What is the information content of dividend changes - Columbia ...
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) threefactor 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 1
Under 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 timeseries, 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 firmdividend 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: TCA j ,t = θ 0,t + θ1,t CFO j ,t −1 + θ 2,t CFO j ,t + θ 3,t CFO j ,t +1 + θ 4,t ∆REV j ,t + θ 5,t PPE j ,t + υ j ,t
(1)
4
Dividend initiations and omissions are not included in our sample as a result of these filters. 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. 5
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 timeseries standard deviation of the residuals, σ (υˆ )t , 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 FamaFrench three-factor model after incorporating the information risk factor returns from FLOS (2004) as follows: R j ,t − R f , t = α j + α j , ∆ DTi ,t + β j ( Rm,t − R f ,t ) + β j , ∆ DTj ,t ( Rm,t − R f ,t ) + s j SMBt + s j ,∆ DTi ,t SMBt + hj HMLt + hj , ∆ DTj ,t HMLt + e j IRt + e j , ∆ DTj ,t IRt + et
(2)
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, R p ,t − R f ,t , for each of the two portfolios for each month (a total of 372 months). The following multi-factor model is estimated for each portfolio:
R p ,t − R f ,t = α p + β p ( Rm ,t − R f ,t ) + s p SMBt + hp HMLt + e p IRt + et
(3a)
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: R p ,t − R f ,t = α p + α pdiv D + β p ( Rm ,t − R f ,t ) + β pdiv D *( Rm,t − R f ,t )
+s p SMBt + s pdiv D * SMBt + hp HMLt + hpdiv D * HMLt + ep IRt + epdiv D * IRt + et (3b) where D is an indicator variable for dividend-paying firms, β pdiv , s pdiv , hpdiv , e pdiv are the differences in factor loadings of dividend-paying firms on ( Rm,t − R f ,t ) , SMB, HML, and IR, respectively, and α pdiv 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: Ei ,t +1 = α i + λi Ei ,t + ε i ,t
(3c)
We test the difference in earnings persistence between dividend-paying and nondividend-paying firms using the following regression: Ei ,t +1 = α i + α pdiv Di ,t + λi Ei ,t + λ pdiv Di ,t ( Ei ,t ) + ε i ,t
(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 λ pdiv is consistent with dividend-paying firms having less information risk. Results reported in Panel A of Table 2 indicate that dividend-paying and nondividend-paying firms have distinct risk characteristics. Dividend-paying firms have higher market and HML betas, and smaller size betas, and the differences ( β pdiv , s pdiv , hpdiv ) between the loadings are significant at 5% or higher. More importantly, dividend-paying firms load negatively on the information risk factor whereas the nondividend-paying firms load positively on the information risk factor, and the difference between the loadings ( e pdiv ) is significant at 1% level. This suggests that dividendpaying 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 λ pdiv 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 firmannouncement 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) riskadjusted 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 dividenddecreasing 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 dividendincreasing firms are sensitive to the samples used. The results in Panel A, based on all 10,597 dividend announcements, generally support the hypothesis that dividendincreasing 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 nonoverlapping 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 dividendincreasing 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 longterm 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 dividenddecrease 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-existingearnings’ 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 shortterm 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 threefactor 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)
∆Div
Mean 29.66
Median 20.00
Standard Deviation 30.33
N 9,110
Mean -46.07
Median -50.00
Standard Deviation 16.62
N 1,487
Mean -75.73*
Median -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
αp
βp
sp
hp
ep
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*
α pdiv
β pdiv
s pdiv
hpdiv
e pdiv
Differences
0.4205*
0.0554**
-0.1100**
0.2356*
-0.6300*
Panel B: Earnings Persistence
αi
λi
Non-dividend Firms
-0.0067*
0.7040*
Dividend Firms
0.0145*
0.8761*
α pdiv
λ pdiv
0.0212*
0.1721*
Differences
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: R p ,t − R f ,t = α p + β p ( Rm ,t − R f ,t ) + s p SMBt + h p HMLt + e p IRt + et R p ,t − R f ,t = α p + α pdiv D + β p ( Rm ,t − R f ,t ) + β pdiv D * ( Rm ,t − R f ,t )
+s p SMBt + s pdiv D * SMBt + hp HMLt + hpdiv D * HMLt + ep IRt + epdiv D * IRt + et 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 NYSEAMEX-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: Ei ,t +1 = α i + λi Ei ,t + ε i ,t Ei ,t +1 = α i + α pdiv Di ,t + λi Ei ,t + λ pdiv Di ,t ( Ei ,t ) + ε i ,t
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
Table 3 Characteristics of Dividend Changes Panel A: Frequency of Dividend Changes Number of Divided Change Events
Number of Unique Firms
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 > 15
860 584 386 315 231 187 124 80 77 53 25 17 15 6 6 15
Total
2,981
Panel B: Industrial Classification Industry
Agriculture, Forestry and Mining Mining Construction Manufacturing Transportation and Communication Utilities Wholesale Retail Financial Services Public Administration Others
SIC
Number of Unique Firms Dividend Increase Dividend Decrease
01-09 10-14 15-17 20-39 40-48 49 50-51 52-59 60-69 70-89 90-97 98-99
Total
390 59 21 975 77 83 70 160 614 128 0 0 2,577
15.1% 2.3 % 0.8% 37.8% 3.0% 3.2% 2.7% 6.3% 23.8% 5.0% 0.0% 0.0%
51 9 4 127 10 48 9 17 114 14 0 1
12.6% 2.2% 1.0% 31.4% 2.5% 11.9% 2.2% 4.3% 28.2% 3.5% 0.0% 0.2%
404
Panel C: Magnitude of Dividend Changes Absolute Percentage Change
Number of Firm-event Pairs Dividend Increase Dividend Decrease
12.5 to 25 25 to 50 50 to 100 100 to 250 250 to 500
5,688 2,359 692 339 32
183 916 388 0 0
Total
9,110
1,487 30
Table 4 Changes in Factor Loadings Surrounding Dividend Change Announcements Panel A: All Dividend Announcement (n=10,597) All Dividend Increases Predicted Sign Alpha
α− j α j ,∆
n/a
N=9,110
Dividend Increase Quartile 1 N=2,374
Dividend Increase Quartile 2 N=2,293
Dividend Increase Quartile 3 N=2,166
Dividend Increase Quartile 4 N=2,277
0.8618*
0.5857*
0.8089*
0.9392*
1.1358*
-0.7219*
-0.4705*
-0.6880*
-0.7704*
-0.9784*
1.0212*
0.9783*
1.0154*
1.0460*
1.0479*
-0.0457*
-0.0144
-0.0593*
-0.0653*
-0.0440*
0.6764*
0.5509*
0.6494*
0.7246*
0.7896*
-0.0733*
-0.0087
-0.0501***
-0.1142*
-0.1250*
0.1937*
0.1642*
0.2041*
0.1616*
0.2401*
0.0061
0.0723*
-0.0102
0.0184
-0.0554**
-0.0226*
-0.0683*
-0.0235**
-0.0302**
0.0336**
-0.0257*
-0.0141
-0.0365**
-0.0136
-0.0373**
Dividend Decreases Predicted Sign
N=1,487
-0.2493* n/a
-0.1325**
Market factor
β− j
β j ,∆
−
0.8860* +
0.0973*
Size factor
s− j s j ,∆
−
0.6003* +
0.0056
Book to market factor
h− j h j ,∆
−
0.3852* +
0.1332*
Information risk factor
e− j e j ,∆
−
-0.0164 +
0.1028*
31
Table 4 (continued) Panel B: Non-overlapping Dividend Announcements (n=1,946) All Dividend Increases Predicted Sign Alpha
α− j α j ,∆
n/a
N=1,458
Dividend Increase Quartile 1 N=365
Dividend Increase Quartile 2 N=366
Dividend Increase Quartile 3 N=379
Dividend Increase Quartile 4 N=348
0.7169*
0.4966*
0.6432*
0.7458*
1.0087*
-0.8152*
-0.5278*
-0.6884*
-0.8673*
-1.2157*
0.9654*
0.9199*
0.9481*
0.9743*
1.0308*
-0.0460*
-0.0290
-0.0411
-0.0534
-0.0604
0.6936*
0.4783*
0.6291*
0.7820*
0.8948*
-0.0481
0.0870
-0.0311
-0.1277***
-0.1203
0.2431*
0.2229*
0.2793*
0.2163*
0.2471*
0.0871*
0.1232*
0.0549
0.1457*
0.0269
-0.0119
-0.0846*
-0.0001
-0.0203
0.0605
0.0141
-0.0031
-0.0251
0.0336
-0.0687
Dividend Decreases Predicted Sign
N=488
-0.5617* n/a
-0.0097
Market factor
β− j β j ,∆
−
0.8576* +
0.1250*
Size factor
s− j s j ,∆
−
0.5542* +
0.0294
Book to market factor
h− j h j ,∆
−
0.4336* +
0.1629*
Information risk factor
e− j e j ,∆
−
-0.0356 +
0.1595*
Notes: Model: R j ,t − R f ,t = α j + α j , ∆ DTi ,t + β j ( Rm ,t − R f ,t ) + β j , ∆ DT j ,t ( Rm ,t − R f ,t ) + s j SMBt + s j , ∆ DTi ,t SMBt + h j HMLt + h j , ∆ DT j ,t HMLt + e j IRt + e j , ∆ DT j ,t IRt + et 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.
32
Table 5 Information Characteristics Surrounding Dividend Change Announcements Dividend Increases (A) (B) -3yr to +1yr to 1yr +3yr
(B)-(A)
(A) -3yr to -1yr
Dividend Decreases (B) (B)-(A) +1yr to +3yr
DISPFYI
Mean Median N
0.0955* 0.0549* 494
0.1243* 0.0545* 598
0.0257 -0.0033** 474
0.2007* 0.0946* 219
0.3432* 0.0737* 217
0.1519* 0.0726* 205
DISPLTG
Mean Median N
0.2752 0.2265 294
0.2423 0.2104 337
-0.0352** -0.0162** 273
0.4040 0.3116 156
0.4688 0.3296 162
0.0801 0.0582*** 143
SD_RET
Mean Median N
10.2540* 9.7651* 1,458
9.8064* 9.2765* 1,458
-0.4476* -0.3076* 1,458
8.7349* 8.1471* 488
11.3480* 10.2485* 488
2.6131* 1.7198* 488
SD_CFO
Mean Median N
0.0313* 0.0231* 308
0.0290* 0.0228* 425
-0.0019* -0.0014* 307
0.0315* 0.0232* 139
0.0294* 0.0213* 170
-0.0018 -0.0005 138
SD_ROA
Mean Median N
0.0139* 0.0093* 910
0.0118* 0.0091* 979
-0.0021* -0.0003* 854
0.0122* 0.0094* 336
0.0115* 0.0091* 356
-0.0009 -0.0003 324
SD_Sales
Mean Median N
0.0417* 0.0301* 1,063
0.0413* 0.0285* 1,149
-0.0001 -0.0003** 1,062
0.0409* 0.0282* 377
0.0388* 0.0271* 398
-0.0022 -0.0005 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.
33
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 Median N
0.0049* 0.0021* 1,176
-0.0004 0.0008* 1,182
-0.0089* -0.0028* 1,230
-0.0141* -0.0063* 1,171
-0.0087* -0.0040* 1,181
-0.0093* -0.0035* 390
-0.0174* -0.0051* 390
0.0026** 0.0014* 400
0.0120* 0.0064* 388
0.0199* 0.0040* 390
PAYOUT
Mean Median N
0.3381* 0.2780* 1,203
0.3459* 0.2682* 1,188
0.4497* 0.3451* 1,229
0.1117* 0.0576* 1,174
0.1138* 0.0509* 1,163
0.7108* 0.5613* 394
0.7544* 0.5505* 274
0.4995* 0.3942* 380
-0.1846* -0.1022* 362
-0.1993* -0.0155* 264
CAPEX
Mean Median N
0.0710* 0.0607* 1,023
0.0781* 0.0655* 1,015
0.0718* 0.0614* 1,030
0.0009 0.0012** 1,014
-0.0059* -0.0010* 1,013
0.0672* 0.0603* 336
0.0567* 0.0458* 333
0.0522* 0.0441* 335
-0.0145* -0.0104* 330
-0.0044* -0.0010*** 332
CashSTI
Mean Median N
0.0930* 0.0630* 1,198
0.0893* 0.0545* 1,211
0.0839* 0.0518* 1,251
-0.0097* -0.0061* 1,196
-0.0057* -0.0019* 1,211
0.0765* 0.0460* 395
0.0716* 0.0365* 398
0.0728* 0.0417* 415
-0.0037 -0.0010 393
0.0011 0.0013** 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
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) threefactor 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 1
Under 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 timeseries, 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 firmdividend 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: TCA j ,t = θ 0,t + θ1,t CFO j ,t −1 + θ 2,t CFO j ,t + θ 3,t CFO j ,t +1 + θ 4,t ∆REV j ,t + θ 5,t PPE j ,t + υ j ,t
(1)
4
Dividend initiations and omissions are not included in our sample as a result of these filters. 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. 5
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 timeseries standard deviation of the residuals, σ (υˆ )t , 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 FamaFrench three-factor model after incorporating the information risk factor returns from FLOS (2004) as follows: R j ,t − R f , t = α j + α j , ∆ DTi ,t + β j ( Rm,t − R f ,t ) + β j , ∆ DTj ,t ( Rm,t − R f ,t ) + s j SMBt + s j ,∆ DTi ,t SMBt + hj HMLt + hj , ∆ DTj ,t HMLt + e j IRt + e j , ∆ DTj ,t IRt + et
(2)
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, R p ,t − R f ,t , for each of the two portfolios for each month (a total of 372 months). The following multi-factor model is estimated for each portfolio:
R p ,t − R f ,t = α p + β p ( Rm ,t − R f ,t ) + s p SMBt + hp HMLt + e p IRt + et
(3a)
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: R p ,t − R f ,t = α p + α pdiv D + β p ( Rm ,t − R f ,t ) + β pdiv D *( Rm,t − R f ,t )
+s p SMBt + s pdiv D * SMBt + hp HMLt + hpdiv D * HMLt + ep IRt + epdiv D * IRt + et (3b) where D is an indicator variable for dividend-paying firms, β pdiv , s pdiv , hpdiv , e pdiv are the differences in factor loadings of dividend-paying firms on ( Rm,t − R f ,t ) , SMB, HML, and IR, respectively, and α pdiv 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: Ei ,t +1 = α i + λi Ei ,t + ε i ,t
(3c)
We test the difference in earnings persistence between dividend-paying and nondividend-paying firms using the following regression: Ei ,t +1 = α i + α pdiv Di ,t + λi Ei ,t + λ pdiv Di ,t ( Ei ,t ) + ε i ,t
(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 λ pdiv is consistent with dividend-paying firms having less information risk. Results reported in Panel A of Table 2 indicate that dividend-paying and nondividend-paying firms have distinct risk characteristics. Dividend-paying firms have higher market and HML betas, and smaller size betas, and the differences ( β pdiv , s pdiv , hpdiv ) between the loadings are significant at 5% or higher. More importantly, dividend-paying firms load negatively on the information risk factor whereas the nondividend-paying firms load positively on the information risk factor, and the difference between the loadings ( e pdiv ) is significant at 1% level. This suggests that dividendpaying 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 λ pdiv 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 firmannouncement 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) riskadjusted 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 dividenddecreasing 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 dividendincreasing firms are sensitive to the samples used. The results in Panel A, based on all 10,597 dividend announcements, generally support the hypothesis that dividendincreasing 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 nonoverlapping 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 dividendincreasing 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 longterm 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 dividenddecrease 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-existingearnings’ 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 shortterm 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 threefactor 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)
∆Div
Mean 29.66
Median 20.00
Standard Deviation 30.33
N 9,110
Mean -46.07
Median -50.00
Standard Deviation 16.62
N 1,487
Mean -75.73*
Median -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
αp
βp
sp
hp
ep
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*
α pdiv
β pdiv
s pdiv
hpdiv
e pdiv
Differences
0.4205*
0.0554**
-0.1100**
0.2356*
-0.6300*
Panel B: Earnings Persistence
αi
λi
Non-dividend Firms
-0.0067*
0.7040*
Dividend Firms
0.0145*
0.8761*
α pdiv
λ pdiv
0.0212*
0.1721*
Differences
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: R p ,t − R f ,t = α p + β p ( Rm ,t − R f ,t ) + s p SMBt + h p HMLt + e p IRt + et R p ,t − R f ,t = α p + α pdiv D + β p ( Rm ,t − R f ,t ) + β pdiv D * ( Rm ,t − R f ,t )
+s p SMBt + s pdiv D * SMBt + hp HMLt + hpdiv D * HMLt + ep IRt + epdiv D * IRt + et 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 NYSEAMEX-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: Ei ,t +1 = α i + λi Ei ,t + ε i ,t Ei ,t +1 = α i + α pdiv Di ,t + λi Ei ,t + λ pdiv Di ,t ( Ei ,t ) + ε i ,t
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
Table 3 Characteristics of Dividend Changes Panel A: Frequency of Dividend Changes Number of Divided Change Events
Number of Unique Firms
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 > 15
860 584 386 315 231 187 124 80 77 53 25 17 15 6 6 15
Total
2,981
Panel B: Industrial Classification Industry
Agriculture, Forestry and Mining Mining Construction Manufacturing Transportation and Communication Utilities Wholesale Retail Financial Services Public Administration Others
SIC
Number of Unique Firms Dividend Increase Dividend Decrease
01-09 10-14 15-17 20-39 40-48 49 50-51 52-59 60-69 70-89 90-97 98-99
Total
390 59 21 975 77 83 70 160 614 128 0 0 2,577
15.1% 2.3 % 0.8% 37.8% 3.0% 3.2% 2.7% 6.3% 23.8% 5.0% 0.0% 0.0%
51 9 4 127 10 48 9 17 114 14 0 1
12.6% 2.2% 1.0% 31.4% 2.5% 11.9% 2.2% 4.3% 28.2% 3.5% 0.0% 0.2%
404
Panel C: Magnitude of Dividend Changes Absolute Percentage Change
Number of Firm-event Pairs Dividend Increase Dividend Decrease
12.5 to 25 25 to 50 50 to 100 100 to 250 250 to 500
5,688 2,359 692 339 32
183 916 388 0 0
Total
9,110
1,487 30
Table 4 Changes in Factor Loadings Surrounding Dividend Change Announcements Panel A: All Dividend Announcement (n=10,597) All Dividend Increases Predicted Sign Alpha
α− j α j ,∆
n/a
N=9,110
Dividend Increase Quartile 1 N=2,374
Dividend Increase Quartile 2 N=2,293
Dividend Increase Quartile 3 N=2,166
Dividend Increase Quartile 4 N=2,277
0.8618*
0.5857*
0.8089*
0.9392*
1.1358*
-0.7219*
-0.4705*
-0.6880*
-0.7704*
-0.9784*
1.0212*
0.9783*
1.0154*
1.0460*
1.0479*
-0.0457*
-0.0144
-0.0593*
-0.0653*
-0.0440*
0.6764*
0.5509*
0.6494*
0.7246*
0.7896*
-0.0733*
-0.0087
-0.0501***
-0.1142*
-0.1250*
0.1937*
0.1642*
0.2041*
0.1616*
0.2401*
0.0061
0.0723*
-0.0102
0.0184
-0.0554**
-0.0226*
-0.0683*
-0.0235**
-0.0302**
0.0336**
-0.0257*
-0.0141
-0.0365**
-0.0136
-0.0373**
Dividend Decreases Predicted Sign
N=1,487
-0.2493* n/a
-0.1325**
Market factor
β− j
β j ,∆
−
0.8860* +
0.0973*
Size factor
s− j s j ,∆
−
0.6003* +
0.0056
Book to market factor
h− j h j ,∆
−
0.3852* +
0.1332*
Information risk factor
e− j e j ,∆
−
-0.0164 +
0.1028*
31
Table 4 (continued) Panel B: Non-overlapping Dividend Announcements (n=1,946) All Dividend Increases Predicted Sign Alpha
α− j α j ,∆
n/a
N=1,458
Dividend Increase Quartile 1 N=365
Dividend Increase Quartile 2 N=366
Dividend Increase Quartile 3 N=379
Dividend Increase Quartile 4 N=348
0.7169*
0.4966*
0.6432*
0.7458*
1.0087*
-0.8152*
-0.5278*
-0.6884*
-0.8673*
-1.2157*
0.9654*
0.9199*
0.9481*
0.9743*
1.0308*
-0.0460*
-0.0290
-0.0411
-0.0534
-0.0604
0.6936*
0.4783*
0.6291*
0.7820*
0.8948*
-0.0481
0.0870
-0.0311
-0.1277***
-0.1203
0.2431*
0.2229*
0.2793*
0.2163*
0.2471*
0.0871*
0.1232*
0.0549
0.1457*
0.0269
-0.0119
-0.0846*
-0.0001
-0.0203
0.0605
0.0141
-0.0031
-0.0251
0.0336
-0.0687
Dividend Decreases Predicted Sign
N=488
-0.5617* n/a
-0.0097
Market factor
β− j β j ,∆
−
0.8576* +
0.1250*
Size factor
s− j s j ,∆
−
0.5542* +
0.0294
Book to market factor
h− j h j ,∆
−
0.4336* +
0.1629*
Information risk factor
e− j e j ,∆
−
-0.0356 +
0.1595*
Notes: Model: R j ,t − R f ,t = α j + α j , ∆ DTi ,t + β j ( Rm ,t − R f ,t ) + β j , ∆ DT j ,t ( Rm ,t − R f ,t ) + s j SMBt + s j , ∆ DTi ,t SMBt + h j HMLt + h j , ∆ DT j ,t HMLt + e j IRt + e j , ∆ DT j ,t IRt + et 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.
32
Table 5 Information Characteristics Surrounding Dividend Change Announcements Dividend Increases (A) (B) -3yr to +1yr to 1yr +3yr
(B)-(A)
(A) -3yr to -1yr
Dividend Decreases (B) (B)-(A) +1yr to +3yr
DISPFYI
Mean Median N
0.0955* 0.0549* 494
0.1243* 0.0545* 598
0.0257 -0.0033** 474
0.2007* 0.0946* 219
0.3432* 0.0737* 217
0.1519* 0.0726* 205
DISPLTG
Mean Median N
0.2752 0.2265 294
0.2423 0.2104 337
-0.0352** -0.0162** 273
0.4040 0.3116 156
0.4688 0.3296 162
0.0801 0.0582*** 143
SD_RET
Mean Median N
10.2540* 9.7651* 1,458
9.8064* 9.2765* 1,458
-0.4476* -0.3076* 1,458
8.7349* 8.1471* 488
11.3480* 10.2485* 488
2.6131* 1.7198* 488
SD_CFO
Mean Median N
0.0313* 0.0231* 308
0.0290* 0.0228* 425
-0.0019* -0.0014* 307
0.0315* 0.0232* 139
0.0294* 0.0213* 170
-0.0018 -0.0005 138
SD_ROA
Mean Median N
0.0139* 0.0093* 910
0.0118* 0.0091* 979
-0.0021* -0.0003* 854
0.0122* 0.0094* 336
0.0115* 0.0091* 356
-0.0009 -0.0003 324
SD_Sales
Mean Median N
0.0417* 0.0301* 1,063
0.0413* 0.0285* 1,149
-0.0001 -0.0003** 1,062
0.0409* 0.0282* 377
0.0388* 0.0271* 398
-0.0022 -0.0005 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.
33
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 Median N
0.0049* 0.0021* 1,176
-0.0004 0.0008* 1,182
-0.0089* -0.0028* 1,230
-0.0141* -0.0063* 1,171
-0.0087* -0.0040* 1,181
-0.0093* -0.0035* 390
-0.0174* -0.0051* 390
0.0026** 0.0014* 400
0.0120* 0.0064* 388
0.0199* 0.0040* 390
PAYOUT
Mean Median N
0.3381* 0.2780* 1,203
0.3459* 0.2682* 1,188
0.4497* 0.3451* 1,229
0.1117* 0.0576* 1,174
0.1138* 0.0509* 1,163
0.7108* 0.5613* 394
0.7544* 0.5505* 274
0.4995* 0.3942* 380
-0.1846* -0.1022* 362
-0.1993* -0.0155* 264
CAPEX
Mean Median N
0.0710* 0.0607* 1,023
0.0781* 0.0655* 1,015
0.0718* 0.0614* 1,030
0.0009 0.0012** 1,014
-0.0059* -0.0010* 1,013
0.0672* 0.0603* 336
0.0567* 0.0458* 333
0.0522* 0.0441* 335
-0.0145* -0.0104* 330
-0.0044* -0.0010*** 332
CashSTI
Mean Median N
0.0930* 0.0630* 1,198
0.0893* 0.0545* 1,211
0.0839* 0.0518* 1,251
-0.0097* -0.0061* 1,196
-0.0057* -0.0019* 1,211
0.0765* 0.0460* 395
0.0716* 0.0365* 398
0.0728* 0.0417* 415
-0.0037 -0.0010 393
0.0011 0.0013** 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
