Resolving the Price Volatility Puzzle: The Role of Earnings

Transcription

1 Resolving the Price Volatility Puzzle: The Role of Earnings Gil Sadka First version: November 23, 2003 This version: January 26, 2005 Abstract In an efficient market, prices should vary only if investors change their expectations about cash flows, discount factors, or both. Previous research showed that the dividend yield varies mostly due to variation in expected returns, and contains little information about cash flows. This literature uses dividend growth variation as its measure of cash-flow information. However, according to Miller and Modigliani (1961) with no taxes, given earnings, dividends are strictly a financing decision and should not affect prices. Consistent with the dividend-policy irrelevance hypothesis, this paper shows that variation in expected profitability growth explains as much as 70% of the variation in the dividend yield. Thus, the dividend yield contains information about cash flows in terms of earnings, not dividends. In addition, this paper finds evidence consistent with a permanent downward shift in the dividend yield in the 1990s. Controlling for this permanent shift, the results indicate that the dividend yield has not lost its ability to predict returns. Keywords: accounting valuation, expected-return variation, profitability, equity premium, variance decomposition I would like to thank Ray Ball, Philip Berger, John Cochrane, Richard Leftwich, Efraim Sadka, Ronnie Sadka, Wendy Rothschild, Douglas Skinner, Kendrew Witt, and the workshop participants at the University of Chicago for comments and suggestions. Any errors are my own. I gratefully acknowledge financial support from the University of Chicago, Graduate School of Business. University of Chicago Graduate School of Business, 5807 S. Woodlawn Ave., Chicago, Illinois 60637; gsadka@chicagogsb.edu.

2 1 Introduction In general, prices are expected discounted cash flows. Thus, there are two factors that may affect prices: the expectations regarding discount factors, and expectations regarding future cash flows. Research on stock price volatility has found that variation in expected returns explains most of the variation in aggregate stock returns (the market portfolio) and the dividend yield (e.g., Campbell and Shiller (1988a, 1988b), Campbell (1991), and Campbell and Ammer (1993)). 1 On the other hand, variation in expected dividends does not explain much of the variation in prices. Consistent with this analysis on stock price volatility, the finance literature has found that returns are predictable and dividends are not (see e.g., Fama and French (1988, 1989), Keim and Stambaugh (1986), Lettau and Ludvigson (2001), Kothari and Shanken (1997), Lamont (1998) and Cochrane (2001)). Exceptions are Ribeiro (2002) and Lettau and Ludvigson (2004), who find that the labor-income-to-dividend and the consumption-to-wealth ratios identify some predictable dividend variation. While these recent studies find evidence of dividend predictability, it remains the case that dividends do not seem to affect the dividend-price ratio (dividend yield). These results are somewhat disturbing. Prices are simply expected discounted cash flows. Thus, one would expect that both cash flow and return variation would generate price variation. Since the dividend yield is stationary and varies, it must predict either returns or cash flows or both. The evidence described above suggests that only expected returns variation affects the aggregate dividend yield. For instance, as Cochrane (2001) points out: "It is nonetheless an uncomfortable fact that almost all variation in price/dividend ratios is due to variation in expected excess returns, How nice it would be if high prices reflected expectations of higher future cash flows." This literature focuses on dividends as cash flow information. However, dividends are not expected to have an effect on prices. According to Miller and Modigliani (1961), given earnings and ignoring taxes, dividend policy is irrelevant and should not affect prices. Dividends are irrelevant because earnings measure the "potential" cash flow that the asset generates and dividends are only an endogenous financing decision made by the firm and its stock holders (when the dividend is 1 When the analysis is applied in the cross-section (e.g. Vuolteenaho (2002), Callen and Segal (2004), Easton (2004) and Cohen, Polk and Vuolteenaho (2003)), the results suggest that variation in expected profitability can explain much of the variation in the firm-level book-to-market, returns and earnings-price ratios. The difference between the aggregate and firm-level results has been attributed in part to the relative strength of the idiosyncratic components of cash flow variation versus the systematic components of expected returns. 2

3 distributed). Earnings, on the other hand, are not an endogenous decision, they are a result of the firms operations and investments and thus represent the ability of firms to distribute dividends. The difference between dividends and earnings approximates the difference between actual cash flow and free cash flow. Investors are not interested in expected short-term dividends. They are interested in the expected ability to pay out cash, i.e., free cash flow. Therefore, the relevant variable to predict, or, in other words, the variable that should be reflected in prices, is profitability growth and not dividend growth. The primary advantage of using accounting income in this setting as opposed to dividends stems from the theory developed by Miller and Modigliani (1961). The theory suggests that given earnings and ignoring taxes, dividend policy is irrelevant. This paper does not claim that dividends do not matter. In the long-run investors are concerned with discounted cash flows (dividends) that the asset produces. However, this paper suggests that on the aggregate level short-run dividend variations do not affect prices. On the other hand, investors will be sensitive to earnings variations because they provide information about the long-run dividend flow. Accounting income must be received in cash or assets in the future, which will eventually be distributed as dividends. In contrast, dividends do not necessarily reflect future cash flows, they are distributed from past and current earnings. The hypothesis of short-term dividend irrelevance is apparent in stock prices, particularly in the 1990s. There are many firms trading with a positive price that do not pay dividends and are not expected to pay dividends in the short-run. Even though these firms may not distribute dividends or are not expected to in the short-term, it does not mean their prices do not vary due to changes in expected cash flows. Thepriceofthestockreflects the long-run dividend stream, which is a function of current and expected earnings - dividends follow earnings. The firm invests, it accrues earnings, the earnings turn into cash flows and when the firm no longer requires the cash to finance its operations, it distributes it as dividends. In sum, in the long-run it does not matter if we choose dividends or earnings because they are the same, but earnings are more appropriate in short-horizon tests of stock price volatility. This role of accounting information has been studied extensively in the literature. For example, Dechow (1994) illustrates that accounting earnings and accruals are better measures of firms performance than cash flows. 2 A common example is accounts receivables. Assume that some of 2 See also Basu (1997), Ball, Kothari and Robin (2000), Dechow, Kothari and Watts (1998), and Callen and Segal 3

4 the firm s sales are made on account. In this case the firm and its assets generate the right to receive cash flows. Because the cash is not yet received, then in order to measure the firm s performance and the cash flows it s entitled to, one must include accounts receivables in the performance measure. Therefore, cash measures alone would not be appropriate as performance measures. Earnings have several advantages over dividends in addition to the fact that dividends are a result of financing decisions. The legal status of earnings makes it the most appropriate measure of future cash flows. Firms distribute dividends from accrued earnings. Legally, earnings represent the firm s verifiable cash flows generated by its investments and assets that belong to its stock holders. While dividends might provide a signal for future profitability (e.g., Watts (1973) and Healy and Palepu (1988) and Nissim and Ziv (2001)), 3 it remains the case that dividends are distributed from accrued earnings. They cannot, legally, exceed the book value of retained earnings. Previous research provides an additional reason for using earnings as a proxy for future cash flows. Previous literature found that prices contain information about expected earnings. For instance, empirical evidence suggests that prices predict earnings better than conventional time series models and that current price changes reflect future expected earnings shocks (see e.g, Beaver, Clarke and Wright (1979), Beaver, Lambert and Morse (1980), Collins, Kothari and Rayburn (1987), Collins and Kothari (1989) and Kothari and Sloan (1992)). This result is also apparent in early research such as Ball and Brown (1968). The accounting literature on prices and earnings implies that the slope on the dividend yield with respect to earnings growth is expected to be negative. That is, higher prices reflect expectations for higher future profitability. Therefore, higher expected earnings push prices up and the dividend yield down. The third reason for using the aggregate earnings and earnings-dividend ratio in tests of price volatility is the evidence concerning their predictability. For instance, Lamont (1998) finds that the dividend-earnings ratio co-integrates with the dividend-price ratio and predicts returns. 4 Notice that earnings growth multiplied by the dividend-earnings ratio growth is equal to dividend growth. The above implies that earnings and the earning-dividend ratio are good candidates to test for (2004). 3 In contrast, DeAngelo, DeAngelo and Skinner (1996) find that dividends are not a reliable signal for profitability. In addition, Watts (1973) finds only weak evidence of the predictive power of dividends with respect to earnings. 4 Vuolteenaho (2000) finds that as much as 40% of the variation in the aggregate book-to-market ratio is due to expected profitability (Return on Equity - ROE). 4

5 the predictability of cash flows, 5 or systematic undiversifiable profitability variation (e.g., Ball and Brown (1967)) that is priced. Based on the discussion above, this paper contributes to the study of price volatility and predictability of earnings and returns by studying the information contained in the aggregate dividend yield with regard to cash flows. Specifically, this paper investigates whether the dividend yield contains information about future cash flows in terms of accounting income. The results show that expected profitability is a major source of dividend yield variation. During the sample period ( ), earnings growth explains as much as 70% of the variation in the aggregate dividend yield. Thus, expected earnings growth is one of the factors that determine the equilibrium dividend yield. This finding is consistent with a large body of research that studies the role of accounting income in the economy and asset prices (e.g., Dechow (1994), Basu (1997), Callen and Segal (2004), Ball, Kothari and Robin (2000), and Penman and Yehuda (2004)). These studies document that earnings and accruals are more strongly associated with stock prices than are dividends and cash flows. In the short-term, the dividend yield is informative about earnings, not dividends. In the longterm, over the life of the firm, earnings and dividends are the same. But, in the short-term, such as the ten to 15-year-ahead horizon commonly used in the literature, earnings are a more appropriate measure of cash flows, because earnings are more timely. In fact, the results suggest that the dividend yield predicts both earnings growth and changes in the dividend-earnings ratio. Due to expected dividend "smoothing," when expected earnings are high the expected dividend-earnings ratio is low, and vice versa.using these results, this paper shows that the dividend yield would only be able to predict dividends in the very long-term. Higher expected earnings are not expected to contemporaneously translate into higher dividends. The implied 40-year-ahead horizon slope coefficient of log dividend growth on the log dividend-price ratio is only Thus, in the shortterm, earnings, rather than dividends, is the more appropriate and more useful measure of cash flows. As discussed above, this paper finds that the dividend yield predicts both expected returns and expected earnings growth. Since the dividend yield predicts both returns and profitability, it is clear that the two are not independent. In fact, the results indicate a negative contemporaneous correlation between returns and earnings growth. However, returns are positively correlated (0.21) 5 See also Ribeiro (2002) and Pastor and Veronesi (2003). 5

6 with the one-year ahead earnings growth (this result is not tabulated in the paper). Since lower dividend yield predicts both low expected returns and high earnings growth, the contemporaneous correlation between earnings growth and returns is expected to be negative. A sharp price increase (high contemporaneous returns) would result in a decline in the dividend-price ratio, and hence, should be positively correlated with long-term earnings growth. Consequently, variations in the dividend yield due to variations in expected returns can be attributable in part to variations in expected earnings. In addition to the lack of dividend predictability, some recent studies (e.g., Lettau and Ludvigson (2004)) find that the dividend yield has declined and lost its ability to predict returns. This paper finds evidence consistent with a permanent downward shift in the 1990s. Controlling for the permanent change, the results suggest that the dividend yield is a good predictor of future excess returns and future earnings growth. In fact, the R 2 of the regression of one-year-ahead horizon returns on the dividend yield is 10%. Moreover, the coefficients are very similar to the ones reported in Cochrane (2001), using a sample period ending at 1996, consistent with a permanent shift in the dividend yield. The remainder of this paper is organized as follows. Section 2 provides a short description of the data and their sources. Section 3 tests whether the dividend yield contains information about cash flows through expected accounting earnings growth. Section 4 includes some additional tests including tests aimed at determining whether the equilibrium dividend yield has declined. Section 5 concludes. 2 Data Thesamplecontainsallfirm-year data in the CRSP monthly and COMPUSTAT annual databases for the period for firms with fiscal-year ends in December. The December fiscal year end requirement avoids temporal misspecifications due to different reporting and different cumulation periods of annual earnings. The returns, dividends, and price data are extracted from the CRSP monthly data set. The earnings item used is the earnings before extraordinary items in COMPU- STAT. The annual financial variables are measured from April of year t until March of year t +1. Table 1 reports summary statistics for the data used in this paper. The table reports the time series averages, medians, and standard deviations of the variables used in the paper. The annual 6

7 returns are the annual value-weighted returns in excess of the risk-free rate. 6 Theriskfreerateis extracted from the Fama and French three factor model data in the WRDS database. The earnings growth measure is defined as growth in the sum of annual earnings for the sample firms. This paper deviates from past literature (e.g., Vuolteenaho (2002) uses return on equity) in its measure of profitability for two reasons. First, accounting conservatism 7 requires timely recognition of expected declines in cash flowsandatthesametimedoesnotallowfirms to recognize unverifiable expected increases in cash flows. This asymmetry in recognition of "economic income" results in a skewed earnings distribution (with relatively large negative values). This skewness is likely to affect profitability measures such as average profitability growth and average return on equity. The summary statistics in Table 1 for earnings growth are consistent with a symmetric distribution. Second, earnings growth is very similar in essence to dividend growth commonly used in the literature (e.g., Cochrane (2001)) and the sum of earnings is a good approximation for the profitability of the market portfolio, which is the variable of interest. 3 The Dividend Yield and the Predictability of Earnings, Dividends and Returns Previous studies suggest that the dividend yield varies mainly due to variations in expected returns. As noted above, the stationarity of the dividend yield implies that it must predict either returns or cash flows or both. The evidence suggests that the dividend-price ratio contains very little information regarding future dividend growth. Table 2 reports the estimation of the regression models and R t t+i = δ 0 + δ 1 D/P t + η t+i (1) D t+i /D t = δ 0 + δ 1 D/P t + η t+i (2) The table provides a short summary of previously recorded results in the literature (e.g., Fama and 6 Tables 4-7 use raw returns. 7 See e.g., Basu (1997), Ball (2001), and Ball, Kothari and Robin (2000). 7

8 French (1988, 1989)). The dividend yield predicts returns. Its predictive power increases over the long term. The adjusted R 2 for the 10-year horizon returns is increasing to 55%. This result is very similar to the results reported in Fama and French (1988). The coefficient on the dividend-price ratio increases with horizon as well. The relation between short and long-term predictability can be interpreted by the following two assumptions: and R t+1 = a D/P t + ε 1,t+1 (3) D/P t+1 = ρ D/P t + ε 2,t+1 (4) Cochrane (2001) shows that these assumptions (where ρ 1) imply both the increase in the coefficient and the increase in R 2 over longer horizons. The coefficients on the dividend yield are all positive, implying that low prices are associated with high expected returns. While the dividend-price ratio predicts returns, it does not predict future dividend growth. The coefficients on the dividend yield are statistically insignificant for all horizons and their sign changes for different horizons. Moreover, the adjusted R 2 for all horizons is negative. 3.1 Predictability of Earnings To summarize up to this point, the dividend yield predicts returns but not dividend growth. This lack of dividend predictability led the finance literature to conclude that expected returns are the main cause for aggregate price movements. Although the dividend-price ratio does not predict dividend growth, it may contain information about expected cash flow through other measures such as accounting earnings. In fact, as discussed above, investors should be more interested in measures of free cash flow or their portfolio s ability to distribute dividends than actual dividends, which represent financing decisions. To test whether the dividend yield predicts earnings growth, the following two regression models were estimated for 1-10 year horizons. 8

9 E t+i /E t = δ 0 + δ 1 D/P t + η t+1 (5) and E t /D t = δ 0 + δ 1 D/P t + η E t+i /D t+1 (6) t+i Table 3 reports the results of OLS estimation of the above two equations. 8 Notice that (E t+i /E t ) [(E t /D t ) / (E t+i /D t+1 )] = D t+i /D t. Thus, the information about dividend growth can be expressed as information about expected earnings growth and information about the expected earningsdividend ratio. 9 This decomposition is not specific for accounting earnings. Dividend growth can be expressed as D t+i /D t =(X t+i /X t ) [(X t /D t ) / (X t+i /D t+1 )] for any X. However, the use of accounting earnings is not arbitrary. Earnings is the most appropriate measure and predictor of cash flows. The results reported in Table 3 appear to confirm the hypothesis that the dividend yield contains information about cash flows in terms of earnings. The dividend yield seems to be predicting longterm earnings growth, especially at the 6-year horizon and longer. This result is consistent with the conservative nature of accounting. Economic gains are not recorded in a timely fashion. Economic growth at period t would result in earnings increases as much as ten years later. The patterns of predictability are similar to those of the returns predictability reported in Table 2. The coefficient on the dividend yield increases in absolute value with the horizon, as does the R 2. The results in Table 3 for the estimation of Equation (6) are consistent with expected dividend "smoothing." While the dividend yield predicts future earnings growth, it also predicts changes in the earnings-dividend ratio. The change in earnings-dividend ratio is also predictable in the longterm and it offsets the effects of expected earnings growth. The results suggest that an increase in expected profitability is associated with an expected decline in the dividend-earnings ratio. In other words, dividends do not vary as strongly as earnings, and an expected earnings increase does not 8 Equation 5 was also estimated using real earnings growth (deflated by GDP deflator). The results do not change qualitatively. 9 It is also possible to include profitability using the Clean Surplus Relation (e.g. Ohlson (1995), Feltham and Ohlson (1995) and Vuolteenaho (2000)). However, as Lo and Lys (1999) points out, accounting rules violate the clean surplus relation and this relation is not necessarily related to accounting. Lo and Lys state that either the book value or earnings can be chosen arbitrarily and still satisfy the Clean Surplus Relation. 9

10 imply an equivalent expectation for a dividend increase. Since (E t+i /E t ) [(E t /D t ) / (E t+i /D t+i )] = D t+i /D t, this result is consistent with the inability of the dividend yield to predict future dividend growth. Hence, the market anticipates dividend smoothing. 3.2 Understanding the Lack of Dividend Predictability The results in Figure 1 summarize and illustrate the relation between earnings growth, expected dividend-earnings ratio, and the expectations for future dividend growth. The figure plots the expected earnings growth and expected dividend-earnings ratio from estimating and ln (E t+i /E t )=δ 0 + δ 1 ln (D/P t )+η t+1 (7) ln (E t+i D t /E t /D t+i )=δ 0 + δ 1 ln (D/P t )+η t+1 (8) The implied expected log dividend growth is the sum of the predicted values from the above regression model. Since E (xy) 6= E (x) E (y), Figure 1 uses logs to make use of the property of expectations, E (x + y) =E (x)+e(y), to calculate the implied log dividend growth rate. Notice, that while there is an increase in expected earnings growth, the expected dividendearnings ratio declines. The resulting implied dividend growth is very weak. There are two different interpretations for this result. First, it is possible that the dividend yield predicts dividend growth in the very long-horizon. Second, it is possible that the dividend "smoothness" is endogenous - implying that dividend growth might be unpredictable. 10 Since some recent studies find some evidence of dividend predictability, the first interpretation, i.e., dividend yield predicts dividends only in the very long horizon, is more likely. The results show that firms are expected to keep reserves when earnings are high and use them when earnings are low, resulting in smooth dividends. To understand the lack of dividend predictability, consider the following equations: 10 Apart from exogenous shocks such as recessions (see Section 4.1). 10 e t+10 = a dp t + t+10 (9) 10 (d e) t+10 = b dp t + ς t+10 (10) 10

11 dp t+1 = ρ dp t + ψ t+1 (11) where i x t+i = x t+i x t,forallx, i, e t =ln(e t ), (d e) t =ln(e t D t+i /E t+i /D t ),anddp t = ln (D/P t ). Since Figure 1 shows that the implied dividend growth appears only over the long term, the analysis begins with long-term predictability of earnings growth (ten years). Equation (11) implies that dp t+10 = ρ 10 dp t + error (12) and for the log dividend growth, 10 d t+10, Equations (9) and (10) imply that 10 d t+10 =(a + b) dp t + error (13) Using Equations (12) and (13), the implied long-term dividend growth is given by for all i. 10 i d t+10 i = h i (a + b)+ρ 10 (a + b) ρ 10 (i 1) (a + b) dp t + error (14) The results for Equations (9) and (10), not reported, are a = 0.84, b =0.76, and ρ = Therefore, a + b = Note, that the implied dividend growth is very low. The following table reports the implied dividend growth, given these results. i Implied Dividend Growth Coefficient The table above represents the lack of dividend predictability. Since earnings growth is only predictable in the long-term about five years ahead horizon and longer, dividends are unpredictable in longer horizons. In fact, the table shows that the implied 40-year-ahead horizon dividend growth coefficient is only The table implies that to find dividend predictability, using the dividend yield, one must use very long-term tests. Such a test is not feasible with the available data, and 11

12 long horizon tests might be unreliable. In sum, earnings are more timely than dividends and they provide a better measure for cash flows than dividends. 3.3 A Variance Decomposition Approach The variance decomposition approach follows the work of Campbell and Shiller (1988a), 11 who decompose the variance of the dividend-price ratio to two major components, expected returns and expected dividend growth. 12 This approach contributes by estimating how variation in expected profitability affects the dividend yield (economic significance). In other words, the method tests how much of the variation in the dividend-price ratio is attributable to information about expected profitability. For brevity, this paper provides only the key steps. Note, R t+1 =(P t+1 + D t+1 ) /P t (15) Equation (15) can be rewritten so that the price-dividend ratio can be written as Taking natural logs yields µ P t = Rt P t+1 Dt+1 (16) D t D t+1 D t p t d t = r t+1 + d t+1 +ln ³1+e p t+1 d t+1 (17) The lowercase letter denotes the natural log and d t+1 = d t+1 d t. Taking a Taylor expansion of the last term yields p t d t = r t+1 + d t+1 + const. + ρ (p t+1 d t+1 ) (18) where ρ =1/ (1 + D/P) Notice in Table 1 Panel B that the average dividend-price ratio is approximately 4%. Iterating forward and assuming that lim j ρ j (p t+j d t+j )=0results in the following expression X d t p t = const. + E ρ j 1 (r t+j d t+j ) (19) j=1 11 Forasimilarbutdifferent approach, see Cochrane (1991). 12 See also Cochrane (2001). 12

13 Equation (18) implies that the variance of the dividend yield can be decomposed into two parts: predictability of returns, and dividend predictability. X X var (d t p t )=cov d t p t, ρ j 1 r t+j cov d t p t, ρ j 1 d t+j (20) j=1 Table 4 reports the results for the estimation of Equation (20). Most of the variation in the j=1 dividend-price ratio is due to variation in expected returns (about 122%). 13 On the other hand, dividend growth variation does not generate variation in the dividend-price ratio. These results are consistent with previous findings (e.g., Campbell and Shiller (1988a) and Cochrane (2001)) and the results in Table 2. When cash flow information is restricted to dividend growth, this result suggests that the dividend yield does not contain much information about cash flows. Equation (19) can be modified slightly to include other sources of information about cash flow. Notice as before that, d t+j = x t+j (x t+j d t+j ) for any x. This paper focuses on one source of cash flow information, accounting earnings (denoted by E and e =ln(e)). Equation (19) can be written as X d t p t = const. + E t ρ j 1 [r t+j ( e t+j (e t+j d t+j ))] (21) j=1 The corresponding variance decomposition can be decomposed into three factors: returns predictability, earnings predictability, and earnings-dividend ratio predictability. The sum of the last two parts is the dividend growth. The variance decomposition is then decomposed further into the same three factors and with additional decomposition for different horizons, one through five, and six through ten periods ahead. 14 The GMM estimator is the covariance 13 Expected returns variation explains 132% of the variation when using excess returns. 14 Thedecompositioninto1-5and6-10yearhorizonsisdue to the results in Table 6. The long run (6 years and longer) earnings growth seems more predictable than the shorter horizon. 13

14 var (d t p t ) = cov d t p t, 5X ρ j 1 r t+j + cov d t p t, j=1 cov d t p t, +cov d t p t, 10X j=6 5X ρ j 1 e t+j cov d t p t, j=1 ρ j 1 r t+j (22) 10X j=6 5X ρ j 1 (e t+j d t+j ) + cov d t p t, j=1 +ρ 10 cov (d t p t,d t+11 p t+11 ) ρ j 1 e t+j 10X j=6 ρ j 1 (e t+j d t+j ) Table 4 reports the results from estimating Equation (22). The results are consistent with the results reported in Table 3. The dividend yield reflects expectations for earnings and earningsdividend ratio. As much as 70% of the dividend yield variation is due to earnings growth variation. In the infinite horizon equation, Equation (19), we can replace dividends with earnings because they are the same. Also, the infinite horizon dividend earnings ratio is equal to 1. Therefore, in long horizon tests it does not matter whether we use dividends or earnings. However, the results in Table 4 indicates that in short horizon tests profitability is a more timely measure of cash flows and changes in expected profitability are priced. On the other hand, short-term dividend variation is not reflected in prices. 3.4 The Determinants of the Dividend Yield The analysis below provides additional inferences on whether dividends or earnings determine the equilibrium dividend yield. To simplify the analysis, denote Rt P 10 j=1 ρj 1 r t+j, Et P 10 j=1 ρj 1 e t+j, EDt P 10 j=1 ρj 1 (e d) t+j,anddt P 10 j=1 ρj 1 d t+j. Table 5 reports the correlations between these variables. The dividend yield is highly correlated with both expected profitability growth (-0.7) and expected returns (0.85). growth does seem to be correlated with the dividend yield. On the other hand, expected dividend The apparent strong correlation between earnings growth and returns is not surprising. First, investors preferences of holding risk varies with business conditions. Second, expected profitability is expected to vary with business conditions. For example, expected returns are high in recessions. On the other hand, expected profitability is low in recessions. Note that this result does not contradict previous findings that unexpected earnings are positively associated with positive unexpected 14

15 returns (e.g., Ball and Brown (1968)). An additional method of testing different determinants of the dividend yield is a simple regression analysis. 15 In particular, Table 6 uses the following regression model d t p t = δ 0 + δ 1 Rt + δ 2 Et + δ 3 Dt + δ 4 ρ 10 (d p) t+11 + ζ t (23) The results in Table 6 are consistent with the hypothesis that the dividend yield is determined by expected earnings growth and not dividend growth. The coefficient on expected profitability growth is negative and statistically significant for all model specifications. The coefficient on dividend growth is not statistically significant in any of the specifications. Moreover, the adjusted R 2 for the regression including earnings alone is 0.5 compared to for dividends. In sum, the results in Table 6 are consistent with the hypothesis that expected earnings growth and expected returns determine the equilibrium aggregate dividend-price ratio. The results in Tables 2, 3, and 5 point to the strong relation between expected returns and expected earnings. Since the same variable (the dividend yield) predicts both earnings and returns (Tables 2 and 3), the two are not independent and, in fact, they are highly correlated (Table 5). Expected returns include a large cash flow component as reflected by earnings (notice, that Rt is highly correlated with Et ). These results suggest that variations in expected returns are due in part to variations in expected earnings growth. Given the results in Tables 2, 3, and 5, the determinants of the dividend yield were decomposed into two orthogonal factors of expected returns and expected cash flows, for both earnings and dividends. Since expected returns include both a cash flow component and an expected returns component, the expected returns determinant was decomposed into returns orthogonal to cash flows as follows. was estimated for dividends and R t = ϕ 0 + ϕ 1 D t + υ D t (24) R t = ϕ 0 + ϕ 1 E t + υ E t (25) 15 Kothari and Shanken (1992) use a similar regression analysis using dividends. 15

16 for earnings. Consider the following linear model of the dividend yield d t p t = b X 0 + b X 1 X t + b X 2 υ X t + ζ X t (26) for X = E and D. The variance of the dividend yield can then be decomposed into Var(d t p t )= b X 2 1 Var(X t )+ b X 2 2 Var υ X t + Var ζ X t (27) Table 7 reports estimation results for Equations (26) and (27). The results support the hypothesis that the equilibrium dividend yield is determined in part by cash flow information as reflected by earnings. However, the dividend yield does not seem to be affected by expectations of future dividend growth. Moreover, the results indicate that expected returns themselves contain information about future cash flows. The adjusted R 2 is similar when using dividends and earnings. That is, the expectations of returns contain information about expectations of cash flows 16 as reflected by earnings. For instance, Table 5 shows a very high correlation between long term returns and long term earnings. 17 Table 7 also shows that expected earnings growth explains as much as 50% of the variation in the dividend yield, compared to only 1% that dividends explain. 4 Additional Considerations This section provides some additional tests and results related to the dividend yield. The evidence explored in this section is consistent with a permanent decline in the dividend yield during the 1990s. Controlling for this shift restores the predictive power of the dividend yield with respect to expected returns. This section also explores some additional related issues such as raw returns versus excess returns, and the information content of dividend growth with respect to earnings and returns. 16 See Menzly et al. (2004). 17 See Easton, Harris and Ohlson (1992). 16

17 4.1 The Dividend Yield in the 1990s Previous studies (e.g., Lettau and Ludvigson (2004)) find that the dividend yield lost its predictive power with respect to returns in the 1990s and is lower than its past mean. This finding suggests that one of the following is true. First, it is possible that the dividend yield is not as informative about expected returns as suggested by prior research. Second, it is possible that the dividend yield will increase to its past mean due to either higher dividends and/or lower returns. Finally, it is possible that the dividend yield has declined to a new lower equilibrium stationary level. The following test supports the latter interpretation of the results. Controlling for a permanent shift in the dividend yield, it seems that the dividend yield is highly informative about expected excess returns (including the 1990s) and it did not lose its predictive power. To control for a permanent shift in the dividend yield, the following regression model was used D/P t = δ 0 + δ 1 DUM90s + τ t (28) where DUM90s is an indicator variable that receives the value of 1 if the year is greater than or equal to 1990 and zero otherwise. The results (not reported) suggest that the dividend yield shifted in the 1990s. δ 1 = and is statistically significant. To test whether the dividend yield has maintained its ability to predict excess returns and earnings growth, Table 8 reports results for the following regression models and R t t+i = δ 0 + δ 1 τ t + η t+1 (29) E t+i /E t = δ 0 + δ 1 τ t + η t+1 (30) The results are consistent with a permanent shift in the dividend yield. The results suggest that the dividend yield did not loose its ability to predict excess returns and earnings growth in the 1990s. The results are also stronger compared to Tables 2 and 3. For example, the adjusted R 2 for the one-year-ahead horizon predicting regression (for expected excess returns) is around 10% Notice that Tables 4-7 are not affected by the permanent shift. Since the tests use 10 year horizon returns and earnings the dividend yields during the 1990s are mostly excluded. 17

18 4.1.1 Why Has the Dividend Yield Declined? The dividend yield seems to have declined during the 1990s. This result is consistent with Fama and French (2001). Fama and French find that fewer firms pay cash dividends. The main reason for the decline in dividends is the change in the firms characteristics. The number of small firms with low earnings and high growth opportunities has increased significantly in the 1990s. Moreover, they find that firms are less likely to pay dividends even when controlling for the change in characteristics. In the 1990s, many firms engaged in R&D activities and stock markets were used more extensively in financing such projects. These projects are high growth projects with low short-term cash flows, resulting in a lower dividends and high prices, i.e., a new lower equilibrium market dividend-price ratio. 4.2 An Impulse Response Function Another method of illustrating the point presented in Figure 1 is using an impulse response function. In particular, using the following set of equations to illustrate the effects of a shock to the dividend yield until it converges back to its equilibrium level. D/P t+1 = ρ D/P t + ε 1,t+1 (31) R t+1 = a D/P t + ε 2,t+1 (32) and E t+i /E t = b D/P t + ε 3,t+1 (33) E t /D t = c D/P t + ε 4,t+1 (34) E t+i /D t+i Unfortunately, Table 3 shows that the predictability in the dividends/earnings ratio begins at the two year horizon. The coefficient on the one-year-ahead horizon is negative. Therefore, it is necessary to extract a more appropriate estimate for c. Since(E t /D t ) / (E t+2 /D t+2 )=(c + ρ c) D/P t + ε 4,t+2,thefigure uses the two-year-horizon regression to extract c. 18

19 The impulse response function is plotted in Figure 2. The pattern in this figure is consistent with Figure 1. A positive shock to the dividend yield results in higher future returns, lower earnings, and higher dividends/earnings ratio. Notice that the effect on earnings growth is higher than the expected effect on the dividends/earnings ratio, implying long-term dividend decline. 4.3 The Information Content of Dividend Growth Previous work, such as Healy and Palepu (1988) and Nissim and Ziv (2001), shows that dividends can provide information about future profitability. To test the implications of the predictive power of dividend increases for future profitability the following model was estimated. E t+i /E t = δ 0 + δ 1 D/P t + δ 2 D t /D t 1 + η t+1 (35) The model was estimated including and excluding the dividend yield. The results (not reported) are not consistent with dividend growth predicting future profitability growth. δ 2 is generally negative and is mostly not statistically significant. In other words, on the aggregate level, dividend growth does not seem to signal profitability growth. This result suggests that the information content of the dividend yield with respect to earnings is generated mostly from the denominator (i.e., prices). 4.4 RawReturnsversusExcessReturns The results in Tables 2 and 3 are estimated using excess returns (in excess of the risk-free rate). These tests were replicated using raw returns. The results do not vary qualitatively. The paper chose excess returns to show that there are time-varying risk premiums. However, the variance decomposition approach requires the use of raw returns. When Table 4 is estimated using excess returns, the sum of the decomposition-components does not add up to 100%. Therefore, Table 4 utilizes the raw returns. For the same reason, Tables 5-7 use raw returns as well. 5 Conclusion The paper investigates the implications of accounting profitability for the information content of the dividend-price ratio. Previous studies suggest that the dividend yield does not contain information about cash flows, i.e., the dividend yield does not predict dividend growth. However, the results 19

20 presented in this paper are consistent with predictability of accounting profitability. These results suggestthatthecashflow information embedded in the dividend-price ratio shows up in terms of profitability growth ("free cash flow") and not dividend growth. Although it is unlikely that dividend policy is irrelevant, as suggested by Miller and Modigliani (1961), it seems that on the aggregate level, investors are more interested in measures of free cash flow than dividends. The dividend growth is much smoother, less predictable, and much less timely than earnings. Thus, especially for short horizons, such as ten to 15 years ahead (the horizon commonly used in the literature), earnings rather than dividends are more appropriate and are likely to provide more insight on the information embedded in prices. As an approximation, over the life of the firm earnings and dividends are the same. However, it is unclear how long it takes for a profitability shock to translate into dividends. Long-term dividend growth is affected by many different profitability shocks and might be difficult to predict. For example, a $100 profitability shock in any year might translate into a $4 dividend shock for 25 years, which would be difficult to detect in the data, particularly given other past and future profitability shocks. In fact, the paper shows that the implied 40-year ahead horizon dividend growth, as reflected in the dividend yield, is relatively weak. This paper also shows that expected returns contain a significant cash flow component. Since the dividend yield predicts both profitability and returns, the two cannot be independent. This means that variation in the dividend yield due to expectations of returns also reflects variation in expected profitability. 20

21 References Ball, Ray, 2001, Infrastructure Requirements for an economically efficient system of public financial reporting and disclosure. Brookings - Wharton Papers on Financial Services. Ball, Ray and Philip Brown, 1967, Some preliminary findings on the association between the earnings of a firm, its industry, and the economy, Journal of Accounting Research 5, Ball, Ray and Philip Brown, 1968, An empirical evaluation of accounting income numbers, Journal of Accounting Research 6, Ball, Ray, S.P. Kothari, and Ashok Robin, 2000, The effect of international institutional factors on properties of accounting earnings, Journal of Accounting and Economics 29, Basu, S., 1997, The conservatism principle and the asymmetric timeliness of earnings, Journal of Accounting and Economics, 24, Beaver, William H., Roger Clarke, William F. Wright, 1979, The information content of security prices, Journal of Accounting and Economics, 17, Beaver, William H, Richard Lambert, and Dale Morse, 1980,The information content of security prices, Journal of Accounting and Economics, 2,3-28. Callen, Jeffrey L., and Dan Segal, 2004, Do accruals drive stock returns? a variance decomposition analysis, Journal of Accounting Research, 42, Campbell, John Y., 1991, A variance decomposition for stock returns, Economic Journal, 101, Campbell, John Y, and John Ammer, 1993, What moves the stock and bond markets? A variance decomposition for long-term asset returns, Journal of Finance, 48, Campbell, John Y., and Robert J. Shiller, 1988a, The dividend-price ratio and expectations of future dividends and discount factors, Review of Financial Studies, 1, Campbell, John Y., and Robert J. Shiller, 1988b, Stock prices, earnings, and expected dividends, The Journal of Finance, 43, Cochrane, John H., 1991, Explaining the variance of price-dividend ratios, Review of Financial Studies, 5, Cochrane, John H., 2001, Asset pricing, Princeton Press. Cohen, Randolph B., Christopher Polk, and Tuomo Vuolteenaho, 2003, The value spread, Journal of Finance, 58, Collins, Daniel W., S.P. Kothari, 1989, An analysis of intertemporal and cross-sectional determinants of earnings response coefficients, Journal of Accounting and Economics, 11, Collins, Daniel W., S.P. Kothari, and Judy D. Rayburn, 1987, Firm size and the information content of prices with respect to earnings, Journal of Accounting and Economics, 9, DeAngelo, Harry, Linda DeAngelo and Douglas J. Skinner, 1992, Dividends and losses, Journal of Finance, 47 (5), DeAngelo, Harry, Linda DeAngelo and Douglas J. Skinner, 1996, Reversal of fortune dividend signaling and the disappearance of sustained earnings growth, Journal of Financial Economics, 40, Dechow, Patricia, 1994, Accounting earnings and cash flowsasmeasuresoffirm performance, Journal of Accounting and Economics, 18, Dechow, Patricia, S.P. Kothari and Ross L. Watts, 1998, The relation between earnings and cash flows, Journal of Accounting and Economics, 25, Easton, Peter D, 2004, PE ratios, PEG ratios, and estimating the implied expected rate of return on equity capital, TheAccountingReview,79 (1),

22 Easton, Peter D, Trevor S. Harris, and James A. Ohlson, 1992, Aggregate accounting earnings can explain most of security returns: The case of long returns intervals, Journal of Accounting and Economics,15, Ertimur, Yonca, 2003, Financial information environment of loss firms, Working Paper, New York University. Fama, Eugene F., and Kenneth R. French, 1988, Dividend yields and expected stock returns, Journal of Financial Economics, 22, Fama, Eugene F., and Kenneth R. French, 1989, Business conditions and expected returns on stocks and bonds, Journal of Financial Economics, 25, Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, Fama, Eugene F., and Kenneth R. French, 2001, Disappearing dividends: changing firm characteristics or lower propensity to pay?, Journal of Financial Economics 60, Fama, Eugene F., and James MacBeth, 1973, Risk, return and equilibrium: empirical tests, Journal of Political Economy 81, Feltham, Gerald, A. and James A. Ohlson, 1995, Valuation and clean surplus accounting for operating and financial activities, Contemporary Accounting Research, 11, Francis, Jennifer, J. Douglas Hanna, and Linda Vincent, 1996, Causes and effects of discretionary asset write-offs, Journal of Accounting Research 34. Healy, Paul M. and Krishna, G. Palepu, 1988, Earnings information conveyed by dividend initiation and omissions, Journal of Financial Economics, 21, Korajczyk, Robert A., and Amnon Levy, 2003, Capital structure choice: macroeconomic conditions and financial constraints, Journal of Financial Economics, 68, Kothari, S.P. and Jay Shanken, 1992, Stock return variation and expected dividends, Journal of Financial Economics, 31, Kothari, S.P. and Jay Shanken, 1997, Book-to-market, dividend yield, and expected market return: a time-series analysis, Journal of Financial Economics, 44, Kothari, S.P. and Richard G. Sloan, 1992, Information in prices about future earnings: implications for earnings response coefficients, Journal of Accounting and Economics, 15, Lamont, Owen, 1998, Earnings and expected returns, Journal of Finance 53, Lettau, Martin, and Sydney C. Ludvigson, 2001, Resurrecting the (C) CAPM: a cross-sectional test when risk premia are time-varying, Journal of Political Economy, 109, Lettau, Martin, and Sydney C. Ludvigson, 2004, Expected returns and expected dividend growth, Journal of Financial Economics, forthcoming. Lo, Kin and Thomas Lys, 1999, The Ohlson model: contribution to valuation theory, limitations, and empirical applications, Journal of Accounting, Auditing & Finance, Menzly, Lior, Tano Santos and Pietro Veronesi, 2004, Understanding predictability, Journal of Political Economy, 112,1-47. Miller, M.H. and F. Modigliani, 1961, Dividend policy, growth and the valuation of shares, Journal of Business, 34, Nissim, Doron and Amir Ziv, 2001, Dividend changes and future profitability, The Journal of Finance, 56, Ohlson, James A., 1995, Earnings, book values, and dividends in security valuation, Contemporary Accounting Research, 11, Pástor, Lubos and Pietro Veronesi, 2003, Stock valuation and learnings about profitability, Journal of Finance, 58,

23 Pástor, Lubos and Pietro Veronesi, 2004, Rational IPO wave, Forthcoming - Journal of Finance. Penman, Stephen H., and Nir Yehuda, 2004, The pricing of earnings and cash flows and an affirmation of accrual accounting, Working Paper - Columbia University. Ribeiro, Ruy M., 2002, Predictable dividends and returns, Working Paper - University of Chicago. Sadka, Gil and Ronnie Sadka, 2003, The post-earnings-announcement-drift and liquidity risk, Working Paper - University of Chicago. Vuolteenaho, Tuomo, 2000, Understanding the aggregate book-to-market ratio and its implications to current equity-premium expectations, working paper - Harvard University. Vuolteenaho, Tuomo, 2002, What drives firm-level stock returns? Journal of Finance, 57, Watts, Ross L., 1973, The information content of dividends, The Journal of Business, 46,