what is the information content of dividend changes · pdf filewhat is the information content...
TRANSCRIPT
What is the information content of dividend changes? A new investigation of an old puzzle
Shuping Chen* Assistant Professor
Department of Accounting University of Washington Business School
Box 353200 Seattle, WA 98195
Email: [email protected]
Terry Shevlin Paul Pigott/PACCAR Professor of Business Administration
University of Washington Business School Box 353200
Seattle, WA 98195 Email: [email protected]
Yen H. Tong Ph.D. Student
University of Washington Business School Box 353200
Seattle, WA 98195 Email: [email protected]
Draft Date: October 6, 2004
*Corresponding author. We thank Jennifer Francis, Ryan LaFond, Per Olsson, and Katherine Schipper for sharing the AQ factor returns data and I/B/E/S for providing us with the analyst data used in this project.
What is the information content of dividend changes? A new investigation of an old puzzle
Abstract
In this paper we extend Grullon, Michaely, and Swaminathan (2002) and examine the hypothesis that dividend increases (decreases) are associated with decreases (increases) in firms’ information risk, which, together with the decreases (increases) in other systematic risks, contributes to the positive (negative) price changes surrounding dividend change announcements. Using Fama-French three-factor models augmented by an Information Risk factor, we find that, as predicted, dividend-decreasing firms exhibit an increase in information risk in the years after decrease announcements. However, our results on dividend-increasing firms are sensitive to the samples we use. Further analysis of the information characteristics of the dividend change firms indicates a reduction in analyst forecast dispersion and stock return volatility for dividend-increase firms and an increase in analyst forecast dispersion and stock return volatility for dividend-decrease firms in the years after dividend changes. Inconsistent with the prediction of traditional signaling models, dividend increase (decrease) firms do not exhibit an increase (a decrease) in future profitability.
1
What is the information content of dividend changes? A new investigation of an old puzzle
1. Introduction
A long-standing literature in accounting and finance has documented that
dividend changes are associated with changes in stock price of the same sign around the
dividend change announcement (e.g., Healy and Palepu, 1988; Michaely et al., 1995).
However, the same literature has consistently failed to produce evidence in support of
another fundamental implication of the dividend signaling models - that dividend changes
and future earnings move in the same direction. If dividend changes do not signal
changes in future profitability, then why do we observe positive (negative) price changes
for dividend increase (decrease) announcements? This lack of congruence between the
intuitively appealing predictions of dividend signaling models and the empirical evidence
to date suggests that dividend changes may convey information other than firms’ growth
in future cash flows, such as changes in risks (Allen and Michaely, 2002; Grullon,
Michaely, and Swaminathan, 2002). Using Fama-French three-factor models Grullon et
al. (2002) find that firms that increase (decrease) dividends exhibit decreases (increases)
in market, size, and book-to-market betas, and conclude that the systematic risks for
dividend-increase (decrease) firms decrease (increase) after their dividend changes. In
this paper, we extend the Grullon et al. (2002) findings: we propose and examine the
hypothesis that dividend changes are also associated with changes in the information risk
of firms; that is, dividend increases (decreases) are associated with decreases (increases)
in firms’ information risk, which contributes to the positive (negative) price changes
surrounding change announcement.
Recent theoretical research shows that firm-specific information risk is priced,
and cannot be diversified away (Easely and O’Hara, 2001, O’Hara, 2003, Leuz and
Verrechia, 2004). These models imply that the precision of information reduces the risk
premium demanded by uninformed traders in a multi-asset market. Based upon these
theoretical models, Francis, LaFond, Olsson, and Schipper (2004, hereafter FLOS) argue
that earnings quality is an important accounting determinant of information risk. They
use accruals quality to capture firms’ accounting information risk and explore the relation
between earnings quality and the cost of capital.1 FLOS (2004) find significant factor
loadings on an “accrual-quality” factor in multi-factor asset pricing regressions, and
interpret their findings as evidence that information risk is a priced risk factor. Aboody,
Hughes, and Liu (2004) lends further empirical support to the theoretical prediction that
information risk is priced by the capital market by simultaneously examining the pricing
of information risk and concomitant information-based trading by insiders. Consistent
with this research, we define information risk as the probability that firm-specific
information pertinent to investor pricing decisions is of poor quality. To the extent that
earnings serves as an important information item investors use in making investment
decisions, firms with poor earnings quality exhibit higher information risk.
To examine our central prediction that dividend changes convey information
about changes in firms’ information risk, we augment the Fama and French (1993) three-
factor model by adding an Information Risk factor (IR factor) based on the FLOS (2004)
accruals quality measure, and examine the factor loadings on this IR factor for dividend
1Under the assumption that cash flow is the primitive element that investors price, FLOS (2004) identify accruals quality as the measure of information risk associated with a key accounting number – earnings. Because accruals quality reflects the mapping of accounting earnings into cash flows, they argue that higher accruals quality indicates higher earnings quality. We use the terms “accruals quality” and “earnings quality” interchangeably in this paper.
3
change firms before and after the dividend change announcement. Consistent with our
prediction, we find that dividend-decreasing firms experience a significant increase in
their information risk beta after dividend change announcement. Consistent with Grullon
et al, we also find significant increases in market beta and B/M beta after dividend
decreases. However, our results concerning the information risks of dividend-increasing
firms are mixed and sensitive to the samples we use. Further examination of the
information characteristics surrounding dividend changes reveals a reduction in analyst
forecast dispersion and stock return volatility for dividend-increasing firms, and an
increase in analyst forecast dispersion and stock return volatility for dividend-decreasing
firms.
Our paper contributes to the existing literature in the following three ways. First,
instead of focusing on earnings changes of dividend change firms, we focus on changes
in the non-diversifiable information risk of firms that announce changes in dividend
payouts. Twenty years after the emergence of the first signaling models (Bhattachrya,
1979; Miller and Rock, 1985; John and Williams, 1985), researchers are still grappling
with the consistent lack of evidence in support of the basic prediction of these dividend
signaling models: that firms adjust dividends to signal their prospects. The new evidence
documented here not only deviates from, but also extends, the long-standing empirical
literature focusing on earnings changes following dividend changes. We provide
evidence that dividend-decreasing firms exhibit significant increases in their information
risk after their change announcements, and thus reconcile the significant price decline in
part with changes in firms’ information risk.
4
Second, our study complements and extends research on the information content
of dividend changes from a risk perspective. Grullon et al. (2002) present an alternative
explanation to the abnormal returns associated with dividend change announcements -
dividend changes contain information about changes in systematic risks. These authors
find that firms that increase (decrease) dividends show decreases (increases) in
systematic risk following the dividend change announcement. They interpret their
findings as consistent with a ‘maturity hypothesis’: as firms mature, firms’ systematic
risk, growth and investment opportunities all start to decline. As a result, more cash is
distributed back to investors in the form of dividends. However, the maturity hypothesis
only explains dividend increases, and is hard to reconcile with dividend decreases. We
show that dividend decreases are accompanied by increases in firms’ information risks.
Thus, our results on dividend decreases complement Grullon et al.’s (2002) maturity
hypothesis. However, our results fail to provide unequivocal support for the prediction
that dividend increases are associated with declines in firms’ information risks. We hope
our research will help open up more discourse in this area, which may ultimately lead to a
better understanding of the information content of dividend changes.
Our paper also relates to the growing accounting literature on earnings quality and
information risk. FLOS (2004) investigate the relation between earnings quality and the
cost of capital for a large sample of firms over the period 1971-2000. They predict and
find that firms with lower quality accruals have higher costs of capital as evidenced by
larger realized costs of debt, lower debt ratings, larger equity betas, and positive loadings
on an accruals quality factor added to one-factor and three-factor asset pricing
regressions. Aboody et al. (2004) confirms the FLOS (2004) findings and reports that
5
information risk is priced by the market in an insider trading setting. They document
positive associations between accounting proxies for asymmetric information and
expected returns, and between these proxies and insider trading profits. Both papers
employ earnings-related proxies for information risk and their findings lend empirical
support to the theoretical argument that information risk should be reflected in the cost of
capital. We build upon this literature by examining the joint hypothesis that information
risk is priced and that dividend changes contain information about changes in firms’
information risk. Our paper also relates to Skinner’s (2004) study on whether dividends
provide information about earnings quality. Skinner (2004) examines earnings
persistence using an AR (1) model and finds that earnings are more persistent for firms
that pay dividends, and the strength of the association between current and future
earnings is positively related to the size of the dividends and the size of the firm. Thus,
our results, together with Skinners’ findings, suggest that dividends contain information
about firms’ information risk related to earnings quality.
The rest of the paper is organized as follows. In section 2 we review relevant
literature and develop our empirical predictions. In section 3 we describe our empirical
design. Section 4 presents our empirical results, and section 5 concludes.
2. Relevant Literature and Hypothesis Development
2.1 Dividend changes and hypothesis
Dividends and share repurchases are important forms of payout to shareholders of
modern corporations.2 The notion “the information content of dividends” originates in
2 Share repurchases have become increasingly more popular as a form of payout to shareholders in recent years. However, share repurchases differ from dividends in that managers do not need to commit to
6
Miller and Modigliani (1961). Miller and Modigliani suggest that if managers’
expectations about future earnings affect their current dividend payout policy, then
changes in dividends will convey information about future earnings. Bhattacharya (1979)
formalizes this notion by treating dividends as ex-ante signals of future cash flows.
Miller and Rock (1985) adopt a slightly different perspective and treat dividends as
providing information about earnings as a description of the sources and uses of funds
identity. This second approach makes the subtle distinction that the fact that dividends
convey information does not necessarily indicate they are being used consciously as a
signal. Consistent with this second approach, Brav, Graham, Harvey, and Michaely
(2003)’s recent survey of corporate CFOs and treasurers indicates that managers believe
dividends convey information, but they do not use dividends explicitly and deliberately
as a costly signal to change market’s perceptions concerning future earnings prospect.
The traditional dividend-signaling models predict that dividends convey
information about firms’ future earnings prospect - dividend increases signal good news
and dividend decreases signal bad news. Two testable implications of these signaling
models are (1) dividend changes are followed by earnings changes in the same direction,
and (2) unanticipated dividend changes are accompanied by stock price changes in the
same direction. Extant empirical results generally support prediction (2), showing that
announcements of dividend increases (decreases) are associated with positive (negative)
abnormal returns. In contrast, researchers have consistently failed to provide empirical
maintaining a certain level of payouts, and repurchases allow managers more flexibility in investment strategies. In this paper we concentrate on dividend changes and do not investigate share repurchases. See Brav, Graham, Harvey, and Michaely (2003) for more detailed discussion on dividends and share repurchases.
7
support for prediction (1). The results documented more often are opposite to what is
predicted by the signaling theories.
For example, although Healy and Palepu (1988) find earnings increases in the two
years following dividend initiation, they also find evidence of earnings increases in the
next several years for firms that omitted dividend payments. Using a large number of
firms from 1979 to1991 and different definitions of earnings innovations, Benartzi,
Michaely and Thaler (1997) find two sets of robust results. First, dividend changes are
positively associated with earnings changes of the same sign in the year of and two years
preceding the dividend change. Second, there is no positive relation between dividend
increases and future earnings changes. However, there is a clear pattern of earnings
increases in the two years following dividend decreases. Grullon et al. (2002) provide
similar findings. In addition, they document that firms’ profitability decreases following
dividend increases. Thus, accumulated empirical evidence does not support the assertion
that dividend increases (decreases) convey information about increases (decreases) in
future earnings.
The contradictory evidence on price changes upon dividend change
announcements versus subsequent earnings changes presents a puzzle that challenges the
central prediction of traditional dividend signaling models. If the positive (negative)
market reaction is not associated with future earnings increases (decreases), what could
be contributing to the observed abnormal returns? In this paper, we re-examine this old
puzzle by presenting evidence on the changes in information risk surrounding dividend
changes. Our investigation is along the same line of Grullon et al. (2002). Grullon et al.
(2002) posit that the fundamental news about a firm must be either about its cash flows or
8
its risk characteristics. If the good news (bad news) in a dividend increase (decrease) is
not about expected increases (decreases) in future cash flow, it may relate to changes in
firms’ systematic risks. Based on this recognition, Grullon et al. (2002) present an
alternative explanation, the “maturity hypothesis,” to explain the price changes upon
dividend change announcement. Specifically, they propose that as firms mature, they
experience a concurrent decline in their investment opportunity set and systematic risk.
The decline in growth opportunities generates an increase in free cash flows, leading to
an increase in dividends and a decrease in future profitability. The positive market
reaction surrounding dividend increases indicate that the news about risk dominates the
news about profitability. The maturity hypothesis represents an important deviation from
the traditional dividend-signaling model. However, the maturity hypothesis cannot be
applied to explain why firms decrease dividend payouts. In this paper we adopt a
different approach, and argue that dividend changes contain information about changes in
firms’ systematic risk, including non-diversifiable information risk, without relating them
to firms’ life cycle. Our objective is to better understand the information content of
dividend changes, and to relate firms’ dividend policy to the cost of capital. We
formalize our hypothesis, in alternative form, as follows:
Hypothesis: Ceteris paribus, dividend increasing/decreasing firms exhibit a decrease/an increase in their information risks, respectively.
2.2 Information risk and earnings quality
Recent development in accounting research shows that information risk, as
proxied by earnings quality, is priced by the market (FLOS 2004; Aboody et al., 2004).
These empirical investigations are premised on recent theoretical research, which
demonstrates that information risk is a non-diversifiable risk factor (e.g., Easley and
9
O’Hara, 2001). Easley and O’Hara (2001) advance the notion that uninformed traders
require a premium to invest in risky assets in a multi-risky asset market with both
informed and uninformed traders. In their model, informed traders are better able to
adjust their portfolio weights in response to new information, some of which is private to
them. As a result, uninformed traders face a form of systematic, non-diversifiable
information risk, and will require a premium for bearing this information risk. The two
implications of their model are: 1) the required risk premium increases with private
information, and 2) the required risk premium decreases with the precision of public and
private information.
Aboody et al. (2004) builds their investigation on the first implication on the
asymmetric information risk, and examine whether insiders exploit their information
advantage to obtain abnormal profits. FLOS (2004) explore the second implication, i.e.,
the information risk resulting from lack of precision of information. Based on the
premise that cash flow is the primitive element that investors price, they argue that a poor
mapping of accruals into cash flows signals poorer earnings quality and increases
information risk, and that investors price this information risk. Thus, precise accounting
information reduces the cost of capital by decreasing the non-diversifiable information
risk to uninformed traders. Both studies use earnings quality measures to proxy for
information risk. Aboody et al. (2004) define earnings quality as “the portion of
accounting earnings susceptible to management discretion” and predict that firms with
low earnings quality present higher information risk. FLOS (2004) use similar measures
of information risk based on accruals quality. Both studies share the same premise in
using earnings/accruals quality metrics to capture information risk: cash flows are less
10
susceptible to manipulation, and investors price cash flows as the ‘non-manipulated’
element. Better accruals quality indicates a better mapping of cash flows into earnings,
and insiders are better able to parse out the ‘noise’ in the public signals contained in
earnings and accruals.
Our definition of information risk follows that of FLOS (2004). We define
information risk as the probability that firm-specific information is of poor quality, and
use earnings quality to capture this information risk. Thus, firms with poor earnings
quality pose higher information risk (IR), and we expect to find an increase in IR for
dividend decreasing firms and a decrease in IR for dividend increasing firms.
3. Empirical Design
3.1 Data and Sample
We form our test sample following Grullon et al. (2002). We start from all dividend
paying firms listed on NYSE, AMEX, and NASDAQ between April, 1974 and March
1999.3 We delete dividend announcements with other nondividend distribution events
such as stock splits, stock dividends, and mergers occurring within 15 trading days
surrounding the dividend announcement. To be included in the sample, a dividend
announcement must relate to quarterly cash dividend paid on ordinary common shares
(not shares of Americus Trust components, closed-end funds, or REITs), and the previous
cash dividend payment was paid within a window of 20-90 trading days prior to the
current dividend announcement. Before calculating dividend changes we adjust dividend
3 We have information risk (AQ) factor returns available from FLOS (2004) from April 1971 to March 2002. Since our primary tests require 73 months of return data centered on the dividend change announcement, we include dividend announcements made between April 1974 and March 1999 (inclusive). Thus, our analysis and inference are not influenced by the Jobs and Growth Tax Relief Reconciliation Act of 2003 that reduces investor tax rates for dividends and long-term capital gains to 15%.
11
per share for any stock splits or stock dividends so that all dividends are on the same
split-adjusted basis. Similar to Grullon et al. (2002) we retain only dividend changes
between 12.5% and 500% to ensure that we include economically significant dividend
changes at the lower bound and exclude outliers at the upper bound.4 Since our time-
series, firm-specific regressions require 36 months of returns before and 36 months of
returns after the dividend change announcement, we require our sample firms to have
corresponding stock return data available for 73 months centered on the dividend change
announcement month on CRSP. This sample selection process yields 10,597 firm-
dividend observations over 25 years from April 1974 to March 1999 (inclusive) made by
2,981 unique firms. Of these, 9,110 are dividend increase announcements and 1,487 are
dividend decrease announcements.
3.2 Information Risk Factor
We use the FLOS (2004) AQ factor as our information risk factor in testing
whether information risk changes surrounding dividend change announcements.5 FLOS
(2004) calculate the information-risk-factor-mimicking returns to a portfolio that takes a
long position in stocks with the poorest accruals quality (AQ) and takes a short position
in the best AQ stocks.6 To obtain the factor mimicking returns, FLOS (2004) follow two
steps. First, they compute the AQ measures based on the Dechow-Dichev (2002) model
augmented by two Jones model variables, change in revenues and PPE, and estimate the
following cross-sectional regression for each industry group:
, 0, 1, , 1 2, , 3, , 1 4, , 5, ,j t t t j t t j t t j t t j t t j t j tTCA CFO CFO CFO REV PPE ,θ θ θ θ θ θ− += + + + + ∆ + +υ
(1)
4 Dividend initiations and omissions are not included in our sample as a result of these filters. 5 We thank Jennifer Francis and her co-authors for sharing the AQ factor returns data for this project. 6 Firms with poorer (better) AQ have higher (lower) information risk and are expected to earn higher (lower) returns to compensate for the risk.
12
where TCA is total current accruals, CFO is cash flow from operations, ∆REV is change
in revenue, and PPE is the level of PPE, all scaled by average total assets. The time-
series standard deviation of the residuals, ˆ( )tσ υ , calculated over t-4 to t, is the AQ
measure. Larger AQ measures indicate poorer mapping of accruals into cash flows and
higher information risk. Second, FLOS rank firms each month into quintiles based on
their most recent AQ value, calculate the average monthly excess returns for each quintile
for each period from April 1971 to March 2002, and form the information risk factor
mimicking returns by taking the difference between the monthly excess returns of the top
two AQ quintiles and the bottom two AQ quintiles.
3.3 Changes in information risks surrounding dividend changes
We test for a change in the information risk of dividend change firms using the
Fama-French (1993) multi-factor asset pricing model. We estimate the following Fama-
French three-factor model after incorporating the information risk factor returns from
FLOS (2004) as follows:
, ,
, , , , , , , , , ,
, , , ,
( ) ( )
(2)
j t f t
j j i t j m t f t j j t m t f t j t j i t t
j t j j t t j t j j t t t
R R
DT R R DT R R s SMB s DT SMB
h HML h DT HML e IR e DT IR e
α α β β∆ ∆ ∆
∆ ∆
−
= + + − + − + +
+ + + + +
We estimate this model from month t*-36 to month t*+36 (73 monthly
observations) surrounding the month of dividend announcement (t*= month 0) for each
dividend-changing firm. DTi,t is a dummy variable coded as 1 for months t >= t*, and 0
otherwise. Rj,t is the monthly stock returns for firm j, Rf,t is the monthly risk-free rates,
and Rm,t is the monthly return on the NYSE-AMEX-NASDAQ value-weighted market
portfolio. SMBt and HMLt are the size and book to market factors, respectively. IRt is
our information-risk factor from FLOS (2004).
13
In equation (2) αj represents the monthly risk-adjusted abnormal return of firm j
before the dividend announcement, and αj,∆ represents the incremental or the change in
risk-adjusted abnormal return after the dividend change announcement. βj, sj, hj, and ej
are factor loadings of firm j with respect to Rmt-Rft, SMB, HML, and the IR factors in the
three years before the dividend change announcement. βj,∆, sj,∆, hj,∆ , and ej,∆ represent
the changes in the factor loadings on the risk factors after the dividend announcement.
Our focus is on the factor loading ej,∆. We predict ej,∆ to be negative for dividend increase
firms and positive for dividend decrease firms. That is, information risk decreases
(increases) for firms that increase (decrease) dividends. For each sample firm with a
dividend change we estimate model (2) over the 73 months and report the average of the
coefficient estimates of these time-series models.
4. Empirical results
4.1 Descriptive statistics on dividend-paying firms
We begin our empirical analysis by presenting descriptive statistics on firm
characteristics of dividend change firms and by comparing the factor loadings of
dividend-paying versus non-dividend-paying firms. We expect dividend-paying firms to
exhibit less information risk than non-dividend-paying firms. Brav et al. (2003) report
that dividend decisions are made very conservatively, as reflected in executives’ extreme
reluctance to start issuing dividends and their extreme reluctance to reduce dividends
once they start paying dividends to shareholders. As a result, dividend-paying firms have
14
more sustainable and stable earnings streams than non-dividend-paying firms and less
volatility in earnings.7
Table 1 tabulates the percentage dividend change, firm size in terms of total assets
and market value of equity, and market to book ratios, fiscal year buy-and-hold returns,
annual sales growth, return on assets, and change in return on assets for both dividend
increasing and dividend decreasing firms, all measured at the beginning of the year. The
mean (median) dividend increase per share in our sample is 30% (20%), whereas the
mean (median) dividend decrease per share is 46% (50%).8 Dividend-increasing firms
appear to be smaller than dividend-decreasing firms both in terms of total assets ($2.5
billon vs. $3.6 billion) and market value of equity ($1.1 billion vs. $1.2 billion).
Dividend-increasing firms also appear to have higher market-to-book ratio, higher sales
growth, higher stock return and return on assets than dividend-decreasing firms at the
beginning of the dividend change year. The mean and median differences between
dividend increasing and dividend decreasing firms on these variables are all significant at
the 1% level.
4.2 Factor loadings of dividend-paying versus non-dividend paying firms
To better understand the information risk betas of dividend-paying firms, we compare
the factor loadings of dividend-paying and non-dividend-paying firms. We also examine
the different earnings persistence of these two sets of firms. We create two portfolios of
firms: (1) firms that never pay dividends over our sampling period (April 1971 to March
7 Brav et al. (2003) document that among firms that do not currently pay out, 70% of the executives indicate that they never plan to issue dividends, and that the most important factor influencing the decision to eventually issue dividends is sustainable increases in earnings. Among firms that currently pay dividends, “executives consider the continuation of the existing level of dividends as (nearly) untouchable,” and “the preservation of dividends equal to, and in some cases more important than, investment decisions.” 8 These % changes are very close to those reported in Grullon et al. (2002) on a sample including only NYSE and AMEX firms.
15
2002), and (2) firms that pay dividends for at least 5 contiguous years (20 contiguous
quarters) from April 1971 to March 2002. We then calculate the equally-weighted
portfolio excess returns, , ,p t f tR R− , for each of the two portfolios for each month (a total
of 372 months). The following multi-factor model is estimated for each portfolio:
, , , ,( )p t f t p p m t f t p t p t p t t (3 )R R R R s SMB h HML e IR e aα β− = + − + + + +
where Rp,t is the equally weighted portfolio monthly returns, t goes from April 1971 to
March 2002, and everything else is defined as in (2). To test for differences in the factor
loadings of dividend-paying firms and non-dividend paying firms, we stack the two
portfolios and estimate the following regression:
, , , , , ,( ) *(p t f t p pdiv p m t f t pdiv m t f t )R R D R R D R Rα α β β− = + + − + −
* * *p t pdiv t p t pdiv t p t pdiv t ts SMB s D SMB h HML h D HML e IR e D IR e+ + + + + + +
(3b)
where D is an indicator variable for dividend-paying firms, , , ,pdiv pdiv pdiv pdivs h eβ are the
differences in factor loadings of dividend-paying firms on , ,( m t f t )R R− , SMB, HML, and
IR, respectively, and pdivα represents the differences in intercepts.
For the earnings persistence test, we estimate the following regression of earnings
at time t+1 on earnings at time t, for dividend-paying and non-dividend-paying firms,
respectively:
, 1 , ,i t i i i t i tE Eα λ ε+ = + + (3c)
We test the difference in earnings persistence between dividend-paying and non-
dividend-paying firms using the following regression:
, 1 , , , , ,( )i t i pdiv i t i i t pdiv i t i t i tE D E D Eα α λ λ+ = + + + + ε (3d)
16
where Ei,t is operating income scaled by total assets, and D is a dummy variable coded as
1 for dividend-paying firms, zero otherwise. We expect to find earnings to be more
persistent for dividend-paying firms versus non-dividend-paying firms, as has been
documented by Skinner (2004). More persistent earnings lead to less uncertainty
surrounding firms’ future earnings realizations, which would in turn lead to more
information precision and less information risk. Thus a finding of a positive pdivλ is
consistent with dividend-paying firms having less information risk.
Results reported in Panel A of Table 2 indicate that dividend-paying and non-
dividend-paying firms have distinct risk characteristics. Dividend-paying firms have
higher market and HML betas, and smaller size betas, and the differences ( pdivβ ,
, ) between the loadings are significant at 5% or higher. More importantly,
dividend-paying firms load negatively on the information risk factor whereas the non-
dividend-paying firms load positively on the information risk factor, and the difference
between the loadings ( ) is significant at 1% level. This suggests that dividend-
paying firms are less risky in terms of information risk than non-dividend paying firms.
The results in Panel B corroborate those in Panel A: dividend-paying firms have higher
earnings persistence than non-dividend paying firms, as
pdivs pdivh
pdive
pdivλ is significantly positive at
1% level. The results also show that dividend-paying firms earn significant positive
alphas (mean of 0.16% per month) whereas non-dividend-paying firms earn significant
negative alphas (mean of -0.26% per month). Thus dividend-paying firms appear to
perform better than non-dividend-paying firms. Overall, the results in Table 2 are
17
consistent with our expectation that dividend-paying firms have lower information risk
than non-dividend-paying firms.
4.3 Changes in information risk surrounding dividend change announcements
Table 3 reports the frequency of dividend changes, industry distribution, and the
magnitude of dividend changes for our sample firms. Panel A shows that of the 2,981
unique firms, 860 (29%) report only one dividend change during the 25-year sampling
period, 1,285 firms (43%) have 2~4 dividend changes, 752 (25%) have 5~10 dividend
changes, and the remaining 84 firms (3%) announce 11~37 dividend changes. This
frequency distribution suggests that some of our dividend changes may be overlapping
during the 73 months estimation period. Thus, we estimate model (2) on all 10,597 firm-
announcement observations and also on a sub-sample of 1,946 dividend changes that are
not overlapping in the 73-month estimation window.
Panel B of Table 3 tabulates the industry distribution of our sample firms by the
sign of dividend change. Manufacturing firms dominate in both the dividend increases
(37.8%) and dividend decreases (31.4%), followed by financial firms (23.8% and 28.2%).
Overall, the industry distribution of firms across dividend increases and decreases are
similar, except for utilities: only 3.2% of dividend increase firms are utilities firms,
versus 11.9% of dividend decrease firms. Panel C reports the magnitude of dividend
changes: the majority of dividend increases fall between 12.5% to 25%, whereas the
majority of dividend decreases fall between 25% to 50%.
Table 4 reports the results testing our central prediction using equation (2) and
tabulates the mean coefficients of the estimation.9 Panel A uses all 10,597 dividend
change announcements, and Panel B restricts the sample to the non-overlapping 1,946 9 Following Grullon et al. (2002) we winsorize the coefficients at the first and ninety-ninth percentiles.
18
dividend-announcements. For dividend-increasing observations we present both the
results on all increases and the results on four dividend-increase quartiles so as to be
comparable to Grullon et al. (2002).
The results on the alphas and market and size betas are largely consistent with those
reported by Grullon et al. (2002).10 Several patterns are noticeable. First, the alphas
indicate that dividend-increasing firms appear to be doing better in the 36 months before
the change announcement, earning a monthly average of 0.86% (0.72% in Panel B) risk-
adjusted returns, whereas dividend-decreasing firms earn a negative monthly average
return of -0.25% (-0.56%in Panel B). Second, larger dividend increases are generally
associated with greater decreases in market beta, size beta, and IR beta, suggesting firms
with larger dividend increases experience greater decrease in market risk, small firm risk,
and information risk. The changes in book-to-market beta are non-linear across dividend
increase quartiles, and not significant in the overall sample of 9,110 dividend increases.
The changes in IR beta are significantly negative in dividend-increase quartiles 2 and 4,
resulting in a significantly negative ej,∆ for all dividend-increase observations. Third, and
most importantly, in both Panels A and B dividend-decreasing firms have a positive
ej,∆, consistent with our prediction that dividend-decreasing firms exhibit an increase in
their information risk in the months after the change announcement. For dividend-
decreasing firms the changes in market and B/M betas are also significantly positive,
whereas the changes in size beta are positive but not significant. Taken together the
results suggest that dividend-decreasing firms experience an overall increase in their
10 Our sample consists of firms on NYSE, AMEX, and NASDAQ that announced dividend changes over the period April 1974 to March 1999. When we restrict our sample to the Grullon et al. (2002) NYSE and AMEX firms over the period 1967 to 1993 we are able to replicate their results very closely.
19
systematic risks, including information risks, in the three years after their decrease
announcements.
While the empirical results on dividend-decreasing firms largely support our
prediction that these firms exhibit an increase in their information risk, in addition to
other systematic risks, in the years after the dividend decrease, the results on dividend-
increasing firms are sensitive to the samples used. The results in Panel A, based on all
10,597 dividend announcements, generally support the hypothesis that dividend-
increasing firms experience decreases in their systematic risks, including information
risks, after the change. This is evidenced in the significantly negative change in market,
size, and information risk betas. However, the results based on the 1,946 non-
overlapping dividend announcements reported in Panel B show that only the changes in
market and size betas remain negative while the change in information risk beta becomes
positive, though not significant. In addition, the change in the B/M beta is significantly
positive. Taken together, our asset-pricing regression results provide support for the
change in information risks for dividend-decreasing firms, but not for dividend-
increasing firms.
Given that our evidence on dividend-increasing firms is sensitive to the two
different samples used, to gain a better understanding of the economic circumstances
surrounding dividend changes we next turn our attention to an examination of the
information and operating characteristics of firms around the dividend change
announcements. We focus our attention on the non-overlapping sample because this is a
‘cleaner’ sample and because our prediction on the information risk change for dividend
increase firms does not hold in this sample.
20
4.4 Analysis of information characteristics surrounding dividend change
announcements
If firms’ information risks change after dividend changes, then we should be able
to observe systematic changes in firms’ information characteristics as well. We examine
the dispersion of fiscal year 1 analyst earnings forecasts (DISPFY1) and the dispersion of
analysts’ long-term growth forecasts (DISPLTG), measured as the standard deviation of
forecasts scaled by the absolute value of the mean forecast. Analyst forecast data are
obtained from the I/B/E/S summary history file. We expect greater forecast dispersion to
indicate greater uncertainty related to firms’ future earnings and growth prospects. In
addition, we also examine the standard deviation of stock returns, the standard deviation
of operating cash flows (ST_CFO), and standard deviations of sales (SD_Sales), and the
standard deviation of ROA (SD_ROA): higher standard deviation reflects higher
volatility and greater uncertainty in the underlying earnings streams.
We compare the information characteristics of dividend change firms three years
before the dividend change (-3 to -1) to three years after the dividend change (+1 to +3).
Year 0 is the year of dividend change announcements. We conduct our empirical
analysis on the 1,946 non-overlapping observations and test the mean (median)
differences using paired two-sample t tests (Wilcoxon signed rank tests). The results of
this analysis are reported in Table 5.
The results on analyst forecast dispersion indicate that both dispersion measures
decrease (with the exception of the mean change in DISPFY1, which is positive but not
significant) for dividend-increase firms, suggesting the uncertainty surrounding future
earnings and growth prospects for dividend-increasing firms decrease after the dividend
21
increase announcements. In contrast, and as expected, for dividend-decreasing firms,
both analyst forecast dispersion measures increase in the three years following dividend
decreases, suggesting an increase in uncertainty with regard to future earnings and long-
term growth for firms that decrease dividends. In addition, stock return volatility
decreases for dividend-increase firms and increases for dividend-decrease firms in the
years following the dividend change. Dividend-increase firms also experience decrease
in cash flow volatility, ROA volatility, and sales volatility. The changes in these three
volatility measures for dividend-decrease firms are not significant. Overall, our evidence
on the information characteristics of firms is consistent with a reduction in information
uncertainty for dividend-increase firms, and with an increase in information uncertainty
for dividend-decrease firms.
4.5 Analysis of changes in profitability, payout, capital expenditure, and cash balance
surrounding dividend change announcements
Our analysis above indicates that dividend change firms experience changes in
information characteristics in the years following dividend changes. In this section we
examine the Grullon et al. (2002) operating characteristics variables to see if their results
hold in our sub-sample. The variables we examine include earnings growth rates, the
dividend payout ratio, capital expenditure, and cash levels of dividend change firms.11
We first examine the change in earnings growth rate of dividend change firms. If
the prediction of traditional signaling models holds, then we would expect to find
increases (decreases) in profitability in the years after dividend increases (decreases).
11 Grullon et al. (2002) conduct their tests on all dividend announcements, including over-lapping dividend announcements.
22
Alternatively, Benartzi et al. (1997) suggests that instead of informing us about future
earnings, dividend increases confirm a permanent increase in past earnings.
The profitability measure we use is ∆ROA, measured as the difference between
two adjacent years’ ROA. For the non-overlapping dividend-increase observations,
∆ROA decrease in the 3 years after the change (the median firm experiences a drop of
0.64% in ∆ROA). For the dividend-decrease observations, on the other hand, ∆ROA
increases after the change (the median firm exhibits an increase of 0.64% in ∆ROA).
Thus the our non-overlapping sample results on firms’ growth rate are qualitatively very
similar to those reported by Grullon et al. (2002) on all dividend change events, including
overlapping events. Taken together, our results suggest that dividend-increase firms
experience a drop in earnings growth rate after the dividend change and the dividend-
decrease firms’ earnings grow at a higher rate after dividend decreases. These results are
contrary to the central prediction of the traditional dividend signaling models, and the
results on dividend-increasing firms are more consistent with the ‘confirming-existing-
earnings’ explanation advanced in Benartzi et al. (1997).
We next examine the behavior of dividend payouts, capital expenditures, and cash
balances. The dividend payout ratio is defined as the ratio of common dividends to net
income before extraordinary items, capital expenditure is calculated as the ratio of capital
expenditures to total assets, and cash balance is measured as the ratio of cash and short-
term investments to total assets. The results reported in Table 6 are largely consistent
with the free cash flow hypothesis: in the three years after the dividend increase, on
average payout ratios increase 11%, cash balances decrease by 1%, and capital
expenditures remain the same when compared to 3 years before the change and decline
23
by 6% when compared to the year of the change. For dividend decrease firms we
observe decreases in payout ratios of 21% from the year of the change to 3 years after the
dividend change (significant at the 1% level). Capital expenditures for these firms also
decrease, and their cash balances remain unchanged. Thus, while the results are
consistent with dividend-increasing firms having less investment need and hence more
free cash flow, they also indicate less investment needs for dividend-decreasing firms.
5. Conclusion
Building upon Grullon et al. (2002), who document that firms experience
decreasing (increasing) market, size, and book-to-market betas after they increase
(decrease) dividends, we investigate whether firms’ information risks decrease after
dividend increases, and increase after dividend decreases. Using the Fama-French three-
factor model augmented by an Information Risk factor, we find that, as predicted, the
dividend-decreasing firms’ information risk beta increases in the three years after the
dividend decrease announcements. We also find that for dividend-decrease firms analyst
forecast dispersion and stock return volatility become larger, suggesting greater
information uncertainty concerning the future prospects of dividend-decrease firms.
However, in our multi-factor return model analysis we do not find consistent evidence
supporting the hypothesis that dividend-increasing firms experience a decrease in
information risks in the years after the dividend change. Further analysis shows that
analyst forecast dispersion and stock return volatility become smaller after dividend
increases. Our analysis on firms’ operating characteristics surrounding dividend changes
24
are consistent with the existing empirical findings that dividend increases (decreases) are
not associated with future increases (decreases) in profitability.
In their recent survey of financial executives, Brav et al. (2003) report that
managers rarely view dividends as a self-imposed costly tool to signal firms’ ability and
separate a company from its competitors. Skinner (2004) suggests that while dividends
may have been a viable signaling tool in the early part of last century when managers had
few means of communicating information other than using the financial statement
themselves, changes in the ways managers communicate with the outside world may
have rendered dividend signaling obsolete. Our results, coupled with the consistent lack
of empirical support for the predictions of traditional dividend signaling models, suggest
that the information contained in dividends is not about future earnings, but more about
changes in firms’ systematic risks, including information risks, and changes in the
uncertainty regarding firms’ future earnings and growth prospects.
25
References:
Aboody, D., J. Hughes, and J. Liu. 2003. Earnings quality, insider trading, and cost of capital. UCLA working paper. Allen, F., R. Michaely, 2002. Payout policy. University of Pennsylvania and Cornell University working paper. Benartzi, S., R. Michaely, and R. Thaler. 1997. Do changes in dividends signal the future or the past? Journal of Finance 52 (3): 1007-1034. Bhattacharya, S., 1979. Imperfect information, dividend policy, and ‘the bird in the hand’ fallacy, Bell Journal of Economics, 10(1): 259-270. Brav, A., J.R. Graham, C. R. Harvey, and R. Michaely, 2003. Payout policy in the 21st century. Duke University, National Bureau of Economic Research, Cornell University, the Inter-Disciplinary Centre Israel, working paper. Dechow, P., I. Dichev, 2002. The quality of accruals and earnings: the role of accrual estimation errors. The Accounting Review 77 (supplement): 35-59. Grullon, G., R. Michaely and B. Swaminathan, 2002. Are dividend changes a sign of firm maturity? Journal of Business 75 (3): 387-424. Easely, D., M. O’Hara, 2001. Information and the cost of capital. Cornell University working paper. Fama, E., K. French, 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33: 3-56. Francis, J., R. LaFond, P. Olsson, and K. Schipper. 2004. The market pricing of accruals quality. Journal of Accounting & Economics, forthcoming. Healy, P. M., and K. G. Palepu, 1988. Earnings information conveyed by dividend initiations and omissions. Journal of Financial Economics 21 (September): 149-75. John, K., J. Williams, 1985. Dividends, dilution, and taxes: a signaling equilibrium. Journal of Finance 40 (4): 1053-1070. Leuz, C, and R. Verrecchia, 2004. Firms’ capital allocation choices, information quality, and the cost of capital. University of Pennsylvania working paper. Miller, M., K. Rock, 1985. Dividend policy under asymmetric information. Journal of Finance 40 (4): 1030-1051.
26
Miller, M., F. Modigliani, 1961. Dividend policy, growth and the valuation of shares. Journal of Business 34: 411-433. O’Hara, M, 2003. Presidential address: liquidity and price recovery. Journal of Finance 58: 1335-1354. Skinner, D., 2004. What do dividends tell us about earnings quality? University of Michigan, working paper.
27
Table 1 Descriptive Statistics on Dividend Change Firms
Dividend Increase (1) Dividend Decrease (2) Differences (2) – (1)
Mean Median
Standard Deviation N
Mean Median
Standard Deviation N Mean
Median
∆Div 29.66 20.00 30.33 9,110 -46.07 -50.00 16.62 1,487 -75.73* -70.00*
TA 2,465.30
282.11 8,312.92 7,652 3,621.03
479.53 10,143.51 1,182 1,155.73*
1,97.42*
MV 1,114.37
150.05 3,350.98 7,830 1,246.60
156.22 3,675.17 1,224 132.23
-133.56
MTB 1.8700
1.4360 1.4643 7,619 1.5305
1.0746 1.3877 1,174 -0.3395*
-0.3614*
BHR 18.22
12.77 42.76 7,802 1.63
-4.92 38.55 1,219 -16.59*
-4.21*
Growth 17.31
14.60 17.57 7,503 9.04
7.47 19.73 1,163 -8.27*
-7.13*
ROA 16.96
17.03 9.46 7,528 11.99
10.53 8.42 1,157 -4.97*
-6.50*
∆ROA 0.90
0.46 3.80 7,412 -1.40
-0.58 4.69 1,146 -2.30*
-1.04*
Notes: ∆Div: Dividend change in percentage between adjacent quarters when dividend change is between 12.5 percent and 500 percent TA: Assets of firms in millions, as of beginning of the dividend change year MV: Market value of equity, in millions, as of beginning of the dividend change year MTB: Ratio of market value of equity to book value of equity, as of beginning of the dividend change year BHR: Fiscal year buy-and-hold annual returns in percentage, as of beginning of the dividend change year Growth: Annual sales growth in percentage, as of beginning of the dividend change year ROA: Returns on assets in percentage measured as operating income before depreciation deflated by total assets, as of beginning of the dividend change year ∆ROA: Change in ROA in percentage measured as the difference between current and prior period ROA, as of beginning of the dividend change year. *, **, *** Significant at the 1%, 5%, and 10% levels, respectively, two-tailed. Mean (median) differences are tested using two-sample t (Wilcoxon signed rank) tests.
28
Table 2 Dividend Paying versus Non-Dividend Paying Firms
Panel A: Factor Loadings
pα
pβ ps
ph pe
Non-dividend Portfolio -0.2618* 0.9223* 0.6050* 0.2623* 0.5210* Dividend Portfolio 0.1587** 0.9777* 0.4950* 0.4979* -0.1090* Differences
pdivα pdivβ pdivs pdivh pdive
0.4205* 0.0554** -0.1100** 0.2356* -0.6300*
Panel B: Earnings Persistence
iα iλ
Non-dividend Firms -0.0067* 0.7040* Dividend Firms 0.0145* 0.8761* Differences
pdivα pdivλ
0.0212* 0.1721*
Notes: Non-dividend paying firms are firms that have never paid any dividends over our sampling period of April 1971 to March 2002. Dividend paying firms are firms that pay dividend for at least 5 contiguous years (20 quarters) over our sampling period. Factor loading Models:
, , , ,( )p t f t p p m t f t p t p t p tR R R R s SMB h HML e IR etα β− = + − + + + +
, , , , , ,( ) * ( )p t f t p pdiv p m t f t pdiv m t f tR R D R R D R Rα α β β− = + + − + − * *p t pdiv t p t pdiv t p t pdiv t* ts SMB s D SMB h HML h D HML e IR e D IR e+ + + + + + + D is dummy variable coded as 1 for dividend paying firms, zero otherwise. Rp,t is the equally weighted portfolio monthly returns. Rf,t is the monthly risk-free rates. Rm,t is the monthly return on the NYSE-AMEX-NASDAQ value-weighted market portfolio. SMBt and HMLt are the size and book to market factors, respectively. IRt is the information-risk factor. Earnings Persistence Models:
, 1 , ,i t i i i t i tE Eα λ ε+ = + +
, 1 , , , , ,( )i t i pdiv i t i i t pdiv i t i t i tE D E D Eα α λ λ+ = + + + + ε D is dummy variable coded as 1 for dividend paying firms, zero otherwise. Ei,t is operating income before depreciation deflated by total assets. *, **, *** Significant at the 1%, 5%, and 10% levels, respectively, two-tailed.
29
30
Table 3 Characteristics of Dividend Changes
Panel A: Frequency of Dividend Changes
Number of Divided Change Events Number of Unique Firms
1 860 2 584 3 386 4 315 5 231 6 187 7 124 8 80 9 77
10 53 11 25 12 17 13 15 14 6 15 6
> 15 15
Total 2,981 Panel B: Industrial Classification
Industry SIC Number of Unique Firms Dividend Increase Dividend Decrease
Agriculture, Forestry and Mining 01-09 390 15.1% 51 12.6% Mining 10-14 59 2.3 % 9 2.2%
Construction 15-17 21 0.8% 4 1.0% Manufacturing 20-39 975 37.8% 127 31.4%
Transportation and Communication 40-48 77 3.0% 10 2.5% Utilities 49 83 3.2% 48 11.9%
Wholesale 50-51 70 2.7% 9 2.2% Retail 52-59 160 6.3% 17 4.3%
Financial 60-69 614 23.8% 114 28.2% Services 70-89 128 5.0% 14 3.5%
Public Administration 90-97 0 0.0% 0 0.0% Others 98-99 0 0.0% 1 0.2%
Total 2,577 404
Panel C: Magnitude of Dividend Changes
Absolute Percentage Change Number of Firm-event Pairs Dividend Increase Dividend Decrease
12.5 to 25 5,688 183 25 to 50 2,359 916
50 to 100 692 388 100 to 250 339 0 250 to 500 32 0
Total 9,110 1,487
Table 4 Changes in Factor Loadings Surrounding Dividend Change Announcements
Panel A: All Dividend Announcement (n=10,597) All Dividend
Increases Dividend Increase
Quartile 1
Dividend Increase
Quartile 2
Dividend Increase
Quartile 3
Dividend Increase
Quartile 4
DividendDecreases
Predicted Sign
N=9,110 N=2,374 N=2,293 N=2,166 N=2,277 PredictedSign
N=1,487
Alpha
jα− 0.8618* 0.5857* 0.8089* 0.9392* 1.1358* -0.2493*
,jα ∆ n/a -0.7219* -0.4705* -0.6880* -0.7704* -0.9784* n/a -0.1325**
Market factor
jβ− 1.0212* 0.9783* 1.0154* 1.0460* 1.0479* 0.8860*
,jβ ∆ − -0.0457* -0.0144 -0.0593* -0.0653* -0.0440* + 0.0973*
Size factor
js− 0.6764* 0.5509* 0.6494* 0.7246* 0.7896* 0.6003*
,js ∆ − -0.0733* -0.0087 -0.0501*** -0.1142* -0.1250* + 0.0056
Book to market factor
jh − 0.1937* 0.1642* 0.2041* 0.1616* 0.2401* 0.3852*
,jh ∆ − 0.0061 0.0723* -0.0102 0.0184 -0.0554** + 0.1332*
Information risk factor
je − -0.0226* -0.0683* -0.0235** -0.0302** 0.0 **336 -0 4.016
,je ∆ − -0.0257* -0.0141 -0.0365** -0.0136 -0.0373** + 0.1028*
31
32
Table 4 (continued) Panel B: Non-overlapping Dividend Announcements (n=1,946)
All Dividend Increases
Dividend Increase
Quartile 1
Dividend Increase
Quartile 2
Dividend Increase
Quartile 3
Dividend Increase
Quartile 4
DividendDecreases
Predicted Sign
N=1,458 N=365 N=366 N=379 N=348 PredictedSign
N=488
Alpha
jα− 0.7169* 0.4966* 0.6432* 0.7458* 1.0087* -0.5617*
,jα ∆ n/a -0.8152* -0.5278* -0.6884* -0.8673* -1.2157* n/a -0.0097
Market factor
jβ− 0.9654* 0.9199* 0.9481* 0.9743* 1.0308* 0.8576*
,jβ ∆ − -0.0460* -0.0290 -0.0411 -0.0534 -0.0604 + 0.1250*
Size factor
js− 0.6936* 0.4783* 0.6291* 0.7820* 0.8948* 0.5542*
,js ∆ − -0.0481 0.0870 -0.0311 -0.1277*** -0.1203 + 0.0294
Book to market factor
jh − 0.2431* 0.2229* 0.2793* 0.2163* 0.2471* 0.4336*
,jh ∆ − 0.0871* 0.1232* 0.0549 0.1457* 0.0269 + 0.1629*
Information risk factor
je − -0 9.011 -0.0846* -0 1.000 -0 3.020 0.0605 -0 6.035
,je ∆ − 0.0141 -0.0031 -0.0251 0.0336 -0.0687 + 0.1595*
Notes: Model:
, , , , , , , , , ,( ) ( )j t f t j j i t j m t f t j j t m t f tR R DT R R DT R Rα α β β∆ ∆− = + + − + − , , , , , ,j t j i t t j t j j t t j t j j t t ts SMB s DT SMB h HML h DT HML e IR e DT IR e∆ ∆ ∆+ + + + + + + DTi,t is a dummy variable coded as 1 for months t >= t*, and 0 otherwise. t* is the month of dividend change announcement. Rj,t is the monthly stock returns for firm j, Rf,t is the monthly risk-free rates, and Rm,t is the monthly return on the NYSE-AMEX-NASDAQ value-weighted market portfolio. SMBt and HMLt are the size and book to market factors, respectively. IRt is the information-risk factor.
, **, *** Significant at the 1%, 5% and 10% levels, respectively, two tailed. *
33
Table 5 Information Characteristics
Surrounding Dividend Change Announcements
Dividend Increases Dividend Decreases (A)
-3yr to -1yr
(B) +1yr to
+3yr
(B)-(A) (A) -3yr to
-1yr
(B) +1yr to
+3yr
(B)-(A)
DISPFYI Mean 0.0955* 0.1243* 0.0257 0.2007* 0.3432* 0.1519* Median 0.0549* 0.0545* -0.0033** 0.0946* 0.0737* 0.0726* N 494 598 474 219 217 205 DISPLTG Mean 0.2752 0.2423 -0.0352** 0.4040 0.4688 0.0801 Median 0.2265 0.2104 -0.0162** 0.3116 0.3296 0.0582*** N 294 337 273 156 162 143 SD_RET Mean 10.2540* 9.8064* -0.4476* 8.7349* 11.3480* 2.6131*
Median 9.7651* 9.2765* -0.3076* 8.1471* 10.2485* 1.7198* N 1,458 1,458 1,458 488 488 488
SD_CFO Mean 0.0313* 0.0290* -0.0019* 0.0315* 0.0294* -0.0018 Median 0.0231* 0.0228* -0.0014* 0.0232* 0.0213* -0.0005 N 308 425 307 139 170 138 SD_ROA Mean 0.0139* 0.0118* -0.0021* 0.0122* 0.0115* -0.0009 Median 0.0093* 0.0091* -0.0003* 0.0094* 0.0091* -0.0003 N 910 979 854 336 356 324 SD_Sales Mean 0.0417* 0.0413* -0.0001 0.0409* 0.0388* -0.0022
Median 0.0301* 0.0285* -0.0003** 0.0282* 0.0271* -0.0005 N 1,063 1,149 1,062 377 398 376
Notes: DISPFY1 :dispersion of analyst forecasts for fiscal year 1, measured as the standard deviation of forecasts for FY1 scaled by the absolute value of the mean forecasts for FY1. DISPLTG: dispersion of analysts’ long term growth forecasts, standard deviation of long term growth forecasts, scaled by the absolute value of the mean long term growth forecasts SD_RET: Standard deviation of monthly stock returns less the risk free rate SD_CFO: Standard deviation of cash flows from operations deflated by total assets, measured over quarterly data SD_ROA: Standard deviation of ROA, measured over quarterly data SD_Sales: Standard deviation of total sales deflated by total assets, measured over quarterly data. *, **, *** Significant at the 1%, 5% and 10% levels, respectively, two-tailed. Mean (median) differences are tested using paired two-sample t (Wilcoxon signed rank) tests.
Table 6 Operating Characteristics
Surrounding Dividend Changes
Dividend Increase
Dividend Decrease
(A)
-3yr to –1yr
(B) 0
(C) +1yr to
+3yr
(C) – (A)
(C) – (B)
(A) -3yr to
–1yr
(B) 0
(C) +1yr to
+3yr
(C) – (A)
(C) – (B)
∆ROA Mean 0.0049* -0.0004 -0.0089* -0.0141* -0.0087* -0.0093* -0.0174* 0.0026** 0.0120* 0.0199*
Median 0.0021* 0.0008* -0.0028* -0.0063* -0.0040* -0.0035* -0.0051* 0.0014* 0.0064* 0.0040* N 1,176 1,182 1,230 1,171 1,181 390 390 400 388 390
PAYOUT Mean 0.3381* 0.3459* 0.4497* 0.1117* 0.1138* 0.7108* 0.7544* 0.4995* -0.1846* -0.1993* Median 0.2780* 0.2682* 0.3451* 0.0576* 0.0509* 0.5613* 0.5505* 0.3942* -0.1022* -0.0155* N 1,203 1,188 1,229 1,174 1,163 394 274 380 362 264
CAPEX Mean 0.0710* 0.0781* 0.0718* 0.0009 -0.0059* 0.0672* 0.0567* 0.0522* -0.0145* -0.0044* Median 0.0607* 0.0655* 0.0614* 0.0012** -0.0010* 0.0603* 0.0458* 0.0441* -0.0104* -0.0010*** N 1,023 1,015 1,030 1,014 1,013 336 333 335 330 332
CashSTI Mean 0.0930* 0.0893* 0.0839* -0.0097* -0.0057* 0.0765* 0.0716* 0.0728* -0.0037 0.0011 Median 0.0630* 0.0545* 0.0518* -0.0061* -0.0019* 0.0460* 0.0365* 0.0417* -0.0010 0.0013** N 1,198 1,211 1,251 1,196 1,211 395 398 415 393 398
Notes: All variables measured using annual data. ROA: Returns on assets measured as operating income before depreciation deflated by total assets ∆ROA: Change in ROA measured as the difference between current and prior period ROA PAYOUT: Ratio of common dividends to net income before extraordinary items CAPEX: Ratio of capital expenditure to total assets CashSTI: Ratio of cash and short-term investments to total assets *, **, *** Significant at the 1%, 5% and 10% levels, respectively, two-tailed. Mean (median) differences are tested using paired two-sample t (Wilcoxon signed rank) tests.
34