Do Dividends Matter more in Declining Markets?

Transcription

1 Do Dividends Matter more in Declining Markets? Kathleen Fuller Terry College of Business, University of Georgia, Athens, GA Michael Goldstein* Finance Department, Babson College, Babson Park, MA October 26, 2004 Abstract Using S&P 500 monthly returns as a proxy for market conditions, we find that dividend-paying stocks outperform non-dividend-paying stocks by more in declining markets than in advancing markets, implying that investors asymmetrically prefer dividends in down markets. These results are robust to risk adjustments using the CAPM, size and book-to-market quartiles, volume, the Fama-French three-factor model, and Fama-McBeth style regressions. Our results also suggest that is the existence of dividends, and not the level of the dividend, drives returns asymmetric behavior relative to market movements. We conclude that a signaling theory explanation is more consistent with our results than a prospect theory explanation. JEL Classification Code: G35 Keywords: Dividend policy, asymmetry, market movements *Corresponding author: Michael Goldstein, Finance Department, 223 Tomasso Hall, Babson College, Babson Park, MA tel: (781) fax: (781) Goldstein@babson.edu. We thank Deepak Agarwal, Jeff Bacidore, Jennifer Bethel, Wayne Ferson, Paul Irvine, Jon Karpoff, Gautam Kaul, Laurie Krigman, Marc Lipson, James Mahoney, Donna Paul, Tyler Shumway, John Scruggs, and Chris Stivers. Kathleen Fuller acknowledges financial support from the Terry-Sanford Research Grant. Michael Goldstein acknowledges support from the Babson College Board of Research.

2 Do Dividends Matter more in Declining Markets? Previous theoretical and empirical research indicates investors preferences for dividends vary across shareholder types. For example, depending on their tax bracket, some shareholders may prefer high dividend-paying stocks while others may prefer non-dividend-paying stocks. Anecdotal evidence also suggests that investors preferences for dividends may vary over time. For example, in 2000, Fidelity began airing advertisements for a mutual fund consisting of only dividend-paying stocks in which an advisor informs his client this fund would help diversity his portfolio and moderate losses in down markets. Also in 2000, Standard & Poor s predicted a revived interest in dividends, stating market weakness may boost interest in dividends as investors begin to see the value of a bird in the hand. These statements imply that investors preferences for dividend-paying stocks over non-dividend-paying stocks vary over time conditional on the state of the market, i.e., advancing and declining markets. There are two reasons why investors might condition their preference for dividend-paying stocks on the state of the market: prospect theory and signaling theory. 1 Prospect theory, developed by Kahneman and Tversky (1979), indicates that people respond differently to certain verses probabilistic gains and losses and care more about losses than they do to gains. Prospect theory suggests therefore that investors may prefer the cash from dividend-paying stocks more when they predict future uncertainty or economic downturns, and less when the market is doing well. 2 Dividend-paying stocks provide a return where at least part of the return is a certain gain over those non-dividend-paying firms for which the entire gain or loss is uncertain. In markets that are moving upward, investors, while still valuing the relatively more certain gain, may value it less so since the preference for loss avoidance is mitigated. During periods of market decline, however, investors prefer dividends as a cushion to their returns, particularly if 1 A third reason why investors would prefer cash dividends, the theory of self-control, was developed by Thaler and Shefrin (1981). As discussed in Shefrin and Statman (1984), investors may want to consume dividend payments so to keep from consuming their long-run wealth (i.e., consuming capital). However, the theory of self-control cannot explain differential investor preferences based on market movements. 2 Markowitz (1952, 1959) noted that investors may care about downside risk differentially. Other theories, such as first-order risk aversion utility functions in Gul (1991), provide a similar result. 1

3 they are downside risk-averse. Such responses are similar to the flight to quality tendency that is seen during market declines documented by Connolly, Stivers, and Sun (2004). To the extent that investors value dividends as a certain return, investors may move from risky to less risky investments, in this case from non-dividend-paying to dividend-paying firms. Another possibility relates to the signaling nature of dividends. Previous research suggests dividend-paying firms are better able to signal managers expectations than non-dividend-paying firms. 3 Again, this ability to signal may be more valuable in declining markets than in advancing markets. In declining markets, dividend-paying firms can signal positive information by just maintaining dividend payments; in such markets the prior commitment to pay dividends is more likely to be binding, increasing the value and importance of the signal. In advancing markets, it is expected that firms will generally perform well and the probability of, and costs associated with, financial distress are lower. The prior commitment to pay is therefore less likely to be binding in advancing markets and thus the value of dividend signals is likely to be lower. It is possible, therefore, that a policy of paying dividends during advancing markets conveys less information than a policy of paying dividends during declining markets. Overall, because non-dividend-paying firms do not credibly provide this signal, investors may discover that their state preferences for dividend payments cause them to value dividend-paying firms higher than non-dividend-paying firms during market downturns. Both prospect theory and signaling theory suggest that the extent of investors preferences for dividend-paying stocks over non-dividend-paying stocks should be stronger in declining markets than in advancing markets. In this paper, we examine if investors have a preferences for dividend-paying firms in declining markets and determine whether prospect or signaling theory is the more likely explanation for this preference. Using S&P 500 returns as a proxy for market conditions, we examine the return 3 Dividends either signal managers private information regarding future earnings [e.g., Bhattacharya (1979), John and Williams (1985), and Miller and Rock (1985)] or signal that managers will not waste excess cash [e.g., Easterbrook (1984), Jensen (1986), and Lang and Litzenberger (1989)]. As noted in Allen and Michaely (2004), empirical tests of these two hypotheses fail to pick an overwhelming winner. Brav, Graham, Harvey, and Michaely (2004) find that managers believe dividends do signal information but are not sure exactly what is being signaled. We do not attempt to distinguish between these two theories of dividend signaling. 2

4 behavior of dividend-paying and non-dividend-paying firms in both up and down markets from January 1970 to December We find that dividend-paying firms outperform non-dividend-paying firms by more in down markets than they do in up markets, implying that dividends do provide a differential benefit depending on market conditions. Our finding that the magnitude of the outperformance of dividend-paying stocks over non-dividend-paying stocks depends on market conditions is robust to a variety of adjustments for risk, including the CAPM, size and book-to-market quartiles, the Fama-French three-factor model, and Fama-McBeth style time-varying regressions, as well as various controls for size, time period, and alternative definitions of up and down markets. Our results are also robust to a number of other controls. For example, these results are robust to exchange listing (NYSE vs. NASDAQ), indicating different market structures as well as the inherent differences across NYSE-listed and NASDAQ-listed stocks do not drive our results. To verify that these results are not primarily due to small stocks, we truncate the sample by examining only the middle 50% of the stocks based on market capitalization and still find that dividend-paying stocks outperform nondividend-paying stocks by more in down markets than in up markets. To verify that down markets do not just proxy for expectations of increased future overall market risk, we segment our data based on estimates of implicit volatility from S&P stock options and find that dividend paying stocks continue to outperform non-dividend-paying stocks by more in down than up markets. We also find that these results hold across different sub-periods. Finally, our finding that dividends matter more in declining markets is not driven by the cash payment itself. Kalay and Michaely (2000) note that time series variation may be related to actual dividend payments itself. They note that Black and Scholes (1974) and Litzenberger and Ramaswamy (1979, 1980, and 1982) have different results primarily due to differences in how they classify a dividendpaying stock: while Black and Scholes ascribe a stock as dividend paying even during months during which a dividend is not being paid, Litzenberger and Ramaswamy do not. Instead, for quarterlydividend-paying stocks, Litzenberger and Ramaswamy classify the stock as a non-dividend paying for 3

5 eight months that a dividend is not paid and as a dividend-paying stock for the other four ex-dividend months. To verify that our results are not driven by the cash payment, we examine the eight months of the year when dividend are not paid by dividend-paying stocks. We find dividend-paying stocks outperform non-dividend-paying stocks by more in down markets than in up markets even during the months when the dividend-paying firms did not pay a dividend, indicating that it is not the receipt of the cash itself that is driving these results. While these broad results are consistent with both a prospect theory explanation and a signaling theory explanation, a number of tests indicate that the signaling theory explanation is more likely. First, we find that the different market reaction in up and down markets is due to the dividend-paying or nondividend-paying nature of the stock and does not vary with dividend yield, implying the ability to pay a dividend and not how much is the cause of the differential performance across market conditions. Thus, the dividend yield itself is not driving the results; just the payment of a dividend results in dividendpaying firms outperforming non-dividend paying firms in down markets. The insignificance of the yield level is more consistent with a signaling theory explanation. Signaling theory focuses more on the ability to signal and not the previously established level while prospect theory would suggest that a larger amount of cash should matter more. Second, while our results hold for all stocks, the results are more pronounced for smaller stocks, even after all risk adjustments. While prospect theory would not suggest a preference across larger and smaller stocks, signaling theory would indicate that the signal is more valuable the lower the overall information environment for that stock. Fuller (2003) finds that firms with less information available to the market have greater price reactions to dividend changes and the signal provided. Since smaller stocks tend to have fewer analysts and less overall information, the extra signaling ability of small dividend-paying stocks over small non-dividend-paying stocks should be even more pronounced in declining markets. The signaling explanation is consistent with our finding of a more pronounced differential between dividend-paying and non-dividend-paying small stocks than large stocks. 4

6 Third, we find the results are also more pronounced for stocks with more trading volume, after controlling for risk. Amihud (2002), Butler, Grullon, and Weston (2004), and Lipson and Mortal (2004) have related liquidity to stock returns, the cost of raising capital, and capital structure. Chordia, Roll, and Subrahmanyam (2001) and Van Ness, Van Ness, and Warr (2004) also find that liquidity varies with the overall market movements; both find a pronounced change in down markets. We examine potential liquidity effects with respect to asymmetric performance for dividend paying stocks and find that dividend-paying stocks continue to outperform non-dividend-paying stocks more in down markets than in up markets even after controlling for volume. Again, prospect theory would not suggest any preference for high or low volume stocks, while signaling theory predicts that for firms with more dispersion in investors beliefs, the greater the value of the signal. For example, Frankel and Froot (1990) find that dispersion Granger-causes volume, and Harris and Raviv (1993) and Shalen (1993) find greater dispersion in investors opinions of the firm s value leads to higher volume. 4 Thus, our results are consistent with the signaling theory prediction that the asymmetric response across dividend-paying and non-dividend-paying stocks should be positively correlated with volume. Collectively, these results provide more support for a signaling theory explanation than one involving prospect theory. The remainder of the paper is organized as follows. Section I expands on how dividend payments relate to market movements and some unique characteristics of dividend payments, as well as providing testable empirical predictions. Section II presents the data and methodology. Section III describes the major empirical results, both overall as well as risk-adjusted using CAPM, Fama-French three-factor model tests, and Fama-McBeth style regressions. Section IV provides some robustness checks by examining alternate definitions for up and down markets, alternate model specifications, and controls for volatility and market listing. Section V examines other issues related to liquidity issues and dividend changes. Section VI examines whether signaling theory or prospect theory is the more likely explanation 4 As Frankel and Froot (1990, p. 182) note The tremendous volume of trading is another piece of evidence that reinforces the idea of heterogenenous expectations, sinceit takes differences among market participants to explain why they trade. 5

7 for the results found in this paper. Section VII concludes. I. Market Movements, Dividends, and Empirical Predictions a. Market movements Although there are several reasons why investors might prefer dividend-paying stocks in down markets, traditional asset pricing models such as the Sharpe (1964) Lintner (1965) CAPM or the Fama- French (1992, 1993) models do not account for state-specific investor preferences. Previous research suggests, however, that investors may have asymmetric preferences; for example, Harvey and Siddique (2000) find that risk-averse investors dislike negative skewness. In addition, an increasing body of work examines asymmetric responses of returns to market movements, and finds different return characteristics in up or down markets. Ang and Chen (2002) find that correlations between stocks returns and the market increase when the market declines. DeBondt and Thaler (1987) find different upside and downside betas for previous winners, although Ang, Chen, and Xing (2004) find that it is downside correlations, and not downside betas, that investors price. Additionally, Ang, Chen, and Xing (2004) find higher expected returns for firms whose correlations with the market increase when the market is declining. Using a Bayesian framework, Hong, Tu, and Zhou (2003) finds that there are significant differences across bull and bear markets, arguing that they should be examined separately. Other studies, such as Goldstein and Nelling (1999), find that the returns on REITs have different risk characteristics, with different correlation structures and betas, depending on whether the market has a positive or negative return. Broadly rising or falling markets tend to indicate albeit imperfectly investors perceptions of the state of the economy in the future. In particular, broadly falling markets indicate either an increase in interest rates or a reduced expectation of future cash flows across most firms. During such times, both prospect theory and signaling theory indicate that investors may prefer cash payments to capital gains. For example, prospect theory directly implies that investors would prefer the certainty of a dividend 6

8 payment relative to the more variable capital gain, especially when markets are moving down. While investors in non-dividend-paying firms could mimic this cash flow by selling stock, they would be doing so at depressed prices (given the market downturn) and incur potentially non-trivial transaction costs (a guaranteed loss). It is possible, therefore, that in these states of the world investors would prefer dividends, while in more positive states of the world the receipt of dividends would be less valuable. Further, as a great deal of stock is now held in tax-deferred accounts, it is no longer clear that dividends are taxdisadvantaged to selling stock. 5 Similarly, signaling theory suggests investors have more demand for signals during declining markets. Bhattacharya (1979), John and Williams (1985), and Miller and Rock (1985) argue that dividends may signal managers private information regarding future earnings. When future projections of the economy are poor or uncertain, investors are likely to look to managers for information regarding their firm s financial health and future prospects. Since the probability and costs of financial distress generally rise in down economies, this information is particularly valuable. Note that in this case it is the existence of a dividend payment that provides such a signal. In fact, it need not be that each individual dividend payment has great signaling value; instead, the maintenance of the commitment to pay regular dividends allows the firm to send a signal that the future prospects for the firm remain positive. A similar signaling argument can be made for the free cash flow hypothesis as developed by Jensen and Meckling (1976) and Jensen (1986), which suggests that managers can credibly signal they will not invest in negative NPV projects through the bonding of maintaining dividend payments. This signal of avoidance of future negative NPV projects should be most valuable when the market is down, since the probability that the bonding will be a binding constraint increases as the state of the economy decreases. Thus, whether the signal is the maintenance or increase in future earnings or the reduction of waste of free cash flows, the ability to signal should be most valuable in periods of poor overall stock market performance. 5 See Clements (2002) for a larger discussion on this point. Recent changes in tax laws have also reduced or 7

9 b. Dividends We concentrate on dividend payments because dividends have a variety of special features that lend themselves nicely to studying investors preferences in declining markets. While Grullon and Michaely (2002) argue for examining the total payout of a firm, dividends have a variety of unique features not shared by repurchases. First, dividend payments are cash payments to all shareholders, while the benefit of share repurchases is a possible capital gain to those retaining their shares. Brennan and Thakor (1990) hypothesize that repurchases only benefit informed investors but dividend payments benefit all investors equally. Second, repurchases are not continuing obligations of the firm and may be suspended at any time, while the payment of a dividend generally signifies a continuing obligation to continue to pay a dividend (for that dollar amount or higher) in the future. Cook, Krigman, and Leach (2004) note that the timing of share repurchases is uncertain, including if and when they are completed, thus limiting their signaling ability. Howe, He and Kao (1992) also indicate that repurchases are discretionary and are not always observable by investors. However, the timing and completion of dividend payments is transparent: dividends are announced and then paid on a certain date. It is immediately clear to shareholders whether or not this obligation has been met in part or in full. Finally, unlike repurchase programs, dividend payments are regularly scheduled. For example, for quarterly dividend payers, investors know in advance when to expect the next dividend payment. Yet, investors in non-dividend-paying firms do not know when they will next receive credible information about their investment. In particular, the value of dividend-paying stocks should be highest in those states of the world where the signal is of the most value, i.e., down markets when there is either increased uncertainty or when expectations of the future have become less rosy. As a result, investors may prefer dividend-paying stocks to non-dividend-paying stocks even in the months when the dividend-paying eliminated the tax benefits of capital gains over dividend payments. 8

10 stock does not pay a dividend, as investors reasonably expect to get a signal within a few months for dividend-paying stocks. 6 Focusing only on dividends is unlikely to bias our findings. Examining a similar time period, Boudoukh, Michaely, Richardson, and Roberts (2004) indicate that the correlation between dividend yield and payout yield is approximately 0.80 and was, for most of the time period, higher than repurchase yields measured similarly. To the extent that repurchases are beneficial in down markets, not considering repurchases will only bias us against finding significant results. (For example, if repurchases are beneficial, they would mitigate our results on non-dividend-paying stocks, since, by definition, nondividend-paying stocks can only increase payout through share repurchases.) If repurchases are highly correlated with dividend payments, our results would still hold. Other papers also support this focus on dividends. Although Fama and French (2001) claim that firms tend to view dividends as less necessary now than in the past, DeAngelo, DeAngelo, and Skinner (2004) find that this trend has occurred due to very small dividend payers omitting dividends; increases by large dividend payers have more than offset these omissions in value. Allen, Bernardo, and Welch (1998) find that firms have continued to pay more of their earnings in dividends than in repurchases, and Brennan and Thakor (1990) note that uninformed investors would prefer dividends to share repurchases due to adverse selection effects. There is also a large body of research that examines dividend and nondividend-paying stocks separately, but that does not consider the state of the market. Papers such as Blume (1980), Litzenberger and Ramaswamy (1980, 1982), and Elton, Gruber, and Rentzler (1983) examine dividend yields in a modified version of Brennan (1970) s after-tax CAPM that explicitly accounts for non-dividend-paying firms. These papers find that although there is a relation between expected returns and dividend yields for those firms that pay dividends, this relation does not hold for 6 For example, if a stock pays dividends in March, June, September, and December, the non-dividend-paying months are January, February, April, May, July, August, October, and November. While Easterbrook (1984) noted that designing an empirical test [of constant-payout policies] is formidable, Section V provides an empirical test of dividend-paying firms in non-dividend-paying months to examine the benefit of such constant dividend-payout policies. 9

11 non-dividend-paying firms. Christie (1990) examines non-dividend-paying firms explicitly and finds that non-dividend-paying firms underperform firms of similar size. None of these papers, however, examine differing effects across different market conditions. 7 c. Empirical implications We examine dividend and non-dividend paying stocks separately to examine their differing responses to advancing and declining markets by testing a variety of hypotheses. We begin by examining the three empirical predictions which should hold under either the prospect theory explanation or the signaling explanation to investigate whether investors differentially prefer dividend-paying stocks in declining markets. First, we test whether dividend-paying stocks outperform non-dividend-paying stocks more in declining markets than in advancing markets. Under either prospect or signaling theory, this result should hold overall, after controlling for risk, for most sub-periods, and across most stocks (large vs. small, NYSE vs. NASDAQ, etc.). Alternatively, according to traditional asset pricing models, we should find no differences. Second, the maintenance of a dividend should cause a favorable response during a declining market but not during an advancing market, and the increase of a dividend should matter more in declining markets than advancing markets. Prospect theory would indicate that investors prefer cash in down markets more than in up markets, while signaling theory proposes that the maintenance of a dividend in a down market is a positive signal and the increase of a dividend in a declining market is a very strong signal. Alternatively, no difference should be seen under symmetric asset pricing models. Third, investors should prefer dividend-paying stocks to non-dividend-paying stocks even during those months between dividend payments during which no dividend is paid. Both prospect theory and the preference for signaling ability indicate that it is not the receipt of a cash payment, but the knowledge that such payments are coming, that matter. Alternative explanations (such 7 Baker and Wurgler (2004) find some evidence that when investors value dividends, managers will initiate dividend payments and when investors value non-dividend-paying stocks, manager omit dividend payments. 10

12 as tax explanations) require cash payments and therefore the results would not hold in non-dividendpaying months for dividend-paying stocks if these explanations were correct. Next, we determine if the signaling explanation is more likely than prospect theory to drive these results. First, while dividend-paying stocks outperform non-dividend-paying stocks in declining markets, this result should not vary with the amount paid. Prospect theory would imply that investors would prefer more cash to less in down markets, while signaling theory would suggest that just the existence of a dividend and not its level is preferred by investors. Therefore, our results should vary significantly with dividend yield if prospect theory is the driving explanation, and should not vary significantly with dividend yield if signaling theory is more correct. Second, although true for all stocks, small dividendpaying stocks should outperform small non-dividend-paying stocks in declining markets more than large dividend-paying stocks outperform large non-dividend-paying stocks in declining markets. Prospect theory does not differentiate across stock types; however, signaling theory implies the value of the ability to signal is more valuable for firms for which there is less information. As a result, smaller stocks should show more pronounced effects under the signaling theory explanation. Finally, the relative difference between up and down markets between dividend-paying and non-dividend-paying stocks may be highest for more liquid, high volume stocks. Although prospect theory does not indicate that there should be any difference across liquidity grouping for the preference of dividend-paying stocks over non-dividendpaying stocks, signaling theory suggests that since more liquid stocks may have more investor dispersion, and therefore, a significant difference between dividend-paying and non-dividend-paying stocks. II. Data and Methodology We identify a sample of dividend-paying and non-dividend-paying firms from the Center for Research in Security Prices (CRSP) monthly master file by examining all NYSE, AMEX, and NASDAQ listed stocks with data in CRSP over the 31-year period from January 1970 to December For each firm, we collect its monthly return, market capitalization and share volume data from CRSP. Since 11

13 studies of returns and dividend yields such as Blume (1980), Litzenberger and Ramaswamy (1980, 1982), Elton, Gruber, and Rentzler (1983), and Christie (1990) note a U-shaped pattern in returns due to the unusual nature of non-dividend-paying stocks, we identify dividends by comparing the CRSP total return to the CRSP return that does not include dividends. If the returns are different, the firm is considered to have paid a dividend in that month, and the difference is considered to be the dividend. Although this method might result in the calculation of negative dividends, these likely errors were retained so as to not impart any bias by correcting errors on only one side. Next, we used the distribution code in CRSP to determine if the dividend was a special, annual, semi-annual, quarterly, or monthly. As we are concerned with the signaling aspect of dividends, we only examine quarterly-dividend-paying stocks. Choosing quarterly-dividend-paying stocks increases the frequency of the possibility of signal observations while decreasing the length of time between potential signals. Thus, only quarterly-dividends were considered when determining whether a firm was a dividend-paying firm. All firms paying dividends other than quarterly dividends were considered nondividend-paying firms. In this way, to the extent that non-quarterly dividends are at all considered positive, we will bias our results against finding any results for quarterly-dividend-paying firms. 8 For each month we classify firms as either dividend paying or non-dividend-paying. If a firm paid a quarterly-dividend in that month, it is classified as a dividend-paying firm. Further, the months between quarterly-dividend payments are also classified as dividend-paying months. If a firm does not pay a dividend and then begins paying a dividend, it is classified as a non-dividend-paying firm until the month after the dividend is paid. That is, if a firm lists on January 1989 but does not pay a quarterlydividend until June 1992, then the firm is considered a non-dividend-firm from January 1989 through June 1992 and is classified as a dividend-paying firm as of July 1992 until the firm stops paying a dividend, the firm is delisted, or the sample period ends. In this way, any positive return in the stock 8 To insure that these rules did not affect our results, we re-ran our tests using alternate definitions of dividendpaying stocks. For example, we also included all regularly scheduled dividend payments (monthly, quarterly, and yearly) when classifying firms as dividend-paying. As another alternative classification method, we dropped all 12

14 price due to the initiation of a dividend will be attributed to the non-dividend-paying stock group, thereby biasing our results against finding outperformance by dividend-paying stocks. If a firm pays a dividend and then stops paying a dividend, it is classified as a dividend-paying firm until the month after the scheduled quarterly-dividend payment. That is, a firm lists on January 1989 and begins paying a quarterly-dividend as of March 1989 but does not pay the June 1992 dividend, then the firm is considered a non-dividend-paying firm for January, February, and March 1989, a dividend-paying firm from April 1989 to June 1992 (the month of the expected quarterly-dividend), and a non-dividend-paying firm from July 1992 until it is either delisted, pays a dividend, or the sample period ends. 9 Thus, any negative surprise due to the cessation of a dividend will be attributed to the dividend-paying group, further biasing our results against finding outperformance by dividend-paying stocks. 10 Because we want to examine if investors value firms that pay dividends, we must look at those periods where dividend payments should be most valuable. We use down markets as a proxy for those times when investors should more highly value dividend payment. Similar to Goldstein and Nelling (1999), we collect the S&P 500 returns for each month from CRSP and classify an up market as a month during which the monthly return on the S&P 500 was positive, while a down market is one where the S&P 500 posted a negative monthly return. Other papers examining asymmetric market responses have defined up/down markets differently: Ang and Chen (2002) define up and down markets by looking at returns in excess of the one-month Treasury bill, while Ang, Chen, and Xing (2004) define up and down markets by whether the excess market return is below its mean in the previous year. Given that the market historically has a positive excess return, these alternate definitions would result in more months non-quarterly-dividend payments from the sample so non-dividend-paying firms never paid any type of cash dividend. Results are qualitatively similar and available upon request. 9 As a robustness check, we also ran the results requiring that for a firm to be called a dividend-paying firm, the firm must pay dividends for the entire time period. All of the results were substantively unchanged. 10 As one area of concern is the response of dividend-paying stocks, we chose a classification method that, if anything, will bias downwards the returns of dividend-paying stocks. In particular, this classification method does not bias upward the results for the dividend-paying stocks due to initiations and omissions. 13

15 being classified as down markets than will our more restrictive definition. As we are primarily interested in different responses in down markets, we use the more conservative definition. 11 Overall, our sample includes 20,315 NYSE, Amex and NASDAQ listed firms for the 372 calendar months from January 1970 to December Each firm was classified as either dividendpaying or non-dividend-paying for every month of the sample period in which data was available, resulting in a total of 2,161,688 firm months in our time period, of which 1,392,422 are non-dividendpaying firm months and 769,266 are dividend-paying firm months. Table 1 describes the dividend- and non-dividend-paying firm months in our sample. Panel A provides averages across all observations for all 372 calendar months in our sample, while Panels B and C provide averages for those observations that occur in the 217 months in our sample where the S&P 500 had a positive return ( up markets ) and the 155 months where it did not ( down markets ). As Panel A indicates, the average market capitalization of firms during the months when they were classified as dividend paying is almost five times that of firms classified as non-dividend-paying. This larger size is due to having twice as many shares outstanding and an average price about 2.5 that of the non-dividend-paying firms. Trading volume was relatively similar for dividend-paying firms and non-dividend-paying firms. Betas for non-dividend-paying firms were slightly larger than those classified as dividend-paying firms. These general results in Panel A are similar for both up and down markets as indicated by Panels B and C, indicating that the relative relationships between dividend-paying and non-dividend-paying do not vary significantly with overall market movements. Insert Table 1 here 11 As a robustness check, we also ran tests defining down markets as negative excess returns. The results on excess returns (not reported here) are substantively similar. In the robustness section, we classified up and down markets based on bull and bear market definitions provided by Ned Davis Research Inc. 14

16 III. Main Results: Overall and Risk-Adjusted To investigate how investor preferences for dividends vary across market conditions, we examine returns of dividend-paying and non-dividend-paying stocks overall and in up and down markets separately. In addition to showing results for all markets, Table 2 presents evidence for the 217 months in our sample where the S&P 500 had a positive return ( up markets ) and the 155 months where it did not ( down markets ). Panel A indicates that we find that dividend-paying firms significantly outperformed non-dividend-paying firms by 0.37% per month across all the months in our sample, with this difference statistically significant at the 1% level for the Student t-test, the Wilcoxon sign-rank test, and the Kruskal- Wallis test. In addition, dividend-paying firms significantly outperformed non-dividend-paying firms in both up and down markets separately at the 1% level. The magnitude of the difference, however, depended on the state of the market. Although dividend-paying firms returned only 0.16% more than non-dividend-paying firms during up markets, they provided 0.90% more than non-dividend-paying firms during down markets. We therefore test to see if dividend-paying stocks statistically outperform non-dividend-paying stocks more in declining markets than in advancing markets as signaling or prospect theory would indicate. Using a difference-ofdifferences test, we find that dividend-paying stocks outperformed non-dividend-paying stocks by 0.74% more in down markets than in up markets, and that this difference is significant at the 1% level, finding support for these theories. 12 To verify that this overall result is not driven by one particular subperiod, Panel B examines three separate decade sub-periods: January 1970 to December 1979, January 1980 to December 1989, and January 1990 to December While the relative outperformance of dividend-paying over nondividend-paying stocks does not hold for two of the three up markets, it does hold for all three of the 12 The difference of differences used throughout the paper is the difference of non-dividend-paying stocks minus dividend-paying stocks in up markets minus the difference of non-dividend-paying stocks minus dividend-paying stocks in down markets. A positive number for this test indicates that dividend-paying stocks outperformed nondividend-paying stocks by more in down markets than in up markets. Throughout the paper, only parametric methods were used to test the significance of the difference of difference test due to the nature of the data (unequal number of observations across all four potential categories). 15

17 down markets. More importantly, the differences-of-differences tests for all three sub-periods indicate that dividend-paying stocks outperform non-dividend-paying stocks by more in down markets than in up markets, and that this relative outperformance is significant at the 1% level. Thus, the difference-ofdifference results are consistent both overall and in each of the three sub-periods. Collectively, the results in Table 2 support the first major empirical prediction that investors differentially prefer dividend-paying stocks over non-dividend-paying stocks more in declining markets than in rising markets. Insert Table 2 here A. CAPM Results Although the state of the market affects the magnitude of the difference, so too does the level of risk of the stock. We examine the abnormal return for each firm i using the capital asset pricing model to determine expected returns; we estimate: Abnormal Return ( r β ( r r )) = ActualReturn F (1) i i F i M where Actual Return i is the return for firm i for that month, r F is the three-month Treasury bill for that month, r M is the return on the CRSP equally-weighted portfolio, and β i is the beta for stock i give by CRSP. We then compare the abnormal returns for dividend-paying and non-dividend-paying firms for all markets, up markets and down markets, as shown in Table 3. Although non-dividend-paying firms outperform dividend-paying firms in up markets by 0.19%, in down markets dividend-paying firms perform significantly better (i.e., significantly less negative returns) than non-dividend-paying firms by 0.60%, more than four times as much. As a result, the difference of differences of 0.79% is highly significant, indicating that the relative abnormal returns for dividend-paying stocks over non-dividendpaying stocks are larger in down markets than in up markets. The CAPM risk adjusted results in Table 3 are consistent with the univariate results found earlier in Table 2: in each case we find that dividend- 16

18 paying stocks outperform non-dividend paying stocks by more in down markets than in up markets. Thus, the answer to our central question do investors prefer dividend-paying firms to non-dividendpaying firms in down markets is yes, even after controlling for risk using beta and the CAPM. Insert Table 3 here B. Size and Market-to-Book Results Fama and French (1992) and others have suggested that book-to-market and size are also important determinants of returns. Christie (1990) also suggests that size is an important factor when examining the returns of dividend-paying and non-dividend-paying stocks. To test if the results are robust to segmentation by book-to-market and size, we replicate the original Fama and French (1992) methodology by dividing our samples of dividend-paying and non-dividend-paying stocks into four market capitalization quartiles and then further sub-dividing those quartiles into four book-to-market quartiles, for a total of 16 sub-groups for each of the dividend-paying and non-dividend-paying stocks. 13 The end result is thirty-two portfolios: sixteen portfolios of dividend-paying stocks based on book-tomarket and size quartiles, and sixteen portfolios for non-dividend-paying stocks. We then calculate the average excess return (return of a firm in month t over the three-month Treasury bill rate in month t) for each portfolio. Insert Table 4 here Table 4 examines the excess return characteristics for the portfolios formed on size and book-tomarket. Overall, size seems not to matter as much as book-to-market in determining the difference between the dividend-paying and non-dividend-paying groups for all markets. As shown in Panel A, non- 17

19 dividend-paying stocks seem to outperform dividend-paying stocks for the lower book-to-market groups across all markets, with the reverse true for the higher book-to-market quartiles. However, as Panel B shows, in up markets the non-dividend-paying stocks outperform the dividend-paying stocks for low book-to-market quartiles, but for the higher book-to-market groups, size becomes a factor and only the smaller stocks in the higher book-to-market quartiles have a significant difference. Panel C, however, presents results that during down markets, dividend-paying stocks do better than non-dividend-paying stocks for all book-to-market categories, but only for the smaller stocks. Thus, the results are strongest for low book-to-market small stocks. 14 Panel D examines the differences of differences and finds reasonably strong support for dividendpaying stocks outperforming non-dividend-paying stocks by more in down markets than in up markets, as predicted by both the signaling and prospect theory hypotheses. However, the results are the weakest for the largest stocks, particularly those with medium to high book-to-market values, and strongest for small stocks and lower book-to-market values. These results therefore provide slightly more support for the signaling hypothesis over prospect theory. Smaller, low book-to-market (i.e., high market-to-book) stocks are likely to have less information provided than larger, high book-to-market firms. As a result, the value of the receipt of a signal may be stronger for these firms, particularly when the general economy (as proxied by the stock market) looks less favorable. Therefore, these results are consistent with a signaling explanation. Prospect theory, on the other hand, would not differentiate across these firms; we would expect to see no systematic difference across these categories. Thus, the results in Table 4 support the overall suggestion that investors prefer dividend-paying stocks to non-dividend-paying stocks in down 13 We thank Ken French for providing the cutoff points for the market capitalization and book-to-market quartiles. 14 To determine if the results in Table 4 were driven by differences in size and book-to-market values for dividendpaying versus non-dividend-paying stocks within each individual size and book-to-market subgrouping, we examined the median size and book-to-market values for each group in the sixteen groupings. We found that only for the smallest firms were there significant differences between the median size of dividend-paying and nondividend-paying firms (dividend paying firms were significantly larger than non-dividend-paying firms). There were no significant differences in median book-to-market ratios for dividend-paying and non-dividend-paying firms across the sixteen groups. 18

20 markets more than in up markets, and also provide some support for the signaling explanation as to why they have this preference. C. Fama-French Three-Factor Model Results Similar to Ang, Chen, and Xing (2004), we also use the Fama and French (1993) (FF) threefactor model to estimate abnormal returns for monthly portfolios. This model controls for nonindependence of returns over time, size, and book-to-market effects. We estimate a modified FF threefactor model as follows: r r = α + b RMRF + s SMB + h HML + d DOWN + ε (2) it Ft it it t it t it t it t it where r it is the monthly return on a portfolio i of dividend-paying or non-dividend-paying firms, r Ft is the monthly return on three-month Treasury bills, RMRF is the excess return on a value-weighted aggregate market proxy, SMB is the difference in the returns of a value weighted portfolio of small stocks and large stocks, HML is the difference in the returns of a value-weighted portfolio of high book-to-market stocks and low book-to-market stocks as in Fama and French (1993), and DOWN is a dummy variable which takes on the value one if the market is down and zero if the market is up. For each month we calculate the excess return on an equally-weighted portfolio composed of either all dividend-paying firms or all nondividend-paying firms. We then regress this portfolio return on the factors in equation (2) and examine the differences in coefficients. We would expect that if investors prefer dividend-paying firms in down markets, that for the dividend-paying portfolio the coefficient on DOWN should be significantly higher than that of the non-dividend-paying portfolio. 1. Econometric issues and weighting methodology To perform these analyses and those that follow, we use equally-weighted portfolios. There are a number of factors in favor of equally weighting of portfolios. First, we are examining the responses of 19

21 dividend and non-dividend portfolios to up and down markets, where up is defined as a positive S&P 500 return. The S&P 500 index is itself a value-weighted portfolio. Thus, the value-weighted dividend and non-dividend portfolios will be very highly correlated with the variable that conditions on the up and down market, namely S&P 500 index. 15 Many of the same stocks will determine the return characteristics of both the portfolios and the index that divides our sample. This effect will be particularly exacerbated for the value-weighted dividend portfolio, given the structure of the S&P 500 index, which will further complicate comparisons across the dividend and non-dividend portfolios. 16 A second issue related to the use of equally-weighted portfolios is related to whether an investor can trade on this information. While an equally-weighted portfolio may incur more transaction costs due to the increased trading from more frequent portfolio rebalancing, the issue in this paper is the differential asymmetric response of dividend and non-dividend-paying stocks in up and down markets. Given that the state of the market is fixed during any one month, investors cannot trade on this information; it is a state of the world for all stocks. Furthermore, even if it were possible to trade on the state of the market, it would be prohibitively expensive for investors to move from an all dividend-paying portfolio to an all non-dividend-paying portfolio based on the state of the market. Therefore, given the focus of the question and the nature of the test, trading considerations are not a primary concern. Third, the results in Table 4 for the sixteen size and book-to-market portfolios indicate that the results vary with size, so making size-weighted portfolios would somewhat obfuscate the results. Ang and Chen (2002) also note that these asymmetric results decrease with size; therefore, value-weighted portfolios would tend to understate the magnitude of the results. In addition, Fama and French (1993, 1996) find that three-factor models have systematic problems explaining returns for small stocks, making this methodology not as useful. 15 We also changed the definition of up/down markets to be based on whether the CRSP equally-weighted index had positive returns or not. The results provided even more support that dividend-paying stock outperform nondividend-paying stocks even when portfolio returns were value-weighted. 16 For example, since the value-weighted dividend-paying portfolio returns are highly correlated with the S&P 500 return, we would expect that dividend-paying portfolios to always do worse than non-dividend-paying firms in down markets. 20

22 Finally, as Fama (1998) notes, the weighting structure of the portfolio should determined by the underlying question. Because the question under investigation in this study is more a question about the particular nature of an individual stock does it or does it not pay a dividend and not particularly about a portfolio, Fama (1998) implies that equally weighting is appropriate. Equally-weighted portfolios allow for an examination of individual characteristics of a stock by treating each stock similarly, while value weighted portfolios can be primarily driven by a limited number of stocks. Results from equallyweighted portfolios therefore better represent the average stock. A number of recent papers, such as Lowry (2003), have therefore presented equally-weighted results. In addition, dividend policy is a management choice variable. As managers do not control a portfolio of firms, but instead a single firm, managers are more concerned with the results for an average stock. As a result, we use equally-weighted portfolios in these analyses. 2. Fama-French overall results We first verify that the FF three-factor model using our sample provides consistent results with prior research. As indicated in Panel A of Table 5, the coefficients on the FF three-factors indicate that the data load properly on the factors, do not have significant alphas, and have very high adjusted R 2 of around 90% for both the non-dividend-paying and dividend-paying portfolios, indicating that the basic FF model works well with this data. In addition, although the factor loadings are significantly different from each other, contrary to the hypotheses in Brennan (1970) or the empirical results in Black and Scholes (1974), Litzenberger and Ramaswamy (1979, 1980, 1982), and Blume (1980), we do not find differences between dividend-paying and non-dividend-paying stocks overall in terms of alpha outperformance. Insert Table 5 here 3. Modified Fama-French results 21