
Citation 
 Permanent Link:
 http://ufdc.ufl.edu/AA00037981/00001
Material Information
 Title:
 Dividend initiation and differential information an empirical investigation
 Creator:
 Wong, Thian Soon, 1958
 Publication Date:
 1991
 Language:
 English
 Physical Description:
 vi, 87 leaves : ill. ; 29 cm.
Subjects
 Subjects / Keywords:
 Data transmission ( jstor )
Dividends ( jstor ) Information content ( jstor ) Interest rates ( jstor ) Investors ( jstor ) Net income ( jstor ) Prices ( jstor ) Quartiles ( jstor ) Statistical discrepancies ( jstor ) Statistics ( jstor ) Dissertations, Academic  Finance, Insurance, and Real Estate  UF Dividends ( lcsh ) Finance, Insurance, and Real Estate thesis Ph. D
 Genre:
 bibliography ( marcgt )
nonfiction ( marcgt )
Notes
 Thesis:
 Thesis (Ph. D.)University of Florida, 1991.
 Bibliography:
 Includes bibliographical references (leaves 8386).
 General Note:
 Typescript.
 General Note:
 Vita.
 Statement of Responsibility:
 by Thian Soon Wong.
Record Information
 Source Institution:
 University of Florida
 Holding Location:
 University of Florida
 Rights Management:
 The University of Florida George A. Smathers Libraries respect the intellectual property rights of others and do not claim any copyright interest in this item. This item may be protected by copyright but is made available here under a claim of fair use (17 U.S.C. Â§107) for nonprofit research and educational purposes. Users of this work have responsibility for determining copyright status prior to reusing, publishing or reproducing this item for purposes other than what is allowed by fair use or other copyright exemptions. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder. The Smathers Libraries would like to learn more about this item and invite individuals or organizations to contact the RDS coordinator (ufdissertations@uflib.ufl.edu) with any additional information they can provide.
 Resource Identifier:
 027082485 ( ALEPH )
25622470 ( OCLC )

Downloads 
This item has the following downloads:

Full Text 
DIVIDEND INITIATION AND
AN EMPIRICAL
DIFFERENTIAL INFORMATION: INVESTIGATION
BY
THIAN SOON WONG
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN
PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1991
ACKNOWLEDGEMENTS
I gratefully
acknowledge
the encouragement,
patience
guidance committee
of
Professor
chairman.
I also
Robert
thank
Radcliffe, Professors
my
dissertation
Michael
Ryngaert,
Anand
Desai,
and Sanford
Berg
whose
help
and guidance
greatly
improved
this
dissertation.
I owe a special
debt
to C. Sloan Swindle
without
whose
support,
been
insight and
completed.
encouragement this
Finally,
I dedicate
this
study study
would
to
not have
my parents:
Grace
Foo Kai Chi and Wong
Cheng
Boon.
ii
and
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ABSTRACT . CHAPTERS
1 INTRODUCT 2 INFORMATI
Prior Emp Causes fo
3 DATA AND
Data . .
Tests on
4 TESTS ON
The Overa
0 0 0 ï¿½ 0 * 0 S ï¿½
ION . . . . .. ON AND DIVIDEND irical Work . . r Returns Volat METHODOLOGY . .
0 00 0
Quarterly QUARTERLY 11 Sample
0 0 500
Earnings EARNINGS
INITIATION i t  *0 as0 ility Decreases
0 0 * 0 0 0 0
page
ii iv
1 5
0 0 0 0 0
0 0 0 * S
0 O
. 0 0 0 S * 0 0 0 0 0 0
ANNOUNCEMENTS
The Sample Grouped by Informatio
The Market Reaction to Earnings
5 RETURNS VOLATILITY AROUND DIVIDE
Variance Changes in the Overall
Variance Ratio Tests . . . * The Timing of Variance Changes
The Effect of Clustered Data
6 SUMMARY AND CONCLUSIONS . . . REFERENCESE..... ... .. ï¿½ï¿½ï¿½
BIOGRAPHICAL SKETCH . . . . . . .
0 0S 0
6
11
15 15 18
35 35 40 51 57 58
61 65 71 80 83
87
S 0 0 0 * 0 0
n Availability Changes . . .
ND INITIATION
Sample.....
* 0 ï¿½ ï¿½ . . S ï¿½
o . 0 5 5 ~ 5 0 O 5 0 5 0 S S S
0 5 0 0 9 ~ ~
iii
Abstract
the Un
of Dissertation Pre iversitv of Florida
Requirements
DIVIDEND
asented to t in Partial
for the Degree
INITIATION
:11
of Doct
AND DIFFERENTIAL
'ie Graduate School Fulfillment of the
.or of Philosophy
INFORMATION:
AN
EMPIRICAL
INVESTIGATION
By
Thian
Soon
Wong
December
Chairman:
Major
Robert
Department:
Radcliffe Finance,
Insurance,
and Real
Estate
The information
transmission
hypothesis
states that
firms
begin
paying
to investors.
is partially
dividends
Evidence
as a means
that
substitutable
of transmitting
and earnings
dividend
has been
used
information information
to support
this
hypothesis.
This
is not
an appropriate
test.
Instead,
testable hypothesis proportion
implication
is that
of
of
dividend
publicly
the
information
initiation
disclosed
to
should
privately
transmission
increase
the
acquired
information.
The informativeness
of quarterly
earnings
announcements
before
and after
dividend
initiation
is examined using
returns
and standardized
variance
measures.
Under
the information
transmission hypothesis,
the standardized variance
of earnings
and dividend
announcements,
should
increase
after
dividend
iv
of
1991
m wo
initiation.
available
I find
Further,
information
that
while
firms should
returns
the least
experience
measures
the greatest
decrease
after
change. dividend
initiation,
the
low
standardized information
variance
measures do
availability
have
not.
Firms
larger
with
price
revaluations.
This
relationship
is not found with standardized
variance.
There
appears
to be
no evidence
that
the proportion
of publicly initiation.
disclosed
information
increases
after
dividend
The observed
decrease
in returns measures
of information
content
may be explained
by
a reduction
in returns
volatility
that
coincides
with
dividend
initiation.
Tests
are conducted
to detect initiation an effect
reducing
if
an information
is causing
could
bidask
effect
the returns
cause
spreads
returns
or noise
associated
with
dividend
volatility to decrease.
volatility
trading.
to decline
Variance
Such
by
ratio
tests
fail
to detect
any
such
decrease
following
dividend
initiation.
Analysis
of the timing
of variance
changes
reveals
that
dividend decreases
are equally
likely before
as after
dividend
initiation.
Therefore,
decreases
in returns
volatility
do not
systematically the contention
occur
that
after
dividend
information
initiation.
conveyed
by
This
weakens
the dividend
initiation
has caused
the volatility
decrease.
The observed
volatility
decrease
appears to be
caused
initiating dividends
by
which
amount
of
nave
firms
in the
proportion
of sample
large
period
19741977.
This
period
also
coincides
with
a decrease
in volatility
for the market
market volatility, t reduced. The general
in general.
he reduction conclusion
in firm is that
After
adjusting
volatility information
for
is much related
effects
do not
appear
to cause
the reduction
in returns
volatility.
vi
CHAPTER 1 INTRODUCTION
firms
pay
dividends
is
an enduring
puzzle
in finance.
One motivation
that
has been
advanced
is that
dividends
convey
information
dividends suggested formalized
Rock
or by
by
(1985)
to
investors.
dividend
Miller
Ross
and John
This
signalling
and Modigliani
(1977)
information hypothesis
(1961)
Bhattacharya
and Williams
(1985)
content
was
first
and subsequently
(1979) ,
among
Miller others.
addition proposed which in refer to
to this
that
signalling
initiating
formation
this
about
expanded
hypothesis,
dividends
some
may augment
firms is transmitted role for dividends as
researchers
the
have
process
to investors.
by
I
the information
transmission
hypothesis.
For example,
Asquith
and Mullins
(1983)
state
that
Dividend
policy
has several
attractive
information transmission mechanism.
focus of other a simple, c interpretation
future
prospects.
announcements must firm must either capital markets t
credibility of cas] visible compared w dividends are i
anticipate management (p. 94)
announcements, .omprehens ive of the firm's :
Unlike most be backed generate th :0 supply j
[ signals,
tith
other
nitiated,
a periodic is forced to
Unlik
dividen
signal recent
aspects as an .e the detailed
Is can be used as of management's
performance
announcements,
wit] is Lto
hi hard cash c
cold
and its dividend
cash.
r
Drconvlr
In addition
dividends
are a
announcements. shareholders
signal submit
by
to
manage
The
ce the to the
lso highly 00. .Once apparently ment, and
a periodic review.
Why
of
and
In
Two recent (1988) report
studies results
by
Venkatesh
that
(1989)
and Healy
can be interpreted
and Palepu
to support
this
view
of why
firms
pay
dividends.
If dividend
information
substitutable announcements
for
before
earnings
dividend
information,
initiation
then
should
earnings
convey
more
information t announcements.
informativeness
by
the amount
:han
postdividend
Venkatesh of quarterly
of price
finds earnings
revaluation)
initiation
that
the
announcements
decreases
after
earnings average (measured dividend
initiation.
Healy
and Palepu
find
that
the price
response
annual
earnings
information
is lower
in the
years
following
dividend
initiation.
I contend
that
the above
results
merely
show
that
earnings
and dividend
substitutability
information
is
substitutable.
is not compelling evidence of
Demonstrating the information
transmission
hypothesis.
Why
would
a firm
engage
in costly
information transmission
via dividend
payments
when
earnings
reports pursue
will
serve
an enhanced
the same
purpose.
information
I argue
transmission
that
a firm
policy
would
only
if
such
a policy
investors. investors
reduces
Firms
the cost
for which
of information
it
to acquire information
is
will
relatively
have
acquisition
costly
greater
for for
incentive
to disclose
information
through
dividends.
Consequently,
dividends
firm,
then
augment
the
the available
relative
public
amount
of
informatio private
about
the
information
acquisition
after
dividend initiation should
decrease.
is
to
if
I test comparing
the above
the
average
information standardize(
transmission d variance oI
hypothesis
by
announcements
from the
pre
and postdividend
initiation
periods.
The
use of
standardized
variances,
i.e.
announcement
period
variance
divided changes
by estimation period
in returns
volatility.
variance,
avoids
Standardized
problems
variance
with
measures
the amount dividends a privately periods. I
publicly
of "new"
i
innouncements acquired d
f dividend available ti
nformati relativ uring initiati
O
standardized variances of
[on
released
re to the amount surrounding nc .on alters the
privately
earnings and
acquired
dividend
at earnings
and
of information onannouncement
proportion
of
information, announcements
should
increase
after
dividend
initiation.
Further,
one would
expect
firms
with
low information
availability
relative
hypothesis,
to experience
informativeness
of public
I use two proxies
increase
announcements.
of information
in the
To test this
availability
(namely,
Journal
market
Index
value
news
of equity
items)
to divide
of Wall
Street
of dividend
initiating differences,
firms
across
into
five
groups, in the
groups.
relative
I then
test
for
of
these
public
announcements
before
and
after
dividend
initiation.
The results
from
this
study
do not support
the hypothesis
that dividend
initiation
changes
the relative
amount of
public
versus
no differential
the greatest
and number
a sample
informativeness
private
information.
Further,
I f ind
changes grouped
in relative
by
information
informativeness
across
of firms
availability.
If changes
findings
that
in information
Venkatesh
and Healy
are not the basis
and Palepu
report,
for the then what
is? Venkatesh postdividend
finds
that
period.
returns
I present
variability
evidence
that
is lower
their
in the
results
are driven volatility
by
a contemporaneous
of firms
initiating
decline
dividends.
in the returns
This
decline
leads
Venkatesh
and Healy
and Palepu
to interpret
postdividend
initiation ea I investigate
rnings
announcements
the nature
of this
as being returns v
less
informative.
olatility
decline.
I find
driven
no evidence
by
that
information
the decrease
related
in returns
volatility
causes.
This
paper
is organized
as follows.
Chapter
2 develops
implications
of the information
transmission
hypothesis.
critique
the Venkatesh
and Healy
and Palepu
studies
within the
framework
of this
for the decline
hypothesis. in returns
I also
discuss
volatility
possible
following
causes
dividend
initiation.
Chapter
3 presents
my
data
and the methodology
used to
examine
the information
content
of quarterly
earnings.
In Chapter
4,
I report
and discuss
the results
of tests
pre
and postdividend
5 presents summarizes
tests
initiation
on changes
quarterly
in returns
earnings.
volatility.
Chapter
Chapter
the results
is
the
on
samples
of this
study .
INFORMATION
CHAPTER 2 AND DIVIDEND
INITIATION
Asquith
and Mullins
(1983)
hypothesize
that
dividends
provide
information
a vehicle
for communicating
the
firm's
management' s
current
and
superior
future
prospects.
Information
about
a firm
can come
from
two sources:
information information)
provided
by
and information
the
firm
privately
(publicly
acquired
by
released investors.
I assume
that
firms
have
a cost
advantage
in producing
information
about
themselves.
Therefore,
managers
prefer
release greater
information publicly
than
have
cost*1
One way
dividends
can transmit
information
to investors
by providing
substantiation
for the cashflows
reported
in the
earnings announcement.
Without dividends,
investors would have
to seek
substantiation
from
other
costly
sources.
Venkatesh
(1989)
alludes
to this
for the postdividend
less
in his
paper.
in returns volatility
is that
'information'
(announcements/rumors)
that could have
induced price reactions
in the predividend
period"
(p. 176).
If dividends
play an
It is not necessary to assume that
altruistic purposes. By managers obtain a lower
managers
reducing the costs of
required rate of retu
do this
for
investigation, rn.
concerning
may
investors
to
produce
it at
is
accord
argument decrease
importance
His explanation
to pieces of
informative
private inf dividend in proportion
role,
ormation itiation
of
announcements,
information
the major tc acquisition
period.2 As
information
investors
privately.
stable will dE the firm
through
have
Therefore,
less
implication
crease
discloses
earnings
incentive
is that
in the post
a greater
and dividend
to acquire
proportion
of
public
versus
private
information
will
increase
following
dividend
initiation.
I make
two assumptions
to test
this
hypothesis.
substitutable announcements.
First,
for
Second,
the information
information the total
conveyed imparted
amount
by
by
is
earnings
information
does
not change
with
dividend
initiation.
Prior
Empirical
Work
Venkatesh
(1989)
studies
a sample
of 72 NYSE
or AMEX
firms
that
initiated
quarterly
dividend
payments
between
19721983.
He collects
the announcement
dates
of quarterly
earnings
dividend
initiation. announcement
that
Using
for 14 quarters
raw and
as a measure
information
excess
before returns
of information
content
and after
from
dividend
the earnings
content,
he finds earnings
announcements
before
dividend
initiation
is significantly
greater.
Based
on this
evidence,
he concludes
that
earnings
The issues
implications framework by Verrecchia (
of information
are formally
Hellwig (19 1982) and Bhu
explored
80) , Diam
ishan
O
substitutability
and its
in a rational expectations ,nd and Verrecchia (1981),
(1989) .
the relative
dividends
of firm
payments
and
the average
of quarterly
and dividend information
information
is transmitted
are partially
by
substitutable
dividends.
Venkatesh
dividend
also
finds
initiation.
that returns volatility
His explanation
for this
is lowe
decrease
r after is that
investors
in the postdividend
initiation
period
attribute
less
importance
to
information
arriving
during
non
announcement
and
lower
investors passively
periods. observed
do not activ(
receive
This
leads
volatility. ely acquire
information
to smaller
Venkatesh information.
and decide
price
reactions
assumes Instead,
whether
that
they
to act
on
this
information.
Before
dividend
initiation,
they
are more
likely
to act
on nonannouncement
period
information.
In the
postdividend
period,
investors
receive
more
publicly
announced
information
and rely less
on nonannouncement period
information.
The main
determine
about
future
focus
whether
of Healy
dividend
earnings.
They
and Palepu's initiations hypothesize
(1988)
convey
that
study
is to
information
investors
view
dividend cash, of enable i forecast
initiation
changes
as management
future earnings
investors
errors
should
changes.
to revise
in earnings
therefore
forecasts,
So,
substantiated
dividend
expectations announcements be reduced.
by
announcements
of earnings.
following
The
dividend
To test
this
hypothesis,
they
obtain
the price
response
(namely, surprise
twoday
cumulative
excess
returns)
and earnings
standardized
and that
earnings
by price)
(the
change
for5
annual
earnings
initiation.
They
announcements
regress
prior
the price
to and fo] response
lowing dividend on the earnings
surprise measures
to obtain
the earnings
the informativeness
response
of the earnings
coefficient.
surprise.
This
Other
things leads values
being equal,
to large
earnings
price
of the earnings
information
revaluations
that
and ther
response coefficient.
is unanticipated efore to larger
They
find
that
the earnings
response
coefficient
is lower
in the post
dividend dividend
initiation initiation
period.
provides
Healy
and Palepu
information
which
argue
that
preempts
subsequent
earnings announcements.
Do Venkatesh's
and Healy
and Palepu's
results
regarding
earnings
provide
support
for the information
transmission
hypothesis?
argue
they
might
not.
These
studies
suggest
that
dividend
and earnings
information
are substitutable.
Managers
have
little
motivation
to initiate
dividends
if doing
merely
shifts
be conveyed
to dividends, by earnings.
information
This
that
would
is especially
otherwise
true
since
dividends
are costly
to investors.
First,
they
are costly
because
firm
dividends
chooses
the extent
instead
that
are taxed
to retain
al gains a
capitz
at ordinary
earnings,
income taxes
re postponed.
rates.
are r
Second,
I f the
reduced to dividends
must used
be paid
out of net income
for investment.
Replacing
that
would
otherwise
this source of
have
capital
been
incurs
transactions
costs
that
are ultimately
borne
by
investors.
dividends also can
so
Third ,
initiating
lead to shifts in
clienteles
costs
must
that
are costly
of administering a
play
documented
to investors.
dividend
a substantially
by
Venkatesh
policy.
greater
or Healy
Finally, Therefore,
informative
and Palepu
there are dividends
role
to justify
than their
cost.
I hypothesize
that
this
greater
role
must
be the increase
in publicly
costly
Healy
two
released
private
and Palepu
reasons.
information
information
studies
First,
and
a consequent
acquisition.
cannot
address
in the postdividend
decrease
The Venkatesh
this
hypothesis
initiation,
there
two routes
through
which
public
information
is transmitted:
dividends public in
comparing neglecting
and earnings.
iformation
only
A priori,
is divided
earnings
the amount
we do
between
announcements,
of information
not know
these ti
how the total
wo routes.
the researcher
conveyed by dividends.
By
is He
can infer
it.
From
the effect inference,
of that
he
information,
can only
but cannot
conclude
that
quantify dividend
information
has substituted
for earnings
information.
As
an illustration,
information
is
a. This
assume amount
the total
remains
amount
constant
of available
for
pre
and
postdividend
initiation periods.
Suppose that before
dividend
initiation,
privately
half
(a/2)
acquired.
is publicly
After
dividend
disclosed an initiation,
id half
is
earnings
disclose accounts
a/3,
dividends
for the remaining
disclose
third.
a/3 and private
We
see that
acquisition
after
the information
dividend publicly
in
and for are
is
initiation,
twothirds
of
10
disclosed.
Yet if
one observes
only
earnings
all
one can
conclude declined
is that
from
a/2
the information
to a/3.
We can
content conclude
of earnings
little
about
much
information
was transmitted
by
dividends.
Any attempt
gauge
the relative
proportion
information must examine the and dividend announcements.
of
public
informativeness
versus
of both
private earnings
The second
deficiency
in the Venkatesh
and Healy
and Palepu
studies
lies
in their
information content measures.
They
base
their
measures
on announcement
period
returns
that
do not
account
for the general reduction in returns
volatility.
This
decrease in returns volatility
may
be due to causes
unrelated
to any
information
effect associated with dividend
initiation.
Suppose
that
announcement period returns
variance
decreases
proportionally
with
the decrease in returns volatility.
Then
one
would
returns
observe
that
are
postdividend
lower
in
initiation
magnitude
than
announcement predividend
initiation distinguish
announcement
a decrease
returns.
If this
is
in announcement period
Sol
returns
one cannot
as being
due to information
related causes or otherwise.
A study
of the
nature of changes initiation is calle
in returns
d for. In
volatility
the next
around
section,
dividend I review
a decrease in returns volatility.
has
how
to
several possible causes for
11
Causes
for Returns
Volatility
Decreases
Information
related
causes.
In informationally
efficient
markets
where
new information
is rapidly
incorporated
into
price,
returns
volatility
is directly related
to the rate
information
reduction possible
arrival.
in the rate
that
when
Lower
volatility
at which
returns
may
information
on the firm's
be caused arrives.
underlying
by
It is
assets
become returns
less
uncertain
volatility
may
and
more
occur.
predictable,
However,
this
a decrease
effect
does
depend
on any changes
disseminated
in the
or processed
process
about
by
which
the firm.
information
The decision
initiate
dividends
may
coincide
with
the firm's
asset
returns
becoming less uncertain.
then
only
coincidentally
Decreases associated
in returns
with
dividend
volatility
are
initiation.
The initiation
of dividends
does
not
cause
the rate
information
An
arrival
alternative
to decrease. information
related
explanation
decreases
reduces
ways
by
in returns
noise
Black
trading. (1986).
volatility
Noise has b( I consider
is that
dividend
een interpreted
noise
initiation in various
as the difference
between
the security's
intrinsic
value
and its observed price.
This
mispricing
occurs
when
traders
overreact
activities
of other
traders
(French
and Roll
(1986)).
Greater
noise could
trading reduce
leads noise
to higher trading
 volatility. I if it induces
Dividend greater
initiation consensus
among traders.
of
in
not
is to
of
for
to
the
12
The bidask
spread.
Firms
with
large
bidask
spreads
tend
to have
greater
in several
ways.
returns
First,
volatility. the larger
This
relationship
the bidask
arises
spread
the
larger t] Observed
he difference
returns
between
are calculated
buy
and sellside
using daily
transactions.
closing prices.
If
successive
closing
transactions
are alternatively buy and sell
transactions,
bidask greater
spread.
they would
Hence,
the observed
be conducted the greater
at opposite the bidask
edges
spread,
of the
the
return.
Bidask (Demsetz,
spreads increase with
1968
Ho and Stoll,
the dealer's
1981)
and
inventory
costs
the degree
of
informational
uninformed
asymmetry
dealer
(Bagehot,
between
1971,
informed Copeland
traders
and Galai,
and the
1983,
Glosten
and Milgrom,
1985).
It is possible
that
dividend
initiation may alter these
two factors
and therefore
reduce
bidask trading
spreads. volume.
For example,
This
reduces
dividend
the dealer's
initiation may
inventory
increase
costs
as
inventory reduce ii
turnover
nformationa
increases.
1 asymmetr)
Dividend y through
initiation greater
may
also
consensus
between
traders
and dealers.
Stock price.
Volatility decreases as stock price
increases.
This relationship
bidask observed
spread, returns
suppose stock
Al
occurs
higher
for two priced
reasons.
stocks
First,
will
and therefore lower volatility. Aas bid and ask prices of 9 7/8
for the
have
same
smaller
For example, and 10 1/8,
13
respectively; stock
B has bid and ask prices
of 19 7/8
and 20
1/8.
sell
The bidask
and buy
spread
transactions
is 2/8
for both
the observed
stocks.
return
For successive
for stocks
Aand
B is 2.5%
and 1.24%,
respectively.
Second,
as a firm's
stock
price
increases
(without
the firm
increasing positively
considering
debt) ,
related a firm
the firm's
leverage decreases.
to leverage.
in
This
a ModiglianiMiller
Volatility
can be easily
world
with
seen
is by
no taxes
and riskless
debt.
The return
on a leveraged
firm
= kv + (kvr)
where
r is the riskless
the unleveraged
firm.
interest
The market
rate and
value
is the return
of debt
and equity
D and S, respectively.
Taking
variances
= or(1
Therefore,
stock
volatility
us depends
positively
on leverage
D/S.
Interest
rates.
leverage. decrease
rates.
Interest
Volatility
rates
An increase
leverage.
indirectly in interest
First,
also
changes
affect
rates
as interest
with
volatility
can both
rates
rise,
interest through
increase the value
or of
both
debt
and
proportionately
equity greater
both than
decline
debt;
this
but
leads
equity
to
declines
an increase
transfer from bond to
is
D
()
S
on is
Os
D + 8
there is a wealth
in leverage.
Second,
14
equity
holders
that
decreases
leverage.
Christie
(1982)
studied
that
the net effect
high
interest
of interest rates on leverage.
rates
are associated
with
high
He finds returns
volatility.
If dividend
initiation
has information
related
effects,
volatility
can decrease
due to reductions
in noise trading and
bidask
spreads.
Volatility
can also
decrease
due to factors
unrelated
to information
effects.
The returns on the
assets
firms
initiating dividends
could
have
become
less uncertain
and thus could e>
prices
experience
are less
increases
volatile.
in stc
Dividend Dck price
initiating which le
firms id to
proportionally
Finally, changes
returns
smaller
bidask
volatility
spreads
and reduced
leverage.
can decrease due to market
wide
in interest
has
of
n.,
rates .
DATA
CHAPTER 3
AND METHODOLOGY
Data
My
data
dividends identified
Security collect
sample
between
consists
19701986
by scanning
Prices
earnings
(CRSP)
of
332
firms
inclusive.
the 1988 NYSE/AMEX
information
that
These
Center
and NASDAQ
for approximately
initiated
firms
for Research
files.
were
in
I also
14 quarters
trading
days)
before
and after
the initial
dividend.
These
were
obtained
from
the 1987
quarterly
Compustat
files
and/or sample,
the Wall
firms
Street
must
meet
Journal
Index.
the following
To be included requirements.
in the
The initiating the firm's cc
dividend )rporate
is either the
history
first dividend
or the resumption
in of
dividend
payments
after
a hiatus
of at least
10
years.
The initial
dividend
cannot
be
an extra
or special
dividend. The firm
three
must
years
continue
after
paying
the initial
dividends dividend.
for at least
The firm
must
have
3 years
of daily
returns
data
before
and after
the initial
dividend.
Not
preceding and
one CRSP
tape
all
the returns
following source. F(
data
dividend ourteen f
for th(
initiation
irms
which
n. 3 year period are obtained from
were
on the NASDAQ
15
(900
(1)
(2)
(3)
16
One year appearance
announcement
must
elapse
between
on the NYSE/AMEX
of it's
initial
the
or NASDAQ dividend.
firm's
tapes
initial and the
In each
initial
14 quarter dividend,
period
preceding
announcement
dates
and following and earnings
price
data
for at least
10 quarters
must
be available.
This
eliminates
firms
that
only
report
annual
earnings
or f irms
merged
out of existence.
Table
dividend
1 presents initiations
an overview occurred in
of the sample.
the middle
Most
seventies.
of the Almost
65% of the sample
initiated
dividends
between
1975
and 1977.
This
could
controls
as posited
be due to the relaxation
or to crosssectional
by
Marsh
and Merton
dependence
(1987).
of the Nixon n. in earnings c
More
than
price ,hanges three
quarters quarters dividend
yearly
of the sample following di
payment frequ
and annual
paid
vidend
ency
payments.
at least
initiation.
was quarterly,
American
9 dividends
The mot followe
Exchange
in the 14 Dt common
by
listed
half firms
form
a slight
maj ority
followed
by
New York
Stock
Exchange
OTC firms. quarterly
More
than
earnings
90% of the sample
had at least
years
data.
tape when NYSE/AMEX tape when
they tape.
initiated
Similarly
they initiated
dividends , 6 firms dividends
had continuing data on the
which were on the NYSE/AMEX had prior NASDAQ data.
(4)
(5)
the and
and
of
17
TABLE 1
OVERVIEW OF SAMPLE CHARACTERISTICS
_______________Icumulative
Freq.
Percent
Freq.
Percent
___________________ ____________________ A  A _____________ L _______________ I _____________I Y Y I
Year of
Dividend Initiation
1970
.8
1.8
1971 4 1.2 10 3.0 1972 18 5.4 28 8.4 1973 31 9.3 59 17.8 1974 27 8.1 86 25.9 1975 62 18.7 148 44.6 1976 70 21.1 218 65.7 1977 49 24.8 267 80.4 1978 23 6.9 290 87.3 1979 12 3.6 302 91.0 1980 9 2.7 311 93.7 1981 6 1.8 317 95.5 1982 5 1.5 322 97.0 1983 6 1.8 328 98.8
19
86
.2
32
100.0
____________________ A L ________________ a ________________ ______________Y F 1 1
Number of Dividendsa
24
26
7.8
26
7.8
58 50 15.1 76 22.9 912 53 16.0 129 38.9
13
16
20
1.1
32
100
.0
___________________ I ____________________ A ï¿½ ______________ _______________ i _____________T 1 1
Dividend Payment Frequencyb
Annual
21
.3
21
6.3
Half 35 10.5 56 16.9 Quarter 217 65.4 273 83.2
Other'
59
17
.8
332
& I I I
100.0
18
TABLE
 continued
Cumulative
, , i
Freq.
Percent
Freq.
Percent
NYSE 104 31.3 104 31.3 Exchanged AMEX 143 43.1 247 74.4
NASDAQ 85 25.6 332 100.0
1619 12 3.6 12 3.6 Number of 2023 42 12.7 54 16.3 Quarterly
Earningse 2427 105 31.6 159 47.9
2830 173 52.1 332 100.0
, , i i i~ iii i iii O
a Number
days f
b The fr
of dividends
ollowing "equency
which u
the initial
of dividend
were
declared
dividend. payments
within
in the 3
900 trading
year
period
following dividend Within the 3 year
firms
changed
frequency. firms chan
from
6 firms
ged
frc
from quarterly tc to half yearly
yearly with d Exchange on
initiated.
e Number
of
declaration information
initiation.
period
follow
an annual
changed
)m half half
froi
yearly yearly.
t(
2
to quarterly.
additional
which
firm
quarters
yearend
ing dividend to a half yE m annual to , o quarterly. ] firms change 1 firm chang
dividends
was trading
within
of the initial was available.
900 tr
dividend
when
initiation,
early quarte
payment rly. 44
firm changed from annual
from
half
to quarterly.
dividends were
ding days
for which
of
the
earnings
Tests
on Quarterl
Earnings
Measurinq
information
content.
The information
content
an earnings
announcement is computed
in the following way.
announcement announcement
(t+1)
period
(day
the announcement
consists
of 3 trading
the day preceding
date.
Beginning
days:
(t1)
at days
the day
of the
and following t2 and t+2,
the announcement
of
The
are
"M "
i
,'a(
period
surrounding
for 60 days
returns
19
collected.
dividends
These
returns
and other
form
earnings
the nonannouncement
announcements
occur
period.
within
If
nonannouncement
period,
the returns
from
these
announcements
are excluded.
The 60 nonannouncement
period
returns
are used to
estimate
the market
model
for firm j:
= () +ij ,
r=i,60
where
are the returns
on day
T for the firm
j, r P
are the
returns
(including
dividends)
of
a valueweighted
index
all firms
on the tape,
and
ej and 3j
are OLS estimated
coefficients.
Excess returns uo, are calculated
 (aj c i ma)
I assume
zero
that
expected
excess
value.
returns
are normally
distributed
with
Three
measures
of information
content
are used.
The first
is the standardized
variance,
a variant
of Beaver's
U (see
Beaver,
1968) .
It is denoted
by
BVU.
This
is the ratio
announcement
period
returns
variance
to nonannouncement
period
returns
or marginal
variance.
information
Beaver' s
conveyed
is
a measure
to the market by
of the "new" the earnings
dividends)
announcement
relative
to
the
average
information
also
disclosed
available
during
of
versus privatel)
the nonannouncement
amounts
formation.
period.
It
of publicly I assume that
returns variances
this
Rj
for
Uj'
as
(and
of
is
a measure
the relative
y acquired inf
= Rjr
nonannouncement
period
capture the
extent
20
of private
true
as firms
information acquisition.
may engage in other
This may not
forms
of public
be strictly disclosure,
for example,
minimized
earnings
if dividend
forecasts.
initiation
However,
does
their
effect
is
to changes
in the firm's
other
public
disclosures.
The
changes
use of standardized in returns volatility.
of circumstances.
First,
variances
This
also
is critical
if the market
were
controls
under two generally
for
sets
more
variable
during
a certain
period,
a direct
comparison
announcement
period
returns
would
bias
the test
of firm
specific
price
response to
earnings
announcements.
This
bias
is exacerbated
if the sample
is clustered
in calendar
time
clustering
is coincident
with
changes
in market
volatility.
Second,
if
volatility Standardizing
there
is
a systematic
contemporaneous the announcement
with period
decrease dividend returns va
in
returns
initiation.
triance
by
the
nonannouncement
arise
previously
from
period
returns
differences
results
variance avoids
in volatility.
may
be driven
by
problems
I suspect
returns
that
that
variance
decreases and
not by changes
in information
transmission.
Let A represent
the set of announcement
period
days
and N
represent estimate
the set of 60 nonannouncement
the announcement
period
returns
period
variance
days .
to be the
sum of the squared excess
returns u
(T
EA)
divided
number
of days
within
the announcement
period,
period returns variance is
TA. Similarly, the sum of 60
not systematically
lead
of
and
may
reported
by
the
the nonannouncement
21
squared
excess
returns
UT
(7r
c N)
divided
standardized
variance
is calculated
Var [ ujj,
Var [uJ
,rETA]
, rcTN]
1 2 TN 1'eTN
Standardized
the announcement
variance
period.
is calculated In the first
for two definitions
definition,
BVUE,
announcement announcement. Venkatesh's r
period
This
esults.
consists
definition
As BVUE
only
is used
measures
of
the
earnings
for comparison
the marginal
with
information
content
of earnings
relative
to the
average
information
in the
nonannouncement substitutability
period,
of
this
earnings
ratio
and
is used
dividend
to examine
the
information.
Decreases
in postdividend
initiation
BVUE's
indicate
that
dividend
announcements
have
preempted
earnings
announcements.
In the second
definition
I BVUD,
the announcement
period
consists
of the earnings announcement
announcement announcement. effect of div
postdividend
occurring
This
ridends
within
definition as a public
initiation
period,
30
and the nearest
days
is used
to
disclosure
of
the
dividend earnings
measure the total mechanism. In the
information
is publicly
disclosed quarter,
through
BVUD
both
measures
dividends
and
the information
earnings.
content
In each
of all public
disclosures
(from
both
earnings
and dividends) in postdividend
relative
by
as
60.
The
BVUj
of
the
to
private
information.
initiation
Increases
22
BVUD's
indicate
that
more
information
is released
publicly
than
acquired
privately.
I also
calculate
two
measures
which
are
based
announcement
period
returns.
They
are similar
to those used by
Venkatesh.
They
are measures of
price
revaluations
induced
information released at earnings announcements.
These measures
do not take
into
account
nonannouncement period
returns
therefore,
the general
level
of returns
variability.
The first
measure,
RAW
. is the absolute
value
of the
sum of earnings
announcement
period
returns.
RAW
The second,
MAJ,
is the absolute
value
of the
sum of earnings
announcement
period
excess
MAJ
returns.
=1
=t+1 :=t1
These
measures
differ
only
to the extent
that
the market
model
regression
accounts
for contemporaneous
market
movements.
Because
the explanatory
power
of market
model
regressions
typically
low,
these
measures
are quantitatively very
similar.
The values
for MAJ are usually
slightly
smaller
in magnitude
than
those
obtained
with
I define
the
pre
and postdividend
windows
to be 900 day
periods
(about
14 quarters)
preceding
and following
initiation of dividends.
on
by
and
=1
RjT
is
RAW.
the
t~l
E
T=ti
UjT
A"j oint" announcement
occurs when
23
a dividend announcement.
is
declared I follow
within
2 days
Venkatesh
in
of
an
excluding
earnings
joint
announcements
from
the analysis.
I require
that
in each
window,
there
are at least
3 usable
earnings
announcements.
This
requirement
eliminates
7 firms
from
the sample.
The final
sample
consists
of 325 firms.
pre
firm
j I
average
the information
initiation
content
quarters.
measures "Matched
differences"
are obtained
by subtracting
the
average
post
dividend dividend
initiation initiation
information information
content content
from
the
for each
average
firm.
DRAWj DMAJ DBVUEJ DBVUAj
= RAWj,PRE =AJi, PRE
= j, PRE = BVUPRE
RAWj, POST
=Jj, POST
BVUEj, POST
 ]D UAj, POST
The distribution
of matched
differences
i s then
analyzed
detect
the effect
of dividend
initiation.
Hvpotheses
and
test
statistics.
There
are
2 main
hypotheses.
The
first
is
that
dividend
information
substitutable
one would
expect
announcements
to
to observe decrease
information.
raw price
in
the
Under
this
reactions
postdividend
hypothesis, of earnings
period.
the appropriate
For each
of all
and postdividend
pre
to
to earnings
is
test is
Therefore
24
DRAW
DMAJ
o0
o0
versus
versus
Ha Ha
DRAW DMAJ
Further,
if the total
amount
of information
or the general
level
of volatility
does
not increase
from
preto
post
dividend measure
initiation pe
of information
n.riods, t content
then
the standardized
should
also
variance
decrease.
: DBVUE
 0
versus
Ha
: DBVUE
second
hypothesis
is that
dividends
augment
transmission
of public
information
about
the firm.
Under
this
information
transmission
hypothesis,
the relative
information
content
of public
announcements
(both
earnings
and dividends)
to private
information
should
increase.
Therefore,
: DBVUD
 0
versus
Ha
DBVUD
These
hypotheses are tested using standard
t statistics
Wilcoxon
signedrank
statistics.
The nonparametric
Wilcoxon
signedrank
test
is particularly
(for
suited
example,
to the
see Hollander
case
of paired
and Wolfe, replicates
1973) data
i.e.
pairs
of
are concerned
prewith
and posttreatment
a shift
in location
observations,
where
we
due to the application
of the treatment.
In this case,
the treatment is
the payment
techniques require
H0 H0
> 0
and
> 0.
H0
The
> 0.
the
H0
<0O.
and
of dividends .
Nonparametric
few assumptions
25
about
the underlying
obtained. I traditional
n particular, assumption
populations
from
nonparametric
that
which
procedures
the underlying
the data
forego
populations
normal.
The Wilcoxon
signedrank
statistic
W is calculated
follows.
ranked
The absolute
in
ascending
values order.
of the matched
The ranks
of
differences
those
are
matched
differences
that
have
the sign
of the hypothesized
direction
(i.e.
DRAW>0,
DMAJ>O
I DBVUE> 0
and DBVUD<0)
are summed.
Denote
this
sum as
0I use the large
sample
approximation
to obtain
W ï¿½
 [
n(n+l)
4
n(n+l) (2n+l)
24
The statistic
has
an asymptotic
standard
normal
distribution.
a two tailed
test,
the null
hypothesis
is rejected
if 1w
> (a/2),
where
z(.)
denotes
the standard
normal
variable
a the significance
level.
Differential
information
availability.
I also
test
information transmission hypothesis by examining whether firms
grouped by content of
information
earnings
availability have different
announcements.
information
If information transmission
a valid
incentive
for initiating
dividends,
then
one would
expect
firms with
low information
availability
to experience
are the
are
as
T
For
and
is
the
26
the greatest
increase
in the relative
informativeness
public
announcements.
To test
this
hypothesis,
I group
firms
using value
two proxies
of equity
of information
and number
of Wall
availability
Street
(namely,
Journal
Index
market
news
items) .
these
I test public
for group
differences
in the
before
and
informativeness
after
of
dividend
initiation.
Previous
studies
(1989) ,
by
Atiase
Shores
(1985)
(1990) , Zeghal
Bhushan
(1984)
(1989) ,
and Grant
Lobo
(1980)
consistently
find
that
the earnings
announcements
for firms
with
low information
availability
have
greater
information
content.
These
studies
group
their
sample
firms
on proxies
information
availability
such
as firm size,
analyst
following,
trading number 4
volume,
of market
exchange makers a
listing,
and bidasi
financial < spread.
media r( The most
eportage, commonly
used
proxy
of information
availability
is market
size.
example,
(1989),
earnings
have
Shores
Atiase
lower
already
(1990) (1985) a
announcements
information
have
more
Lobo
nnd Zeghal of large
content.
information
nd Mahmoud
(1984) al.
market
This availah
(1989), 1 report
Bhushan
that
capitalization
large
them ai
the
firms firms nd the
additional
contribution
of earnings
information
is smaller.
Large several
firms h reasons.
iave
Large
firms
information
may
have
availability
economies
of sc
due to ale in
producing
and disseminating
information
about
themselves.
Large
information to meet
of
announcements
Mahmoud
and
of
For
is because )le about t
greater
c o pr;, n e
also
produce
more
27
regulatory
requirements.
Large
firms
tend
to have
more
shareholders thirdparty
a A large sh information
areholder providers
base
such
provides
brokerage
incentive
services,
financial
newsletters
etc.
, to produce
information about
firm.
I standardize
market
Standardization capitalization.
value takes
to
into
market
obtain
account
Standardization
may
value
of equity
firm
general
by
the
size .
market
important
if
information collection depends
not
on the magnitude
of market
capitalization
total total
universe
market
but
on the ranking
of firms.
value
changes
Relative
over
of the firms
rankings
time.
Since
will
relative
change
our sample
to the as the covers
a 1 ong
period,2
any
secular
change
in total
market
value
be important.
group
It turns
rankings
out that
greatly.
standardizatior
Out of 332 firms
i does 32 fir
not alter ms change
groups
when
ranked
by
market
value
of equity
alone.
The firm's
adjusted
firm
size (SHR)
is calculated
multiplied
as the total
by the price
days
before
the first
dividend
is declared
This
figure
is then
divided
by
the total
dollar
value
of all nonADR
securities
1,000,000.
(TOTMV)
and multiplied
by
a scaling
factor
Studies
which
have
used
firm
size
have
covered
shorter period over which minimal.
changes
in total market value may
for
total
the firm's
the
adjusted
the
be
the
may
number
of shares outstanding
two
of
be
(Pt2) ï¿½
28
SHR
Adjusted
Firm Size
X Pt2
1,000,000
TOTMV
Price
data
the CRSP for NYSE
and number
tapes.
and AMEX
Prior
of shares to 1973,
listed
outstanding
total
securities.
market
From
1973,
are obtained
dollar
from
value
is
NASDAQtraded
stocks
are included.
The number
of
news
items
appearing
in the Wall
Street
Journal
Index
measures
the
occurrence
of events
that
have
informative
value.
It is also
a measure
of financial
media
coverage.
A firm
with
a high
number
of
news
items
is assumed
to have sample, Journal
higher
information
the number
Index
of
during
news
the
availability.
items appearing
year
prior to the
For each
firm
in the Wall
declaration
in the Street
o f the
first
dividend is collected.
The firms
adjusted
divided
firm
into
in my size
five
sample
are ranked
and the number
groups.
Note
that
in ascending
of
news
the groups
items
based
order
by
and then
on news
items
number
example,
do not have
of
news
firms
characteristics
equal
items
numbers
of firms.
is discrete
in the third
group
of the information
This
causing
all have
is because
clustering.
news
availability
items. groups
the
For The are
shown
in Tables
2 and 3. There
appears
to be sufficient
variation
size
with
groups,
thE
respec =. mean
to firm
market
value
size.
For the adjusted
of equity
firm
is approximately
the preceding
I
two times
greater
than
group .
29
TABLE
CHARACTERISTICS
OF GROUPS
BASED
ON ADJUSTED
FIRM
SIZE
a SIZE
two
b NEWS
d
is the markE lays before is the numb(
value
of equity
the initial er of Wall S
dividend
street
(in millions
of dollars)
is announced.
Journal
Index news
items
in the year dividend.
prior
to the announcement
of the initial
Table
size yield
items
findings
and
4 shows
news
identical reported
the distribution
items.
While
groupings,
about
of Thompson,
them. Olsen
of firms
the information
large
firms
This
tend
grouped proxies
to have
is consistent
and Dietrich
(1987)
Wall Street Journal Index
by firm do not
more with
news
the
who examine news items.
GroupiN Var Mean SD Min Med Max
SIZEs 6.563 2.396 2.860 6.399 15.587
1 66
NEWSb 6.833 2.271 2 6 16
SIZE 13.144 3.558 6.274 13 .052 24.820
2 67 ... ...
NEWS 7.667 2.420 4 7 18
SIZE 23.303 6.088 10.621 22.727 38.900
3 67
NEWS 9.552 4.328 5 9 33
SIZE 50.667 17.750 26.837 45.794 127.630
4 66
NEWS 10.797 5.262 5 9 28
SIZE 308.83 584.906 50.474 128.265 3631.92
5 66
NEWS 13.078 7.895 5 10 50
the characteristics of
30
TABLE 3
CHARACTERISTICS
OF GROUPS
BASED
ON NUMBER
OF NEWS
ITEMS
a SIZE
is the market
two days
NEWS
before
is the numb
value
the init er of Wal
of equity (in ial dividend
ii
Street
millions
is
Journal
of dollars)
announced.
Index
news
items
in the year
dividend.
prior
to the announcement
of the initial
DISTRIBUTION
OF FIRMS
TABLE IN GROUPS
BASED
ON ADJUSTED
FIRM
SIZE
AND NUMBER
OF NEWS
ITEMS
ADJUSTED MARKET VALUE
NEWS
ITEMS
3
 ii
JI A A I A dl II ii v Y I Y II
17
28
10
Total
66
2 11 27 13 13 3 67 3 9 13 7 26 12 67 4 8 11 11 17 19 66
13
14
31
66
Total 48 92 46 79 67 332
GroupI N _Var Mean SD Min Med Max
SIZEs 23.727 26.846 2.860 14.018 138.52
1 48
_LINEWSb 4.826 0.5698 2 5 5
SIZE 43.899 91.479 2.880 13.312 512.290
2 92
92___ NEWS 6.424 0.497 6 6 7
SIZE 25.242 22.562 4.020 15.970 82.357
3 46
NEWS 8 0 8 8 8
SIZE 46.317 47.877 3.294 27.547 297.024
4 79
__ __NEWS 9.397 0.952 9 10 12
SIZE 238.728 586.243 10.621 55.544 3631.920
5 67
LNEWS 17.657 6.383 13 16 50
31
To test
for differential
information
contents
across
five
groups
the data
sample
is characterized
as a oneway
layout
design
experiment.
The data
consists
of N
nk (for
k=l, . . 15) information
observations availability
with
group.
observations
Note
that
in the kth
the number
of
observations
in each
group
need
not be equal
to each
other.
This
oneway
layout
is illustrated
in Figure
3.1.
D(j,k)I's
the matched observations.
differences
from
group
k with
individual
Information
Availability
Groups
1 2 3 4 5
D(1, 1)
D(2, 1)
D(1,2) D(2,2)
D(l,3) D(2,3)
D(1,4) D(2,4)
(1,5) (2,5)
D(n4,4)
D(n2, 2)
D(nl, 1)
D(n5 ,4)
D(n3,3)
Figure Sample Data
1
Design
the
are
= Ek
The basic
model
D(j ,k)
= M + Tk + ejk,
j=i, . . * ink
=1, ... ,5
where
is the unknown
overall
mean,
Tk
is the unknown
information
effect
in
group
k. I assume
the
errors
e are
mutually
independent
and
have
the
same
continuous
distribution.
The hypothesis
to be tested
: T
 7 3
 74
: 71 7_ 2>r3 >T4
= T5
versus
T5
where
at least
one of the inequalities
is strict.
The firms
lower
information
availability
groups
have
larger
information
effects.
I use the JonckheereTerpstra
statistic
(see
Hollander
Wolfe,
1973)
to test
the hypothesis.
It is based
on the number
group
observations which
are less than
each
of the
group
observations,
where
> j
For 5
groups,
there
are 10 of
these
MannWhitney
counts Uii.
ni nj s=l t=l
i) D(t,j) ]
32
is
is
H0
Ha
the
in
of
and
T
33
where
1 if a
MannWhitney
counts.
0 otherwise.
The stronger
Let JT
are the
be the
group
sum of the
effects,
larger
is JT
The large
sample
approximation
J is
nk 2]/4
JT I [N2k1
k=l
[N2(2N+3)
i
k=l
nk2 (2nk+3) ]
It has
an asymptotic
N(0, 1)
distribution.
In addition,
use the KruskalWallis
(see
Hollander
Wolfe,
1973)
statistic
to detect
i f there
any
differences
r across
groups.
The alternative
hypothesis
is that
the r's
are not all equal.
observations. in ascending
groups
It is based
All matched difference:
order.
and summed.
These
Let rk
ranks
be the
on the ranks s in the sample
are then
divided
of sample are ranked
into
sum of the ranks
their
of the
observations
in
group
k. The KruskalWallis
statistic
H is
given
5
N121
N (N+I)k=l
 3 (N+l)
large
samples,
has
an
asymptotic
chisquared
distribution
with
k1 degrees
of freedom.
The above
tests are equivalent
to parametric
analysis
variance
F tests.
However,
when the
distributions
of matched
differences a inappropriate.
re not normal,
Further, the
the
use of F statistics
JonckheereTerpstra
is
the
1 72
and
in
by
For
of
test is a
0 (a ,b) =
34
comprehensive
problems
test
associated
which with
does
multiple
not suffer
from
comparisons
the usual
such
as the
Bonferroni
or Tukey
methods.
CHAPTER
TESTS
ON QUARTERLY
EARNINGS
ANNOUNCEMENTS
The Overall
Sample
The overall
sample consists
of 325 firms.
Each
firm
must
have at
least
3 usable quarterly
earnings
(joint
announcements
are omitted)
For each
firm,
or postdividend
initiation quarterly
periods. earnings
announcements in each
period are averaged.
Summary statistics
for the crosssectional
distribution
of average
information
content measures for
pre
and postdividend
initiation periods
are shown
in Table
5. The information
content measures
based
(MAJ),
on raw returns
the ratio
nonannouncement
(RAW)
market
adjusted
excess
of earnings announcement returns
returns
variance
(BVUE)
returns
variance
and the ratio
to of
earnings
and dividend
announcement
returns
variance
to non
announcement
dividend relevant
period
initiation
and
the
returns period,
ratio
variance
only
is
(BVUD) ï¿½
earnings
with
In the
announcements
respect
to
pre
are
earnings
announcement returns variance
(BVU).
The information
content
measures
initiation
RAW and MAJ have period. The pred
larger
ividend
initiation
values
of the minimum,
the first
quartile,
median,
third
quartile
and maximum
always
exceed
those
from
the post
35
in both pre
the information content of
are
values
in the predividend
36
TABLE
DISTRIBUTION EARNINGS ANN
OF THE AVERAGE
OUNCEMENTS
BEFOR
INFORMATION E AND AFTER
CONTENT OF QUARTERLY DIVIDEND INITIATION.
Sample Size = 325 firms
PreDividend Period PostDividendPeriod
q q i i
RAW
IMAJ
BVU
RAW
MAJ
BVUE
BVUD
Mean 5.79 5.67 2.007 4.52 4.39 1.908 1.862 SD 2.13 2.15 1.139 1.62 1.62 1.178 1.444 Minimum 1.14 1.52 0.459 0.02 0.85 0.230 0.528 Qja 4.29 4.13 1.309 3.42 3.34 1.173 1.205 Median 5.61 5.43 1.684 4.36 4.15 1.609 1.573 Q3a 6.98 6.89 2.263 5.35 5.25 2.367 2.036 Maximum 13.31 12.82 7.663 11.65 11.43 9.589 15.081
a Ql=first
quartile;
Q3=third
quartile.
dividend quarterly revaluatio
initiation earnings
n is 5.79%.
period.
For
announcements
predividend
the
For postdividend
mean
initiation
initiation
raw
price
quarterly
earnings
dividend
announcements,
initiation
mean
the mean
market
is 4.52%.
adjusted
Similarly,
price
the
revaluation
pre
is
5.67%
with
the postdividend
initiation mean being
4.39%.
These
figures
For raw returns,
are similar
Venkatesh
to those
obtains
reported
by
a predividend
Venkatesh.1 initiation
1
study
75 NYS
The differences are as follows. '
3E
or AMEX
firms
between Jenkatesh
which
pa
my study and the Venkatesh examines a smaller sample of
id quarterly dividends. His
estimation period precedes the quarterly earnings announcement date. He uses a lagged ScholesWilliams technique to obtain his market model regressions.
37
mean
of 5.84%.
In the postdividend
initiation
period,
reports
following
a mean
of 4.91%
(preceding)
(4.01%)
dividends
for earnings announcements.
announcements
Results
for
market
adjusted
excess
returns
are similarly
comparable.
The standardized variance
information content measures
show
a lesser for BVUE
dividend
initiation
decrease.
and BVUD
The postdividend
are 1. 908
initiation mean value
period,
the
average
and 1. 862
of 2. 007. dividend
initiation
compared
mean
with
values
the
pre
For the postdividend
announcement
variance
is less
than
the average
earnings
announcement
variance.
Thus
the combined announcement
average
(BVUD)
variance
is less
than
of
earnings
for earnings
and dividend announcements
alone
(BVUE) ï¿½
Whether
these
differences
are statistically
significant
examined
by comparing matched
differences.
The distribution
for matched statistics
statistics greater in
differences
and
the
to test whether
is tabulated
nonparametric
the
the predividend
average
initiation
in Table Wilcoxon
information period. TI
6. I use t signedrank
content
is
ie evidence
for RAW and MAJ is unambiguous.
Both
t and Wilcoxon
signed
rank
tests
support
the hypothesis
that
the
average
post
dividend
initiation
quarterly
earnings
announcement
is less
informative.
About
70% of the sample
experiences
a decrease
earnings
price
informativeness.
The
average
reduction
is about
and dividend
he
is
in
1. 2%. This
earnings
information
suggests
that
38
might
be substitutable.
My sample
yields
results
entirely
consistent
with
those
reported
by
Venkatesh.
The results
from
the standardized
variance
measures
information
content
measures
are in less
accordance
with
raw price decrease
measures. following
Just
over
dividend
half
the sample
initiation.
experience
DBVUE
has
insignificant
decline.
Thus,
there
appears to be
a reduction
in the informativeness
of earnings announcements
relative
average
informativeness
in nonannouncement
days.
More
importantly,
in
however,
a direction opposite
DBVUD
has
to that
a significant hypothesized.
change
but it is
The variance
of
excess
returns
for public
announcements
(both
earnings
dividends
announcements)
relative
to the nonannouncement
periods
is decreased
once
dividends
have
been
initiated.
Contradictory
to the information
transmission
hypothesis,
proportion
privately
of information
acquired
decreases
publicly
after
released
dividend
relative
to that
initiation.
I also
test
the normality
of the distribution
of the
matched differences
the value
probability distribution significance
of information
of the ShapiroWilk
of
rejecting
is normal.
level
the
content measures.
D statistic
null
The null
for all distributions.
The lower
the greater
hypothesis
is rejected
that
the
the
at the 5%
The nonnormality of
the distributions
of matched
differences
justifies
the
use of
nonparametric test statistics.
of
the
the
an
to
and
the
39
CROSSSECTIONAL
TABLE
DISTRIBUTION
OF MATCHED
DIFFERENCES
Sample Size = 325 firms
i'
DRAWa
DMAJa
DBVUEa
DBVUDa
Mean 1.27 1.29 0.099 0.145 Std. Dev. 2.53 2.51 1.546 1.729 Minimum 6.99 6.51 7.098 13.632 Q1b 0.15 0.36 0.785 0.503 Median 1.37 1.23 0.048 0.114 Q3b 2.76 2.95 0.873 0.907 Maximum 9.17 9.80 6.237 6.518 Dc 0.9654 0.9695 0.9364 0.8367 Prob < D 0.0368 0.0421 0.0000 0.0000 No. Positive 241 226 170 177 % Positive 77.2 69.5 52.3 54.5 Tde 9.05 9.24 1.15 1.52 Prob > T 0.0001 0.0001 0.1250 0.0654 Wdf 8.42 8.46 1.17 2.82 Prob > W 0.0001 0.0001 0.1210 0.0024
a Difference
content of for the ra
of average
pre
quarterly earnings,
w price
information
and postdividend
for eaci content
firm.
information For example,
measure,
DRAWj
 R j, PRE
 RAWj, POST
b Ql=first quartile; Q c ShapiroWilk test of d Test of H0: DRAW=0
H0:
DMAJ=0
Ho: DBVUE= Ho: DBVUD=
=0
=0
e tstatistic. f Wilcoxon signedrank
3=third H0: Disi versus
versu Versus I versus
quartile. tribution
is normal.
Ha: DRAW> 0,
Ha:
DMAJ>I0,
Ha: DBVUE>O, H a: DBVUD>0.
statistic.
I
40
The Sample
Grouped
by Information
Availability
In this
section,
I analyze
the matched
differences
across
information transmission
availability hypothesis,
groups.
one
Under
would
the
expect
information
to
see
the
informativeness
of earnings announcements to decrease
as one
goes
from
low information
availability
to high
information
availability
firms.
Table
7 shows
the distribution
of DRAW
grouped
by adjusted
firm
size.
For all
groups,
the differences
are significantly positive.
The Wilcoxon signedrank tests and
ttests
are significant at
KruskalWallis
H statistic
levels well below
the 5% level.
indicates that the group
means
DRAW
do differ
across
groups.
The JonckheereTerpstra
statistic smaller
indicates as market
that value
the
group
increases.
means This
are decreasingly
relationship
is
significant
at
the
5%
level.
Postdividend
initiation
quarterly
earnings
average
price revaluations
are less
than
the predividend group means shows
initiation
that
period.
the decrease
An examination
tends
to be less
of the
for high
market
value
firms.
Low market
value
firms
experience
a drop
of 1.5%
in the
announcements.
decrease. monotonic. decreases
average
High
However,
Firms
than
market
this
revaluation
value
in the lowest
those
occurring
at earnings
firms experience only
relationship
market value
in the next
is
not
group
have
a 0.6%
strictly
smaller
quintile.
Table
8 shows
the distribution
o f DMAJ
grouped
by adjusted
the same general results with
The for
in
firm size. We obtain
DMAJ. The
41
group group group
means decrease
1 is less
than
from that
means is less strong
Tables
9 and 10 show the
group
of
group
than
2 to
group
5. The mean
2. The difference
for
between
for DRAW.
distributions
for the standardized
variance
significant
measures.
change
In general,
in
means
there after
does
not
dividend
appear
to be
initiation.
any
P
values
for the ttests
and Wilcoxon
signed
rank
tests
generally
insignificant.
The only
exception
to this
is
group
2 for DBVUD
which
shows
a decrease.
The KruskalWallis
statistics
fail
to reject the
null
hypothesis
of equal
group
means.
The JonckheereTerpstra statistics do
evidence that
changes in relative informativen
not provide any ess increases as
one goes
from
smaller
to larger
firms.
Tables
11
differences
to
14
for the
show
groups
the
based
distributions
on number
of
of Wall
matched Street
Journal
Index
news
items.
In general,
the results
obtained
from
using
this
information
availability
proxy
are similar
those
obtained
from
adjusted
firm
size.
Within
groups,
the t
and W statistics
show
that
the
mean
matched
differences
DRAW
and DMAJ
are all greater
than
zero
at conventional
significance levels.
The KruskalWallis
statistic
shows
that
the
group
statistic
means
shows
are not all equal.
that
The JonckheereTerpstra
groups with more news items
have
smaller
mean decreases.
are
to
for
42
DISTRIBUTION
OF DRAW
TABLE 7
FOR PORTFOLIOS
BASED
ON FIRM
SIZE
FIRM SIZE PORTFOLIOS (1=SMALL, 5=LARGE)
1 2 3 4 5 N 65 64 67 66 63 Minimum 6.9927 4.7255 5.6859 5.3322 3.6112 Qla 0.0597 0.1516 0.1686 0.2734 0.8354 Median 1.7415 1.7134 1.7880 1.0386 0.4658 Q3a 3.2472 3.5687 2.8520 2.2331 1.7973 Maximum 9.1723 8.4802 7.7009 6.5617 6.2081 Mean 1.4319 1.6038 1.5852 1.0865 0.6213 SD 3.1280 2.5617 2.3125 2.3807 2.0822 T 3.69 5.01 5.61 3.71 2.37 Pr > ITI 0.0005 0.0001 0.0001 0.0001 0.0210 W 3.76 4.29 4.87 3.39 2.18 Pr > IWl 0.0001 0.0001 0.0001 0.0006 0.0292 Hb 9.6564* jc 2.6670*
a Ql=first
b Test
not
rej e(
c Test
one
quartile;
Q3=
of H0: 'r =T2=...=7" the same. H is th ct the null at the of H0: '1== 5
Sl
trict
inequality.
third
5 versu
quartile.
Ha:
le WallisKru 5% level if
versus
at least
one mean
skal statistic. H<9.49.
Ha: 71>_T2>. ï¿½.>5
J is
the
with
(r) is Cannot
at least
JonckheereTerpstra
statistic.
J<1.1645.
* Significant
Cannot
rej ect
the null
at the 5% level
if
at the 5% level
using
a onetailed
test .
DISTRIBUTION
OF DMAJ
43
TABLE 8
FOR PORTFOLIOS
BASED
ON FIRM
SIZE
FIRM SIZE PORTFOLIOS (1=SMALL,_5=LARGE)
____1 2[13 2 4 15 N 65 64 67 66 63 Minimum 6.1745 6.5091 3.5399 4.8611 2.8960 Qva 0.1286 0.5404 0.1559 0.5313 0.8236 Median 1.3908 1.8688 1.4322 0.7621 0.5403 Q3b 2.5956 3.8226 3.0869 2.9983 1.9154 Maximum 9.8017 8.5145 7.4536 6.4123 5.4624 Mean 1.2713 1.6674 1.6423 1.1678 0.6591 SD 2.9604 2.7884 2.2282 2.4313 1.9437 T 3.46 4.78 6.03 3.90 2.69 Pr > ITI 0.0010 0.0001 0.0001 0.0002 0.0091 W 3.52 4.19 5.08 3.35 2.36 Pr > IWI 0.0004 0.0001 0.0001 0.0008 0.0182 Hc 7.2795 1jid L_2.0676*
a Ql=first
quartile;
Q3=third
quartile.
Test
of H0:
not the
sam
r 1=T 2e. H
is
versus
=T5
the
Ha:
at least
WallisKruskal
one mean
statistic.
(T) is Cannot
reject null at
C Test
one
the 5% level if H<9.49.
of H0: T ='T 2= ...=T5
strict
versus
inequality.
Ha: '1T2>'" 0> T 5
J is
the
with
at least
JonckheereTerpstra
statistic.
J<1. 645.
* Significant
Cannot
rej ect
the null
at the 5% level
if
at the 5% level
using
a onetailed
test .
44
DISTRIBUTION
OF DBVUE
TABLE 9
FOR PORTFOLIOS
BASED
ON FIRM
SIZE
FIRM SIZE PORTFOLIOS (1=SMALL,_5=LARGE)
N 65 64 67 66 63 Minimum 7.0975 2.8549 3.3461 2.3720 2.8119 Qla 1.0527 0.7551 0.9388 0.6734 0.5887 Median 0.1229 0.2910 0.0761 0.1267 0.0641 Q3a 0.9738 1.0409 0.9714 0.6136 0.8085 Maximum 5.4849 4.8623 6.2366 3.5934 2.4277 Mean 0.2467 0.3007 0.2396 0.1151 0.0835 SD 2.2479 1.4633 1.6395 1.0479 0.9478 T 0.88 1.64 1.20 0.89 0.70 Pr > ITI 0.3797 0.1051 0.2359 0.3756 0.4868 W 0.49 1.26 0.67 0.61 0.55 Pr > IWI 0.6242 0.2076 0.5028 0.5418 0.5824 Hb 1.6994 jc 0.3855
Ql=first
quartile;
Q3=third
quartile.
Test not t
of Ho: 1r=IT2=. ..=I5 :he same. H is th
versus Ha:
at least
e WallisKruskal
one mean
statistic.
(T) is Cannot
reject the
C Test
one
of H0: strict
statistic. J<1.645.
null
at the 5% level if H<9.49.
TIT2...=5 versus
inequality.
Cannot
rej ect
J the
Ha: r>r2>.>. .>r5 with at least is the JonckheereTerpstra n, null at the 5% level if
45
DISTRIBUTION
OF DBVUD
TABLE FOR POR
10
,TFOLIOS
BASED
ON FIRM
SIZE
FIRM SIZE PORTFOLIOS (1=SMALL,_5=LARGE)
, i iU
SI i
N 65 64 67 66 63 Minimum 7.2114 2.7647 13.6318 10.3739 1.0485 Q1a 0.7629 0.3808 0.5740 0.5135 0.3751 Median 0.0467 0.2756 0.0064 0.2192 0.0104 Q3a 0.8941 1.1312 0.9512 0.8536 0.7990 Maximum 5.3639 4.8751 6.5185 3.4881 3.0709 Mean 0.0263 0.4357 0.0094 0.1248 0.1936 SD 1.9576 1.3613 2.3879 1.6614 0.8251 T 0.11 2.56 0.03 0.61 1.86 Pr > ITI 0.9140 0.0129 0.9745 0.5438 0.0673 W 0.09 2.27 0.92 1.74 1.14 Pr > IWI 0.9282 0.0232 0.3576 0.0818 0.2542 Hb 2.5566 jc 0. 1831
Ql=first
Test
not t
reject Test o one s
quartile;
of H0 .: 1='r2=. .. the same. H is
the null
Q3=third =7,,, versu
quartile.
, Ha:
at least
the WallisKruskal
one mean
statistic.
(T) is Cannot
at the 5% level if H<9.49.
f H0: 71= "2= .ï¿½. . =75 versus
trict
inequality.
Ha:
is
r >r2>. ï¿½ï¿½1>r5 with at least the JonckheereTerpstra
statistic.
J<1.645.
Cannot
rej ect
the null
at the 5% level
if
46
The results
Tables
13
and
for the distributions
14,
respectively,
of DBVUE
provide
and DBVUD
in
no discernable
pattern.
With
the exception
of
group
3,
none
of the
group
means Terpst
decrease
are significantly ra J statistic i
in the relative
different ndicates t
from
that
information
i zero. there
The Jonckheere
is
content
no monotonic of earnings
announcements
as one goes
from
low to
high
news
item.
However,
the WallisKruskal
H statistic
indicates
that
the relative
information
content
between
groups
are not all equal.
In general,
supportive and DRAW,
the results
of each
other.
from Based
these
stratified
on the returns
the decrease in group means
tests are not
measures,
DMAJ
increases as one goes
from
high
market value
(news
item)
firms to low market
value
(news
items)
firms.
This
is consistent with
the information
transmission
informativeness
hypothesis.
of
Based on returns measures alone,
quarterly
earnings
reports
information
decrease
availability
after
dividend
firms
initiation.
experience
However,
the
once
greatest we adjust
for the contemporaneous
level
of returns
volatility,
this
information
effect
no longer
persists.
The
mean
standardized
variance
measures,
DBVUE
and DBVUD,
are not statistically
different exhibit an informatior
following
dividend
y significant availability
change
initiation.
in
groups.
going
There
Neither
from 1 appears
do they
Low to high
to be
no
monotonic
difference
in the
mean
change
in total
information
by information availability.
the low
for
for firms grouped
content
( DBVUD )
47
TABLE
DISTRIBUTION
OF DRAW
11
FOR PORTFOLIOS
ITEMS
BASED
ON NO. OF NEWS
FIRM SIZE PORTFOLIOS (1=LEAST, 5=MOST)
1 2 3 4 5 N 45 92 40 80 68 Minimum 5.6416 5.0269 6.9927 5.6859 4.2648 Q1a 0.7078 0.1753 0.4185 0.0847 0.9469 Median 1.7415 1.5886 1.7555 1.3645 0.5221 Q3a 3.3723 3.2141 2.7990 2.5838 1.6834 Maximum 9.1723 7.7009 8.9110 5.9192 6.5617 Mean 1.5995 1.5428 1.4653 1.2798 0.5568 SD 3.0830 2.4669 2.8481 2.1635 2.3436 T 3.48 6.00 3.25 5.29 1.96 Pr > ITI 0.0011 0.0001 0.0024 0.0001 0.0543 W 3.30 5.32 3.20 4.72 1.80 Pr > wI 0.0010 0.0001 0.0014 0.0001 0.0588 Hb 10.2878* jc 2.6193*
Ql=first
b Test
quartile;
of H0: T1=72=.
not the
rej ect C Test o:
one s
same.
Q3=third
quartile.
versus Ha:
* =T5
at least
H is the WallisKruskal
the null at
f HO T 1:72
trict
statistic.
J<1. 645.
* Significant
inequa Cannot
the 5% level =r5 versus Ha lity. J is
rej ect
one mean
statistic.
if H<9.49.
: 1>'T2>... >_r5
the
the null
with
(r) is Cannot
at least
JonckheereTerpstra at the 5% level if
at the 5% level
using
a onetailed
t~est.
48
TABLE
DISTRIBUTION
OF DMAJ
FOR PORTFOLIOS
BASED
ON NO. OF NEWS
ITEMS
FIRM SIZE PORTFOLIOS (1=LEAST,_5=MOST)
1 2 3 4 5
___ Ii ~ 23 1 4 1 5
N 45 92 40 80 68 Minimum 6.5091 5.2964 6.1745 3.5399 4.0065 Q1a 0.3792 0.0396 0.0547 0.1967 0.8519 Median 1.8883 1.3843 2.0159 1.2342 0.3210 Q3a 3.4979 3.1635 3.3842 2.4626 1.7938 Maximum 9.8017 7.4536 7.1067 6.1621 5.9768 Mean 1.8230 1.4624 1.6080 1.1782 0.6300 SD 3.3569 2.4725 2.7286 2.0246 2.1987 T 3.64 5.67 3.73 5.20 2.36 Pr > ITI 0.0007 0.0001 0.0006 0.0001 0.0210 W 3.41 5.67 3.73 5.20 2.36 Pr > IWI 0.0006 0.0001 0.0003 0.0001 0.0182 Hb 9.4949* jc 2.7262*
Ql=first
quarti e;
Q3=third
quartile.
of H0:
not the
rej ect c Test o:
one s
sam
T 1=T2=
ie. H
the null
f H0: trict
statistic.
* Significant
is
versus
the
Ha:
at least
WallisKruskal
at the 5% level
1=T 2.. .=T5 versus inequality. J
Cannot
rej ect
one mean
statistic.
(r) is Cannot
if H<9.49.
Ha: Ti>T2>...>T5
is
the null
with
at least
the JonckheereTerpstra if J<1.645.
at 5% level using
12
Test
a onetailed
test .
49
TABLE
DISTRIBUTION
OF DBVUE
FOR PORTFOLIOS
BASED
ON NO. OF NEWS
ITEMS
FIRM SIZE PORTFOLIOS (1=LEAST,_5=MOST)
1[ 2 3 4 5 N 45 92 40 80 68 Minimum 7.0975 3.9293 1.5222 3.3461 7.0369 Q1a 0.0897 1.0469 0.2689 0.4310 0.8351 Median 0.0336 0.0293 0.7221 0.0897 0.1813 Q3a 1.0041 1.0610 1.3258 0.6437 0.4573 Maximum 4.4923 6.2366 4.8623 5.1343 2.8748 Mean 0.0024 0.1548 0.6755 0.0632 0.2103 SD 2.1131 1.7655 1.1946 1.1790 1.2715 T 0.01 0.84 3.58 0.48 1.36 Pr > ITi 0.9940 0.4024 0.0009 0.6328 0.1771 W 0.45 0.38 3.28 0.70 1.46 Pr > awl 0.6528 0.7040 0.0010 0.4840 0.1442 Hb 11.8148* jc 1.4030
Ql=first
Test
of H0
not the
rej ect Test o one s
quartile; " 1= T2="
same.
Q3=third =T5 versus
quartile.
Ha:
at least
H is the WallisKruskal
one mean
statistic.
(T) is Cannot
the null if H<9.49.
ï¿½H: T1T2=. 7
trict
statistic.
J>l. 645.
* Significant
versus
inequality.
Cannot
rej ect
J
Ha: T1 >T2>.ï¿½.>T5
is the
the null
with
at least
JonckheereTerpstra at the 5% level if
at the 5% level
13
using
a onetailed
test .
50
TABLE
DISTRIBUTION
OF DBVUD
FOR PORTFOLIOS
BASED
ON NO.
OF NEWS
ITEMS
FIRM SIZE PORTFOLIOS (1=LEAST, 5=MOST)
1 2 3 4 5 N 45 92 40 80 68 Minimum 7.2114 13.6318 1.3339 5.7861 3.1993 Q1a 0.7629 0.5921 0.1358 0.3647 0.5277 Median 0.1868 0.0714 0.7632 0.1048 0.0717 Q3a 1.0446 1.1932 1.2732 0.7652 0.4795 Maximum 4.0260 6.5185 4.8751 3.4881 3.1993 Mean 0.0482 0.1039 0.7711 0.0608 0.0028 SD 1.9836 2.4677 1.1524 1.0976 1.0661 T 0.16 0.40 3.58 0.48 1.36 Pr > ITI 0.8713 0.6873 0.0001 0.6223 0.9828 W 0.80 1.04 3.78 1.39 0.48 Pr > IWI 0.4238 0.2984 0.0002 0.1646 0.6312 Hb 11.8098* jc 111.2855
Ql=first
Test
S
quartile; : T 2"** ame. H is
the null at
H0ï¿½
rict
statistic.
J<1. 645.
* Significant
Cannot
Q3=third
versus
the
rej ect
quartile.
Ha:
at least
WallisKruskal
one mean
(7'r)
is
Cannot
the 5% level if H<9.49.
sus Ha: J is the nL
at least
the ill
14
of H0
not the
reject I o Test of
one st]
T1=T2=. . . =T5 VE inequality.
statistic.
r1>r2>... >T5 with
JonckheereTerpstra at the 5% level if
at the 5% level
e r
using
aonetailed
test .
51
This
contradicts
one would released
expect
the information the increase in
transmission
the proportion
for low
hypothesis
of publicly information
availability
firms.
This
returns
inconsistency information
investigation
of
between content
returns
the standardized
measures
volatility
calls
variance
for
and
further dividend
initiation.
returns
It is possible
volatility
that a contemporaneous
the results
decrease
obtained.
This
investigated
in the next
chapter.
The Market
Reaction
to Earnins
Chanaes
going
contemporaneous
on
decline
to
examine
in returns
the
variance,
effect
of
I analyze
price
reaction
to reported
quarterly
earnings
change.
This
section
provides
further
proof
that
what
has been
previously
interpreted
as a decrease
in earnings
informativeness
caused
dividend
initiation
may
be due to
a contemporaneous
decrease
in returns
volatility.
The relationship
between market price
reaction and
reported
earnings
is modeled as
CERjq
 Po
/31D
+ i3AEjq
+)34DAEjq
The 3day
cumulated
excess
returns
for the day prior
of and the day following
the earnings
announcement
date
for quarter
information
to be greatest
as
around
is producing
in is
Before
the the
by
day
(1)
to
the
and f irm
is denoted
by CERjq.*
The dummy
52
variable
D takes
on the value
1 if the quarterly
earnings
announcement
comes
from
the postdividend
initiation
period
and is
zero
otherwise.
The earnings
surprise
or standardized
earnings deflated earnings
is the quarterly
by
the firm's
share
price
in earnings
two days
per
before
announcement.
EPSjq
 EPSjq1
Pj ,t2
The change randomwalk
in quarterly expectations
earnings model. D
is appropriate
eflating
by
based
the share
on a
price
captures
formed
any expectations
through
about
the forthcoming
gathered
between
earnings earnings
announcements.
regression
approach
allows
us
to
incorporate
information
contained
in the earnings
report
(namely,
change
above
in earnings
analysis
per share)
based
on
which
groups.
we could
This
not do in the
for the
potential
bias
that
could
occur
if earnings
changes
smaller
similar
or less
to
variable
the approach
in the postdividend
used
by Healy
period.
and Palepu
This
(1988) .
The average
price
response
to earnings announcement
for the
sample
is
in the postdividend
go. Any change
initiation
in the period
average
price
is captured
response
by j1,
average
market
price
reaction
to
a given level
of earnings
is 92 It is expected
change
share
the
AEjq
information
The
the
controls
whole
are
is
the
and
indicating
to be positive
that
surprise
53
the market improvements
responds (declines)
positively
The
averag(
(negatively) e postdividen
to earnings Ld initiation
adjustment
to 82
If dividends
is 83 " enable
investors
to revise
their
expectations
of earnings,
will
their
be reduced.
forecast However,
errors
measures
in the postdividend period
of unexpected
earnings
which
ref lec
only incorporate t this additional
quarterly informatio
earnings changes n obtained from c
will
not
dividends.
Standardized estimates of
quarterly unexpected
earnings earnings.
changes Therefore,
will
the m
be noisier arket price
response reduced.
to
our measure
Coefficient
of earnings
92 is expected
surprise
should
to be negative.
I estimate
each
firm
in my
equation
sample,
(1) using 8 quarters
ordinary
least
preceding and
squares. following
initial
dividend
are collected.
A quarter
is excluded
if the
announcement announcements
is
within
joint,
2 days
i.e.
of each
earnings
other.
and
A quarter
dividend
is also
excluded
i f there
was no data
on net income,
share
price
shares
outstanding
to calculate
the standardized
earnings.
require
that
a firm
had 4 usable
pre
and postdividend
initiation
quarters.
This
requirement
excludes
58
firms
leaving
274 firms
for the regression analysis.
For comparison,
I also
ran
the regression
using
measure BVUE.
be
For the
or
the
adjusted
variance
54
BVUEjq
= P
+ 31 AEjq
+ 94DI AEjq I
This regression market response and the reported
initiation.
thus
analyzes
adjusted
quarterly
Because
the relationship
for the firms's
earnings
the earnings
: before surprise
between
returns
and after
BVUE
the
volatility
dividend
is based
on
squared absolute
excess value
returns
and is always
of the earnings
surprise
positive,
use the
as the independent
variable.
The interpretation
on P2
and 93
remains
the same.
The results
of the regressions
are shown
in Table
15. For
the 3day
cumulated excess returns
CER
, the sample
mean
5.4832
and
is
highly
dividendinitiating
revaluation
firms
at quarterly
significant. experienced
earnings
This
indicates
an average
during
5.5%
the 2 year
that price period
preceding
dividend
initiation.
The coefficient
fl
is 0.8237
and is highly
significant.
For the quarters
in the 2 year
period
reaction
following
is reduced
dividend
by
initiation
0.82%.
These
the
results
average are of
price course
restatements
of those
obtained
in Table
The coefficient
02
is 0. 1925
and significant,
indicating
that
the announcement
of
a 1%
increase
in standardized
earnings However,
on average
leads
to
the coefficient
For postdividend
a 0.2%
has
quarterly
increase
a positive
earnings
in stock
value
announcement
price.
of 0.1828. nts, a 1%
increase
in standardized
earnings
on average
gives
rise
in stock
(2)
g0
is
5.
to
+81GD
0.38%
increase
price.
This
result
directly
55
MARKET
PRICE
TABLE RESPONSE TO
15
STANDARDIZED EARNINGS
a Model estimated
+PD
+ 3A Eq
+ 04DAEq
Model estimated is
BVUEq
= PO
+PD + P3
+ D I AEq
CER is the cumulated 3 day excess returns; one day prior to, the day of and the day following the quarterly earnings announcement. D is a dummy variable which takes on the value 1 if the quarterly earnings occurs after dividend initiation and is zero otherwise. The standardized earnings is the change in earnings per share from the prior quarter divided by the stock price two trading days before the
earnings
announcement.
AE =
EPSq
 EPSq_1
Pt2
tstatistics
* Significant
** Significant
a
a
in parentheses. t the 1% level t the 5% level
using using
twotailed twotailed
test. test.
Dep. Var. CERa BVUEb 80 5.4832 (47.538**)c 1.9508 (24.385**) 11 0.8237 (5.025**) 0.0698 (0.613)
P2 0.1925 (2.174*) 0.0131 (0.214) P3 0.1828 (0.480) 0.1232 (0.465)
F 10.376** 0.224
Adj. R2 0.0072 0.0006
No. of Firms 274 No. of Obs. 3901
is
CERq
=O
56
contradicts
the notion that
dividends
convey
information
that
leads
investors to better
forecast
earnings.
The coefficient
93 is, however,
not statistically
significant.
My regression
results
yield
results
diametrically
opposite
to those
obtained
significantly
by Healy
negative
930.
and Palepu I suspect
(1988) .
that
They
this
obtain
difference
arises
out of
a difference
in regression
technique
and model
specification.
unrelated
Healy
regressions
and Palepu technique
apparently use for estimation.
a seemingly
This
their these
estimates
firm
of 0
specific
and /32
coefficients,
specific.
model
In obtaining specification
excludes
volatility negative.
the
Al1
is thus
coefficient.
forced
onto
Any
P3
decrease
which
in
biases
returns
it to be
variance results.
measure
for the regression of informativeness
The relative
informativeness
using
is consistent
of quarterly
with
prior
earnings
response
announcement
after
at quarterly
periods
dividend
earnings
information.
announcements
is not significantly
The
relative related
market
to
non
to the
earnings
reported,
nor does
this
responsiveness
change
after
dividend initiation.
to be firm
allows
their
The results
does
the standardized
not change
CHAPTER
RETURNS
VOLATILITY
AROUND
DIVIDEND
INITIATION
In the previous
chapter,
I examined
the information content
of quarterly initiation.
r earnings announcements before a The information content measures
Ind after
used
dividend
are directly
related
to the volatility
for the observed
decrease
of returns. in the pric
A possible e reactions
explanation associated
with
postdividend
initiation
earnings
announcements
is that
returns
volatility
has decreased.
Declines the process
in returns
by
which
volatility information
may
be caused
is impounded
by changes in in price. For
example,
dividend
initiation
may
induce
greater
consensus
among
traders.
This
decreases
noise
trading
and bidask
spreads
and consequently returns variance.
However,
volatility
also
may
decrease
due to causes that
are independent
information motivated experience increases proportionally smaller
Finally,
reason.
in
Dividend
stock
bidask
the observed volatility
price
spreads
decreases
initiating which lE
and reduced
in
Eac
firms I to
leverage.
my sample
could
be due to unrelated
market
wide
changes
in interest
rates.
this
chapter,
I perform
several
tests
to distinguish
causes of
the observed
volatility
decrease.
I conclude
that
57
of
any
In
the
58
information
is not the only
plausible
cause
for volatility
decline.
Variance
Changes
in the Overall
Sample
Table
16 shows
the distribution
of percentage
changes
variances
calculated
for different
sampling
periods
around
dividend and 100
initiation.
trading
The sampling
days
periods
immediately
cover
before
900,
500,
and after
declaration
of the
initial
dividend.
Fewer
firms
are available
for the longer
sample
periods
because
of the longer
sequence
of returns
required.
For the 100,
250,
500 and 900 day
sampling returns,
periods,
I require
respectively.
at least
Variances
90, 240,
are calculated
480 and 850 as the sum of
the squared approximately
The percentage
continuously
equals change
E[R2]
compounded for short
in variance
returns
since
measurement
Var[R]
intervals.
is
Percentage variance
change
= 100lx [i
VARPOST ] VARPRE
A variance
decrease
after
dividend
initiation
gives
rise
positive
percentage
change.
The
mean
100 day) measured
percentage
sampling returns vo
change
periods.
latility.
is positive
This
for all
indicates
The median
is
(except
a decline
a better
the in
statistic
since the mean percentage variance
to
in
250
the
to
is
of location
change
59
DISTRIBUTION OF
TABLE 16
PERCENTAGE VARIANCE CHANGES
a Percentage
variance
changes are
calculated as
Percentage variance change =
loo [1 VARPOST] 10 0 x [ 1 VARpsT
Variances are calculated using daily continuously compounded stock returns from sample periods of 900, 500, 250 and 100 trading days before and after the declaration
day of the initial dividend.
b For 900, 500, 250, and 100 day sampling periods, a firm is
included if it has 850, 480, 240, and 90 valid returns,
respectively.
C Number of firms in sample with positive percentage variance
changes, i.e. variance decreases after dividend initiation. d ShapiroWilk test of normality. e Test of H0: Percentage variance change = 0.
Perioda 900 500 250 100 Sampleb 225 304 327 331 Min 260.12 310.63 382.41 1196.15 Q1 13.05 12.14 20.03 25.86 Median 39.44 33.65 20.33 9.68 Q3 55.20 54.58 43.73 36.83 Max 80.59 86.77 80.31 81.33 Mean 26.94 13.50 3.52 10.06 SD 46.29 59.78 61.69 92.38 No. Pos.c 182 212 216 186 % Pos. 80.9 69.7 66.1 56.2 Dd 0.8884 0.8354 0.8152 0.6024 Pr>IDI 0.0 0.0 0.0 0.0 Te 8.47 3.94 1.03 1.98 Pr>ITI 0.0001 0.0001 0.3035 0.0484 We 8.59 5.79 4.09 1.14 Pr>I W1 0.0001 0.0001 0.0001 0.2559
A
60
downwardly
biased.1
The medians
are positive
for all sampling
periods sampling
and always
period,
the
greater
more
than
the means.
apparent is the
The longer
volatility
the
decline.
The median
percentage
change
is 39% for the 900 day sample
period,
this
declines
to 10% for the 100 day sample
period.
The proportion
of firms
experiencing
volatility
declines
also
increases
with
longer
sampling
periods:
56% for the 100 day
sample
period
increasing
to 81% for the 900 day sample
period.
The Wilcoxon percentage
(except
signedrank
variance
the 100 day)
tests
change sample
of the null
is
zero
hypothesis
are rej ected
periods.
When
examining
changes
in stock
return
variance,
it is
important
to consider
changes
in market
volatility
through
time.
This
is particularly
so for my
sample which
has dividend
initiation
dates
dispersed
over
17 years
and where
there
also
considerable
possible
that
any
clustering
observed
in the middle
change
in market
seventies.
variance
It is around
time
of
contemporaneous
divider changes
initiation
in market
is
To address
this
possibility,
I divide
firm
variances
estimate
of market
volatility.
This
estimate
is obtained
from
the contemporaneous
returns
on an equally
weighted
market
index
of all
firms
on the relevant
data
tape.
This
adjustment
is in the spirit
of
a heteroskedasticity
adjustment,
where
From Jensen's
that
the
for all
the
is
explained
volatility.
by
by
an
the
>E [VpRE]/E [VposT] ï¿½
inequality,
E [ VpRE/VposT ]
61
stock' s market. changes
return
Table where
variance is assumed
17 shows
the firm
proportional
the distribution
variances
have
to that
of the
of percentage variance
been
adjusted
contemporaneous estimate
of the market
variance.
It
appears
that
changes
in market
volatility
explain
part
of the variance
decline.
Median
standardization
percentage
but
are still
changes positive.
are
smaller
The Wilcoxon
after
signed
rank
sample
tests can
period.
only
This
rej ect
provides
equal some
variances
evidence
for the 900 day
that
clustering
the data
sample
may explain
the observed
variance
decreases.
Variance
Ratio
Tests
Consider
returns
calculated
over
intervals
of k days
where
If daily
returns
follow
a random
walk,
the variance
these
kday returns
should
be k times
the variance
of daily
returns.
French
and Roll
(1986)
model
a stock's
return
having
3 components:
a rational information
(intrinsic
value)
component, component.
a noise
If noise
or mispricing
component
and bid/ask error
and
components
a bid/ask
error
are temporary
components,
over
time,
they
will
be corrected
and induce
negative
autocorrelat ions.
This
negative
autocorrelation
causes
the kday
variance to be
less
than
k times
the daily
variance. variance
A variance
ratio
(VR)
is the ratio
of kday
returns
to daily returns variance
by
of
k>l.
of
as
and Roll
(1986)
(French
62
DISTRIBUTION
OF MARKET
TABLE ADJUSTED
17
PERCENTAGE VARIANCE
CHANGES
a Percentage
variance
changes are calculated as
Percentage variance change = 100 x [1
VARPOST]
VARpR
Variances are calculated using daily continuously compounded stock returns from sample periods of 900, 500, 250 and 100 trading days before and after the declaration day of the initial dividend. Firm variances are divided by the returns variance of a valueweighted market index. b For 900, 500, 250, and 100 day sampling periods, a firm is
included if it has 850, 480, 240, and 90 valid returns,
respectively.
C Number of firms in sample with positive percentage variance
changes, i.e. variance decreases after dividend initiation. d ShapiroWilk test of normality. e Test of H0: Percentage variance change = 0.
Perioda 900 500 250 100 Sampleb 225 304 327 331 Min 198.86 272.93 406.08 803.86 Q1 0.36 31.79 31.85 43.60 Median 28.73 13.29 6.28 5.22 Q3 48.10 45.73 39.83 30.41 Max 90.54 80.75 79.44 79.81 Mean 17.73 2.74 6.50 10.06 SD 46.33 66.64 66.25 92.15 No. Pos.c 169 181 184 176 % Pos. 75.1 59.5 56.3 53.2 Dd 0.8452 0.8544 0.8367 0.6862 Pr>IDI 0.0 0.0 0. 0.0 Te 5.74 0.72 1.77 3.85 Pr>ITI 0.0001 0.4736 0.0769 0.0001 We 7.21 1.84 0.75 1.70 Pr>IWI 0.0001 0.0666 0.4556 0.1359
63
call
these
actualtoimplied
variance
ratios) .
1 Var
k
Var (Rl)
where
Rk is the kday
return.
Under
the null
hypothesis
that
returns
follow
a random
walk,
the VR should
be equal
If there
are transitory
components,
the VR should
be smaller
than
one.
the transitory
used
French
in recent
(1988) ,
The VR therefore
bid/ask studies
Lo
allows
and noise
by
and
French
Mackinlay
us to gauge components.
and Roll
(1988)
the effect
VRs have
(1986) ,
and
Fama
Kaul
of
been
and and
Nimalendran
and Richardson
(1990) .
and S
Cochrane (1 mith (1991)
L988)
Lo and Mackinlay
analyze
the properties
(1989),
of the
VR test
statistic.
In general,
they
find
that
the VR test
statistic
is robust
and has higher
relative
power
compared
with
alternative
test statistics.
If dividend
initiation
reduces
the bid/ask
error
component
(either
due to improved
liquidity
or reduced
information
asymmetry)
or noise
trading
(better
informed
traders
with
greater consensus)
then postdividend
initiation VRs
should be
closer to
the summary
one than
predividend
statistics
for 10day
initiation
and 20day
VRs.
Table
return
18 shows
variances
It
of es basic
can be shown
;timated
that
VR(k)
autocorrelations
measurement
interval
can be written
of returns
(daily
in
as a function
measured
over
our case).
+ [ (kl) P
k
+ (k2) P2
0 a + Pki]
(Rk)
to
one.
VRk
the
VRk =
64
TABLE
PRE
DISTRIBUTION
AND POSTDIVIDEND
OF VARIANCE RATIOS
INITIATION
FOR
500 DAY SAMPLE
PERIODS.
Sample = 310
VRo 1IVR20
PRE
POST
DIFFa
PRE
POST
DIFFa
Minimum 0.294 0.326 0.867 0.215 0.250 1.093 Q1 0.775 0.794 0.180 0.743 0.709 0.178 Median 0.949 0.950 0.018 0.915 0.917 0.030 Q3 1.166 1.142 0.171 1.211 1.148 0.234 Maximum 1.942 1.921 1.114 2.532 2.078 1.504 Mean 0.993 0.983 0.010 0.985 0.960 0.025 Tb 0.61 (0.5406d) 1.18 (02392 Wc 0.51 (0.6077d) 1.31 (0.1921a)
a For each
dividend
firm,
the difference
initiation variance ratios
between
pr
is DIFF
ee and = PRE 
post
 POST.
t test
of H0:
DIFF
 0.
Wilcoxon signedrank
test
of H0:
DIFF
d Significance
level
of twotailed
test.
obtained
from
500 day sample
periods
before
and after
dividend
initiation. multiplying
The 10day
10 (20)
500 day sample
20day
returns.
and postdividend
period
(20day)
successive
there
The VRs are
initiation
returns
returns.
10day
daily
are fifty
all less than
periods.
Mean
are obtained
Therefore, an twenty
one for both
and median
by
in
five prepre
dividend
initiation significant
initiation
VRs but (Wilcoxon
VRs this
are
smaller
difference
signedrank test).
than
postdividend
is not statistically
Therefore,
I cannot
18
= 0.
65
conclude
that
variance decreases are due
to reductions
in bid
ask spreads
or noise
trading.
The Timing
of Variance
Changes
results
show
that
the
decrease
in
the
price
revaluations
at earning
announcements
in the postdividend
initiation
returns
information
period
may
variability.
effect
be related
This
associated
decline
with
to the general
may
be
dividend
decline
caused
by
initiation.
ascribe variance
causality
changes
to this should
information
occur
after
effect
the
requires
initiation
that
of
dividends.
systematically
On the other
occur
hand, after
if variance
dividend
changes
initiation,
do not
the
plausibility questionable.
of
an dividend
initiation
information
effect
In this
section
I I analyze
the timing
of variance
changes.
I first successil
examine percentage ve 250 day sample p
varian( periods
changes
beginning
calculated 750 trading
over days
prior
to dividend
initiation.
Due to the long
series
returns
required,
only
272 firms
are included
in the sample.
Table
changes. declines the year
19 shows
the distribution
The sample
in the
year
of these
of firms experience
prior
to (period
percentage
significant
[250,i])
variance variance
as well
as
the initiation
The
in an To
is
of
      I I  I
I
of dividends.
(period
[ 1,250] )
af ter
66
TABLE
SUMMARY (250
a Percentage
calculated
STATISTICS DAY SAMPLE I
variance
OF PERCENTAGE
PERIOD)
changes
VARIANCE
FOR SUCCESSIVE
for period
CHANGES ?ERIODS
(500,251)
as
Percentage
variance
change
= 00x [i
VAR(500,
251), ]
VAR(750,501)
The period
dividend ir 15. Day 0 i b Number of f
changes, i. C Test of H0:
(1,250) nitiation
is comparable percentage va
Ls the declaration date
irn
e.
to the
pre
Lriance changes of the initial
and postin Table dividend.
is in sample with positive percentage variance variance decreases after dividend initiation.
Percentage
variance
change
 0.
Therefore
the variance
decline
does
not
occur
strictly
after
dividend
initiation.
There
is significant
variance
decline
year
prior
to dividend
initiation.
To better
successive
examine
overlapping
when variance declines
periods
of 100 trading
occur,
days.
I examine For each
firm,
I collect
400 trading
days
of returns:
200 before
initiation of dividends
19
Sample 272 Perioda 500,251 250,1 1,250 251,500 Median 7.266 13.29 6.28 5.22 Mean 5.052 2.74 6.50 10.06 No. pos.b 129 164 161 136 % Pos. 47.4 60.2 59.1 50.0 Wc 0.98 4.40 2.75 1.20 Pr>IWI 0.3297 0.0001 0.0026 0.2324
are
the
in
the
.0
I divide
these
400 days
and 200 after.
67
into seven 100
day samples
with each
sample overlapping
prior from
from from
sample
day
day day
200
150
100
by
50 days.
to 101;
to 51;
Sample sample
sample
period period
period
one consists two consists
three
to 1; and so on. The final
consists
sample
of returns of returns of returns
period
seven
consists
of returns
from
day
101 to 200.
The sample
periods
are shown
in Figure
200
2.
100
I I H I
Sample
periods
Sample
Periods
Figure 2 for Detecting
Variance
Changes
Within
each
sample
period,
I test
whether
a significant
variance
decrease
has occurred.
For each
firm,
I count
number
of significant variance
decreases
occurring
in the
pre
dividend dividend
initiation initiation
sample sample
periods
periods
1, 2 and 3 and the post
5, 6,
and 7. Denote
these
counts
as COUNT1
and COUNT2,
respectively.
If
change is more likely
a variance initiation,
the
100
I
200
1
the
I I , i
to occur following
dividend
68
COUNT2
is going
to be greater
than
COUNT1.
The test
therefore
COUNT2COUNT1
I test
this
= 0
hypothesis
versus
using
H1:
the
COUNT2COUNTI
Wilcoxon
> 0
signedrank
statistic.
To detect
a variance
decrease
within
a sample
period,
I use
a test statistic (1977, 1979). LE daily returns R1
normal
variates.
T* which et M be a
,R2 f * ,RM
Further,
has been
extensively
consecutively
which
examined
observed
are assumed
let the variances of
by
sequence
Hsu of
to be independent
these
variables
be represented
by a121(T22 10*1
respectively.
The problem
to test
H0:
whether
2 2 2 U.1 = 2
a variance
0 0 &
shift
= cM = 02
6 has occurred,
(ao2
is unknown)
against
* * 0 = kM2
= 2
co
 Uk+2
0 0 *
= cM 
c0 +6,
16 >0,
where
k is
(c02+6) ! (0,o) .
unknown Define
(k=1,2,
S Ml)ï¿½
Xi= (RiA)
6 is
unknown
The
and
test
statistic
T is defined
as
S
l=l
(i1)
xi
I O
(MI)
is
is
2
Ha:
U~k+l1
xi
i=1
il, 2, ...,M.
69
Under
the null
hypothesis
of
no variance
shift,
this
test
statistic 1) (M+2) ].
has mean
use
E[T]=
the
and variance
sample
VAR[T]= (M+I) / [6 (M
approximation
0.5)/(VAR[T] )1
which
is distributed
asymptotically
standard
normal.
In this
test,
I calculate
T
for each
sample
period
with
M=100. sample
level
firms3.
I deem
period of just
that
a variance
if T*<2.6.
under
0.5%.
This
decrease
corresponds
firm
to
sample
83 had
in that
is 328
variance variance
changes
increases.
in the 7 sample
189 firms
(58%
periods.
56 firms
of the sample)
had only
had at least
one variance decrease
in the 7 sample
periods.
The results
analysis
are
shown
in
Table
20.
The
frequency
distribution
show
a preponderance
dividend
initiation
(i.e.
decreases f variance
in sample
by sample changes
periods
periods
does
occurring
5, 6 and 7).
not
after This
is confirmed
by
the Wilcoxon
signedrank
statistic's
value
0.70.
I conclude
from
this
analysis
that
the decrease
in returns
variability
initiation.
does
not
Variance
occur
decreases
systematically
occur
with
after equal
dividend frequency
before
and after
dividend
initiation.
Four
an uninterru subsequently
OTC firms
pted
were
series
excluded
because
they
of consecutive returns.
did not have These firms
traded on either the AMEX
large
T*= (T
has occurred
Of the 328 firms
The total examined,
a significance
size
no significant
this
of variance
of
of
or NYSE .
70
ANALYSIS
TABLE 20 OF THE TIMING OF
VARIANCE
DECREASES
a The announcement
A variance
date
decrease is
of the initial
deemed
to have
dividend n. occurred
is day if T*
'1.
<2.6
for that
sample
period.
Variance before a
decrease nd after
occur with
dividend
equal
initiation.
COUNT2
 COUNT 1
against
Variance
dividend
decreases
occur
more
initiation.
COUNT2
 COUNT1
where
COUNT1
TOTAL
OCCURRING
NUMBER
OF
IN SAMPLE
VARIANCE PERIODS 1
DECREASES 2 OR 3 FOR
FIRM j.
TOTAL
OCCURRING FIRM j.
NUMBER
OF
IN SAMPLE
VARIANCE PERIODS 5
DECREASES 6 OR 7 FOR
The Wilcoxon
signedrank
statistic
W is 0.70.
Cannot Sample
reject null at the
is 328 firms.
Sample 1 2 3 4 5 6 7
Period
Beginning 200 150 100 50 1 51 101
Returns ____________ ____Ending 101 51 1 50 100 150 200 Returna
Decrease sb 36 41 31 33 44 39 35
H0:
frequency
=0
frequently
after
> 0
COUNT2
5% significance level.
size
71
The Effect
of Clustered
Data
I examine
the possibility
that
specific
periods are causing
the observed
decrease
in firm
volatility.
Just
over
75% of
sample
firms
initiated
dividends
in the period
197277,
with
over
50% occurring
between
19751977.
The effect
of interest
rates on volatility
is particularly
relevant
to
my sample.
seventies experienced
volatile
changes
in interest
rates.
particular,
interest
rates
rose
from
197273
and fell
from
197476. It is plausible my sample, predividend
that
for
periods
a large
number
coincided
witt
of firms in
ia rising
interest
rate
volatility)
environment
and postdividend
(and
periods
coincided
high with
returns falling
interest
rates
(low
returns
volatility).
If the decrease
in firm
volatility
is independent
secular variance
influences,
declines
the proportion
should
of firms
experiencing
Table
21 shows percentage variance changes
calculated
for 500
sample
periods
distributed
by
the
year
of dividend
initiation. experiencing
The table variance
shows
that
declines
the proportion
is not constant
of firms
across
years.
In particular,
the periods
197273
and 197879
have
majority of the period
firms
197576
experiencing variance increases. Conversely,
which
accounts
for almost
40% of
my sample
of dividend
initiating
firms
has above average
proportion
4
(1989)
See,
for example,
Livingston
p . 37.
my
The
In
consequently
day
of
not cluster around
certain
periods.
the
of
(1990)
p .,
or Madura
72
TABLE
21
OF PERCENTAGE
BY YEAR
VARIANCE
OF DIVIDEND
CHANGES
INITIATION
Year No.a a pos.b % Pos . Mean Median
1970 6 5 83.3 20.4 24.5 1971 4 4 100.0 34.2 33.2 1972 18 1 5.6 73.1 33.2 1973 29 4 13.8 57.6 63.9 1974 24 16 66.7 4.5 19.1 1975 58 57 98.3 54.5 56.8 1976 62 62 100.0 53.4 57.2 1977 47 33 70.2 13.1 26.6 1978 20 9 45.0 22.4 4.3 1979 11 4 36.4 7.2 4.6 1980 9 6 66.7 8.0 31.1 1981 6 5 83.3 21.8 35.0 1982 3 3 100.0 19.8 16.8 1983 5 4 80.0 23.7 21.7 1986 3 0 0.0 60.1 69.0
a Number b Number
changes C Sample
of sample of sample
firms firms
i.e. variance
firms
with
of sample
initiating
with posit decreases.
tive
percentage
firms initiatin
in
that
percentage
variance ch ig dividends
year.
variance
a
nges as in that
(500
DISTRIBUTION
DAY SAMPLE
PERIOD)
a percentage year.
dividends
positive
73
variance
decreases.
This
pattern
is also
found
for variances
calculated
for 900,
250 and 100 day sample
periods.
Table changes
22 shows
adjusted
by
the distribution
a estimate
of percentage
of the contemporaneou
variance s market
variance.
Note
that
after
adjusting
for market
variance,
mean and median
period 197576.
experience
percenta Firms
decreases
ge variance initiating
in variances
change
dividen
but o f
is negative f ds in those
a smaller
for the years
magnitude
than
the market.
The behavior
of sample
firm
variances
through
time
plotted variance weighted
in Figures
of
an index
average
3 through 5. of dividend
of all firms
Figure 3 plots initiating firms on the NYSE/AMEX
the monthly
(VAR)
and
and NASDAQ
tapes. monthly
For each returns
month
(m=1/70
variance
for
to 12/86),
VAR is the average
a set of dividend
initiating
firms
within
Nm 0
A firm
24 months
is included of month m.
in Nm
if it initiated
dividends
Varj,
The monthly
returns
variance
for the firm
is calculated
Varj ,m
where
Tm
is the number
of trading
days
in month
m. If
a firm
has less
from VAR.
than
15 valid
returns
in any
month,
it is excluded
the
is
VARm
Nm
Nmj=
Tm
=1
R2j
as
VAR thus represents
variance
monthly
for
the mean
74
DISTRIBUTION
OF MARKET
TABLE ADJUSTED
22
PERCENTAGE
VARIANCE
CHANGES
DAY SAMPLE
PERIOD)
BY YEAR
OF DIVIDEND INITIATION
Year No. pos.b % Pos . Mean Median 1970 6 2 33.3 3.9 9.7 1971 4 1 25.0 14.9 15.6 1972 18 14 77.8 28.8 58.1 1973 29 29 100.0 51.4 51.8 1974 24 22 91.7 31.0 37.7 1975 58 9 15.5 48.4 36.5 1976 62 24 39.7 50.5 16.2 1977 47 35 74.5 18.6 32.0 1978 20 18 90.0 45.2 53.3 1979 11 11 100.0 39.9 42.5 1980 9 6 66.7 17.3 34.5 1981 6 5 83.3 25.0 33.9 1982 3 2 66.7 6.5 8.8 1983 5 1 20.0 38.6 22.9 1986 3 3 100.0 63.7 61.8
a Number b Number
changes C Sample
of sample of sample
firms
firms
firms
initiating with posit decreases.
rive
in
that
year.
variance
hanges
as
a percentage year.
of sample firms initiating
dividends
in that
(500
i.e. variance
with
dividends
positive
percentage
percentage variance c
75
the sample
of dividend
initiating
firms
through
a moving
window
of 24 months.
From
1/70
to 12/72,
the equallyweighted
VarM
is the monthly
NYSE/AMEX
market
returns
index.
variance
From
1/73
12/86,
VarM
is
a weighted
average
of the monthly
returns
variance indices.
of the equallyweighted
NYSE/AMEX
and NASDAQ
market
VarMm
= wVarNYSE/AMEX,m
+ (1W) VarNASDAQ,m
The weight NYSE/AMEX variances
w corresponds
firms
to the
in the sample
of the market
indices
proportional
(w
= 247/332) .
are calculated
representation
of
The monthly
in the same
way
as for individual
firms.
Figure
3 shows
the wide
variations
in returns
variance
the sample.
The returns
variance
is particularly
high
in the
years
1973
to 1976.
The sample
variance
behaves
similarly
the wider market.
Therefore,
variations
in return variance
not peculiar
to the sample
of dividend
initiating
firms.
Figure
4 plots
the monthly
variance
and the distribution of
firms
total
initiating dividends,
sample
INIT) .
This
expressed
figure
as a percentage
illustrates
of the
the effect
of
clustering
initiating
in the sample.
dividends
between
The large
1975
and 1977
proportion
follows
of firms
a period
of
above
average
returns
variance.
This potentially could
explain
the reported
decrease
in returns
variance
and quarterly
revaluations.
of
to
of
to
are
price
earnings
76
Figure
Monthly
Returns
Firms
(VAR
Variance
.) and a
of
an Index
Weighted
of Dividend
Market
Index
Initiating
(VarM) .
Figure
rate. Rates:
Reserve
The
5 plots source
Money
the monthly variance and
for the 3 month
and Capital
Bulletin.
Market
The figure
Tbill
Rates
shows
the 3month
rate
tables
that
Tbill
is the Interest
from
the Federal
the increase
in
returns preceded
variance
by increases
in the middle
seventies
in interest rates.
This
appears
to be
is not to
say
that interest
10
9
8
7
3
2
1 _  N 7001 7101 7201 7301 7401 7501 7601 7701 7801 7901 8001 8101 8201 8301 8401 8 6601
VAR  VarM
 I
rates
have caused
the changes
in returns
77
volatility.
They
could
both
be reactions
some
other
macro
economic
event,
such
as the oil shock
of 1973.
Figure
Monthly
Firms
Returns S(VAR)
Variance
of
an Index
and the Distribution c
Dividends
of Dividend )f Firms Ini
Initiating
Ltiating
INIT)
a Values o: b Multiply
n the vertical axis values on the verti
are in percentages. cal axis by 103.
7001 7101 7201 7301 7401 7501 7601 7701 7801 7901 8001 8101 8201 8301 8401 8501 8801
b
% INITbVAR
I
Monthly
Firms
Returns (VAR) a
Variance
78
Figure
of
nd the 3Mont
an Index o h Treasury
f Dividend
Bill
Rate
Initiating (TBill)
a Source:
Federal
Reserve
Bulletin.
Values
on the vertical
axis are b Multiply
in percentages. values on the vertical
axis
The general
conclusion
that
can be drawn
from
the above
analysis
dividend factors. negative
is that
variance
initiation Variance
decreases
i are not caused decreases do not
seemingly
by
associated
information
reduce
with
related
the degree
of
autocorrelation
7001 7101 7201 7301 7401 7501 7601 7701 7801 7901 8001 8101 8201 8301 8401 8501 8601 TBilla VAR b
by
I
103.
reductions in
would indicate
that
79
noise
trading
or
bidask
spreads.
Neither
do
they
systematically
occur
following
dividend
initiation.
variance seventies coincide
decreases
when with
appear
a large periods
to be
occur
proportion
mostly
of dividend
of low interest
rates
in the middle
initiations
preceded
by
periods
of high
interest
rates.
Variance
decreases
could
therefore
be driven
by changes
in interest
rates
or other
macroeconomic
events.
The evidence
in this
and the previous
chapter
tends
to reject
the hypothesis
that there are long
effects associated with dividend
The
run
information
initiation.
CHAPTER
SUMMARY
AND CONCLUSIONS
It has been
suggested
that
dividends
play
a role
transmitting
information
to investors.
This
hypothesis
motivated
the examination of the
relationship between
dividend
initiation
found
that
and earnings
announcements.
Previous
the price reactions at earnings
research
announcements
has
are
on average
result
lower
has been
after
dividends
interpreted
have
to support
been
initiated.
the contention
This that
dividends preempted
has conveyed
information
information.
that
would
Dividend
otherwise
information
have
been
has
conveyed
by earnings
reports.
I contend
that
this
conclusion
is invalid.
Showing
that
price
reactions
have
declined
merely
shows
that
earnings
dividends implication
dividends r appropriate
information of dividends
is
substitutable.
The
as an information source
educe costly private
empirical
measure
information
is
therefore
testable is whether
acquisition.
The
not earnings
returns
variance
alone
but standardized
returns variance.
nonannouncement
activities
of
: returns investors,
reflects then s
the private tandardiz ing
information announcement
returns gives a
variance
measure
by
the surrounding
of the importance
nonannouncement
of the public
returns
announcement.
80
in
has
and
If
81
If private dividends,
information
investors
acquisition
should
place
is reduced
greater
reliance
by having
on the
firm's
public
announcements,
both
dividends
and earnings.
addition,
this
reliance
should
be greatest
for firms
for which
information
is least
readily
available.
I do not detect
any
difference
in the relative
measures
the information
content
of earnings announcements
before
after sample
dividend
of firms
initiation.
as well
This
as when
result
holds
the sample
for the total
is grouped
by
proxies
of
information
availability.
The
information
transmission
hypothesis
is not supported.
What
is puzzling
is that
there
is
a significant
reduction
raw price
reactions
for postdividend
initiation earnings
announcements.
I contend
that
this
reduction
is caused
genera
with eviden
1 decrease dividend ce that th
in the volatility
initiation. e reduction
of returns
Regression
which
analysis
in postdividends
earni
coincides provides .ngs price
reactions reduction
is related
to the variance
in the information
content
decrease
and not to
of earnings.
It can be argued
that
an information
effect
associated
with
dividend initiation causes the
decrease
in returns
volatility.
For example,
i f the
initiation of
dividends
induces greater
consensus
among
traders,
noise trading
or bidask
spreads
be reduced.
I test
for reductions
in variance
ratios
before
and after
dividend
initiation.
I do not find
any significant
Further, one should
In
of
and
in
by
may
change.
o f variance
observe
the frequency
82
declines timing of decreases dividend
to be higher
variance appear
after
dividend
changes show with equal
initiation.
no evidence frequency
Tests
of this.
before
of the
Variance
and
after
initiation.
The findings
hypothesis
that
of this dividends
study play
are not supportive
an informational
of the
role .
The
search
for a valid reason
why
firms
initiate
dividend
payments
must
still
continue.
However,
it does
provide
evidence
that
market
is
sufficiently
efficient
such
that
from dividends are minor.
the
any
informational effects
83
REFERENCES
Asquith, P.
Dividend Business
and D. W. Mullins, 1983, "The Payments on Shareholders' 56, 7796.
Impact Wealth,
of
of
Initiating
Journal of
Atiase, R., 1985,
Capitalization, Announcements,"
"Predisclosure Information, Firm and Security Price Behavior Around Earnings Journal of Accounting Research 23, 2136.
Bagehot, W., 1971,
Analysts Journal
"The 22,
Only Game in Town," 1214.
Financial
Black,
F. , 1986,
"Noise", Journal of Finance 41,
529543.
Beaver, W.
Earnings Selected Research
H., 1968, "The Information Content Announcements," Empirical Research Studies, Supplement to the Journal 6, 6792.
of in
of
Annual Accounting:
Accountinc
Bhattacharya, S.,
Policy, and the of Economics 10
1979 ," 'Bird 2592
Imperfect Information, in the Hand' Fallacy,", 70.
Dividend Bell Journal
Bhushan, R., 1989, "Collection of Information
Traded Firms: Theory and Evidence," Journal
and Economics 11, 183206.
About Publicly of Accounting
Christie,
Stock
432.
A. A., 1982, "The
Variances,1" Journal
Stochastic Behavior of of Financial Economics
Common 10, 407
Cochrane,
Journal
J. H., 1988, "How Big of Political Economy
is the Random Walk in GNP?," 96, 893920.
COMPUSTAT, Standard and Poor's Compustat Services, Englewood,
Colo.
Copelandr
BidAsk
T. , and Spread,
D.
Galai, Journal
1983, "Information Effects of Finance 38, 14571469.
on the
CRSP Daily Stock Master Returns and Index, Center for Research
in Security Prices, University of Chicago, Chicago, Ill.
Demsetz, H., 1968, "The
Journal of Economics
Cost 82,
of
335
Transacting," Quarterly 13.
84
Diamond, D. W.,andaRE. Verrecchia,
Aggregation in a Noisy Rational Journal of Financial Economics 9,
1981, "Informational Expectations Economy," 221235.
Grant, E. B., 1980, "Market
Amounts Information of Accounting Research 18,
Implications of Differential Interim Information," Journal 255268.
Fama, E. F. and K. R. French, .
Components of Stock Prices,
96, 246273.
1988, "Permanent and Temporary " Journal of Political Economy
Federal Reserve Bulletin, Board of Governors of the Federal
Reserve System, Washington, D. C.
French, K.,
Arrival Journal
an of of
d R. Roll, 1986, "S Information and Financial Economics
tock
the 17,
Return Variances: The Reaction of Traders," 526.
Glosten, L. ,
Prices in Traders,"
and P. Milgrom, 1985,
a Specialist Market wi Journal of Financial
"Bid, Ask and Transaction th Heterogeneously Informed Economics 14, 71100.
Healy, P., and K. Palepu, 1988
by Dividend Initiations
Financial Economics 21, 14
"Earnings Information Conveyed and Omissions," Journal of 9175.
Hellwig, M. F.,
Competitive
498.
1980, "
Markets,
'On
the Aggregation of Journal of Economic
Information Theory 22,
in 477
T., and H. Transaction Economics 9
Stoll, 1981, "Optimal Dealer Pricing under and Return Uncertainty," Journal of Financial 4774.
Hollander, M. and
Methods, (New
D. A.
York,
Wolfe, 1973, Nonparametric Statistical NY: Wiley).
Hsu, D. A.,
Point, "
1977,
Appliec
"Tests of Variance Shift at an Unknown Time d Statistics 26, 279284.
Hsu, D. A., 1979,
Sequences with
Flow Analysis, Association 74,
"Detecting Applicatio " Journal 3140.
Shifts of n. to Stock of the
Parameter Price and American
in Gamma
Air Traffic Statistical
John, K.
Taxes: 10531
and J. Williams, 1985, "Dividends, Dilution and A Signalling Equilibrium," Journal of Finance 40,
070.
of
Ho
I
85
Kaul G. and M
Errors of Economics
. Nimalendran, 1990, "Price Reversals:
Market Overreaction?," Journal of 28, 6793.
BidAsk Financial
Livingston, M., 1990, Money and
Instruments and Their Uses,
PrenticeHall).
Capital Markets: Financial
(Englewood Cliffs, N. J.:
Lo, A. W
Not
Speci
* and A. Follow fication
C. MacKinlay, Random Walks Test," Review
1988, "Stock
Evidence of Financial
Market I from Studies
Prices Do a Simple 1, 4166.
Lo, A. W. and A.
the Variance Investigation
C. MacKinlay, Ratio Test in ,1" Journal of
1989, "The Size Finite Samples: Econometrics 40,
and Power of A Monte Carlo 203238.
Lobo, G. J. and A. A. W. Mahmoud, 1989, "Relationship Between
Differential Information and Security Return Variability,"
Journal of Accounting Research 27, 116134.
Madura,
(St.
Paul
1989, Financial Markets , Minn.: West Publishing
Marsh, T., and R. Merton,
Aggregate Stock Market,
and Institutions,
Co.).
1987, "Dividend Behavior
" Journal of Business 60,
for the 140.
Miller, M.
Growth
34, 411
H. and and the
433.
F. Modigliani, 1961, Valuation of Shares,
"Dividend " Journal
Policy, of Business
Miller, M. H.
Asymmetric
and K. Rock, 1985, "Dividend Policy Information," Journal of Finance 40,
Under 10311051.
Richardson, M. and T. Smith, 1991,
with Tests for Serial Correlati Paper, Rodney L. White Center Wharton School, University of
"Robust Power on in Stock Pr: for Financial Pennsylvania.
Ross, S. A., 1976, "The Determination of
The IncentiveSignalling Approach,
Economics Spring, 2340.
Financi " Bell
Calculations ices," Working Research, The
al Structure:
Journal of
Shores, D., 1990, "The Associa"tion
Security Returns Surrounding Journal of Accounting Research
Thompson, R. B., II, C. Olsen, and J.
"Attributes of News About Firms: Specific News Reported in the Wall Journal of Accounting Research 25,
Between Interim and Earnings Announcements," 28g 164181.
R. Dietrich, 1987, An Analysis of FirmStreet Journal Index," 245274.
86
Venkatesh, P. C., 1989, "The Impact of Dividend Initiation
the Information Content of Earnings Announcements Returns Volatility", Journal of Business 62, 1751970
Verrecchia, R. E., 1982, "Information Acquisition
Rational Expectations Economy," Econometrica 50,
in a Noisy
14151430.
Wall Street Journal Index,
Dow Jones and Co., New York, N.
Zeghal, D., 1984, "Firm Size
Financial Statements,"
Quantitative Analysis 19
and the Informational Content of Journal of Financial and
299310.
on and
Y e
87
BIOGRAPHICAL
SKETCH
Thian
Soon
Wong was
born
in Johor
Bahru,
Malaysia,
on March
1958.
He received
a bachelor's
degree
in electrical
engineering Malaysia, ir
from
n 1982
the University
and a Master
of Malaya,
of Business
Kuala
Lumpur,
Administration
from
the University
of Florida
in
1986.
He expects
to receive
Doctor
of Philosophy
degree
in finance
from
the University
Florida in 1991.
19,
of
I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and its fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.
/ I  / / ,
Robert C. Radcliffe, Chairman Professor of Finance, Insurance, Real Estate
and
I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and its fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy.
I
Anand Desai Assistant Profe Insurance, and R(
.ssor of eal Estate
Finance,
I certify that I have read this study and that in opinion it conforms to acceptable standards of schola presentation and its fully adequate, in scope and quality, a dissertation for the degree of Doctor of Philosophy.
Michael D Assistant Insurance
2.
Ryngaert Professor of and Real Estate
I certify that I have read this study and that in opinion it conforms to acceptable standards of schola presentation and its fully adequate, in scope and quality, a dissertation for the degree of Doctor of Philosophy.
my rly as
SanforVBerg Professor of Economics
This dissertation was sulaitted to the Graduate Faculty of the Department of Finance, Insurance, and Real Estate in the College of Business Administration and to the Graduate School and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy.
Dean, Graduate School
my
rly as
Finance,

Full Text 
PAGE 1
DIVIDEND INITIATION AND DIFFERENTIAL INFORMATION AN EMPIRICAL INVESTIGATION BY THIAN SOON WONG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1991 Â• Â•
PAGE 2
ACKNOWLEDGEMENTS I gratefully acknowledge the encouragement, patience and guidance of Professor Robert Radcliffe, my dissertation committee chairman. I also thank Professors Michael Ryngaert, Anand Desai, and Sanford Berg whose help and guidance greatly improved this dissertation. I owe a special debt to C. Sloan Swindle without whose support, insight and encouragement this study would not have been completed. Finally, I dedicate this study to my parents: Grace Foo Kai Chi and Wong Cheng Boon. Â«r 11
PAGE 3
TABLE OF CONTENTS page ACKNOWLEDGEMENTS ii ABSTRACT iv CHAPTERS 1 INTRODUCTION 1 2 INFORMATION AND DIVIDEND INITIATION 5 Prior Empirical Work 6 Causes for Returns Volatility Decreases .... 11 3 DATA AND METHODOLOGY 15 Data 15 Tests on Quarterly Earnings 18 4 TESTS ON QUARTERLY EARNINGS ANNOUNCEMENTS ... 35 The Overall Sample 35 The Sample Grouped by Information Availability . 40 The Market Reaction to Earnings Changes .... 51 5 RETURNS VOLATILITY AROUND DIVIDEND INITIATION . 57 Variance Changes in the Overall Sample 58 Variance Ratio Tests 61 The Timing of Variance Changes 65 The Effect of Clustered Data 71 6 SUMMARY AND CONCLUSIONS 80 REFERENCES 83 BIOGRAPHICAL SKETCH 87 Â• Â• Â• 111
PAGE 4
Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy DIVIDEND INITIATION AND DIFFERENTIAL INFORMATION: AN EMPIRICAL INVESTIGATION By Thian Soon Wong December 1991 Chairman: Robert Radcliffe Major Department: Finance, Insurance, and Real Estate The information transmission hypothesis states that firms begin paying dividends as a means of transmitting information to investors. Evidence that dividend and earnings information is partially substitutable has been used to support this hypothesis. This is not an appropriate test. Instead, a testable implication of the information transmission hypothesis is that dividend initiation should increase the proportion of publicly disclosed to privately acquired information . The informativeness of quarterly earnings announcements before and after dividend initiation is examined using returns and standardized variance measures. Under the information transmission hypothesis, the standardized variance of earnings and dividend announcements, should increase after dividend IV
PAGE 5
initiation. Further, firms which have the least amount of available information should experience the greatest change. I find that while returns measures decrease after dividend initiation, standardized variance measures do not. Firms with the low information availability have larger price revaluations. This relationship is not found with standardized variance. There appears to be no evidence that the proportion of publicly disclosed information increases after dividend initiation. The observed decrease in returns measures of information content may be explained by a reduction in returns volatility that coincides with dividend initiation. Tests are conducted to detect if an information effect associated with dividend initiation is causing the returns volatility to decrease. Such an effect could cause returns volatility to decline by reducing bidask spreads or noise trading. Variance ratio tests fail to detect any such decrease following dividend initiation . Analysis of the timing of variance changes reveals that dividend decreases are equally likely before as after dividend initiation. Therefore, decreases in returns volatility do not systematically occur after dividend initiation. This weakens the contention that information conveyed by the dividend initiation has caused the volatility decrease. The observed volatility decrease appears to be caused by a large proportion of sample firms initiating dividends in the v
PAGE 6
period 19741977. This period also coincides with a decrease in volatility for the market in general. After adjusting for market volatility, the reduction in firm volatility is much reduced. The general conclusion is that information related effects do not appear to cause the reduction in returns volatility. vi
PAGE 7
CHAPTER 1 INTRODUCTION Why firms pay dividends is an enduring puzzle in finance. One motivation that has been advanced is that dividends convey information to investors. This information content of dividends or dividend signalling hypothesis was first suggested by Miller and Modigliani (1961) and subsequently formalized by Ross (1977), Bhattacharya (1979), Miller and Rock (1985) and John and Williams (1985) among others. In addition to this signalling hypothesis, some researchers have proposed that initiating dividends may augment the process by which information about firms is transmitted to investors. I refer to this expanded role for dividends as the information transmission hypothesis. For example, Asquith and Mullins (1983) state that Dividend policy has several attractive aspects as an information transmission mechanism. Unlike the detailed focus of other announcements, dividends can be used as a simple, comprehensive signal of management's interpretation of the firm's recent performance and its future prospects. Unlike most announcements, dividend announcements must be backed with hard cold cash. The firm must either generate this cash or convince the capital markets to supply it. In addition to the credibility of cash signals, dividends are also highly visible compared with other announcements. . . . Once dividends are initiated, shareholders apparently anticipate a periodic signal by management, and management is forced to submit to a periodic review, (p. 94) 1
PAGE 8
2 Two recent studies by Venkatesh (1989) and Healy and Palepu (1988) report results that can be interpreted to support this view of why firms pay dividends. If dividend information is substitutable for earnings information, then earnings announcements before dividend initiation should convey more information than postdividend initiation earnings announcements. Venkatesh finds that the average informativeness of quarterly earnings announcements (measured by the amount of price revaluation) decreases after dividend initiation. Healy and Palepu find that the price response to annual earnings information is lower in the years following dividend initiation. I contend that the above results merely show that earnings and dividend information is substitutable. Demonstrating substitutability is not compelling evidence of the information transmission hypothesis. Why would a firm engage in costly information transmission via dividend payments when earnings reports will serve the same purpose. I argue that a firm would pursue an enhanced information transmission policy only if such a policy reduces the cost of information acquisition for investors. Firms for which it is relatively costly for investors to acquire information will have greater incentive to disclose information through dividends. Consequently, if dividends augment the available public information about the firm, then the relative amount of private information acquisition after dividend initiation should decrease.
PAGE 9
3 I test the above information transmission hypothesis by comparing the average standardized variance of announcements from the preand postdividend initiation periods. The use of standardized variances, i.e. announcement period variance divided by estimation period variance, avoids problems with changes in returns volatility. Standardized variance measures the amount of "new" information released at earnings and dividends announcements relative to the amount of information privately acquired during surrounding nonannouncement periods. If dividend initiation alters the proportion of publicly available to privately acquired information, standardized variances of earnings and dividend announcements should increase after dividend initiation. Further, one would expect firms with low information availability to experience the greatest increase in the relative informativeness of public announcements. To test this hypothesis, I use two proxies of information availability (namely, market value of equity and number of Wall Street Journal Index news items) to divide a sample of dividend initiating firms into five groups. I then test for differences, across groups, in the relative informativeness of these public announcements before and after dividend initiation. The results from this study do not support the hypothesis that dividend initiation changes the relative amount of public versus private information. Further, I find no differential
PAGE 10
4 changes in relative informativeness across samples of firms grouped by information availability. If changes in information are not the basis for the findings that Venkatesh and Healy and Palepu report, then what is? Venkatesh finds that returns variability is lower in the postdividend period. I present evidence that their results are driven by a contemporaneous decline in the returns volatility of firms initiating dividends. This decline leads Venkatesh and Healy and Palepu to interpret postdividend initiation earnings announcements as being less informative. I investigate the nature of this returns volatility decline. I find no evidence that the decrease in returns volatility is driven by information related causes. This paper is organized as follows. Chapter 2 develops the implications of the information transmission hypothesis. I critigue the Venkatesh and Healy and Palepu studies within the framework of this hypothesis. I also discuss possible causes for the decline in returns volatility following dividend initiation. Chapter 3 presents my data and the methodology used to examine the information content of guarterly earnings. In Chapter 4 , I report and discuss the results of tests on preand postdividend initiation guarterly earnings. Chapter 5 presents tests on changes in returns volatility. Chapter 6 summarizes the results of this study.
PAGE 11
CHAPTER 2 INFORMATION AND DIVIDEND INITIATION Asquith and Mullins (1983) hypothesize that dividends may provide a vehicle for communicating management's superior information concerning the firm's current and future prospects. Information about a firm can come from two sources: information provided by the firm (publicly released information) and information privately acquired by investors. I assume that firms have a cost advantage in producing information about themselves. Therefore, managers prefer to release information publicly than have investors produce it at greater cost. 1 One way dividends can transmit information to investors is by providing substantiation for the cashflows reported in the earnings announcement. Without dividends, investors would have to seek substantiation from other costly sources. Venkatesh (1989) alludes to this argument in his paper. His explanation for the postdividend decrease in returns volatility is that "investors accord less importance to pieces of 'information' (announcements/rumors) that could have induced price reactions in the predividend period" (p. 176) . If dividends play an 1 It is not necessary to assume that managers do this for altruistic purposes. By reducing the costs of investigation, managers obtain a lower required rate of return. 5
PAGE 12
6 informative role, the major testable implication is that private information acquisition will decrease in the postdividend initiation period . 2 As the firm discloses a greater proportion of information through earnings and dividend announcements, investors have less incentive to acquire information privately. Therefore, the relative proportion of public versus private information will increase following dividend initiation. I make two assumptions to test this hypothesis. First, the information conveyed by dividends is substitutable for information imparted by earnings announcements. Second, the total amount of firm information does not change with dividend initiation. Prior Empirical Work Venkatesh (1989) studies a sample of 72 NYSE or AMEX firms that initiated quarterly dividend payments between 19721983. He collects the announcement dates of quarterly earnings and dividend payments for 14 quarters before and after dividend initiation. Using raw and excess returns from the earnings announcement as a measure of information content, he finds that the average information content of quarterly earnings announcements before dividend initiation is significantly greater. Based on this evidence, he concludes that earnings 2 The issues of information substitutability and its implications are formally explored in a rational expectations framework by Hellwig (1980), Diamond and Verrecchia (1981), Verrecchia (1982) and Bhushan (1989).
PAGE 13
7 and dividend information are partially substitutable and that information is transmitted by dividends. Venkatesh also finds that returns volatility is lower after dividend initiation. His explanation for this decrease is that investors in the postdividend initiation period attribute less importance to information arriving during nonannouncement periods. This leads to smaller price reactions and lower observed volatility. Venkatesh assumes that investors do not actively acquire information. Instead, they passively receive information and decide whether to act on this information. Before dividend initiation, they are more likely to act on nonannouncement period information. In the postdividend period, investors receive more publicly announced information and rely less on nonannouncement period information . The main focus of Healy and Palepu's (1988) study is to determine whether dividend initiations convey information about future earnings. They hypothesize that investors view dividend changes as management forecasts, substantiated by cash, of future earnings changes. So, dividend announcements enable investors to revise expectations of earnings. The forecast errors in earnings announcements following dividend initiation should therefore be reduced. To test this hypothesis, they obtain the price response (namely, twoday cumulative excess returns) and earnings surprise (the earnings change standardized by price) for 5
PAGE 14
8 annual earnings announcements prior to and following dividend initiation. They regress the price response on the earnings surprise to obtain the earnings response coefficient. This measures the informativeness of the earnings surprise. Other things being equal, earnings information that is unanticipated leads to large price revaluations and therefore to larger values of the earnings response coefficient. They find that the earnings response coefficient is lower in the postdividend initiation period. Healy and Palepu argue that dividend initiation provides information which preempts subsequent earnings announcements. Do Venkatesh ' s and Healy and Palepu' s results regarding earnings provide support for the information transmission hypothesis? I argue they might not. These studies suggest that dividend and earnings information are substitutable. Managers have little motivation to initiate dividends if doing so merely shifts to dividends, information that would otherwise be conveyed by earnings. This is especially true since dividends are costly to investors. First, they are costly because dividends are taxed at ordinary income rates. If the firm chooses instead to retain earnings, taxes are reduced to the extent that capital gains are postponed. Second, dividends must be paid out of net income that would otherwise have been used for investment. Replacing this source of capital incurs transactions costs that are ultimately borne by investors. Third, initiating dividends also can lead to shifts in
PAGE 15
9 clienteles that are costly to investors. Finally, there are costs of administering a dividend policy. Therefore, dividends must play a substantially greater informative role than documented by Venkatesh or Healy and Palepu to justify their cost . I hypothesize that this greater role must be the increase in publicly released information and a consequent decrease in costly private information acquisition. The Venkatesh and Healy and Palepu studies cannot address this hypothesis for two reasons. First, in the postdividend initiation, there are two routes through which public information is transmitted: dividends and earnings. A priori . we do not know how the total public information is divided between these two routes. By comparing only earnings announcements, the researcher is neglecting the amount of information conveyed by dividends. He can infer the effect of that information, but cannot quantify it. From inference, he can only conclude that dividend information has substituted for earnings information. As an illustration, assume the total amount of available information is a. This amount remains constant for preand postdividend initiation periods. Suppose that before dividend initiation, half (a/2) is publicly disclosed and half is privately acquired. After dividend initiation, earnings disclose a/3, dividends disclose a/3 and private acquisition accounts for the remaining third. We see that after dividend initiation, twothirds of the information is publicly
PAGE 16
10 disclosed. Yet if one observes only earnings all one can conclude is that the information content of earnings has declined from a/2 to a/3. We can conclude little about how much information was transmitted by dividends. Any attempt to gauge the relative proportion of public versus private information must examine the informativeness of both earnings and dividend announcements. The second deficiency in the Venkatesh and Healy and Palepu studies lies in their information content measures. They base their measures on announcement period returns that do not account for the general reduction in returns volatility. This decrease in returns volatility may be due to causes unrelated to any information effect associated with dividend initiation. Suppose that announcement period returns variance decreases proportionally with the decrease in returns volatility. Then one would observe postdividend initiation announcement returns that are lower in magnitude than predividend initiation announcement returns. If this is so, one cannot distinguish a decrease in announcement period returns as being due to information related causes or otherwise. A study of the nature of changes in returns volatility around dividend initiation is called for. In the next section, I review several possible causes for a decrease in returns volatility.
PAGE 17
11 Causes for Returns Volatility Decreases Information related causes. In informationally efficient markets where new information is rapidly incorporated into price, returns volatility is directly related to the rate of information arrival. Lower volatility may be caused by a reduction in the rate at which information arrives. It is possible that when returns on the firm's underlying assets become less uncertain and more predictable, a decrease in returns volatility may occur. However, this effect does not depend on any changes in the process by which information is disseminated or processed about the firm. The decision to initiate dividends may coincide with the firm's asset returns becoming less uncertain. Decreases in returns volatility are then only coincidentally associated with dividend initiation. The initiation of dividends does not cause the rate of information arrival to decrease. An alternative information related explanation for decreases in returns volatility is that dividend initiation reduces noise trading. Noise has been interpreted in various ways by Black (1986). I consider noise as the difference between the security's intrinsic value and its observed price. This mispricing occurs when traders overreact to the activities of other traders (French and Roll (1986)). Greater noise trading leads to higher volatility. Dividend initiation could reduce noise trading if it induces greater consensus among traders.
PAGE 18
12 The bidask spread. Firms with large bidask spreads tend to have greater returns volatility. This relationship arises in several ways. First, the larger the bidask spread the larger the difference between buy and sellside transactions. Observed returns are calculated using daily closing prices. If successive closing transactions are alternatively buy and sell transactions, they would be conducted at opposite edges of the bidask spread. Hence, the greater the bidask spread, the greater the observed return. Bidask spreads increase with the dealer's inventory costs (Demsetz, 1968 , Ho and Stoll, 1981 ) and the degree of informational asymmetry between informed traders and the uninformed dealer (Bagehot, 1971 , Copeland and Galai, 1983, Glosten and Milgrom, 1985 ) . It is possible that dividend initiation may alter these two factors and therefore reduce bidask spreads. For example, dividend initiation may increase trading volume. This reduces the dealer's inventory costs as inventory turnover increases. Dividend initiation may also reduce informational asymmetry through greater consensus between traders and dealers. Stock price . Volatility decreases as stock price increases. This relationship occurs for two reasons. First, for the same bidask spread, higher priced stocks will have smaller observed returns and therefore lower volatility. For example, suppose stock A has bid and ask prices of 9 7/8 and 10 1/8,
PAGE 19
13 respectively; stock B has bid and ask prices of 19 7/8 and 20 1/8. The bidask spread is 2/8 for both stocks. For successive sell and buy transactions the observed return for stocks A and B is 2.5% and 1.24%, respectively. Second, as a firm's stock price increases (without the firm increasing debt) , the firm's leverage decreases. Volatility is positively related to leverage. This can be easily seen by considering a firm in a ModiglianiMiller world with no taxes and riskless debt. The return on a leveraged firm k s is k s = k v + (k v r) (^) where r is the riskless interest rate and k v is the return on the unleveraged firm. The market value of debt and equity is D and S, respectively. Taking variances Â°S = a V^ + g) Therefore, stock volatility a s depends positively on leverage D/S. Interest rates. Volatility also changes with interest rates. Interest rates indirectly affect volatility through leverage. An increase in interest rates can both increase or decrease leverage. First, as interest rates rise, the value of both debt and equity both decline but equity declines proportionately greater than debt; this leads to an increase in leverage. Second, there is a wealth transfer from bond to
PAGE 20
14 equity holders that decreases leverage. Christie (1982) has studied the net effect of interest rates on leverage. He finds that high interest rates are associated with high returns volatility . If dividend initiation has information related effects, volatility can decrease due to reductions in noise trading and bidask spreads. Volatility can also decrease due to factors unrelated to information effects. The returns on the assets of firms initiating dividends could have become less uncertain and thus prices are less volatile. Dividend initiating firms could experience increases in stock price which lead to proportionally smaller bidask spreads and reduced leverage. Finally, returns volatility can decrease due to market wide changes in interest rates.
PAGE 21
CHAPTER 3 DATA AND METHODOLOGY Data My data sample consists of 332 firms that initiated dividends between 19701986 inclusive. These firms were identified by scanning the 1988 Center for Research in Security Prices (CRSP) NYSE/AMEX and NASDAQ files. I also collect earnings information for approximately 14 quarters (900 trading days) before and after the initial dividend. These were obtained from the 1987 quarterly Compustat files and/or the Wall Street Journal Index . To be included in the sample, firms must meet the following requirements. (1) The initiating dividend is either the first dividend in the firm's corporate history or the resumption of dividend payments after a hiatus of at least 10 years. The initial dividend cannot be an extra or special dividend . (2) The firm must continue paying dividends for at least three years after the initial dividend. (3) The firm must have 3 years of daily returns data before and after the initial dividend. 1 1 Not all the returns data for the 3 year period preceding and following dividend initiation are obtained from one CRSP tape source. Fourteen firms which were on the NASDAQ 15
PAGE 22
16 (4) One year must elapse between the firm's initial appearance on the NYSE/AMEX or NASDAQ tapes and the announcement of it's initial dividend. (5) In each 14 quarter period preceding and following the initial dividend, announcement dates and earnings and price data for at least 10 quarters must be available. This eliminates firms that only report annual earnings or firms merged out of existence. Table 1 presents an overview of the sample. Most of the dividend initiations occurred in the middle seventies. Almost 65% of the sample initiated dividends between 1975 and 1977. This could be due to the relaxation of the Nixon price controls or to crosssectional dependence in earnings changes as posited by Marsh and Merton (1987). More than threequarters of the sample paid at least 9 dividends in the 14 quarters following dividend initiation. The most common dividend payment frequency was quarterly, followed by half yearly and annual payments. American Exchange listed firms form a slight majority followed by New York Stock Exchange and OTC firms. More than 90% of the sample had at least 5 years of quarterly earnings data. tape when they initiated dividends had continuing data on the NYSE/AMEX tape. Similarly, 6 firms which were on the NYSE/AMEX tape when they initiated dividends had prior NASDAQ data.
PAGE 23
17 TABLE 1 OVERVIEW OF SAMPLE CHARACTERISTICS Cumulative Freq . Percent Freq. Percent Year of Dividend Initiation 1970 6 1.8 6 1.8 1971 4 1.2 10 3 . 0 1972 18 5.4 28 8.4 1973 31 9 . 3 59 17.8 1974 27 8 . 1 86 25.9 1975 62 18 . 7 148 44 . 6 1976 70 21.1 218 65.7 1977 49 24 . 8 267 80.4 1978 23 6.9 290 87.3 1979 12 3 . 6 302 91.0 1980 9 2.7 311 93.7 1981 * 6 1.8 317 95.5 1982 5 1 . 5 322 97.0 1983 6 1.8 328 98.8 1986 4 1.2 332 100.0 Number of Dividends 3 24 26 7.8 26 7.8 58 50 15.1 76 22.9 912 53 16.0 129 38.9 1316 203 61.1 332 100.0 Dividend Payment Frequency b Annual 21 6.3 21 6.3 Half 35 10.5 56 16.9 Quarter 217 65.4 273 83.2 Other c 59 17 . 8 332 100.0
PAGE 24
18 TABLE 1 continued Cumulative Freq . Percent Freq. Percent NYSE 104 31.3 104 31.3 Exchange d AMEX 143 43 . 1 247 74.4 NASDAQ 85 25.6 332 100.0 1619 12 3 . 6 12 3 . 6 Number of 2023 42 12 . 7 54 16.3 Quarterly Earnings 6 2427 105 31.6 159 47.9 2830 173 52 . 1 332 100.0 a Number of dividends which were declared within 900 trading days following the initial dividend. b The frequency of dividend payments in the 3 year period following dividend initiation. c Within the 3 year period following dividend initiation, 5 firms changed from an annual to a half yearly payment frequency. 6 firms changed from annual to quarterly. 44 firms changed from half yearly to quarterly. 1 firm changed from quarterly to half yearly. 2 firms changed from annual to half yearly to quarterly. 1 firm changed from half yearly with additional yearend dividends to quarterly. d Exchange on which firm was trading when dividends were initiated . e Number of quarters within 900 trading days of the declaration of the initial dividend for which earnings information was available. Tests on Quarterly Earnings Measuring information content. The information content of an earnings announcement is computed in the following way. The announcement period consists of 3 trading days: the day of the announcement (day t) , the day preceding (t1) and following (t+1) the announcement date. Beginning at days t2 and t+2, returns for 60 days surrounding the announcement period are
PAGE 25
19 collected. These returns form the nonannouncement period. If dividends and other earnings announcements occur within this nonannouncement period, the returns from these announcements are excluded. The 60 nonannouncement period returns are used to estimate the market model for firm j : R jr = <*j + PjKr' r =1 , 60 where R jT are the returns on day r for the firm j, R mT are the returns (including dividends) of a valueweighted index for all firms on the tape, and and (3^ are OLS estimated coefficients. Excess returns u jT are calculated as u . = R. (a, + 8Â±RÂ„) JT JT ' J TUT/ I assume that excess returns are normally distributed with zero expected value. Three measures of information content are used. The first is the standardized variance, a variant of BeaverÂ’s U (see Beaver, 1968) . It is denoted by BVU. This is the ratio of announcement period returns variance to nonannouncement period returns variance. Beaver's U is a measure of the "new" or marginal information conveyed to the market by the earnings (and dividends) announcement relative to the average information available during the nonannouncement period. It is also a measure of the relative amounts of publicly disclosed versus privately acguired information. I assume that nonannouncement period returns variances capture the extent
PAGE 26
20 of private information acquisition. This may not be strictly true as firms may engage in other forms of public disclosure, for example, earnings forecasts. However, their effect is minimized if dividend initiation does not systematically lead to changes in the firm's other public disclosures. The use of standardized variances also controls for changes in returns volatility. This is critical under two sets of circumstances. First, if the market were generally more variable during a certain period, a direct comparison of announcement period returns would bias the test of firmspecific price response to earnings announcements. This bias is exacerbated if the sample is clustered in calendar time and clustering is coincident with changes in market volatility. Second, if there is a systematic decrease in returns volatility contemporaneous with dividend initiation. Standardizing the announcement period returns variance by the nonannouncement period returns variance avoids problems that may arise from differences in volatility. I suspect that previously reported results may be driven by returns variance decreases and not by changes in information transmission. Let A represent the set of announcement period days and N represent the set of 60 nonannouncement period days. I estimate the announcement period returns variance to be the sum of the squared excess returns u T (r e A) divided by the number of days within the announcement period, T A . Similarly, the nonannouncement period returns variance is the sum of 60
PAGE 27
21 squared excess returns u T (r e N) divided by 60. The standardized variance is calculated as BVU, J Var[u jT , t eT A ] Var [ u jT , t Â£ T N ] ^ E u JT A reT A E V u. : rp JT reT N Standardized variance is calculated for two definitions of the announcement period. In the first definition, BVU E , the announcement period consists only of the earnings announcement. This definition is used for comparison with Venkatesh's results. As BVU E measures the marginal information content of earnings relative to the average information in the nonannouncement period, this ratio is used to examine the substitutability of earnings and dividend information. Decreases in postdividend initiation BVU E 1 s indicate that dividend announcements have preempted earnings announcements. In the second definition, BVU D , the announcement period consists of the earnings announcement and the nearest dividend announcement occurring within 30 days of the earnings announcement. This definition is used to measure the total effect of dividends as a public disclosure mechanism. In the postdividend initiation period, information is publicly disclosed through both dividends and earnings. In each quarter, BVU D measures the information content of all public disclosures (from both earnings and dividends) relative to private information. Increases in postdividend initiation
PAGE 28
22 BVU d 's indicate that more information is released publicly than acquired privately. I also calculate two measures which are based on announcement period returns. They are similar to those used by Venkatesh. They are measures of price revaluations induced by information released at earnings announcements. These measures do not take into account nonannouncement period returns and therefore, the general level of returns variability. The first measure, RAW, is the absolute value of the sum of earnings announcement period returns. t.i raw j = I E R J T =tl The second, MAJ, is the absolute value of the sum of earnings announcement period excess returns. T= t + 1 MAJj = I E U J.I T =tl These measures differ only to the extent that the market model regression accounts for contemporaneous market movements. Because the explanatory power of market model regressions is typically low, these measures are quantitatively very similar. The values for MAJ are usually slightly smaller in magnitude than those obtained with RAW. I define the preand postdividend windows to be 900 day periods (about 14 quarters) preceding and following the initiation of dividends. A "joint" announcement occurs when
PAGE 29
23 a dividend is declared within 2 days of an earnings announcement. I follow Venkatesh in excluding joint announcements from the analysis. I require that in each window, there are at least 3 usable earnings announcements. This requirement eliminates 7 firms from the sample. The final sample consists of 325 firms. For each firm j, I average the information content measures of all preand postdividend initiation quarters. "Matched differences" are obtained by subtracting the average postdividend initiation information content from the average predividend initiation information content for each firm. DRAW. = KX^.PRE RAW. vJ DMAJ . J MAJ^pre M7 CT J dbvu ej = BVUj.PRE BVU^ dbvu a j BVU^pre BVU; j , POST E j , POST A j , POST The distribution of matched differences is then analyzed to detect the effect of dividend initiation. Hypotheses and test statistics. There are 2 main hypotheses. The first is that dividend information is substitutable to earnings information. Under this hypothesis, one would expect to observe raw price reactions of earnings announcements to decrease in the postdividend period. Therefore the appropriate test is
PAGE 30
2 4 H 0 : DRAW = 0 versus H a : DRAW > 0 and H 0 : DMAJ = 0 versus H a : DMAJ > 0 . Further, if the total amount of information or the general level of volatility does not increase from preto postdividend initiation periods, then the standardized variance measure of information content should also decrease. H 0 : DBVU e = 0 versus H a : DBVU E > 0 . The second hypothesis is that dividends augment the transmission of public information about the firm. Under this information transmission hypothesis, the relative information content of public announcements (both earnings and dividends) to private information should increase. Therefore, H 0 : DBVU d = 0 versus H a : DBVU d < 0 . These hypotheses are tested using standard t statistics and Wilcoxon signedrank statistics. The nonparametric Wilcoxon signedrank test (for example, see Hollander and Wolfe, 1973) is particularly suited to the case of paired replicates data i.e. pairs of preand posttreatment observations, where we are concerned with a shift in location due to the application of the treatment. In this case, the treatment is the payment of dividends. Nonparametric techniques require few assumptions
PAGE 31
25 about the underlying populations from which the data are obtained. In particular, nonparametric procedures forego the traditional assumption that the underlying populations are normal . The Wilcoxon signedrank statistic W is calculated as follows. The absolute values of the matched differences are ranked in ascending order. The ranks of those matched differences that have the sign of the hypothesized direction (i.e. DRAW>0, DMA J > 0 , DBVU E >0 and DBVU d < 0) are summed. Denote this sum as T + . I use the large sample approximation to obtain W. T . _ ^ n (n+1) W = N n (n+1) ( 2n+l) 24 The statistic has an asymptotic standard normal distribution. For a two tailed test, the null hypothesis is rejected if w > z(a/2), where z(.) denotes the standard normal variable and a the significance level. Differential information availability. I also test the information transmission hypothesis by examining whether firms grouped by information availability have different information content of earnings announcements. If information transmission is a valid incentive for initiating dividends, then one would expect firms with low information availability to experience
PAGE 32
26 the greatest increase in the relative informativeness of public announcements. To test this hypothesis, I group firms using two proxies of information availability (namely, market value of equity and number of Wall Street Journal Index news items) . I test for group differences in the informativeness of these public announcements before and after dividend initiation . Previous studies by Shores ( 1990 ) , Bhushan (1989) , Lobo and Mahmoud ( 1989 ), Atiase ( 1985 ), Zeghal ( 1984 ) and Grant (1980) consistently find that the earnings announcements for firms with low information availability have greater information content. These studies group their sample firms on proxies of information availability such as firm size, analyst following, trading volume, exchange listing, financial media reportage, number of market makers and bidask spread. The most commonly used proxy of information availability is market size. For example, Shores ( 1990 ), Lobo and Mahmoud (1989), Bhushan (1989), Atiase ( 1985 ) and Zeghal ( 1984 ) all report that the earnings announcements of large market capitalization firms have lower information content. This is because large firms already have more information available about them and the additional contribution of earnings information is smaller. Large firms have greater information availability due to several reasons. Large firms may have economies of scale in producing and disseminating information about themselves. Large companies also produce more information to meet
PAGE 33
27 regulatory requirements. Large firms tend to have more shareholders. A large shareholder base provides incentive for thirdparty information providers such brokerage services, financial newsletters, etc., to produce information about the firm. I standardize the firm's market value of equity by the total market value to obtain adjusted firm size. Standardization takes into account the general market capitalization. Standardization may be important if information collection depends not on the magnitude of market capitalization but on the ranking of the firms relative to the total universe of firms. Relative rankings will change as the total market value changes over time. Since our sample covers a long period, 2 any secular change in total market value may be important. It turns out that standardization does not alter the group rankings greatly. Out of 332 firms, 32 firms change groups when ranked by market value of equity alone. The firm's adjusted firm size is calculated as the total number of shares outstanding (SHR) multiplied by the price two days before the first dividend is declared (P t 2 ) Â• This figure is then divided by the total dollar value of all nonADR securities (TOTMV) and multiplied by a scaling factor of 1 , 000 , 000 . 2 Studies which have used firm size have covered a shorter period over which changes in total market value may be minimal .
PAGE 34
28 Adjusted Firm Size SHR X P t _ 2 TOTMV X 1,000,000 Price data and number of shares outstanding are obtained from the CRSP tapes. Prior to 1973, total market dollar value is for NYSE and AMEX listed securities. From 1973, NASDAQtraded stocks are included. The number of news items appearing in the Wall Street Journal Index measures the occurrence of events that have informative value. It is also a measure of financial media coverage. A firm with a high number of news items is assumed to have higher information availability. For each firm in the sample, the number of news items appearing in the Wall Street Journal Index during the year prior to the declaration of the first dividend is collected. The firms in my sample are ranked in ascending order by adjusted firm size and the number of news items and then divided into five groups. Note that the groups based on news items do not have egual numbers of firms. This is because the number of news items is discrete causing clustering. For example, firms in the third group all have 8 news items. The characteristics of the information availability groups are shown in Tables 2 and 3. There appears to be sufficient variation with respect to firm size. For the adjusted firm size groups, the mean market value of eguity is approximately two times greater than the preceding group.
PAGE 35
29 TABLE 2 CHARACTERISTICS OF GROUPS BASED ON ADJUSTED FIRM SIZE Group N Var Mean SD Min Med Max 1 66 SIZE 3 6.563 2 . 396 2 .860 6.399 15.587 NEWS b 6.833 2 . 271 2 6 16 2 67 SIZE 13 . 144 3 . 558 6.274 13.052 24.820 NEWS 7.667 2.420 4 7 18 3 67 SIZE 23.303 6.088 10.621 22.727 38.900 NEWS 9 . 552 4 .328 5 9 33 4 66 SIZE 50.667 17 .750 26.837 45.794 127.630 NEWS 10.797 5.262 5 9 28 5 66 SIZE 308 .83 584 .906 50.474 128.265 3631.92 NEWS 13 . 078 7 . 895 5 10 50 a SIZE is the market value of equity (in millions of dollars) two days before the initial dividend is announced. b NEWS is the number of Wall Street Journal Index news items in the year prior to the announcement of the initial dividend . Table 4 shows the distribution of firms grouped by firm size and news items. While the information proxies do not yield identical groupings, large firms tend to have more news items reported about them. This is consistent with the findings of Thompson, Olsen and Dietrich (1987) who examine the characteristics of Wall Street Journal Index news items.
PAGE 36
30 TABLE 3 CHARACTERISTICS OF GROUPS BASED ON NUMBER OF NEWS ITEMS Group N Var Mean SD Min Med Max 1 48 SIZE 3 23.727 26.846 2 .860 14.018 138.52 NEWS b 4 . 826 0.5698 2 5 5 2 92 SIZE 43 . 899 91.479 2 . 880 13 .312 512.290 NEWS 6.424 0.497 6 6 7 3 46 SIZE 25.242 22 . 562 4 . 020 15.970 82.357 NEWS 8 0 8 8 8 4 79 SIZE 46.317 47 . 877 3 .294 27.547 297.024 NEWS 9 . 397 0 .952 9 10 12 5 67 SIZE 238 . 728 586.243 10.621 55.544 3631.920 NEWS 17 . 657 6. 383 13 16 50 SIZE is the market value of equity (in millions of dollars) two days before the initial dividend is announced. NEWS is the number of Wall Street Journal Index news items in the year prior to the announcement of the initial dividend . TABLE 4 DISTRIBUTION OF FIRMS IN GROUPS BASED ON ADJUSTED FIRM SIZE AND NUMBER OF NEWS ITEMS NEWS ITEMS i 2 3 4 5 Total ADJUSTED MARKET VALUE 1 17 28 10 9 2 66 2 11 27 13 13 3 67 3 9 13 7 26 12 67 4 8 11 11 17 19 66 5 3 13 5 14 31 66 Total 48 92 46 79 67 332
PAGE 37
31 To test for differential information contents across the five groups the data sample is characterized as a oneway layout design experiment. The data consists of N = E k n k (for k=l,...,5) observations with n k observations in the kth information availability group. Note that the number of observations in each group need not be equal to each other. This oneway layout is illustrated in Figure 3.1. D(j,k) 's are the matched differences from group k with j individual observations . Information Availability Groups 1 2 3 4 5 D ( 1 , 1 ) D ( 2 , 1 ) D(ni,l) D ( 1 , 2 ) D ( 2 , 2 ) D ( n 2 , 2 ) D ( 1 , 3 ) D ( 2 , 3 ) D (n 3 , 3 ) D(l,4) D ( 2 , 4 ) D(n 4/ 4) D ( 1 , 5 ) D ( 2 , 5) D(n 5 , 4) Figure 1 Sample Data Design
PAGE 38
32 The basic model is D ( j , k) = n + r k + e jk , j=l , . . . , n k =1 , . . . , 5 where /i is the unknown overall mean, r k is the unknown information effect in group k. I assume the errors e are mutually independent and have the same continuous distribution . The hypothesis to be tested is H 0 : t 1 = t z =t 3 = t 1i = r 5 versus H a : t 1 > r 2 > r 3 > r A > r 5 where at least one of the inequalities is strict. The firms in the lower information availability groups have larger information effects. I use the JonckheereTerpstra statistic (see Hollander and Wolfe, 1973) to test the hypothesis. It is based on the number of group j observations which are less than each of the group i observations, where i > j. For 5 groups, there are 10 of these MannWhitney counts . n i n j u ij = Â£Â£0[ D (s,i) D(t, j) ] S = 1 t = l
PAGE 39
33 where
PAGE 40
34 comprehensive test which does not suffer from the usual problems associated with multiple comparisons such as the Bonferroni or Tukey methods.
PAGE 41
CHAPTER 4 TESTS ON QUARTERLY EARNINGS ANNOUNCEMENTS The Overall Sample The overall sample consists of 325 firms. Each firm must have at least 3 usable quarterly earnings (joint announcements are omitted) in both preor postdividend initiation periods. For each firm, the information content of quarterly earnings announcements in each period are averaged. Summary statistics for the crosssectional distribution of average information content measures for preand postdividend initiation periods are shown in Table 5. The information content measures are based on raw returns (RAW) , market adjusted excess returns (MAJ) , the ratio of earnings announcement returns variance to nonannouncement returns variance (BVU E ) and the ratio of earnings and dividend announcement returns variance to nonannouncement period returns variance (BVU D ) . In the predividend initiation period, only earnings announcements are relevant and the ratio is with respect to earnings announcement returns variance (BVU) . The information content measures RAW and MAJ have larger values in the predividend initiation period. The predividend initiation values of the minimum, the first quartile, median, third quartile and maximum always exceed those from the post35
PAGE 42
36 TABLE 5 DISTRIBUTION OF THE AVERAGE INFORMATION CONTENT OF QUARTERLY EARNINGS ANNOUNCEMENTS BEFORE AND AFTER DIVIDEND INITIATION. Sample Size = 325 firms PreDividend Period PostDividend Period RAW MAJ BVU RAW MAJ BVU e BVU d Mean 5.79 5.67 2 . 007 4 . 52 4 . 39 1.908 1.862 SD 2 . 13 2 . 15 1.139 1.62 1.62 1.178 1.444 Minimum 1.14 1 . 52 0.459 0.02 0.85 0.230 0.528 Ql a 4 .29 4 . 13 1.309 3 . 42 3 . 34 1.173 1.205 Median 5.61 5.43 1.684 4 .36 4 . 15 1.609 1.573 Q3 a 6.98 6.89 2 .263 5.35 5.25 2 .367 2 . 036 Maximum 13.31 12 . 82 7 . 663 11 . 65 11.43 9 . 589 15.081 Ql=first quartile; QS^third quartile. dividend initiation period. For predividend initiation quarterly earnings announcements, the mean raw price revaluation is 5.79%. For postdividend initiation quarterly earnings announcements, the mean is 4.52%. Similarly, the predividend initiation mean market adjusted price revaluation is 5.67% with the postdividend initiation mean being 4.39%. These figures are similar to those reported by Venkatesh. 1 For raw returns, Venkatesh obtains a predividend initiation 1 The differences between my study and the Venkatesh study are as follows. Venkatesh examines a smaller sample of 75 NYSE or AMEX firms which paid quarterly dividends. His estimation period precedes the quarterly earnings announcement date. He uses a lagged ScholesWilliams technique to obtain his market model regressions.
PAGE 43
37 mean of 5.84%. In the postdividend initiation period, he reports a mean of 4.91% (4.01%) for earnings announcements following (preceding) dividends announcements. Results for market adjusted excess returns are similarly comparable. The standardized variance information content measures show a lesser decrease. The postdividend initiation mean values for BVU e and BVU D are 1.908 and 1.862 compared with the predividend initiation mean value of 2.007. For the postdividend initiation period, the average dividend announcement variance is less than the average earnings announcement variance. Thus the combined average variance of earnings and dividend announcement (BVU D ) is less than for earnings announcements alone (BVU E ) . Whether these differences are statistically significant is examined by comparing matched differences. The distribution for matched differences is tabulated in Table 6. I use t statistics and the nonparametric Wilcoxon signedrank statistics to test whether the average information content is greater in the predividend initiation period. The evidence for RAW and MAJ is unambiguous. Both t and Wilcoxon signedrank tests support the hypothesis that the average postdividend initiation quarterly earnings announcement is less informative. About 70% of the sample experiences a decrease in earnings price informativeness. The average reduction is about 1.2%. This suggests that earnings and dividend information
PAGE 44
38 might be substitutable. My sample yields results entirely consistent with those reported by Venkatesh. The results from the standardized variance measures of information content measures are in less accordance with the raw price measures. Just over half the sample experience a decrease following dividend initiation. DBVU E has an insignificant decline. Thus, there appears to be a reduction in the informativeness of earnings announcements relative to the average informativeness in nonannouncement days. More importantly, however, DBVU d has a significant change but it is in a direction opposite to that hypothesized. The variance of excess returns for public announcements (both earnings and dividends announcements) relative to the nonannouncement periods is decreased once dividends have been initiated. Contradictory to the information transmission hypothesis, the proportion of information publicly released relative to that privately acquired decreases after dividend initiation. I also test the normality of the distribution of the matched differences of information content measures. The lower the value of the ShapiroWilk D statistic the greater the probability of rejecting the null hypothesis that the distribution is normal. The null is rejected at the 5% significance level for all distributions. The nonnormality of the distributions of matched differences justifies the use of nonparametric test statistics.
PAGE 45
39 TABLE 6 CROSSSECTIONAL DISTRIBUTION OF MATCHED DIFFERENCES Sample Size = 325 firms DRAW 3 DMAJ a DBVU E a DBVU D a Mean 1.27 1.29 0.099 0.145 Std. Dev. 2 . 53 2 . 51 1.546 1.729 Minimum 6.99 6.51 7.098 13.632 Ql b 0. 15 0.36 0.785 0.503 Median 1.37 1.23 0.048 0.114 Q3 b 2 .76 2 .95 0.873 0.907 Maximum 9 . 17 9 . 80 6 .237 6.518 DÂ° 0.9654 0 .9695 0.9364 0.8367 Prob < D 0.0368 0 . 0421 0.0000 0.0000 No. Positive 241 226 170 177 % Positive 77 . 2 69 . 5 52 . 3 54.5 rpd, e 9 . 05 9 .24 1.15 1.52 Prob > T 0 . 0001 0 . 0001 0.1250 0.0654 W d Â’ f 8 .42 8.46 1.17 2.82 Prob > W 0 . 0001 0 . 0001 0.1210 0.0024 a Difference of average preand postdividend information content of quarterly earnings, for each firm. For example, for the raw price information content measure, DRAWj = HSW j>PRE RAWj ( P0ST b Ql=first quartile; Q3=third quartile. c ShapiroWilk test of H 0 : Distribution is normal. d Test of H 0 : DRAW=0 versus H a : DRAW>0, H 0 : DM A J = 0 versus H a : DM A J > 0 , H 0 : dbvu e =o versus H a : DBVU e >0 / H 0 : DBVU d =0 versus H a : DBVU d >0. tstatistic . Wilcoxon signedrank statistic.
PAGE 46
40 The Sample Grouped by Information Availability In this section, I analyze the matched differences across information availability groups. Under the information transmission hypothesis, one would expect to see the informativeness of earnings announcements to decrease as one goes from low information availability to high information availability firms. Table 7 shows the distribution of DRAW grouped by adjusted firm size. For all groups, the differences are significantly positive. The Wilcoxon signedrank tests and ttests are significant at levels well below the 5% level. The KruskalWallis H statistic indicates that the group means for DRAW do differ across groups. The JonckheereTerpstra J statistic indicates that the group means are decreasingly smaller as market value increases. This relationship is significant at the 5% level. Postdividend initiation quarterly earnings average price revaluations are less than in the predividend initiation period. An examination of the group means shows that the decrease tends to be less for high market value firms. Low market value firms experience a drop of 1.5% in the average revaluation occurring at earnings announcements. High market value firms experience only a 0.6% decrease. However, this relationship is not strictly monotonic. Firms in the lowest market value group have smaller decreases than those in the next quintile. Table 8 shows the distribution of DMAJ grouped by adjusted firm size. We obtain the same general results with DMAJ. The
PAGE 47
41 group means decrease from group 2 to group 5. The mean for group 1 is less than that of group 2. The difference between group means is less strong than for DRAW. Tables 9 and 10 show the distributions for the standardized variance measures. In general, there does not appear to be any significant change in means after dividend initiation. Pvalues for the ttests and Wilcoxon signed rank tests are generally insignificant. The only exception to this is group 2 for DBVU d which shows a decrease. The KruskalWallis statistics fail to reject the null hypothesis of equal group means. The JonckheereTerpstra statistics do not provide any evidence that changes in relative informativeness increases as one goes from smaller to larger firms. Tables 11 to 14 show the distributions of matched differences for the groups based on number of Wall Street Journal Index news items. In general, the results obtained from using this information availability proxy are similar to those obtained from adjusted firm size. Within groups, the t and W statistics show that the mean matched differences for DRAW and DMAJ are all greater than zero at conventional significance levels. The KruskalWallis statistic shows that the group means are not all equal. The JonckheereTerpstra statistic shows that groups with more news items have smaller mean decreases.
PAGE 48
42 TABLE 7 DISTRIBUTION OF DRAW FOR PORTFOLIOS BASED ON FIRM SIZE FIRM SIZE PORTFOLIOS ( 1=SMALL, 5=LARGE) 1 2 3 4 5 N 65 64 67 66 63 Minimum 6.9927 4 . 7255 5.6859 5.3322 3 . 6112 Ql a 0.0597 0. 1516 0.1686 0.2734 0.8354 Median 1.7415 1.7134 1.7880 1.0386 0.4658 Q3 a 3 . 2472 3 . 5687 2 . 8520 2 .2331 1.7973 Maximum 9 . 1723 8 . 4802 7 .7009 6.5617 6.2081 Mean 1.4319 1.6038 1.5852 1.0865 0.6213 SD 3 . 1280 2 . 5617 2 .3125 2 . 3807 2 . 0822 T 3 . 69 5.01 5.61 3 .71 2.37 Pr >  T 0.0005 0.0001 0.0001 0.0001 0.0210 w 3 . 76 4 . 29 4 . 87 3 .39 2 . 18 Pr >  W 0 . 0001 0 . 0001 0.0001 0.0006 0.0292 H b 9 . 6564 * J c 2 . 6670* a Ql=first quartile; Q3=third quartile. b Test of H 0 : r 1 =r 2 : =. . . =r 5 versus H a : at least one mean (r) is not the same. H is the WallisKruskal statistic. Cannot reject the null at the 5% level if H<9.49. 0 Test of H 0 : r 1 =r 2 =...=r 5 versus H a : r 1 >r 2 >...>r 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null at the 5% level if J<1. 645. * Significant at the 5% level using a onetailed test.
PAGE 49
43 TABLE 8 DISTRIBUTION OF DMAJ FOR PORTFOLIOS BASED ON FIRM SIZE FIRM SIZE PORTFOLIOS ( 1=SMALL, 5=LARGE) 1 2 3 4 5 N 65 64 67 66 63 Minimum 6.1745 6.5091 3 . 5399 4 . 8611 2.8960 Ql a 0.1286 0.5404 0.1559 0.5313 0.8236 Median 1.3908 1.8688 1.4322 0.7621 0.5403 Q3 b 2 . 5956 3 . 8226 3 . 0869 2.9983 1.9154 Maximum 9 .8017 8 . 5145 7 .4536 6.4123 5.4624 Mean 1.2713 1.6674 1.6423 1.1678 0.6591 SD 2 .9604 2 .7884 2 .2282 2.4313 1.9437 T 3 .46 4 . 78 6.03 3 .90 2.69 Pr >  T 0.0010 0 . 0001 0.0001 0.0002 0.0091 W 3 . 52 4 . 19 5.08 3 .35 2.36 Pr >  W  0.0004 0.0001 0 . 0001 0.0008 0.0182 H c 7 . 2795 J d 2 . 0676* a Ql=first quartile; Q3=third quartile. b Test of H 0 : r 1 =T 2 =...=r 5 versus H a : at least one mean (r) is not the same. H is the WallisKruskal statistic. Cannot reject null at the 5% level if H<9.49. c Test of H 0 : r 1 =r 2 =. . .=r 5 versus H a : t 1 >t 2 >. . . >r 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null at the 5% level if J<1 . 64 5 . * Significant at the 5% level using a onetailed test.
PAGE 50
44 TABLE 9 DISTRIBUTION OF DBVU E FOR PORTFOLIOS BASED ON FIRM SIZE FIRM SIZE PORTFOLIOS ( 1=SMALL, 5=LARGE) 1 2 3 4 5 N 65 64 67 66 63 Minimum 7 . 0975 2 . 8549 3 . 3461 2 . 3720 2.8119 Ql a 1.0527 0.7551 0.9388 0.6734 0.5887 Median 0.1229 0.2910 0.0761 0.1267 0.0641 Q3 a 0.9738 1. 0409 0.9714 0.6136 0.8085 Maximum 5.4849 4 .8623 6.2366 3 . 5934 2.4277 Mean 0 . 2467 0 .3007 0 .2396 0.1151 0.0835 SD 2 . 2479 1.4633 1.6395 1.0479 0.9478 T 0.88 1.64 1.20 0.89 0.70 Pr > T 0.3797 0.1051 0.2359 0.3756 0.4868 w 0 . 49 1.26 0 . 67 0.61 0.55 Pr > W 0.6242 0 .2076 0.5028 0.5418 0.5824 H b 1.6994 J c 0.3855 Ql=first quartile; Q3=third quartile. Test of H 0 : r 1 =r 2 =...=r 5 versus H a : at least one mean (r) is not the same. H is the WallisKruskal statistic. Cannot reject the null at the 5% level if H<9.49. Test of H 0 : t 1 =t 2 =. . .=r 5 versus H a : r 1 >r 2 >. . ,>t 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null at the 5% level if Jcl.645.
PAGE 51
45 TABLE 10 DISTRIBUTION OF DBVU d FOR PORTFOLIOS BASED ON FIRM SIZE FIRM SIZE PORTFOLIOS ( 1=SMALL, 5=LARGE) 1 2 3 4 5 N 65 64 67 66 63 Minimum 7 .2114 2 .7647 13 . 6318 10.3739 1.0485 Ql a 0.7629 0.3808 0.5740 0.5135 0.3751 Median 0.0467 0.2756 0.0064 0.2192 0.0104 Q3 a 0.8941 1. 1312 0.9512 0.8536 0.7990 Maximum 5.3639 4 . 8751 6.5185 3 .4881 3.0709 Mean 0 . 0263 0.4357 0.0094 0.1248 0.1936 SD 1.9576 1 . 3613 2 . 3879 1.6614 0.8251 T "0. 11 2 . 56 0.03 0.61 1.86 Pr > T 0 .9140 0 . 0129 0.9745 0.5438 0.0673 w 0.09 2 . 27 0 . 92 1.74 1.14 Pr > w 0.9282 0.0232 0.3576 0.0818 0.2542 H b 2 . 5566 J c 0 . 1831 a Ql=first quartile; Q3=third quartile. b Test of H 0 : r 1 =r 2 =...=r 5 versus H a : at least one mean (r) is not the same. H is the WallisKruskal statistic. Cannot reject the null at the 5% level if H<9.49. c Test of H 0 : t 1 =t 2 =. . .=r 5 versus H a : r 1 >r 2 >. . .>r 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null at the 5% level if J<1.645.
PAGE 52
46 The results for the distributions of DBVU E and DBVU d in Tables 13 and 14, respectively, provide no discernable pattern. With the exception of group 3, none of the group means are significantly different from zero. The JonckheereTerpstra J statistic indicates that there is no monotonic decrease in the relative information content of earnings announcements as one goes from low to high news item. However, the WallisKruskal H statistic indicates that the relative information content between groups are not all equal. In general, the results from these stratified tests are not supportive of each other. Based on the returns measures, DMAJ and DRAW, the decrease in group means increases as one goes from high market value (news item) firms to low market value (news items) firms. This is consistent with the information transmission hypothesis. Based on returns measures alone, the informativeness of quarterly earnings reports for low information availability firms experience the greatest decrease after dividend initiation. However, once we adjust for the contemporaneous level of returns volatility, this information effect no longer persists. The mean standardized variance measures, DBVU E and DBVU d , are not statistically different following dividend initiation. Neither do they exhibit any significant change in going from low to high information availability groups. There appears to be no monotonic difference in the mean change in total information content (DBVU d ) for firms grouped by information availability.
PAGE 53
47 TABLE 11 DISTRIBUTION OF DRAW FOR PORTFOLIOS BASED ON NO. OF NEWS ITEMS FIRM SIZE PORTFOLIOS ( 1=LEAST , 5=MOST) 1 2 3 4 5 N 45 92 40 80 68 Minimum 5.6416 5.0269 6.9927 5.6859 4.2648 Ql a 0.7078 0.1753 0.4185 0.0847 0.9469 Median 1.7415 1.5886 1.7555 1.3645 0.5221 Q3 a 3 . 3723 3.2141 2 . 7990 2 . 5838 1.6834 Maximum 9 . 1723 7 .7009 8 .9110 5.9192 6.5617 Mean 1 . 5995 1.5428 1.4653 1.2798 0.5568 SD 3 . 0830 2 .4669 2 . 8481 2 . 1635 2 .3436 T 3 .48 6 . 00 3 .25 5.29 1.96 Pr > T 0 . 0011 0 . 0001 0.0024 0.0001 0.0543 w 3 .30 5. 32 3 .20 4 .72 1.80 Pr >  W 0.0010 0.0001 0 . 0014 0.0001 0.0588 H b 10 . 2878* JÂ° 2 . 6193* a Ql=first quartile; Q3=third quartile. b Test of H 0 : t 1 =t 2 =. . .=r 5 versus H a : at least one mean (r) is not the same. H is the WallisKruskal statistic. Cannot reject the null at the 5% level if H<9.49. c Test of H 0 : r 1 =r 2 =. . .=r 5 versus H a : r 1 >r 2 >. . . >r 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null at the 5% level if Jcl.645. * Significant at the 5% level using a onetailed test.
PAGE 54
48 TABLE 12 DISTRIBUTION OF DMAJ FOR PORTFOLIOS BASED ON NO. OF NEWS ITEMS FIRM SIZE PORTFOLIOS ( 1=LEAST, 5=MOST) 1 2 3 4 5 N 45 92 40 80 68 Minimum 6.5091 5.2964 6. 1745 3 . 5399 4 . 0065 Ql a 0.3792 0.0396 0.0547 0.1967 0.8519 Median 1.8883 1.3843 2 . 0159 1.2342 0.3210 Q3 a 3 .4979 3 . 1635 3 .3842 2 .4626 1.7938 Maximum 9 .8017 7 .4536 7 . 1067 6.1621 5.9768 Mean 1 . 8230 1.4624 1.6080 1.1782 0.6300 SD 3 .3569 2 .4725 2 . 7286 2 . 0246 2 . 1987 T 3 . 64 5.67 3 .73 5 .20 2 .36 Pr > T 0.0007 0.0001 0 . 0006 0.0001 0.0210 W 3.41 5.67 3 . 73 5.20 2.36 Pr > j W 0 . 0006 0 . 0001 0.0003 0.0001 0.0182 H b 9 . 4949* J c 2 . 7262* a Ql=first quartile; Q3=third quartile. b Test of H 0 : t 1 =t 2 =. . .=r 5 versus H a : at least one mean (r) is not the same. H is the WallisKruskal statistic. Cannot reject the null at the 5% level if H<9.49. c Test of H 0 : t 1 =t 2 =. . .=r 5 versus H a : r 1 >r 2 >. . .>t 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null if J<1.645. * Siqnificant at 5% level usinq a onetailed test.
PAGE 55
49 TABLE 13 DISTRIBUTION OF DBVU E FOR PORTFOLIOS BASED ON NO. OF NEWS ITEMS FIRM SIZE PORTFOLIOS ( 1=LEAST, 5=M0ST) 1 2 3 4 5 N 45 92 40 80 68 Minimum 7 . 0975 3 .9293 1.5222 3 . 3461 7.0369 Ql a 0.0897 1.0469 0.2689 0.4310 0.8351 Median 0.0336 0. 0293 0.7221 0.0897 0.1813 Q3 a 1.0041 1.0610 1.3258 0.6437 0.4573 Maximum 4 .4923 6.2366 4 .8623 5.1343 2.8748 Mean 0 . 0024 0.1548 0.6755 0.0632 0.2103 SD 2 . 1131 1.7655 1. 1946 1.1790 1.2715 T 0.01 0 . 84 3 . 58 0.48 1.36 Pr >  T 0.9940 0 .4024 0.0009 0.6328 0.1771 w 0.45 0 . 38 3 . 28 0.70 1.46 Pr > w 0 . 6528 0 .7040 0 . 0010 0.4840 0.1442 H b 11.8148* J c 1.4030 a Ql=first quartile; Q3=third quartile. b Test of H 0 : r 1 =r 2 =...=r 5 versus H a : at least one mean (r) is not the same. H is the Wall isKruskal statistic. Cannot reject the null if H<9.49. c Test of H 0 : t 1 =t 2 =. . .=r 5 versus H a : r 1 >r 2 >. . .>r 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null at the 5% level if J>1 .645. * Siqnificant at the 5% level usinq a onetailed test.
PAGE 56
50 TABLE 14 DISTRIBUTION OF DBVU d FOR PORTFOLIOS BASED ON NO. OF NEWS ITEMS FIRM SIZE PORTFOLIOS (1=LEAST, 5=MOST) 1 2 3 4 5 N 45 92 40 80 68 Minimum 7 .2114 13 . 6318 1.3339 5.7861 3.1993 Ql a 0.7629 0.5921 0.1358 0.3647 0.5277 Median 0.1868 0.0714 0.7632 0.1048 0.0717 Q3 a 1.0446 1. 1932 1.2732 0.7652 0.4795 Maximum 4 . 0260 6.5185 4 . 8751 3 .4881 3 . 1993 Mean 0.0482 0. 1039 0 . 7711 0.0608 0.0028 SD 1 .9836 2 .4677 1 . 1524 1.0976 1.0661 T 0.16 0.40 3 . 58 0.48 1.36 Pr > T 0.8713 0.6873 0 . 0001 0.6223 0.9828 w 0.80 1.04 3 . 78 1.39 CO Â• 0 1 Pr > w 0 .4238 0 . 2984 0.0002 0.1646 0.6312 H b 11.8098* J c 1.2855 a Ql=first quartile; Q3=third quartile. b Test of H 0 : r 1 =r 2 =...=r 5 versus H a : at least one mean (r) is not the same. H is the WallisKruskal statistic. Cannot reject the null at the 5% level if H<9.49. c Test of H 0 : r 1 =r 2 =. . .=t 5 versus H a : . .>r 5 with at least one strict inequality. J is the JonckheereTerpstra statistic. Cannot reject the null at the 5% level if J<1. 645. * Significant at the 5% level using a onetailed test.
PAGE 57
51 This contradicts the information transmission hypothesis as one would expect the increase in the proportion of publicly released information to be greatest for low information availability firms. This inconsistency between the standardized variance and returns information content measures calls for further investigation of returns volatility around dividend initiation. It is possible that a contemporaneous decrease in returns volatility is producing the results obtained. This is investigated in the next chapter. The Market Reaction to Earnings Changes Before going on to examine the effect of the contemporaneous decline in returns variance, I analyze the price reaction to reported quarterly earnings change. This section provides further proof that what has been previously interpreted as a decrease in earnings informativeness caused by dividend initiation may be due to a contemporaneous decrease in returns volatility. The relationship between market price reaction and reported earnings is modeled as CER jq = 0 O + ( 3 , D + 03 AE jq + ( 3 , DAE jq (1) The 3day cumulated excess returns for the day prior to, the day of and the day following the earnings announcement date for quarter q and firm j is denoted by CER jq . The dummy
PAGE 58
52 variable D takes on the value 1 if the quarterly earnings announcement comes from the postdividend initiation period and is zero otherwise. The earnings surprise or standardized earnings is the quarterly change in earnings per share deflated by the firm's share price two days before the earnings announcement. EPS jq EPS jq i P j.t2 The change in quarterly earnings is appropriate based on a randomwalk expectations model. Deflating by the share price captures any expectations about the forthcoming earnings formed through information gathered between earnings announcements . The regression approach allows us to incorporate information contained in the earnings report (namely, the change in earnings per share) which we could not do in the above analysis based on groups. This controls for the potential bias that could occur if earnings changes are smaller or less variable in the postdividend period. This is similar to the approach used by Healy and Palepu (1988). The average price response to earnings announcement for the whole sample is (3 0 . Any change in the average price response in the postdividend initiation period is captured by (3 lr and the average market price reaction to a given level of earnings surprise is /3 2 . It is expected to be positive indicating that
PAGE 59
53 the market responds positively (negatively) to earnings improvements (declines) . The average postdividend initiation adjustment to / 3 Z is /? 3 . If dividends enable investors to revise their expectations of earnings, their forecast errors in the postdividend period will be reduced. However, measures of unexpected earnings which only incorporate quarterly earnings changes will not reflect this additional information obtained from dividends. Standardized quarterly earnings changes will be noisier estimates of unexpected earnings. Therefore, the market price response to our measure of earnings surprise should be reduced. Coefficient /3 2 is expected to be negative. I estimate equation (1) using ordinary least squares. For each firm in my sample, 8 quarters preceding and following the initial dividend are collected. A quarter is excluded if the announcement is joint, i.e. earnings and dividend announcements within 2 days of each other. A quarter is also excluded if there was no data on net income, share price or shares outstanding to calculate the standardized earnings. I require that a firm had 4 usable preand postdividend initiation quarters. This requirement excludes 58 firms leaving 274 firms for the regression analysis. For comparison, I also ran the regression using the variance adjusted measure BVU E .
PAGE 60
54 BVU E, jq = Po + + /3
PAGE 61
55 TABLE 15 MARKET PRICE RESPONSE TO STANDARDIZED EARNINGS Dep. Var. CER a BVU E b Po 5.4832 (47 . 538**) c 1.9508 (24.385**) Pi 0.8237 (5.025**) 0.0698 (0.613) P 2 0.1925 (2.174*) 0.0131 (0.214) P 3 0.1828 (0.480) 0.1232 (0.465) F 10.376** 0.224 Adj . R 2 0.0072 0.0006 No. of Firms 274 No. of Obs. 3901 a Model estimated is CER = P 0 + P X D + P 3 AE + P 4 DAE, Model estimated is BVU e = p 0 + P,D + P 3 AE q  + P 4 D  AE q CER is the cumulated 3 day excess returns; one day prior to, the day of and the day following the quarterly earnings announcement. D is a dummy variable which takes on the value 1 if the quarterly earnings occurs after dividend initiation and is zero otherwise. The standardized earnings is the change in earnings per share from the prior quarter divided by the stock price two trading days before the earnings announcement. EPS q EPS ql P t ; 0 tstatistics in parentheses. * Significant at the 1% level using a twotailed test. ** Significant at the 5% level using a twotailed test.
PAGE 62
56 contradicts the notion that dividends convey information that leads investors to better forecast earnings. The coefficient ;0 3 is, however, not statistically significant. My regression results yield results diametrically opposite to those obtained by Healy and Palepu (1988). They obtain a significantly negative /3 3 . I suspect that this difference arises out of a difference in regression technique and model specification. Healy and Palepu apparently use a seemingly unrelated regressions technique for estimation. This allows their estimates of /3 0 and f3 z to be firm specific. In obtaining these firm specific coefficients, their model specification excludes the f3 1 coefficient. Any decrease in returns volatility is thus forced onto (3 3 which biases it to be negative . The results for the regression using the standardized variance measure of informativeness is consistent with prior results. The relative informativeness of quarterly earnings does not change after dividend information. The market response at quarterly earnings announcements relative to nonannouncement periods is not significantly related to the earnings reported, nor does this responsiveness change after dividend initiation.
PAGE 63
CHAPTER 5 RETURNS VOLATILITY AROUND DIVIDEND INITIATION In the previous chapter, I examined the information content of quarterly earnings announcements before and after dividend initiation. The information content measures used are directly related to the volatility of returns. A possible explanation for the observed decrease in the price reactions associated with postdividend initiation earnings announcements is that returns volatility has decreased. Declines in returns volatility may be caused by changes in the process by which information is impounded in price. For example, dividend initiation may induce greater consensus among traders. This decreases noise trading and bidask spreads and consequently returns variance. However, volatility also may decrease due to causes that are independent of any information motivated reason. Dividend initiating firms experience increases in stock price which lead to proportionally smaller bidask spreads and reduced leverage. Finally, the observed volatility decreases in my sample could be due to unrelated market wide changes in interest rates. In this chapter, I perform several tests to distinguish the causes of the observed volatility decrease. I conclude that 57
PAGE 64
58 information is not the only plausible cause for volatility to decline . Variance Changes in the Overall Sample Table 16 shows the distribution of percentage changes in variances calculated for different sampling periods around dividend initiation. The sampling periods cover 900, 500, 250 and 100 trading days immediately before and after the declaration of the initial dividend. Fewer firms are available for the longer sample periods because of the longer sequence of returns required. For the 100, 250, 500 and 900 day sampling periods, I require at least 90, 240, 480 and 850 returns, respectively. Variances are calculated as the sum of the squared continuously compounded returns since Var[R] approximately equals E[R 2 ] for short measurement intervals. The percentage change in variance is Percentage variance change = 100 x [1 VAR POST VAR PRE ] A variance decrease after dividend initiation gives rise to a positive percentage change. The mean percentage change is positive for all (except the 100 day) sampling periods. This indicates a decline in measured returns volatility. The median is a better statistic of location since the mean percentage variance change is
PAGE 65
59 TABLE 16 DISTRIBUTION OF PERCENTAGE VARIANCE CHANGES Period 3 900 500 250 100 1 Sample b 225 304 327 331 Min 260. 12 310.63 382 .41 1196.15 Q1 13 . 05 12 . 14 20.03 25.86 Median 39.44 33 . 65 20.33 9.68 Q3 55.20 54 . 58 43.73 36.83 Max 80.59 86.77 80.31 81.33 Mean 26.94 13 . 50 3 . 52 10.06 SD 46.29 59 .78 61.69 92 . 38 No. Pos . c 182 212 216 186 % Pos . 80 . 9 69 . 7 66.1 56.2 D d 0 . 8884 0.8354 0.8152 0.6024 Pr>  D  0.0 0.0 0.0 0.0 rpe 8 . 47 3 . 94 1.03 1.98  Pr>  T  0.0001 0.0001 0 .3035 0.0484 w e 8 . 59 5 . 79 4 . 09 1.14 Pr>  W  0.0001 0. 0001 0 . 0001 0.2559 Percentage variance changes are calculated as Percentage variance change = 100 x [1 VAR POST VAR PRE ] Variances are calculated using daily continuously compounded stock returns from sample periods of 900, 500, 250 and 100 trading days before and after the declaration day of the initial dividend. b For 900, 500, 250, and 100 day sampling periods, a firm is included if it has 850, 480, 240, and 90 valid returns, respectively . 0 Number of firms in sample with positive percentage variance changes, i.e. variance decreases after dividend initiation. d ShapiroWilk test of normality. Â® Test of H 0 : Percentage variance change = 0.
PAGE 66
60 downwardly biased. 1 The medians are positive for all sampling periods and always greater than the means. The longer the sampling period, the more apparent is the volatility decline. The median percentage change is 39% for the 900 day sample period, this declines to 10% for the 100 day sample period. The proportion of firms experiencing volatility declines also increases with longer sampling periods: 56% for the 100 day sample period increasing to 81% for the 900 day sample period. The Wilcoxon signedrank tests of the null hypothesis that the percentage variance change is zero are rejected for all (except the 100 day) sample periods. When examining changes in stock return variance, it is important to consider changes in market volatility through time. This is particularly so for my sample which has dividend initiation dates dispersed over 17 years and where there is also considerable clustering in the middle seventies. It is possible that any observed change in market variance around the time of dividend initiation is explained by contemporaneous changes in market volatility. To address this possibility, I divide firm variances by an estimate of market volatility. This estimate is obtained from the contemporaneous returns on an equally weighted market index of all firms on the relevant data tape. This adjustment is in the spirit of a heteroskedast icity adjustment, where the i From Jensen's inequality, E [ V PRE /V P0ST ] > E [V pre ]/E [Vpo ST ] .
PAGE 67
61 stock's return variance is assumed proportional to that of the market. Table 17 shows the distribution of percentage variance changes where the firm variances have been adjusted by a contemporaneous estimate of the market variance. It appears that changes in market volatility explain part of the variance decline. Median percentage changes are smaller after standardization but are still positive. The Wilcoxon signedrank tests can only reject equal variances for the 900 day sample period. This provides some evidence that clustering of the data sample may explain the observed variance decreases. Variance Ratio Tests Consider returns calculated over intervals of k days where k>l . If daily returns follow a random walk, the variance of these kday returns should be k times the variance of daily returns. French and Roll (1986) model a stock's return as having 3 components: a rational information (intrinsic value) component, a noise or mispricing component and a bid/ask error component. If noise and bid/ask error components are temporary components, over time, they will be corrected and induce negative autocorrelations. This negative autocorrelation causes the kday variance to be less than k times the daily variance. A variance ratio (VR) is the ratio of kday returns variance to daily returns variance (French and Roll (1986)
PAGE 68
62 TABLE 17 DISTRIBUTION OF MARKET ADJUSTED PERCENTAGE VARIANCE CHANGES Period 3 900 500 250 100 Sample 13 225 304 327 331 Min 198 .86 272 . 93 406.08 803.86 Q1 0.36 31.79 31.85 43.60 Median 28.73 13 . 29 6.28 5.22 Q3 48 . 10 45.73 39.83 30.41 Max 90.54 80.75 79 . 44 79.81 Mean 17 .73 2 . 74 6.50 10.06 SD 46.33 66 . 64 66.25 92.15 No. Pos . c 169 181 184 176 % Pos. 75. 1 59 . 5 56.3 53.2 D d 0.8452 0 .8544 0 . 8367 0.6862 Pr> D 0 . 0 0 . 0 0. 0 0.0 5.74 0 . 72 1.77 3.85 Pr> T 0 . 0001 0 .4736 0 . 0769 0.0001 w e 7.21 1 . 84 0.75 1.70 Pr> W 0.0001 0 . 0666 0.4556 0.1359 Percentage variance changes are calculated as Percentage variance change = 100 x [1 post^ VAR pre Variances are calculated using daily continuously compounded stock returns from sample periods of 900, 500, 250 and 100 trading days before and after the declaration day of the initial dividend. Firm variances are divided by the returns variance of a valueweighted market index. b For 900, 500, 250, and 100 day sampling periods, a firm is included if it has 850, 480, 240, and 90 valid returns, respectively . c Number of firms in sample with positive percentage variance changes, i.e. variance decreases after dividend initiation. d ShapiroWilk test of normality. Â® Test of H 0 : Percentage variance change = 0.
PAGE 69
63 call these actualtoimplied variance ratios) . ^ Var (R k ) VR k = Â— Var (R 1 ) where R k is the kday return. Under the null hypothesis that returns follow a random walk, the VR should be equal to one. If there are transitory components, the VR should be smaller than one. 2 The VR therefore allows us to gauge the effect of the transitory bid/ask and noise components. VRs have been used in recent studies by French and Roll (1986) , Fama and French ( 1988 ), Lo and Mackinlay ( 1988 ), and Kaul and Nimalendran ( 1990 ). Cochrane ( 1988 ), Lo and Mackinlay (1989), and Richardson and Smith ( 1991 ) analyze the properties of the VR test statistic. In general, they find that the VR test statistic is robust and has higher relative power compared with alternative test statistics. If dividend initiation reduces the bid/ask error component (either due to improved liquidity or reduced information asymmetry) or noise trading (better informed traders with greater consensus) then postdividend initiation VRs should be closer to one than predividend initiation VRs. Table 18 shows the summary statistics for 10day and 20day return variances 2 It can be shown that VR(k) can be written as a function of estimated autocorrelations of returns measured over the basic measurement interval (daily in our case) . VR k 1 + [ (k1) P x + (k2) P 2 + ... + p k .J Jx.
PAGE 70
64 TABLE 18 DISTRIBUTION OF VARIANCE RATIOS FOR PREAND POSTDIVIDEND INITIATION 500 DAY SAMPLE PERIODS. Sample = 310 VRio VR Z0 PRE POST DIFF a PRE POST DIFF 3 Minimum 0.294 0.326 0.867 0.215 0.250 1.093 Q1 0.775 0.794 0.180 0.743 0.709 0.178 Median 0.949 0.950 0 . 018 0.915 0.917 0.030 Q3 1. 166 1. 142 0. 171 1.211 1.148 0.234 Maximum 1.942 1.921 1. 114 2 . 532 2 . 078 1.504 Mean 0.993 0.983 0 . 010 0.985 0.960 0.025 rpb 0.61 ( 0 . 54 06 d ) 1.18 ( 0 . 2 392 d ) w c 0.51 ( 0 . 6077 d ) 1.31 ( 0 . 192 l d ) a For each firm, the difference between preand postdividend initiation variance ratios is DIFF = PRE POST. b t test of H 0 : DIFF = 0. c Wilcoxon signedrank test of H 0 : DIFF = 0. d Significance level of twotailed test. obtained from 500 day sample periods before and after dividend initiation. The 10day (20day) returns are obtained by multiplying 10 (20) successive daily returns. Therefore, in a 500 day sample period there are fifty 10day an twenty five 20day returns. The VRs are all less than one for both preand postdividend initiation periods. Mean and median predividend initiation VRs are smaller than postdividend initiation VRs but this difference is not statistically significant (Wilcoxon signedrank test) . Therefore, I cannot
PAGE 71
65 conclude that variance decreases are due to reductions in bidask spreads or noise trading. The Timing of Variance Changes The results show that the decrease in the price revaluations at earning announcements in the postdividend initiation period may be related to the general decline in returns variability. This decline may be caused by an information effect associated with dividend initiation. To ascribe causality to this information effect requires that variance changes should occur after the initiation of dividends. On the other hand, if variance changes do not systematically occur after dividend initiation, the plausibility of an dividend initiation information effect is questionable . In this section, I analyze the timing of variance changes. I first examine percentage variance changes calculated over successive 250 day sample periods beginning 750 trading days prior to dividend initiation. Due to the long series of returns required, only 272 firms are included in the sample. Table 19 shows the distribution of these percentage variance changes. The sample of firms experience significant variance declines in the year prior to (period [250,1]) as well as the year after (period [1,250]) the initiation of dividends.
PAGE 72
66 TABLE 19 SUMMARY STATISTICS OF PERCENTAGE VARIANCE CHANGES (250 DAY SAMPLE PERIOD) FOR SUCCESSIVE PERIODS Sample 272 Period 3 500,251 250,1 1,250 251,500 Median 7 .266 13 . 29 6.28 5.22 Mean 5.052 2 .74 6.50 10.06 No. Pos . b 129 164 161 136 % Pos . 47 . 4 60 . 2 59 . 1 50.0 W c 0.98 4.40 2.75 1.20 Pr>  W  0 . 3297 0 . 0001 0.0026 0.2324 Percentage variance changes for period (500,251) are calculated as Percentage variance change = 100 x [1 ( ~ 500 r : ? jil ] VAR (? so / 501) The period (1,250) is comparable to the preand postdividend initiation percentage variance changes in Table 15. Day 0 is the declaration date of the initial dividend. Number of firms in sample with positive percentage variance changes, i.e. variance decreases after dividend initiation. Test of H 0 : Percentage variance change = 0. Therefore the variance decline does not occur strictly after dividend initiation. There is significant variance decline in the year prior to dividend initiation. To better examine when variance declines occur, I examine successive overlapping periods of 100 trading days. For each firm, I collect 400 trading days of returns: 200 before the initiation of dividends and 200 after. I divide these 400 days
PAGE 73
67 into seven 100 day samples with each sample overlapping the prior sample by 50 days. Sample period one consists of returns from day 200 to 101; sample period two consists of returns from day 150 to 51; sample period three consists of returns from day 100 to 1; and so on. The final sample period seven consists of returns from day 101 to 200. The sample periods are shown in Figure 2 . 200 100 0 100 200 Sample periods 135 7 2 4 6 Figure 2 Sample Periods for Detecting Variance Changes Within each sample period, I test whether a significant variance decrease has occurred. For each firm, I count the number of significant variance decreases occurring in the predividend initiation sample periods 1, 2 and 3 and the postdividend initiation sample periods 5, 6, and 7. Denote these counts as C0UNT1 and C0UNT2 , respectively. If a variance change is more likely to occur following dividend initiation,
PAGE 74
68 C0UNT2 is going to be greater than C0UNT1. The test is therefore , H 0 : C0UNT2C0UNT1 = 0 versus H x : C0UNT2 C0UNT1 > 0 I test this hypothesis using the Wilcoxon signedrank statistic. To detect a variance decrease within a sample period, I use a test statistic T* which has been extensively examined by Hsu (1977, 1979). Let M be a consecutively observed sequence of daily returns R 1 ,R 2 ,...,R M which are assumed to be independent normal variates. Further, let the variances of these variables be represented by a : 2 , a 2 2 , . . . , a M 2 , respectively. The problem is to test whether a variance shift 6 has occurred, H 0 : aL 2 = a 2 2 = . . . = a M 2 a 0 2 i (a 0 2 is unknown) against H a : ? 2 (Ji = Go = ak M 2 = a 0 2 ; 2 _ k+1 = a k+2 a 2 _ M a 0 2 + 0 , where k is unknown (k=l , 2 , . . . , Ml) ; is unknown and (a 0 2 +
PAGE 75
69 Under the null hypothesis of no variance shift, this test statistic has mean E[T] =! $ and variance VAR[T] = (M+l) / [ 6 (Ml)(M+2)]. I use the large sample approximation T*=(T0. 5)/ {VAR[T] } % which is distributed asymptotically standard normal . In this test, I calculate T* for each sample period with M=100. I deem that a variance decrease has occurred in that sample period if T*<2.6. This corresponds to a significance level of just under 0.5%. The total firm sample size is 328 firms 3 . Of the 328 firms examined, 83 had no significant variance changes in the 7 sample periods. 56 firms had only variance increases. 189 firms (58% of the sample) had at least one variance decrease in the 7 sample periods. The results of this analysis are shown in Table 20. The frequency distribution of variance decreases by sample periods does not show a preponderance of variance changes occurring after dividend initiation (i.e. in sample periods 5, 6 and 7). This is confirmed by the Wilcoxon signedrank statistic's value of 0.70. I conclude from this analysis that the decrease in returns variability does not occur systematically after dividend initiation. Variance decreases occur with equal frequency before and after dividend initiation. 3 Four OTC firms were excluded because they did not have an uninterrupted series of consecutive returns. These firms subsequently traded on either the AMEX or NYSE.
PAGE 76
70 TABLE 20 ANALYSIS OF THE TIMING OF VARIANCE DECREASES Sample Period 1 2 3 4 5 6 7 Beginning Return 3 200 150 100 50 1 51 101 Ending Return 3 101 51 1 50 100 150 200 Decreases 13 36 41 31 33 44 39 35 a The announcement date of the initial dividend is day 1. b A variance decrease is deemed to have occurred if T <Â—2.6 for that sample period. H 0 : Variance decreases occur with equal frequency before and after dividend initiation. COUNT2 COUNT1 = 0 against H a : Variance decreases occur more frequently after dividend initiation. C0UNT2 C0UNT1 > 0 where COUNT1 = COUNT2 = TOTAL NUMBER OF VARIANCE DECREASES OCCURRING IN SAMPLE PERIODS 1, 2 OR 3 FOR FIRM j . TOTAL NUMBER OF VARIANCE DECREASES OCCURRING IN SAMPLE PERIODS 5, 6 OR 7 FOR FIRM j . The Wilcoxon signedrank statistic W is 0.70. Cannot reject null at the 5% significance level. Sample size is 328 firms.
PAGE 77
71 The Effect of Clustered Data I examine the possibility that specific periods are causing the observed decrease in firm volatility. Just over 75% of my sample firms initiated dividends in the period 197277, with over 50% occurring between 19751977. The effect of interest rates on volatility is particularly relevant to my sample. The seventies experienced volatile changes in interest rates. In particular, interest rates rose from 197273 and fell from 197476.'' It is plausible that for a large number of firms in my sample, predividend periods coincided with a rising interest rate environment (and consequently high returns volatility) and postdividend periods coincided with falling interest rates (low returns volatility). If the decrease in firm volatility is independent of secular influences, the proportion of firms experiencing variance declines should not cluster around certain periods. Table 21 shows percentage variance changes calculated for 500 day sample periods distributed by the year of dividend initiation. The table shows that the proportion of firms experiencing variance declines is not constant across the years. In particular, the periods 197273 and 197879 have a majority of firms experiencing variance increases. Conversely, the period 197576 which accounts for almost 40% of my sample of dividend initiating firms has above average proportion of A See, for example, Livingston (1990) p. 3 or Madura (1989) p. 37.
PAGE 78
72 TABLE 21 DISTRIBUTION OF PERCENTAGE VARIANCE CHANGES (500 DAY SAMPLE PERIOD) BY YEAR OF DIVIDEND INITIATION Year No . a Pos . b % Pos . c Mean Median 1970 6 5 83 . 3 20.4 24.5 1971 4 4 100 . 0 34.2 33.2 1972 18 1 5.6 73 . 1 33.2 1973 29 4 13 . 8 57 . 6 63.9 1974 24 16 66 . 7 4 . 5 19.1 1975 58 57 98 . 3 54 . 5 56.8 1976 62 62 100 . 0 53 . 4 57.2 1977 47 33 70.2 13 . 1 2 6.6 1978 20 9 45 . 0 22 . 4 4 . 3 1979 11 4 36.4 7 . 2 4 . 6 1980 9 6 66 . 7 8 . 0 31.1 1981 6 5 83 . 3 21.8 35.0 1982 3 3 100 . 0 19 . 8 16.8 1983 5 4 80 . 0 23.7 21.7 1986 3 0 0 . 0 60.1 69 . 0 a Number of sample firms initiating dividends in that year. b Number of sample firms with positive percentage variance changes, i.e. variance decreases. c Sample firms with positive percentage variance changes as a percentage of sample firms initiating dividends in that year.
PAGE 79
73 variance decreases. This pattern is also found for variances calculated for 900, 250 and 100 day sample periods. Table 22 shows the distribution of percentage variance changes adjusted by a estimate of the contemporaneous market variance. Note that after adjusting for market variance, the mean and median percentage variance change is negative for the period 197576. Firms initiating dividends in those years experience decreases in variances but of a smaller magnitude than the market. The behavior of sample firm variances through time is plotted in Figures 3 through 5. Figure 3 plots the monthly variance of an index of dividend initiating firms (VAR) and a weighted average of all firms on the NYSE/ AMEX and NASDAQ tapes. For each month (m=l/70 to 12/86) , VAR is the average monthly returns variance for a set of dividend initiating firms N m . A firm is included in N m if it initiated dividends within 24 months of month m. VAR m = irS Var h.n iN m i =1 N m The monthly returns variance for the firm j is calculated as m Var bm = Â£ R % X =1 where T m is the number of trading days in month m. If a firm has less than 15 valid returns in any month, it is excluded from VAR. VAR thus represents the mean monthly variance for
PAGE 80
TABLE 22 DISTRIBUTION OF MARKET ADJUSTED PERCENTAGE VARIANCE CHANGES (500 DAY SAMPLE PERIOD) BY YEAR OF DIVIDEND INITIATION Year No . a Pos . b % Pos . c Mean Median 1970 6 2 33 . 3 3 . 9 9.7 1971 4 1 25 . 0 14 . 9 15.6 1972 18 14 77 . 8 28 . 8 58 . 1 1973 29 29 100 . 0 51.4 51.8 1974 24 22 91.7 31.0 37.7 1975 58 9 15 . 5 48 . 4 36.5 1976 62 24 39 . 7 50 . 5 16.2 1977 47 35 74 . 5 18 . 6 32 . 0 1978 20 18 90 . 0 45.2 53 . 3 1979 11 11 100 . 0 39 . 9 42 . 5 1980 9 6 66.7 17 . 3 34.5 1981 6 5 83 . 3 25 . 0 33.9 1982 3 2 66.7 6.5 8.8 1983 5 1 20 . 0 38 . 6 22.9 1986 3 3 100 . 0 63 . 7 61.8 a Number of sample firms initiating dividends in that year. b Number of sample firms with positive percentage variance changes, i.e. variance decreases. 0 Sample firms with positive percentage variance changes as a percentage of sample firms initiating dividends in that year .
PAGE 81
75 the sample of dividend initiating firms through a moving window of 24 months. From 1/70 to 12/72, VarM is the monthly returns variance of the equallyweighted NYSE/AMEX market index. From 1/73 to 12/86, VarM is a weighted average of the monthly returns variance of the equallyÂ— weighted NYSE/AMEX and NASDAQ market indices . VarM m = wVarjjYgg/^gx p, + (1w) Var NASDAQ(m The weight w corresponds to the proportional representation of NYSE/AMEX firms in the sample (w = 247/332) . The monthly variances of the market indices are calculated in the same way as for individual firms. Figure 3 shows the wide variations in returns variance of the sample. The returns variance is particularly high in the years 1973 to 1976. The sample variance behaves similarly to the wider market. Therefore, variations in return variance are not peculiar to the sample of dividend initiating firms. Figure 4 plots the monthly variance and the distribution of firms initiating dividends, expressed as a percentage of the total sample (% INIT) . This figure illustrates the effect of clustering in the sample. The large proportion of firms initiating dividends between 1975 and 1977 follows a period of above average returns variance. This potentially could explain the reported decrease in returns variance and quarterly earnings price revaluations.
PAGE 82
76 10 X10 7001 7101 7201 7301 7401 7501 7001 7701 7801 7901 8001 8101 8201 8301 8401 8501 8001 VAR VarM Figure 3 Monthly Returns Variance of an Index of Dividend Initiating Firms (VAR) and a Weighted Market Index (VarM) . Figure 5 plots the monthly variance and the 3month Tbill rate. The source for the 3 month Tbill rate is the Interest Rates: Money and Capital Market Rates tables from the Federal Reserve Bulletin . The figure shows that the increase in returns variance in the middle seventies appears to be preceded by increases in interest rates. This is not to say that interest rates have caused the changes in returns
PAGE 83
77 volatility. They could both be reactions some other macroeconomic event, such as the oil shock of 1973. Figure 4 Monthly Returns Variance of an Index of Dividend Initiating Firms (VAR) and the Distribution of Firms Initiating Dividends (% INIT) 1 Values on the vertical axis are in percentages. Â’ Multiply values on the vertical axis by 10~ 3 .
PAGE 84
78 Figure 5 Monthly Returns Variance of an Index of Dividend Initiating Firms (VAR) and the 3Month Treasury Bill Rate (TBill) a Source: Federal Reserve Bulletin . Values on the vertical axis are in percentages. b Multiply values on the vertical axis by 10~ 3 . The general conclusion that can be drawn from the above analysis is that variance decreases seemingly associated with dividend initiation are not caused by information related factors. Variance decreases do not reduce the degree of negative autocorrelation that would indicate reductions in
PAGE 85
79 noise trading or bidask spreads. Neither do they systematically occur following dividend initiation. The variance decreases appear to be occur mostly in the middle seventies when a large proportion of dividend initiations coincide with periods of low interest rates preceded by periods of high interest rates. Variance decreases could therefore be driven by changes in interest rates or other macroeconomic events. The evidence in this and the previous chapter tends to reject the hypothesis that there are long run information effects associated with dividend initiation.
PAGE 86
CHAPTER 6 SUMMARY AND CONCLUSIONS It has been suggested that dividends play a role in transmitting information to investors. This hypothesis has motivated the examination of the relationship between dividend initiation and earnings announcements. Previous research has found that the price reactions at earnings announcements are on average lower after dividends have been initiated. This result has been interpreted to support the contention that dividends has conveyed information. Dividend information has preempted information that would otherwise have been conveyed by earnings reports. I contend that this conclusion is invalid. Showing that price reactions have declined merely shows that earnings and dividends information is substitutable. The testable implication of dividends as an information source is whether dividends reduce costly private information acquisition. The appropriate empirical measure is therefore not earnings returns variance alone but standardized returns variance. If nonannouncement returns reflects the private information activities of investors, then standardizing announcement returns variance by the surrounding nonannouncement returns gives a measure of the importance of the public announcement. 80
PAGE 87
81 If private information acquisition is reduced by having dividends, investors should place greater reliance on the firm's public announcements, both dividends and earnings. In addition, this reliance should be greatest for firms for which information is least readily available. I do not detect any difference in the relative measures of the information content of earnings announcements before and after dividend initiation. This result holds for the total sample of firms as well as when the sample is grouped by proxies of information availability. The information transmission hypothesis is not supported. What is puzzling is that there is a significant reduction in raw price reactions for postdividend initiation earnings announcements. I contend that this reduction is caused by a general decrease in the volatility of returns which coincides with dividend initiation. Regression analysis provides evidence that the reduction in postdividends earnings price reactions is related to the variance decrease and not to a reduction in the information content of earnings. It can be argued that an information effect associated with dividend initiation causes the decrease in returns volatility. For example, if the initiation of dividends induces greater consensus among traders, noise trading or bidask spreads may be reduced. I test for reductions in variance ratios before and after dividend initiation. I do not find any significant change. Further, one should observe the frequency of variance
PAGE 88
82 declines to be higher after dividend initiation. Tests of the timing of variance changes show no evidence of this. Variance decreases appear with equal frequency before and after dividend initiation. The findings of this study are not supportive of the hypothesis that dividends play an informational role. The search for a valid reason why firms initiate dividend payments must still continue. However, it does provide evidence that the market is sufficiently efficient such that any informational effects from dividends are minor. /
PAGE 89
83 REFERENCES Asquith, P. and D. W. Mullins, 1983, "The Impact of Initiating Dividend Payments on Shareholders' Wealth," Journal of Business 56, 7796. Atiase, R. , 1985, "Predisclosure Information, Firm Capitalization, and Security Price Behavior Around Earnings Announcements," Journal of Accounting Research 23, 2136. Bagehot, W. , 1971, "The Only Game in Town,", Financial Analysts Journal 22, 1214. Black, F. , 1986, "Noise", Journal of Finance 41, 529543. Beaver, W. H. , 1968, "The Information Content of Annual Earnings Announcements," Empirical Research in Accounting: Selected Studies, Supplement to the Journal of Accounting Research 6, 6792. Bhattacharya , S., 1979, "Imperfect Information, Dividend Policy, and the 'Bird in the Hand' Fallacy," Bell Journal of Economics 10, 259270. Bhushan, R. , 1989, "Collection of Information About Publicly Traded Firms: Theory and Evidence," Journal of Accounting and Economics 11, 183206. Christie, A. A., 1982, "The Stochastic Behavior of Common Stock Variances," Journal of Financial Economics 10, 407432 . Cochrane, J. H., 1988, "How Big is the Random Walk in GNP? , " Journal of Political Economy 96, 893920. COMPUSTAT . Standard and Poor's Compustat Services, Englewood, Colo . Copeland, T., and D. Galai, 1983, "Information Effects on the BidAsk Spread," Journal of Finance 38, 14571469. CRSP Daily Stock Master Returns and Index , Center for Research in Security Prices, University of Chicago, Chicago, 111. Demsetz, H. , 1968, "The Cost of Transacting," Quarterly Journal of Economics 82, 3353.
PAGE 90
84 Diamond, D. W. , and R. E. Verrecchia, 1981, "Informational Aggregation in a Noisy Rational Expectations Economy," Journal of Financial Economics 9, 221235. Grant, E. B. , 1980, "Market Implications of Differential Amounts Information of Interim Information," Journal of Accounting Research 18, 255268. Fama, E. F. and K. R. French, 1988, "Permanent and Temporary Components of Stock Prices," Journal of Political Economy 96, 246273. Federal Reserve Bulletin . Board of Governors of the Federal Reserve System, Washington, D. C. French, K. , and R. Roll, 1986, "Stock Return Variances: The Arrival of Information and the Reaction of Traders," Journal of Financial Economics 17, 526. Glosten, L. , and P. Milgrom, 1985, "Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders," Journal of Financial Economics 14, 71100. Healy, P. , and K. Palepu, 1988, "Earnings Information Conveyed by Dividend Initiations and Omissions," Journal of Financial Economics 21, 149175. Hellwig, M. F. , 1980, "On the Aggregation of Information in Competitive Markets," Journal of Economic Theory 22, 477498 . Ho, T. , and H. Stoll, 1981, "Optimal Dealer Pricing under Transaction and Return Uncertainty," Journal of Financial Economics 9, 4774. Hollander, M. and D. A. Wolfe, 1973, Nonparametric Statistical Methods . (New York, NY: Wiley) . Hsu, D. A., 1977, "Tests of Variance Shift at an Unknown Time Point," Applied Statistics 26, 279284. Hsu, D. A., 1979, "Detecting Shifts of Parameter in Gamma Sequences with Applications to Stock Price and Air Traffic Flow Analysis," Journal of the American Statistical Association 74, 3140. John, K. and J. Williams, 1985, "Dividends, Dilution and Taxes: A Signalling Equilibrium," Journal of Finance 40, 10531070.
PAGE 91
85 Kaul G. and M. Nimalendran, 1990, "Price Reversals: BidAsk Errors of Market Overreaction?," Journal of Financial Economics 28, 6793. Livingston, M. , 1990, Money and Capital Markets: Financial Instruments and Their Uses . (Englewood Cliffs, N. J.: PrenticeHall) . Lo, A. W. and A. C. MacKinlay, 1988, "Stock Market Prices Do Not Follow Random Walks: Evidence from a Simple Specification Test," Review of Financial Studies 1 , 41 66 . Lo, A. W. and A. C. MacKinlay, 1989, "The Size and Power of the Variance Ratio Test in Finite Samples: A Monte Carlo Investigation," Journal of Econometrics 40, 203238. Lobo, G. J. and A. A. W. Mahmoud, 1989, "Relationship Between Differential Information and Security Return Variability," Journal of Accounting Research 27, 116134. Madura, J., 1989, Financial Markets and Institutions . (St. Paul, Minn.: West Publishing Co.). Marsh, T. , and R. Merton, 1987, "Dividend Behavior for the Aggregate Stock Market," Journal of Business 60 , 1 40 . Miller, M. H. and F. Modigliani, 1961, "Dividend Policy, Growth and the Valuation of Shares," Journal of Business 34, 411433. Miller, M. H. and K. Rock, 1985, "Dividend Policy Under Asymmetric Information," Journal of Finance 40 , 1031 1051 . Richardson, M. and T. Smith, 1991, "Robust Power Calculations with Tests for Serial Correlation in Stock Prices," Working Paper, Rodney L. White Center for Financial Research, The Wharton School, University of Pennsylvania. Ross, S. A., 1976, "The Determination of Financial Structure: The IncentiveSignalling Approach," Bell Journal of Economics Spring, 2340. Shores, D. , 1990, "The Association Between Interim and Security Returns Surrounding Earnings Announcements," Journal of Accounting Research 28, 164181. Thompson, R. B. , II, C. Olsen, and J. R. Dietrich, 1987, "Attributes of News About Firms: An Analysis of FirmSpecific News Reported in the Wall Street Journal Index," Journal of Accounting Research 25, 245274.
PAGE 92
86 Venkatesh, P. C. , 1989, "The Impact of Dividend Initiation on the Information Content of Earnings Announcements and Returns Volatility", Journal of Business 62, 175197. Verrecchia, R. E. , 1982, "Information Acquisition in a Noisy Rational Expectations Economy," Econometrica 50, 14151430. Wall Street Journal Index . Dow Jones and Co. , New York, N. Y. Zeghal, D. , 1984, "Firm Size and the Informational Content of Financial Statements," Journal of Financial and Quantitative Analysis 19, 299310.
PAGE 93
87 BIOGRAPHICAL SKETCH Thian Soon Wong was born in Johor Bahru, Malaysia, on March 19, 1958. He received a bachelor's degree in electrical engineering from the University of Malaya, Kuala Lumpur, Malaysia, in 1982 and a Master of Business Administration from the University of Florida in 1986. He expects to receive a Doctor of Philosophy degree in finance from the University of Florida in 1991.
PAGE 94
I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and its fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. / '/ / \ c /) f / Robert C. Radcliffe, Chairman Professor of Finance, Insurance, and Real Estate I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and its fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Anand Desai Assistant Professor of Finance, Insurance, and Real Estate I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and its fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Michael D. Ryngae'rt Assistant Professor of Insurance and Real Estate Finance , I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and its fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. This dissertation was submitted to the Graduate Faculty of the Department of Finance, Insurance, and Real Estate in the College of Business Administration and to the Graduate School and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. Dean, Graduate School

