement. Results reported in section 3.1 show that there are significant abnormal returns before the dividend announcements, indicating that information regarding the sample companies’ future performance is relepp- A LL d&&g this period. This information could be from prior quarterly earnings announcements or anticipation of the dividend news. The market-adjusted retu& ~1 cumulated from one day after the earnings announcemezit for year - 1 to two days before the dividend date is used as a proxy for this information.

We estimate the following cross-*sectional regression separately for each year t (t=O,...,4):

where AEjl is the standardized earnings change for firm j in year t as defined in eq. (I), DREq is the market-adjusted return for one trading day before and the day of the d%dend initiation or omiss$n announcement, and PREq is the cumulative market-adjusted return from one trading day after the earzzgs announ~ment for year - 1 to two trading days before the dividend initiation or omission announcement.

The coefficient of primary interest in eq. (2) is & If dividend initiations and omissions convey information about future earnings, this coefficient will-be positive and significant. In addition, if earnings changes in year t - 1 can be used to forecast the change in year t earnings, the coeficient & will be non=zero. Finally, if information released before the dividend announcement but after the p~&sus earnings announcement is related to subsequent eam- ings performance, the coefficient & will be positive.

IQ- (2) is estimated cross-sectionally, using the 131 dividend initiating firms for - a& of the ff-ie years following the dividend mnk~u~ceae~t.~~ Ffe c&-

?‘he actuclr number of observatiotis used to estimate the regression iks each year is difkaent from the number of observations reported by year in i&r; 5 for ‘;lw’r CZGGZ: xc l 352 e&r&g changes in two successive years in each regression and we require the ~niiti ear~&qp announce- mefit date p&r to & AZ*: j - WYWXI~ anoouncemeut to calculate PMT. These additional data requirements reduce the numbe; of usabie observations.

P.M. Healy and K.G. Palepu, Dividend poky and earnings information 165

Table 7

Tests of the relation between s+adardized changes in e=rnings fo!!owing announcements of dividend initiations/omissions and the dividend announcement return. The sample comprises 131 firms that initiated dividends in the period VW-1979 and 172 firms that omitted dividends in the

period 1969-1980 (? staGstics in parentheses)?~

AEj,=~+aDRE’I;;+aAEj,,_, +&PRE;F;-+Ej,.

Year (in relation ?o (dividend

announcement Number of &iilll I% a a & R2

Panel A: Dividend initiation sample

0 124 0.029 0.197 (2.67)’ (1.72)=

11 123 0.014 0.356 (0‘89) (2.06)”

2 123 0.047 - 0.056 (3.47)C ( - S;35)

3 121 0.016 0.028 (0.87) (0.13)

4 ‘120 - 0.030 0.033 ( - 1.28) (0.12)

0.204 (1.84)’

- 0.208 (-1.57)

-0.406 ( -n.g3)=

-0.049 (- 1.45)

0.139 - 0.014 0.012 (1.16) (-0.20)

0.117 0.140 (3.77)

0.065 0.053 (1.35)

0.005 0.171 (0.13)

- 0.162 0.025 (-0.90)

0

1

Panel B: Dividhd omission sample

161 - 0.085 0.423 0.031 0.131 0.055 ( - 3.12)’ (2.24)d $24) (2.13)d

I59 0.013 0.386 - 0.683 O.lO!l 0.261 (O-40) (1.74)e ( - 7.4l)C (1.37)

2 158 0.091 - 0.095 - 0.350 - 0.077 0.179 (3.31)= (-0.48) ( - 5.63)’ (- 1.19)

3 152 - 0.036 - 0.624 - 0.017 -0.023 0.059 (-1.28) (-3.06)" (-0.24) (-0.35)

4 141 - 0.039 - 0.327 0.151 0.030 0.023 (-0.90) (-1.02) (1.12) (0.30)

“The number of observations for each year is Merent from the numbers reported in table 3 for two reasons: (1) we use earnings changes in two successive years in each regression, and (2) we require the annual earnings announcement date prior to the dividend announcement to calculate PRET.

bA E/I is firm j’s change in earnings standardized by its stock price two trading days prior to the dividend announcement; DREIj is the market-adjusted return for firm j for one trading day before and the day of the dividend anno~uncement; and PRET is t&e market-adjusted return for firm j from one tmding day fo!!ov+&g year -i’s earnings announcement to two trading days prior to the dividend an..ou.~c~ment,

CF&nit?~~t at the 1% lever using a twc+tailed test. dSigni&.nt at &z >% level using r :XG Sled test. %gniE;~t at the iO% level using a two=tailed test.

166 P.M. Healy and KG. Palepu, Dividend poky and earnings infmnation

mated co&icients are reported in panel A of of table 7.14 The estimates for the dividend announcement return coeficients (&) in years 0 and 1 are 0.197 and 0.356 and are significant a* the 10% and 5% levels, respectively, using a two-tailed test.15 This evidence is consistent with the hpothesis that dividend initiation announcements convey information about firms’ earnings prospects in the year of and the year following the dividend initiation. For a Cm with a stock price of $50, a 1% abnormal price reaction to a dividend initiation implies a 10 cent and and 18 cent increase in earnings per share in years 0 and 1, respectively. The coefficients for years 2 to 4 are insignificant, indicating that &vidend _ --se_ - announcements convey no information on earnings changes beyond year 1.

We also replicate the test of the relation between the earnings change in year 0 and the market reaction to the announcement of the dividend policy change using annualized quarterly earnings. Eq. (2) is estimated cross-section- ally for year 0 for the 129 initiating test firms for which quarterly earnings data are available. PREq is redefined as the market-adjusted return for firm Ji cumulated from one day following the previous quarterly earnings announce- ment to two days &S-z-e the dividend announcement? Table 8 reports the estimated regression coefficients.

The estimate of the dividend announcement return coefficient is 0.31 and is statistically significant at the 5% level using a two-tailed test. The estimate implies that a 1% abnormal price increase for a $50 stock at a dividend initiation announcement 3s associate0 with a 15 cent change in earnings per share in year 0. The earnings change measure for year 0 corresponds to information strictly released after the dividend announcement. Therefore, this finding provides further support for the hypothesis that dividend initiations provide information about subsequent earnings.

3.3.2. Dividend omission rewh

The parameters of regression eq. (2) are estimated cross-sectionahy using the v2 chidend omitting -f-b-- w _-rrla~ h iix sample. Separate eoauations are estimated

id White tests for hettrosk&a&&y are not sij&ficaut for my of f&e five remessi-jns [SC< &--iie (1980) for a description of this test]. lBefs!ey, Km& SC %#eisch &~anneti~ -pz~*h acwgg tic effect

of extreme observati--- -““-J; sis..e-- i__

V~ on the regression coeficients, are also examined [see BMey, Ksuh, and -Welsch (1980) for a description of these diagnostics]. - - appear to &e infl~~~~--J ’

me reported coefficient estimates do not --..ur~~ ey extreme observations.

‘%nce the information bjpothesis predicts the sign of the dividend return coefficient to be positive, a one-tailed test is probably more appropriate. Under a one-tailied test ihe coefficient for years 0 and 1 are significant at the 5% level. “Ike significance levels reported in the tables are for a mnra “nn=vative tyo-tai!ed test. U.V.V -.a-.

P6Quarterly awnings announcements for 16 initiating firms are reported on the same date as the dividend announcement. Eq. (2) is estimated after excluding these Arms and the rewlts do not differ from those reported.

P.M. Hea& and K.G. PaIepu, Dividend poiiq afir’ tti~%~gs i$cwnation 167

Table 8

Tests of the :elation between standardized changes in z!nmu&ed quarterly earnings one year following announcements of dividend initiations/omissions and the dividend announcement return. The sample comprises 129 firms that initiated dividends in the period 1970-1979 and 140

firms that omitted dividends in the period l%9-1980 (t statistics in parentheses).8.b

AEj,=&+blDREq+&AEj, -1 +&PREq+&jt. -_.._ _

.Nujmbtr of firms &J & a a R2

129

Panel A: Dividend initiation sample

0.029 0.315 - 0.178 0.123 0.086 (2.fqC (2.49)” (- 2.11)d (1 98)’

iro0 Panel B: Dividend om!ssion .cawg!e

- izi.079 0.394 -_ G.&c: < 0.032 ( - 3.42)’ (2.26Jd ; GS1j

a~u+.l:__~ - ~xning,s Me estimated using earnings for the f’c;ur quarters prior 4s the dividend announcement and the folur subsequent quarters.

bA Ej, is firm jr& ~harmse in :UII@ ~tddkd by its S~O& price two trading days prior to the dividGad initiation (emission) announcement; DRE? is the market-adgusted return for firm j for one trading day prior to and the day of the dividend initiation (omission) announcement; and PREq is the rnnrket-adjusted return for firm j from one trading day following the quarterly earnings announcement immediately prior to the dividend announcement to two trading days prior to the dividend date.

~Significant at the 1% level using a two-tailed test. “Significant at the 5% level using a two-tied test. eSignificant at the 6% level using a two-tailed test.

for each of the fiscal years 0 to 4. I7 The estimated coefficients are reported in panel B of table 7. ‘* The estimates for &, the dividend omission return coefbient, are 0.42 and 0.39 in years 0 and 1 and are significant at the 5% and 10% levels, respectivelv. in 8 ttvfi- the infcznztion $p&hesis:

. ..~aikii test. These findings are consistent with the estimates indicate that a 1% unexpected

decline in price of a $50stock at the omission announccmcnt is acccmpanied by about a 20 cent decrease in earnings per share in the year of the dividend announcement WA ruAU the following year. The estimate for year 3 is negative and significant, reflecting the earnings Ron _-very after the dividend omission docu- mented a*bove. The estimated coefficients for years 2 and 4 are not significant.

Again, we replicate the test of the relation between the earnings change in year Q and the market reaction to t 25 announcement of the dividend policy change using annuali& quarterly earnings. Eq. (2) is reestimated cross+ec-

19% rUe ~%tual number of observations used to estimate the regression in each year varies and is reported in table 7. The number for each year is di?Terent from ‘&e number of observations reported by year in table 4 because of additional data requirements discussed earlkz.

‘*Apia, White heteroskedasticity tests and Belsley, Kuh, and Welsch intIuew&d dkgqostics indicate that the regressions are well-specified.

f68 I? M. Healy and KS G Palepu5 Lbidend p&y and earnings information

tionally for year 0 for the 140 omission test firins for which quarterly earnings data are availabb.19 Table 8 reports the estimated regression coefficients. The estimate of the dividend announcement return coefficient is 8.39 and is statistically significant. Since year 0 earnings are announeed strictly after the dividend date, this finding provides further support for the dividend informa- tion hypothesis.

In summary, for both the dividend initiation and the dividend omission samples, there is a positive relation between the market reaction to the di-vidend announcements and earnings changes in the year of the dividend policy change and the year following. These results are obGned after control- ling for the earnings changes in prior years, and information on future earnings available be?zre the dividend announcement. They arc consistent with the hypothesis that dividend initiations or omissions convey informat% on future earnings performance.

3.4. Market reaction to earnings announcements after dislidend initiations and omissions

A number of accounting studies have documented that there is a significant stock price reaction to the announcement of unexpected earnings [see Ball and Brown (1968), Beaver, Clarke, and Wright (1979), and Beaver, Lambert, and Morse (198V)]. We use the following model to represent the usual relation between earnings announcement retams and the size of unexpected earnings:

where ARET Yjr is the market-adjusied return at the time of the annual earnings annottncementforfirm j(j=l,...,N)inyear t (t= -S,...,4),andbEi, is unexpected earnings based on a random-walk earnings expectation model deflated by the firm’s equity price two days bef~= *b _ ___pI ,z earn@ tilouncement. The parameters &, and & are assumed to be firm~specific. /3i is the elasticity bZ the market reaction to unexpected earnings and is expected to be positive, consistent with the findings of the earlier studies-

The previous section shows that dividend initiation or omission announce- ments convey information on subsequent earnings. These announcements emrablz investors to revise their expectations of earnings and thus reduce their forecast Errors. IIowever, measures of unexpected earnings based on a ran- dom-walk model do not reflect this additional information in earnings fore- casts. In years after the dividend initiation or omission, therefore, the elasticity

P.M. Hera@ und KS. Palepu, Dividend policy and earnings information 169

of the market reaction to unexpected earnings based on a random-walk model will be less than ‘normal.

T’o test the above prediction , wz use the following modified form of eq. (3):

The five parameters lo to h, are cross-sectional average adjustments to ,61j, the elasticity of the ma&et reaction to unexpected earnings, in each of the five years foIlowing the dividend initiation or omission. The multiplicative dummy variable D+; ties tie value 1 in ye”* w T folIowing the dividend initiation or omission and 0 nn other years. If the dividend policy announcement Ieads investors to revise their forecasts of subsequent earnings, changes in earnings wiII be noiser estimates of unexpected earnings in years 0 to 4 than in years - 5 to - 1, and the parameters X, to X, wiIl be negative.

The sample distribution of estimated t statistics is used to test the sign& canoe of the con~pan~_~pecifi~ coefkients &j and pij. For each parameter the foIlowiig Z statistic is computed

where tj is the t-statistic for firm j associated with the estimate of the parameters & or &, kj is the degrees of freedom in the regression for firm j, and N is the number of firms in the sample. The t srtatistic for firm j is distributed Student t with variance kj/(kj - 2). Under the central limit theorem, the sum of the standardized t statistics is noamaIly distributed with a variance of N. The 2 statistic for each parameter is therefore a standard normaI variate urder the null hypothesis that the parameter (& or &) is not significantly different from zero. 20 A Student t test is used to test the significance of the parameters that are assumed constant across firms (A, to h4).

X4.1. Dividend iniiiisiim ma&s

The parameters of regression eq. (4) are estimated jointly using the obsexva- tions over ten years for the 131 dividend initiation e distribution of the estimated regression coefficients ~*j anJ pi, and the estimar values of the X coefficients are shown in table 9. The sampk mean value of & is 0.2894

‘“Tbe test is parameters are

basal on th sample i dqxndent across

distributiion of the pargmetea estimates. Pt is as!!UrnP~ - -____ that tlx

170 P.M. Mealy and U. G. Palepu, Dividend policy and earnings information

Table 9

Tests of the relation between unexpected stock returns at Wall Street Journaf earn@ announce- ments and standardized changes in earnings for *years surrounding dividend initiations. The

sample comprises 131 firms that initiated drvrdends in the period 1970-1979.ab

AREqt * &j + BljAEjt + xz_,XT DrtA E/t + ejl.

Mean % statistic’: 1 sta*&tic First quartiie Me&II Third quartile Adjusted A’

0.0039 0.2894 -0.1280 -0.0422 - 0.0957 - 0.0223 - 0.X34 10.79d 7.26d

- 2.04= -0.64 - 1.49 - 0.45 - 0.27 -0.OW2 0.0003

0.0013 0.1418 0.0194 0.3738 0.272

‘The results for coefficients /& and & are for the cross-sectional distribution of time-series regression coefficients. The coefficients h, (7 = 0,. . . q 4) are assumed to be corNant across firms.

bAREql is the market-adjusted return for one trading day prior to aud the day of the WaII S&CL Journul amml eamiqs atmouncement; AEif is the change in eamiugs per share in year t standardized by the stock price two trading days pnor to the eamings announcement; aud a>,, is ci dummy variable that takes the value OEE F ‘;-- U-CT fogowing the dividend initiation announcement aud zero otherwise.

%ader the null hype____, _ the&s each 2: statistic is distibuted unit normal. dSiguificaut at the 1% level using a two-tailed test. eSigui&aut at the 5% ieve using a two-tailed test.

and is significant, indicating that the announcement of a 10% increase in standardized earnings normally leads to a 2.9% increase in stock price. However, in the year of the dividend initiation, there is a signScant reduction in the price response to eamings announcemen.ts. The estimate of this reduc- tion, AO, is -0.128 and i&cats that a iG% &=~se h standardi& earnings produm only a 1.6% stock price increase. The estimates of A,, A,, A,, and A, are not significantly &Kerent from zero, impi+g +&iS-&qti-z i-~i&on to earnings announcements returns to normal in years 1 to 4.*l

3.4.2. Dividend omission results

The coeficients of regression eq. (4) are estimated for the 172 dividend omission firms using ten years observations. These results are reported in table 10. The normal relation between unexpected earnings and stock prices is very similar to that for the dividend initiation sample, as indicated by the mean values of & and &. Also similar to what we found for the dividend initiation sample is the significant reduction in the price response to earnings

‘“An alternative interpretation of these findings is that there is abnom;lllly high noise in earnings in year 0, reducing the eamings c&Gent in this year. However, we find no evidence that tk cross-sectional variaxe of the standardized change in earnings iacreases in this year.

P.M. Meaty and K.G. Palepu, Didend policy and earnings informaZion 171

Table lu

Tests of the r&ion between unexpected stock returns at Wall Street Jountal earnings announce- ments and standardized changes in earnings for years surrounding dividend omissions. The

sample comprises 172 firms that omitted dividends in the period 1%9-1980.a*b

Mean 0.0038 0.2714 - 0.1148 - 0.1182 - 0.1323 - 0.1194 - 0.1048 z st&tic’ 4.43d 17.79d % statistic - 2.76d - 2.81d - 2.%d - 2.66’ - 2.33e First quartile -0.094 0.0794 Median 0.0032 0.1791 Thirdquartile O,OlQ? 0.3664 Adjusted R2 0,2%

“The results au1 -__~_____- cfiv nh~ffiri~nta R. and R are fcbr th* c~g~~+z+-~n ry __ ri __ --- _-- -. ,d distribution of time-series regression coefficients. The coefficients & (7 = 0,. . . , 4) are assumed to be constant across firms.

%IREqt is the market-tijusted return for one trading day prior to and +he day of the Wail Street Journd amuaI emings announcement; A Ejt is the chauge in earnhgtt pr share in year t standardized by the stock price two trading days prior to the earnings announcement; and D,, is a dummy varkble that takes the value one 7 years following the dividend omission announcement and zwo o*th\2rwise.

cLJn&~ tic? null hypothesis, each 2 statistic is distributed unit aormal. dQnificant at the 14% kvcl using a two-taiied test. significant at the 5% level using a two&&d test.

in year 0. However, for omissions this reduction seems to persist beyond year 0, as indicated by the c@ifkant negative estimates of A, to A,.

The persistence of smaller price responses to earnings in years 1 to 4 is particularly noteworthy given that the earnings changes for the dividend omission sample switch from negative to posit+ after year 0. The negative value of X in later years for the dividend omission sample cannot be attributed to information revealed by the dividend omission announcement. One possible explanation for the negative ~&F&nts is that the market receives rnoze uwuatiuubu~

*nnm wwbm~n*krr tiformation than usual on the firm’s perfor- mance once a dividend omitting firm begins to turn its performance azound.If so, earnings announcements in these years are likely to convey less informa- tion than usual. Another explanation is that there is unusually high noise in earnings in years 1 to 4, reducing the earrk~gs coefficients in these years. Tests of the cross-sectional variance of standardized earnings changes show that the variances in YWaiS 1 to 4 are significantly -higher than the preomission vari- ances.22

22The cross-sectional variance of standard* rzed earnings changes in year 0 is also higher than in years - 5 to - 1. However, the variance in this year is substantially mall,er than the variances in years I to 4.

172 P.M. Healy and K. G. PaIepu, Dividend p&y ad earnings information

3.5. Sample selection bias

As &cussed e&ier, hms inch&xi in the final sample are rewired to have edgs data available for g of the 11 years surrounding &the dividend announcement. Twenty-two percent of the original initiating sample and 18% of the initial omission sample fail to satisfy this condition. The exclusion of these companies leads to a potential selection bias. If the dividend initiation $kms that are excluded are the worst performers, the sample mean earnings haeases swrouding the dividend announcement overstate mean earnings changes for the population. Similarly, if the best-performing kms are ex- cluded from the omission sample, the sample mean earnings decline overstates the population rn=G change iu earnings.

The most frequent reasons for exclusion are that Erms are not covered on Cmnpwia ix have missing earnings announcement dates in the Wizll Street Journal. ‘I’wenty-=four initiating firms and 17 omitting kms (65% and 45% of those excluded, resptively) are excluded for these reasons. There is little evidence on the characteristics of ffirms that are not covered by Compustat or the Wall Street Journal, IIt seems plausible that Compustat and the Wall Street Journal cover fkms that are of interest to the investment community. There is no obvious reason to believe that the worst-performing initiating firms and the best-performing omitting fkms are systematically excluded from the set of kms covered by these services.

Firms with insufkient data because of new listings before and acquisitions after the dividend date account for 32% and 37% of the observations excluded from the initiation and omission samplc~, respectively. Tkre eff& UC excluding these tkns is uncertain: they could be perfor~ning either well or poorly. If the newly listed and acQuired Urms are poor performers in the initiation sample and good performers in the omission sampk, our results are biased. Once again, there is no obvious reason to believe that this is the case.

The one systematic bias that can be identified comes from excluding firms with insufficient data because they were delisted after the dividend date. It is ‘EkcZy that these fkn6 are poor performers. Since only one initiating fkm (3% of the excluded initiating kms) is excluded for this reason, the &ect of this bias on the initiating sample results is likely to be small. However, seven omitting firms (19% of the excluded omitting firms) are excluded for this reason, leading to a potentially significant understatement of the mean eam- ings decline surrounding omissions. The earnings recovery documented for dividend omission firms may, in part, be attributed to this bias.

‘5~ sscss tk p,ffwt d the above bias on the dividend otisslon findings, we dkct earnings data for the seven excluded firms for all available years from Compustat, Moodv’s Industrial Manuals, the Wall Street Journal Index, and the ?k+ ?k& IE&w I’rr&x, Mean standardized earnings changes for these seven firms are - 39.2% in year 0 and - .2% in year 1. In years 2 to 4 six of

s go out of business. he= results wnfirm that excluding these firms

P.M. Healy and K. G. Palepu, Dividend policy and earnivtgs infcmation 173

leadstoanupwardbi5~si~he p~~~~~~~~~~~~~~~rn~ni~~ngs results for the omission sample.

In summary, the eRect of the selection bias on our results is uncertain. The most identifiable bias is for tie omission sample and may explain the post- omission earnings reversal. It is puzzling that large, listed Sms appear to omit dividends after a temporary earnings decline, especially given the substantial market reaction to the omission announcement. Qne possible explanation for these nuzzles is the survival bias in our sample.

The tests reported in this paper examine whether there are significant changes in firm’ earnings performance surrounding either a dividend ini- tiaticn or omission and, if so, whether these changes are consistent with the market reaction to the dividend policy changes. We examine a sample of 136 firms that pay dividends for the first time or after a hiatus of ten years, and a sample of 172 firms that omit dividends for the East +&e or after paving dividertds continuously for at least ten years. All the dividend policy changes examined occur between 1969 andi 1980,

The statistical tests and results presented in the paper lead to four conclu- sions. First, there are signScant earnings increases/decreases for at least one year before dividend i&iat~on/omission announcements. Second, Grms have earnings increases for the y&r of, and two years following, a dividend in%Aon; these increases appear to be permanent F%ms that omit dividends have earnings declines for only one year prior to the dividend date; sub=- quently, the omission Grms’ earnings recover. Third, the abnormal stock price reactions to the dividend initiations or omissions ze correlated tith the firms’ earnings changes in the year of and the year after *the dividend announce- ments+ We find that this relation exists after controrUing for prior earnings changes and info_rmation available to the stock market at the time of the dividend announcement. Therefore, dividend initiations and omissions seem to provide incremental information on fiarms” future earnings performance. Fi- nally9 the market reaction to earnings changes is less than usual in the year following dividend initiation announcements, and for five years f&owing announcements of dividend omissions. Once again, this is consistent with the hypothesis that the dividend initiation or omission announcements anticipate subsequent earnings changes.

There are three caveats on the interpretation of our results. tests use ex post earnings time-series data, the results fiSoBe~Sflced b-f “J a sampie survivaj *bias. TICS bias could e recovelry for omission firms soon after

decisions solely to communicate their earnings forecasts. Finally5 our sample comprises only dividend initiations and omissions, relatively rare changes in dividend policy. Although they are interesting, since they have the largest

I____ ---_--- i &v&jend ip?ul,U~~m~& r\biUllD, &q ;;;&> ;AzL L’L zLFzGz_ -n** +” -4 La csWsQPra*Qa;wr s,f 6hP nfit,aja.. .JXY;r;_-=--~ -_,- _-..I ___ _ __

tion of dividend policy changes. However, Ofer and Siegel (198$,-who examine a sample of unexpected dividend changes, report analysts’ earnings forecast revisions that are c01Asteiit v4th cur Endings.

The fklings reported in this pape- r provide strong support for Lintxxr’s (1956) description of managers’ dividend decision-making process, and the &&knd information hmothesis prqosed by Maer and Modigliani (1961). Managers appear to conGder past, current, and future earnings perf~rman~ when hey da& to tieate or 0 l t cash dividends. Dividend initiation and omktion decisions are therefore interpreted by the market as managers’ forecasts of future earnings increases and decreases.

References Aiwoney, J. and I. Swary, 1980, Quarterly divide& and eamimzs anno&emenk and stock-

holders’ returns: An empirical analysis, Journal af Firnance 31, T-12. ---_ -_---

&q&h, P. and D.W. Mullins, 1983, Th e impact of initiating dividend payments on shareholders’ wealth, Journal of Business 56, ‘77-96.

Beaver: W.H., R Clarke, atid W. Writit. 1979. The association between usystematic security returns and the magnitude of ear&gs forecast erroxs, Journal of Accounting Researck 27, 316-340.

Beaver, W.H., R. Lambert, and D. Morse, 1980, The info_rmatio~ conten,! of secz+y ~:ices, Joumai of Accouut&g and Economics 2,3-28.

Belsley, D.A., E. Kuh, and R.E. Welsch, 1980, Regression diagnostics (Wiley, New York, NY). Bhartacharya, S., 1979, Imperfect information, dividend policy, and the ‘bird in the hand’ fallacy,

Bell Jour& of Economics 10,2§9-270. Bhattacharya, S., 1980, Nondissipative s&Ming structrrre azd dividend policy, Quarterly Journal

of Economics 95,1-24. Brickley, J., 1983, Shareholder wealth, information signaling and the specially dr”sigated div-

idend, TX mal of Financial Economics 12, : 37-259. mehan, ;r T .E. and XR OppmbGmer, 1984, An examination of investor behavior during periods

of large dividend changes3 Journal of Financial and Quaatittative Aza&jsis, 197-216. Fama, E,F. and H. Babiak, 1968, Dividend policy: An empirical ana@sis, Journal of the American

Statistical Association 63,113~~1161. Gonedes, NJ., 1978, Corporate signaling, external wxunting, and capital market equilibrium:

Evidence on dividends, incomz and extraordinq itetis, Journd of Accounting Re.sem& 16, 26-79.

Day, A. and U. Lowatein, 1985, Predictable eveats and excess returns: The case of dividend announcements, Journal of Financial Ikunoticf; 14; 423-450,

Lintier . rf., 1956, Distribution of income’! of corporations among dividends, retained earnings, and taxes, Am&can Econotic Review 46, 97-413.

P.M. Healy and K.G. Palepu, Dividend policy and earnings information 115

Miller, M.H. and F. Modi~ani, 1961, Dividend uoliq, growth and the valuation of shares, * burna! lef Business 34,4111-433.

Miller, M.H. and K. Rock, 1985, Dividend p&y -under asymmetric information, Journ& of Finance 40,1031-10X

Ofer, A. and D. Siegel, 1987, Tests of rational sigmlling modeis using expectations data: An application to dividend signalling, Journal of Finance 42,889-911.

Ofer, A. and W. Thakw, i987, A theory of price responses to alternative corporate cash disbursement methods: Stock repurchases and dividends, Journal of Finance 42,365-394.

Watts, R., 1973, The information content of dividends, Journal of Business 46,191-211. Watts* R. and Rr Left&h, 1977, The time series of annual accounting earnings, J~urnaf of

Accoun-Gng Rcsear& 15,253-271. White, Habert, 1980, A heteroskedasticity-consistent covariance matrix estimator and a direct test

for heteroskedasticity, Econometrica 48,817-838.