University of Southern California Dissertations and Theses

Which of stock returns or dividend growth is predictable? A defense of stock return predictability

(USC Thesis Other)

Which of stock returns or dividend growth is predictable? A defense of stock return predictability

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Which of Stock Returns or Dividend Growth is Predictable? 1

Which of Stock Returns or Dividend Growth is Predictable? A Defense of Stock
Return Predictability

Miae Park

Master of Arts (ECONOMICS)
USC Graduate School
University of Southern California

May 2018

Abstract

With CRSP return index widely used to compute the dividend-price ratio in the finance
literature, the lack of dividend growth predictability has been considered as stronger evidence
than does the presence of return predictability - “the dog that did not bark: a defense of return
predictability”, Cochrane (2008). However, CRSP’s completion of Pre62 Daily Data Series
Project made small changes in their historical returns especially before 1947s and a substantial
difference in their implied dividends is generated, which dramatically change the ability of the
dividend-price ratio to forecast dividend growth – now it forecasts both future return and future
dividend growth. In this paper, I comprehensively re-evaluate the empirical evidence of
predictability of stock returns and dividend growth splitting the sample period pre- and pro-1947
and adopting the two additional market returns: return from Robert J. Shiller online website and
return constructed with independently measured NIPA dividends. This paper shows that (1)
depends on the time periods (especially pre-1947s) and data resources taken from, predictability
dividend growth is present. (2) all of the long-horizon return regressions give greater statistical
evidence for return predictability. In addition, the same variables that produce dividend growth
predictability steadily show the identical patterns of return predictability at long horizons, is
stronger evidence of return predictability.

Which of Stock Returns or Dividend Growth is Predictable? 2
Table of Contents
Abstract 1
1 Introduction 3
2 Related Literature 8
3 Study Design & Data 11
3.1 Methodology 11
3.2 Data Description 12
3.3 CRSP Data Change 15
3.4 Graph Visualization 15

4 Empirical Result 21

5 Contribution & Limitations 27

6 Conclusion 28

7 Reference 31

Which of Stock Returns or Dividend Growth is Predictable? 3
1 Introduction
Are stock returns predictable? Inconsistent with the early interpretation of the efficient market
hypothesis, in which it was assumed that expected returns were largely constant. If that were
true, then a regression of future dividend growth on dividend yields should produce a negative
and significant coefficient, whereas a regression of future returns on dividend yields should be
insignificant. Cochrane (2008) reports, like many before him, that this is not what U.S. data
shows. He cites a regression of future returns and future dividend growth on current dividend
yields as empirical evidence, supporting that with respect to U.S. data, a long-horizon predictive
regression with a dividend yield as a regressor is a significant predictor of future returns, not
dividend-growth. Much of the literature has focus on empirical evidence related to stock return
predictability with long-horizon regressions that give greater statistical evidence with horizons,
and the evidence with regard to predictability is interpreted based on the significance of t-
statistics and high R
2
values (Fama and French (1988), Cochrane (1992), Cochrane (2008),
Lettau and Nieuwerburgh (2008), Cochrane (2011)).
The intuition behind the long-horizon return predictability is just a connection between highly
persistent forecasting variables and the variation in expected returns (Cochrane (2008)).
Dividends are a convenient way to detrend prices, which allows low prices relative to dividends
to signal higher future returns, and high prices to signal lower future returns. Thus, numerous
studies report evidence of U.S. equity risk premium predictability adopting various persistent
financial ratios, such as the price-earnings ratio, the market-to-book ratio (Rozeff (1984), Fama
and French (1988), Campbell and Shiller (1988), Cochrane (1991), Goetzman and Jorion (1993),
Hodrick (1992), Lewellen (2004), and others), default spread between Baa and Aaa yield
(Domian and Reichenstein (1998)), the term spread between long-term government yields and T-
Which of Stock Returns or Dividend Growth is Predictable? 4
bill yields (Boudoukh, Richardson and Whitelaw (2005)), the stochastically de-trended risk free
rate (Guo (2002)), and credit spreads (Han, Subrahmanyam, Zhou (2015)), among others etc.
The success of long-horizon regression tests, however, has been subject to great statistical
scrutiny. Studies such as Valkanov (2003), Boudoukh, Richardson and Whitelaw (2005),
Hjalmarsson (2006) suggest that long-horizon return regressions have no greater statistical power
to reject the null hypothesis of no predictability than their short-horizon regressions. They argue
that testing predictability with long-horizon regressions (as opposed to one-period regressions)
would appear to be of little value, concluding that estimates of long-horizon returns do not, in
fact, have better statistical power than one-period returns. Ang and Bekaert (2006) and
Bollerslev and Zhou (2007) also point out that return predictability is best visible at short
horizons using monthly frequencies with an additional regressor such as interest rate or variance
risk premium. In addition, Ang and Bekaert (2006) report that long-horizon stock return
predictability by dividend yields may be a statistical fluke before 1990’s, as the majority of the
literature establishing strong evidence of long-horizon return predictability use data before or up
to the early 1990’s. They fail to reject the null hypothesis of no long-horizon return predictability
after adding the 1990’s decade to the sample.
Cochrane (2008) defends the return predictability of the dividend-price ratio, stating the
absence of dividend growth predictability in the U.S. data gives stronger evidence than does the
presence of return predictability. His statistical estimation is included in Table 1, which reports a
long-horizon forecasting regression of stock returns (left-side column) and dividend growth
(right-side column) using dividend yield extracted from CRSP value-weighted index over 1926-
2004.

Which of Stock Returns or Dividend Growth is Predictable? 5

Table 1. Forecasting Regressions of John Cochrane

Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

B t(b) R
2
b t(b) R
2

1 3.8** 2.7 0.08 0.07 0.06 0.0001
2 7.9** 3.0 0.12 -0.42 -0.22 0.001
3 12.5** 4.5 0.21 0.16 0.09 0.0001
4 17.7** 3.5 0.26 1.37 0.72 0.007
5 20.6** 2.6 0.22 2.42 1.11 0.02
Note. The regression equations are R
y
t, t+k
= α

+ β(D
t
/P
t
)
y
+ ε
t+k
and R
y
t, t+k
= α

+ β(D
t+1
/D
t
)
y
+ ε
t+k
. The
dependent variable R
y
t, t+k
is the CRSP value-weighted return less the three-month Treasury bill return.
D
t
/P
t
is the dividend-price ratio of the CRSP value-weighted portfolio, and D
t+1
/D
t
is real dividend growth.
Data are annual, 1926-2004. P-values less than 0.05 (0.01) level are doenoted by * (**).

The economic size of return-forecast coefficients becomes really interesting over the horizons,
even though the statistical significance of the return-forecast regression is not very large with a t-
statistic that is not above a value of 5. However, the right-side column of Table 1 reports that
dividend-growth forecast regression fails to be significant at all. In fact, the small point estimates
have the wrong sign; a high dividend yield means a low price, which should signal lower, not
higher, future dividend growth. Cochrane reports this lack of dividend growth predictability of
the dividend-price ratio as a stronger evidence to reject the null hypothesis of no return
predictability than does the presence of return predictability. He explains that this is because, in
the language of co-integration, one of dividend growth or price growth must be forecastable to
bring the dividend yield back following a shock. A null hypothesis in which returns are not
predictable therefore must specify that dividend growth is forecastable. If both returns and
dividend growth are unpredictable, then present value logic implies that the dividend yields
should be constant, which it is not.
In this paper, I revisit this empirical evidence using a larger and updated dataset as the
Center for Research in Security Prices (CRSP) completed Pre62 Daily Data Series Project, which
made small changes to their history return series. I find that these small changes in returns have
Which of Stock Returns or Dividend Growth is Predictable? 6
generated substantial differences in their implied dividends. Figure 1 compares the two datasets
of John Cochrane data and Updated CRSP data plotting three panels: returns with dividends,
returns without dividends and dividend yields extracted from those two returns. Two returns
with/without dividends are almost identical while the dividend yield extracted from those two
returns shows a huge difference, especially before 1947s. Therefore, empirically re-testing the
predictability of dividend yield with the updated CRSP-extracted dividend yield is necessary.
Figure 1. John Cochrane Data vs Updated CRSP Data

Figure 1.1: Stock Returns with Dividends

Figure 1.2: Stock Returns without Dividends

-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1/1/26
1/1/28
1/1/30
1/1/32
1/1/34
1/1/36
1/1/38
1/1/40
1/1/42
1/1/44
1/1/46
1/1/48
1/1/50
1/1/52
1/1/54
1/1/56
1/1/58
1/1/60
1/1/62
1/1/64
1/1/66
1/1/68
1/1/70
1/1/72
1/1/74
1/1/76
1/1/78
1/1/80
1/1/82
1/1/84
1/1/86
1/1/88
1/1/90
1/1/92
1/1/94
1/1/96
1/1/98
1/1/00
1/1/02
1/1/04
Stock Returns with Dividends
John Cochrnae R Updated CRSP R
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Stock Returns without Dividends
John Cochrane RX Updated CRSP RX
Which of Stock Returns or Dividend Growth is Predictable? 7

Figure 1.3: Dividend Yields

These figures plot the CRSP value-weighted index by S&P 500 composite portfolio before and after
Pre62 Daily Data Series Project, namely John Cochrane Data and Updated CRSP Data. Figures plot three
time series of stock returns with dividends, returns without dividends and dividend yields calculated from
those two returns. Data is annual from 1926 to 2004.

Replicating the predictive regression using the updated CRSP-extracted dividend-price ratio
over the same period of time 1926-2004 is present in Table 2, which reports the return forecast
(left-side column) and dividend-growth forecast (right-side column) with the updated dividend-
price ratio using annual frequency.
Table 2. Forecasting Regressions of Updated CRSP Data

Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 2.4** 2.2 0.05 -5.47 -5.7** 0.29
2 6.3** 3.9 0.17 -6.19 -4.56** 0.21
3 7.9** 4.1 0.18 -5.18 -3.33** 0.13
4 11.7** 4.9 0.25 -5.19 -2.87** 0.10
5 15.04** 5.6 0.30 -4.24 -2.57** 0.08
Note. The regression equations are R
y
t, t+k
= α

+ β(D
t
/P
t
)
y
+ ε
t+k
and R
y
t, t+k
= α

+ β(D
t+1
/D
t
)
y
+ ε
t+k
. The
dependent variable R
y
t, t+k
is the CRSP value-weighted return less the three-month Treasury bill return.
D
t
/P
t
is the dividend-price ratio of the CRSP value-weighted portfolio, and D
t+1
/D
t
is real dividend growth.
Data is after Pre62 Daily Data Series Project is complete, 1926-2004. P-values less than 0.05 (0.01) level
are denoted by * (**).

I compare the estimated coefficients and t-statistics after Pre62 Daily Data Series Project is
applied (Table 2) with the initial estimation (Table 1). First, the return-forecast coefficients in
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Dividend Yields
John Cochrane D/P Updated CRSP D/P
Which of Stock Returns or Dividend Growth is Predictable? 8
Table 2 are increasing over horizons similar to those coefficients in Cochrane data, but I find the
statistical significance of the return-forecast regression is strongly improved, e.g., a t-statistic at
the 5-year is now over a value of 5. Second, I find that after Pre62 Daily Data Series Project is
completed, the ability of the dividend-price ratio to forecast future dividend growth is
dramatically changed. The right-column of Table 2 shows that dividend-growth forecast appears
to be considerably significant, and all of the coefficients have “correct” sign, which is negative; a
high dividend yield means a low price, which should signal lower future dividend growth. To
sum up, the small changes in the CRSP value-weighted returns by CRSP’s Pre62 Daily Data
Series Project have improved the statistical significance on return predictability and at the same
time, overturned the result of dividend growth predictability. It now shows both return and
dividend growth predictability. This is economically important as it implies, according to a
standard present value model, that not only the expected return news but also the cash flow news
affects stock prices.

2 Related Literature

Dividend-price ratio’s ability to predict dividend growth is equivalent to stating that the cash
flow news affect stock prices (Cochrane, 2008). A few recent papers discuss evidence of
dividend growth predictability in the U.S data. Sabbatucci (2015) shows that if M&A cash flows
are taken into account for the dividend yield extracted from the CRSP value-weighted returns,
expected dividend growth explains approximately 60% of the variation in the dividend-price
ratio, while time-varying discount rates account for the remaining 40%. Chen (2009) reports that
dividend growth predictability from the dividend-price ratio is present only pre-1945, if
dividends are measured without reinvestment. However, he does not find evidence of dividend
Which of Stock Returns or Dividend Growth is Predictable? 9
growth predictability post-1945 or over the full sample. van Binsbergen and Koijen (2010)
model expected returns and dividend growth rates as a latent process with filtering techniques to
show that both of them are predictable within a present value model. However, they do not find
any dividend growth predictability using predictive regressions. Golez (2014) adjusts the
standard dividend-price ratio for changes in expected dividend growth using estimates implied
by the derivatives market and finds that dividend growth is predictable by an implied dividend
growth rate, but not by the unadjusted dividend-price ratio. Maio and Santa-Clara (2014) find
that in a predictive regression framework, dividend growth is predictable using the standard
dividend-price ratio only for small and value stocks, but not for large and growth stocks.
Sabbatucci (2015) reports that the cash flow component in dividends has substantial impact
on dividend growth predictability. The cash flow component, which tends to generate more
spikes (higher volatility) in the series of the dividend-price ratio, is the main factor of dividend
yield’s ability to forecast dividend growth (Hordford (2005)). According to CRSP, dividend
amount is defined as the cash adjustment factor in a holding period return time period used to
calculate returns, which is an adjusted summation of all distribution cash amounts available in
the distribution history with ex-distribution dates. Looking at the top two panels in Figure 1, the
updated CRSP returns with and without dividends show a very small difference from the initial
CRSP returns, but the last panel in Figure 1 shows the dividend yields extracted from the two
returns appear substantially different; the small difference in returns leads to substantial
differences in implied dividends. I see that the the updated dividend yield tends to have more
volatility with more distinctive spikes over the sample periods, which is expected since cash flow
component in dividends is volatile and tends to happen in waves (Hordford (2005)). I also note
that the volatility is especially higher pre-1947, while it tends to decrease after 1947s. This is
Which of Stock Returns or Dividend Growth is Predictable? 10
because CRSP backfilled shares outstanding data for some stocks prior to 1947. To sum up,
CRSP’s Pre62 Daily Data Series Project made small changes in their returns, which produce
substantial differences in their implied dividends. Dividend-price ratios extracted from the
returns, therefore, completely overturn the non-predictability of dividend growth, that now both
stock returns and dividend growth are significantly predictable. Therefore, our evidence of
dividend growth predictability of the dividend-price ratio extracted from the updated CRSP
value-weighted returns also contributes to the existing literature of dividend growth
predictability.
However, it is important to notice a few points about CRSP dividend measurement.
Sabbatucci (2015) notes that the dividend measure from CRSP is potentially biased because the
CRSP value-weighted index is a value-weighted portfolio built using all issues listed on the
NYSE, NASDAQ and AMEX exchanges, except for ADRs. This implies that CRSP value-
weighted index is a proxy for the overall market index, in which the constituents of the index are
not only U.S. common stocks, but other securities such as certificates, SBIs, ETFs and closed-
end funds. As a consequence, Sabbatucci (2015) describes the dividend measure from the CRSP
index as a noisy proxy for common stock dividends that resembles more a measure of the
aggregate dividends within “the CRSP universe”. Second, dividends extracted from the CRSP
value-weighted index only include ordinary and liquidation dividends, which implies that a
series of cash flows, such as cash dividends received during a takeover, are not taken into
account in the dividend measure. However, Boudoukh et al. (2007) argue that excluding such
sources of dividends, especially in recent years where cash dividends have been substituted by
other forms of cash flows is hardly justifiable.
Therefore, testing asset pricing and present value models with a more comprehensive measure
Which of Stock Returns or Dividend Growth is Predictable? 11
of dividends is important in order to assess dividend yield’s predictability of return and dividend
growth. In this paper, I incorporate the two additional stock market returns in order to compare
predictability of dividend yields composed with different dividend measures, (1) return from
Robert J. Shiller’s online website and (2) return from National Institute of Pension
Administrators (NIPA). Moreover, I test the predictive ability of dividend yield in different time
periods to observe the effect of Pre62 Daily Data Series Project to the sample, splitting the total
sample period based on the year 1947: 1926-1946, 1947-2004 and 1947-2016. This will also help
us resolve the concern for interpreting the evidence on excess return predictability using the
dividend yield in regard to a time period, i.e., the trend towards low dividend yields in the 1990’s
created a statistical fluke of predictability of dividend yield.

3 Study Design & Data

3.1 Methodology

We adopt a linear regression model, which is less prone to overfit the data. Our linear regression
to measure the predictive power of equity premium and stock market time-series forecasting
regressions is evaluated with a simple, recursive residuals (out-of-sample) approach. Out-of-
sample R
2
is a statistic that gives us a tool to distinguish alternatives than conventional full-
sample hypothesis tests (Cochrane, 2008). The regressions to be estimated are given by
Equations (4),
R
y
t, t+k
= α

+ β(D
t
/P
t
)
y
+ ε
t+k
(4)

Which of Stock Returns or Dividend Growth is Predictable? 12
Where the left-hand side R
y
t, t+k
is the real stock return on the S&P 500 minus the three-year
Treasury bond yield over 1- and 5-year horizon. The right-hand side D
t
/P
t
is dividend-price ratio
of the S&P 500.
The regressions are run over several time periods over forecasting horizons of 1- and 5-year:
1926-1946, 1947-2004 and 1947-2016. All of the reported t-statistics are based on the
heteroskedasticity and serial correlation consistent standard errors that explicitly take account of
the overlap in the regression following Hodrick (1992). The central issue is forecasting power.
Each forecasting regression is evaluated by the corresponding adjusted R
2
s, statistical
significance by t-statistics and the estimated slope coefficients β.

3.2 Data Description

The empirical analysis presented here is based on the aggregate S&P 500 composite index over
the sample periods 1926-2004, 1947-2004 and 1947-2016. While most of literatures only rely on
the well-known the CRSP value-weighted return on the stock market, I adopt two more market
returns in our analysis, returns from Robert J. Shiller website and returns constructed using NIPA
dividends. Hence, four stock return series are analyzed in this paper: CRSP value-weighted
return before Pre62 Daily Data Series Project is conducted, namely John Cochrane R, CRSP
value-weighted return after Pre62 Daily Data Series Project is applied, namely Updated CRSP R,
return obtained from Robert J. Shiller’s website, shortly Shiller R and return constructed using
NIPA dividends, NIPA R. Data are downloaded directly from the Center for Research in
Security Prices (CRSP), Online Data Robert J. Shiller
(http://www.econ.yale.edu/~shiller/data.htm) and the Bureau of Economic Analysis (BEA)
website.
Which of Stock Returns or Dividend Growth is Predictable? 13
Market returns measure changes in the asset value, providing useful information about the
probability distribution of asset prices (Cochrane (2008)). This is essential for investors and
portfolio mangers as they use this information to value assets and manage their risk exposure.
Stock returns are constructed using the official closing price adjusted for capital actions and
dividends. Each stock price (P
t
) and dividends (D
t
) are calculated as identities:

Return including dividends, R
t+1
= (P
t+1
+ D
t+1
)/P
t
(1)
Return excluding dividends, RX
t+1
= P
t+1
/P
t
(2)

In Equations (1) and (2), P
t
is the adjusted closing price of day t and P
t+1
is the adjusted close
price of the next day. Both R
t+1
and RX
t+1
are subtracted with the three-month Treasury bill
return to obtain excess stock return. However, as Cochrane noted (2008), using continuously
compounded returns or reducing returns by the risk-free rate does not alter the basic nature of the
results.
Dividend yield is obtained by extracting from the two returns (1) and (2). A typical dividend
yield is specified as

D
t+1
/P
t+1
= (1+R
t+1
)/(1+RX
t+1
) – 1 (3)

We aggregate monthly observations to a year to avoid the seasonality, and annual data let us
observe the predictability of dividend yields over the long horizons.
Table 3 provides the descriptive statistics of four datasets. Each dataset contains three series
of variables: return including dividends, return excluding dividends and the dividend yield
extracted from the two returns.
Which of Stock Returns or Dividend Growth is Predictable? 14
Table 3. Descriptive Statistics

Mean Sdev. Skew Kurt
John Cochrane Data

R 0.12 0.2 -0.35 -0.22
RX 0.08 0.2 -0.36 -0.22
D/P 0.04 0.015 0.18 -0.37
Updated CRSP Data

R 0.12 0.19 -0.67 0.64
RX 0.08 0.19 -0.71 0.78
D/P 0.04 0.02 1.85 6.03
Shiller Data

R 0.12 0.17 -0.51 0.62
RX 0.07 0.17 -0.57 0.74
D/P 0.04 0.016 0.53 0.54
NIPA Data

R 0.1 0.17 -0.61 0.87
RX 0.07 0.16 -0.58 0.86
D/P 0.03 0.009 0.19 -1.19
Note. This table reports the mean, standard deviation, skewness and kurtosis of four data sets of Cochrane
data, Updated CRSP data, Shiller data, and NIPA data. Each dataset contains three variables: R, RX and
D/P. R denotes return with dividend, and RX denotes return without dividend. D/P denotes dividend yield
which is calculated as (1+R)/(1+RX) – 1. All stock returns are deflated with 3-month T-bill rate. The
sample period extends from 1926 to 2004.

First, the mean values of the original- and the updated- CRSP returns with and without dividends
are the same, respectively 12% and 8%, corresponding to the same mean value of dividend yield
with 4%. On the other hand, the standard deviation of the dividend-price ratio extracted from the
updated CRSP returns is about two percentage points, almost as 0.5% larger than the standard
deviation of the original CRSP-extracted dividend yield in this sample. It is important to note
that standard deviation is equivalent to our measured volatility. In the second column of Table 3,
all of the four dividend-price ratios (d/p) have different standard deviation. The order of most
volatile to least volatile is the updated CRSP d/p of 2%, Shiller d/p of 1.6%, Cochrane d/p of
1.5% and NIPA d/p of 0.9%. In addition, the dividend-price ratio constructed with NIPA
dividends excluding many of the private-sector estimates moderately decrease its mean and
volatility.
Which of Stock Returns or Dividend Growth is Predictable? 15
3.3 CRSP Data Change
When CRSP research team finds new sources that allow them to make corrections and quality
improvements to their data, CRSP make it a practice to incorporate these changes. Recently,
CRSP has initiated a project to backfill shares outstanding data for some stocks prior to 1947,
and these changes affect the early history of their return series. Edits were applied to 2947
observations for 181 securities between December 31, 1925 and December 31, 1946. These
changes reflect a very small impact in the early 1925-1946 period and from 1947 to present,
levels overlay. A number of securities that did not have shares data prior to the late 1947 now
reflect more accurate start dates. These changes, according to CRSP, mainly resulted in many
small changes to their historical CRSP value-weighted returns and have no impact on CRSP
equal-weighted indexes. The cumulative value of $1 over the life of the value-weighted series
shows a 2% impact due to the changes made in the 1925-1946 range. For a description of these
projects and the types of changes made, please refer to thier website
(http://www.crsp.com/documentation) and refinement documentation
(http://crsp.com/files/images/release_notes/mdaz_201402.pdf).

3.4 Graph Visualization

To illustrate the variables graphically, Figure 1.2.3.4 plot the time series of returns with
dividends, returns without dividends, and dividend yields extracted from the two returns. As
immediately evident from the Figure 1, the updated CRSP returns with and without dividends are
almost identical with the original CRSP returns, but the dividends yields extracted from the two
returns generate a substantial difference, especially pre-1947. The variance measure of the
updated CRSP-extracted dividend yield is somewhat higher during the 1926-1946 part of the
Which of Stock Returns or Dividend Growth is Predictable? 16
sample, while the distinct spikes in the series of dividend yield generally tend to decrease after
the 1947 years. This is an empirical evidence cited above of an indicative of the impact of Pre62
Daily Data Series Project on mainly changes in their implied dividends pre-1947. Figure 2 plots
Shiller data against updated CRSP data in our sample period 1926-2004. Shiller-extracted
dividend yield does a better job in sharing the similar patter with the original CRSP-extracted
dividend yield. However, it also has a few spikes in the period pre-1947 that generally decease
and experience the similar level with the original CRSP-extracted dividend yield after the 1947
years. This is the consistent empirical phenomenon that is present in dividend yields extracted
from market returns. Figure3 illustrates the comparison of NIPA data with original CRSP data in
the sample period 1926-2004. Since NIPA independently measures its dividends and excludes
many of the private-sector for dividend estimates, NIPA dividend yield excludes a few additional
points; the associated mean value is lowest with 3%. As Figure 3 shows, NIPA-extracted
dividend yield has the lowest level of volatility, and it does not vary a lot over time. Therefore, it
is a pure variable for interpreting the evidence of return and dividend growth predictability
regardless of the trend that other dividend yields have in time periods. At the end, I combine all
of the four data series in the figure 4. The variation of dividend yields extracted from the returns
is more obvious, even though the difference between the four return series is small. The former
period 1926-1946 in which CRSP mainly revised its data makes it apparent that there exist some
fluctuations in the dividend ratios, while over the later part 1947-2004 all of four dividend ratios
share the similar movement. On the other hand, regardless of the time period, NIPA-extracted
dividend yield is almost stable with low volatility.

Which of Stock Returns or Dividend Growth is Predictable? 17
Figure 2. John Cochrane Data vs Shiller Data

Figure 2.1: Stock Returns with Dividends

Figure 2.2: Stock Returns without Dividends

-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
Stock Returns with Dividends
John Cochrnae R Shiller R
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Stock Returns without Dividends
John Cochrane RX Shiller RX
Which of Stock Returns or Dividend Growth is Predictable? 18
Figure 2.3: Dividend Yields

These figures plot the original CRSP value-weighted index of S&P 500 and stock index from Robert
Shiller’s website. Figures plot stock returns with dividends, returns without dividends and dividend yields
calculated from those two returns. Data is annual from 1926 to 2004.

Figure 3. John Cochrane Data vs NIPA Data

Figure 3.1: Stock Returns with Dividends

0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
1/1/26
1/1/28
1/1/30
1/1/32
1/1/34
1/1/36
1/1/38
1/1/40
1/1/42
1/1/44
1/1/46
1/1/48
1/1/50
1/1/52
1/1/54
1/1/56
1/1/58
1/1/60
1/1/62
1/1/64
1/1/66
1/1/68
1/1/70
1/1/72
1/1/74
1/1/76
1/1/78
1/1/80
1/1/82
1/1/84
1/1/86
1/1/88
1/1/90
1/1/92
1/1/94
1/1/96
1/1/98
1/1/00
1/1/02
1/1/04
Dividend Yields
John Cochrane D/P Shiller D/P
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
Stock Returns with Dividends
John Cochrnae R NIPA R
Which of Stock Returns or Dividend Growth is Predictable? 19
Figure 3.2: Stock Returns without Dividends

Figure 3.3: Dividend Yields

These figures plot the original CRSP value-weighted index of S&P 500 and S&P 500 stock return
constructed with estimates of NIPA dividends. Figures plot stock returns with dividends, returns without
dividends and dividend yields calculated from those two returns. Data is annual from 1926 to 2004.

-0.6
-0.4
-0.2
0
0.2
0.4
0.6
1/1/30
1/1/32
1/1/34
1/1/36
1/1/38
1/1/40
1/1/42
1/1/44
1/1/46
1/1/48
1/1/50
1/1/52
1/1/54
1/1/56
1/1/58
1/1/60
1/1/62
1/1/64
1/1/66
1/1/68
1/1/70
1/1/72
1/1/74
1/1/76
1/1/78
1/1/80
1/1/82
1/1/84
1/1/86
1/1/88
1/1/90
1/1/92
1/1/94
1/1/96
1/1/98
1/1/00
1/1/02
1/1/04
Stock Returns without Dividends
John Cochrane RX NIPA RX
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Dividend Yields
John Cochrane D/P NIPA D/P
Which of Stock Returns or Dividend Growth is Predictable? 20
Figure 4. John Cochrane Data vs Updated CRSP Data vs Shiller Data vs NIPA Data

Figure 4.1: Stock Returns with Dividends

Figure 4.2: Stock Returns without Dividends

Figure 4.3: Dividend Yields

These figures plot four S&P 500 stock return: return from the original CRSP value-weighted index, return
from the updated CRSP value-weighted index, return from Rober J. Shiller online website and return
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
Stock Returns with Dividends
John Cochrnae R Updated CRSP R Shiller R NIPA R
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
Stock Returns without Dividends
John Cochrane RX Updated CRSP RX Shiller RX NIPA RX
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
1/1/30
1/1/32
1/1/34
1/1/36
1/1/38
1/1/40
1/1/42
1/1/44
1/1/46
1/1/48
1/1/50
1/1/52
1/1/54
1/1/56
1/1/58
1/1/60
1/1/62
1/1/64
1/1/66
1/1/68
1/1/70
1/1/72
1/1/74
1/1/76
1/1/78
1/1/80
1/1/82
1/1/84
1/1/86
1/1/88
1/1/90
1/1/92
1/1/94
1/1/96
1/1/98
1/1/00
1/1/02
1/1/04
Dividend Yields
John Cochrane D/P Updated CRSP D/P Shiller D/P NIPA D/P
Which of Stock Returns or Dividend Growth is Predictable? 21
constructed with estimates of NIPA dividends. Figures plot stock returns with dividends, returns without
dividends and dividend yields calculated from those two returns. Data is annual from 1926 to 2004.

4 Empirical Result

In this section, we start by focusing on the long-horizon out-of-sample forecasting regressions
with the dividend-price ratio as the regressor over three sample periods: 1926-1946, 1947-2004
and 1947-2016. This several sample periods help us to observe the impact of Pre62 Daily Data
Series Project to the sample, especially pre-1947. Further, it helps us observe if a time trend in
dividend yields is a concern for interpreting the evidence on excess return predictability using the
dividend yield. All of the tables are composed of two parts: the left-hand columns are forecasting
regressions of the annual S&P stock return on the lagged dividend-price ratio from 1- to 5-year
horizons and the right-hand columns are of the same with dividend growth. A measure of
predictability considered in this paper is the magnitude and pattern of β-coefficients and
corresponding t-statistics across horizons.
As I mentioned above in Table 1 and 2, after CRSP conducted Pre62 Daily Data Series
Project and made small changes in their returns, the statistical estimation of the predictive
regression with CRSP-extracted dividend yield produce different predictability of stock returns
and dividend growth. In this section, the full sample period of 1926-2004 is split into two sub-
samples based on the year 1947, respectively Table 4 and 5, which allows us to observe the
impact of Pre62 Daily Data Series Project to the sample. Table 4 reports the forecasting
regression with the dividend yield extracted from the original CRSP value-weighted returns over
1926-1946 (Table 4A) and 1947-2004 (Table 4B).

Which of Stock Returns or Dividend Growth is Predictable? 22
Table 4. John Cochrane Data

A. 1926-1946
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2
1 10.22 1.75* 0.15 -1.42 -0.37 0.008
2 21.24 2.41** 0.25 -3.68 -0.62 0.022
3 38.68 4.37** 0.54 1.05 0.17 0.002
4 54.75 5.46** 0.67 7.20 1.16 0.08
5 46.78 3.43** 0.46 5.72 0.73 0.07
B. 1947-2004
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2
1 3.75 2.29** 0.087 0.95 0.75 0.01
2 6.96 2.93** 0.14 1.08 0.64 0.008
3 10.48 3.74** 0.21 1.55 0.86 0.014
4 14.47 4.21** 0.25 2.23 1.22 0.08
5 20.78 7.73** 0.31 4.22 2.23 0.09
Note. We estimate regressions of cumulated and annualized 1-5 years ahead stock returns less risk-free
rates by dividend yields and dividend growth over the two sub-samples, 1926-1946 (A) and 1947-2004
(B). John Cochrane data is the CRSP value-weighted stock index before the Pre62 Daily Data Series
Project is applied. P-values less than 0.05 (0.01) are denoted by * (**).

I find that the left column (return predictability) of Table 4A and 4B show both increasing
coefficients over horizons, but it is important to note that especially t-statistic of Table 4B goes
up all the way until the factor of 7. Using the sample period excluding the pre-1947 years shows
considerable statistical significance of return predictability. This is important information to
Cochrane (2008) who highlights only economic significance of return predictability stating the
relatively weak statistical significance. On the other hand, dividend growth predictability (the
right-hand column) of Table 4 does not show any significance regardless of the time periods.
Table 5 reports the predictive regression on the CRSP-extracted dividend yield after Pre62
Daily Data Series Project is completed over 1926-1946 (Table 5A) and 1947-2004 (Table 5B).

Which of Stock Returns or Dividend Growth is Predictable? 23
Table 5. Updated CRSP Data

A. 1926-1946
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2
1 1.68 0.71 0.03 -9.01 -5.88** 0.66
2 6.45 1.87* 0.17 -11.81 -4.44** 0.54
3 8.50 2.09** 0.21 -10.32 -2.97** 0.36
4 15.42 3.27** 0.40 -9.95 -2.19** 0.24
5 18.47 4.67** 0.61 -5.43 -1.48* 0.14
B. 1947-2004
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2
1 3.93 2.74** 0.12 -3.42 -2.56** 0.11
2 7.99 3.77** 0.21 -2.19 -1.36* 0.06
3 10.27 3.82** 0.20 -0.93 -0.55 0.005
4 12.25 3.50** 0.19 -0.58 -0.35 0.002
5 17.93 4.12** 0.25 -1.76 -0.85 0.01
Note. We estimate regressions of cumulated and annualized 1-5 years ahead stock returns less risk-free
rates by dividend yields and dividend growth over the two sub-samples, 1926-1946 (A) and 1947-2004
(B). The Updated CRSP data is the CRSP value-weighted stock index after the Pre62 Daily Data Series
Project is applied, especially before 1947s. P-values less than 0.05 (0.01) are denoted by * (**).

I find that the multiple-horizon estimated slope coefficients on return predictability (the left-hand
columns) of Table 5A and 5B are both increasing of similar size, which is different from the
return-forecast coefficients for Table 4A and 4B that has the discrepancy in the size such that the
coefficient size of Table 4B is one-third of smaller than Table 4A. Most interestingly, neither
Table 4A nor Table 4B showing dividend growth predictability is entirely overturned in Table
5A and 5B, resulting in dividend predictability that is economically and statistically significant.
It is also important to note that Table 5A from the pre-1947 years reports stronger significant
dividend growth predictability than the predictability after the 1947 in Table 5B: the t-statistics
of the 1- to 5-year estimators are -5.88, -4.44, -2.97, -2.19 and -1.48, respectively in Table A,
while there still remains significant dividend growth predictability in Table B, but the
predictability gets weaken only showing the 1- and 2-year significance. To sum up, dividend
growth predictability is significantly robust with the updated CRSP-extracted dividend yield
Which of Stock Returns or Dividend Growth is Predictable? 24
mainly in the pre-1947 years, and some of the 1- and 2-year significance after the 1947 years. In
addition, this same variable steadily produces the identical patterns of return predictability at
long horizons, is another strong evidence of return predictability.
Table 6 reports the forecasting regression with Shiller-extracted dividend yield over 1926-
1946 (A), 1947-2004 (B) and 1926-2004 (C).
Table 6. Shiller Data

A. 1926-1946
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 1.16 0.26 0.004 -8.12 -4.17** 0.49
2 7.19 1.04 0.06 -9.99 -2.49** 0.27
3 15.36 1.87* 0.18 -7.04 -1.33 0.10
4 28.97 4.13** 0.53 -1.69 -0.29 0.006
5 30.96 5.75** 0.70 3.30 0.56 0.022
B. 1947-2004
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 3.12 2.37** 0.09 1.00 1.976 0.07
2 6.92 3.20** 0.11 2.16 2.272 0.09
3 10.95 3.74** 0.21 3.05 2.329 0.09
4 15.50 4.15** 0.25 3.71 2.442 0.10
5 22.26 4.73** 0.31 4.23 2.67 0.12
C. 1926-2004
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 2.43 1.94** 0.05 -1.42 -2.12** 0.06
2 6.20 3.08** 0.11 -1.61 -1.37 0.02
3 10.28 3.92** 0.0002 -1.17 -0.77 0.008
4 15.65 4.99** 0.25 -0.52 -0.30 0.001
5 19.58 5.33** 0.28 0.23 0.13 0.0002
Note. We estimate regressions of cumulated and annualized 1-5 years ahead stock returns less risk-free
rates by dividend yields and dividend growth over two sub-samples 1926-1946 (A) and 1947-2004 (B)
including the whole sample period, 1926-2004 (C). The Shiller data is the S&P 500 stock returns obtained
from Robert Shiller’s website. P-values less than 0.05 (0.01) are denoted by * (**).

Firstly, I find that all of the left-hand columns (return predictability) of Table 6 A.B.C confirm
the strong return predictability that all of the coefficients and t-statistics increases significantly
over long horizons. Second, Table 6C with the full sample period 1926-2004 indicates that
Which of Stock Returns or Dividend Growth is Predictable? 25
dividend growth can be predictable over the horizon of 1 year. I therefore exclude the period of
pre-1947 and then apply this sub-period to check predictability of dividend growth. Once without
the pre-1947 years, Table 7B shows that dividend growth predictability is not present anymore,
while Table 6A (1926-1946) reports dividend growth predictability with the forecast horizon of
1- and 2-year. In short, replicating the statistical estimation with Shiller-extracted dividend yield
provides much of the same empirical evidence that the updated CRSP-extracted dividend yield
provides, which is strong return predictability regardless of the presence of dividend growth that
is mostly present before the 1947 years.
Table 7 shows the estimates from the predictive regression with the dividend yield
constructed using NIPA dividends and then examine the predictability of stock returns and
dividend growth. The full sample period for this case is not divided based on the year 1947, as
there is no time trend in the dividend-price ratio contracted with NIPA dividends.
Table 7. NIPA Data, 1926-2004

Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 5.06 2.47** 0.08 -0.39 -0.25 0.0008
2 11.48 3.64** 0.16 -1.04 -0.44 0.003
3 18.26 4.56** 0.23 -2.57 -0.88 0.01
4 26.62 5.45** 0.30 -2.28 -0.75 0.008
5 35.28 6.31** 0.37 -1.66 -0.60 0.005
Note. We estimate regressions of cumulated and annualized 1-5 years ahead stock returns less risk-free
rates by dividend yields and dividend growth from 1926 to 2004. The S&P 500 stock returns are
constructed with the estimate of NIPA dividends that is independently measure its dividends from the
market. P-values less than 0.05 (0.01) are denoted by * (**).

The left-hand column (return predictability) in Table 7 shows the slope coefficients significantly
increase over horizons with strong robust statistics i.e., the t-value at the 5-year is 6.31. The
right-hand column, on the other hand reports that dividends are not predictable at all. The
estimate of NIPA dividends, the pure dividend measure without any distortion included, strongly
coincides with Cochrane (2008) finding on predictability of stock returns that is economically
Which of Stock Returns or Dividend Growth is Predictable? 26
and statistically significant over long horizons without any presence of dividend growth
predictability.
Finally, I show the long-horizon predictive regressions over the extended sample period until
2016 excluding the period before 1947 in Table 8. Since pre-1947 dividend growth predictability
is present, the sample period of 1947-2016 allows us to determine if dividend growth
predictability still exists over recent years.
Table 8. 1947-2016
A. Updated CRSP Data
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 3.58 2.70** 0.10 -3.66 -3.05** 0.12
2 7.09 3.64** 0.17 -3.33 -2.24** 0.07
3 8.67 3.48** 0.16 -2.56 -1.54 0.04
4 10.67 3.36** 0.15 -2.34 -1.39 0.03
5 16.56 4.35** 0.23 -3.63 -1.75 0.05
B. Shiller Data
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 2.67 2.26** 0 0.07 0.079 0.13 0.0002
2 5.87 2.97** 0.12 0.015 0.01 0.000003
3 9.47 3.51** 0.16 -0.34 -0.21 0.0007
4 14.22 4.15** 0.21 -1.049 -0.52 0.004
5 21.32 5.14** 0.30 -1.81 -0.85 0.01
C. NIPA Data
Horizon k
(years)
R
t->t+k
→ a+ bD
t
/P
t
+ e
t+k
D
t+k
/D
t
→ a+ bD
t
/P
t
+ e
t+k

b t(b) R
2
b t(b) R
2

1 3.79 2.25** 0.06 -0.51796 -0.387 0.6572
2 9.01 3.42** 0.12 -0.52365 -0.252 0.0007631
3 14.35 4.22** 0.18 -1.14093 -0.434 0.002291
4 21.25 4.99** 0.24 -0.83578 -0.296 0.001077
5 29.06 5.80** 0.30 -0.61271 -0.219 0.0005986
Note. We estimate regressions of cumulated and annualized 1-5 years ahead stock returns less risk-free
rates by dividend yields and dividend growth over 1947-2004 with three data sets. Data sets are the
Updated CRSP data, Shiller data and NIPA data. P-values less than 0.05 (0.01) are denoted by * (**).

The left-hand columns (return predictability) of Table 8 show the exact pattern one should expect
under the null hypothesis of no return predictability by now. All of the forecast produce
Which of Stock Returns or Dividend Growth is Predictable? 27
increasing coefficient estimates and higher t-statistics at the longest time periods; the t-statistic is
significant, and that significance really go up all that much at longer horizons. However, the
issue of predictability of dividend growth is controversial. The CRSP-extracted dividend yield
shows dividend growth predictability over the horizon of 1- and 2-year, with a t-statistics of -
3.05 and -2.24, respectively, while Shiller- and NIPA-extracted dividend yields do not show
dividend growth predictability at all. Thus, the long-horizon regression equations of the
dividend-price ratios do not provide convincing evidence to reject the no-predictability
hypothesis of dividend growth. However, the null hypothesis of no return predictability is
strongly rejected with the empirical evidence that is consistently shown and clarifies the
controversial concerning the return predictability with showing weak statistical significance or
due to statistical fluke due to the year 1990s. Our empirical evidence proves stock return
predictability is statistically and economically significant regardless of data sources and time
periods. In addition, it steadily shows the identical pattern of increasing long-horizon coefficients
whether dividend growth predictability is present or not is stronger evidence of return
predictability.
5 Contribution & Limitations
To the best of my knowledge, this paper is the first to explicitly analyze the predictability of
dividend-price ratios on stock returns and dividend growth with various dividend yields over
several time periods considering effect of Pre62 Daily Data Series Project to the sample, pre- and
pro-1947. In so doing, I make several contributions to the predictability literature. First, dividend
growth predictability is present depends on the time period (especially pre-1947) and data
sources taken from. Using the CRSP-extracted dividend yield, dividend growth predictability is
Which of Stock Returns or Dividend Growth is Predictable? 28
present in all periods: 1926-1946, 1947-2004 and 1947-2016, but its significance is especially
strong pre-1947. I also proceed to show that Shiller-extracted dividend yield shows dividend
growth predictability but only pre-1947. Second, all of the return predictive regressions of
dividend yields over different time periods produce a consistent empirical result. All of the return
forecasting coefficients are increasing at 1-year to 5-year horizons, and statistical significance
has strong power advantages to long-horizon regressions that are maximized at very long
horizons. In addition, the same variables that produce dividend growth predictability steadily
show identical patterns of return predictability at long horizons, is stronger evidence of return
predictability.
Our limitation is that spurious rejections are most likely to occur with a short sample
evaluation sample. Even with the size distortions, we propose that is robust to the effect of
considering multiple sample split points (Timmermann 2012). Empirical applications to
predictability of stock returns and dividend demonstrate that out-of-sample forecast evaluation
can be a great tool to critically compare the forecasting power depend on the time period.

6 Conclusion

Long-horizon stock return predictability is considered as one of the most important pieces of
evidence in the empirical asset pricing literature over the last couple of decades (Campbell, Lo,
and MacKinlay (1997) and Cochrane (2001)). This paper re-evaluates the evidence of long-
horizon predictive regression of the dividend-price ratio, taking into account the completion of
Pre62 Daily Data Series Project by CRSP that has produced the small changes in their value-
weighted returns. These changes have resulted in substantial differences in their implied
Which of Stock Returns or Dividend Growth is Predictable? 29
dividends, which dramatically changes the ability of the dividend-price ratio to forecast future
dividend growth.
Out of all the regressions performed over the periods: 1926-2004, 1926-1946 and 1947-2004,
dividend growth predictability is present depends on the time period (especially the pre-1947
years) and the type of sources data is taken from. The predictive regression of CRSP-extracted
dividend yield is strongly significant over the pre-1947 years, and some of the 1- and 2-year
significance after the pro-1947 years. In addition, the predictive regression of Shiller-extracted
dividend yield shows the presence of dividend growth predictability but only for pre-1947. To
further understand the reality of dividend growth predictability, I provide a visual representation
comparing the movement of dividend yield series: the higher the volatility (more spikes) of the
dividend yield series is within its known persistent movement, the more the dividend is captured
to be predictable. Therefore, during the time periods when the movement of dividend yield series
is relatively stable, for example, Shiller-extracted dividend yield after the 1947 years and NIPA-
extracted dividend yield over the full sample period, do not show any dividend growth
predictability.
While dividend growth predictability is this controversy, we obtain a nice resolution of the
long-horizon return predictability. All of the dividend yields consistently show statistical and
economic significance of return predictability, even though the predictive power (t-statistics)
depends on the time period and estimation detail. The size of coefficients increases over the
long-horizon regardless of the sample size and time periods, and the statistical significance of
return forecasts gets as considerably significant as its economic significance such that the t-
statistics at the fifth horizon reaches a factor of 5. In addition, the same variables that produce
Which of Stock Returns or Dividend Growth is Predictable? 30
dividend growth predictability steadily show the identical patterns of return predictability at long
horizons, is stronger evidence of return predictability.
To sum up, the argument that the absence of dividend growth predictability is stronger
evidence of return predictability is not valid anymore with the CRSP value-weighted index, as
CRSP revised its stock returns and the dividend-price ratio extracted from the returns now
predict both stock returns and dividend growth. However, it does not mean that the power of
return predictability is diminished by the presence of dividend growth predictability. Instead, the
updated CRSP-extracted dividend yield itself and other several dividend yields incorporated
regressions with different time periods gives us stronger evidence of return predictability that is
consistently shown whether dividend growth predictability is present or not.

Which of Stock Returns or Dividend Growth is Predictable? 31
Reference
Ang, Andrew., & Bekaert, Geert. (2006). Stock Return Predictability: Is it There?. The Review of
Financial Studies, 20(3), 651-707.

Ang, Andrew., & Timmermann, Allan. (2012). Regime Changes and Financial Markets. Annual
Review of Financial Economics, vol. 4(1), pages 313-337.

Boudoukh, Jacob., & Whitelaw, F. Robert. (2005). The Myth of Long-Horizon Predictability.
The Review of Financial Studies, vol 21(4), pages 1577–1605.

Boudoukh, Jacob., Michaely, Roni., Richardson, Matthew., & Roberts, R. Michael. (2007). On
the Importance of Measuring Payout Yield: Implications for Empirical Asset Pricing, The
Journal of Finance, vol 62(2), pages 877–915.

Bollerslev, Tim., Tauchen, George., & Zhou, Hao. (2007). Expected Stock Returns and Variance
Risk Premia. The Review of Financial Studies, vol 22(11).

Campbell, J., A. Lo, & C. MacKinlay. (1997). The Econometrics of Financial Markets.
Princeton University Press.

Campbell, Y, John., & Shiller, J, Robert. (2001). Valuation Ratios and the Long-Run Stock
Market Outlook: An Update. Cowles Foundation Discussion Papers 1295, Cowles Foundation
for Research in Economics, Yale University.

Campbell, Y, J., & Thompson, B, S. (2005). Predicting the equity premium out of sample: Can
anything beat the historical average?. The Review of Financial Studies 21(4): 1509-1531.

Chen, Long. (2009). On the reversal of return and dividend growth predictability: A tale of two
periods, Journal of Financial Economics vol. 92(1), pates 128–151.

Cornell, Bradford. (2014). Dividend-price ratios and stock returns: International evidence.
California Institute of Technology. Available at
SSRN: https://ssrn.com/abstract=2072178 or http://dx.doi.org/10.2139/ssrn.2072178

Cochrane, H.J. (1992). Explaining the Variance of Price-Dividend Ratios. Review of Financial
Studies, vol 5(2), 243-280.

Cochrane, H. J. (2008). The Dog That Did Not Bark: A Defense of Return Predictability. The
Review of Financial Studies, vol 21(4), Pages 1533–1575.

Da, Zhi., & Jagannathan, Ravi. (2014). Growth expectations, dividend yields, and future stock
returns. National Bureau of Economic Research (NBER).

Domian, L. Dale., & Reichenstein, William. (1998). Term Spreads and Predictions of Bond and
Stock Excess Returns. Financial Services Review, vol 7(1), pages 25-44.
Which of Stock Returns or Dividend Growth is Predictable? 32

Fama, E., & R. Bliss. (1987). The Information in Long Maturity Forward Rates. American
Economic Review, vol 77, pages 680–92.

Fama, F. Eugene., & French, R. Kenneth. (1988). Dividend yields and expected stock returns.
Journal of Financial Economics, vol 22(1), pages 3-25.

Golez, Benjamin. (2014). Expected returns and dividend growth rates implied by derivatives
markets, The Review of Financial Studies vol. 27(3), pages 790–822.

Golez, Benjamin., & Koudijs, Peter. (2017). Four Centuries of Return Predictability. Journal of
Financial Economics. Forthcoming, Stanford University Graduate School of Business Research
Paper No. 17-12

Goyal, Amit., & Welch, Ivo. (2002). Predicting the equity premium with dividend ratios. Journal
of Empirical Finance, vol 17(4), pages 539-551

Goyal, Amit., & Welch, Ivo. (2004). A comprehensive look at the empirical performance of
equity premium prediction. The Review of Financial Studies, vol 21(4), pages 1455-1508

Guo, Hui. (2002). Time-Varying Risk Premia and the Cross Section of Stock Returns. Journal of
Banking & Finance, vol 30(7), pages 2087-2107.

Han, Bing., Subrahmanyam, Avanidhar., and Yi Zhou. (2015). The Term Structure of Credit
Spreads and the Cross-Section of Stock Returns. Journal of Financial Economics, vol 124, pages
147-171.

Hodrick, J. Rober. (1992). Dividend Yields and Expected Stock Returns: Alternative Procedure
for Inference and Measurement. The Review of Financial Studies, vol. 5, no. 3, pages 357-386.

Hardford, Jarrad. (2005). What drives merger waves?. Journal of Financial Economics, vol 77,
pages 529–560.

Hjalmarsson, E. (2010). Predicting global stock returns. Journal of Financial and Quantitative
Analysis, vol 45, pages 49–80. Hjalmarsson, E. (2006). New Methods for Inference in Long-Run Predictive Regressions.
International Finance Discussion Papers 853. Board of Governors of the Federal Reserve System

Jones, Mark., & Neuberger, Anthony. (2011). Improved Inference and Estimation in Regression
with Overlapping Observations. Journal of Business Finance & Accounting, vol 38, no 5-6,
pages 657-683.

Maio, Paulo, & Pedro Santa-Clara. (2015). Dividend yields, dividend growth, and return pre-
dictability in the cross-section of stocks, Journal of Financial and Quantitative Analysis, vol. 50,
pages 33-60.
Which of Stock Returns or Dividend Growth is Predictable? 33

Mark, N. C. (1995). Exchange Rates and Fundamentals, Evidence on Long Horizon
Predictability. American Economic Review, vol 85(1), pages 201–218.

McMillan, G. David. (2012). Stock Return Predictability: Risk Premium or Dividend Growth?.
Journal of Accounting & Marketing, J Account Mark 2:e111.doi:10.4172/2168-9601.1000e111.

Lewellen, Jonathan. (2002). Predicting Returns with Financial Ratios. Journal of Financial
Economics, vol 74, pages 209–235.

Lettau, Martin., & Nieuwerburgh, Stijin. (2006). Reconciling the return predictability evidence.
The Review of Financial Studies, vol 21(4), pages 1607–1652.

Neely, J. Christopher., & Rapach, E. David. (2013). Forecasting the equity risk premium: The
role of technical indicators. Available at SSRN: https://ssrn.com/abstract=1787554
or http://dx.doi.org/10.2139/ssrn.1787554

Rapach, David E., & Wohar, Mark E. (2006). In-sample vs. out-of-sample tests of stock return
predictability in the context of data mining. Journal of Empirical Finance, vol 13(2), pages 231-
247.

Sabbatucci, Riccardo. (2015). Are Dividends and Stock Returns Predictable? New Evidence
using M&A Cash Flows. Available at SSRN: https://ssrn.com/abstract=2578393
or http://dx.doi.org/10.2139/ssrn.2578393

Shiller, J. Robert. (1996). Price–earnings ratios as forecasters of returns: The stock market
outlook in 1996. Yale Economics. http://www.econ.yale.edu//~shiller/data/peratio.html.

Phan, Ding., Sharma, Susan., & Narayan, Paresh. (2015). Stock Return Forecasting: Some New
Evidence. International Review of Financial Analysis, vol 40, pages 38-51

Phillips, C.B. Peter., & Lee, Ji Hyung. (2013). Predictive regression under various degrees of
persistence and robust long-horizon regression. Journal of Econometrics. vol 177. pages 250-
264.

van Binsbergen, Jules H., & Ralph S.J. Koijen. (2010). Predictive regressions: A present-value
approach, Journal of Finance vol. 65, pages 1439–1471.

Valkanov, Rossen. (2003). Long-Horizon Regressions: Theoretical Results and Applications.
Journal of Financial Economics. vol. 68(2), pages 201-232.

Zhou, Chunsheng. (1996). Forecasting Long- and Short-Horizon Stock Returns in a United
Framework. Federal Reserve Board.

Wolf, Michael. (1997). Stock Returns and Dividend Yields Revisited: A New Way to Look at an
Old Problem. Journal of Business & Economic Statistics, vol. 18(1), pages 18-30.
Which of Stock Returns or Dividend Growth is Predictable? 34

Wu, Jyh-Lin, & Hu, Yu-Hau. (2011). Price–Dividend Ratios and Stock Price Predictability.
Journal of Forecasting, vol31(5), pages 423–442.

Abstract (if available)

Abstract With CRSP return index widely used to compute the dividend‐price ratio in the finance literature, the lack of dividend growth predictability has been considered as stronger evidence than does the presence of return predictability—“the dog that did not bark: a defense of return predictability”, Cochrane (2008). However, CRSP’s completion of Pre62 Daily Data Series Project made small changes in their historical returns especially before 1947s and a substantial difference in their implied dividends is generated, which dramatically change the ability of the dividend‐price ratio to forecast dividend growth—now it forecasts both future return and future dividend growth. In this paper, I comprehensively re‐evaluate the empirical evidence of predictability of stock returns and dividend growth splitting the sample period pre‐ and pro‐1947 and adopting the two additional market returns: return from Robert J. Shiller online website and return constructed with independently measured NIPA dividends. This paper shows that (1) depends on the time periods (especially pre‐1947s) and data resources taken from, predictability dividend growth is present. (2) all of the long‐horizon return regressions give greater statistical evidence for return predictability. In addition, the same variables that produce dividend growth predictability steadily show the identical patterns of return predictability at long horizons, is stronger evidence of return predictability.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Expectation dynamics and stock returns

PDF

The application of machine learning in stock market

PDF

Predicting and modeling human behavioral changes using digital traces

PDF

A spatially explicit approach to measuring carbon dynamics for reducing emissions from deforestation and forest degradation: a case study of Chinese forests

PDF

Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes

PDF

Geospatial analysis of the Round Fire: a replication of burn severity analyses in the Sierra Nevada

PDF

Developmental etiology of reactive and proactive aggression, electrodermal reactivity and their relationships: a twin study

PDF

Germanium and silicon isotope geochemistry in terrestrial and marine low-temperature environments

Asset Metadata

Creator Park, Miae (author)

Core Title Which of stock returns or dividend growth is predictable? A defense of stock return predictability

School College of Letters, Arts and Sciences

Degree Master of Arts

Degree Program Economics

Publication Date 05/08/2018

Defense Date 05/08/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag CRSP data change,dividend growth predictability,dividend yield,dividends,long‐horizon return regression,OAI-PMH Harvest,stock return predictability

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kocer, Yilmaz (committee chair), Imrohoroglu, Selahattin (committee member), Rashidian, Manochehr (committee member)

Creator Email miaepark@usc.edu,miaepark1669@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-502531

Unique identifier UC11267315

Identifier etd-ParkMiae-6321.pdf (filename),usctheses-c40-502531 (legacy record id)

Legacy Identifier etd-ParkMiae-6321.pdf

Dmrecord 502531

Document Type Thesis

Format application/pdf (imt)

Rights Park, Miae

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA