Dividend Initiations, Increases and Idiosyncratic Volatility

Transcription

1 Dividend Initiations, Increases and Idiosyncratic Volatility Bong Soo Lee Florida State University (850) Nathan Mauck University of Missouri - Kansas City mauckna@umkc.edu (816) ABSTRACT We examine three aspects of the relation between dividend initiation and increase announcements and idiosyncratic volatility. First, consistent with dividend signaling, we find that firms with higher levels of idiosyncratic volatility are associated with higher announcement abnormal returns when initiating or increasing dividends. Second, high idiosyncratic volatility firms are associated with stronger positive post event return drift. Finally, firms on average experience an ex post reduction in idiosyncratic volatility following dividend initiations that is associated with announcement and long-term abnormal returns. JEL classification: G30, G32, G35 Keywords: Dividends, Signaling, Volatility, Idiosyncratic Volatility 1

2 1. Introduction We conduct a comprehensive analysis of the link between dividend initiations and increases and firm idiosyncratic volatility in U.S. firms over the period Our theoretical motivation is based on the long established dividend signaling literature and on Eades (1982) in particular. While dividend announcements and idiosyncratic volatility have received some attention with respect to determinants of payout policy, considerably less is known about the relation between dividend events and idiosyncratic volatility in the context of abnormal returns as well as ex post volatility changes. In order to better understand the link between idiosyncratic volatility and dividend initiations and increases, we examine three aspects of the relation. First, we provide evidence on the link between idiosyncratic volatility and the market reaction to dividend initiations and increases, which builds on the model of Eades (1982). This analysis relates to the dividend signaling hypothesis. The literature has thoroughly examined the dividend signaling hypothesis, yet controversy remains about whether it holds empirically. Dividend signaling suggests that firms with higher levels of information asymmetry would be able to reduce such asymmetry through dividends (Modigliani and Miller, 1961). However, the literature has documented that such firms are less likely to pay dividends (Li and Zhao, 2008), suggesting the signaling hypothesis does not hold. We introduce firm idiosyncratic volatility as an information asymmetry proxy in the context of dividend changes. While high idiosyncratic volatility firms have been found to be less likely to pay dividends (Hoberg and Prabhala, 2009), little is known about the market reaction to dividends from such firms. In short, even though high information asymmetry and idiosyncratic volatility firms are less likely to pay dividends, a stronger positive reaction to dividend changes by such firms would be consistent with the signaling hypothesis. Our results indeed indicate a strong 2

3 positive link between dividend initiation and increase announcements and idiosyncratic volatility. Further, our results suggest that idiosyncratic volatility, and not systematic risk, drives the market response to dividends. Second, we examine the changes in idiosyncratic volatility following dividend initiations and increases and their link to announcement returns. Grullon et al. (2002) find that systematic risk declines in the three years following dividend increases. Similarly, Venkatesh (1989) finds that total volatility (i.e., standard deviation of daily returns) declines following dividend initiations. Dyl and Weigand (1998) find similar evidence and also establish a positive link between volatility decline and announcement returns for dividend initiations. However, we are the first to our knowledge to directly examine idiosyncratic volatility in this context. Using the definition of idiosyncratic volatility from Hoberg and Prabhala (2009) (i.e., standard deviation of residuals from daily market model), we find that idiosyncratic volatility declines following dividend initiations. We find similar results for dividend increases, but only when restricting the sample to large (i.e., greater than 12.5%) increases to dividends. Additionally, we find that firms with ex post declines in idiosyncratic volatility are associated with stronger positive abnormal returns at the time of announcement for dividend initiations. However, the market is unable to identify firms that experience volatility declines at announcement for dividend increases. Our results differ from prior literature in that we find that it is idiosyncratic volatility changes, not systematic risk changes, that best explain market reactions to dividends. Third, we test the degree to which idiosyncratic volatility and other information asymmetry proxies are related to the positive return drift documented by Michaely et al. (1995) and Benartzi et al. (1997). If one source of the drift is information asymmetry (i.e., difficult to measure the impact of an event at announcement for relatively high asymmetry firms), we would expect 3

4 relatively stronger longer-term abnormal returns for firms with high information asymmetry. Consistent with this explanation, we find that firms with above median levels of idiosyncratic volatility show between five and ten times the magnitude of positive drift in one and three-year long-term event studies for dividend initiations and increases. The results are similar for firm size and bid-ask spread which are our other information asymmetry proxies. Finally, we find that dividend event firms that realize ex post reductions in idiosyncratic volatility show much stronger positive long-term abnormal returns than firms with increases in idiosyncratic volatility. Our results regarding idiosyncratic volatility hold when controlling for other information asymmetry measures as well as a host of firm variables identified by the literature as related to payout policy. The results for other information asymmetry measures, namely firm size and bidask spread, confirm that information asymmetry is related to announcement abnormal returns and long-term abnormal returns. Our results add to the literature on dividend signaling. While firms with high information asymmetry have been found to be less likely to pay dividends, we find that when such firms do pay, the signal is quite strong. Our results are also related to the literature on idiosyncratic volatility. Consistent with Ang et al. (2006) who document a negative relation between idiosyncratic risk and return, we find that firms with reductions in idiosyncratic volatility show relatively stronger positive long-term abnormal returns following the dividend announcement. The rest of our paper is organized as follows. Section 2 presents the literature review and theoretical motivation. Section 3 presents our data and and methodology. Section 4 presents our results, and section 5 concludes. 4

5 2. Literature Review and Theoretical Motivation 2.1 Dividend Signaling Literature Dividend signaling theory is established by Modigliani and Miller (1961). They show that payout policy does not affect shareholder value and therefore conclude that payout policy is irrelevant. However, they recognize that changes in dividend policy are accompanied by abnormal returns surrounding the announcement date. They suggest that the information content of dividends may explain the result. Specifically, they argue that markets respond to changes in dividends because investors interpret the change in payout as a signal of information about firm value. This paper along with Lintner (1956) and Miller and Modigliani (1958) laid the foundations for dividend signaling theory. Dividend signaling theory suggests that there is asymmetric information between inside managers and outside investors. To resolve the difference in information sets, managers take actions to signal information to investors. Dividends are viewed as a signal of firm stability, health, and as an indication of relatively predictable future earnings. Bhattacharya (1979) shows that dividends represent costly signals. If a signal has no related cost then it may not be credible. John and Williams (1985) show that there is a theoretical signaling equilibrium for taxable dividends. This is driven by asymmetric information between insiders and outsiders. Miller and Rock (1985) provide a framework where inside managers know more about the firm and its prospects than outside investors. Ambarish et al. (1987) provide a model where firms signal with both dividend and investment decisions. Dividend signaling is usually linked to signals about future earnings. Healy and Palepu (1988) show that earnings changes precede and follow dividend changes. They further show that dividends forecast future earnings changes. DeAngelo et al. (1992) show that dividend policy has 5

6 information content. They show reduced dividends improve the ability of current earnings to predict future earnings. In addition to the evidence that firms signal future earnings with dividends, there is considerable evidence that firms use dividend changes to signal other attributes as well. Asquith and Mullins (1986) compare dividends, stock repurchases, and equity issues as communicators of firm value. They argue that dividends are credible signals and asymmetric information leads to signaling. Yoon and Starks (1995) show that investment policy information is not associated with dividend announcements. Dividend increases lead to capital expenditure increases. Dividend changes are associated with revisions in analyst earnings forecasts. Guay and Harford (2000), Jagannathan et al. (2000) and Lee and Rui (2007) show that dividends are related to permanent cash flows and repurchases to nonpermanent cash flows (see also Lee, 1996). The market interprets the payout action as a way to update opinions on cash flows. Nissim and Ziv (2001) show dividend changes provide information about profitability of firms in subsequent years. Jensen et al. (2010) show dividend reductions signal reductions in real growth options. Venkatesh (1989) shows the information content of earnings announcements decreases after dividend initiations. Dividend changes and earnings announcements are partial information substitutes. This implies that each event adds incrementally to the public information set for a firm and dividends reduce information asymmetries. Dewenter and Warther (1998) compare dividend payouts of U.S. and Japanese firms. Japanese firms are shown to have a lower announcement reaction and less asymmetric information. Japanese firms adjust dividends up and down more quickly. They are able to do this because there is less asymmetric information so the market reactions to these policy changes are diminished. Therefore, information asymmetries and announcement reaction are positively related. Amihud and Li (2006) suppose dividends are 6

7 disappearing due to reduced efficacy of dividends as a signal. They examine institutional ownership as a proxy for information asymmetry. More institutional ownership leads to lower information asymmetries. Cumulative abnormal announcement returns are shown to decrease in institutional holdings. Kale et al. (2012) examine the decision to pay dividends of newly listed firms and confirm that information signaling helps to explain payout decisions. More support for the role of dividends in reducing information asymmetries is found in Woolridge (1983). He shows that signaling is the primary factor influencing security prices around dividend change announcements. Michaely et al. (1995) use a one year post-event study to show initiations and omissions have drift distinct from earnings announcement drift. Denis et al. (1994) support the signaling hypothesis, confirming that dividends convey information. Taken together, these studies reveal that dividends convey information even in the presence of competing information generating events such as earnings announcements. However, the literature also documents evidence counter to the signaling hypothesis. DeAngelo et al. (1996) find no support that dividend changes can identify firms with superior future earnings. Dividends are not seen as good signals because they might incorporate behavioral bias of managers that overestimate firms prospects. Benartzi et al. (1997) show that earnings increase in years -1 and 0 of a dividend increase. Firms that cut dividends have reduced earnings in years -1 and 0, but increased earnings in year 1. This relationship does not support signaling as dividends should positively predict future earnings. Grullon et al. (2003) find that dividend changes contain no information about future earnings or profitability. Lie (2005) confirms that dividend omissions and reductions are not followed by reductions in earnings. Khang and King (2006) use insider trading as a proxy for information asymmetries. Dividends are shown to be negatively related to insider returns. Higher dividends are linked to lower insider returns. This does 7

8 not support signaling, and shows that high dividend paying firms have low information asymmetry. Li and Zhao (2008) show firms that are most subject to information asymmetries are less likely to pay, initiate, or increase dividends, which is counter to dividend signaling Theoretical Motivation Eades (1982) established the Relative Signaling Strength hypothesis. The hypothesis suggests that firms with higher volatility will experience larger abnormal announcement returns when compared to firms of the same true value but with lower volatility. The intuition behind the hypothesis is simple. Stock return volatility is shown to contribute to the cost of the signal, therefore, firms with higher volatility are less likely to pay dividends. When high volatility firms initiate or increase dividends, the signal is more meaningful. We propose a slight extension of this interpretation. Volatility may be used as a measure of asymmetric information; firms with high volatility have high asymmetric information, making issuing a signal much more effective because of the greater need of an information producing event. This interpretation is consistent with Dewenter and Warther (1998), who show that abnormal returns surrounding dividend announcements are positively related to asymmetric information. We believe that idiosyncratic volatility will be important to signaling strength because it is a measure of firm specific asymmetric information. Our intuition is that firms do not signal to resolve systematic informational asymmetries; rather, firms signal to resolve asymmetric information that relates specifically to them. Therefore, firms with higher idiosyncratic volatility will experience greater announcement returns because their signals will be more effective at closing the information gap. 1 Dividend signaling is, of course, not the only theory of payout policy. Elton and Gruber (1970) establish the dividend clientele effect. DeAngelo et al. (2006) suggest the dividend life-cycle theory. 8

9 Firms also signal risk changes with dividends. In Dielman and Oppenheimer (1984), betas are shown to decrease (increase) for firms that initiate (omit) or increase (decrease) dividends by 25%. Venkatesh (1989) shows total volatility decreases after dividend initiations, and most of this comes from decreases in firm specific volatility. Dyl and Weigand (1998) show firms that initiate dividends become fundamentally less risky. The reduction in risk is related to the size of announcement reaction. Grullon et al. (2002) show that firms that increase dividends experience a significant decline in systematic risk. Positive market reactions to dividend changes are related to future reduction in systematic risk and the largest declines in risk have stronger positive drift. Li and Zhao (2008) and Hoberg and Prabhala (2009) show that risk is an important component of the propensity to pay dividends. While the above papers do not specifically examine idiosyncratic risk, they do suggest an expectation that idiosyncratic risk will decline following dividend initiations and increases. One exception is Bessembinder and Zhang (2013), who examine long-run returns following corporate events and find that event firms differ from control firms in terms of idiosyncratic risk, among other firm factors. However, with respect to dividend initiating firms, they find that event firms have similar idiosyncratic volatility to benchmark firms. 2.3 Theoretical Model Volatility has been shown to be linked to asymmetric information. Jiang et al. (2009), He and Wang (1995), and Fishman and Hagerty (1989) describe the relation between asymmetric information and volatility. Firms that disclose less information experience relatively more heterogeneous beliefs. While firms leak positive information on a continual basis, they withhold negative information for as long as possible. The withholding of negative information by firms leads to larger idiosyncratic volatility. Thus, firms with higher idiosyncratic volatility have higher levels of asymmetric information. 9

10 Bhattacharya (1979) and Eades (1982) establish a model that allows not only the mean of the firm s value to influence signaling costs but also includes variance. Eades demonstrates theoretically that dividend yield and variance of returns should be negatively correlated. He also shows that signal strength and variance of returns should be positively correlated. where D = σ 2 (1) (f(d ) ), (1) 4 F(D ) F(D ) = cumulative distribution function of X evaluated at D, f(d ) = density function of X evaluated at D, X = Actual time 1 liquidation value with X~N(M, σ 2 ), D = Equilibrium dividends, and σ 2 = Firm risk. Equation (1) corresponds to Eades equation (6). Taking the derivative of equilibrium dividends with respect to firm risk yields a negative definite result on the left hand side. This is due to the fact that both f(d ) and F(D ) are positive. From this, Eades proves that signaling costs are a function of risk. Increased risk leads to increased signaling costs, which in turn lowers equilibrium dividends: 2 D M σ 2 = (1 4 ) ( 1 βσ 2) (f(d ) F(D ) ) (D M), (2) f(x a )~N(M, σ a 2 ) and f(x b )~N(M, σ b 2 ), (3) If σ a 2 < σ b 2. Then D a > D b given that M changes by an equivalent amount, where subscripts a and b denote values for two different hypothetical firms. Equation (2) corresponds to Eades equation (7). Taking the second derivative of equilibrium dividends with respect to mean firm value and then with respect to firm risk yields a 10

11 result driven by the relationship (D M). When equilibrium dividends are less than M, then the final term is negative. Thus, two firms with identical values for M but where risk is different will have different equilibrium dividends. The firm with higher risk will have lower dividends. Another way to view this relationship is to note that for equal changes in M, the higher risk firm will have a smaller change in dividends. Therefore, a firm with higher risk but identical M should see larger announcement returns holding change in dividends constant. 2 We extend the model by decomposing total risk into systematic risk (SYS) and idiosyncratic risk (IDIO), which does not alter the validity of the model and provides additional insight into the Relative Signaling Strength hypothesis. D = (SYS 2 +IDIO 2 ) (1) (f(d ) ), (4) 4 F(D ) 2 D M (SYS 2 +IDIO 2 ) = (1 4 ) ( 1 β(sys 2 +IDIO 2 ) ) (f(d ) F(D ) ) (D M), (5) r i = α i + β i (MKTRF) + ε i, (6) E[r i ] = α i + β i (MKTRF) (7) where σ 2 = VAR(r i ) = VAR(E[r i ]) + VAR(ε i ), (8) STD = VAR(r i ) = Total risk, (9) IDIO = VAR(ε i ), and (10) SYS = VAR(E[r i ]). (11) 2 For a more detailed analysis, we refer readers to Eades (1982). 11

12 Equations (4) and (5) are identical to equations (1) and (2) except that they replace the measure for total risk with a decomposition into systematic risk (SYS) and idiosyncratic risk (IDIO). Following Hoberg and Prabhala (2009), we use the market model to differentiate between the two types of risk. Equation (6) shows our estimation. Equation (7) is our expected return given the model. Equation (8) is the decomposition of total risk into two parts. We are able to make this distinction given that the expected value of return on stock i is orthogonal to the error term. Therefore, the variance of returns is equal to the sum of the variance of expected returns and the variance of the error term. The usual interpretation of Eades model is that volatility is a proxy for risk. We propose that additional insight can be gained when volatility is also allowed to proxy asymmetric information. This is consistent with Dierkens (1991) who shows information asymmetry explains equity issue announcement returns. Higher asymmetric information leads to stronger negative reactions, and asymmetric information fluctuates significantly through time. Proxies that can be measured over short periods of time are preferable. Idiosyncratic volatility then may be an ideal information asymmetry proxy. Interpreting volatility as a measure of asymmetric information, in addition to a measure of risk, strengthens the intuition of Eades hypothesis that volatility is positively related to announcement returns. This is the case because higher volatility implies greater information asymmetry, which in turn implies a greater need to signal. We also would expect this given two firms with identical true value. If firm A has less asymmetric information than firm B, firm A will have a lower announcement return than firm B given that both have an identical dividend change. Traditional finance theory suggests that systematic risk is the only risk that matters to returns. Asset pricing theory is unclear on the impact that idiosyncratic volatility should have on 12

13 event returns. We believe that our results are significant in that we show idiosyncratic volatility does matter, which is consistent with recent empirical findings (e.g., Ang et al., 2006; Jiang and Lee, 2006; Fu, 2009; Huang et al., 2010). 3. Data and Methodology 3.1 Dividends We focus on two dividend variables, divided initiations and dividend increases. While Officer (2011) notes that dividend increases may be less informative than dividend initiations, we nonetheless consider both increases and initiations consistent with Kothari et al. (2009), Hoberg and Prabhala (2009), and Charitou et al. (2011), among others. We define dividend increases as positive dividend percent changes [Div(t) Div(t-1)]/Div(t-1). CRSP data is used to identify dividend announcement dates and dividend changes over the period Our data is subject to the following screens based on the literature in order to identify significant events, avoid confounding effects of earnings announcements, and ensure compatibility with prior studies 3 : 1) Only regular quarterly dividends (CRSP code 1232) are included. 2) Only ordinary common shares are included (CRSP codes 10, 11, and 12). 3) There have been no missed dividend payments or changes in dividend payments in the prior year. 4) The dividend percent change is at least 1% and not greater than 500%. 5) The dividend announcement date is at least three days from an earnings announcement date. 3 See for example: Lee and Yan (2003), Kothari et al. (2009), and Charitou et al. (2011). 13

14 6) At least 65 daily return observations are available prior to and after the dividend announcement date. This ensures enough days for both the event study results as well as the change in idiosyncratic volatility results. 7) Financial firms and utilities are excluded. The sample restrictions we impose ensure that our sample includes firms comparable to the prior literature and that neither outliers nor earnings announcements drive the results. Our sample includes 11,870 dividend increases and 1,691 dividend initiations. Our sample size, when adjusting for different sample years, is comparable with prior studies. 3.2 Announcement Abnormal Returns We establish cumulative abnormal returns (CARs) for the announcement period for event days -1 to +1 for dividend initiations and increase, separately. The market model is estimated for the period beginning one year prior to announcement until 45 days prior to the announcement to establish normal returns. Abnormal returns are the actual daily firm return less the predicted return based on the market model regression. In our regressions CAR is the dependent variable and we follow Officer (2011) in constructing control variables. Specifically, our model is: CAR i = i + IV i + MV i + BA SPREAD i + DIV i + YIELD i + Q i + CF i + REP IND i + ROA i + RE/TE i + IV i + SYST i + e i, (12) where CAR is the cumulative abnormal return. Announcement CARs are calculated for the window beginning one day before the announcement and ending one day after the announcement. The market model is used to estimate normal returns and abnormal returns are the daily realized returns less the predicted normal return. IV is the idiosyncratic volatility of the announcing firm. IV is calculated as the standard deviation of residuals from a market model regression for each 14

15 firm in the one year prior to the announcement using daily data. MV is the natural log of market value of the firm prior to the announcement. BA SPREAD is the bid-ask spread. It is calculated as the average of the daily closing bid-ask spread in the year prior to announcement. ΔDIV is the size of the dividend change in percent. YIELD is the dividend yield. Q is total assets plus fiscal-yearend market value of equity minus book value of equity minus balance sheet deferred taxes all divided by total assets (as in Officer, 2011 and Kaplan and Zingales, 1997). Data on this and all accounting variables is from Compustat. CF is cash flow from operations and is created following Bushman et al. (2008). Specifically, CF is earnings before extraordinary items plus depreciation and amortization minus working capital accruals all divided by total assets. Working capital accruals are defined as the change in current assets minus the change in cash holdings minus the change in current liabilities plus the change in short-term debt plus the change in tax payable. REP IND is a repurchase indicator equal to one if the firm repurchased stock in the five years prior to the event and is zero otherwise. ROA is return on assets and is measured as income before extraordinary items plus interest expense and income statement deferred taxes all divided by total assets. RE/TE is retained earnings divided by book value of equity. ΔIV and ΔSYST is the yearly change in idiosyncratic risk and systematic risk, respectively. We follow Grullon et al. (2002) in defining systematic risk as the standard deviation of the product of beta and daily market return over the period. Following Petersen (2009), robust standard errors which cluster by firm and include time fixed effects are used in all specifications. 3.3 Long-term Abnormal Returns Long-term abnormal returns are calculated using the Fama and French (1993) three-factor model augmented for momentum. We form equally-weighted calendar time portfolios for firms 15

16 initiating or increasing dividends, separately, in the past 12 or 36 months depending on the horizon of interest. We then run the following regression for the time series of portfolio returns: (R i,t R f,t ) = α j + b j (R m,t R f,t ) + c j SMB t + d j HML t + m j MOM t + ε i,t, (13) where R i,t is the monthly return of firm i in calendar month t. R f,t and R m,t are the risk-free rate (one-month T-Bill) and the return on the market (equally weighted CRSP return index) for calendar month t, respectively. SMB t, HML t and MOM t are the return on the size, book-tomarket, and momentum factors in calendar month t, respectively. The intercept is the monthly abnormal return. Portfolios are formed for the full sample as well as numerous bifurcations based on information asymmetry proxies. 4. Results Tables 1 and 2 present summary statistics for our sample. In Table 1, we find that the mean and median CAR for dividend initiations (dividend increases) is 2.66% and 1.87% (1.82% and 1.54%), respectively. The mean and median CAR is significant at the 1% level in both cases. The sign, significance and magnitude of the CARs, as well as the sample size is consistent with prior studies (Asquith and Mullins, 1983; Kothari et al., 2009; Officer, 2011). [Insert Table 1 Here] Table 2 presents summary statistics for variables used in regressions. We note that the bidask spread limits the sample considerably (we lose over 50% of the initiation sample) due to lack of availability in the early years of our sample. Similarly, CF imposes some constraints on the sample as about 30% of the dividend increase sample is lost due to its inclusion. [Insert Table 2 Here] 16

17 Table 3 presents regression results in which the announcement CAR is the dependent variable and proxies for information asymmetry are the independent variables. In both the dividend initiation and dividend increase samples, we find that IV is positively related to announcement CAR (significant at the 1% level). This result holds even when controlling for other information asymmetry proxies (i.e., firm size and bid-ask spread). 4 Thus, the greater the firm information asymmetry, as measured by IV, the greater the positive response to dividend initiations and increases. This suggests that positive changes to dividends are particularly strong signals in the case of firms with relatively high levels of information asymmetry. This result is consistent with Dewenter and Warther (1998), although we are the first, to our knowledge, to show the link between IV and announcement CARs in the context of dividend initiations and increases. Li and Zhao (2008) find that firms with higher levels of information asymmetry are less likely to pay dividends. Similarly, Hoberg and Prabhala (2009) find that decreasing propensity of firms to pay is largely driven by increases in idiosyncratic volatility. This lower propensity to pay may further explain the strong positive response to dividend initiations and increases in such firms. Specifically, if the probability of dividend paying of high information asymmetry firms is low, the paying of dividends by such firms is a particularly strong signal. Firm size is another potential information asymmetry proxy. Consistent with the IV results, the MV results indicate that firms with higher levels of information asymmetry are associated with relatively more positive announcement CARs. In particular, firm size is inversely related both to announcement CARs and firm information asymmetry. Still, IV captures information not contained in MV as IV remains significant when MV is included. The economic significance for IV and MV is nearly identical as 4 We considered but did not include proxies for information asymmetry based on analyst estimates. Preliminary analysis indicated that over 50% of the sample is lost due to including such variables and they are strongly correlated with the well-populated measure of firm size. 17

18 a one standard deviation increase in either is associated with a roughly 1% increase in CAR. However, we note that IV is relatively more dynamic compared to MV, which means that IV may provide a more up to date measure of information asymmetry. The results for the bid-ask spread are mixed as BA SPREAD is insignificant (marginally significant) in the specification which includes all three information asymmetry proxies at the same time for dividend initiations (dividend increases). Overall, the results in Table 3 indicate a positive relation between information asymmetry and announcement CARs for dividend initiating and dividend increasing firms. [Insert Table 3 Here] It may be that the observed results in Table 3 are due to factors other than information asymmetry. In order to control for this possibility, we follow Officer (2011) in the construction of our control variables in Table 4. The control variables proxy for the size of the dividend change, the dividend yield, Q, cash flow, a repurchase indicator, return on assets, and return on equity divided by total equity. Despite the inclusion of control variables which are designed to capture factors related to firm payout, the results in Table 4 show that IV remains positive and significant (at the 1% level) in all eight specifications. Similarly, MV is negative and significant (at the 1% level) in all specifications. Thus, even after controlling for other factors related to the payout decision, firm information asymmetry is positively related to announcement CARs. We note that the inclusion of additional control variables renders BA SPREAD statistically insignificant. The control variables are generally not significant. The exceptions include the size of the change in dividends and dividend yield in the dividend increase sample. Both the size of the dividend change and the dividend yield are positively related to announcement CARs. Our control variable results are consistent with Officer (2011). 18

19 [Insert Table 4] Table 5 presents results which examine the change in measures of risk and information asymmetry in the year following the dividend event. Grullon et al. (2002) document that systematic risk falls following dividend increases. Venkatesh (1989) finds that total volatility (standard deviation of returns) drops following dividend initiations. The paper notes that this is mostly due to firm specific (i.e., idiosyncratic) volatility. However, the result for firm specific volatility is based on a three day window which does not directly measure idiosyncratic volatility. Rather, the paper shows that total risk drops and beta doesn t explain much of this drop which leads the author to conclude that firm risk must drive the results. Similarly, Dyl and Weigand (1998) find that volatility drops following dividend initiation, although idiosyncratic risk is not examined. Outside of the volatility literature, Kothari et al. (2009) find that firm information asymmetry declines following dividend initiation. While the prior literature suggests a decline in idiosyncratic volatility is likely following the dividend events examined in this paper, the result has not been documented to the best of our knowledge. The dividend initiation results in Table 5 are consistent with predictions from the literature. Specifically, systematic and idiosyncratic risk drop in the one year following the dividend initiation. The statistical significance of the decline in idiosyncratic volatility is particularly strong and nearly 60% of all initiating firms see a decline in idiosyncratic volatility. Thus, the initiation of dividends appears to reduce firm information asymmetry/firm risk. The results for the bid-ask spread are statistically insignificant. [Insert Table 5 Here] 19

20 The dividend increase results initially appear counter to the results for dividend initiations. Specifically, the dividend increase sample reveals an increase to systematic risk and no change to idiosyncratic volatility. The proportion of firms showing a decline in both measures of risk is roughly 50/50. The bid-ask spread results are conflicting with a mean (median) decrease (increase) and a roughly 50/50 split between decreasing/increasing occurrence. These results appear to be counter to Grullon et al. (2002) who document volatility declines following dividend increases. Our results and those of Grullon et al. (2002) can be reconciled by noting that the minimum dividend increase percentage in Grullon et al. (2002) is 12.5% while our minimum is 1%. In unreported results, we restrict our sample to only dividend increases greater than or equal to 12.5% and find results that are qualitatively identical to Grullon et al. Nonetheless, dividend increases appear to be less volatility reducing than initiations. This is perhaps not surprising, considering that moving from non-payer to payer is a more significant strategic change than already being a payer and increasing the dividend size, particularly for relatively small increases. Figure 1 presents the average changes in idiosyncratic volatility (as a percentage) for dividend initiations and increases for announcements in a given year. Specifically, for all announcements in a given year, we take the equal-weighted average of the ex post one year change in idiosyncratic volatility. The trend for initiations and increases is very similar. Of interest, we note that the strongest declines are in periods of high uncertainty. Specifically, the NBER defined recession periods of 1970, 2001, and most recently as well as the Black Monday period beginning October 1987 are associated with the largest declines in idiosyncratic volatility. Other NBER recession periods demonstrate little significant pattern. Nonetheless, the figure indicates that dividend initiations and increases are particularly strong signals in times of high uncertainty. 20

21 This is consistent with Fuller and Goldstein (2011) who find that dividends matter more in declining markets. [Insert Figure 1 Here] Regardless of the experience of the average firm with respect to volatility/information asymmetry changes, we would expect that larger declines in such measures would be positively related to announcement CARs. Indeed, Grullon et al. (2002) find that changes in beta are negatively related to CARs (i.e., a decline in beta is associated with economically greater CAR) and Dyl and Weigand (1998) find that total volatility changes are negatively related to CARs. We extend this analysis to include idiosyncratic volatility in Table 6. First, we note that including the change in idiosyncratic volatility and/or the change in systematic risk does not change our conclusions with respect to the level of IV prior to the event in the regressions (significant at 1% in 11 of 12 specifications and at 5% in 1 of 12 specifications). 5 Second, the change in idiosyncratic volatility is related to announcement CARs in all 12 specifications, although the direction of the relation depends on whether the dividend event is an initiation or an increase. Specifically, change in idiosyncratic volatility is negatively (positively) related to announcement CARs for dividend initiations (dividend increases). Thus, a reduction in volatility is associated with a relatively more positive (negative) CAR in the initiation (increase) sample. The initiation results are consistent with the negative relation between idiosyncratic volatility and returns documented by Ang et al. (2006). The interpretation of the dividend increase results is not as straightforward. Table 5 demonstrates that in the full sample of dividend increases, idiosyncratic volatility is on average the same in the year following the increase as it was in the 5 As changes in IV are related to levels of IV, we orthogonalize IV levels in specifications which include both IV measures. 21

22 year prior. Li and Zhao (2008) find that dividend paying firms have lower information asymmetry than non-payers. Combining these facts, dividend increasing firms already have relatively low information asymmetry which was reduced upon dividend initiation and dividend increases do little to change this fact. Thus, a lack of relation between idiosyncratic volatility and announcement CARs for dividend increasing firms is not surprising. However, the inverse of the initiation result is surprising. It may be that the market is not able to identify firms with volatility reductions at the time of the dividend increase, but subsequent firm performance may still indicate a positive relation between idiosyncratic volatility reduction and returns. We defer further discussion to the long-term event study results in Table 7. Third, the change in systematic risk is not related to announcement CARs when controlling for change in idiosyncratic risk. Thus, consistent with earlier results in this paper, the results suggest that changes in idiosyncratic volatility are more informative about dividend signaling than changes in systematic risk. [Insert Table 6 Here] Long-run positive abnormal returns (i.e., drift) following dividend initiations and increases have been established by the literature (Michaely et al. 1995; Benartzi et al. 1997). While high information asymmetry firms see higher announcement returns following dividend initiations and increases, it may be that such reactions are not large enough to fully reflect the information of the event. In short, drift may be more likely in firms with high information asymmetry as predicting the future for such firms may be especially difficult. The results of Table 6 which show that the market does not on average identify idiosyncratic volatility reducing dividend increasing firms at announcement is further evidence that projecting the future for such firms is difficult. Table 7 presents the results of our long-term event study. We form calendar time portfolios for each month beginning the month after announcement. Portfolio returns are regressed on the 22

23 Fama and French factors and a momentum factor. The intercept, alpha is the monthly abnormal return. We report the annualized abnormal return for one and three year event studies for dividend initiations and increases as well as bifurcations based on levels of idiosyncratic volatility, firm size, bid-ask spread, and changes in idiosyncratic volatility. The full sample results in Table 7 are consistent with prior literature (i.e., Michaely et al. 1995; Benartzi et al. 1997; Charitou et al. 2011) in that we find a positive drift following dividend initiations and increases. The magnitude of the drift is greater for the dividend increase sample, which indicates that the market particularly underreacts to these announcements. We split the sample of dividend initiation and increase firms into two groups based on the median value (except for changes in idiosyncratic volatility) of various firm characteristics. Portfolios are then formed for each sub-sample. The first such split is on the level of idiosyncratic volatility prior to the event announcement. In both the dividend initiation and increase sample we find that the economic magnitude of the drift is much stronger (i.e., five to ten times the annualized abnormal return) for firms with higher levels of idiosyncratic volatility. Idiosyncratic volatility is positively related to announcement abnormal returns, however, the reaction appears to be insufficient to fully incorporate the information of the dividend event for firms with greater idiosyncratic volatility. This result is similar to what we find in Table 7 in the sub-sample analysis based on firm size and firm bid-ask spread. Specifically, firms with higher levels of information asymmetry show an economically stronger drift. However, the magnitude of the difference between groups is clearly strongest for the idiosyncratic volatility split. Thus, idiosyncratic volatility seems to best capture information asymmetry in the case of dividend initiations and increases. 23

24 The final split we evaluate is based on whether or not ex post idiosyncratic volatility declined or increased. In Table 5, we found that idiosyncratic volatility drops on average for dividend initiation firms and for firms with large dividend increases, but not for firms with small dividend increases. Table 6 shows that the market reacts more (less) positively to dividend initiations (increases) at the announcement of the event. In short, the market appears able (unable) to identify firms likely to see an idiosyncratic volatility decline following dividend initiation (dividend increase). In both the dividend initiation and increase sample, we find an economically much stronger positive drift for firms with ex post idiosyncratic volatility declines. Thus, both samples appear consistent with Ang et al. (2006), who document a negative relation between idiosyncratic volatility and returns. The fact that the magnitude of the drift is stronger for dividend increases than initiations is consistent with the market being unable to identify idiosyncratic volatility decreasing firms during such events. [Insert Table 7 Here] 5. Conclusion Dividend signaling has received a great deal of attention in the literature, but the empirical evidence remains mixed. Firms with high levels of information asymmetry should have a greater need to signal. However, such firms have been found by the literature to be less likely to pay, which suggests that signaling theory does not hold. Yet, it may be that although high information asymmetry firms are less likely to pay, when they do pay the market reaction is stronger. In other words, dividends do serve as a signal. We find empirical evidence of this as firms with higher levels of idiosyncratic volatility and smaller firms are associated with more positive market reactions at dividend initiation and increase announcements. 24

25 We find further evidence of signaling in our results which show that idiosyncratic risk declines following dividend initiations. Specifically, firms that initiate dividends are associated with declines in information asymmetry. Such declines are positively (negatively) related to dividend initiation (increase) announcement returns. Thus, the market is able (unable) to identify firms likely to experience ex post declines in idiosyncratic volatility for dividend initiation (increase) events. Finally, we examine the post dividend event drift. Our results indicate that the economic magnitude of the drift is between five and ten times larger for dividend initiations and increases for firm with higher levels of idiosyncratic volatility. Thus, higher information asymmetry is associated with stronger positive post event drift. Firms that experience declines in idiosyncratic volatility are associated with particularly strong positive post event drift. Overall, our results provide evidence for dividend signaling. Further, we offer new results on the relation between idiosyncratic volatility and dividends. Further, we provide a theoretical extension in this context which establishes idiosyncratic volatility as a timely and relevant proxy for firm information asymmetry. 25

26 Appendix A - Variables Variable CAR Definition The cumulative abnormal return. Announcement CARs are calculated for the window beginning one day before the announcement and ending one day after the announcement. The market model is used to estimate normal returns and abnormal returns are the daily realized returns less the predicted normal return. IV The idiosyncratic volatility of the announcing firm. IV is calculated as the standard deviation of residuals from a market model regression for each firm in the one year prior to the announcement using daily data. MV BA SPREAD Δ DIV YIELD Q The natural log of market value of the firm prior to the announcement. The bid-ask spread. It is calculated as the average of the daily closing bid-ask spread in the year prior to announcement. The size of the dividend change in percent. The dividend yield. Total assets plus fiscal-year-end market value of equity minus book value of equity minus balance sheet deferred taxes all divided by total assets. CF Cash flow from operations and is created following Bushman et al. (2008). Specifically, CF is earnings before extraordinary items plus depreciation and amortization minus working capital accruals all divided by total assets. Working capital accruals are defined as the change in current assets minus the change in cash holdings minus the change in current liabilities plus the change in short-term debt plus the change in tax payable. REP IND ROA A repurchase indicator equal to one if the firm repurchased stock in the five years prior to the event and is zero otherwise. Return on assets measured as income before extraordinary items plus interest expense and income statement deferred taxes all divided by total assets. RE/TE Δ IV Δ SYST Retained earnings divided by book value of equity. The yearly change in idiosyncratic risk. The yearly change in systematic risk where systematic risk is the standard deviation of the product of beta and daily market return. 26

27 Figure 1 Time Series of IV Changes and Dividend Initiations and Increases This figure presents the equal-weighted average percentage decline in idiosyncratic volatility for dividend initiation or increasing firms over the period Dividend Initiations - ΔIV Dividend Increases - ΔIV 27

28 REFERENCES Ambarish, Ramasastry, Kose John, and Joseph Williams Efficient signaling with dividends and investments. Journal of Finance 42(2): Amihud, Yakov, and Kefei Li The Declining Information Content of Dividend Announcements and the Effects of Institutional Holdings. Journal of Financial and Quantitative Analysis 41(3): Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang The cross section of volatility and expected returns. Journal of Finance 61(1): Asquith, Paul, and David Mullins The impact of initiating dividend payments on shareholder s wealth. Journal of Business 56: Asquith, Paul, and David Mullins Signaling with dividends, stock repurchases, and equity issues. Financial Management 15(3): Benartzi, Shlomo, Roni Michaely, and Richard Thaler Do changes in dividends signal the future or the past? Journal of Finance 52(3): Bessembinder, Hendrik, and Feng Zhang Firm characteristics and long-run stock returns after corporate event. Journal of Financial Economics 109: Bhattacharya, Sudipto Imperfect information, dividend policy, and the bird in the hand fallacy. The Bell Journal of Economics 10(1): Bushman, R., A. Smith, and F. Zhang Investment-cash flow sensitivities are really investment-investment sensitivities. Working Paper, University of North Carolina. Charitou, Andreas, Neophytos Lambertides, and Giorgos Theodoulou Dividend increases and initiations and default risk in equity returns. Journal of Financial and Quantitative Analysis 46: DeAngelo, Harry, Linda DeAngelo, and Douglas Skinner Dividends and Losses. Journal of Finance 47(5): DeAngelo, Harry, Linda DeAngelo, and Douglas J Skinner Reversal of fortune dividend signaling and the disappearance of sustained earnings growth. Journal of Financial Economics 40(3): DeAngelo, Harry, Linda DeAngelo, and Rene M. Stulz Dividend policy and the earned/contributed capital mix: a test of the life-cycle theory. Journal of Financial Economics 81(2):