While the irrelevance theorem of Miller and Modigliani (1961) implies that there is no reason to suspect that dividends play a role in determining equity price levels or equity returns, the theorem is silent on the usefulness of dividends in explaining these variables. It is then, perhaps, not surprising that there is a considerable literature exploiting the properties of dividends and dividend yields to better understand the fundamentals of asset pricing both in the time series and in the cross section. Motivation for the former comes from variations of the Gordon growth model in which dividend yields can be written as the return minus the dividend's growth rate (see, e.g., Fama and French (1988)), from consumption-based asset pricing models in which the firm's dividends covary with aggregate consumption (e.g., Lucas (1978) and Shiller (1981)), and so forth. Additional motivation comes from cross-sectional heterogeneity in tax, agency, and asymmetric information considerations (e.g., Litzenberger and Ramaswamy (1979), Jensen (1986), John and Williams (1985), Allen, Bernardo, and Welch (2000), and Grullon, Michaely, and Swaminathan (2002)).

We propose that this underlying motivation really refers to distributed cash flow going to equity holders, be it in the form of dividends or anything that substitutes for dividends, such as repurchases. To the extent that researchers find dividends to be a useful variable for empirically characterizing asset pricing models (e.g., Fama and French (1988), Campbell and Shiller (1988b), Hodrick (1992), Cochrane (1998), Charest (1978), and Benartzi, Michaely, and Thaler (1997)), two potentially important questions are how well do dividends proxy for total payout, and what are the implications of any mismeasurement? This issue is not vacuous as a substantial amount of recent evidence suggests that repurchases have substituted for dividend payments over the last 15 to 20 years (see, e.g., Fama and French (2001), Grullon and Michaely (2002), Dittmar and Dittmar (2002), and Brav et al. (2005)). Thus, there is reason to believe that dividend and repurchase policies are not independent.

Whether these changes in payout policy are relevant to both time-series and cross-sectional tests of asset pricing is an empirical question. Anecdotally, an emerging literature argues that the dividend yield has lost some of its allure as a key empirical variable in asset pricing (e.g., Stambaugh (1999), Valkanov (2003), Lettau and Ludvigson (2002), Cochrane (2001), and Goyal and Welch (2003)). This paper provides a comprehensive analysis of the impact of measuring dividends versus payouts on existing empirical asset pricing model results. We show that the loss of the predictive power of dividends is related to the definition of payouts in asset pricing tests.

Though the definition of total payout is conceptually straightforward, measuring this variable is a challenge. For example, identifying the fraction of repurchases meant to substitute for dividends is difficult, if not impossible, to discern. Since our focus here is on asset pricing implications, we examine several measures of total payout, leaving the debate over which measure may be “more appropriate” to future research. We also consider a measure of total *net* payout yield, which accounts for cash flows from investors to the firm (e.g., seasoned equity offerings). We examine cash inflows since ex ante there is the possibility that cash is raised to maintain dividends, in which case a correction to account for true economic dividends, the net inflows and outflows, needs to be examined (e.g., Allen and Michaely (2003)). As such, our analysis may be viewed more broadly as an examination of payout-based measures in general.

Figure 1 graphs aggregate series for common dividends, repurchases of common stocks, and sales of common stock by nonfinancial corporations in the merged CRSP/Compustat database from 1971 to 2003.^{1} Consistent with the literature, Figure 1 shows that payout yields are systematically underestimated if repurchases are ignored. The figure also shows that equity issuances represent a significant negative cash transfer to shareholders. While dividends comprise the majority of cash flows during the first part of the sample period, its relative share declined through the mid- to late-1980s. For example, the ratio of repurchases to total payouts (dividends plus repurchases) hovers between 5% and 15% through the early 1980s, after which the ratio rises to nearly 50% by the end of the sample.^{2}

We show that using dividends alone to describe payout is not just a bias per se (as illustrated in Figure 1). Rather, it also has potential cross-sectional effects as the rank correlation between firms' dividend yields and firms' payout yields generally decrease over the sample. Moreover, the time-series process for the dividend yield is different from that for the payout (and net payout) yield, carrying important implications for asset pricing in the context of the existing literature. Interestingly, the time-series processes for the dividend yield prior to the emergence of repurchases as a significant form of distributing cash and that of the payout yield after repurchases became dominant look remarkably similar. This supports the paper's thesis that repurchases should be taken into account when relating yields to expected returns.^{3}

The omission of alternatives to dividends as a means of payout introduces a measurement error problem both in the time series and in the cross section. While this measurement error is potentially an important issue from a theoretical perspective, the focus of the paper is on documenting the empirical importance of measuring *total payouts* (dividends plus repurchases) and *total net payouts* (dividends plus repurchases less equity issuances), or more succinctly *payouts* and *net payouts,* on asset pricing tests. In particular, this paper looks at time-series and cross-sectional regressions of asset returns on various measures of payout yields. The basic strategy is to first document the results using dividend yields and then to show how the results change as we incorporate repurchases and, ultimately, issuances. We report several findings.

First, the evidence of stock return predictability in the time series is much stronger using the payout (net payout) yield. For example, for our full sample period (1926 to 2003) the regression of returns on dividend yields at an annual frequency and horizon generates an *R ^{2}* of 5.5% and a coefficient of 0.116 with a

*t*-statistic of 2.240. The total payout yield regressions, depending on the measurement of repurchases, exhibit

*R*s of 8.0% and 9.1%, an increase of 45% and 65%, respectively. The net payout yield regression exhibits an

^{2}*R*of 26%, an almost fivefold increase. Moreover, while the bias-adjusted (Stambaugh (1999)) dividend yield coefficient is insignificant, those of the payout and net payout yields are strongly significant. In a horse race between dividend yield and (net) payout yield we see that any association between dividends and returns disappears, captured entirely by the other payout variable. Finally, using the out-of-sample predictability framework of Goyal and Welch (2003), we show that our payout measures exhibit positive and robust predictability in spite of model uncertainty due to repeated rolling estimation.

^{2}Insight into this improved predictability is found in the dynamic properties of the individual yield series. Structural break tests reveal instability in the dividend yield series around the time of the enactment of SEC rule 10b-18, which provides a legal safe harbor for firms repurchasing their shares in accordance with the rule's provisions. In contrast, no such instability is detected in the payout measures. Furthermore, regression results over the period 1926 to 1985 show that our payout yield coefficient is very similar to those found in the full sample regressions. Thus, in sum this evidence suggests that explanations of the dividend yield's apparent decline as a predictive variable based on arguments such as spurious statistics, learning, etc. may not be the dominant force behind the reduced predictive power of the dividend yield. Rather, the result may simply be an outcome of using the dividend yield instead of the payout yield.

Second, we find that the payout yield measures have a stronger correlation with returns than do dividend yield measures in the cross-section. For example, the average monthly returns on low, medium, and high payout (net payout) yield portfolios are 1.28% (1.24%), 1.40% (1.36%), and 1.56% (1.57%), respectively. In contrast, similar dividend yield portfolios exhibit average monthly returns of 1.15%, 1.28%, and 1.33%, respectively. Thus, the cross-sectional relation between total payout yields and returns is more distinct than the relation between dividend yields and returns. This conclusion is reinforced by Fama-MacBeth (1973) regressions of returns on beta, size, book-to-market, and our yield variables. In these regressions, dividend yields show an insignificant association with returns, whereas our payout measures exhibit highly significant associations with returns. Interestingly, book-to-market is subsumed within payouts when we confine our attention to those firms that actually pay out cash via dividends.

Moreover, while there is a consistent relation between average returns and payout yields in the context of Fama-French (1993) three-factor model regressions, this is not the case for dividend yields. Most important, asset pricing restrictions of the Fama-French three-factor model can be rejected for a cross-section of portfolios sorted by these factors and payout yield. However, when a payout yield factor is added to the mix we cannot reject the restrictions of the model but for one of the three sets of portfolios.

Finally, based on these previous results, we devise a simple self-financed trading strategy that goes long a portfolio of high-yield stocks and short a portfolio of low-yield stocks, and that rebalances these holdings on an annual basis (Figure 3). The strategy based on the net payout yield exhibits an average annual return of 4.44% compared with 3.36% for the payout portfolio, and 2.16% for the strategy based on the dividend yield. These strategies result in portfolios with negative market betas and negative loading on the size factor, suggesting that these returns are not likely to be explained by standard risk measures.

This paper is organized as follows. In Section II, we describe the data, including definitions, sources, and statistical properties. In Section III, we investigate the time-series and cross-sectional implications of the measurement problem from an empirical viewpoint. Section IV concludes.