Electronic copy available at: http://ssrn.

com/abstract=1271607
Expected Returns and Dividend Growth Rates Implied in
Derivative Markets
Benjamin Golez

University of Notre Dame

September, 2012
y
Abstract
The dividend-price ratio is a noisy proxy for expected returns in the presence of time-varying
expected dividend growth rates. This paper uses a new and forward-looking measure of dividend
growth rates extracted from S&P 500 futures and options to correct the dividend-price ratio for
changes in expected dividend growth rates. For the period from January 1994 through June 2011,
I nd that the dividend growth rates implied in derivative markets reliably forecast future dividend
growth rates, and the corrected dividend-price ratio predicts S&P 500 returns substantially better
than the uncorrected dividend-price ratio. The forecasting relation between the corrected dividend-
price ratio and returns is statistically signicant and remarkably stable. Decomposing variance of
the dividend-price ratio, I show that a substantial portion of price movements is attributable to
the time-varying value of expected dividend growth rates. Expected dividend growth rates are
also highly correlated with expected returns. The empirical results depend importantly on the
simultaneous use of futures and options in estimating implied dividend growth rates.
Keywords: present value models, dividend-price ratio, return predictability, derivatives
JEL classication: G12, G13

Department of Finance, 256 Mendoza College of Business, Notre Dame, Indiana 46556-5646. Tel. (574) 631-1458.
Email: bgolez@nd.edu.
y
I would like to thank Mascia Bedendo, Geert Bekaert, Francesco Corielli, Martijn Cremers, Zhi Da, Jens Carsten
Jackwerth, Ralph Koijen, Gueorgui I. Kolev, Jos M. Marn, Francisco Penaranda, Christopher Polk, Jesper Rangvid,
Fabio Trojani, and Robert Zymek for invaluable comments. I also thank seminar participants at Universitat Pompeu
Fabra, Swiss Banking Institute, Stockholm School of Economics, Bocconi University, Carlos III University of Madrid,
CNMV, Erasmus Research Institute of Management, FED Board, University of Notre Dame, Oslo BI, New Economic
School, Einaudi Institute for Economics and Finance as well as participants at Campus for Finance, 12th Conference of
the Swiss Society for Financial Market Research, European Finance Association Doctoral Tutorial, Chicago Quantitative
Alliance Conference, and Eastern Finance Association. Earliest versions of the paper were circulated under the title
Options Implied Dividend Yield and Market Returns.
1
Electronic copy available at: http://ssrn.com/abstract=1271607
1 Introduction
The predictability of market returns is of great interest to market practitioners and has important
implications for asset pricing. At rst, any sign of market return predictability was understood as
evidence against market eciency. As evidence has accumulated that price multiples, such as the
dividend-price ratio, predict market returns (Fama and French, 1988), time-varying expected returns
have become an integral part of equilibrium models with ecient markets (Campbell and Cochrane,
1999; Bansal and Yaron, 2004). Yet return predictability remains debatable. The forecasting relation
between returns and the dividend-price ratio is subject to statistical biases (Stambaugh, 1999). It is
unstable and it also has poor out-of-sample forecasting power (Goyal and Welch, 2008). Koijen and
Van Nieuwerburgh (2011) provide a comprehensive survey of return predictability evidence.
This paper argues that the inconsistent performance of the dividend-price (DP) ratio in predicting
returns is attributable largely to the time-varying nature of expected dividend growth rates. I introduce
a novel measure of expected dividend growth rates that is extracted from index options and futures,
and derive a simple present value model to guide the empirical analysis. Using dividend growth rates
implied in derivative markets to correct the DP ratio for variation in expected dividend growth rates,
I nd that market returns are strongly predictable. Indeed, the corrected DP ratio predicts market
returns both in-sample and out-of-sample. It is robust as well to statistical biases that have been
shown to hinder the return predictive ability of the uncorrected DP ratio.
That time-varying expected dividend growth rates can reduce the ability of the DP ratio to predict
returns has long been recognized in the predictability literature (Campbell and Shiller, 1988; Fama
and French, 1988). According to the textbook treatment, the DP ratio may vary over time not only
because of changes in expected returns but also because of changes in expected dividend growth rates.
Therefore, Fama and French (1988) note that the DP ratio is a noisy proxy for expected returns
in the presence of time-varying expected dividend growth rates (see also Cochrane, 2008a; Rytchkov,
2008; Binsbergen and Koijen, 2010; Piatti and Trojani, 2012). Moreover, since DP ratio increases with
expected returns and decreases with expected dividend growth rates, problems caused by time-varying
expected dividend growth rates are pronounced when expected returns and expected dividend growth
rates are positively correlated (Lettau and Ludvigson, 2005). This positive correlation, which arises in
2
the general equilibrium model as a natural consequence of dividend growth rate predictability (Menzly
et al., 2004), osets changes in expected returns and in expected dividend growth rates, which further
hampers the DP ratios ability to predict returns.
Thus, if the task is to predict returns, and expectations as to dividend growth rates vary over
time, the DP ratio is insucient: We must also account for the changes in expected dividend growth
rates. Yet expected dividend growth rates are dicult to estimate because they aggregate investors
expectations about future growth opportunities.
Recent studies on return predictability typically assume that the future will be similar to the past,
and extract expected dividend growth rates from historical data. For example, Binsbergen and Koijen
(2010) use a latent variable approach within the present value model to lter out both expected returns
and expected dividend growth rates from the history of dividends and prices (see also Rytchkov, 2008).
Lacerda and Santa-Clara (2010) use a simple average of historical dividend growth rates as a proxy
for expected dividend growth rates. These authors conclude that improved prediction of dividend
growth rates enhances the predictability of longer-term (annual) returns. Nevertheless, their methods
use only the information that can be derived from past dividends and prices. Investors, however, base
expectations about future cash ows on a much richer and forward-looking information set.
This paper takes a dierent approach to estimating expected dividend growth rates. Rather than
rely on historical data, I extract a proxy for investors expected dividend growth rates from index
options and futures. Derivative markets provide a unique laboratory for estimating markets expec-
tations about future cash ows. As prices of derivatives depend on future dividend payments and are
related through no-arbitrage, one can invert pricing relations to extract a proxy for expected dividends
from observable market prices of options and futures. Because index derivatives are highly liquid, new
information about future cash ows should be rapidly incorporated into estimated implied dividends.
To provide an analytical framework for the empirical analysis, I rst derive a simple present value
model. Like Binsbergen and Koijen (2010), I combine the Campbell and Shiller (1988) present value
identity with a simple rst-order autoregressive process for expected returns and expected dividend
growth rates. In this model, future returns are a linear function of the DP ratio and the expected
dividend growth rates. We can therefore predict returns through a multivariate regression of returns
3
on the DP ratio and an estimate of expected dividend growth rates, or we can combine both values in
a single predictor, which I call the corrected DP ratio. The corrected DP ratio can be interpreted as
the dividendprice ratio adjusted for variation in expected dividend growth rates.
Following the present value model, I estimate the proposed proxy for expected dividend growth
rates. To extract dividend growth rates implied in index options and futures, I rst estimate an
implied dividend yield. Because any mispricing in options is easiest to arbitrage away by trading
futures, I estimate implied dividend yield simultaneously from options and futures. By combining
the no-arbitrage cost-of-carry formula for index futures and the putcall parity condition for index
options, I derive an expression that enables estimation of the implied dividend yield in terms of the
observable prices of derivatives and the underlying asset. Then I combine the implied dividend yield
with the realized DP ratio to calculate implied dividend growth rates and the corrected DP ratio.
I apply the empirical analysis to the S&P 500 index. As we require data on both options and
futures, the main analysis runs from January 1994 through June 2011.
1
Like other studies, I nd
the standard DP ratio is a rather poor predictor of both future returns and dividend growth rates at
horizons ranging from one month to one year. Implied dividend growth rates, however, are reliable
predictors of future dividend growth rates for all the horizons. Accordingly, the ability to predict
market returns improves considerably when we include implied dividend growth rates as an additional
regressor in the standard DP ratio predictive regression. The results also conrm that the DP ratio
and the implied dividend growth rates can be replaced by a single predictor: the corrected DP ratio.
The predictive coecient on the corrected DP ratio is always statistically signicant. In the predictive
regressions with monthly returns, the corrected DP ratio has an in-sample adjusted 1
2
of 4.21% and
an out-of-sample 1
2
OS
of 5.33%, compared to 0.44% and 0.04% (respectively) for the uncorrected
DP ratio. Because the corrected DP ratio is less persistent than the uncorrected DP ratio, and as
innovations to the corrected DP are related only weakly to returns, the corrected DP ratio has the
additional advantage that it is robust to the small sample bias that has been shown to hinder the
predictive ability of the uncorrected DP ratio.
Contrary to the standard view that almost all the variation in the DP ratio corresponds to changes
1
Note that this period is not aected by structural breaks in the mean of the DP ratio, which have been shown to
aect the forecasting relation of returns and the DP ratio over longer periods (Lettau and Van Nieuwerburgh, 2008;
Favero et al., 2011).
4
in discount rates (Binsbergen and Koijen, 2010; Cochrane, 2011), a variance decomposition reveals
that a substantial portion of variation in the DP ratio is attributable to the time-varying value of
expected dividend growth rates. Expected dividend growth rates account for 34% of variation in
the DP ratio, and the discount rate eect accounts for 134% of its variation. I also nd that expected
returns and expected dividend growth rates are highly correlated at 0.74. This positive correlation
dampens the volatility of the DP ratio, making it less able to predict either returns or dividend growth
rates (see also Menzly et al. 2004; Lettau and Ludvigson, 2005). Once the DP ratio is corrected for
changes in the expected dividend growth rates, it becomes more volatile and better able to predict
returns.
The results are robust to a number of variations, such as exclusion of the recent nancial crisis or
inclusion of other return predictors, and are not driven by seasonality of dividend payments, dividend
risk premium or synchronicity issues in the end-of-day derivatives data. Further, the same results
cannot be obtained using historical dividend growth rates or information on announced dividends.
The results depend importantly, however, on the simultaneous use of options and futures in estimating
the implied dividend growth rates, which conrms that the arbitrage relations for S&P 500 derivatives
should best be analyzed simultaneously.
By emphasizing the role of dividend growth rates for price movements, the results have important
implications for the theoretical models (Campbell and Cochrane 1999; Bansal and Yaron, 2004). That
both time-varying expected returns as well as time-varying expected dividend growth rates should be
featured in the asset pricing models is also advocated by Koijen and Van Nieuwerburgh (2011) and
Chen et al. (2012).
The paper also draws on studies using implied dividends. Dividends implied in derivative markets
have been used as an input in the calculation of risk-neutral densities (Ait-Sahalia and Lo, 1998),
and to study empirical properties of dividend strips (Binsbergen et al. 2012a). I believe, however,
this paper is the rst to use implied dividends for predicting market returns. I also introduce a
new technique that enables extraction of implied dividends simultaneously from options and futures
without resort to the use of proxies for the implied interest rate.
Binsbergen et al. (2012a) document a substantial dividend risk premium in longer-maturity implied
5
dividends. Similar conclusion is provided by Binsbergen et al. (2012b), who analyze a growing
market for dividend futures.
2
The dividend risk premium, however, is not a concern in our case
because I focus on derivatives with shorter maturities (six months), where a substantial portion of
dividends is preannounced and hence, riskless. Accordingly, I nd that the risk premium documented
by Binsbergen et al. (2012a) is absent in the 6-month implied dividends. The estimated implied
dividend ratios also exhibit very low correlations with other proxies for risk.
This work relates as well to the literature that uses other options-implied predictors to forecast
market returns. Bollerslev et al. (2009) use the variance risk premium and Bakshi et al. (2011)
employ forward variances. I focus instead on a cash ow-based predictor and show that information
about dividends implied in derivative markets conveys information about future returns beyond the
information in the variance risk premium. The proposed predictor is also unique in the sense that it
simultaneously exploits information embedded in two closely related derivatives markets, the options
market and the futures market.
Finally, several papers analyze equity options as opposed to index options and document that
measures of violations of no-arbitrage relations and implied skewness embody useful information for
predicting returns on individual stocks (Cremers and Weinbaum, 2010; Conrad et al. 2012; Xing et
al., 2010). In line with Pan and Poteshman (2006), however, I do not nd evidence that these results
carry over to the aggregate market. Also, while ndings on equity options suggest that informed
trading takes place in the derivative markets, results of this paper do not support that implication
and should be interpreted in the light of evidence on market integration. That is, information about
dividends implied in derivative markets can be useful for predicting returns on the underlying asset
only if markets are suciently integrated and investors use the same information to set prices in both
markets.
2
Dividend futures allow to trade dividends directly, are available across dierent regions, and have maturities up to 10
years. Thus, they are very suitable to study implied dividends. The disadvantage is the short sample. Dividend futures
were introduced around 2000 and are exchange traded since 2008.
6
2 Theoretical considerations
To provide an analytical framework for the empirical analysis, this section derives a simple log-linear
present value model and describes theoretical aspects of estimating dividend growth rates implied in
derivative markets.
2.1 Present value model
The model combines the Campbell and Shiller (1988) present value identity with time-series processes
for expected returns and expected dividend growth rates. Rytchkov (2008) and Binsbergen and Koijen
(2010) use a similar approach. The main innovation of this study lies in the empirical estimation of
this framework.
Dene log return r
t+1
, log dividend growth rate d
t+1
, and log dividend-price ratio dj
t
as
r
t+1
= log
_
1
t+1
+ 1
t+1
1
t
_
, d
t+1
= log
_
1
t+1
1
t
_
, dj
t
= log
_
1
t
1
t
_
. (1)
Rewrite return as
r
t+1
' i + dj
t
+ d
t+1
jdj
t+1
, (2)
where i = log
_
1 + exp(dj)

+ jdj and j =
exp(oj)
1+exp(oj)
are constants related to the long-run average
of the dividend-price ratio, dj. Iterate (2) forward to obtain the Campbell and Shiller (1988) present
value identity
dj
t
'
i
1 j
+ 1
t
1

)=0
j
)
(r
t+1+)
) 1
t
1

)=0
j
)
(d
t+1+)
). (3)
Let j
t
= 1
t
(r
t+1
) be the conditional expected return, and let q
t
= 1
t
(d
t+1
) be the conditional
expected dividend growth rate. Suppose that j
t
and q
t
follow some time-series processes. For simplicity
assume that both j
t
and q
t
follow 1(1) processes:
3
3
The AR(1) structure for returns and dividend growth rates is considered in several recent contributions (Rytchkov,
2008; Binsbergen and Koijen, 2010; Piatti and Trojani, 2012) and is motivated by growing evidence that both expected
7
j
t+1
= c
0
+ c
1
(j
t
) + -
j
t+1
, (4)
q
t+1
=
0
+
1
(q
t
) + -
j
t+1
, (5)
d
t+1
= q
t
+ -
o
t+1
, (6)
where -
j
t+1
, -
j
t+1
, and -
o
t+1
are zero mean errors. Combine the present value identity in (3) with the
1(1) assumptions to nd the dividend-price ratio:
dj
t
' , +
_
1
1 jc
1
_
j
t

_
1
1 j
1
_
q
t
, (7)
where , is a constant related to i, j, c
0
, c
1
,
0
, and
1
(details are provided in the Appendix).
Equation (7) states that the log dividend-price ratio is related to expected returns and is therefore
a good candidate for predicting future returns. However, the dj
t
ratio also incorporates information
about expected dividend growth rates. Thus, if those rates vary over time, the dj
t
ratio is only a
noisy proxy for expected returns and an imperfect predictor of future returns (Fama and French,
1988; Rytchkov, 2008; Binsbergen and Koijen, 2010; Lacerda and Santa-Clara, 2010). Since the dj
t
ratio increases with expected returns and declines with expected dividend growth rates, the problem
is pronounced when expected returns and expected dividend growth are positively correlated (Menzly
et al., 2004; Lettau and Ludvigson, 2005). This positive correlation osets the changes in expected
returns and in expected dividend growth rates, further reducing its ability to predict returns.
This implies that the dj
t
ratio is insucient to capture variation in expected returns. We must also
account for the time-varying value of the expected dividend growth rates. To present this formally, I
combine (2), (6), and (7) to derive a return forecasting equation
r
t+1
' c + (1 jc
1
)dj
t
+
_
1 jc
1
1 j
1
_
q
t
+
v
t+1
, (8)
returns and expected dividend growth rates are time-varying and persistent (Menzly et al., 2004; Bansal and Yaron, 2004;
Lettau and Ludvigson, 2005). Assuming dierent time-series processes would preserve the main intuition (see Cochrane
2008b for a discussion).
8
where
v
t+1
= -
o
t+1
j
_
.

I+1
1
1jc
1

.

I+1
1j
1
_
, and c is a constant related to i, j, c
0
, c
1
,
0
, and
1
. According
to equation (8), to predict returns, we need both the dj
t
ratio and an estimate for the expected
dividend growth rates. Because dj
t
ratio and expected dividend growth rates are linearly related to
future returns, we can also replace them by a single predictor:
r
t+1
' c + (1 jc
1
)dj
Ccvv
t
+
v
t+1
, (9)
where dj
Ccvv
t
= dj
t
+ q
t
_
1
1j
1
_
is the corrected dividend-price ratio, which can be interpreted as
the dj
t
ratio adjusted for variation in the expected dividend growth rates. The adjustment factor is
a constant that depends on the linearization parameter (j) and the persistence of expected dividend
growth rates (
1
). Because the linearization parameter is typically close to one and fairly stable over
time, the adjustment constant depends mainly on the persistence of the expected dividend growth
rates. The more persistent the expected dividend growth rates, the more important the role of the
expected dividend growth rates in the corrected dividend-price ratio.
4
2.2 Estimating implied dividend growth rates
The present value model implies that the dividend-price ratio is not enough to capture variation in
expected returns. We also need an estimate for expected dividend growth rates.
I propose to extract an estimate for expected dividend growth rates from index derivatives. Prices
of derivatives, such as options and futures, depend on, among other factors, the dividends that the
index pays until expiration of the contracts. Therefore, we can invert pricing relations and estimate a
proxy for expected dividend growth rates from market prices of derivatives. All we need is absence of
arbitrage opportunities.
Following Ait-Sahalia and Lo (1998), I assume that the index pays a continuously compounded
dividend yield and describe how it can be extracted from derivatives using no-arbitrage relations. I
start with a simple case when future dividend yield is known, and then discuss the case of a stochastic
4
Lacerda and Santa-Clara (2010) derive a similar correction for their adjusted dividend-price ratio, dj
/u
I
= djI +
j
I

1
1p
t

. In their version, the adjusted dividend-price ratio (dj

/u
I
) does not depend on the persistence of expected
dividend growth rates because they assume that expected dividend growth rates are equal to average historical dividend
growth rates (j
I
).
9
dividend yield.
Under the assumption of a known dividend yield, the no-arbitrage futures price is given by the
cost-of-carry formula
1
t
(t) = o
t
exp[(r
t
(t) `
t
(t))t] , (10)
where 1
t
is the futures price, o
t
is the price of the underlying, r
t
(t) is the annualized continuously
compounded interest rate from t to t + t, and `
t
(t) is the annualized continuously compounded
dividend yield between t and t +t. Similarly, by no-arbitrage, the dierence between a European call
and a European put written on the index is given by the put-call parity
C
t
(1, t) 1
t
(1, t) = o
t
exp[`
t
(t)t] 1 exp[r
t
(t)t] , (11)
where C
t
(1, t) and 1
t
(1, t) are the prices of a European call and a European put option with the
same maturity t and the same strike price 1.
Both no-arbitrage conditions relate prices of derivatives to the future dividend yield and the risk-
free rate. With a proxy for the implied risk-free rate, we can therefore estimate implied dividend
yield from either options or futures. Nevertheless, implied interest rate may be dierent from the
observable proxies for the interest rate (Naranjo, 2009), and choice of the proxy may adversely aect
results. Further, given that the basket of stocks in the index is dicult to trade, any mispricing of
options is easiest to arbitrage away by trading futures (Kamara and Miller, 1995). This suggests that
the no-arbitrage conditions in equations (10) and (11) should best be analyzed simultaneously. Thus,
I propose to combine both no-arbitrage relations, thereby also isolate the eect of the interest rate,
and estimate implied dividend yield simultaneously from index options and index futures.
5
I rst solve for the implied interest rate by combining the no-arbitrage relations
r
t
(t) =
1
t
log
_
1
t
(t) 1
C
t
(1, t) 1
t
(1, t)
_
. (12)
5
Section 8.2 considers alternative ways for estimating implied dividend yield and conrms that results are the strongest
when implied dividend yield is estimated simultaneously from futures and options.
10
Then I plug (12) back in to (11) to obtain a formula for the implied dividend yield
`
t
(t) =
1
t
log
__
C
t
(1, t) 1
t
(1, t)
o
t
_
+
1
o
t
_
C
t
(1, t) 1
t
(1, t)
1
t
(t) 1
__
. (13)
Following equation (13), all we need to estimate the implied dividend yield is a European call option
and a European put option with the same strike and the same maturity, the futures price with the
same expiration date as the options, and the price of the underlying asset.
Equation (13), however, is derived under the assumption of a known dividend yield. When the
future dividend yield is stochastic, as assumed in the present value model, the dividend yield in (13)
not only reects expectations about the future dividend yield, but may also include a dividend (yield)
risk premium.
6
Recent studies show that the dividend risk premium plays an important role for
horizons longer than one year (Binsbergen et al. 2012a; 2012b). I focus, however, on derivatives with
six months to maturity, where a substantial portion of dividends is preannounced and hence riskless.
I later show that even at an annual horizon, dividend growth rates can be predicted with considerable
precision, given information on announced dividends. This largely mitigates the eect of the dividend
risk premium. Indeed, I nd that the risk premium documented by Binsbergen et al. (2012a) is absent
in the 6-month implied dividends. Therefore, in the main analysis, I use implied dividend yield as
an uncontaminated measure of expected dividend yield. Sensitivity analysis conrms that the risk
premium is not a concern for short-maturity implied dividend yields.
Once I have an estimate for the implied dividend yield, I combine it with the realized dividend-price
ratio to calculate implied dividend growth rates.
3 Data
I use the S&P 500 index as a proxy for the aggregate market. I obtain the daily S&P 500 price index
(without dividends) and the S&P 500 total return index (with dividends reinvested) from Datastream.
I purchase the S&P 500 futures data from the Chicago Mercantile Exchange and the S&P 500 options
6
Lioui (2006) derives the put-call parity relation in the case of stochastic dividend yield by assuming complete markets
and a mean-reverting stochastic price of market risk, CI(1, t) 1I(1, t) = oI exp

`I(t)t 12(o
2
S
(t) (t)
2
)

1 exp[vI(t)t] , where 12(o

2
S
(t) (t)
2
) is a correction for the dividend risk premium.
11
data from Market Data Express. The futures data run from April 1982 through June 2011 and the
options data run from January 1990 through June 2011.
To analyze synchronicity issues in the derivative markets, I obtain for part of the sample intra-daily
data from Tick Data. The intra-daily cash index data cover the period from February 1983 through
June 2011. The intra-daily trade and quote data for S&P 500 options run from July 2004 through June
2011, and the newly released intra-daily trade and quote data for S&P 500 futures run from January
2010 to June 2011. To analyze demand imbalances in the options market, I also obtain for the sample
from January 1990 through June 2011 the SPX OpenClose data from Market Data Express. These
data provide separate daily observations for each option of opening and closing volume for dierent
investor groups (small customers, middle customers, large customers, and rm proprietary traders).
To compare implied dividends with dividend forecasts based on announced dividends, I obtain from
Standard & Poors the indicated dividend series for the S&P 500 index from February 1995 through
June 2011. To analyze dividend risk premium, I download implied dividend prices of Binsbergen et
al. (2012a) from Ralph Koijens website. To compare the period analyzed in this paper with longer
time periods, I download dividends data from Amit Goyals webpage. In some parts of the analysis I
also make use of other variables, I download daily constant-maturity 3-month and 6-month Treasury
bill rates, the default spread (dierence between Moodys BAA and AAA corporate bond spreads),
and the term spread (dierence between 10-year Treasury bond yield and 3-month Treasury bill rate)
from the Federal Reserve Bank of St. Louis. I obtain daily S&P 500 earnings-price ratio and 6-
month LIBOR data from Datastream. I download implied skewness from the Chicago Board Options
Exchange (CBOE), end-of-month variance risk premiums from Hao Zhous webpage, and the quarterly
consumption-to-wealth ratio from Sydney C. Ludvigsons website.
While European options on the S&P 500 index have been traded since April 1986, the Market
Data Express options data go back only to January 1990. Also, until 1994, the settlement procedures
for S&P 500 options and futures diered. While futures are settled in the opening value of the index
since June 1987, the most liquid S&P 500 options initially expired in the closing value of the index.
As liquid options and futures with matching expiration times are needed to estimate implied dividend
growth rates, I conne the main analysis to the period from January 1994 through June 2011.
7
7
Section 8.3 considers extending the time period back to 1990.
12
For all the derivatives data I regularly eliminate observations with missing values for any of the
main variables. With the help of a Market Data Express support team, I eliminate all but standard
S&P 500 options (that is, I eliminate quarterlies, weeklies, and mini options). I also eliminate options
that violate the basic arbitrage relations. All variables are sampled at the end-of-month.
3.1 Empirical estimation
Here I dene the main variables used in the empirical analysis, rst variables based on the realized
data and then variables related to dividends implied in the derivative markets.
Because dividend ratios are calculated by aggregating dividends over a given horizon (typically
one year), we need to take a stance on whether or not to reinvest the dividends in the market. Chen
(2009) and Binsbergen and Koijen (2010) show that when dividends are reinvested in the market, part
of the return properties are imputed in the dividend growth rates, and this adversely aects return
and dividend growth rate predictability results (see also Koijen and Van Nieuwerburgh, 2011). To
avoid such contamination, I aggregate realized dividends without reinvestment as in Ang and Bekaert
(2007). To ensure comparability with the implied dividend yield, as discussed below, I transform the
continuously compounded implied dividend yield in a raw dividend yield.
Monthly returns. Monthly returns are dened as
r
A
t
= log
_
1
t
+ 1
t
1
t1
_
, (14)
where 1
t
and 1
t
denote price and monthly dividends at the end of month t. Realized dividends are
calculated from the S&P 500 price index and the S&P 500 total return index. Since Datastream
reinvests dividends daily, I rst calculate daily dividends and then sum them over the past month.
Dividend-price ratio. The dividend-price ratio is dened as a 12-month trailing sum of realized
dividends (1
12
t
) over the current price:
dj
t
= log [11
t
] = log
_
1
12
t
1
t
_
. (15)
13
Dividend growth rates. Like Ang and Bekaert (2007), I calculate monthly dividend growth
rates based on a 12-month trailing sum of dividends:
d
A
t
= log
_
1
12
t
1
12
t1
_
. (16)
This approach to calculating dividend growth rates diminishes the eect of seasonality in dividend
payments, but introduces overlapping in dividend growth rates for any frequency higher than annual.
Implied dividend yield. I estimate implied dividend yield according to equation (13). I use
daily settlement prices for futures, the mid-point between the last bid and the last ask price for options,
and closing values for the S&P 500 price index.
To minimize eects of microstructure noise and market frictions, I calculate implied dividend yield
by aggregating information from a wide set of options and futures. At the end of each month, I use 10
days of (backward-looking) data and construct option pairs (put-call pairs with the same strike and
the same maturity) from all the reliable options (options with positive volume or open interest greater
than 200 contracts).
8
Then I combine option pairs with futures of matching maturity and the current
value of the underlying index. To eliminate some extreme observations, I discard observations where
1I(t)1
CI(1,t)1I(1,t)
is smaller than 0.5 or greater than 1.5, and where C
t
(1, t) 1
t
(1, t) is zero.
9
This
matching procedure results in several estimates of the implied dividend yield for a given maturity at
the end of each month (227 on average, with a minimum of 37 and a maximum of 935). The number of
estimates generally increases over time and declines with the maturity of the derivatives. I aggregate
these estimates at the end of each month by taking the median across all the implied dividend yields
with the same maturity.
As derivatives are most liquid at the short end, implied dividend yields estimated from near-
to-maturity options and futures should provide the most reliable estimates. Within year dividends
are highly seasonal, however, and a common approach to alleviate seasonality issues is to aggregate
dividends over one year. This suggests that the implied dividend yield should rather be calculated
from options and futures with one year to expiration. Considering both aspects, the choice of the
8
Untabulated results show that there is no strike price eect, i.e., the implied dividend yield does not depend on the
moneyness level.
9
This lter eliminates a bit under 2% of the observations.
14
appropriate maturity of the implied dividend yield depends on the trade-o between the within-year
seasonality of dividend payments and the liquidity of derivatives.
To examine this trade-o, Table 1 reports summary statistics for annualized realized dividends
summed over either 1 month, 3 months, 6 months, or 12 months, and Table 2 reports summary
statistics on the liquidity of derivatives.
Table 1 conrms that within-year dividends are highly seasonal. The standard deviation of annual-
ized monthly dividends is 7.36, which compares to a standard deviation of 4.61 for dividends summed
over 12 months. Thus, annualized monthly dividends are 60% more volatile than annual dividends.
Nevertheless, the volatility of dividends decays rapidly with increase of the horizon. When dividends
are summed over 6 months, the standard deviation of annualized dividends is 4.70, which is only 2%
higher than the standard deviation of dividends summed over 12 months. This suggests that the
problem of seasonality in dividend payments has largely diminished at a semi-annual frequency.
[Insert Table 1 about here]
Table 2 conrms that short-maturity derivatives are the most liquid. The tilt toward short matu-
rities is especially pronounced for futures. While 60-80% of options trading is concentrated on options
with maturities of less than 3 months, as much as 95% of futures trading is in short-maturity futures.
The market for S&P 500 futures, however, is among the most liquid derivative markets, and in notional
terms, there is a considerable amount of trading in the longer-maturity futures. The average daily
open interest of futures with times to maturity of between 3 and 6 months is $5.93 billion. Even open
interest of futures with maturities between 6 and 9 months amounts to $839 million per day. Hence,
although derivatives become less liquid with maturity, especially futures, there is a substantial amount
of trading in S&P 500 options and futures with maturities of 6 months.
[Insert Table 2 about here]
Given the trade-o between the seasonality of dividends and the liquidity of derivatives, I conduct
the main analysis on options and futures with six months to expiration.
10
Because options expire
10
Section 8.1 shows that the results are robust to using derivatives with either 5 or 7 months to expiration. Implied
dividend ratios based on shorter-maturity derivatives are more volatile and yield somewhat weaker results.
15
monthly and futures expire quarterly, there are only four dates each year when options and futures
expire simultaneously (the third Friday morning in March, June, September, and December). To
obtain monthly implied dividend yields with a constant 6-month maturity, I proceed as follows. For
months on a quarterly cycle, I calculate implied dividend yield using options and futures with 6 months
to maturity. For the other two months in the quarter, I interpolate linearly between implied dividend
yields with maturities closest to 6 months.
By construction, the estimated implied dividend yield is annualized and continuously compounded.
To make it comparable to the realized dividend-price ratio, I transform it into a raw (eective) implied
dividend yield, 111
t
= exp(

`
t
) 1. The log implied dividend yield is simply idj
t
= |oq(111
t
).
Implied dividend growth rates. The implied dividend growth rate is dened as the dierence
between the log implied dividend yield and the log dividend-price ratio. Given that the implied
dividend yield is based on derivatives with maturities of 6 months, it should ideally be matched with
the 6 month dividend-price ratio construct. As the analysis of seasonality suggests that annualized
dividends calculated over 6 months behave much like dividends summed over 12 months, however, I
rely on the standard dividend-price ratio and calculate implied dividend growth rates by combining
annualized 6 month implied dividend yield with the annual dividend-price ratio:
11
idq
t
= log
_
111
t
11
t
_
= idj
t
dj
t
. (17)
Corrected dividend-price ratio. The corrected dividend-price ratio is given by
dj
Ccvv
t
= dj
t
+
_
1
1 j
1
_
idq
t
, (18)
where j is the estimated linearization constant, and
1
is the 1(1) coecient of the implied dividend
growth rates.
11
Section 8.1 shows that the return predictability results are robust to using a dividend-price ratio based on an
annualized 6-month trailing sum of dividends.
16
3.2 Data description
Table 3 reports summary statistics for the main variables. All the variables are sampled monthly and
expressed in logs. Annualized returns and dividend growth rates are on average 7.84% and 3.78%,
respectively.
Figure 1 depicts construction of the implied dividend growth rates. Panel A plots implied dividend
yield along with the dividend-price ratio, and Panel B plots the dierence between both of them
the implied dividend growth rates. The implied dividend growth rates nicely reect general market
conditions. They are positive during the market booms of 1994-1997 and 2002-2007, when investors
were optimistic about future growth opportunities, and generally negative during stock market busts,
such as the Asian-Russian-LTCM crisis (1998), the dot.com bubble burst (2001), and the recent
nancial crisis (2008-2009), when investors were pessimistic about growth opportunities. The average
implied dividend growth rate is higher than the average realized dividend growth rate (7.31%), but
the dierence is not statistically signicant.
A somewhat disconcerting feature is rather high volatility, especially around the turn of the century
when implied dividend growth rate swings from 0.76 to +0.49. These uctuations are hard to justify
by changes in investors expectations and could be partially driven by data problems.
12
Because
implied dividend yield is estimated from no-arbitrage relations that involve highly leveraged positions,
any errors would be magnied in the implied dividend growth rates (Boguth et al., 2012). The period
of these high uctuations also coincides with historically low dividend yields (approximately 1% per
year in raw terms) when even tiny deviations of implied dividend yield from the realized dividend-
price ratio caused big swings in the implied dividend growth rate. Thus, at least part of the observed
uctuations could be explained by a scaling eect, exacerbated by the log scale. Most important,
however, as documented later, implied dividend growth rates reliably predict future dividend growth
rates, and do not seem to be inuenced by either demand imbalances in derivative markets or other
measures of risk. Thus, despite volatility and possible noise in the data, implied dividend growth rates
appear to be a good proxy for expected dividend growth rates.
12
Golez and Jackwerth (2012) note an unusually high number of missing observations of S&P 500 options in 1998-2002.
This pattern of missing observations is found in end-of-day options data provided by Market Data Express as well as in
the OptionMetrics data.
17
The AR(1) coecient of the implied dividend growth rates at monthly frequency is 0.53. Note
that because of the overlapping structure in the implied dividend growth rates, which invalidates the
AR(1) assumption of data, monthly AR(1) cannot be used to infer annual AR(1). Therefore, in the
the variance decomposition of the dividend-price ratio (Section 6), which is typically based on annual
variables, I use non-overlapping annual observations. In the predictability analysis, however, given
the relatively short time period, I use data sampled at monthly frequency, and make appropriate
adjustments for the overlap when necessary.
[Insert Table 3 about here]
[Insert Figure 1 about here]
The corrected dividend-price ratio is given by a linear combination of the dividend-price ratio and
the implied dividend growth rates
dj
Ccvv
t
= dj
t
+
_
1
1 j
1
_
idq
t
= dj
t
+
_
1
1 (0.9823 0.5335)
_
idq
t
= dj
t
+ 2.1011 idq
t
, (19)
where the linearization constant j and the persistence of the implied dividend growth rates
1
are
based on the whole sample.
13
Figure 2 plots dj
Ccvv
t
ratio along with the dj
t
ratio. Both dividend
ratios exhibit strong comovement (a pairwise correlation coecient of 0.72), but the patterns dier in
three important aspects.
[Insert Figure 2 about here]
First, the dj
Ccvv
t
ratio is on average higher than the dj
t
ratio in the boom periods and lower in the
bust periods. This means that the dj
t
ratio tends to predict returns that are too low to match the
markets optimism about growth opportunities during the boom periods. At the same time, the dj
t
13
Because bj and c
1
are estimated using the whole sample, the corrected dividend-price ratio is subject to a look-ahead
bias. Out-of-sample predictability results in Section 5.1 show, however, that look-ahead bias has a marginal inuence on
the results.
18
ratio tends to forecast returns that are too high during the crisis periods. This is particularly apparent
at the end of the sample, when the market experienced one of the greatest drops in the history of the
U.S. stock market, but the uncorrected dividend-price ratio rose and therefore predicted unrealistically
high returns.
Second, the corrected dividend-price ratio is notably more volatile than the uncorrected dividend-
price ratio. The standard deviation is 0.26 for the dj
t
ratio compared to 0.52 for the dj
Ccvv
t
ratio. While
part of this dierence could be accounted for by measurement errors, in the context of the present
value model, such an increase in volatility implies that expected returns and expected dividend growth
rates are highly correlated. To see this formally, expand the variance of the corrected dividend-price
ratio as
ar(dj
Ccvv
t
) = ar(dj
t
) + 2
_
1
1 j
1
__
1
1 jc
1
_
co(j
t
, q
t
)
_
1
1 j
1
_
2
ar(q
t
). (20)
Equation (20) says that the variance of the dj
Ccvv
t
ratio can be higher than the variance of the dj
t
ratio only if expected returns and expected dividend growth rates co-vary and the co-variation is high
enough
_
2
_
1
1jc
1
_
co(j
t
, q
t
)
_
1
1j
1
_
ar(q
t
)
_
. Since dj
t
ratio increases with expected returns
and declines with expected dividend growth rates, this positive co-variation osets shocks to expected
returns and to expected dividend growth rates, and reduces the volatility of the dj
t
ratio (Lettau and
Ludvigson, 2005, Rytchkov, 2008; Binsbergen and Koijen, 2010). Once we correct the dividend-price
ratio for changes in the expected dividend growth rates, we restore part of the variation, which is
otherwise oset by comovement of the expected returns and the expected dividend growth rates (see
also Lacerda and Santa-Clara, 2010).
Finally, the dj
Ccvv
t
ratio is also less persistent than the dj
t
ratio. The rst-order autocorrelation
coecient for the dividend-price ratio is 0.98, and the null hypothesis of a unit root cannot be rejected,
according to an augmented Dickey-Fuller (ADF) test. The AR(1) for the corrected dividend-price ratio
is 0.74, and the null of a unit root in the corrected dividend-price ratio is rejected at the 5% level. This
reduced persistence is important because highly autocorrelated predictors aggravate small-sample bias
(Stambaugh, 1999) and produce inaccurate inference results in the case of overlapping observations
19
(Boudoukh et al., 2008).
14
By applying equation (12) and following the same estimation procedure as for the implied dividend
yield, I also estimate the implied interest rate (111
t
). Although 111
t
is not of direct interest for this
study, it is worth noting that the 111
t
exhibits the expected behavior. Figure 3 shows that 111
t
strongly co-varies with the T-bill rate and LIBOR, and is on average closer to LIBOR (see also
Naranjo, 2009). Still, 111
t
is more volatile than the T-bill rate and LIBOR at the beginning of the
period analyzed, and it diverges from both proxies for the interest rate during the recent nancial
crisis, when it is notably lower than LIBOR. This points at the importance of isolating the eect of
interest rates when estimating the implied dividend yield.
[Insert Figure 3 about here]
4 Results
The present value model assumes an AR(1) structure for expected returns and expected dividend
growth rates. This implies that all the information is incorporated in the short-term (one horizon)
prices and dividends. But the model is silent about the appropriate length of the horizon. I consider
predicting dividend growth rates and market returns over horizons varying from one month to one
year. I use standard predictive regressions, in which returns or dividend growth rates are regressed on
lagged predictors. All regressions are based on monthly frequencies. The Hodrick (1992) t-statistics
reported explicitly take into account the overlapping structure of regressions.
15
I also report adjusted
1
2
s. Note, however, that the 1
2
s in the context of overlapping observations need to be interpreted
with caution because they tend to increase with the extent of the overlap even in the absence of true
predictability (Kirby, 1997; Valkanov, 2003; Boudoukh et al., 2008).
14
The reduced persistence also makes the corrected dividend-price ratio less prone to data snooping problems because
data mining tends to produce highly persistent predictors (Ferson et al., 2003).
15
Ang and Bekaert (2007) show that the Hodrick (1992) standard errors, which are based on summing predictors in
the past, perform better than other standards errors frequently employed in the literature, such as Newey-West (1987),
or Hansen and Hodrick (1980) standard errors.
20
4.1 Predicting dividend growth rates
Figure 1 shows that implied dividend growth rates track general market conditions and therefore seem
to be a good proxy for expected dividend growth rates. Now I test whether implied dividend growth
rates also predict future dividend growth rates. For a comparison, I use annual realized dividend
growth rates and the dividend-price ratio. I use the latter as a competitive predictor for two reasons.
First, the dividend-price ratio is itself a function of expected dividend growth rates and could therefore
predict not only future returns but also future dividend growth rates. Second, implied dividend growth
rates are dened as the dierence between the implied dividend yield and the dividend-price ratio.
Therefore, it is necessary to show that implied dividend growth rates do not predict future dividend
growth rates simply because they duplicate information contained in the dividend-price ratio.
The main dividend growth rates regression is
d
t+I
= a
0
+ a
1
(A
t
) + -
t+1
, (21)
where d
t+I
=
I

i=1
d
A
t+I
is the dividend growth rate with / = 1, 3, 6, or 12 months, and A
t
is a set
of predictor variables. Because of the overlapping structure in the realized dividend growth rates, I
use lagged annual dividend growth rates as a competitive predictor only in predictive regressions for
annual dividend growth rates.
Table 4 reports the results. The dividend-price ratio is always negatively related to future dividend
growth ratesjust as the theory suggestsbut the associated t-statistics are relatively low and vary
between 1.88 and 1.54. Accordingly, the associated ad,. 1
2
s are low, ranging from 1.59% for
predicting monthly dividend growth rates to 4.39% for predicting annual dividend growth rates. The
implied dividend growth rates by contrast are positively related to future dividend growth rates,
explaining 4.99% of the variation in monthly dividend growth rates and 19.05% of the variation in
annual dividend growth rates. Also, all the estimated parameters on the implied dividend growth
rates are highly statistically signicant, with t-statistics varying between 3.70 and 4.98.
The implied dividend growth rates remain signicant in bivariate predictive regressions with the
dividend-price ratio, which conrms that the implied dividend growth rates provide an independent
21
source of information about the future dividend growth rates. Moreover, in bivariate regressions, the
statistical signicance of both predictors (the implied dividend growth rates and the dividend-price
ratio) is higher than in the univariate regressions, and the ad,. 1
2
rises to 9.73% in a regression for
monthly dividend growth rates and to 32.86% in a regression for annual dividend growth rates.
16
Further, the implied dividend growth rates predict future dividend growth rates also better than
do the lagged realized dividend growth rates. The implied dividend growth rates exhibit higher ad,.
1
2
and t-statistics than the realized dividend growth rates in the univariate regressions, and remain
highly signicant in the bivariate regression with realized dividend growth rates. While the t-statistic
on the estimated parameter on realized dividend growth rates in the bivariate regression is 2.50, it is
4.48 for implied dividend growth rates.
[Insert Table 4 about here]
4.2 Predicting market returns
I employ three specications for the return predictive regressions. I start with the standard predictive
regression, in which returns are regressed on the lagged dividend-price ratio. Then I augment the
standard predictive regression with the implied dividend growth rates. Finally, I replace the dividend-
price ratio and the implied dividend growth rates with the corrected dividend-price ratio. The main
return regression is
r
t+I
= /
0
+ /
1
(A
t
) + -
t+1
, (22)
where r
t+I
=
I

i=1
r
A
t+i
is the market return with / = 1, 3, 6, or 12 months, and A
t
is a set of predictor
variables.
Note that the adjustment factor in the corrected dividend-price ratio depends on the persistence
16
This improvement in predicting dividend growth rates can be accounted for by the partial resolution of the omitted
variable bias. According to the present value model, a regression of dividend growth rates on the dividend-price ratio
is subject to the omitted variable bias because it lacks a control for expected returns. When expected dividend growth
rates are highly correlated with expected returns, a proxy for expected dividend growth rates (implied dividend growth
rates) can help to diminish the omitted variable bias. That omitted variable bias plays an important role in the dividend
growth rate predictive regressions is also documented in Binsbergen and Koijen (2010) and Koijen and Van Nieuwerburgh
(2011).
22
of expected dividend growth rates, which, according to the present value model, is measured over
the same horizon as returns. As all return regressions are based on monthly frequency, I let the
adjustment factor in the corrected dividend-price ratio vary with the return horizon as dj
Ccvv
t
=
dj
t
+
_
1,
_
1 j
\

1
(/)
__
idq
t
, where
\

1
(/) = 1(/) is estimated at a monthly frequency. Thus, in the
predictive regression for monthly returns, the adjustment factor is based on the AR(1) of the implied
dividend growth rates. The corrected dividend-price ratio in this case is the same as in Table 3 and
Figure 2. For longer return horizons, the impact of the implied dividend growth rates on the corrected
dividend-price ratio decreases with the length of the horizon in line with the decline in the persistence
of the implied dividend growth rates.
Table 5 reports the results. In accordance with the theory, the estimated parameter on the
dividend-price ratio is always positive, but the associated t-statistics are relatively low, especially
at the short end. The estimated parameter on the dividend-price ratio is signicant at the conven-
tional 5% signicance level only in the predictive regression for annual returns. The associated ad,.
1
2
s are relatively low and vary from 0.44% for predicting monthly returns to 17.62% for predicting
annual returns.
When the implied dividend growth rates are included as an additional regressor to the dividend-
price ratio, return predictability improves for all the horizons considered. The ad,. 1
2
rises from
0.44% to 4.49% in the regression for monthly returns and from 17.62% to 19.87% in the regressions
for annual returns. Results also conrm that we can replace the dividend-price ratio and the implied
dividend growth rate by a single predictor, the corrected dividend-price ratio. The corrected dividend-
price ratio predicts returns approximately as well as the dividend-price ratio and the implied dividend
growth rates together. The ad,. 1
2
amounts to 4.21% at the monthly horizon and to 19.04% at the
annual horizon. Estimated parameters on the dj
Ccvv
t
ratio are always statistically signicant, with
t-statistics varying between 2.42 and 3.10.
17
Because the corrected dividend-price ratio incorporates expectations about future dividend growth
rates, it should also be less prone to statistical biases found in regressions with the dividend-price ratio.
17
Untabulated results show that the F-test fails to reject restrictions regarding predictive coecients in the bivariate
return regressions with the dividend-price ratio and the expected dividend growth rates
\
/
u
t
1
(I) = (1bj
\

1
(I))
\
/
.u
t
1
(I),
at all the considered horizons.
23
Specically, positive news about future dividends produces high returns and low dividend-price ratios.
This induces negative correlation between shocks to returns and shocks to the dividend-price ratio,
contributing to an upward bias in the predictive coecient (Fama and French, 1988; Stambaugh,
1999; Amihud and Hurvich, 2004, Polk et al. 2006; Boudoukh et al., 2008). The more persistent the
predictor, the greater the bias. The dividend-price ratio is highly persistent and shocks to returns and
shocks to the dividend-price ratio are highly negatively correlated at 0.97. Simulations indicate that
the upward bias in this case can be just as high, or even higher, then the predictive coecient on the
dividend-price ratio found in the data.
18
Unlike the dividend-price ratio, the corrected dividend-price
ratio incorporates expectations about future dividends, which reduces negative correlation between
shocks to returns and shocks to the predictor variable. I nd that the correlation between shocks
in the case of the corrected dividend-price ratio is only 0.23. Since the corrected dividend-price
ratio is also less persistent, this evidence suggests (and simulations conrm) a very low bias in the
predictive coecient (2%3%, depending on the return horizon). Simulations also conrm that the
predictive coecient on the realized dividend-price ratio is insignicant, with bootstrapped p-values
between 29% and 48%, depending on the horizon, compared to p-values of between 0% and 1% for
the corrected dividend-price ratio.
19
Further, the same results cannot be obtained using lagged dividend growth rates as opposed to the
implied dividend growth rates. Inclusion of the realized dividend growth rates in the return predictive
regressions with the dividend-price ratio and the implied dividend growth rates leaves the ad,. 1
2
largely unchanged. The realized dividend growth rate is statistically insignicant at all the horizons.
[Insert Table 5 about here]
18
The simulated model is
v
I+I
= c + ,
0
rI + &I+1,
rI+1 = 0 + crI + I+1,
where I is either 1, 3, 6, or 12, and rI is the predictor variable. Error terms are bivariate-normal. I simulate monthly
returns under the null of no predictability (, = 0). Longer-horizon returns are constructed from simulated monthly
returns. In 100,000 simulations, bias is measured as the mean of the predictive coecients. Bootstrapped p-values are
calculated as the percentage of simulated coecients that are farther away from zero than the coecients found in the
data.
19
For the case of non-overlapping monthly return regressions, approach proposed by Amihud and Hurvich (2004)
conrms that the bias-adjusted estimated parameter on the corrected dividend-price ratio is essentially identical to the
OLS estimated parameter (0.0183 versus 0.0188), and the adjusted t-statistic is 3.10, which is identical to the Hodrick
(1992) t-statistic. This compares to a slightly negative estimated parameter on the realized dividend-price ratio, after
adjustment (0.0010) and an adjusted t-statistic of 0.07.
24
5 Additional tests
To further validate the ability of the corrected dividend-price ratio to predict returns, I next examine
stability of in-sample forecasting relations and consider predicting returns out-of-sample. I also analyze
how the corrected dividend-price ratio compares to other popular return predictors, including measures
of risk and demand imbalances in the derivative markets. In addition, I report results for the subsample
analysis. To avoid statistical problems inherent in the use of overlapping observations (Kirby, 1997;
Boudoukh et al., 2008), but to maximize the number of observations, I restrict the analysis henceforth
to prediction of non-overlapping monthly returns.
5.1 Parameter stability and out-of-sample predictability
The forecasting relation of returns and the realized dividend-price ratio is found to be unstable, and
the dividend-price ratio typically fails to predict returns out-of-sample (Lettau and Van Nieuwerburgh,
2008; Goyal and Welch, 2008).
20
To examine whether the corrected dividend-price ratio produces a more stable forecasting relation,
Figure 4 shows predictive coecients and t-statistics for predictive regressions of monthly returns on
either the dividend-price ratio or the corrected dividend-price ratio using 120-month rolling windows.
As expected, the forecasting relation between returns and the dividend-price ratio is very unstable.
During the recent nancial crisis, the dividend-price ratio even became negatively related to future
returns. The predictive coecient on the dividend-price ratio can be anywhere between 0.03 and
0.07. The estimated parameter on the corrected dividend-price ratio, however, is much more stable,
as well as always positive, between 0.01 and 0.05. Its t-statistics are always well above the t-statistics
associated with the dividend-price ratio. Thus, the forecasting relation between returns and the
corrected dividend-price ratio is both strong and stable.
[Insert Figure 4 about here]
Next, I examine the ability of the corrected dividend-price ratio to predict returns in real time. I
calculate the out-of-sample 1
2
as in Campbell and Thompson (2008) and Goyal and Welch (2008).
20
An in-depth discussion on the possibility of predicting returns out-of-sample is provided in a special issue of the
Review of Financial Studies (Spiegel, 2008).
25
1
2
OS
= 1
'o1
1\
'o1
1
, (23)
where 'o1
1\
=
1
T1
T

t=1
(r
t+1
j
t
)
2
is the mean-squared error based on the tted values ( j
t
) from
a predictive regression estimated through period t, and 'o1
1
=
1
T1
T

t=1
(r
t+1
r
t
)
2
is the mean-
squared error based on the historical average return (r
t
) estimated through period t. A positive
out-of-sample 1
2
indicates that the predictive regression has a lower mean-squared prediction error
than the historical average return. I measure the statistical signicance of results using the 'o11
statistic proposed by McCraken (2007).
To make out-of-sample forecasts, I split the sample to two subperiods. I use the period from
January 1994 through December 1999 for estimation of the initial parameters and the period from
January 2000 through June 2011 for calculation of the 1
2
OS
. All out-of-sample forecasts are based on
a recursive scheme using all the available information up to time t.
Recall that the corrected dividend-price ratio is dened as dj
Ccvv
t
= dj
t
+
_
1
1bjc
1
_
idq
t
, where
j (a linearization constant) and
1
(persistence of the implied dividend growth rates) are estimated
using the whole sample period, therefore introducing a slight look-ahead bias. To alleviate the concern
that the look-ahead bias could inuence results, I additionally estimate a so-called no-look-ahead-bias
corrected dividend-price ratio, dj
Ccvv.11
t
= dj
t
+
_
1
1bjc
1
_
idq
t
, where j
t
and
t
are time-varying
and estimated using the same recursive scheme as in calculation of the out-of-sample 1
2
OS
.
Table 6 reports the results. The dj
t
ratio with its poor ability to predict returns in-sample also fails
to predict returns out-of-sample. The 1
2
OS
for the dj
t
ratio is 0.04%, compared to an out-of-sample
1
2
OS
for the dj
Ccvv
t
of 5.33% (statistically signicant at the one percent signicance level). Thus, the
dj
Ccvv
t
ratio both predicts returns in-sample and delivers better out-of-sample forecasts of monthly
returns than forecasts based on the historical average.
21
Moreover, approximately the same 1
2
OS
, if
not even slightly higher, is obtained with a corrected dividend-price ratio adjusted for the look-ahead
bias (1
2
OS
5.35%). Hence, look-ahead bias is not a concern, and the dj
Ccvv
t
ratio can be eectively be
21
Untabulated results show that the dj
Corr
I
ratio predicts returns out-of-sample better than a constant of 6% per year,
which has been found to be the best forecasting variable in Simin (2008).
26
used in real time for portfolio allocation decisions.
22
[Insert Table 6 about here]
[Insert Figure 5 about here]
To illustrate the relative success of the dj
Ccvv
t
ratio in predicting returns out-of-sample, Figure 5
plots out-of-sample forecasts and realized returns. Although realized returns are signicantly more
volatile than any of the forecasted returns, there are considerable dierences across the forecasts.
The forecasts based on the realized dividend-price ratio and the forecasts based on the historical
average return are both relatively smooth. Forecasts based on the corrected dividend-price ratio vary
signicantly more, and the changes of the forecasts are often of the same sign as the changes of the
realized returns.
5.2 Alternative predictors
Next I analyze how corrected dividend-price ratio compares to other return predictors. For the rst
alternative predictor, I construct a dividend-price ratio corrected for average historical dividend growth
rates. Lacerda and Santa-Clara (2010) show that correcting the dividend-price ratio for the 10-year
moving average of dividend growth rates improves the predictability of longer-horizon (annual) returns.
As my focus is on short-horizon predictability, and the 10-year moving average of dividend growth rates
evolves rather slowly, I calculate average historical dividend growth rates as a moving average of 12
months of annualized monthly dividend growth rates. For comparability with the dividend-price ratio
corrected for implied dividend growth rates, the persistence of the historical dividend growth rates is
presumed the same as the persistence of the implied dividend growth rates. Thus, the dividend-price
ratio adjusted for the historical dividend growth rates is dened as dj
1ict
t
= dj
t
+ 2.10 dq
A
t
, where
dq
A
t
is the moving average of annualized monthly dividend growth rates over the past 12 months. In
addition, I use two other standard return predictors, the earnings-price ratio cj
t
, and the consumption-
22
The rather small dierence in the 1
2
CS
between the two corrected dividend-price ratios occurs because the persistence
of the implied dividend growth rates b
I
and the linearization constant bj
I
are very stable (b
I
ranges from 0.4713 to 0.6185,
and bj
I
is between 0.9814 and 0.9833). This makes dj
Corr
I
and dj
^1/TCorr
I
highly correlated (0.9988) and almost
indistinguishable.
27
to-wealth ratio caj
t
proposed by Lettau and Ludvigson (2001).
23
Because the implied dividend growth rates and consequently the corrected dividend-price ratio
could be inuenced by the risk premium, I analyze how the corrected dividend-price ratio relates to
other risk related predictors, such as the default spread dc)
t
, the term spread tcr:
t
, and the variance
risk premium as implied in the S&P 500 index. The variance risk premium is especially relevant
because it is also based on information implied in the S&P 500 options. I employ two versions of
the variance risk premium. The rst version, dened as in Bollerslev et al. (2009), measures the
dierence between the variance implied in the S&P 500 options and the realized variance based on the
intra-daily data for the S&P 500 Index rj
1\ 1
t
. The second version is based on the dierence between
the implied variance and the forecasted variance (forecasts are based on the VIX and the intra-daily
realized variance) rj
11\ 1
t
.
Implied dividend yield is estimated from no-arbitrage relations. Consequently, any violations of
no-arbitrage relations would impact implied dividend growth rates. Since violations of put-call parity
and related measures, such as implied skewness, are shown to predict returns in the cross-section (Xing
et al. 2010; Cremers and Weinbaum, 2010; Conrad et al., 2012), I consider whether these results carry
over to the aggregate market. For this purpose, I use the skewness implied in 30 days-to-maturity
S&P 500 options as provided by the CBOE :/cn
t
. Because violations of no-arbitrage relations are
related to the transactions volume in puts and calls to open new positions (Cremers and Weinbaum,
2010), I also calculate open-buy volume based on the open-close data for S&P 500 options. I dene
an open-buy variable o/
t
as the dierence between the opening volume for calls and puts.
If demand imbalances aect prices of derivatives, then these imbalances could aect not only
implied dividend growth rates, but also the estimates of implied interest rate, especially because they
are estimated simultaneously and are based on the identical data. Therefore, I additionally use the
spread between 6-month LIBOR and the 6-month implied interest rate :jr
111
t
. Because LIBOR is an
indicated average rate for all the large banks, and the implied interest rate reects the interest rate for
arbitrageurs (presumably those with the cheapest access to money), the spread may not reect only
23
Lettau and Van Nieuwerburgh (2008) adjust the dividend-price ratio for structural breaks in its mean. Using data
from 1927 through 2004, they nd strong evidence in favor of a structural break in the early 1990s (and possibly another
one in the 1950s). When I consider the period from January 1994 until June 2011 in isolation, however, I do not nd
evidence for structural breaks, according to tests developed by Bai and Perron (1998). SupF tests of zero versus one
break, or zero versus two breaks, are both insignicant, and sequential test chooses the number of breaks as zero.
28
market frictions, but also credit risk.
Summary statistics. Table 7 reports the basic summary statistics and the unconditional cor-
relation structure for all the variables sampled monthly.
24
The standard return predictors are highly
persistent, and the hypothesis of a unit root for most of them cannot be rejected. The persistence
of the dividend-price ratio corrected for average historical dividend growth rates is only marginally
lower than the persistence of the realized dividend-price ratio. Dividend-price ratio corrected for im-
plied dividend growth rates, variance risk premiums, and measures of market frictions, however, are
substantially less persistent, and the null of a unit root is strongly rejected in most cases.
The realized dividend-price ratio, the dividend-price ratio corrected for historical dividend growth
rates, and the dividend-price ratio corrected for implied dividend growth rates are all highly correlated.
Also, they are all positively correlated with the earnings-price ratio and the consumption-to-wealth
ratio. Importantly, the dividend-price ratio corrected for implied dividend growth rates exhibits overall
much lower correlations with the risk-related predictors than the realized dividend-price ratio.
Both dividend ratios also exhibit low correlations with the variables measuring demand imbalances
and market frictions. The only exception is the relation with the spread between LIBOR and the
implied interest rate. While :jr
111
t
is positively correlated with the realized dividend-price ratio,
it is negatively correlated with the corrected dividend-price ratio. This dierence, however, can be
explained by the :jr
111
t
capturing information about the credit risk. Indeed, :jr
111
t
is strongly
correlated with the default spread, which is related to the dividend-price ratio, but unrelated to the
corrected dividend-price ratio.
[Insert Table 7 about here]
Predicting market returns. Table 8 reports results for predicting monthly returns. Since in-
sample and out-of-sample results are largely consistent, I evaluate predictors mainly on the in-sample
evidence. The traditional predictors based on the realized data explain only a small portion of the
variation in future monthly returns. The univariate return regressions with either the earnings-price
24
Since coI is available only quarterly, monthly observations of coI are dened by the most recently available quarterly
observation.
29
ratio, the consumption-to-wealth ratio, the term spread, or the default spread exhibit ad,. 1
2
s between
minus 0.5% and 1%. The dividend-price ratio, as already documented, explains 0.44% of the variation
in future monthly returns.
Correcting the dividend-price ratio for average historical dividend growth rates does not seem to
improve the predictability of monthly market returns in the period analyzed. The ad,. 1
2
in the return
regression with the dj
1ict
t
is approximately the same, if not slightly lower, than the ad,. 1
2
in the
regression with the realized dividend-price ratio.
25
Furthermore, predictive regressions with the skewness and open-buy exhibit slightly negative ad,.
1
2
s. Thus, the results as to the predictive ability of those variables from the cross-section do not seem
to carry over to the aggregate market. This is in line with Pan and Poteshman (2006), who nd that
put-call ratios from option volume initiated by buyers to open new positions predict stock returns,
but they fail to predict returns on the aggregate market. Like the default spread, the spread between
LIBOR and the implied interest rate is negatively related to future returns. It is marginally signicant
at 10% level and explains 1.89% of the variation in monthly returns.
The dividend-price ratio corrected for implied dividend growth rates dj
Ccvv
t
and the variance risk
premium based on the realized variance rj
1\ 1
t
, however, explain substantially more of the variation
in future monthly returns. The rj
1\ 1
t
is positively related to future returns, with an ad,. 1
2
of
4.62%. The dj
Ccvv
t
, as already documented, has an ad,. 1
2
of 4.21%. The rj
1\ 1
t
and the dj
Ccvv
t
are also the only predictors that are signicant at conventional levels of statistical signicance. The
t-statistics are 2.13 on the variance risk premium and 3.10 on the corrected dividend-price ratio. Note,
however, that the predictability results using the variance risk premium depend on how it is dened.
Unlike the variance risk premium based on the realized variance rj
1\ 1
t
, the variance risk premium
based on the forecasted variance rj
11\ 1
t
fails to predict returns in the period analyzed.
All in all, the corrected dividend-price ratio and the variance risk premium based on the realized
variance are the strongest return predictors of the variables considered. Since they are both based
on information extracted from the derivative markets, the relative success of the dj
Ccvv
t
in predicting
25
Untabulated results show that this result is robust to using alternative proxies for historical dividend growth rates,
such as lagged monthly dividend growth rates or the moving average of 10 years of monthly dividend growth rates.
Furthermore, results are as well robust to using the adjusted dividend-price ratio proposed by Lacerda and Santa-Clara
(2010), in which the adjustment factor does not depend on the persistence of expected dividend growth rates.
30
future returns could be driven by the fact that the dj
Ccvv
t
is duplicating information in the variance
risk premium. To address this concern, I show that the corrected dividend-price ratio remains highly
signicant in a bivariate regression with the variance risk premium. Moreover, in a bivariate regression,
the statistical signicance of both predictors increases and the ad,. 1
2
rises to as much as 10.03%.
Both predictors also predict returns out-of-sample with an 1
2
OS
of 11.37%. The corrected dividend-
price ratio as well remains highly signicant in bivariate regressions with other measures of risk and
market frictions.
[Insert Table 8 about here]
5.3 Subsample Analysis
The period of analysis runs from January 1994 through June 2011 and covers several of turmoil periods,
including the recent nancial crisis. Such periods have an important impact on forecasting relations.
Figure 4 has shown that the forecasting relation between returns and the dividend-price ratio at the
peak of the nancial crisis momentarily even changed sign.
Table 9 repeats all the predictive regressions by truncating the period at December 2007. The
variance risk premium based on the realized variance rj
1\ 1
t
, which performs very well in the full
sample, now becomes an insignicant predictor for future returns and loses its advantage to the variance
risk premium based on the forecasted variance.
Cash-ow related predictors, however, such as the realized dividend-price ratio, the earnings-price
ratio, and the consumption-to-wealth ratio, now become signicantly related to future monthly returns.
The realized dividend-price ratio alone exhibits an ad,. 1
2
of 2.83%. Correcting the dividend-price
ratio for the historical dividend growth rates improves the predictability of returns, with an ad,. 1
2
of 4.02%, although still lower than the ad,. 1
2
of 6.49% for the dividend-price ratio corrected for
implied dividend growth rates. Interestingly, just as in the full period, the corrected dividend-price
ratio predicts returns especially well jointly with the variance risk premium rj
1\ 1
t
. The in-sample
ad,. 1
2
and the out-of-sample 1
2
OS
in the bivariate regression are 9.50% and 12.64%, respectively.
All in all, the corrected dividend-price ratio emerges as the strongest predictor of monthly returns
when the nancial crisis is excluded. But its relative advantage over the realized dividend-price
31
ratio is greater when the nancial crisis is included. This is intuitive, as the outbreak of the crisis
completely altered expectations about future cash ows and made lagged realized dividend growth rates
an overoptimistic proxy for expected dividend growth rates.
26
As a result, the standard dividend-price
ratio as well as the dividend-price ratio corrected for the historical dividend growth rates are both too
high to be justied by the general pessimism in the market. The dividend-price ratio corrected for the
implied dividend growth rates, however, incorporates forward-looking expectations, and is thus better
able to predict future returns.
[Insert Table 9 about here]
6 What moves the prices?
The general conclusion in the literature is that price movements are due primarily to changes in dis-
count rates and not so much to changes in expected dividend growth rates (Cochrane, 1992; Cochrane,
2005; Ang and Bekaert, 2007; Binsbergen and Koijen, 2010; Koijen and Van Nieuwerburgh, 2011).
Yet the predictability results of this paper suggest that variation in expected dividend growth rates is
important for price movements. Therefore, I now explore further the implications of the present value
model to provide additional insights into the interpretation of results.
According to equation (9), the estimated parameter on the corrected dividend-price ratio equals
1 j
t

c
1
. The persistence of expected returns

c
1
can therefore be calculated from the estimated
parameter on the corrected dividend-price ratio and the linearization constant j
t
.
27
Further, we can
use equation (7) to decompose the variance of the dividend-price ratio to two covariance terms as
ar(dj
t
) '
_
1
1 jc
1
_
co(dj
t
, j
t
)
_
1
1 j
1
_
co(dj
t
, q
t
), (24)
where the rst term on the right-hand side represents the contribution of expected returns (discount
26
Untabulated results conrm that the crisis also had an important impact on the dividend growth rate predictability.
While implied dividend growth rates reliably predict future dividend growth rates (including or excluding the crisis), the
forecasting relation between the dividend-price ratio and the dividend growth rates changes sign and the dividend-price
ratio becomes positively (wrongly) related to future dividend growth rates when the crisis is excluded. Further, as
expected, lagged dividend growth rates predict future dividend growth rates better when the crisis is excluded.
27
Alternatively, the persistence of expected returns can be identied from the estimated parameter on the dividend-
price ratio or the implied dividend growth rates, according to equation (8). Both appraoches yield very similar results.
32
rates), and the second term the contribution of expected dividend growth rates (cash ow). I stan-
dardize the right-hand side by the left-hand side, so that both terms sum up 100%. If we alternatively
decompose the variance of the dividend-price ratio to two variance terms and a covariance term, we
can also calculate the correlation between expected returns and expected dividend growth rates.
28
The variance of expected returns can be estimated by inverting equation (7).
So far all results have been based on data sampled at monthly frequencies. Dividend ratios, how-
ever, are based on realized dividends summed over 12 months, which invalidates the AR(1) assumption
of data sampled monthly. Therefore, the analysis should best be conducted on non-overlapping an-
nual observations (see also Binsbergen and Koijen, 2010). As the time period in the main analysis is
rather short, but the data are available monthly, I rely on non-overlapping annual observations and
estimate all parameters 12 times using successive annual samples starting at the end of each month
(end of Januarynext January, end of Februarynext February, and so on). I report the mean and
the interquartile range for the estimated parameters. This approach does not avoid the problem of
overlapping observations, but lets us use more data and examine how stable the estimated parameters
are.
The relatively short time period also raises a concern that the results may be sample specic.
Therefore, for a comparison with the implied dividend growth rates, I rst estimate the parameters
using lagged annual dividend growth rates as a proxy for expected dividend growth rates. Using
realized dividend growth rates I also examine how the sample period analyzed in this paper compares
to longer time periods. In Model (1), I report results based on the period from January 1926 through
June 2011. Model (2) collects results for the post-war period from January 1946 through June 2011
and Model (3) reports results for the period from January 1994 through June 2011.
Next I use implied dividend growth rates as a measure of expected dividend growth rates. As dis-
cussed earlier, implied dividend growth rates could be contaminated by market frictions, risk premium,
and other measurement errors. Because measurement errors aect the volatility and the persistence of
expected dividend growth rates, they could importantly inuence the variance decomposition results.
28
The variance decomposition of the dividend-price ratio in this case is
ov(djI) '

1
1p
1

2
ov(j
I
) +

1
1p
1

2
ov(jI) 2

1
1p
1

1
1p
1

co(j
I
, jI).
Note that assigning half of the covariance term to expected returns and half to expected dividend growth rates would
yield the same results as the variance decomposition in equation (24).
33
To address this concern, I use four dierent versions of implied dividend growth rates. In Model (4),
I use the implied dividend growth rate as is. In Model (5), I clean implied dividend growth rates
of variation due to demand imbalances and market frictions in the options market. Specically, I
use residuals in the regression of the implied dividend growth rates on open-buy volume and implied
skewness. In Model (6), I clean implied dividend growth rates of co-variation with risk measures.
Here I use residuals from the regression of implied dividend growth rates on the default spread, term
spread, and variance risk premiums. In Model (7), I decompose implied dividend growth rates into
a cyclical component and a trend component according to the Hodrick-Prescott algorithm with a
smoothing parameter of 100. I take an extreme view, treat the entire cyclical component as noise, and
report results based on the trend component only. Note, however, that this could lead to ltering out
important information from implied dividend growth rates.
Table 10 reports the results. Using lagged dividend growth rates as a measure of expected dividend
growth rates, I nd the results consistent with ndings of recent studies (e.g. Binsbergen and Koijen,
2010; Koijen and Van Nieuwerburgh, 2011).
29
As reported in Columns 1 and 2, in the period from
Janaury 1926 through 2011 as well as in the period from January 1946 through 2011, most of the
variation in the dividend-price ratio is due to the discount rate eect (98% and 107%), and almost
none due to the dividend growth eect (2% and 7%). Further, expected returns are much more
persistent (0.97 and 0.92) than expected dividend growth rates (0.34 and 0.37). Expected returns are
as well less volatile than expected dividend growth rates. Both are positively correlated (0.30 and
0.48).
In the period from January 1994 through June 2011 (Column 3) expected returns and expected
dividend growth rates are somewhat less persistent (0.75 and 0.32), but remain positively correlated
at 0.37. Like in the period from Janaury 1926 through June 2011, I nd that 98% of variation in the
dividend-price ratio is accounted by the discount rate eect and only 2% of the variation is driven by
the dividend growth eect. Thus, the main period analyzed in this paper is by and large comparable
to longer time periods.
When I use implied dividend growth rates (rather than lagged dividend growth rates) as a measure
29
Binsbergen and Koijen (2010) estimate the present value model by means of the maximum likelihood. Ma and Wohar
(2011) raise a concern that estimates depend on the modelling assumptions.
34
of expected dividend growth rates, expected returns and expected dividend growth rates become
more volatile (Column 4). The volatility of expected returns increases from 7.67% to 9.34% and
the volatility of expected dividend growth rates increases from 8.13% to 18.09%. The persistence
of expected returns rises to 0.78 and the persistence of expected dividend growth rates declines to
0.22. Despite the reduced persistence of expected dividend growth rates, however, the covariance
eect outweighs and the dividend growth eect becomes much more important for price movements.
Expected dividend growth rates account for 34% of variation in the dividend-price ratio, and the
discount rate eect accounts for 134% of its variation.
30
The negative sign on the contribution of the
dividend growth eect is driven by the positive covariance between the dividend-price ratio and the
expected dividend growth rates, which occurs in the present value model only when expected returns
and expected dividend growth rates are highly correlated. Indeed, I nd the correlation coecient to
be 0.74, which is twice the value of the correlation coecient documented in the model with lagged
dividend growth rates (Column 3).
Thus, in comparison to the benchmark case and the typical values reported in the literature,
expected dividend growth rates have a more important impact on price movements and are much
more correlated with expected returns. This suggests that the osetting eects of expected returns
and expected dividend growth rates on the dividend-price ratio documented by Menzly et al. (2004)
and Lettau and Ludvigson (2005) are magnied and the omitted variable bias in the standard dividend-
price ratio predictive regression is more severe than implied by the ndings of the recent studies.
Simulations show that adding an additional layer of noise on the implied dividend growth rates
preserves the point estimates.
31
The point estimates also remain largely unchanged if implied dividend
growth rates are cleaned of variation due to either market frictions in the options market (Column 5)
or other measures of risk (Column 6).
The results, nevertheless, need to be interpreted with caution as there is considerable variation in
30
Using a three-way decompostion as in Binsbergen and Koijen (2010), the results are even more dramatic. In this
case 270% of the variation in the dividend-price ratio is due to variation in expected returns, 102% is due to time-varying
expected dividend growth rates, and 272% is accounted for by the covariance between expected returns and expected
dividend growth rates.
31
In 100, 000 simulations, contaminating implied dividend growth rates with a layer of noise with volatility equal
to 10% (50%) of the standard deviation of the implied dividend growth rates results in the dividend growth eect of
34.14% (32.02%) and the correlation coecient between expected returns and expected dividend growth rates of
0.7362 (0.7383).
35
the estimated parameters between annual samples. The interquartile range suggest that the dividend
growth eect in the main model (Column 4) can be anywhere between 75% and 7%. The correlation
between expected returns and expected dividend growth rates varies between 0.54 and 0.96.
As reported in Column 7, the interquartile range for all the parameters decreases substantially when
implied dividend growth rates are cleaned of the cyclical component. The variance of expected dividend
growth rates declines and their persistence increases. The dividend growth rate eect becomes more
pronounced and varies between 36% and 64%. Similarly, the correlation between expected returns
and expected dividend growth rates increases and exhibits lower variability (0.82 to 0.90). Hence,
insofar Hodrick-Prescott algorithm is useful in disentangling measurement errors from the genuine
uctuations, this evidence suggests that measurement errors could be understating the importance of
the dividend growth rate eect.
[Insert Table 10 about here]
7 Announced dividends and dividend risk premium
Companies typically announce dividends one to three months in advance, and we can use this infor-
mation for predicting aggregate dividend growth rates. Although publicly available information about
dividends should already be incorporated in prices and thus unrelated to future returns, the fact that
dividends are announced in advance has important implications because it aects the dividend risk
premium embedded in the implied dividend growth rates. As discussed earlier, when future dividend
yield is stochastic, implied dividend growth rates may be contaminated by the dividend risk premium.
The dividend risk premium is mitigated, however, when part of dividends is known. Holding the price
of risk constant, the larger the part of the dividends that is known, the smaller the role for the dividend
risk premium. Therefore, it is instructive to analyze to what extent annual dividend growth rates can
be predicted from the information on announced dividends.
For this purpose, I obtain data on 12-month indicated dividends for the S&P 500 index from
Standard & Poors. Indicated dividends are calculated by annualizing announced dividends of all
the constituents of the S&P 500, where announced dividends enter into the calculation at the time
36
of announcement. The data are monthly and run from February 1995 onward. I dene indicated
dividend growth rates dq
1.1
t
as the log dierence between the 12-month indicated dividends and the
12-month trailing sum of realized dividends.
The results summarized in Table 11 show that indicated dividend growth rates are a strong pre-
dictor of future dividend growth rates. While the dividend-price ratio predicts dividend growth rates
in the period analyzed with an ad,. 1
2
of 6.62%, indicated dividend growth rates explain as much
as 72.50% of the variation in future dividend growth rates. Indicated dividend growth rates are also
highly signicant with a Hodrick t-statistic of 4.63. Thus, even at an annual horizon, a large part
of dividends is easy to predict, and hence riskless. Therefore, the dividend risk premium embedded
in implied dividend growth rates, which is based on short-maturity derivatives, should be largely
diminished.
32
To formally test for the presence of the dividend risk premium, I repeat the analysis of Binsbergen
et al. (2012a) using data for 6-month dividend prices. In particular, I consider a trading strategy
that involves buying a claim on 6-month dividends every June and December, holding the claim until
maturity, and collecting the accrued dividends. Based on Binsbergen et al. (2012a) data for 6-month
dividend prices in the period from February 1996 through October 2009, the return on the trading
strategy is 4.01%, which is approximately the same as the risk-free rate, and 2 to 3 times lower than
the return on a similar trading strategy based on longer-maturity dividend prices (12- to 24-months).
Thus, the high risk premium for longer-maturity dividend claims documented by Binsbergen et al.
(2012a) is absent in the short-horizon dividend claims where a large part of dividends is known in
advance. I also nd similar results based on the dividend prices estimated from options and futures.
Table 11 also conrms that indicated dividend growth rates are unrelated to future returns. In-
dicated dividend growth rates in a bivariate regression with the dividend-price ratio are insignicant.
Moreover, the ad,. 1
2
in a bivariate regression is lower than in the standard dividend-price ratio
predictive regression without the indicated dividend growth rates.
[Insert Table 11 about here]
32
Untabulated results show that indicated dividend growth rates predict semi-annual dividend growth rates with the
od). 1
2
of 81.52%.
37
8 Robustness checks
Next I analyze how sensitive results are to several methodological changes in calculation of the implied
dividend growth rates. In all the tests below, I report summary statistics for implied dividend growth
rates. I also test how well implied dividend growth rates predict annual dividend growth rates and
how well the corrected dividend-price ratio predicts monthly returns. To enhance comparability across
versions of the corrected dividend-price ratios, I impose the persistence of the implied dividend growth
rates to be the same as the persistence of the implied dividend growth rates the main analysis. In
other words, the corrected dividend-price ratio is always calculated as the dividend-price ratio plus
2.10 times the implied dividend growth rate. I also report the variance decomposition results and the
correlation coecient between expected returns and expected dividend growth rates.
8.1 Seasonality of dividends and maturity of derivatives
The implied dividend growth rates analyzed throughout have been dened as the dierence between
the log implied dividend yield and the log realized dividend-price ratio idq
t
= idj
t
dj
t
. Given the
trade-o between seasonality in dividend payments and the liquidity of derivatives, results so far have
been based on the annualized implied dividend yield with 6 months to maturity and the realized
dividend-price ratio estimated as usual by summing dividends over the past 12 months. To address
a concern that the maturity mismatch between the implied dividend yield and the realized dividend-
price ratio as well as seasonality in dividend payments could be driving the documented improvement
in predicting returns, I consider two additional tests.
In the rst test, I re-estimate implied dividend growth rates using implied dividend yields with
maturities of between 3 and 7 months. In the second test, I repeat that exercise, but instead of the
standard dividend-price ratio with dividends summed over the past 12 months I use the dividend-price
ratio based on annualized dividends summed over the past 6 months, dj
6n
t
= log
_
21
6
I
1I
_
, where 1
6
t
is the 6 month trailing sum of dividends.
Table 12 and Table 13 report results for the rst and the second test, respectively.
33
I analyze
33
Because implied dividend yields with short maturities are very volatile, the implied dividend yield at maturity of 3
months has a negative value on three occasions (December 1999, May 2000, and July 2000). I replace these observations
with the implied dividend yield at maturity of 4 months.
38
results in Table 12 rst. The results seem to support two general conclusions. First, whatever the
maturity of the derivatives, implied dividend growth rates remain a signicant predictor for annual
dividend growth rates, and the corrected dividend-price ratio predicts returns better than the realized
dividend-price ratio. Second, maturity of the implied dividend yield matters. Implied dividend growth
rates with longer maturities (5 to 7 months) are half as volatile as the implied dividend growth rates
with shorter maturities (3 and 4 months). Also, implied dividend growth rates with longer maturities
predict future dividend growth rates much better than implied dividend growth rates with shorter
maturities. Accordingly, the corrected dividend-price ratio based on longer-maturity implied dividend
growth rates predicts returns better than the corrected dividend-price ratio based on shorter-maturity
implied dividend growth rates. Since implied dividend growth rates with longer maturities are less
sensitive to the seasonality in dividend payments, the results imply that the documented improvement
in the predictability of returns in the baseline case (based on 6 month implied dividend growth) is
unlikely to be driven by seasonality in dividend payments.
[Insert Table 12 about here]
[Insert Table 13 about here]
Results reported in Table 13 further reveal that the documented predictability is robust to the
alternative way of constructing the realized dividend-price ratio. Specically, comparison of results
reported in Table 12 and Table 13 shows that the implied dividend growth rates in both cases ex-
hibit very similar time-series characteristics. While dividend growth predictive regressions are slightly
aected, return predictive regressions yield virtually the same results.
34
8.2 Alternative ways to estimate implied dividend yield
In the main analysis I estimate implied dividend yield simultaneously from options and futures. I
motivate this approach by the notion that no-arbitrage relations hold better when options and futures
with matching maturities coexist, so any mispricing in options can be easily arbitraged away by
34
Note that here, given that dividends are summed over last 6 months, to better match the data, I consider predicting
semi-annual dividend growth rates as opposed to annual dividend growth rates.
39
trading futures. Combining two no-arbitrage relations has an additional advantage because it enables
estimating implied dividend yield without specifying a proxy for the implied interest rate.
Now I consider alternative ways to estimate implied dividend yield. I estimate implied dividend
yields either from options (based on put-call parity) or from futures (based on cost-of-carry formula).
As a proxy for the interest rate I use either LIBOR (closest to maturity), or the interest rate implied
by the derivative markets (see equation (12) and Figure 3).
35
The results reported in Table 14 demonstrate that it is important to estimate implied dividends
simultaneously from options and futures. When LIBOR is used in conjunction with either options
data or futures data, dividend growth rate predictability results are rather weak, and the corrected
dividend-price ratio does not predict returns better than the realized dividend-price ratio. When I
use implied interest rates rather than the LIBOR, the predictability results improve. In this case, the
implied dividend growth rates based on either options or futures predict dividend growth rates almost
as well as in the baseline case. Also, return predictability improves, and the corrected dividend-price
ratio becomes a signicant predictor of future returns. The ad,. 1
2
in the predictive regression with
monthly returns is 1.67% when corrected dividend-price ratio is based on options, and it is 2.18%
when the correction is based on dividends implied in the futures. Still, this is much lower than the ad,.
1
2
of 4.21% when the corrected dividend-price ratio is based on the implied dividend growth rates
estimated simultaneously from options and futures. Thus, it is indeed advantageous to combine no-
arbitrage relations and extract information about implied dividends from two closely related markets
for S&P 500 derivatives.
[Insert Table 14 about here]
8.3 Synchronicity issues
The main analysis is based on end-of-day futures and option prices. This could be problematic because
spreads tend to widen toward the end of the trading day. Also, options and futures have a dierent
35
When estimating implied dividend yield from options, one can circumvent the need to use a proxy for the interest
rate by combining options with dierent strike prices. In untabulated results I nd the results comparable to the case
when implied dividend yield is estimated from options in combination with the interest rate implied in options and
futures.
40
time stamp from the underlying index. While the stock exchange closes at 15:00 (central time), options
and futures trade until 15:15 (central time).
To address the synchronicity issues, I obtain intra-daily data from Tick Data for a part of the
sample. Specically, I obtain S&P 500 options intra-daily trades and quotes data, S&P 500 futures
intra-daily trades and quotes data, and S&P 500 cash index intra-daily data. While cash index data
are available for the whole time period, options and futures data are available for only a part of the
sample. Options data run from July 2004 onward and the newly released intra-daily futures trades
and quotes data run from January 2010 onward.
In all the estimations I rely on the derivatives quotes data. I rst estimate implied dividend
growth rates from January 2010 through June 2011. I use intra-daily cash index data and the mid-
point between the bid and the ask price for options and futures. I consider only pit trades between
10:00 and 14:00. For consistency with the end-of-day data, I focus on quarterly expirations and use
10 days of data at the end of each month.
I rst make within minute matches between the call and the put prices and retain, within each
minute, the match with the closest time stamp (most of the matches include quotes from the same
second with the average time distance of 0.02 seconds). Then I match option pairs with within-minute
observations for the cash index. Again, within each minute, I retain only the match with the closest
time stamp (here the average time distance is 4.2 seconds). Finally, I match these data with the
futures quotes. Note that unlike options, where the ratio of quotes to trades is some 100:1, the same
ratio for futures is closer to 1:1, with more trades on the short end and more quotes on the long end.
Hence, in comparison to options quotes, futures quotes are relatively scarce. Therefore, I now allow
for wider intervals for futures by making within-10-minute matches. Within every interval I keep the
match with the shortest time dierence. The average time dierence between the futures and the
options observations is 56.2 seconds.
Results are reported in Table 15. I use the full period starting in January 1994, but replace implied
dividend growth rates based on end-of-day data with the implied dividend growth rates based on the
intra-daily data for the period from January 2010 through June 2011. As reported in Column 2, the
intra-daily data seem to provide added information about future returns over and above the end-of-day
41
data. The predictive regression for monthly returns with the corrected dividend-price ratio has an
ad,. 1
2
of 5.26% (compared to 4.21% when only end-of-day data are used). Surprisingly, however,
implied dividend growth rates based on the intra-daily data are slightly more volatile than the implied
dividend growth rates based on the end-of-day data (19.20% versus 18.09%). This increase in the
volatility of the implied dividend growth rates is robust to alternative ways for matching intra-daily
data.
Given that futures quotes are less frequent than options quotes, I next consider calculating implied
dividend growth rates without futures data by combining options intra-daily data with the implied
interest rate. Here I use all the options intra-daily data going back to July 2004. I then replace the
implied dividend growth rates based on the end-of-day data with the implied dividend growth rates
based on the intra-daily data and report results in Column 3. Volatility of the implied dividend growth
rates is now lower, but still at approximately the same level as when implied dividend growth rates
are based solely on the end-of day data (17.97% versus 18.09%). The predictability results remain
strong, but deteriorate slightly, presumably because part of the sample is based on options data only
(see Section 8.2).
All in all, although options and futures markets close 15 minutes after the equity market, this
non-synchronicity does not seem to play an important role. This is in line with Kamara and Miller
(1995), who nd that violations of put-call parity in end-of-day S&P 500 options data do not appear
to be more frequent than in intra-daily data. Also, in the study of dividend strips implied in S&P
500 options, Binsbergen et al. (2012a) report that daily data yields similar results as intra-daily
data, except that the estimates are more volatile. Binsbergen et al. (2012a), however, use only one
day of data at the end of each month, while I use 10 days, which substantially smooths estimates of
the implied dividend growth rates. Indeed, untabulated results show that using 10 days of data as
opposed to one day reduces the standard deviation of the implied dividend growth rates from 40.63%
to 18.09%. This drop in the volatility of implied dividend growth rates goes along with the improved
predictability of dividend growth rates and returns (as further analyzed below).
[Insert Table 15 about here]
42
8.4 Options moneyness and backward-looking data
To mitigate the eects of market frictions, I estimate implied dividend yields from a wide set of options
and futures. At the end of month, I use 10 days of data and options across all moneyness levels.
Use of such a wide set of data is necessary to smooth dividend yield estimates, but it may lead
to inclusion of unreliable data. For example, the wider the moneyness level, the more observations
we have for calculation of the implied dividend yields. Yet deep out-of-the money options are less
liquid and thus less reliable. In the main analysis, I use all options across all moneyness levels. Now,
I consider two lters. Because options are most liquid slightly below the at-the-money strike price,
I employ asymmetric lters with respect to the moneyness level. A broader lter eliminates options
with moneyness levels below 0.6 and above 1.3, and a tighter lter eliminates options with moneyness
levels below 0.8 and above 1.1.
Similarly, one should expect that the most recent data are the most important for forecasting
purposes. However, more data may be needed to smooth the implied dividend yield estimates. In the
baseline case, I use 10 days of data at the end of each month. Now, I consider either 5 or 15 days of
data at the end of each month.
Table 16 reports the results. As expected, ltering out deep out-of-the money options makes the
corrected dividend-price ratio better able to predict monthly returns. For example, the corrected
dividend-price ratio based on options with moneyness levels between 0.8 and 1.1 explains as much
as 5.37% of the variation in the future monthly market returns. Yet this lter reduces the number
of observations for estimation of the implied dividend yield, which makes implied dividend growth
rates more volatile and also less able to predict dividend growth rates at the longer (annual) horizon.
Similar patterns are observed in the second change. Using 15 days as opposed to 10 days of data at
the end of each month further reduces the volatility of the implied dividend growth rates and preserves
the strong predictability results. Using 5 days of backward-looking data makes the implied dividend
growth rates more volatile. This also reduces the ability of the corrected dividend-price ratio to predict
returns, although the estimated parameter on the dj
Ccvv
t
remains signicant.
[Insert Table 16 about here]
43
8.5 Extended sample (January 1990-June 2011)
Estimation of implied dividends requires liquid options and futures with matching expiration times.
While futures and options data are available from January 1990, options and futures in the early years
of the sample are not very liquid. The average number of futures contracts outstanding during the
rst 4 years of the sample is approximately one-third of the same average for the last 4 years of the
sample, and the average number of options contracts at the beginning of the period is more than 11
times smaller than the average at the end of the period. Moreover, settlement procedures for S&P
500 options and futures dier in the early years of the sample. While futures have been settled in the
opening value of the index (A.M. settlement) since June 1987, both P.M.- and A.M.-settled S&P 500
options coexist between June 1987 and December 1993. The P.M. settled options have been the most
liquid, and A.M. settled options were initially barely traded. For these reasons, I have conned the
main analysis to the period from January 1994 through June 2011.
Now I extend the time period back to January 1990 when Market Data Express S&P 500 options
data coverage starts. Given that A.M.-settled options in the initial years of the sample were little
traded, relying exclusively on A.M. options, makes estimation of implied dividend growth rates impos-
sible. Therefore, I use both P.M.- and A.M.- settled options simultaneously, despite the mismatch in
expiration times. Even in this case, liquidity is relatively low, contributing to the volatility of implied
variables in the initial years of the sample. Implied dividend growth rates occasionally take values
above 100% and below 100% (in log terms). Moreover, the implied dividend growth rate cannot be
calculated one time (July 1993), so I replace this value by the observation from the previous month.
Because these uctuations are too large to be reconciled with changes in expectations about dividend
growth rates, I restrict implied dividend growth rates to be between 0.5 and 0.5. The restriction is
binding in 10 of 48 times in years 1990 through 1994.
Table 17 reports the results. Recall that even in the main analysis, implied dividend growth rates
twice took a value below 0.5. Therefore, in Column 1, I rst establish that restricting implied
dividend growth rates to be between 0.5 and 0.5 leaves results largely unaected in the baseline
case. Column 2 then reports results for the extended sample. We see that, despite the problems in
estimating implied dividend growth rates in the early sample, implied dividend growth rates reliably
44
predict future dividend growth rates. While the dividend-price ratio is an insignicant predictor for
both annual dividend growth rates and monthly returns, implied dividend growth rates reliably predict
annual dividend growth rates with t-statistic of 4.38 and an ad,. 1
2
of 10.66%. Also, the corrected
dividend-price ratio reliably predicts monthly returns with an ad,. 1
2
of 3.15% and a Hodrick t-statistic
of 2.96. Untabulated results show that imposing somewhat higher or lower bounds on implied dividend
growth rates does not change the results substantially.
[Insert Table 17 about here]
45
9 Conclusions
It has long been accepted that changes in expected dividend growth rates make the dividend-price
ratio less able to predict returns. Yet empirical analysis of this issue is complicated, because expected
dividend growth rates are an aggregate of investors expectations about future growth opportunities
and are therefore dicult to estimate.
I propose extracting expected dividend growth rates from derivative markets (index options and
futures). Because prices of derivatives depend on, among other factors, the dividends that the underly-
ing asset pays until expiration of the contracts, derivatives provide a unique laboratory for estimating
the dividend growth rates that investors expect to realize in the near future. Indeed, I nd that im-
plied dividend growth rates reliably forecast future dividend growth rates, and the dividend-price ratio
corrected for variation in implied dividend growth rates predicts S&P 500 returns in the period from
January 1994 through June 2011 substantially better than the uncorrected dividend-price ratio. The
forecasting relation between the corrected dividend-price ratio and returns is statistically signicant
and remarkably stable.
Contrary to the standard view that variation in the dividend-price ratio is primarily due to discount
rates, I nd that a substantial amount of price movements is attributable to the time-varying value
of expected dividend growth rates. Further, expected returns and expected dividend growth rate
are highly correlated. This makes the dividend-price ratio very smooth and unable to predict either
dividend growth rates or returns. Once the DP ratio is corrected for variation in expected dividend
growth rates, it becomes more volatile, less persistent, and better able to predict returns.
The results are robust to a number of variations, and are not driven by seasonality of dividend
payments, or dividend risk premium. In line with the argument that any mispricing in options is
easiest to arbitrage away by trading futures, I nd that the results are the strongest when implied
dividend growth rates are estimated simultaneously from options and futures.
46
References
[1] Ait-Sahalia, Y., and A. W. Lo. 1998. Nonparametric estimation of state-price densities implicit in nancial
asset markets. Journal of Finance 53: 499547.
[2] Amihud, Y., and C. M. Hurvich. 2004. Predictive regressions: A reduced-bias estimation method. Journal
of Financial and Quantitative Analysis 39: 813841.
[3] Ang, A., and G. Bekaert. 2007. Stock return predictability: Is it there? Review of Financial Studies 20:
651707.
[4] Bakshi, G., G. Panayotov, and G. Skoulakis. 2011. Improving the predictability of real economic activity
and asset returns with forward variances inferred from option portfolios. Journal of Financial Economics 100:
475495.
[5] Bai, J., and P. Perron. 1998. Estimating and testing linear models with multiple structural changes. Econo-
metrica 66: 4778.
[6] Bansal, R., and A. Yaron. 2004. Risks for the long run: A potential resolution of asset pricing puzzles.
Journal of Finance 59: 14811509.
[7] Binsbergen, J. H. van, M. W. Brandt, and R. S. J. Koijen. 2012a. On the timing and pricing of dividends.
American Economic Review, forthcoming.
[8] Binsbergen, J. H. van, W. Hueskes, R. S. J. Koijen, and E. B. Vrugt. 2012b. Equity yields. Working paper,
Univeristy of Chicago.
[9] Binsbergen, J. H. van, and R. S. J. Koijen. 2010. Predictive regressions: A present-value approach. Journal
of Finance 65: 14391471.
[10] Boguth, O., M. Carlson, A. Fisher, and M. Simutin. 2012. Leverage and the limits of arbitrage pricing:
Implications for dividend strips and the term structure of equity risk premia. Working paper, Arizona State
University.
[11] Bollerslev, T., G. Tauchen, and H. Zhou. 2009. Expected stock returns and variance risk premia. Review
of Financial Studies 22: 44634492.
[12] Boudoukh, J., M. Richardson, and R. F. Whitelaw. 2008. The myth of long-horizon predictability. Review
of Financial Studies 21: 15771605.
[13] Campbell, J. Y., and J. H. Cochrane. 1999. By force of habit: A consumption-based explanation of stock
market behavior. Journal of Political Economy 107: 205251.
[14] Campbell, J. Y., and R. J. Shiller. 1988. The dividend-price ratio and expectations of future dividends and
discount factors. Review of Financial Studies 1: 195227.
[15] Campbell, J. Y., and S. B. Thompson. 2008. Predicting excess stock returns out of sample: Can anything
beat the historical average? Review of Financial Studies 21: 15091531.
[16] Chen, L. 2009. On the reversal of return and dividend growth predictability: A tale of two periods. Journal
of Financial Economics 92: 128151.
47
[17] Chen, L., Z. Da, and X. Zhao. 2012. What drives stock price movements? Working paper, Washington
University.
[18] Cochrane, J. H. 1992. Explaining the variance of price-dividend ratios. Review of Financial Studies 5:
243280.
[19] Cochrane, J. H. 2005. Asset pricing, revised. Princeton, NJ: Princeton University Press.
[20] Cochrane, J. H. 2008a. The dog that did not bark: A defense of return predictability. Review of Financial
Studies 21: 15331575.
[21] Cochrane, J. H. 2008b. State-space vs. VAR models for stock returns. Working paper, Univeristy of Chicago.
[22] Cochrane, J. H. 2011. Presidential address: Discount rates. Journal of Finance 66: 10471108.
[23] Conrad, J., R. F. Dittmar, and E. Ghysels. 2012. Ex ante skewness and expected stock returns. Journal of
Finance, forthcoming.
[24] Cremers, M., and D. Weinbaum. 2010. Deviations from put-call parity and stock return predictability.
Journal of Financial and Quantitative Analysis, 45: 335367.
[25] Fama, E. F., and K. R. French. 1988. Dividend yields and expected stock returns. Journal of Financial
Economics 22: 327.
[26] Favero, C. A., A. Gozluklu., and A. Tamoni. 2011. Demographic trends, the dividend-price ratio and the
predictability of long-run stock market returns. Journal of Financial and Quantitative Analysis 46: 1493
1520.
[27] Ferson, W. E., S. Sarkissian, and T. T. Simin. 2003. Spurious regressions in nancial economics? Journal
of Finance 58: 13931414.
[28] Golez, B., and J. C. Jackwerth. 2012. Pinning in the S&P 500 futures. Journal of Financial Economics,
forthcoming.
[29] Goyal, A., and I. Welch. 2008. A comprehensive look at the empirical performance of equity premium
prediction. Review of Financial Studies 21: 14551508.
[30] Hansen, L. P., and R. J. Hodrick. 1980. Forward exchange rates as optimal predictors of future spot rates:
An econometric analysis. Journal of Political Economy 88: 829853.
[31] Hodrick, R. J. 1992. Dividend yields and expected stock returns: Alternative procedures for inference and
measurement. Review of Financial Studies 5: 357386.
[32] Kamara, A., and T. Miller Jr. 1995. Daily and intradaily tests of European put-call parity. Journal of
Financial and Quantitative Analysis 30: 519539.
[33] Kirby, C. 1997. Measuring the predictable variation in stock and bond returns. Review of Financial Studies
10: 579630.
[34] Koijen, R. S. J., and S. Van Nieuwerburgh. 2011. Predictability of returns and cash ows. Annual Review
of Financial Economics 3: 467491.
48
[35] Lacerda, F., and P. Santa-Clara. 2010. Forecasting dividend growth to better predict returns. Working
paper, Univeristy of Chicago.
[36] Lettau, M., and S. Ludvigson. 2001. Consumption, aggregate wealth, and expected stock returns. Journal
of Finance 56: 815849.
[37] Lettau, M., and S. Ludvigson. 2005. Expected returns and expected dividend growth. Journal of Financial
Economics 76: 583626.
[38] Lettau, M., and S. Van Nieuwerburgh. 2008. Reconciling the return predictability evidence. Review of
Financial Studies 21: 16071652.
[39] Lioui, A. 2006. Black-Scholes-Merton revisited under stochastic dividend yields. Journal of Futures Markets
26: 703732.
[40] Ma, J., and M. E. Wohar. 2011. How much do expected returns and expected dividend growth contribute
to movements in stock returns? Issues of weak identication make existing estimates unreliable. Working
paper, University of Alabama.
[41] McCraken, M. W. 2007. Asymptotics for out of sample tests of Granger causality. Journal of Econometrics
140: 719752.
[42] Menzly, L., T. Santos, and P. Veronesi. 2004. Understanding predictability. Journal of Political Economy
112: 147.
[43] Naranjo, L. 2009. Implied interest rates in a market with frictions. Working paper, New York University.
[44] Newey, W. K., and K. D. West. 1987. A simple positive semi-denite, heteroskedasticity and autocorrelation
consistent covariance matrix. Econometrica 55: 703708.
[45] Pan, J., and A. M. Poteshman. 2006. The information in option volume for future stock prices. Review of
Financial Studies 19: 871908.
[46] Piatti, I. and F. Trojani. 2012. Predictable risks and predictive regression in present-value models. Working
paper, University of Alabama.
[47] Polk, C., S. Thompson, and T. Vuolteenaho. 2006. Cross-sectional forecasts of the equity premium. Journal
of Financial Economics 81: 101141.
[48] Rytchkov, O. 2008. Filtering out expected dividends and expected returns. Working paper, University of
Texas at Austin.
[49] Simin, T. 2008. The (poor) predictive performance of asset pricing models. Journal of Financial and
Quantitative Analysis 43: 355380.
[50] Spiegel, M. 2008. Forecasting the equity premium: Where we stand today. Review of Financial Studies 21:
14531454.
[51] Stambaugh, R. F. 1999. Predictive regressions. Journal of Financial Economics 54: 375421.
[52] Valkanov, R. 2003. Long-horizon regressions: Theoretical results and applications. Journal of Financial
Economics 68: 201232.
[53] Xing, Y., X. Zhang, and R. Zhao. 2010. What does individual option volatility smirk tell us about future
equity returns? Journal of Financial and Quantitative Analysis 45: 641662.
49
Appendix: Derivation of the present value model
Dene log return r
|+1
, log dividend growth rate d
|+1
, and log dividend-price ratio dj
|
as
r
|+1
= log

1
|+1
+ 1
|+1
1
|

, d
|+1
= log

1
|+1
1
|

, dj
|
= log

1
|
1
|

. (25)
Let j
|
= 1
|
(r
|+1
) be conditional expected return and let q
|
= 1
|
(d
|+1
) be conditional expected dividend
growth rate. Assume that j
|
and q
|
follow 1(1) processes:
j
|+1
= c
0
+ c
1
(j
|
) + -

|+1
, (26)
q
|+1
=
0
+
1
(q
|
) + -

|+1
, (27)
d
|+1
= q
|
+ -
J
|+1
. (28)
where -

|+1
, -

|+1
and -
J
|+1
are zero mean errors.
Rewrite log returns as
r
|+1
= dj
|
+ d
|+1
+ log [1 + exp(dj
|+1
)] . (29)
Use rst order Taylor expansion to linearize log [1 + exp(dj
|+1
)] around dj = 1(dj
|
):
log [1 + exp(dj
|+1
)] ' log

1 + exp(dj)

+
exp(dj)
1 + exp(dj)

dj
|+1
+ dj

. (30)
Dene j =
exp(J)
1+exp(J)
and i = log

1 + exp(dj)

+ jdj, such that

log [1 + exp(dj
|+1
)] ' i jdj
|+1
. (31)
Plug (31) into (29) to get an expression for the one-period return:
r
|+1
' i + dj
|
+ d
|+1
jdj
|+1
. (32)
Iterate equation (32) forward:
dj
|
' i + jdj
|+1
+ r
|+1
d
|+1
(33)
' i + j (i + jdj
|+2
+ r
|+2
d
|+2
) + r
|+1
d
|+1
(34)
' i ij + j
2
dj
|+2
+ r
|+1
+ jr
|+2
d
|+1
jd
|+2
(35)
'
i
1 j
+ j
1
dj
|+1
+
1
X
=0
j

(r
|+1+
d
|+1+
). (36)
Assume that lim
!1
j

dj
|+
= 0 to obtain the Campbell and Shiller (1988) approximation for the log
dividend-price ratio (since the relationship holds ex-ante and ex-post, an expectation operator can be added to
the right hand sight).
50
dj
|
'
i
1 j
+ 1
|
1
X
=0
j

(r
|+1+
) 1
|
1
X
=0
j

(d
|+1+
). (37)
Iterate equations (26) and (27) forward:
1
|
(r
|+1+
) = c
0
1 c

1
1 c
1
+ c

1
j
|
(38)
1
|
(d
|+1+
) =
0
1

1
1
1
+

1
q
|
. (39)
Work out the expectations of 1
|
P
1
=0
j

(r
|+1+
) and the 1
|
P
1
=0
j

(d
|+1+
).
1
|
1
X
=0
j

(r
|+1+
) =
1
X
=0
j

(c
0
1 c

1
1 c
1
+ c

1
j
|
) (40)
=
c
0
1 c
1
1
X
=0
j

c
0
1 c
1
1
X
=0
j

1
+ j
|
1
X
=0
j

1
(41)
=
c
0
(1 c
1
)(1 j)

c
0
(1 c
1
)(1 jc
1
)
+ j
|

1
1 jc
1

, (42)
1
|
1
X
=0
j

(d
|+1+
) =
1
X
=0
j

(
0
1

1
1
1
+

1
q
|
) (43)
=

0
1
1
1
X
=0
j

0
1
1
1
X
=0
j

1
+ q
|
1
X
=0
j

1
(44)
=

0
(1
1
)(1 j)


0
(1
1
)(1 j
1
)
+ q
|

1
1 j
1

. (45)
Finally, insert (42) and (45) in the Campbell and Shiller (1988) present value identity to nd the dividend-
price ratio:
dj
|
' , + j
|

1
1 jc
1

q
|

1
1 j
1

, (46)
where , =
r
1
+
o0
(1o1)(1)

o0
(1o1)(1o1)

~
0
(1~
1
)(1)
+
~
0
(1~
1
)(1~
1
)
.
51
Table 1: Seasonality of realized dividends
This table reports summary statistics for annualized realized S&P 500 dividends aggregated over either 1 month Div 1M,
3 months Div 3M, 6 months Div 6M, or 12 months Div 12M. The period is from January 1994 through June 2011 and
the frequency is monthly.
1i 1' 1i 3' 1i 6' 1i 12'
Mean 19.2610 19.2347 19.2342 19.2465
Min 6.7289 12.0813 12.8994 13.1741
Max 41.7732 30.8628 29.7071 29.0362
Stdev 7.3644 4.7743 4.6980 4.6094
52
Table 2: Liquidity of S&P 500 derivatives
This table reports average daily open interest and volume by maturity for S&P 500 options and S&P 500 futures. Open
interest and volume are expressed either in number of contracts or in notional value. Notional value for open interest
(volume) is calculated as the open interest (volume) times the spot price of the index times the multiplier. Options
multiplier is 100 and futures multiplier is 250. The period is from January 1994 through June 2011.
Maturity in months (M) M<3 3<M<6 6<M<9
Panel A: Options
Volume in number of contracts 84,561 14,418 7,338
Open interest in number of contracts 1,423,791 595,876 363,913
Volume in notional value 9,785,820,724 1,659,107,660 846,344,457
Open interest in notional value 162,676,482,320 67,976,582,059 42,304,187,941
Panel B: Futures
Volume in number of contracts 54,497 2,390 142
Open interest in number of contracts 414,146 21,353 3,132
Volume in notional value 14,037,995,757 662,610,862 39,141,370
Open interest in notional value 116,944,132,142 5,930,752,412 839,421,108
53
Table 3: Summary statistics
This table reports summary statistics (Panel A) and unconditional correlations (Panel B) for the S&P 500 monthly
returns r
1
|
, the monthly dividend growth rates d
1
|
, the dividend-price ratio dj
|
, the implied dividend growth rates
idq
|
, and the corrected dividend-price ratio dj
co::
|
. All variables are expressed in logs. The corrected dividend-price
ratio is dened as dj
co::
|
= dj
|
+2.10 idq
|
. Sample means and standard deviations (Stdev) for monthly returns and
monthly realized dividend growth rates are multiplied by 12 (
p
12). Augmented Dickey-Fuller Test (ADF Test) tests
the null hypothesis of a unit root. The critical values corresponding to p-values of 0.01, 0.05, and 0.10 are -3.99, -3.42,
-3.14, respectively. The period is from January 1994 through June 2011 and the frequency is monthly.
r
1
|
d
1
|
dj
|
idq
|
dj
co::
|
Panel A: Summary statistics
Mean 0.0784 0.0378 -4.0150 0.0731 -3.8614
Stdev 0.1570 0.0415 0.2580 0.1809 0.5212
AR (1) 0.1107 0.1066 0.9780 0.5335 0.7357
ADF Test -10.0492 -8.0654 -2.1339 -5.5926 -3.9477
Panel B: Unconditional correlations
r
1
|
1.0000 -0.0353 -0.0774 0.0706 0.0132
d
1
|
1.0000 -0.0867 0.1746 0.0844
dj
|
1.0000 0.3090 0.7204
idq
|
1.0000 0.8822
dj
co::
|
1.0000
54
Table 4: Dividend growth rate regressions
This table reports regression results for predicting S&P 500 dividend growth rates. Predictor variables include the
dividend-price ratio dj
|
, the implied dividend growth rates idq
|
, and the annual realized dividend growth rates d
.
|
.
All variables are expressed in logs. Dividend growth rates horizon / is in months. Hodrick (1992) t-statistics accounting
for the overlap are reported in parentheses. All regressions include a constant term. The period is from January 1994
through June 2011 and the frequency is monthly.
Dividend growth horizon / 1 3 6 12
N (209) (207) (204) (198)
dj
|
-0.0067 -0.0193 -0.0389 -0.0680
(-1.88) (-1.74) (-1.69) (-1.54)
ad,. 1
2
0.0159 0.0413 0.0491 0.0439
idq
|
0.0155 0.0492 0.1033 0.1952
(3.70) (4.38) (4.98) (4.81)
ad,. 1
2
0.0499 0.1422 0.1805 0.1905
dj
|
-0.0111 -0.0331 -0.0680 -0.1214
(-2.69) (-2.58) (-2.54) (-2.40)
idq
|
0.0203 0.0639 0.1337 0.2490
(4.04) (4.54) (4.91) (4.84)
ad,. 1
2
0.0973 0.2621 0.3262 0.3286
d
.
|
0.3807
(3.13)
ad,. 1
2
0.1322
idq
|
0.1667
(4.48)
d
.
|
0.2926
(2.50)
ad,. 1
2
0.2637
55
Table 5: Return regressions
This table reports regression results for predicting S&P 500 returns. Predictor variables include the dividend-price ratio
dj
|
, the implied dividend growth rates idq
|
, the corrected dividend-price ratio dj
co::
|
, and the annual realized dividend
growth rates d
.
|
. All variables are expressed in logs. Return horizon / is in months. Hodrick (1992) t-statistics
accounting for the overlap are reported in parentheses. All regressions include a constant term. The period is from
January 1994 through June 2011 and the frequency is monthly.
Return horizon / 1 3 6 12
N (209) (207) (204) (198)
dj
|
0.0168 0.0594 0.1408 0.3158
(1.15) (1.41) (1.78) (2.21)
ad,. 1
2
0.0044 0.0292 0.0767 0.1762
dj
|
0.0047 0.0395 0.1164 0.2766
(0.29) (0.85) (1.36) (1.88)
idq
|
0.0558 0.0915 0.1120 0.1828
(2.55) (1.76) (1.27) (1.24)
ad,. 1
2
0.0449 0.0604 0.0949 0.1987
dj
co::
|
0.0188 0.0501 0.0891 0.2071
(3.10) (2.44) (2.42) (2.55)
ad,. 1
2
0.0421 0.0636 0.0955 0.1904
dj
|
0.0169 0.0592 0.1402 0.3151
(1.15) (1.40) (1.75) (2.18)
d
.
|
0.0029 -0.0079 -0.0302 -0.0484
(0.07) (-0.07) (-0.12) (-0.11)
ad,. 1
2
-0.0004 0.0245 0.0725 0.1724
dj
|
0.0036 0.0372 0.1127 0.2703
(0.21) (0.78) (1.28) (1.79)
idq
|
0.0588 0.0976 0.1216 0.2014
(2.52) (1.77) (1.29) (1.35)
d
.
|
-0.0272 -0.0569 -0.0907 -0.1619
(-0.65) (-0.45) (-0.35) (-0.37)
ad,. 1
2
0.0424 0.0586 0.0934 0.1986
56
Table 6: Out-of-sample predictability
This table reports out-of-sample 1
2
OS
for predicting monthly S&P 500 returns r
1
|+1
. Predictor variables include the
dividend-price ratio dj
|
, the corrected dividend-price ratio dj
co::
|
, and the corrected dividend-price ratio adjusted for
the look-ahead bias dj
co::J.1
|
. All variables are expressed in logs. The 1
2
OS
is calculated over the period January
2000 through June 2011 (the period from January 1994 through December 1999 is used to make the rst forecast).
Statistical signicance is measured according to McCraken (2007). One, two, and three asterisks (*,**,***) denote
signicance at the 10%, 5%, and 1% levels, respectively.
Dependent variable: r
1
|+1
1
2
OS
dj
|
0.0004
dj
co::
|
0.0533***
dj
co::J.1
|
0.0535***
Correlation(dj
co::
|
, dj
co::J.1
|
)=0.9988
57
T
a
b
l
e
7
:
A
l
t
e
r
n
a
t
i
v
e
p
r
e
d
i
c
t
o
r
s
:
S
u
m
m
a
r
y
s
t
a
t
i
s
t
i
c
s
T
h
i
s
t
a
b
l
e
r
e
p
o
r
t
s
s
u
m
m
a
r
y
s
t
a
t
i
s
t
i
c
s
(
P
a
n
e
l
A
)
a
n
d
u
n
c
o
n
d
i
t
i
o
n
a
l
c
o
r
r
e
l
a
t
i
o
n
s
(
P
a
n
e
l
B
)
f
o
r
t
h
e
d
i
v
i
d
e
n
d
-
p
r
i
c
e
r
a
t
i
o
d
j
|
,
t
h
e
d
i
v
i
d
e
n
d
-
p
r
i
c
e
r
a
t
i
o
c
o
r
r
e
c
t
e
d
f
o
r
t
h
e
i
m
p
l
i
e
d
d
i
v
i
d
e
n
d
g
r
o
w
t
h
r
a
t
e
s
d
j
c
o
:
:
|
,
t
h
e
d
i
v
i
d
e
n
d
-
p
r
i
c
e
r
a
t
i
o
c
o
r
r
e
c
t
e
d
f
o
r
t
h
e
a
v
e
r
a
g
e
r
e
a
l
i
z
e
d
d
i
v
i
d
e
n
d
g
r
o
w
t
h
r
a
t
e
s
d
j
1
I
s
|
|
,
t
h
e
e
a
r
n
i
n
g
s
-
p
r
i
c
e
r
a
t
i
o
c
j
|
,
t
h
e
c
o
n
s
u
m
p
t
i
o
n
-
t
o
-
w
e
a
l
t
h
r
a
t
i
o
c
a
j
|
(
m
o
n
t
h
l
y
o
b
s
e
r
v
a
t
i
o
n
s
o
f
c
a
y
a
r
e
d
e

n
e
d
b
y
t
h
e
m
o
s
t
r
e
c
e
n
t
l
y
a
v
a
i
l
a
b
l
e
q
u
a
r
t
e
r
l
y
o
b
s
e
r
v
a
t
i
o
n
)
,
t
h
e
d
e
f
a
u
l
t
s
p
r
e
a
d
d
c
)
|
,
t
h
e
t
e
r
m
s
p
r
e
a
d
t
c
r
:
|
,
t
h
e
v
a
r
i
a
n
c
e
r
i
s
k
p
r
e
m
i
u
m
b
a
s
e
d
o
n
t
h
e
r
e
a
l
i
z
e
d
v
a
r
i
a
n
c
e

r
j
1
\
1
|
,
t
h
e
v
a
r
i
a
n
c
e
r
i
s
k
p
r
e
m
i
u
m
b
a
s
e
d
o
n
t
h
e
f
o
r
e
c
a
s
t
e
d
v
a
r
i
a
n
c
e

r
j
J
1
\
1
|
,
t
h
e
d
i

e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
o
p
e
n
-
b
u
y
v
o
l
u
m
e
f
o
r
c
a
l
l
s
a
n
d
p
u
t
s
i
n
t
h
o
u
s
a
n
d
s
o
f
c
o
n
t
r
a
c
t
s
o
/
|
,
t
h
e
s
k
e
w
n
e
s
s
i
m
p
l
i
e
d
i
n
t
h
e
S
&
P
5
0
0
o
p
t
i
o
n
s
:
/
c
n
|
,
a
n
d
t
h
e
s
p
r
e
a
d
b
e
t
w
e
e
n
t
h
e
s
i
x
m
o
n
t
h
L
i
b
o
r
r
a
t
e
a
n
d
t
h
e
s
i
x
m
o
n
t
h
i
m
p
l
i
e
d
i
n
t
e
r
e
s
t
r
a
t
e
:
j
r
J
1
1
|
.
A
u
g
m
e
n
t
e
d
D
i
c
k
e
y
-
F
u
l
l
e
r
T
e
s
t
(
A
D
F
T
e
s
t
)
t
e
s
t
s
t
h
e
n
u
l
l
h
y
p
o
t
h
e
s
i
s
o
f
a
u
n
i
t
r
o
o
t
.
T
h
e
c
r
i
t
i
c
a
l
v
a
l
u
e
s
c
o
r
r
e
s
p
o
n
d
i
n
g
t
o
p
-
v
a
l
u
e
s
o
f
0
.
0
1
,
0
.
0
5
,
a
n
d
0
.
1
0
a
r
e
-
3
.
9
9
,
-
3
.
4
2
,
-
3
.
1
4
,
r
e
s
p
e
c
t
i
v
e
l
y
.
T
h
e
p
e
r
i
o
d
i
s
f
r
o
m
J
a
n
u
a
r
y
1
9
9
4
t
h
r
o
u
g
h
J
u
n
e
2
0
1
1
a
n
d
t
h
e
f
r
e
q
u
e
n
c
y
i
s
m
o
n
t
h
l
y
.
d
j
|
d
j
c
o
:
:
|
d
j
1
I
s
|
|
c
j
|
c
a
j
|
d
c
)
|
t
c
r
:
|

r
j
1
\
1
|

r
j
J
1
\
1
|
o
/
|
:
/
c
n
|
:
j
r
J
1
1
|
P
a
n
e
l
A
:
S
u
m
m
a
r
y
s
t
a
t
i
s
t
i
c
s
M
e
a
n
-
4
.
0
1
5
0
-
3
.
8
6
1
4
-
3
.
9
4
2
1
-
3
.
2
0
3
2
-
0
.
0
0
2
3
0
.
9
6
1
5
1
.
6
3
0
7
1
8
.
5
2
1
3
1
9
.
2
2
9
0
-
5
1
.
0
2
2
1
1
1
6
.
2
5
0
9
-
0
.
0
2
8
6
S
t
d
e
v
0
.
2
5
8
0
0
.
5
2
1
2
0
.
2
9
8
8
0
.
4
2
0
6
0
.
0
2
3
0
0
.
4
7
6
1
1
.
2
4
3
4
2
1
.
7
5
1
4
2
3
.
5
6
3
7
7
6
.
4
6
5
9
4
.
9
2
6
9
0
.
2
8
6
4
A
R
(
1
)
0
.
9
7
8
0
0
.
7
3
5
7
0
.
9
7
3
7
0
.
9
3
7
6
0
.
9
7
6
1
0
.
9
5
7
5
0
.
9
6
3
4
0
.
2
8
7
6
0
.
2
6
9
8
0
.
1
7
4
7
0
.
6
2
5
8
0
.
5
4
7
6
A
D
F
T
e
s
t
-
2
.
1
3
3
9
-
3
.
9
4
7
7
-
1
.
8
3
5
7
-
2
.
6
4
0
9
-
3
.
2
9
6
2
-
3
.
5
9
9
1
-
2
.
4
2
7
9
-
7
.
2
7
0
6
-
1
1
.
1
3
6
7
-
8
.
3
6
4
7
-
5
.
3
9
1
6
-
5
.
6
5
0
0
P
a
n
e
l
B
:
U
n
c
o
n
d
i
t
i
o
n
a
l
c
o
r
r
e
l
a
t
i
o
n
s
d
j
|
1
.
0
0
0
0
0
.
7
2
0
4
0
.
8
3
0
6
0
.
1
4
6
0
0
.
1
5
2
2
0
.
4
0
5
2
0
.
3
7
0
0
-
0
.
0
8
4
2
0
.
0
5
0
6
-
0
.
1
2
0
2
-
0
.
0
7
8
7
0
.
1
9
5
1
d
j
c
o
:
:
|
1
.
0
0
0
0
0
.
6
8
9
8
0
.
3
3
0
9
0
.
1
3
7
2
-
0
.
0
3
2
2
0
.
1
8
5
5
-
0
.
1
0
6
5
-
0
.
0
4
3
3
-
0
.
1
5
9
6
-
0
.
0
4
3
0
-
0
.
2
6
5
8
d
j
1
I
s
|
|
1
.
0
0
0
0
0
.
4
5
2
4
0
.
1
1
5
9
0
.
2
3
6
8
0
.
0
8
9
5
-
0
.
2
3
4
6
-
0
.
0
6
3
9
-
0
.
1
9
3
1
-
0
.
1
2
0
1
0
.
1
0
2
6
c
j
|
1
.
0
0
0
0
0
.
0
8
0
2
-
0
.
2
4
4
4
-
0
.
3
2
9
8
-
0
.
2
9
0
9
-
0
.
1
4
3
7
-
0
.
1
7
4
9
-
0
.
0
4
6
9
-
0
.
2
2
7
0
c
a
j
|
1
.
0
0
0
0
-
0
.
2
8
4
9
-
0
.
0
9
2
1
0
.
0
9
6
2
0
.
0
2
8
0
0
.
4
1
1
7
-
0
.
3
8
9
1
-
0
.
3
6
6
6
d
c
)
|
1
.
0
0
0
0
0
.
3
6
7
0
-
0
.
0
3
0
3
0
.
2
6
8
4
-
0
.
0
3
5
3
-
0
.
0
0
9
2
0
.
5
2
8
8
t
c
r
:
|
1
.
0
0
0
0
-
0
.
0
0
2
6
0
.
0
8
8
2
0
.
1
5
4
6
0
.
0
3
1
8
0
.
3
4
6
3

r
j
1
\
1
|
1
.
0
0
0
0
0
.
0
7
2
1
0
.
1
2
2
5
-
0
.
0
4
3
1
-
0
.
0
3
2
0

r
j
J
1
\
1
|
1
.
0
0
0
0
-
0
.
0
2
2
7
-
0
.
0
2
9
5
-
0
.
0
2
1
3
o
/
|
1
.
0
0
0
0
-
0
.
2
5
3
7
-
0
.
0
4
2
0
:
/
c
n
|
1
.
0
0
0
0
0
.
1
2
6
0
:
j
r
J
1
1
|
1
.
0
0
0
0
58
T
a
b
l
e
8
:
A
l
t
e
r
n
a
t
i
v
e
p
r
e
d
i
c
t
o
r
s
:
M
o
n
t
h
l
y
r
e
t
u
r
n
r
e
g
r
e
s
s
i
o
n
s
T
h
i
s
t
a
b
l
e
r
e
p
o
r
t
s
i
n
-
s
a
m
p
l
e
a
n
d
o
u
t
-
o
f
-
s
a
m
p
l
e
r
e
s
u
l
t
s
f
o
r
p
r
e
d
i
c
t
i
n
g
m
o
n
t
h
l
y
S
&
P
5
0
0
r
e
t
u
r
n
s
.
I
n
-
s
a
m
p
l
e
r
e
s
u
l
t
s
a
r
e
b
a
s
e
d
o
n
t
h
e
p
e
r
i
o
d
f
r
o
m
J
a
n
u
a
r
y
1
9
9
4
t
h
r
o
u
g
h
J
u
n
e
2
0
1
1
.
H
o
d
r
i
c
k
(
1
9
9
2
)
t
-
s
t
a
t
i
s
t
i
c
s
a
r
e
r
e
p
o
r
t
e
d
i
n
p
a
r
e
n
t
h
e
s
e
s
.
O
u
t
-
o
f
-
s
a
m
p
l
e
1
2 O
S
i
s
c
a
l
c
u
l
a
t
e
d
f
o
r
t
h
e
p
e
r
i
o
d
f
r
o
m
J
a
n
u
a
r
y
2
0
0
0
t
h
r
o
u
g
h
J
u
n
e
2
0
1
1
(
t
h
e
p
e
r
i
o
d
f
r
o
m
J
a
n
u
a
r
y
1
9
9
4
t
h
r
o
u
g
h
D
e
c
e
m
b
e
r
1
9
9
9
i
s
u
s
e
d
t
o
m
a
k
e
t
h
e

r
s
t
f
o
r
e
c
a
s
t
)
.
A
l
l
v
a
r
i
a
b
l
e
d
e

n
i
t
i
o
n
s
a
r
e
i
d
e
n
t
i
c
a
l
t
o
T
a
b
l
e
7
.
d
j
|
0
.
0
1
6
8
(
1
.
1
5
)
d
j
c
o
:
:
|
0
.
0
1
8
8
0
.
0
2
1
1
0
.
0
1
8
5
0
.
0
2
0
6
0
.
0
1
6
4
(
3
.
1
0
)
(
3
.
3
7
)
(
3
.
0
3
)
(
3
.
4
5
)
(
2
.
4
8
)
d
j
1
I
s
|
|
0
.
0
1
2
5
(
1
.
0
2
)
c
j
|
-
0
.
0
0
0
8
(
-
0
.
1
2
)
c
a
j
|
0
.
1
7
1
9
(
1
.
6
0
)
d
c
)
|
-
0
.
0
1
1
5
-
0
.
0
1
0
8
(
-
1
.
0
8
)
(
-
1
.
0
2
)
t
c
r
:
|
-
0
.
0
0
2
5
-
0
.
0
0
4
1
(
-
1
.
0
3
)
(
-
1
.
7
1
)

r
j
1
\
1
|
0
.
0
0
0
5
0
.
0
0
0
5
(
2
.
1
3
)
(
2
.
3
3
)

r
j
J
1
\
1
|
-
0
.
0
0
0
1
(
-
0
.
2
6
)
o
/
|
0
.
0
0
0
0
(
0
.
2
3
)
:
/
c
n
|
-
0
.
0
0
0
2
(
-
0
.
3
5
)
:
j
r
J
1
1
|
-
0
.
0
2
4
3
-
0
.
0
1
6
4
(
-
1
.
8
0
)
(
-
1
.
1
3
)
a
d
,
.
1
2
0
.
0
0
4
4
0
.
0
4
2
1
0
.
0
0
2
0
-
0
.
0
0
4
8
0
.
0
0
2
8
0
.
0
0
9
8
-
0
.
0
0
0
1
0
.
0
4
6
2
-
0
.
0
0
3
8
-
0
.
0
0
4
7
-
0
.
0
0
4
3
0
.
0
1
8
9
0
.
1
0
0
3
0
.
0
5
0
5
0
.
0
4
9
7
0
.
0
4
7
5
1
2 O
S
0
.
0
0
0
4
0
.
0
5
3
3
-
0
.
0
0
2
7
-
0
.
0
1
6
0
-
0
.
0
0
2
6
-
0
.
0
3
7
4
-
0
.
0
1
5
1
0
.
0
5
1
7
-
0
.
0
4
1
1
-
0
.
0
0
6
9
-
0
.
0
0
5
1
0
.
0
0
4
7
0
.
1
1
3
7
0
.
0
0
4
6
0
.
0
4
3
4
0
.
0
3
9
7
59
T
a
b
l
e
9
:
S
u
b
s
a
m
p
l
e
a
n
a
l
y
s
i
s
:
M
o
n
t
h
l
y
r
e
t
u
r
n
r
e
g
r
e
s
s
i
o
n
s
T
h
i
s
t
a
b
l
e
r
e
p
o
r
t
s
i
n
-
s
a
m
p
l
e
a
n
d
o
u
t
-
o
f
-
s
a
m
p
l
e
r
e
s
u
l
t
s
f
o
r
p
r
e
d
i
c
t
i
n
g
m
o
n
t
h
l
y
S
&
P
5
0
0
r
e
t
u
r
n
s
.
I
n
-
s
a
m
p
l
e
r
e
s
u
l
t
s
a
r
e
b
a
s
e
d
o
n
t
h
e
p
e
r
i
o
d
f
r
o
m
J
a
n
u
a
r
y
1
9
9
4
t
h
r
o
u
g
h
D
e
c
e
m
b
e
r
2
0
0
7
.
H
o
d
r
i
c
k
(
1
9
9
2
)
t
-
s
t
a
t
i
s
t
i
c
s
a
r
e
r
e
p
o
r
t
e
d
i
n
p
a
r
e
n
t
h
e
s
e
s
.
O
u
t
-
o
f
-
s
a
m
p
l
e
1
2 O
S
i
s
c
a
l
c
u
l
a
t
e
d
f
o
r
t
h
e
p
e
r
i
o
d
f
r
o
m
J
a
n
u
a
r
y
2
0
0
0
t
h
r
o
u
g
h
D
e
c
e
m
b
e
r
2
0
0
7
(
t
h
e
p
e
r
i
o
d
f
r
o
m
J
a
n
u
a
r
y
1
9
9
4
t
h
r
o
u
g
h
D
e
c
e
m
b
e
r
1
9
9
9
i
s
u
s
e
d
t
o
m
a
k
e
t
h
e

r
s
t
f
o
r
e
c
a
s
t
)
.
A
l
l
v
a
r
i
a
b
l
e
d
e

n
i
t
i
o
n
s
a
r
e
i
d
e
n
t
i
c
a
l
t
o
T
a
b
l
e
7
.
d
j
|
0
.
0
3
1
1
(
2
.
4
5
)
d
j
c
o
:
:
|
0
.
0
1
8
9
0
.
0
2
1
6
0
.
0
1
8
2
0
.
0
2
0
0
0
.
0
1
6
8
(
3
.
1
5
)
(
3
.
4
0
)
(
2
.
7
6
)
(
3
.
3
7
)
(
2
.
5
5
)
d
j
1
I
s
|
|
0
.
0
3
0
3
(
2
.
6
1
)
c
j
|
0
.
0
2
8
1
(
2
.
2
7
)
c
a
j
|
0
.
2
1
4
3
(
2
.
1
0
)
d
c
)
|
-
0
.
0
2
0
7
-
0
.
0
0
7
0
(
-
1
.
1
0
)
(
-
0
.
3
4
)
t
c
r
:
|
-
0
.
0
0
2
8
-
0
.
0
0
4
1
(
-
1
.
0
7
)
(
-
1
.
5
9
)

r
j
1
\
1
|
0
.
0
0
0
3
0
.
0
0
0
5
(
1
.
3
6
)
(
1
.
8
8
)

r
j
J
1
\
1
|
0
.
0
0
0
3
(
1
.
2
5
)
o
/
|
0
.
0
0
0
0
(
0
.
1
1
)
:
/
c
n
|
-
0
.
0
0
0
4
(
-
0
.
5
2
)
:
j
r
J
1
1
|
-
0
.
0
2
5
4
-
0
.
0
1
1
5
(
-
2
.
1
8
)
(
-
0
.
9
0
)
a
d
,
.
1
2
0
.
0
2
8
3
0
.
0
6
4
9
0
.
0
4
0
2
0
.
0
2
7
3
0
.
0
0
7
4
0
.
0
0
4
9
0
.
0
0
0
5
0
.
0
1
1
5
0
.
0
0
8
4
-
0
.
0
0
6
0
-
0
.
0
0
4
5
0
.
0
2
1
8
0
.
0
9
5
0
0
.
0
6
0
4
0
.
0
7
3
0
0
.
0
6
4
1
1
2 O
S
0
.
0
4
8
6
0
.
1
0
0
8
0
.
0
6
4
5
0
.
0
3
9
7
0
.
0
0
2
0
-
0
.
0
1
3
9
-
0
.
0
2
3
7
0
.
0
0
4
5
0
.
0
0
5
6
-
0
.
0
1
1
8
-
0
.
0
0
3
5
0
.
0
1
0
6
0
.
1
2
6
4
0
.
0
4
9
1
0
.
0
7
9
1
0
.
0
7
6
3
60
Table 10: Present value parameter estimates
This table reports the present value parameter estimates. Delta one c
1
is AR(1) coecient for expected returns j
|
and
gama one
1
is AR(1) coecient for expected dividend growth rates q
|
. Expected dividend growth rates are measured
by the realized annual dividend growth rates (Models 1 through 3), the implied dividend growth rates (Model 4), the
implied dividend growth rates cleaned of variation due to demand imbalances and market frictions in the options markets
(Model 5), the implied dividend growth rates cleaned of variation due to risk measures (Model 6), and the annual trend
component of implied dividend growth rates according to Hodrick-Prescott lter (Model 7). Table reports the mean and
the interquartile range (in square brackets) for the parameters estimated on 12 successive annual samples. The period
is from January 1926 through June 2011 (Model 1), from January 1946 through June 2011 (Model 2), or from January
1994 through June 2011 (Models 3 through 7).
(1) (2) (3) (4) (5) (6) (7)
otdc(j
|
) 0.0320 0.0505 0.0767 0.0934 0.0963 0.0908 0.0994
[0.01] [0.01] [0.02] [0.03] [0.03] [0.03] [0.01]
otdc(q
|
) 0.1084 0.0643 0.0813 0.1809 0.1788 0.1686 0.1307
[0.01] [0.00] [0.00] [0.05] [0.04] [0.05] [0.01]
c
1
0.9661 0.9221 0.7487 0.7839 0.7730 0.7890 0.8092
[0.03] [0.01] [0.07] [0.08] [0.08] [0.09] [0.02]

1
0.3410 0.3662 0.3160 0.2160 0.2069 0.1419 0.5706
[0.12] [0.03] [0.15] [0.31] [0.33] [0.45] [0.04]
Decomposition of the dividend-price ratio
Discount rates 0.9827 1.0687 0.9783 1.3425 1.3156 1.4465 1.5022
[0.01] [ 0.01] [0.02] [0.41] [0.40] [0.50] [0.14]
Dividend growth 0.0173 -0.0687 0.0217 -0.3425 -0.3156 -0.4465 -0.5022
[0.01] [0.01] [0.02] [0.41] [0.40] [0.50] [0.14]
Correlation between expected returns and expected dividend growth
j

t
t
0.2975 0.4845 0.3681 0.7362 0.7259 0.7498 0.8638
[0.02] [0.02] [0.05] [0.22] [0.21] [0.27] [0.04]
61
Table 11: Announced dividends
This table reports regression results for predicting S&P 500 annual dividend growth rates d
.
|+12
and monthly returns
r
1
|+1
. Predictor variables include the dividend-price ratio dj
|
and the indicated dividend growth rates idq
|
. All variables
are expressed in logs. Hodrick (1992) t-statistics accounting for the overlap are reported in parentheses. The period is
from February 1995 through June 2011 and the frequency is monthly.
Dependent variable: d
.
|+12
d
.
|+12
r
1
|+1
r
1
|+1
dj
|
-0.0914 0.0204 0.0211
(-1.68) (1.15) (1.23)
dq
11
|
1.3628 0.0100
(4.63) (0.14)
ad,. 1
2
0.0662 0.7250 0.0062 0.0009
62
Table 12: Robustness check: Seasonality of dividends and maturity of derivatives I
This table reports summary statistics for the implied dividend growth rates (Panel A), regression results for predicting
annual S&P 500 dividend growth rates (Panel B), regression results for predicting S&P 500 monthly returns (Panel
C), and the dividend-price ratio variance decomposition results (Panel D). Predictive regressions are based on monthly
frequency. Hodrick (1992) t-statistics are reported in parentheses. Panel D reports the mean and the interquartile range
(in square brackets) for the parameters estimated on 12 successive annual samples. Implied dividend growth rates are
dened as: idq
|
(T) = idj
T
|
dj
|
, where idj
T
|
is annualized log implied dividend yield with maturity T and dj
|
is
the standard dividend-price ratio dened as the 12-month trailing sum of dividends divided by the current price. T is
either 3 months (Model 1), 4 months (Model 2), 5 months (Model 3), 6 months (Model 4), or 7 months (Model 5). The
corrected dividend-price ratio is dened as dj
co::
|
(T) = dj
|
+2.10idq
|
(T). The period is from January 1994 through
June 2011.
(1) (2) (3) (4) (5)
Panel A: Summary statistics for the idq
|
Mean 0.0680 0.0538 0.0710 0.0731 0.0677
Stdev 0.4875 0.4145 0.2145 0.1809 0.1773
AR(1) 0.2349 0.1798 0.4333 0.5335 0.4690
ADF Test -8.8820 -8.5192 -6.2074 -5.5926 -6.0375
Panel B: Predicting annual dividend growth rates (N=198)
idq
|
0.0347 0.0465 0.1442 0.1952 0.2001
(3.16) (3.20) (4.52) (4.81) (4.92)
ad,. 1
2
0.0401 0.0538 0.1457 0.1905 0.1922
Panel C: Predicting monthly returns (N=209)
dj
|
dj
co::
|
0.0069 0.0070 0.0158 0.0188 0.0197
(2.23) (1.91) (2.90) (3.10) (3.05)
ad,. 1
2
0.0247 0.0181 0.0370 0.0421 0.0436
Panel D: Decomposition of the dividend-price ratio
Discount rates 1.5924 1.5430 1.4126 1.3425 1.2557
[0.79] [0.63] [0.57] [0.41] [0.37]
Dividend growth -0.5924 -0.5430 -0.4126 -0.3425 -0.2557
[0.79] [0.63] [0.57] [0.41] [0.37]
Correlation between expected returns and expected dividend growth
j

t
t
0.8315 0.8057 0.7742 0.7362 0.7088
[0.20] [0.23] [0.24] [0.22] [0.23]
63
Table 13: Robustness check: Seasonality of dividends and maturity of derivatives II
This table reports summary statistics for the implied dividend growth rates (Panel A), regression results for predicting
semi-annual S&P 500 dividend growth rates (Panel B), regression results for predicting S&P 500 monthly returns (Panel
C), and the dividend-price ratio variance decomposition results (Panel D). Predictive regressions are based on monthly
frequency. Hodrick (1992) t-statistics are reported in parentheses. Panel D reports the mean and the interquartile range
(in square brackets) for the parameters estimated on 12 successive annual samples. Implied dividend growth rates are
dened as: idq
|
(T) = idj
T
|
dj
6n
|
, where idj
T
|
is annualized log implied dividend yield with maturity T and dj
6n
|
is annualized dividend-price ratio dened as 2 times the 6-month trailing sum of dividends divided by the current price.
T is either 3 months (Model 1), 4 months (Model 2), 5 months (Model 3), 6 months (Model 4), or 7 months (Model 5).
The corrected dividend-price ratio is dened as dj
co::
|
(T) = dj
6n
|
+2.10 idq
|
(T). The period is from January 1994
through June 2011.
(1) (2) (3) (4) (5)
Panel A: Summary statistics for the idq
|
Mean 0.0593 0.0451 0.0623 0.0644 0.0590
Stdev 0.4864 0.4136 0.2122 0.1782 0.1753
AR(1) 0.2250 0.1683 0.4026 0.4912 0.4213
ADF Test -9.0326 -8.6749 -6.5518 -6.0000 -6.4688
Panel B: Predicting semi-annual dividend growth rates (N=204)
idq
|
0.0269 0.0378 0.1164 0.1586 0.1655
(2.42) (2.71) (3.44) (3.72) (3.98)
ad,. 1
2
0.0379 0.0563 0.1478 0.1947 0.2053
Panel C: Predicting monthly returns (N=209)
dj
6n
|
dj
co::
|
0.0069 0.0070 0.0158 0.0188 0.0197
(2.23) (1.91) (2.90) (3.09) (3.04)
ad,. 1
2
0.0247 0.0181 0.0371 0.0422 0.0436
Panel D: Decomposition of the dividend-price ratio
Discount rates 1.6359 1.5861 1.4516 1.3735 1.2790
[0.87] [0.64] [0.62] [0.48] [0.36]
Dividend growth -0.6359 -0.5861 -0.4516 -0.3735 -0.2790
[0.87] [0.64] [0.62] [0.48] [0.36]
Correlation between expected returns and expected dividend growth
j

t
t
0.8309 0.8094 0.7856 0.7492 0.7183
[0.21] [0.24] [0.25] [0.23] [0.21]
64
Table 14: Robustness check: Alternative ways to estimate implied dividend yield
This table reports summary statistics for the implied dividend growth rates (Panel A), regression results for predicting
annual S&P 500 dividend growth rates (Panel B), regression results for predicting S&P 500 monthly returns (Panel
C), and the dividend-price ratio variance decomposition results (Panel D). Predictive regressions are based on monthly
frequency. Hodrick (1992) t-statistics are reported in parentheses. Panel D reports the mean and the interquartile range
(in square brackets) for the parameters estimated on 12 successive annual samples. Implied dividend growth rates are
dened as: idq
|
= idj
|
dj
|
, where idj
|
is log implied dividend yield and dj
|
is the standard dividend-price ratio. In
Model (1) implied dividend yield is estimated from options in combination with LIBOR. In Model (2) implied dividend
yield is estimated from options in combination with the implied interest rate. In Model (3) implied dividend yield is
estimated from futures in combination with LIBOR. In Model (4) implied dividend yield is estimated from futures in
combination with the implied interest rate. In Model (5) implied dividend yield is estimated simultaneously from futures
and options (as in the baseline case). The period is from January 1994 through June 2011.
(1) (2) (3) (4) (5)
Panel A: Summary statistics for the idq
|
Mean 0.0535 0.0766 0.0493 0.0731 0.0731
Stdev 0.1765 0.1768 0.1664 0.1747 0.1809
AR(1) 0.5666 0.4825 0.6060 0.4964 0.5335
ADF Test -5.1038 -5.7080 -5.4847 -5.7499 -5.5926
Panel B: Predicting annual dividend growth rates (N=198)
idq
|
0.0885 0.1781 0.1136 0.1956 0.1952
(2.06) (4.52) (2.45) (4.59) (4.81)
ad,. 1
2
0.0324 0.1506 0.0493 0.1779 0.1905
Panel C: Predicting monthly returns (N=209)
dj
co::
|
0.0050 0.0129 0.0058 0.0145 0.0188
(0.78) (2.07) (0.88) (2.37) (3.10)
ad,. 1
2
-0.0012 0.0167 -0.0002 0.0218 0.0421
Panel D: Decomposition of the dividend-price ratio
Discount rates 1.5727 1.3682 1.4827 1.3118 1.3425
[0.32] [0.57] [0.39] [0.47] [0.41]
Dividend growth -0.5727 -0.3682 -0.4827 -0.3118 -0.3425
[0.32] [0.57] [0.39] [0.47] [0.41]
Correlation between expected returns and expected dividend growth
j

t
t
0.8588 0.7097 0.8229 0.7094 0.7362
[0.10] [0.30] [0.16] [0.22] [0.22]
65
Table 15: Robustness check: Synchronicity issues
This table reports summary statistics for the implied dividend growth rates (Panel A), regression results for predicting
annual S&P 500 dividend growth rates (Panel B), regression results for predicting S&P 500 monthly returns (Panel
C), and the dividend-price ratio variance decomposition results (Panel D). Predictive regressions are based on monthly
frequency. Hodrick (1992) t-statistics are reported in parentheses. Panel D reports the mean and the interquartile range
(in square brackets) for the parameters estimated on 12 successive annual samples. Implied dividend growth rates are
dened as: idq
|
= idj
|
dj
|
, where idj
|
is log implied dividend yield and dj
|
is the standard dividend-price ratio.
The period is from January 1994 through June 2011. In Model (1) implied dividend yield is estimated from end-of-day
options and futures data (as in the baseline case). In Model (2) implied dividend yield for the period from January 2010
through June 2011 is replaced by the implied dividend yield estimated from intra-daily options and futures data. In
Model (3), implied dividend yield for the period from July 2004 through June 2011 is replaced by the implied dividend
yield estimated from intra-daily options data in combination with the implied interest rate.
(1) (2) (3)
Panel A: Summary statistics for the idq
|
Mean 0.0731 0.0611 0.0735
Stdev 0.1809 0.1920 0.1797
AR(1) 0.5335 0.4864 0.5278
ADF Test -5.5926 -6.2674 -5.6685
Panel B: Predicting annual dividend growth rates (N=198)
idq
|
0.1952 0.1774 0.1884
(4.81) (4.71) (4.78)
ad,. 1
2
0.1905 0.1664 0.1752
Panel C: Predicting monthly returns (N=209)
dj
co::
|
0.0188 0.0202 0.0182
(3.10) (3.38) (3.01)
ad,. 1
2
0.0421 0.0526 0.0393
Panel D: Decomposition of the dividend-price ratio
Discount rates 1.3425 1.3409 1.3588
[0.41] [0.42] [0.46]
Dividend growth -0.3425 -0.3409 -0.3588
[0.41] [0.42] [0.46]
Correlation between expected returns and expected dividend growth
j

t
t
0.7362 0.7628 0.7370
[0.22] [0.24] [0.25]
66
Table 16: Robustness check: Options moneyness and backward-looking data
This table reports summary statistics for the implied dividend growth rates (Panel A), regression results for predicting
annual S&P 500 dividend growth rates (Panel B), regression results for predicting S&P 500 monthly returns (Panel
C), and the dividend-price ratio variance decomposition results (Panel D). Predictive regressions are based on monthly
frequency. Hodrick (1992) t-statistics are reported in parentheses. Panel D reports the mean and the interquartile range
(in square brackets) for the parameters estimated on 12 successive annual samples. Implied dividend growth rates are
dened as: idq
|
= idj
|
dj
|
, where idj
|
is log implied dividend yield and dj
|
is the standard dividend-price ratio. In
Model (1) implied dividend yield is estimated as in the baseline case using 10 days of data at the end of each month and
options across all moneyness levels. In Model (2) implied dividend yield is estimated using 10 days of data at the end
of each month and options with moneyness levels between 0.6 and 1.3. In Model (3) implied dividend yield is estimated
using 10 days of data at the end of each month and options with moneyness levels between 0.8 and 1.1. In Model (4)
implied dividend yield is estimated using 15 days of data at the end of each month and options across all moneyness
levels. In Model (5) implied dividend yield is estimated using 5 days of data at the end of each month and options across
all moneyness levels. The period is from January 1994 through June 2011.
(1) (2) (3) (4) (5)
Panel A: Summary statistics for the idq
|
Mean 0.0731 0.0737 0.0598 0.0696 0.0748
Stdev 0.1809 0.1826 0.2224 0.1674 0.2232
AR(1) 0.5335 0.5207 0.4073 0.5955 0.2869
ADF Test -5.5926 -5.7242 -5.9410 -5.0804 -7.2887
Panel B: Predicting annual dividend growth rates (N=198)
idq
|
0.1952 0.1892 0.1244 0.2255 0.1178
(4.81) (4.78) (4.30) (4.69) (5.18)
ad,. 1
2
0.1905 0.1819 0.1150 0.2191 0.1041
Panel C: Predicting monthly returns (N=209)
dj
co::
|
0.0188 0.0189 0.0182 0.0195 0.0129
(3.10) (3.07) (3.03) (3.10) (2.45)
ad,. 1
2
0.0421 0.0433 0.0537 0.0409 0.0235
Panel D: Decomposition of the dividend-price ratio
Discount rates 1.3425 1.3294 1.3826 1.3324 1.2963
[0.41] [0.36] [0.29] [0.43] [0.42]
Dividend growth -0.3425 -0.3294 -0.3826 -0.3324 -0.2963
[0.41] [0.36] [0.29] [0.43] [0.42]
Correlation between expected returns and expected dividend growth
j

t
t
0.7362 0.7319 0.7486 0.7453 0.7723
[0.22] [0.22] [0.14] [0.21] [0.15]
67
Table 17: Robustness check: Extended sample (January 1990-June 2011)
This table reports summary statistics for the implied dividend growth rates (Panel A), regression results for predicting
annual S&P 500 dividend growth rates (Panel B), regression results for predicting S&P 500 monthly returns (Panel
C), and the dividend-price ratio variance decomposition results (Panel D). Predictive regressions are based on monthly
frequency. Hodrick (1992) t-statistics are reported in parentheses. Panel D reports the mean and the interquartile range
(in square brackets) for the parameters estimated on 12 successive annual samples. Implied dividend growth rates are
restricted to be between -0.5 and 0.5. Time period is either from January 1994 through June 2011 (Model 1) or from
January 1990 through June 2011 (Model 2 and 3).
(1) (2) (3)
Panel A: Summary statistics for the idq
|
Mean 0.0744 0.0821
Stdev 0.1758 0.1998
AR(1) 0.5375 0.4315
ADF Test -5.3872 -6.8784
Panel B: Predicting annual dividend growth rates
dj
|
-0.0441
(-1.69)
idq
|
0.2041 0.1204
(4.91) (4.38)
ad,. 1
2
0.1969 0.0331 0.1066
Panel C: Predicting monthly returns
dj
|
0.0137
(1.45)
dj
co::
|
0.0189 0.0141
(3.10) (2.96)
ad,. 1
2
0.0409 0.0060 0.0315
Panel D: Decomposition of the dividend-price ratio
Discount rates 1.3297 1.2651
[0.36] [0.42]
Dividend growth -0.3297 -0.2651
[0.36] [0.42]
Correlation between expected returns and expected dividend growth
j

t
t
0.7334 0.6796
[0.21] [0.34]
68
Figure 1: Dividend ratios
Panel A: Implied dividend yield and dividend-price ratio
Dec-94 Dec-96 Dec-98 Dec-00 Dec-02 Dec-04 Dec-06 Dec-08 Dec-10
-5.5
-5
-4.5
-4
-3.5
-3
-2.5
Dividend-price ratio
Implied dividend yield
Panel B: Implied dividend growth rates
Dec-94 Dec-96 Dec-98 Dec-00 Dec-02 Dec-04 Dec-06 Dec-08 Dec-10
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
Implied dividend growth
This gure plots dividend ratios for the S&P 500 index. Panel A plots log implied dividend yield along with the log
dividend-price ratio. Panel B plots the dierence between both of themthe implied dividend growth rates. The period
is from January 1994 through June 2011.
69
Figure 2: Dividend-price ratio and corrected dividend-price ratio
Dec-94 Dec-96 Dec-98 Dec-00 Dec-02 Dec-04 Dec-06 Dec-08 Dec-10
-6.5
-6
-5.5
-5
-4.5
-4
-3.5
-3
-2.5
Dividend-price ratio
Corrected dividend-price ratio
This gure plots the log dividend-price ratio along with the log corrected dividend-price ratio for the S&P 500 index.
The period is from January 1994 through June 2011.
70
Figure 3: Implied interest rate, T-bill rate, and LIBOR
Dec-94 Dec-96 Dec-98 Dec-00 Dec-02 Dec-04 Dec-06 Dec-08 Dec-10
0
1
2
3
4
5
6
7
8
Implied interest rate
T-bill
LIBOR
This gure plots 6-month implied interest rate along with the 6-month T-bill rate and the 6-month LIBOR. All rates are
in percentages. The period is from January 1994 through June 2011.
71
Figure 4: Parameter (in-)stability in return predictive regressions
Panel A: Predictive coecients
Dec-04 Dec-05 Dec-06 Dec-07 Dec-08 Dec-09 Dec-10
-0.05
0
0.05
0.1
Dividend-price ratio
Corrected dividend-price ratio
Panel B: Hodrick (1992) t-statistics
Dec-04 Dec-05 Dec-06 Dec-07 Dec-08 Dec-09 Dec-10
-1
0
1
2
3
4
Dividend-price ratio
Corrected dividend-price ratio
This gure plots predictive coecients (Panel A) and Hodrick (1992) t-statistics (Panel B) for univariate predictive
regressions of monthly returns on either the dividend-price ratio or the corrected dividend-price ratio using 120-month
rolling windows. The period is from January 1994 through June 2011. The rst rolling window ends in December 2003.
72
Figure 5: Realized and forecasted returns
Panel A: Realized returns
Dec-00 Dec-02 Dec-04 Dec-06 Dec-08 Dec-10
-20
-15
-10
-5
0
5
10
15
Panel B: Return forecasts
Dec-00 Dec-02 Dec-04 Dec-06 Dec-08 Dec-10
-0.5
0
0.5
1
1.5
2
2.5
Forecasts based on the historical mean
Forecasts based on the dividend-price ratio
Forecasts based on the corrected dividend-price ratio
Panel A plots monthly S&P 500 log returns. Panel B plots monthly return forecasts based on either the historical average
return, the dividend-price ratio, or the corrected dividend-price ratio. All returns are in percentages. The period is from
January 1994 through June 2011. Note the dierence in scales.
73