Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Disappearing Dividends: A Rational Explanation and Implications
Min Maunga Vikas Mehrotrab
Abstract
Fama and French (2001) document a startling and secular decline in both the proportion of firms paying dividends and the residual propensity to pay dividends over the preceding two decades. We offer a simple explanation for this phenomenon based on two concurrent trends. First, we show that over this period the average information content of stock prices has increased, diminishing the relative usefulness of dividend-based signaling – we call this the supply side effect, and show that it is able to explain approximately 30% of the residual propensity to pay dividends over this time period. Second, we find that when firms are included in the S&P500 Index, their stock price response to dividend changes and initiations, dividend yields, and the propensity to pay dividends, decline significantly. Under the assumption that indexers are less concerned with idiosyncratic information, inclusion in the index suppresses their demand for dividend-based signaling – we call this the demand side effect. Our study provides a rational explanation for the “disappearing” dividend puzzle.
JEL Classification: G14, G35
a University of Saskatchewan, Edwards School of Business, Department of Finance and Management Science,
Saskatoon, SK S7N 5A7, Canada. Email: [email protected]. b University of Alberta, School of Business, Department of Finance and Statistical Analysis, Edmonton, AB T6G
2R6, Canada. Email: [email protected]
We thank the seminar and conference participants at the University of Saskatchewan and Financial Management
Association Annual Meeting 2011 for helpful comments. All errors are our own.
1
1. Introduction
Dividends represent a puzzle whether they disappear or stay. Fama and French (2001,
henceforth FF) document a large and secular decline in both the fraction of firms that
pay dividends, and in the propensity to pay dividends by all firms, over the last two
decades.1 In this paper, we point to two concurrent trends in an effort to explain the
declining popularity of dividends. The first is an overall increase in the information
content of stock prices over this period, reducing the relative efficacy of signaling firm-
specific information via dividends. The second trend is the tremendous growth in
indexing coinciding with this time period – to the extent indexers care less about
idiosyncratic information, we posit that the demand for such information, including the
demand for dividend-based signaling, would decline.
We measure stock price informativeness (SPI) by estimating market model
residuals as a proxy for idiosyncratic information, as in Roll (1988) and Morck, Yeung
and Yu (2001), and relate changes in the propensity to pay dividends to changes in SPI
under the premise that the relative efficacy of dividend-based signaling diminishes
when the external information environment improves. Firms respond by curbing
dividends as a costly and non-conservative means of signaling.
Our proxy for indexing effects is membership in the S&P500 index. We examine
what happens to dividend yields when firms are included in the S&P500 Index. We also
1 DeAngelo, DeAngelo and Skinner (2004) show that in the aggregate, the total payout by firms has not declined
over time; rather, they point to increasing concentration in earnings and dividends by a relatively smaller coterie
of large firms.
2
examine the differential response of stock prices to dividend changes when firms are in
the index vs. when they are not.
We are not the first to suggest that dividends have lost their dominant role in
signaling. Amihud and Li (2006) argue that dividends are disappearing because their
signaling role is diminished today. They base their argument on the declining dividend
response coefficients (DRCs) in the face of increasing institutional holdings, on the
premise that institutional shareholders have the scale to conduct private research, and
hence have little use for the arguably blunter signals provided by slow-adjusting
dividends. In equilibrium, dividend response coefficients decline as more private
research gets impounded into prices prior to dividend announcements. Firms oblige by
cutting back on dividends.
We take a slightly different tack and focus instead on dividend demand from
investors when firms are added to the S&P500 Index. An advantage of our setting is that
firms do not choose to be added to this index and hence the dividend change associated
with this event is a response to an exogenous event. By contrast, the Amihud and Li
(2006) setting may be associated with slow-moving changes in institutional ownership,
making it difficult to disentangle whether institutions gravitate to low dividend paying
stocks, or firms with high institutional ownership choose to reduce dividends.
Indexers have reduced incentive to invest in idiosyncratic research since they are
evaluated primarily on tracking error. Regardless of earnings surprises, indexers pursue
a mechanical investment strategy whose primary, often sole, purpose is to mimic the
index at hand. We exploit this to control for the demand side of idiosyncratic
3
information. Our main thesis goes through as long as firm-specific information demand
declines following the addition of a firm to a widely followed index.
We find that both dividend response coefficients as well as yields decline
significantly following index additions. Furthermore, we find that the coefficients are
declining with calendar year as in Amihud and Li (2006), but only for indexed stocks.
For non-index stocks, there is no systematic pattern of dividend response coefficients or
yields over time. These results are novel in that they tie dividends to their demand,
though not in the catering sense as in Baker and Wurgler (2004a, 2004b). Rather, firms
respond to a reduced demand for information signaling by cutting yields in the face of
shifting investor preferences. To the extent indexing has grown almost monotonically
over the last two decades, our results point to a hitherto overlooked facet of their
demand.
We begin our analysis by documenting that dividends in general are losing their
importance as a signaling device as indicated by the declining abnormal announcement
returns associated with dividend initiations. This is an extension of Amihud and Li (2006)
who showed that dividend increases and decreases are also accompanied by declining
abnormal returns. We attribute this decline to rising SPI as proxied for by the R-square
(R2) measure of Roll (1988) and Morck, Yeung, and Yu (2000). Our results show that
firms with lower firm-level R2s (higher SPI) are less likely to pay dividends, and that the
aggregate annual R2 explains the time series variation in the propensity to pay dividends.
This finding implies that changes in firms’ information environment have a significant
impact on the information content of dividend change announcements. For instance, we
4
find that firms that initiate dividends in periods when aggregate stock prices are less
informative earn higher abnormal returns.
By using a somewhat different set of controls and methodology, we also obtain
results that are consistent with the risk-based explanation of Hoberg and Prabhala
(2009, henceforth H-P). H-P add risk to a logit model for dividend payers, and find that
B-W’s catering hypothesis is no longer significant in explaining the time series variation
in the propensity to pay dividends. As in H-P, we find that the premium for dividend
payers vs. non-payers becomes insignificant after controlling for risk. However, SPI
variables remain both economically and statistically significant in explaining the
declining propensity to pay dividends.
Our main tests rely on changes in dividends associated with the S&P index
membership. We first document reduced dividend response coefficients in periods when
a firm is part of the S&P index, and repeat the finding for dividend yields. These results
are consistent with Amihud and Li (2006) who show a similar relation between
institutional ownership and dividends. Our results go further on two counts. First, we
show a plausible channel via which institutional indifference to dividends might work –
indexing as an investment strategy makes fewer demands on firm level information.
Second, our index membership result persists after controlling for institutional
ownership, indicating that it is indexing rather than any comparative advantage in
gathering information by institutions that is driving the reduced demand for dividends.
The remainder of the paper is organized as follow. Section 2 reviews the literature.
Section 3 describes our data and choice of variables. Section 4 discusses the results and
5
offers some robustness checks. Section 5 concludes.
2. Literature Review
The literature starts with M&M (1961) establishing the irrelevance of dividend policy in
perfect markets. Relaxing the perfect market assumptions generates important roles for
dividends such as controlling agency costs or information signaling. In the latter camp
are seminal articles by Ross (1977), Bhattacharya (1979), and Miller and Rock (1985),
advocating equilibriums where dividend changes signal information about firm value,
typically via information about a firm’s future earnings.2 In effect, dividends signal the
quality of earnings to shareholders (Nissim and Ziv, 2001).3
The agency theory of dividend policy suggests that dividends also serve as a
mechanism for reducing agency costs (see Jensen and Meckling, 1976; and Easterbrook,
1984). Jensen and Meckling (1976) argue that managers of firms can allocate resources
to activities that enhance their private benefits. Thus, dividend payouts may increase
firm value by reducing over-investment.4 In particular, the agency theory posits that
distributing resources in the form of dividends forces firms to return to capital markets
to raise additional funds for their investment needs, thereby subjecting firms to frequent
monitoring by outside stakeholders. In summary, dividends serve the dual purpose of
reducing agency costs of free cash flows and revealing inside information to outsiders.
2 See Asquith and Mullins (1983) and Healy and Palepu (1988).
3 Arguments regarding whether dividend increases (decreases) indeed signal future profit increases (decreases)
remain part of a heated debate. See, among several others, Benartzi, Michaely, and Thaler (1997), Allen and
Michaely (2003), and Grullon, Michaely, Benartzi, and Thaler (2005). 4 See Lang and Litzenberger (1989) for some empirical evidence.
6
The secular decline in dividends between 1978 and 1999 was brought to light by
an influential article by Fama and French (2001), who document a large decline in the
proportion of payers between 1978 and 1999, as well as in the propensity to pay based
on a logit model of dividend payers. Grullon and Michaely (2002) argue that, at the
aggregate level, firms have become more likely to substitute dividends with share
repurchases, and hence it makes more sense to examine payouts rather than dividends
alone. DeAngelo, DeAngelo and Skinner (2004) show that the aggregate level of real
dividends paid by dividend-paying firms has not declined over the period during which
firm-level dividends have disappeared. They document that real dividend payments
increased among the highest payers and that the observed decline in number is due to
omissions by the increase in listings of smaller firms that usually pay meager amount of
dividends.
Baker and Wurgler (2004a) and Baker and Wurgler (2004b) propose a ‘catering’
theory where firms cater to investor demands or sentiments when deciding whether to
initiate dividends. A prediction of the catering hypothesis is that the propensity to pay
dividends increases when dividend premium is high – this is supported by B-W. Hoberg
and Prabhala (2009) show that idiosyncratic risk explains a significant fraction of the
disappearing dividend puzzle. When risk is added to the dividend payer model, H-P find
that the propensity to pay is no longer related to dividend premium, inconsistent with
the catering hypothesis.5 The literature has not examined the precise channels through
which the propensity to pay has diminished over time.
5The terms idiosyncratic risk and firm-specific risk are used interchangeably in this paper.
7
3. Data
3.1 Variables from Prior Studies
We closely follow the variable construction proecdures as described in Fama and French
(2001), Baker and Wurgler (2004a, 2004b), and Hoberg Prablaha (2009) and exclude
utilities and financial firms. We also exclude firms with book equity below $250,000 or
assets below $500,000.
A firm is defined as a payer if it has positive dividends per share on the day before
the ex-dividend date. As in FF, we define the propensity to pay (PTP) as the difference
between the actual percentage of firms paying dividends in a given year and the
expected percentage, which is the average predicted probability from the given logit
model. M/B is the market-to-book ratio, defined as book assets minus book equity plus
market equity scaled by book assets. M/B is included in all regressions. The dividend
premium (Premium) is defined as the lagged value of the difference between average
M/B of dividend payers and nonpayers (in log-form, consistent with the definition in
BW). NYP is the NYSE size percentile, defined as the percent of firms that have the same
or smaller market capitalization (based on market equity value; stock price times shares
outstanding) than firm i in a given year. dAssets is the change in assets scaled by total
assets, and E/A is earnings (income before extraordinary items plus interest expense
plus income statement deferred taxes) scaled by total assets.
Hoberg and Prabhala (2009) include proxies for idiosyncratic and systematic risks
to show that catering loses its appeal once these risk measures are controlled for. A
firm’s idiosyncratic risk in a given year is the standard deviation of residuals from the
8
market model regression using daily returns for that calendar year. We impose the 200
minimum trading day requirements for calculation of idiosyncratic risk. To be consistent
with Hoberg and Prabhala (2009), firm volatilities are not averaged across all firms in a
given year since the main interest lies in examining idiosyncratic risk at the firm level.
3.2 Measures of the Information content of Stock Price
Our primary proxy for SPI is the R2 measure of Roll (1988). Following Morck et al.
(2000), R2 measure is being increasingly used as a measure of SPI. Morck et al. (2000)
provide evidence of higher R2s in poor economies than in rich economies and argue that
stronger public investor property rights in rich economies promote informed arbitrage,
which allows the incorporation of firm-specific information into asset prices. Durnev,
Morck, Yeung, and Zarowin (2003) show that firms with lower R2s exhibit a higher
association between current returns and future earning: i.e., lower R2 indicates greater
stock price informativeness. During periods when more firm-specific news is released,
one would expect the idiosyncratic component of returns to be high. We estimate
market model R2s using monthly and weekly data. For monthly measures, we remove
firms that do not have twelve months of data in a given year. Final results are
qualitatively similar whether monthly or weekly data are used.
We use the following to calculate aggregate R-squares for a calendar year.
∑
∑ (1)
Here, R2t is the weighted-average R2s of all firms i in a given year τ. The weights are the
total return variations (SST).
9
Our second measure of SPI is the average yearly abnormal announcement returns,
CAARs, associated with dividend initiations. If dividends were to signal private (inside)
information, the abnormal announcement returns associated with dividend initiations
convey the magnitude of this inside information. High abnormal announcement returns
imply that the initiation conveys incrementally important news to the market, ceteris
paribus. We therefore argue that the magnitude of abnormal dividend announcement
return is a useful proxy for SPI.
Unlike the first measure which uses a panel of firm-level R2s from the universe of
all CRSP firms, CAARs only provides us with a general level of SPI inferred from market
reactions for a much smaller sample of initiating firms. While the firm-level R2s can be
observed for both initiating and non-initiating firms, CAARs cannot be observed for non-
initiating firms. Baker and Wurgler (2004a) also use a similar announcement date
measure, though they do not plot this over time and use the measure primarily as a
sentiment proxy.
We define dividend initiation as the first appearance on Compustat of an ordinary
cash dividend of non-monthly frequency (as in Bulan, Subramanyam and Tanlu, 2006).
The event period for CAARs is defined as days –1 to +l, where day 0 is the dividend
announcement date from CRSP. We impose a 200 minimum trading day requirements as
in Michaley, Thaler and Womack (1995) and use months –14 through month –2 to
estimate the parameters for excess return calculations. We do this for the period 1964-
2006.
10
3.3 Rising SPI and the Declining Information Content of Dividend Initiations
In Figure 1, we use the monthly stock returns data to illustrate the annual equal- and
value-weighted average R2s of all Compustat firms from 1964 to 2006.6 Average R2 has
been declining over this period, a trend also noted in Morck et al. (2001). There is some
tendency for R2 to rise just prior to stock market crashes, perhaps due to asset pricing
bubbles being strengthened during these periods. The general decline in R2s supports
our thesis that SPI has improved over this period.
Indeed, we find that the CAARs associated with dividend initiations have declined
over this time period. In Table 1, average CAAR (market model) declines from 4.46% in
the pre-1978 period to 2.46% in the post-1978 period (the difference is significant with
a p-value less than 1%). The results are similar with market-adjusted and Fama and
French three-factor models. The results also remain robust when we repeat the analysis
with different sample periods that end in 1998 instead of 2006. The declining CAAR
trend is consistent with our thesis that the information conveyed by dividends has
declined over a period where the overall SPI has improved.
To test this further, we calculate the yearly average R2 of all firms (dividend-
initiating and non-initiating) and sort the years (1963 to 2006) into quintiles based on
average R2s. Thus the top quintile contains years with the highest average R2 values and
corresponds to a period of low SPI. The bottom quintile contains years with the lowest
average R2 values and corresponds to a period of high SPI. We repeat the dividend
6 Using weekly stock return data also yields a similar pattern. We do not report this to conserve space.
11
initiation tests for each quintile with the expectation that periods of high SPI would elicit
low dividend initiation CAARs. Table 2 reports the results.
Dividend initiations in periods of low SPI (quintile 5) are associated with an
average CAAR of 4.67%, compared with an average CAAR of 2.78% in periods of high SPI.
The difference is statistically significant (p-value<0.001). Going from quintile 1 to
quintile 5 we see a monotonic increase in the CAAR associated with dividend initiations.
This pattern is consistent with the notion that the information content of dividend
initiations is weaker when the external information environment is stronger, and lends
support to our hypothesis that the demand for dividend-based signaling is lower when
the external information environment is strong.
3.4 SPI and Dividend Changes
We have seen in the previous section that the abnormal returns associated with
dividend initiation announcements are higher during periods where stock prices are less
informative. We now extend this analysis to dividend change announcements. The same
reasoning applies here: we expect to see a more pronounced stock price reaction to
dividend change announcements during periods of low SPI, and vice versa. However,
since dividend changes might not convey much of ‘surprise’ news as initiations do, one
would expect the reaction to dividend changes to be more muted. Following the
literature, we assemble a list of dividend change announcements from 1963 to 2006 for
all NYSE, AMEX, and NASDAQ firms. For this sample, we remove dividend initiations (as
defined by the criteria in section 3.2), omissions, and resumptions. We again remove all
financials and utilities. We only consider firms that pay regular quarterly dividends and
12
require that no other distributions are made in the -15/+15 day window covering the
dividend announcement. Dividend change announcements that fall within -3/+3
window of earnings announcements are also removed. Finally, we exclude dividend
changes (in absolute values) that are lower than 10%. Dividend change is defined as the
change in dividend from Qt-1 to Qt scaled by the dividend at Qt-1. The final sample
contains 17,745 dividend change announcements, out of which 5,097 makes up the
dividend decrease sample.
Table 3 reports the results of the impact of SPI on dividend change abnormal
announcement returns.7 Results for dividend change announcements are generally
comparable to those for initiations. Starting with dividend increases, we note that the
abnormal return for the lowest SPI quintile is 1.47%, compared with 0.85% for the
highest SPI quintile. The difference between abnormal returns for the two SPI periods is
significant at the 1% level. Similarly, for dividend decreases, we find that the abnormal
return magnitude is greater in periods when SPI is low: –1.61% for SPI quintile 1, vs. –
0.91% for SPI quintile 5.
These results reinforce the idea that the information content of dividend changes
is a function how informative stock prices are in general. In periods when stock prices
are least informative, dividend change announcements are received as strong “news”.
When SPI is high, our results suggest that dividend convey less information. Since SPI
levels have trended up over the last two decades, firms have discovered that dividends
may be less efficacious in conveying firm specific news to the market.
7 Results are based on the market model. Other estimation methods produce qualitatively similar results.
13
3.5 S&P500 Index Membership and Dividends
In this section, we examine the abnormal returns associated with dividend change
announcements for firms that are included in the S&P500 Index. Our main premise for
this exercise is that indexers are primarily focused on minimizing tracking error for
their portfolios, and are less concerned about idiosyncratic information. Since firms do
not choose either to be added or to be deleted from the index, and since the S&P
corporation has no set views on dividend policy as a condition for Index membership,
this event provides a direct means and an exogenous setting to study the effects of
changes in the demand for dividend-signaling on a firm’s dividend policy.
We obtain the data for S&P firms from CRSP, which provides the dates firms are
added to (and deleted from) the index. We limit our analysis to dividend increases. First,
dividend initiations are generally not applicable to our sample: most firms that are
added to the index are dividend-payers. Second, dividend decreases are rare for S&P
Index firms, so we cannot get a good estimate of the dividend decrease announcement
returns for S&P Index firms.
In Table 4, we separate firms into S&P and non-S&P groups. The first three
models (I-III) report results for the group of firms that have been an index component at
some point. Thus, this group consists of firms that are currently in the index, were in the
index, or have always been in the index. This way, we can observe the same set of firms
as they are added and deleted from the index. For dividend yield, we average the annual
dividend yields of firms based on whether they are in the index for that particular year
14
or not. For CAAR calculation, a firm is considered to be index if the dividend
announcement date falls within the period when a firm is in the S&P index.
Our findings indicate that during periods when firms are in the S&P index,
dividend yields decline and CAARs associated with dividend changes are smaller in
magnitude. The difference between S&P and non-S&P period is statistically significant
at the 1% level. Similar results are obtained when we expand the sample to include all
firms. We test for the S&P index membership effect more formally using the following
pooled regression:
CARit = 0 + 1 SP.PERIODit + Controlsit + εit (2)
where CAR is the firm-level dividend change abnormal announcement returns (for τ =-1
to τ =1, using the Fama and French three-factor model). As before, we initially limit our
sample to firms that have been an index component at some point in time, and then test
the same using all firms.
Table 5 reports the results for all firms (results for the restricted are materially
similar). We include both firm and time subscripts to note that the same firm could
change its dividends more than once at different points in time. We do not scale
dividend change by price as price may contain information that is correlated with our
dependent variable, but subsequently include price for robustness. SP_DUMMY is an
indicator variable that equals one if the dividend announcement date falls within the
period when a firm is in the S&P index, and is zero otherwise. We scale dividend changes
by earnings volatility (∆DIV/EPS.VOL) instead of beginning dividend as an additional test.
We find that SP_DUMMY is significantly negative in all model specifications, indicating
15
that announcement returns are indeed lower in periods when firms are in the index.
Of particular interest to us is the time series dividend response coefficient (DRC)
of ∆DIV/P estimated from the regression model in Table 5. We calculate the average DRC
cross-sectionally for each year. Plotting DRC in calendar time does not show any clear
trends (these results are available from the authors upon request). We further estimate
DRCt = 0 + 1 Institutiont + εt (3)
Institution is defined as the equal-weighted average institutional holdings in each year τ.
There are 27 time series observations (1980 to 2006). The coefficient of Institution is
negative but insignificant (-5.17 with a t-stat of -0.55). Therefore, controlling for change
in a firm’s information environment in the first-stage regression not only takes away the
trend in DRC but also makes institutional holdings insignificant. This is not to say that
institutional holdings does not matter. However, use of institutional holding as proxy of
information is questionable – perhaps institutions are motivated by other factors.
We also interact the change in dividends with S&P Index dummy, and find that
the DRC is indeed smaller when firms are part of the index.
What we have shown thus far is that changes in firms’ informational
environments and stock price informativeness affect announcement returns for both
dividend initiations and dividend changes. During periods of high SPI, announcement
returns for initiations and dividend changes are lower. Also, as firms move into
environments where demand for signaling is lower, dividend change announcement
returns also decline in magnitude. For the rest of the paper, we mainly focus on dividend
16
payments and initiations as the disappearing dividend phenomenon originated from
dividend payments and not from changes.
4 Additional analysis of the propensity to pay dividends
4.1 Original F-F Variables & Idiosyncratic Risk
We first replicate Hoberg and Prabhala (2009) results by using coefficients derived from
Fama and MacBeth (1973) style regressions to generate the propensity to pay dividends.
We regress changes in the propensity to pay dividends on Nixon dummy (Nixon) and the
dividend premium (Premium). First differencing mitigates challenges arising from the
stationarity concerns of the series.8 The Nixon dummy is equal to one for years 1972 to
1974, zero otherwise. The dividend premium variable is standardized as in Baker and
Wurgler (2004b). Table 6 reports the results for the 1978-2006 out-of-sample period
(results for the full sample are similar). Consistent with Hoberg and Prabahala (2009),
we find that the dividend premium is insignificant in explaining changes in PTP. The
Nixon control (Nixon), while positive, is also insignificant. When we use the level of PTP
instead of changes in PTP, the dividend premium variable remains insignificant.
Thus, regardless of the sample period and functional form, dividend premium
loses its significance in the second-stage regressions in explaining either the level of PTP
or its change. These results are inconsistent with the catering hypothesis of dividends,
8 The level form of PTP series has first-order autocorrelation of approximately 0.49 for 1978-2006. The unit root
test does not reject the null at 5% level or lower. However, we can reject the presence of unit roots for all
independent variables at the 1% levels.
17
and point to risk measures as an important factor in explaining the decline in dividends
(H-P conclude the same).
We next introduce the SPI measure into the model and re-estimate the results
(presented in Table 7). Dividend premium does not affect the main results, and in fact,
remains statistically insignificant. To be consistent with Baker and Wulgler (2004b), we
initially standardize all variables to have unit variances. With the change in PTP as the
dependent variable, the coefficient SPI_RSQ9 is 3.66, significant at the 1% level. Thus,
increasing the R2 measure of SPI by one standard variation (approximately 6%) results
in approximately 3.66% change in propensity to pay. Similarly, when we use the level of
PTP as the dependent variable, the SPI measure remains significant, with the model R-
square at 30.4%.
To sum, we find that the average change and level of PTP responds inversely to
an improvement in the information environment. These results are consistent with our
hypothesis that firms cut back on dividends when their relative efficacy declines – that is,
an overall improvement in firm level information is associated with a reduced use of
dividends for signaling the same information.
4.2 Robustness checks
As Petersen (2009) points out, Fama-MacBeth statistics can be inflated in the presence
of fixed firm effects. Accordingly, we re-estimate the regressions using standard errors
that are robust to clustering and to the inclusion of year dummies. We also include the
9 Average R2s estimated using weekly stock returns data.
18
firm-level SPIs for these estimates.10 This replication provides very similar test statistics
and the declining trend. For both methods of estimation, size, profitability, and
investments are all highly significant and take on expected signs. Risk measures are both
negatively significant and their coefficients are also similar to those reported in Hoberg
and Prabhala (2009). Table 8 (I-III) reports the replication results. The firm-level SPIs
either maintain or improve their significance: they are significant at the 1% level in all
sample periods. Table 8 (IV-VI) reports the results with Cash and Retained. The
coefficients for SPI remain qualitatively similar.
4.3 Institutional Holdings & Repurchases
Since institutions are generally better informed than individual investors, firms with
high levels of institutional ownerships should possess lower information asymmetry
problems (see Allen, Bernardo, and Welch, 2000). On the other hand, the answer to how
much incremental private information could be extracted from institutional holding is a
prior unclear due to their herding behavior (Sias, 2004). Institutional ownerships could
also serve as a monitoring mechanism. In this case, we could expect firms with high
institutional ownership to pay higher dividends as they are better-governed firms.11
Institutional holdings data are collected from Thomson Financial database which
provides information from institutional 13F SEC filings. The data are available from
1980. Our measure of institutional holdings is the year-end aggregate institutional stock
10
These are the annual panel data of firm-level R2s derived from equation 1 by using the weekly stock return
data. 11
Jensen (1986) argues that with enhanced monitoring, firms are more likely to pay out their free cash flow. An
alternative argument would be that since these firms have better monitoring mechanisms, they do not have the
need to reduce agency costs of free cash flows via dividend payments.
19
holdings (for each firm) scaled by total value of outstanding shares. We remove all
observations that have the ratios above 1. We use lagged holdings to address the
concern that institutions are attracted to dividend –paying stocks. On average,
institutions (Institutions) hold 35% of total outstanding shares.
From both textbook theory and empirical evidence, shares repurchases are
substitutes for dividends, and some authors have shown that repurchases surged in
1980s (Grullon and Michaely (2002)). However, share repurchases, unlike dividends,
are temporary and firms repurchasing shares do not need to commit their future
operating cash flows to divided payments. For instance, Jagannathan, Stephens, and
Weisbach (2000) argue that dividends are paid by firms with higher ‘permanent’
cash flows. Grullon and Michaely (2002) also show that, at the aggregate level, firms are
substituting dividends with repurchases, and that younger firms are becoming more
likely to pay out cash in the form of repurchases. However, as Fama and French (2001)
points out, in general, firms repurchasing are also dividend-paying firms. Unlike that of
Fama and French (2001), our regression directly controls for the repurchase variable.
Following Grullon and Michaely (2002), we define repurchase (Repurchases) as net of
total expenditure on purchase of common and preferred stocks and reduction in value of
preferred stocks. Repurchases is a dummy variable equal to one if a firm has made a
repurchase, zero otherwise. We then match the institutional-holding and repurchase
data with other variables from CRSP and Compustat databases. The final sample (1981
to 2006) is reduced to under 43,000 observations.
Table 8 (VII) reports the results. The coefficient of Institutions is significantly
20
negative. In unreported results, we also use dividend yield and payout as dependent
variables and the significantly negative coefficient of Institutions remain. Thus, evidence
suggests that firms with higher institutional holdings are less likely to pay dividends,
and among payers, firms with high institutional holdings pay less.12 From marginal
effects (not reported here), shifting from lowest to highest institutional holdings reduces
the probability of being a payer by 24%. Shifting the holdings one standard deviation
reduces the probability by 6%. It is likely that firms with higher institutional holdings
prefer other methods of payments such as stock repurchases (Grinstein and Michaely
(2005)). The repurchase dummy is significantly positive for all sample periods: firms
that are repurchasing are also dividend payers. Being a repurchaser increases the
probability to pay dividends by about 4%. This finding is consistent with that of Fama
and French (2001) who show that firms repurchasing are mainly in the domain of
dividend payers. Our SPI measure remains significantly positive at the 1% level even
after controlling for these additional variables.
We also consider other proxies of agency costs such insider holdings and
managerial compensation. However, insider holding and executive compensation data
are available only from 1992, and the coverage is not comprehensive even in the years
reported. Since we are mainly interested in investigating the disappearing dividend
trend by comparing the characteristics in base and forecast periods, adding these data
create little value for our purpose. Accordingly, we leave these data for future exercises.
12
Grinstein and Michaely (2005) also have similar findings. The authors find that institutions prefer low-
dividend stocks to high-dividend stocks.
21
5. Concluding Remarks
Between 1978 and 1999, Fama and French (2001) document a remarkable decline in the
fraction of listed firms that pay dividends. Much of this is explained by changing asset
characteristics of listeds, but FF show that even after controlling for such differences, the
residual propensity to pay dividends has diminished sharply over this period. This
dividend disappearance in turn has led to several attempts to explain the puzzle, such as
the catering theories of Baker and Wurgler (2004) and the institutional explanations of
Amihud and Li (2009). We do not take issue with these explanations, though we do cast
doubt on the main premise in these explanations. Rather, we offer a simple rational
explanation based on two concurrent trends over the FF examination period.
The first is a sustained increase in the informativeness of stock prices lessening
the relative benefit of costly signaling via dividends. The second trend is the tremendous
growth in indexing that has occurred over the last three decades. Under the assumption
that indexers care primarily about tracking error, their demand for idiosyncratic
information may have also declined. We are not saying that dividends have outlived
their usefulness; rather, we point to two concurrent trends that have reduced the
marginal benefits of the signaling task entrusted to dividend policy.
22
References
Allen, Franklin, and Roni Michaely, 2003. Handbook of the Economics of Finance: Payout Policy (Elsevier, North Holland).
Allen, Franklin, Antonio E. Bernardo, and Ivo Welch, 2000, A theory of dividends based on tax clienteles, Journal of Finance 55, 2499–2536.
Amihud, Yakov, and Kefei Li, 2006, The declining information content of dividend announcements and the effects of institutional holdings, Journal of Financial and Quantitative Analysis 41, 637-660.
Asquith, Paul, and David Mullins, 1983, The impact of initiating dividend payments on shareholders’ wealth, Journal of Business 56, 77–96.
Baker, Malcolm, and Jeffery Wurgler, 2004a, A catering theory of dividends, Journal of Finance 59, 1125–1165.
Baker, Malcolm, and Jeffery Wurgler, 2004b, Appearing and disappearing dividends: The link to catering incentives, Journal of Financial Economics 73, 271–288.
Benartzi, Shlomo, Roni Michaely, and Richard H. Thaler, 1997, Do changes in dividends signal the future or the past? Journal of Finance 52, 1007–1034.
Bhattacharya, Sudipto, 1979, Imperfect information, dividend policy, and the “bird-in-hand” fallacy, Bell Journal of Economics 10, 259–270.
23
Bulan, Laarni, Narayanan Subramanian, and Lloyd Tanlu, 2007, On the timing of dividend initiations, Financial Management 36, 31-65.
DeAngelo, Harry, Linda DeAngelo, and Douglas J. Skinner, 2004, Are dividends disappearing? Dividend concentration and the consolidation of earnings, Journal of Financial Economics 72, 425–456.
Durvev, Artyon, Randall Morck, Bernard Yeung, and Paul Zarowin, 2003, Does greater firm-specific return variation mean more or less informed stock pricing? Journal of Accounting Research 41, 797-836.
Easterbrook, Frank, 1984, Two agency cost explanations of dividends, American Economic Review 74, 650–659.
Fama, Eugene F., and Kenneth R. French, 2001, Disappearing dividends: Changing firm characteristics or lower propensity to pay? Journal of Financial Economics 60, 3–44.
Fama, Eugene F., and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal of Political Economy 81, 607-636.
Grinstein, Yaniv, and Roni Michaely, 2005, Institutional holdings and payout policy, Journal of Finance 60, 1389-1425.
Grullon, Gustavo, and Roni Michaely, 2002, Dividends, repurchases, and the substitution hypothesis, Journal of Finance 57, 1649–1684.
Grullon, Gustavo, Roni Michaely, Shlomo Benartzi, and Richard H. Thaler, 2005, Dividend changes do not signal changes in future profitability. Journal of Business 78, 1659-1682.
Healy, Paul M., and Krishna G. Palepu, 1988, Earnings information conveyed by dividend initiations and omissions, Journal of Financial Economics 21, 149–176.
Hoberg, Gerard, and Nagpurnanand R. Prabhala, 2009, Disappearing dividends: The importance of idiosyncratic risk and the irrelevance of catering, Review of Financial Studies 22, 79-116.
Jagannathan, Murali, Clifford Stephens, and Michael S. Weisbach, 2000, Financial flexibility and the choice between dividends and stock repurchases, Journal of Financial Economics 57, 55-384.
Jensen, Michael C., 1986, Agency costs of free cash flow, corporate finance and takeovers, American Economic Review 76, 323–329.
24
Jensen, Michael, and William Meckling, 1976, Theory of the firm: Managerial behavior, agency costs, and capital structure, Journal of Financial Economics 3, 305–360.
Lang, Larry H. P., and Robert H. Litzenberger, 1989, Dividend announcements: Cash flow signaling versus free cash flow hypothesis? Journal of Financial Economics 24, 181-192.
Michaely, Roni, Richard H. Thaler, and Kent L. Womack, 1995, Price reactions to dividend initiations and omissions: Overreaction or drift? Journal of Finance 50, 573–608.
Miller, Merton H., and Franco Modigliani, 1961, Dividend policy, growth and the valuation of shares, Journal of Business 34, 411–433.
Miller, Merton H., and Kevin Rock, 1985, Dividend policy under asymmetric information, Journal of Finance 40, 1031–1051.
Morck, Randall, Bernard Yeung, and Wayne Yu, 2000, The information content of stock markets: Why do emerging markets have synchronous stock price movements? Journal of Financial Economics 58, 215-260.
Nissim, Doron, and Amir Ziv, 2001, Dividend changes and future profitability, Journal of Finance 56, 2111–2133.
Petersen, Mitchell A., 2009, Estimating standard errors in finance panel data sets: Comparing approaches, Review of Financial Studies 22, 435-480.
Roll, Richard, 1988, R2, Journal of Finance 43, 541-566.
Ross, Stephen A., 1977, The determination of financial structure: The incentive-signalling approach, The Bell Journal of Economics 8, 23-40.
Sias, Richard W, 2004, Institutional Herding, Review of Financial Studies 17, 165-206.
25
Figure 1
Time Series of Market Model R2s
R2s are estimated from market model regressions using monthly data. Equal-weighted R2s (solid line) are calculated by averaging over all firms in a given year (excluding financials and utilities). The value-weighted R2s (short and long dashes) are derived from equation 1. Short-dashed line plots the 5-year moving averages of value-weighted R2s.
26
Table 1
Comparisons of Announcement Day Abnormal Returns for Different Sample Periods
The announcement day abnormal returns are based on the market, market-adjusted, and Fama and French three-factor models. Dividend initiations are defined as the first appearance on Compustat of an ordinary cash dividend. The ‘event period’ is defined as days τ = -1 to τ = l, where τ = 0 is the dividend announcement date documented in CRSP. Cumulative Average Abnormal Returns (CAARs) are the averages of cumulative returns over 3-day periods. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 1%, 5%, and 10% levels, respectively
Market Model Market-Adjusted FF Model
N CAAR Positive: Negative t-Stat N CAAR
Positive: Negative t-Stat N CAAR
Positive: Negative t-Stat
1 1964-2006 1,140 3.33% 726:414 16.43*** 1,140 3.58% 761:379 17.82*** 1,137 3.28% 724:413 16.82***
2 1964-1998 926 3.58% 585:341 15.77*** 926 3.81% 617:309 16.81*** 923 3.55% 583:340 16.24***
3 1964-1977 493 4.46% 333:160 15.85*** 493 4.72% 349:144 16.73*** 490 4.37% 326:164 16.06***
4 1978-1998 433 2.57% 252:181 8.30*** 433 2.77% 268:165 8.95*** 433 2.62% 257:176 8.62***
5 1978-2006 647 2.46% 393:254 9.82*** 647 2.71% 412:235 10.97*** 647 2.46% 398:249 9.92***
(3)-(1)
(1.13%)***
(1.14%)***
(1.09%)***
(3)-(2)
(0.88%)***
(0.91%)***
(0.82%)***
(3)-(4)
(1.89%)***
(1.95%)***
(1.75%)***
(3)-(5)
(2.00%)***
(2.01%)***
(1.91%***)
27
Table 2
Dividend Initiation Returns and Stock Price Informativeness
Quintile 1 contains years with the highest average R2s (lowest SPI), and quintile 5 contains years with the lowest average R2s (highest SPI). The announcement day abnormal returns are based on the market, market-adjusted, and Fama and French three-factor models. Dividend initiations are defined as the first appearance on Compustat of an ordinary cash dividend of non-monthly frequency. The ‘event period’ is defined as days τ = -1 to τ = l, where τ = 0 is the dividend announcement date documented in CRSP. Cumulative Average Abnormal Returns (CAARs) are the averages of cumulative returns over the 3-day period. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
Market Model Market-Adjusted FF Model
Quintile N CAAR t-Stat N CAAR t-Stat N CAAR t-Stat
1(High SPI) 209 2.78% 5.609*** 209 2.94% 5.871*** 207 2.77% 5.626***
2 250 2.86% 7.314*** 250 3.10% 7.987*** 250 2.78% 7.199***
3 236 3.02% 7.256*** 236 3.31% 7.893*** 236 3.05% 7.494***
4 236 3.44% 9.097*** 236 3.94% 10.390*** 236 3.36% 9.078***
5(Low SPI) 208 4.67% 9.736*** 208 4.72% 9.816*** 208 4.56% 9.962***
(5)-(1)
1.89%***
1.78%***
1.79%***
28
Table 3
Dividend Change Announcement Returns and Stock Price Informativeness
For this measure, sample years are divided into quintiles based on the yearly average R2s (reported in Figure 1). Quintile 1 contains years with the highest average R2s (lowest SPI), and quintile 5 contains years with the lowest average R2s (highest SPI). Dividend change announcement returns are based on the market model. Dividend change is defined as the change in dividend from Qt-1 to Qt scaled by dividend at Qt-1. The ‘event period’ is defined as days τ = -1 to τ = l, where τ = 0 is the dividend announcement date in CRSP. Cumulative Average Abnormal Returns (CAARs) are the averages of cumulative returns over 3-day periods. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
SPI Quintile
(1= Lowest SPI quintile;
5=Highest SPI quintile)
Dividend Increases Dividend Decreases
CAAR Positive: Negative t-Stat CAAR Positive: Negative t-Stat
1 1.47% 1542:957 18.527*** -1.61% 478:682 -13.846***
5 0.85% 978:711 8.756*** -0.91% 349:422 -5.763***
Low SPI-High SPI 0.62%***
-0.70%***
29
Table 4
Dividend Yields and Dividend Change Announcement Returns for S&P500 Firms
Dividend yield is defined as the sum of total dividends paid over a 12-month period preceding the announcement month scaled by the average end-of month prices for 3-month period preceding the 12-month period. The announcement day abnormal returns are based on Fama and French three-factor model. The ‘event period’ is defined as days τ = -1 to τ = l, where τ = 0 is the dividend announcement date documented in CRSP. For CAAR calculation, a firm is considered to be in the index if the dividend announcement date falls within the period when a firm is in the S&P index. The numbers reported are the averages for respective groups. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
S&P Firms Only All Firms
I II III IV V VI
S&P Period
Non-S&P Period I – II
S&P Period
Non-S&P Period IV – V
Dividend Yield 0.019 0.021 -0.002*** 0.019 0.022 -0.003***
CAAR 0.007 0.009 -0.002*** 0.007 0.011 -0.004***
30
Table 5
Effect of SPI on Dividend Increase Announcement Returns
The dependent variable is the 3-day announcement abnormal return (CAR). The announcement day abnormal returns are based on Fama and French three-factor model. The ‘event period’ is defined as days τ = -1 to τ = l, where τ = 0 is the dividend announcement date documented in CRSP. Independent variables include: R∆DIV, rate of change of dividend, defined as change in dividend from Qt-1 to Qt scaled by dividend at Qt-1; ∆DIV/EPS.VOL, change in dividend scaled by EPS volatility; ∆DIV/P, change in dividend scaled by end-of-month closing price prior to the announcement month; EPS.VOL, EPS volatility; S&P DUMMY, an indicator variable that equals one if the dividend announcement date falls within the period when a firm is in the S&P index, zero otherwise; Size, SIZE, log of total assets scaled by 1962 dollars; Dividend yield, DIVYIELD, sum of total dividends paid over a 12-month period preceding the announcement month scaled by the average end-of month prices for 3-month period preceding the 12-month period. Pooled regression is used. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
I
1971-2006
II
1971-2006
III
1971-2006
IV
1963-2006
Coefficient Coefficient Coefficient Coefficient
Intercept 0.007*** 0.012*** 0.012*** 0.004***
(7.78) (9.34) (9.1) (3.73)
∆DIV/EPS.VOL 0.028*** 0.024*** 0.024***
(7.55) (6.27) (6.27)
∆DIV/P 1.53***
(17.31)
S&P DUMMY -0.006*** -0.003*** -0.003*** -0.002*
(-5.09) (-2.75) (-2.74) (-1.69)
FIRM SIZE -0.000*** -0.000*** -0.000***
(-5.41) (-5.41) (-2.77)
DIV YIELD -0.000 -0.10
(-0.04) (1.05)
Adjusted R2 0.0091 0.016 0.016 0.038
N 6,235 6,196 6,196 9,726
31
Table 6 Time Series Regression of Dividend Premium on PTP and Change in PTP
The dependent variables are the changes in propensity to pay (columns I and II) and the propensity to pay (column III). Propensity to pay is the difference between the actual percentage of firms paying dividends in a given year and the expected percentage, which is the average predicted probability from a given logit model that controls for FF and H-P variables. The independent variables include: dividend premium (Premium) and the Nixon dummy. The dividend premium is defined as the lagged value of log of the difference between book-value-weighted average M/B of dividend payers and nonpayers. Premium is lagged once and standardized to have unit variance. The results are based on time-series regression coefficients with Newey-West t-statistics (two lags) in parentheses. ***, **, and * imply the significance of coefficient at the 1%, 5%, and 10% levels, respectively.
I:1963-2006 II:1978-2006 III:1978-2006
Dependent= Change in PTP Dependent=PTP
Variable Coefficient Coefficient Coefficient
Intercept -1.012 -0.803 -18.41***
(-1.17) (-0.76) (-8.92)
Premium -0.101 0.411 0.641
(-0.12) (0.41) (0.41)
Nixon 4.028
(1.17)
Adjusted R2 -0.030 -0.034 -0.031
32
Table 7 The Effect of Stock Price Informativeness on the Propensity to Pay Dividends
The dependent variable is defined as the propensity to pay dividends (both in levels and in changes), which is the difference between the actual percentage of firms paying dividends in a given
year and the expected percentage based on a logit model specified in FF and control for FF and H-P variables . The independent variables include: dividend premium (Premium), R2 measure of SPI (SPI_RSQ). Dividend premium is defined as the lagged value of log of the difference between book-value-weighted average M/B of dividend payers and nonpayers. All independent variables are standardized to have unit variances. Premium is lagged once. The results are based on time-series regression coefficients with Newey-West t-statistics (two lags) in parentheses. ***, **, and * imply the significance of coefficient at the 1%, 5%, and 10% levels, respectively.
Dependent Variable: Change in PTP PTP
Variable Coefficient Coefficient
Intercept -0.803 -31.89
(-0.63) (-13.24)
Premium 1.393 2.007
(1.29) (1.47)
SPI_RSQ 3.656*** 1.176***
(2.94) (3.95)
N 29 29
Adjusted R2 0.185 0.304
33
Table 8
Panel Logit Regressions of the Effects of Firm Characteristics on Dividend Pay Status
The dependent variable is a dummy value (DIVPAY) that is equal to one if firm pays dividends, zero otherwise. The independent variables include: firm size (NYP), the percent of firms that have the same or smaller market capitalization; asset growth (dAssets), change in assets scaled by assets; profitability (E/A), income before extraordinary items plus interest expense plus income statement deferred taxes) scaled by total assets s; growth opportunity (M/B), the ratio of market value of assets to book value of assets; firm-specific risk (Idiosyncratic), sum of squared residuals from the market model regression using daily data; systematic (Systematic) standard deviation of the predicted values from the idiosyncratic risk regression; SPI.RSQ.W, firm-level R2 measure of stock price informativeness from weekly data; cash (Cash), the ratio of cash and short-term investments to assets; retained earnings (Retained), the ratio of retained earnings to assets. We remove all financials and utilities. The table reports estimates of panel logit models with standard errors that are robust to clustering at the firm level. Year dummies are included but not reported here. The t-statistics are given in the parentheses: ***, **, and * imply the significance of coefficient at the 1%, 5%, and 10% levels, respectively.
34
I:
1963-2006
II:
1963-1977
III:
1978-2006
IV:
1963-2006
V:
1963-1977
VI:
1978-2006
VII:
1981-2006
Dependent=DIVPAY
Variable Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient
Intercept 2.134*** 2.882*** 1.879*** 0.455** 0.990*** 0.615*** -.003
(13.56) (14.65) (17.48) (2.15) (3.94) (5.19) (-0.02)
NYP 3.614*** 2.956*** 3.765*** 3.766*** 3.207*** 3.908*** 4.063***
(30.94) (15.67) (28.41) (30.28) (15.95) (27.48) (19.91)
dAssets -0.972*** -1.162*** -0.965*** -0.947*** -0.777*** -0.998*** -1.20***
(-8.40) (-8.27) (-7.48) (-12.05) (-5.31) (-11.04) (-9.68)
E/A 4.496*** 7.412*** 4.168*** 1.340*** 2.603*** 1.131*** 0.551*
(21.23) (12.36) (19.24) (5.5) (3.85) (4.47) (1.70)
M/B -0.387*** -0.362*** -0.405*** -0.365*** -0.382*** -0.372*** -0.323***
(-16.33) (-9.18) (-14.54) (-14.55) (-9.04) (-12.65) (-9.67)
Systematic -115.002*** -111.007*** -112.885*** -101.253*** -95.833*** -97.791*** -49.328***
(-23.31) (-13.38) (-20.09) (-20.21) (-11.16) (-17.09) (-12.97)
Idiosyncratic -71.106*** -97.770*** -64.386*** -52.347*** -82.675*** -44.843*** -91.339***
(-28.54) (-22.47) (-23.16) (20.88) (-19.56) (-16.14) (-12.49)
SPI.RSQ.W 1.471*** 0.561*** 1.692*** 1.166*** 0.288 1.379*** 1.342***
(13.20) (2.72) (13.52) (10.11) (1.36) (10.62) (7.71)
Cash -1.766*** -0.187 -1.971*** -2.389***
(-9.69) (-0.45) (-10.09) (-9.90)
Retained 3.788*** 4.538*** 3.675*** 3.603***
(30.63) (18.94) (27.45) (21.41)
Institutions -0.613***
(-3.71)
Repurchases 0.245***
(5.31)
Firm Years 130,388 27,789 102,499 118,561 25,995 92,566 42,327
Firms 13,400 3,635 12,583 12,376 3,593 11,585 7,915
Year Dummies Yes Yes Yes Yes Yes Yes Yes
Log Pseudo-likelihood -50440 -10696 -39553 -44545 -9335 -34947 -16854
Pseudo-R2 0.432 0.386 0.405 0.454 0.430 0.434 0.413