Вы находитесь на странице: 1из 35

The Financial Review 47 (2012) 531564

Do Decomposed Financial Ratios Predict


Stock Returns and Fundamentals Better?
Xiaoquan Jiang
Florida International University

Bong Soo Lee


Florida State University

Abstract
We investigate the prediction of excess returns and fundamentals by financial ratios, which
include dividend-price ratios, earnings-price ratios, and book-to-market ratios, by decomposing
financial ratios into a cyclical component and a stochastic trend component. We find both
components predict excess returns and fundamentals. Cyclical components predict increases
in future stock returns, while stochastic trend components predict declines in future stock
returns in long horizons. This helps explain previous findings that financial ratios in the
absence of decomposition find weak predictive power in short horizons and some predictive
power in long horizons. We also find both components predict fundamentals.
Keywords: financial ratios, return predictability, decomposition, fundamentals
JEL Classifications: G12, G14

Corresponding author: Florida International University, 11200 SW 8th Street, RB 203B, Miami, FL
33199; Phone: (305) 348-7910; Fax: (305) 348-4245; E-mail: jiangx@fiu.edu.

We would like to thank Robert Van Ness (the editor) and two anonymous referees for constructive
and detailed suggestions. Helpful comments were received from the participants at seminar at Florida
International University, 2008 SUERF Colloquium (European Money and Finance Forum), Munich,
Germany, 2009 SFA Conference, Captiva FL, and 2010 FMA Conference, New York, NY. All remaining
shortcomings are our own responsibility.

C 2012, The Eastern Finance Association

531

532

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

1. Introduction
Whether financial ratiossuch as dividend-price ratios, earnings-price ratios,
and book-to-market ratiospredict future stock returns has drawn much attention
from both practitioners and academia. The prediction of stock returns using financial
ratios is theoretically rooted in present value models (e.g., the dividend discount
model, earnings discount model, and residual income model). These models imply
that stock return is determined by fundamentals embedded in financial ratios. Using
a log-linear present value model, Campbell and Shiller (1988a) demonstrate that
the dividend-price ratio is either positively associated with future stock returns or
negatively associated with future dividend growth rates. Empirical evidence generally
suggests that financial ratios can predict future stock returns, especially in long
horizons.1 In contrast, short-horizon returns and the growth of fundamentals are
more difficult to predict.
A consensus view of a quarter century of empirical work has been summarized
in Cochranes book, Asset Pricing (2001, p. 388): 2. Returns are predictable. In
particular, (a) Variables including the dividend/price ratio and term premium can in
fact predict substantial amounts of stock return variation. This phenomenon occurs
over business cycle and longer horizons. Daily, weekly, and monthly stock returns
are still close to unpredictable . . .
Recently, this view has been questioned and challenged on several grounds. It is
well known that financial ratios are very persistent, and return shocks are negatively
correlated with financial ratios. Because of the near-unit-root property of financial
ratios, the statistical inference leads to uninformative inference on the prediction
relation (e.g., Nelson and Kim, 1993; Stambaugh, 1999; Ferson, Sarkissian and
Simin, 2003; Valkanov, 2003; Lewellen, 2004). Ang and Bekaert (2007) provide
evidence that dividend yields do not significantly predict excess returns at long
horizons after carefully accounting for small-sample properties of standard tests.
Goyal and Welch (2003, 2008) show that financial ratios have poor out-of-sample
(OOS) forecast power.
The prediction relation between returns and financial ratios appears to suffer
from structural instability over time. Particularly in the late 1990s, the prediction
relation does not seem robust (see Viceira, 1996; Lettau and Ludvigson, 2001; Goyal
and Welch, 2003; Paye and Timmermann, 2006; Lettau and Van Nieuwerburgh, 2008).
We reexamine the predictive power of financial ratios in predicting stock returns
and fundamentals.2 For this purpose, we propose decomposing financial ratios into a

1 See Rozeff (1984), Fama and French (1988), Pontiff and Schall (1998), Campbell and Shiller (1988a,
1988b, 2005), Cochrane (1992, 2008), Goetzman and Jorion (1993), Hodrick (1992), Lewellen (2004),
Lettau and Ludvigson (2005), Hecht and Vuolteenaho (2006), Jiang and Lee (2007), Lettau and Van
Nieuwerburgh (2008), and others.
2 Jiang and Lee (2009) reexamine the intertemporal risk-return relation and find a positive risk-return
relation by measuring expected returns and conditional variance in a consistent manner using firm fundamentals. As measures of fundamentals, they use earnings and dividends.

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

533

cyclical component and a stochastic trend component using Hodrick and Prescotts
(1997) Kalman filter procedure. To address a finite sample bias issue in the forecasting
regressions, we use the bootstrap simulation procedure in testing the null hypothesis
of no predictability. To circumvent the look-ahead bias in the OOS forecasts, we
provide an alternative decomposition.
Using the decomposed components of financial ratios, we find that both excess returns and fundamentals are significantly predictable for the sample periods
from 1926:Q1 to 2008:Q4 and from 1952:Q1 to 2008:Q4. The cyclical components
of financial ratios predict an increase in stock returns, while the stochastic trend
components of financial ratios predict a decrease in stock returns. Their predictive
power peaks at horizon of around 12 quarters. At the horizon of 12 quarters for the
sample period from 1926:Q1 to 2008:Q4, the explanatory powers (R2 s) of the two
components of dividend-price ratio, earnings-price ratio, and book-to-market ratio
are 45%, 72%, and 77%, respectively. The cyclical components tend to dominate the
stochastic trend component in the explanatory power. Consistent with present value
models, we also find that both components predict fundamentals. For example, for
dividend-price ratio, we find the cyclical component tends to predict dividend growth
with a negative coefficient while the stochastic trend component tends to predict
dividend growth with a positive coefficient.
We also aim to explore the channels through which returns are predicted so
that we can provide better understanding of previous findings that suggest financial
ratios are weak predictors of stock returns in short horizons, while they show stronger
predictive power in long horizons. For example, the log dividend ratio model anticipates that an increase in this financial ratio is associated with future increases in
stock returns and/or future declines in dividend growth rates. Return forecast studies
generally attribute the forecast power of financial ratios to (slow) mean reversion
(see Campbell and Shiller, 1988a, 2005).3 We interpret the cyclical component as
reflecting a local mean reversion effect, while the stochastic trend component reflects
a long-run persistence (slow mean reversion) effect. Ferson and Xie (2009) find that
the long-run risk models better capture momentum effect, which is consistent with
this persistence effect.4 By decomposing the financial ratios into two components,
we show that the two components of financial ratios can better predict both stock
returns and fundamentals than the financial ratios alone.
Therefore, we find returns and fundamentals are predictable through two channels. In particular, we find that the stochastic trend components are persistent and
have higher explanation powers over long horizons. This indicates that the longrun persistence in the stochastic trend components of financial ratios is particularly
effective over long horizons. Since the stochastic trend components tend to be more

In fixed income research, Fama (2006) also proposes local mean reverting and slow mean reverting
components to study the behavior of interest rates. Park (2010) shows the ratio of the short-term-moving
to the long-term-moving average has significant predictive power for future returns.

Our model can be viewed as the cointegrated version of LRR model in Ferson and Xie (2009).

534

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

important in the financial ratios over long horizons, this helps explain previous findings that raw financial ratios tend to show some predictive power for stock returns
over long horizons. In addition, our finding that the two components predict stock returns in the opposite direction suggests that over short horizons, the two components
of financial ratios tend to offset much of each others prediction. This helps explain
the failure of previous studies to find strong predictive power of raw financial ratios
over short horizons in the absence of decomposition. Our evidence also helps explain
why the dividend-price ratio does not predict dividend growth. Lettau and Ludvigson
(2005) argue that the failure of dividend-price ratio to predict dividend growth is due
to the positive correlation between the expected return and the expected dividend
growth, suggesting that time-varying investment opportunities are poorly captured
by dividend-price ratio. Our research suggests that these investment opportunities
are captured separately by each component of dividend-price ratio, and the failure of
dividend-price ratio to predict dividend growth is due to the offsetting effect between
the cyclical and stochastic trend components.
In their recent paper, Lettau and Van Nieuwerburgh (2008) suggest that the
changes in the steady-state mean of financial ratios cause the puzzling empirical patterns in return prediction. They estimate regime-switching models for the steady-state
mean of financial ratios. In regime-switching or structural break models, however,
the number of parameters to be estimated grows rapidly with the number of regimes
or breaks. Generally, two or, at most, three regimes are allowed, and more regimes
imply a lower power. The permanent change in the financial ratios can be attributed
to such factors as technological innovations, changes in market participations, and
tax codes. As such, two or three regimes may not suffice.
In our research, we employ Hodrick and Prescotts (1997) filter procedure in
decomposing financial ratios into stochastic trend and cyclical components, which is
more flexible than allowing for regime changes. To link Hodrick and Prescotts (1997)
Kalman filter decomposition to the regime-switching models, the cyclical component (stochastic long-run trend component) in Hodrick and Prescotts Kalman (1997)
filter is similar to the stochastic component (steady-state value) in regime-switching
settings. The stochastic long-run trend component is, however, time-varying in low
frequency and can be used to forecast stock returns and growth in fundamentals
as an additional predictor. Therefore, Hodrick and Prescotts (1997) filter allows
us to generate the time-varying long-run stochastic trend component, which can
be viewed as a time-varying conditional mean. As a result, we do not directly examine the structural break issue as in Lettau and Van Nieuwerburgh (2008). Instead, we apply Hodrick and Prescotts (1997) filter method to decompose financial
ratios.
Intuitively, the stochastic trend component is measured as a weighted average
of financial ratio over time, and cyclical component is measured as residuals. We use
Hodrick and Prescotts (1997) filter for two reasons. Their decomposition reflects
new classical economic theory (see Lucas, 1980, 1981). In terms of methodology,
Hodrick and Prescotts (1997) filter procedure is very flexible. It can be used to

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

535

Figure 1
Financial ratios (DP, EP, and BP) and their cyclical components and stochastic trend components
with NBER business cycles

decompose highly persistent stationary series (e.g., value premium in Chen, Petkova
and Zhang, 2008), as well as potentially nonstationary series (e.g., gross domestic
product [GDP]). This is particularly useful for financial ratios because it is still
debated whether financial ratios contain a unit root. In theory, they are expected to
be stationary as we can see from various present value models. However, Figure 1
shows that they are highly persistent with a downward trend.

536

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

In terms of economic and finance intuition, Blanchard and Watson (1982) and
Froot and Obstfeld (1991) show that, if stock price is allowed to deviate from its
intrinsic value (i.e., if bubbles are allowed), regardless of whether they are rational or irrational, financial ratios can be decomposed into two parts: the random
walk and bubbles.5 Campbell and Vuolteenaho (2004) show that returns generated by discount rate risk are offset by lower returns in the future (mean reversion), whereas returns generated by cash flow risk are never reversed subsequently
(persistence). Motivated by this insight, we interpret the predictive power of cyclical components of financial ratios as due to a local mean reversion, and that of
stochastic trend components of financial ratios as due to a long-run persistence
effect. Similarly, Fama and French (2007), in studying average returns on value
and growth portfolios, employ a decomposition of capital gains into three sources:
growth in book value, convergence in market-to-book ratios, and drift in market-tobook. Another advantage is that besides the cyclical component, it also generates
the long-run stochastic trend component, which is time-varying and can be used as a
predictor.
Fama and French (2002) argue that average stock return of the last half century is
a lot higher than expected. The observation that 1990s financial ratios are low, while
stock returns remain high makes the return prediction more difficult. The decomposed
financial ratios help us better predict returns both in short and long horizons. Using
the consumption-wealth ratio (CAY), Lettau and Ludvigson (2001, 2005) find that
excess returns are predictable. Consistent with Lettau and Ludvigson (2001, 2005)
and Ang and Bekaert (2007), we find both CAY and relative T-bill rates able to
predict stock returns in the univariate regression. However, when we include the two
components of financial ratios in the prediction regression, the predictive power of
CAY and relative T-bill rates tends to disappear, while both the cyclical and stochastic
trend components of financial ratios predict stock returns significantly in all horizons
considered. We find that the two components of financial ratios have more predictive
power than CAYs and relative T-bill rates.
We also find that dividend growth, earnings growth, and accounting returns are
predictable using the cyclical and stochastic trend components of financial ratios.
Dividend growth is typically predictable by the cyclical component of dividendprice ratio, especially over short horizons. Accounting returns are predictable by the
stochastic trend component of book-to-market ratio, in particular, over long horizons.
Earnings growth is predictable by the cyclical component of the earning-price ratio
over short horizons and by the stochastic trend component of the earning-price ratio
over long horizons.

5 Wu (1997) uses a similar Kalman filter to decompose price-dividend ratio to account for U.S. stock price
volatility.

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

537

2. Methodology
2.1. Decomposition procedure
In view of Figure 1, which shows highly persistent financial ratios, and considering the insights provided by Fama (2006) and Lettau and Ludvigson (2005) and
methodologies developed by King, Plosser, Stock and Watson (1991) and Hodrick
and Prescott (1997), we decompose financial ratios into a stochastic trend component
and a cyclical component. To implement the decompositions, we use the Hodrick
and Prescott (1997) filter procedure. Hodrick and Prescotts (1997) statistical approach does not utilize standard time series analysis. Instead, Hodrick and Prescott
(1997) proceed in a more cautious manner that requires only prior knowledge that
can be supported by economic theory. Hodrick and Prescott (1997, p. 2) state, The
maintained hypothesis, based upon growth theory considerations, is that the growth
component of aggregate economic time series varies smoothly over time. Furthermore, it does not require the underlying series to be necessarily nonstationary and
is flexible enough to capture potential regime changes in a parsimonious manner.
Therefore, it is widely adopted in existing studies.
Let FRt be a log financial ratio at time t. FRt can be log dividend-price ratio,
log earnings-price ratio, or log book-to-market ratio. Following Hodrick and Prescott
(1997), we model FRt as
FRt = Gt + Ct ,

(1)

where Gt is a stochastic trend component that varies smoothly over time, and Ct is a
cyclical component. Since the cyclical component is the residual, it should be close
to zero on average in the long run.
The stochastic trend component can be obtained by solving the following programming problem:
 T

T


2
2
min
Ct +
(2)
[Gt 2Gt1 + Gt2 ] ,
Gt

t=1

t=1

where is a positive parameter that penalizes variability in the stochastic trend


component. The larger the value of , the smoother the stochastic trend component.
There are two attractive properties in this decomposition for predicting returns.
First, the cyclical component, Ct , is stationary and not persistent. It is well known
that highly persistent predictors tend to lead to a small-sample problem (Stambaugh,
1986, 1999) and spurious predictions (Ferson, Sarkissian and Simin, 2003). A stationary cyclical component is less subject to these criticisms.6 Second, the stochastic
trend component is not a constant, instead it is time-varying and smooth (Hodrick
and Prescott, 1997). Furthermore, it captures fundamental structural changes well.
6

In forecasting regressions, we use the first difference of the stochastic trend component.

538

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Hodrick and Prescott (1997) recommend using the Kalman filter procedure to identify
the unobservable stochastic trend component. We employ the following state-space
representation:
Measurement equation



Gt
+ Ct .
FRt = [1, 0]
Gt1

(3)

Transition equation


Gt
Gt1

2 1
=
1 0




Gt1
.
Gt2

(4)

A nice property of the AR(2) process of Gt is that it allows the cyclical component
to be periodic with a peak in its spectral density function, which implies a positive
short-run autocorrelation (momentum) and a negative long-run autocorrelation (mean
reversion).7 Based on Equations (3) and (4), we employ the Kalman filter procedure
to estimate the stochastic trend and cyclical components.

2.2. Models of financial ratios


Campbell and Shiller (1987) develop the following log-linear dividend-price
ratio model:

(5)
j rt+j +1 dt+j +1 .
t = dt pt = Et
j =0

Equation (5) states that the spread, the log dividend-price ratio, is an expected
discounted value of all future returns and dividend growth rates discounted at the
discount rate . That is, the log dividend-price ratio is an expected discounted value
of all future one-period growth-adjusted discount rate, rt+j dt+j . As such, the
log dividend-price ratio provides the optimal forecast of the discounted value of all
future returns or future dividend growth rates, or both.
Lintner (1956), Marsh and Merton (1986), and Lee (1996a, 1996b, 1998) suggest
that dividends are associated with permanent earnings (i.e., the permanent earnings
hypothesis of dividends).8 Similarly, Pindyck and Rotemberg (1993) express stock
price as a present value of future earnings. Combining Pindyck and Rotembergs
(1993) present value model of earnings and Campbell and Shillers (1988a, 1988b)
dynamic present value model, we obtain
7 In macroeconomics studies, Harvey (1985) and Clark (1987), among others, decompose real GDP as
AR(2) process.
8

For the theoretical modeling of the idea of permanent earnings, see Lee (1996a, 1996b, 1998).

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

et pt = Et

j rt+j +1 et+j +1 ,

539

(6)

j =0

where et denotes log earnings. Equation (6) states that the log earnings-price ratio
provides the optimal forecast of the discounted value of all future returns or future
earnings growth rates, or both.
Given unstable dividend policy and controversies about the appropriateness of
using dividends as a proxy for cash flows, Vuolteenaho (2000, 2002) proposes an
alternative, accounting-based, approximate present value model, the log-linear bookto-market ratio model

j rt+j +1 art+j +1 + kt ,
(7)
bt pt = Et
j =0

where art denotes accounting returns, and kt is an approximation error. Equation (7)
states that the log book-to-market ratio is an infinite discounted sum of expected
future stock returns and profitability. Analogous to the Campbell-Shiller model, the
book-to-market ratio provides the optimal forecast of the discounted value of all
future returns or future accounting returns, or both.
In the following discussions, the choice of financial ratios (dividend-price ratio,
earnings-price ratio, and book-to-market ratio) and the prediction of returns and
fundamentals are based on Equations (5), (6), and (7).

2.3. Predictive regressions


To examine whether the decomposition of financial ratios helps predict stock
returns and growth in fundamentals, we consider the following regressions:

yt,h = + FRt + t,t+h ,

(8a)

yt,h = + Ct + t,t+h ,

(8b)

yt,h = + Gt + t,t+h ,

(8c)

yt,h = + Ct + Gt + t,t+h ,

(8d)

here yt,h denotes j =1 (rt+j rf,t+j ) for return regressions or j =1 (Ft+j ) for fundamentals regressions. FRt denotes a financial ratio, which can be DPt (=dt pt ),
EPt (=et pt ), or BPt (=bt pt ). Ct and Gt represent the cyclical and stochastic
trend components of financial ratios, respectively. Ft denotes fundamentals such as
dividend growth, earnings growth and accounting returns. We choose fundamentals
to match financial ratios based on the present value models (Equations [5], [6], and
[7]). For example, dividend growth is used as fundamentals when the predictor is
dividend-price ratio.

540

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

2.4. Econometric issues


Our forecast regression can be subject to a finite sample bias, although the
persistence of the cyclical and stochastic trend components is lower than that of
financial ratios. As documented by Kendall (1954), Stambaugh (1999), and Ferson,
Sarkissian and Simin (2003), the finite sample bias in the forecast regressions can be
substantial and misleading. Long-horizon returns may have potential autocorrelation
and heteroskedasticity. We address this issue by applying a wild bootstrap procedure.
Our bootstrap follows Goncalves and Kilian (2004) and imposes the null of no
predictability in calculating the critical values. We posit that the data are generated
by the following system under the null hypothesis of no predictability:9
rt+1 = + 1t+1 ,
xt+1 = 0 + xt + 2t+1 .
+ ut+1 (under H0 ) where ut+1 = t et+1 while e is the
We resample rt+1 =
residual from the above regression, is any random distribution with zero mean
and variance 1, and is the OLS estimator. Calculate the t-statistics T from the
regression rt+1 = + xt + ut+1 .
. The bootstrap
Repeat this 10,000 times to have bootstrap distribution {T }10000
1
that
is
greater
than T in
critical values can be obtained as the proportion of {T }10000
1
absolute value (in the two-tailed test).
The stochastic trend component reflects the long-run relation and requires a
longer sample to estimate the component. However, there is a tradeoff. Similar to the
construction of CAY in Lettau and Ludvigson (2001), our decomposition uses full
sample and the OOS forecast is subject to the look-ahead bias. We address this issue
by using the moving average of the most recent 25 years financial ratios as a proxy
for the stochastic trend component.10 Then, we use the difference between financial
ratio and the stochastic trend component as a proxy for the cyclical component. By
doing so, we try to circumvent the look-ahead bias as much as we can and show
whether stochastic trend and cyclical components are better able to predict stock
returns and fundamentals.

3. Data
We use stock prices (P), dividends (D, four-quarter moving sum of dividends),
and earnings (E, four-quarter moving sum of earnings) of the Standard and Poors
(S&P) 500 index from 1926:Q1 to 2008:Q4. Following suggestions of Graham and
Dodd (1934) and Campbell and Shiller (2005), we employ a ten-year moving average to smooth earnings to circumvent the problems associated with seasonality and
9

We also consider AR(1) as the null hypothesis and the bootstrap p-values are similar.

10

Fama (2006) uses a similar approach in calculating the long-term expected spot rate.

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

541

earnings manipulation. Since the book value of the S&P 500 index for such a long
period is not available, we rely on the clean surplus accounting relation to generate
book value (Bt ) and assume that the book and market values are identical at the
beginning.11 The risk-free rates (Rf ) are measured as one-month T-bill rates from
CRSP. The CAY is from Lettau and Ludvigson (2001). Inflation rates are based on
the Consumer Price Index from the Bureau of Labor Statistics.
We construct three financial ratios: log dividend-price ratio (DP), log earningsprice ratio (EP), and log book-to-market ratio (BP). All the ratios are measured as the
difference between the log of fundamentals and log of prices.12 Based on Equations
(5), (6), and (7), we choose fundamentals DD, EE, and EB, which are constructed
as the difference in log dividends, the difference in log earnings, and the difference
between log earnings and log book value, respectively.
The decomposition of financial ratios into the two components requires the
selection of the smoothing parameter . We follow the suggestion by Hodrick and
Prescott (1997) in selecting . Assuming that the cyclical components and the second
differences of the stochastic trend components are identically and independently
distributed normal variables with means zero and variances 12 and 22 , they suggest
= 12 /22 . We set = 1,600 as an initial value to solve the programming problem
in Equation (2). Then, we estimate variances 12 and 22 . We then choose as the
ratio of the variances 12 /22 in Hodrick and Prescott (1997) decomposition.13
In Table 1, we report summary statistics of the three financial ratios (DP, EP, and
BP), their cyclical components (CDP, CEP , and CBP ), their stochastic trend components
(GDP, GEP, and GBP ), CAY, relative T-bill rates (RTB), and log excess returns (R).
In Panel A, we report mean, standard deviation, autocorrelation, and unit-root tests
based on the augmented Dickey-Fuller test and the Philips-Perron test. We find that
R, CAY, and RTB are stationary, while all financial ratios are highly persistent. The
first-order autocorrelation of each financial ratio is greater than 0.96, and the null
hypothesis of unit root in DP and BP cannot be rejected by either the augmented
Dickey-Fuller test or the Philips-Perron test.14 Furthermore, Panel A in Figure 1
shows a downward trend for all three financial ratios.
In Panel B, we report a correlation matrix between variables. It is interesting
to note that all financial ratios are positively and closely correlated, but they are

11

The clean surplus accounting relation requires that all gains and losses affecting book value also are
included in earnings; that is, the change in book value is equal to earnings minus dividends. This approach
has been used in both accounting and finance research (see, Vuolteenaho, 2002; Jiang and Lee, 2006,
2007).

12

See Campbell and Shiller (2005).

For the robustness of our results, we also use = 1,600 to decompose the financial ratios. The results
are very similar. We do not report the result to save space.

13

14

This evidence is consistent with the finding of Campbell and Shiller (1987), Froot and Obstfeld (1991),
Jiang and Lee (2007), among others. The null hypothesis of unit root in EP is marginally rejected.

542

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Table 1
Summary statistics
This table reports summary statistics for the financial ratios (log dividend-price ratio (DP), log earningsprice ratio (EP), and log book-to-market ratio (BP)) and their cyclical components (CDP, CEP , and CBP )
and stochastic trend component (GDP, GEP, and GBP ), consumption-wealth ratio (CAY), relative T-bill
rates (RTB), and log excess returns (R). The is the first-order autocorrelation. ADF and PP denote the
augmented Dickey-Fuller test and the Philips-Perron test with four lags. The critical values for ADF and
PP are: 1%, 3.453; 5%, 2.871; and 10%, 2.572, respectively.
Panel A: Summary statistics and unit-root tests for return and ratios
Series

Sample period

Mean

R
DP
EP
BP
CAY
RTB

1926:Q12008:Q4
1926:Q12008:Q4
1926:Q12008:Q4
1926:Q12008:Q4
1951:Q22008:Q4
1926:Q12008:Q4

0.014
3.325
2.912
0.349
0.000
0.000

Std
0.099
0.455
0.379
0.468
0.014
0.002

ADF

0.059
0.974
0.965
0.977
0.882
0.714

8.161
2.558
3.000
2.335
2.732
7.221

PP
19.018
2.418
2.751
2.147
3.513
7.475

Panel B: Correlation matrix for returns and ratios


R
DP
EP
BP
CAY
RTB

DP

EP

BP

1.00

0.044
1.00

0.097
0.952
1.00

0.094
0.963
0.992
1.00

CAY
0.006
0.162
0.086
0.064
1.00

RTB
0.175
0.033
0.005
0.006
0.161
1.00

Panel C: Summary statistics and unit-root tests for cyclical and trend components
Series
CDP
CEP
CBP
GDP
GEP
GBP

Sample period
1926:Q12008:Q4
1926:Q12008:Q4
1926:Q12008:Q4
1926:Q12008:Q4
1926:Q12008:Q4
1926:Q12008:Q4

Mean
0.000
0.000
0.000
3.324
2.912
0.349

Std
0.174
0.169
0.164
0.408
0.326
0.429

0.827
0.827
0.818
1.000
0.999
1.000

ADF
7.438
7.236
7.198
2.030
3.097
3.204

PP
6.146
6.198
6.231
0.027
0.455
0.337

Panel D: Correlation matrix for returns and cyclical components


R
CDP
CEP
CBP
CAY
RTB

CDP

CEP

CBP

1.00

0.356
1.000

0.353
0.943
1.000

0.355
0.952
0.995
1.000

CAY
0.006
0.381
0.371
0.387
1.000

RTB
0.175
0.134
0.189
0.176
0.161
1.000

Panel E: Correlation matrix for returns and trend components


R
GDP
GEP
GBP
CAY
RTB
, ,

GDP

GEP

GBP

1.00

0.234
1.000

0.227
0.922
1.000

0.241
0.944
0.970
1.000

CAY
0.006
0.587
0.675
0.608
1.000

indicate statistical significance at the 0.01, 0.05 and 0.10 level, respectively.

RTB
0.175
0.059
0.059
0.042
0.161
1.000

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

543

weakly and negatively correlated with excess returns.15 In Panel C, we report summary statistics and unit-root tests for the cyclical and stochastic trend components.
Not surprisingly, autocorrelations of the cyclical components of financial ratios are
lower than those of financial ratios. The null hypothesis of unit root in the cyclical
components is rejected at conventional levels by both the augmented Dickey-Fuller
test and the Philips-Perron test. The low autocorrelation in the cyclical components
of financial ratios is very attractive in prediction regressions, since small-sample bias
in predictive regression is primarily due to highly persistent predictors. In contrast,
autocorrelations of the stochastic trend components of financial ratios are close to
one and the null hypothesis of unit root in the trend components is not rejected at
conventional levels by the Philips-Perron test.
In Panel D, we report a correlation matrix between excess returns (or CAY, RTB)
and cyclical components of financial ratios. As in Panel B, the cyclical components of
financial ratios are closely and positively correlated with each other. In contrast, the
cyclical components also are highly and negatively correlated with excess returns.
The correlations between the cyclical components of DP, EP, and BP and excess
returns are 0.36, 0.35, and 0.36, respectively. They are much stronger in their
absolute value than the correlations between excess returns and both CAY and RTB
(0.006 and 0.175, respectively). Panel B in Figure 1 shows that there is no trend
for all three cyclical components of financial ratios. In sum, the cyclical components
of financial ratios are less persistent and more closely correlated with excess returns.
In Panel E, we find that the stochastic trend components of financial ratios also
are associated with excess returns: the correlation between excess returns and the
stochastic trend components of dividend-price ratio, earning-price ratio and bookto-market are 0.23, 0.23, and 0.24, respectively. The absolute values of these
correlations also are much higher than the correlations between excess returns and
both CAY and RTB.

4. Long-horizon prediction
We examine three issues in this section: (1) Are excess returns or fundamentals
predictable? (2) Between the cyclical component and the stochastic trend component,
which component predicts excess returns and fundamentals growth? (3) What is the
mechanism (channel) of the prediction, if there is any?

4.1. Prediction of returns


Theoretical models in Equations (5), (6), and (7) imply that financial ratios (DP,
EP, and BP) predict either stock returns or fundamentals. Here, we investigate em-

15

This is consistent with the findings in the return prediction research that CAY and RTB have more
predictive power than financial ratios (see Lettau and Ludvigson, 2001, 2005; Ang and Bekaert, 2007).

544

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

pirically whether financial ratios predict excess returns in various horizons. First, we
decompose financial ratios (DP, EP, and BP) into a cyclical component and a stochastic trend component using the Kalman filter method recommended by Hodrick and
Prescott (1997). In Table 2, we report the estimation results of univariate prediction
regressions using each component separately. Since the stochastic trend is highly
persistent, we use the first difference (dG) of the stochastic trend in the regression.
We report forecast regression coefficient, Newey-West corrected standard errors in
parentheses, bootstrap p-value in squared brackets, and adjusted R2 statistics in curly
brackets. The sample period is from the first quarter of 1926 to the fourth quarter of
2008.
We find that each component of financial ratios predicts future excess returns
significantly.16 The cyclical components predict future excess returns with a positive coefficient. The adjusted R2 s increase monotonically as the prediction horizon
increases and reach the highest level at the prediction horizon of 16 quarters for CDP ,
and 12 quarters for CEP and CBP , respectively. The adjusted R2 s are 30%, 53%, and
52% for CDP , CEP , and CBP , respectively, at the peak. The coefficients and adjusted
R2 s in regressions of future returns on the cyclical components increase in proportion
to horizons of eight quarters (two years). This implies that the variation in the cyclical
components contains a similar variation of one-period expected returns up to eight
quarters. The coefficients and adjusted R2 s in the regressions increase at a decaying
rate for longer return horizons, which suggests that the cyclical component contains
less variation of one-period expected return in more distant horizons. This pattern
suggests that the cyclical component is associated with the expected return with a
mean-reverting component around a business-cycle-length.17
The stochastic trend components predict future excess returns with a negative
coefficient. The bootstrap p-values and adjusted R2 s show that the stochastic trend
components are able to forecast future stock returns, particularly in long horizons.
It suggests that the stochastic trend components tend to forecast long-horizon future
stock returns. To our knowledge, this is new evidence. Previous research (e.g., Campbell and Shiller, 1988a, 1988b; Lettau and Van Nieuwerburgh, 2008) assumes that
the long-run stochastic trend is constant and tends to ignore its predictive power. The
Hodrick and Prescott (1997) filter allows us to generate the time-varying long-run
stochastic trend component. The adjusted R2 s are 16%, 17%, and 25% at the 16quarter horizon for the stochastic trend components of DP, EP, and BP, respectively.
A similar observation is made when we run multivariate forecast regression
using both cyclical and stochastic trend components as predictors in Table 3. The
adjusted R2 s of two components for DP, EP, and BP at the one-quarter horizon are
5%, 13%, and 13%, respectively, which are much higher than those of raw financial

16
17

The exception is the one-quarter-ahead forecast for CDP .

Fama and French (1988) and Summers (1986) show that the expected return is highly autocorrelated
with a long-run slow mean reversion.

545

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564


Table 2
Univariate forecast of quarterly stock returns: Decomposition regressions (1926:Q12008:Q4)

This table reports univariate estimates from regressions of excess stock returns on the lagged decomposed
financial ratios. The dependent variable is the sum of H-period log excess returns on the S&P 500 index
over a three-month T-bill rate. Financial ratios are log dividend-price ratio (DP), log earnings-price ratio
(EP), and log book-to-market (BP), respectively. They all are from the S&P 500 index. C and dG are the
cyclical and stochastic growth rate (the first difference of trend component) components of financial ratios.
The table reports OLS estimates of regressors, Newey-West corrected standard errors (in parentheses),
bootstrapping p-value (in brackets) and adjusted R2 (in curly brackets). The sample period is from 1926:Q1
to 2008:Q4.
H

CDP

CEP

CBP

dGDP

dGEP

dGBP

0.103
(0.073)
[0.154]
{0.029}

0.183
(0.070)
[0.014]
{0.093}

0.188
(0.070)
[0.010]
{0.092}

1.424
(0.403)
[0.000]
{0.025}

1.595
(0.626)
[0.014]
{0.033}

1.500
(0.569)
[0.008]
{0.036}

0.188
(0.069)
[0.007]
{0.055}

0.333
(0.061)
[0.000]
{0.168}

0.340
(0.062)
[0.000]
{0.164}

2.708
(0.558)
[0.000]
{0.051}

3.055
(0.762)
[0.000]
{0.068}

2.888
(0.688)
[0.000]
{0.074}

0.312
(0.069)
[0.000]
{0.097}

0.513
(0.060)
[0.000]
{0.254}

0.524
(0.060)
[0.000]
{0.248}

3.921
(0.702)
[0.000]
{0.068}

4.411
(0.984)
[0.000]
{0.090}

4.194
(0.876)
[0.000]
{0.099}

0.464
(0.085)
[0.000]
{0.152}

0.713
(0.062)
[0.000]
{0.345}

0.729
(0.062)
[0.000]
{0.338}

5.106
(0.840)
[0.000]
{0.080}

5.686
(1.182)
[0.000]
{0.104}

5.441
(1.060)
[0.000]
{0.117}

0.850
(0.082)
[0.000]
{0.262}

1.223
(0.110)
[0.000]
{0.517}

1.253
(0.107)
[0.000]
{0.509}

9.057
(1.311)
[0.000]
{0.125}

9.649
(1.693)
[0.000]
{0.152}

9.568
(1.508)
[0.000]
{0.183}

12

1.053
(0.131)
[0.000]
{0.292}

1.455
(0.135)
[0.000]
{0.530}

1.495
(0.136)
[0.000]
{0.524}

11.956
(1.589)
[0.000]
{0.153}

12.137
(1.803)
[0.000]
{0.172}

12.584
(1.656)
[0.000]
{0.226}

16

1.169
(0.113)
[0.000]
{0.295}

1.470
(0.097)
[0.000]
{0.443}

1.525
(0.099)
[0.000]
{0.447}

13.558
(1.782)
[0.000]
{0.156}

13.537
(1.950)
[0.000]
{0.173}

14.686
(1.829)
[0.000]
{0.249}

20

1.229
(0.102)
[0.000]
{0.285}

1.334
(0.112)
[0.000]
{0.318}

1.406
(0.112)
[0.000]
{0.331}

13.702
(2.033)
[0.000]
{0.135}

14.257
(2.097)
[0.000]
{0.165}

16.041
(1.962)
[0.000]
{0.255}

546

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Table 3
Multivariate forecast of quarterly stock returns: Decomposition regressions (1926:Q12008:Q4)
This table reports multivariate estimates from regressions of excess stock returns on the lagged decomposed
financial ratios. The dependent variable is the sum of H-period log excess returns on the S&P 500 index
over a three-month T-bill rate. Financial ratios are log dividend-price ratio (DP), log earnings-price ratio
(EP), and log book-to-market (BP), respectively. They are all from the S&P 500 index. C and dG are the
cyclical and stochastic growth rate (the first difference of trend component) components of financial ratios.
The table reports OLS estimates of regressors, Newey-West corrected standard errors (in parentheses),
and bootstrapping p-value (in brackets). Adjusted R2 also is reported. The sample period is from 1926:Q1
to 2008:Q4.
H

CDP

dGCP

R2

CEP

dGEP

R2

CBP

dGBP

R2

0.103
(0.073)
[0.151]

1.433
(0.397)
[0.000]

0.054

0.185
(0.071)
[0.013]

1.629
(0.599)
[0.005]

0.128

0.189
(0.072)
[0.013]

1.521
(0.539)
[0.005]

0.129

0.189
(0.068)
[0.009]

2.714
(0.542)
[0.000]

0.106

0.335
(0.061)
[0.000]

3.111
(0.723)
[0.000]

0.239

0.341
(0.062)
[0.000]

2.918
(0.641)
[0.000]

0.241

0.313
(0.066)
[0.000]

3.927
(0.668)
[0.000]

0.166

0.517
(0.056)
[0.000]

4.502
(0.884)
[0.000]

0.349

0.527
(0.056)
[0.000]

4.243
(0.771)
[0.000]

0.351

0.464
(0.083)
[0.000]

5.106
(0.780)
[0.000]

0.233

0.718
(0.062)
[0.000]

5.812
(0.986)
[0.000]

0.455

0.732
(0.062)
[0.000]

5.509
(0.859)
[0.000]

0.459

0.858
(0.071)
[0.000]

9.227
(1.073)
[0.000]

0.394

1.239
(0.077)
[0.000]

10.074
(0.955)
[0.000]

0.686

1.268
(0.072)
[0.000]

9.879
(0.833)
[0.000]

0.706

12

1.066
(0.113)
[0.000]

12.224
(1.212)
[0.000]

0.453

1.480
(0.086)
[0.000]

12.750
(0.840)
[0.000]

0.722

1.519
(0.085)
[0.000]

13.053
(0.772)
[0.000]

0.770

16

1.183
(0.098)
[0.000]

13.861
(1.423)
[0.000]

0.460

1.498
(0.070)
[0.000]

14.184
(0.921)
[0.000]

0.636

1.555
(0.076)
[0.000]

15.195
(0.911)
[0.000]

0.716

20

1.242
(0.088)
[0.000]

14.005
(1.661)
[0.000]

0.427

1.361
(0.084)
[0.000]

14.804
(1.287)
[0.000]

0.498

1.436
(0.081)
[0.000]

16.480
(1.262)
[0.000]

0.603

ratios.18 The adjusted R2 s increase monotonically as the prediction horizon increases


and reach the highest level at the horizon of around 12 quarters. The adjusted R2 s
fall as the prediction horizon increases beyond 12 or 16 quarters. These adjusted R2 s
18 In an unreported table, we forecast regression market excess return on lagged financial ratios separately. For one-quarter-ahead forecast regression, the adjusted R2 s are 1%, 2%, and 2% DP, EP, and BP,
respectively.

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

547

are substantially high relative to the findings in previous studies.19 We believe the
higher adjusted R2 s are due to the decomposition of financial ratios; each component
may capture different component of expected returns, local mean reversion and longrun slow mean reversion. For financial ratio alone (without decomposition), the two
effects may cancel out partially, resulting in a relative low adjusted R2 . This pattern
of the adjusted R2 is interesting. It is worthwhile to note that the predictive power
of the cyclical and stochastic trend components of financial ratios is at its best over
the horizon of three to four years, which is similar to the duration of business cycles.
Our evidence suggests that the expected returns are highly autocorrelated with a slow
mean reversion.
Lettau and Ludvigson (2001, 2005) demonstrate that CAY has a strong predictive
power for excess returns at business cycle frequencies, showing direct linkage that
risk premia vary countercyclically.20 Ang and Bekaert (2007) provide evidence that
the dividend-price ratio has little predictive power for future excess returns at long
horizons. At short horizons, the dividend-price ratio is able to predict future excess
returns only with short rates. They claim that the strongest predictive power comes
from the short rate rather than from the dividend yield.
A natural question is whether each variable has independent information about
future excess returns. To address this question, in Table 4 we report the results of the
comparisons using the two components of financial ratios, CAY and the relative T-bill
rate for the sample period from 1952:Q1 to 2008:Q4 (CAY is available from 1952:Q1).
The cyclical component (C) and the stochastic trend component (dG) exhibit a similar
predictive pattern as in Table 2. For example, at the one-quarter horizon, the cyclical
components of DP, EP, and BP predict excess returns significantly. Both NeweyWest corrected standard errors and bootstrap p-values are significant at the 1% level.
The adjusted R2 s are about 8% for the cyclical components and around 5% for the
stochastic trend components in one-quarter horizon. The predictive power reaches its
peak at eight (16-)-quarter horizon for the cyclical (stochastic trend) components of
financial ratios.
We also confirm the findings of Lettau and Ludvigson (2001, 2005) and Ang
and Bekaert (2007) that CAY predicts an increase in excess returns, and RTB predicts
a decline in excess returns. The adjusted R2 s of CAY (RTB) at prediction horizons of
1, 2, 3, 4, 8, 12, 16, and 20 quarters are 4% (2%), 8% (2%), 11% (3%), 14% (4%),
23% (1%), 30% (1%), 32% (1%), and 31% (1%), respectively. It is interesting to note
that the adjusted R2 s at the one-quarter horizon are 4% and 2% for CAY and RTB,
respectively, which are much higher than those in the raw financial ratios but much

19

Lettau and Van Nieuwerburgh (2008) report the adjusted R2 of 5% (13%) for prediction regression of
excess returns on dividend-price ratio (with one break) using annual data from 1927 to 2004.

20

Idiosyncratic volatility also has been found associated with expected returns (see Goyal and Santa-Clara,
2003; Ang, Hodrick, Xing and Zhang, 2006; Jiang and Lee, 2006).

548

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Table 4
Univariate forecast of quarterly stock returns: Decomposition regressions (1952:Q12008:Q4)
This table reports univariate estimates from regressions of excess stock returns on the lagged decomposed
financial ratios, lagged CAY and lagged RTB. The dependent variable is the sum of H-period log excess
returns on the S&P 500 index over a three-month T-bill rate. Financial ratios are log dividend-price ratio
(DP), log earnings-price ratio (EP), and log book-to-market (BP), respectively. They all are from the S&P
500 index. C and dG are the cyclical and growth rate (the first difference of trend component) components
of financial ratios. CAY is the consumption-wealth ratio, RTB is the relative T-bill rates. The table reports
OLS estimates of regressors, Newey-West corrected standard errors (in parentheses), bootstrapping p-value
(in brackets) and adjusted R2 (in curly brackets). The sample period is from 1952:Q1 to 2008:Q4.
H

CDP

CEP

CBP

dGDP

dGEP

dGBP

CAY

RTB

0.194
(0.049)
[0.000]
{0.077}

0.193
(0.048)
[0.000]
{0.077}

0.192
(0.048)
[0.000]
{0.076}

1.301
(0.374)
[0.001]
{0.051}

1.397
(0.449)
[0.002]
{0.046}

1.392
(0.419)
[0.001]
{0.051}

1.027
(0.266)
[0.000]
{0.035}

3.770
(1.834)
[0.044]
{0.016}

0.444
(0.056)
[0.000]
{0.190}

0.432
(0.058)
[0.000]
{0.183}

0.430
(0.058)
[0.000]
{0.181}

2.404
(0.545)
[0.000]
{0.080}

2.622
(0.652)
[0.000]
{0.076}

2.614
(0.615)
[0.000]
{0.085}

2.217
(0.414)
[0.000]
{0.081}

5.485
(2.949)
[0.067]
{0.016}

0.654
(0.058)
[0.000]
{0.278}

0.634
(0.063)
[0.000]
{0.264}

0.630
(0.064)
[0.000]
{0.262}

3.421
(0.636)
[0.000]
{0.108}

3.769
(0.765)
[0.000]
{0.105}

3.758
(0.730)
[0.000]
{0.118}

3.110
(0.513)
[0.000]
{0.107}

9.359
(3.724)
[0.012]
{0.034}

0.865
(0.070)
[0.000]
{0.357}

0.842
(0.077)
[0.000]
{0.343}

0.837
(0.078)
[0.000]
{0.340}

4.440
(0.748)
[0.000]
{0.132}

4.908
(0.876)
[0.000]
{0.130}

4.905
(0.848)
[0.000]
{0.147}

4.201
(0.634)
[0.000]
{0.144}

12.321
(4.504)
[0.009]
{0.043}

1.314
(0.112)
[0.000]
{0.440}

1.304
(0.112)
[0.000]
{0.437}

1.304
(0.113)
[0.000]
{0.438}

8.093
(1.012)
[0.000]
{0.226}

8.967
(1.157)
[0.000]
{0.230}

9.123
(1.140)
[0.000]
{0.267}

7.423
(0.861)
[0.000]
{0.235}

9.456
(4.576)
[0.039]
{0.011}

12

1.378
(0.136)
[0.000]
{0.360}

1.394
(0.136)
[0.000]
{0.370}

1.402
(0.136)
[0.000]
{0.376}

10.938
(1.157)
[0.000]
{0.292}

12.291
(1.193)
[0.000]
{0.315}

12.524
(1.181)
[0.000]
{0.363}

9.967
(0.982)
[0.000]
{0.297}

9.151
(4.491)
[0.041]
{0.006}

16

1.272
(0.157)
[0.000]
{0.258}

1.333
(0.148)
[0.000]
{0.285}

1.353
(0.149)
[0.000]
{0.295}

12.821
(1.363)
[0.000]
{0.322}

14.837
(1.236)
[0.000]
{0.377}

15.031
(1.200)
[0.000]
{0.426}

11.697
(1.035)
[0.000]
{0.324}

10.205
(5.720)
[0.071]
{0.007}

20

1.502
(0.171)
[0.000]
{0.289}

1.510
(0.168)
[0.000]
{0.292}

1.532
(0.169)
[0.000]
{0.303}

14.246
(1.658)
[0.000]
{0.304}

16.987
(1.437)
[0.000]
{0.382}

17.235
(1.379)
[0.000]
{0.431}

13.080
(1.150)
[0.000]
{0.314}

14.100
(6.803)
[0.042]
{0.012}

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

549

lower than (similar to) those of cyclical (stochastic trend) components of financial
ratios.
It is also interesting to note that when we run the multivariate regression of
excess returns on cyclical component, stochastic trend component, CAY, and RTB
in Table 5, the coefficient of CAY at the one-quarter horizon decreases from 1.03 to
0.49, and it becomes no longer significant, while the coefficients of both cyclical
and stochastic trend components of financial ratios change little, and their t-statistics
remain highly significant. The coefficient of RTB also decreases in its absolute value,
and becomes insignificant. At long horizons, we find the coefficients of CAY to be
unstable. This implies that the predictive power of CAY and RTB is partially captured
by the cyclical and stochastic trend components of financial ratios.
In sum, in the subsample period from 1952:Q1 to 2008:Q4, the univariate
forecast regression shows that at horizons from one to 12 quarters, the adjusted R2 s are
higher using cyclical components as predictors. At horizons beyond 12 quarters, the
adjusted R2 s are higher using stochastic trend components as predictors. This evidence
indicates that the cyclical components of financial ratios dominate the predictive
power in relative short horizons while the stochastic trend components of financial
ratios dominate the predictive power in the relative long horizons. In multivariate
forecast regressions, we compare the ratio of total variance of return explained by
each component, which also confirms that the cyclical (stochastic trend) components
have more predictive power in relative short (long) horizons.21 Furthermore, the
components of financial ratios outperform CAY and RTB in all horizons.

4.2. Prediction of fundamentals


The present value models in Equations (5), (6), and (7) also imply that financial
ratios (DP, EP, and BP) may predict fundamentals. However, empirical evidence suggests that financial ratios have little predictive power for fundaments (see Campbell,
1991; Lettau and Ludvigson, 2001, 2005; Campbell and Shiller, 2005).22 Cochrane
(2001, p. 398) points out: It is nonetheless an uncomfortable fact that almost all
variation in price/dividend ratios is due to variation in expected excess returns. How
nice it would be if high prices reflected expectations of higher future cash flows.
Alas, that seems not to be the case.
Here, we investigate whether the decomposed financial ratios predict such fundamentals as log dividend growth (DD), log earnings growth (EE), and log accounting
returns (EB) in long horizons. Again, we choose these fundamentals based on the

21
22

We thank the referees suggestion in this direction.

Ang and Bekaert (2007) find the marginal predictability of dividend growth using dividend-price ratio.
Using annual data, Lettau and Ludvigson (2005) find that dividend-price ratio significantly predicts
dividend growth but with the wrong (positive) sign. Ang and Bekaert (2007) find that earnings-price ratio
predicts future cash flows.

0.207
(0.056)
[0.000]
0.465
(0.064)
[0.000]
0.693
(0.062)
[0.000]
0.901
(0.061)
[0.000]
1.349
(0.106)
[0.000]
1.315
(0.108)
[0.000]
1.104
(0.125)
[0.000]
1.335
(0.134)
[0.000]

20

16

12

CDP

CAY

0.493
(0.484)
[0.309]
0.740
(0.540)
[0.171]
1.296
(0.643)
[0.047]
1.365
(0.681)
[0.049]
0.739
(0.842)
[0.392]
1.403
(0.965)
[0.155]
3.503
(1.124)
[0.002]
3.292
(1.152)
[0.005]

dGDP

1.554
(0.562)
[0.005]
2.802
(0.649)
[0.000]
4.149
(0.760)
[0.000]
5.157
(0.797)
[0.000]
8.522
(0.882)
[0.000]
9.743
(1.073)
[0.000]
10.302
(1.384)
[0.000]
12.045
(1.538)
[0.000]

2.728
(1.686)
[0.110]
3.033
(2.588)
[0.240]
5.537
(3.046)
[0.071]
6.787
(3.457)
[0.055]
1.045
(2.823)
[0.710]
2.487
(2.729)
[0.362]
0.656
(3.161)
[0.837]
2.105
(3.816)
[0.589]

RTB

0.597

0.592

0.656

0.662

0.507

0.405

0.273

0.131

R2
0.218
(0.054)
[0.000]
0.474
(0.066)
[0.000]
0.702
(0.071)
[0.000]
0.917
(0.070)
[0.000]
1.469
(0.096)
[0.000]
1.543
(0.098)
[0.000]
1.444
(0.104)
[0.000]
1.649
(0.105)
[0.000]

CEP
1.957
(0.749)
[0.009]
3.523
(0.821)
[0.000]
5.313
(0.948)
[0.000]
6.654
(0.959)
[0.000]
11.445
(1.132)
[0.000]
13.816
(1.131)
[0.000]
15.540
(1.350)
[0.000]
18.489
(1.311)
[0.000]

dGEP
0.761
(0.573)
[0.186]
1.186
(0.605)
[0.055]
1.991
(0.718)
[0.007]
2.263
(0.728)
[0.002]
2.788
(0.961)
[0.005]
1.726
(1.011)
[0.087]
0.826
(1.154)
[0.476]
1.910
(1.222)
[0.121]

CAY
2.380
(1.674)
[0.160]
2.222
(2.576)
[0.387]
4.318
(3.024)
[0.158]
5.078
(3.485)
[0.145]
3.880
(2.814)
[0.161]
5.702
(2.719)
[0.039]
4.168
(2.834)
[0.143]
1.702
(3.285)
[0.600]

RTB

0.702

0.705

0.738

0.697

0.491

0.389

0.263

0.129

R2
0.213
(0.055)
[0.000]
0.468
(0.066)
[0.000]
0.695
(0.069)
[0.000]
0.911
(0.067)
[0.000]
1.451
(0.087)
[0.000]
1.515
(0.085)
[0.000]
1.417
(0.097)
[0.000]
1.621
(0.095)
[0.000]

CBP
1.833
(0.644)
[0.005]
3.372
(0.736)
[0.000]
5.094
(0.842)
[0.000]
6.456
(0.849)
[0.000]
11.110
(0.869)
[0.000]
13.414
(0.842)
[0.000]
15.106
(1.017)
[0.000]
17.981
(0.993)
[0.000]

dGBP
0.678
(0.520)
[0.192]
1.094
(0.571)
[0.055]
1.862
(0.659)
[0.007]
2.154
(0.655)
[0.001]
2.524
(0.789)
[0.001]
1.282
(0.799)
[0.113]
0.200
(0.900)
[0.829]
1.039
(0.990)
[0.292]

CAY

2.576
(1.665)
[0.120]
2.627
(2.552)
[0.305]
4.898
(2.991)
[0.100]
5.811
(3.467)
[0.097]
2.728
(2.814)
[0.335]
4.309
(2.699)
[0.118]
2.579
(2.702)
[0.342]
0.144
(3.184)
[0.963]

RTB

0.751

0.746

0.771

0.727

0.513

0.405

0.273

0.134

R2

This table reports multivariate estimates from regressions of excess stock returns on the lagged decomposed financial ratios, lagged CAY and lagged RTB. The
dependent variable is the sum of H-period log excess returns on the S&P 500 index over a three-month T-bill rate. Financial ratios are log dividend-price ratio
(DP), log earnings-price ratio (EP), and log book-to-market (BP), respectively. They all are from the S&P 500 index. C and dG are the cyclical and growth rate
(the first difference of trend component) components of financial ratios. CAY is the consumption-wealth ratio, RTB is the relative T-bill rates. The table reports
OLS estimates of regressors, Newey-West corrected standard errors (in parentheses), and bootstrapping p-value (in brackets). Adjusted R2 also is reported. The
sample period is from 1952:Q1 to 2008:Q4.

Multivariate forecast of quarterly stock returns: Decomposition regressions (1952:Q12008:Q4)

Table 5

550
X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

551

present value models in Equations (5), (6), and (7). The other reason to examine the
prediction of fundamentals is that it helps to understand the mechanism and driving
forces of the return prediction, which we will discuss later.
We present the univariate estimation results in Table 6. CDP significantly predicts
dividend growth, DD, with a negative coefficient up to 12 quarters. From the onequarter horizon to the 20-quarter horizon, the adjusted R2 s for CDP are 25%, 32%,
33%, 30%, 11%, 4%, 1%, and 0%, respectively. The stochastic trend component of DP
does not seem to have significant predictive power for dividend growth. The cyclical
component of EP (BP) also predicts decreasing earnings growth (accounting returns)
in short horizon while the stochastic trends of EP (BP) tend to predict decreasing
earnings growth (accounting returns) in long horizons.
In a multivariate forecast regression in Table 7, we find that the cyclical (stochastic trend) component of DP predicts decreasing (increasing) dividend growth in
relative short horizons. The adjusted R2 is highest at three quarters for DD. The
predictability is not stable for the cyclical components of EP and BP although they
tend to predict decreasing fundamental growth in short horizons. Both stochastic
trends of EP and BP tend to predict decreasing fundamental growth, particularly in
long horizons. Both univariate and multivariate regressions show that cyclical and
stochastic trend have some predictability of fundamental growths.
Lettau and Ludvigson (2001, 2005) document that there is a common variation
between expected returns and expected dividend growth, which offsets the predictive
power of dividend yield, but not of the CAY. As a robustness check, we investigate the
prediction of fundamentals using the data from 1952:Q1 to 2008:Q4. We also compare
the predictive power using multivariate regressions incorporating the two components
of financial ratios, CAY, and RTB. In Table 8, we find that the cyclical component
of the dividend-price ratio predicts dividend growth with a significantly negative
coefficient up to four quarters. However, the stochastic trend component of the
dividend-price ratio predicts dividend growth with a positive coefficient. The adjusted
R2 s of the stochastic trend component is higher than that of cyclical component and
increases as the horizon increases. Our evidence helps explain why the dividendprice ratio does not predict dividend growth. Lettau and Ludvigson (2005) argue that
the failure of dividend-price ratio to predict dividend growth is due to the positive
correlation between expected return and expected dividend growth, suggesting that
time-varying investment opportunities are poorly captured by dividend-price ratio.
Our evidence suggests that these investment opportunities are possibly captured
separately by each component of dividend-price ratio, and the failure of dividendprice ratio to predict dividend growth is due to the offsetting effect between the
cyclical and stochastic trend components. CAY predicts dividend growth weakly in
univariate regression, while RTB seems to predict dividend growth at short horizons.
When the dependent variable is earnings growth (EE), both the cyclical and
stochastic trend components of the earnings-price ratio predict earnings growth with
a positive coefficient at long horizons. When the dependent variable is accounting
returns (EB), we find that the cyclical component of book-to-market seems to predict

552

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Table 6
Univariate forecast of quarterly fundamentals: Decomposition regressions (1926:Q12008:Q4)
This table reports estimates from regressions of fundamentals on the lagged decomposed financial ratios.
The dependent variable is the sum of H-period log dividend growth (DD), log earnings growth (EE), and
log accounting return (EB). Financial ratios are log dividend-price ratio (DP), log earnings-price ratio
(EP), and log book-to-market (BP), respectively. They all are from the S&P 500 index. C and dG are the
cyclical and stochastic growth rate (the first difference of trend component) components of financial ratios.
The table reports OLS estimates of regressors, Newey-West corrected standard errors (in parentheses),
bootstrapping p-value (in brackets) and adjusted R2 (in curly brackets). The sample period is from 1926:Q1
to 2008:Q4.
Dependent variable: DD
dGDP

Dependent variable: EE

Dependent variable: EB

CDP

CEP

dGEP

CBP

dGBP

0.098
(0.013)
[0.000]
{0.245}

0.169
(0.105)
[0.112]
{0.000}

0.052
(0.025)
[0.041]
{0.006}

1.310
(0.493)
[0.007]
{0.025}

0.035
(0.010)
[0.001]
{0.043}

0.474
(0.107)
[0.000]
{0.051}

0.198
(0.027)
[0.000]
{0.320}

0.307
(0.189)
[0.104]
{0.000}

0.048
(0.048)
[0.300]
{0.000}

2.225
(0.726)
[0.002]
{0.032}

0.068
(0.020)
[0.001]
{0.042}

0.969
(0.215)
[0.000]
{0.055}

0.282
(0.038)
[0.000]
{0.327}

0.404
(0.275)
[0.144]
{0.000}

0.006
(0.069)
[0.930]
{0.003}

3.067
(0.973)
[0.001]
{0.034}

0.098
(0.031)
[0.002]
{0.039}

1.496
(0.323)
[0.000]
{0.059}

0.341
(0.047)
[0.000]
{0.302}

0.458
(0.363)
[0.214]
{0.001}

0.094
(0.092)
[0.291]
{0.002}

3.791
(1.197)
[0.001]
{0.036}

0.120
(0.042)
[0.006]
{0.033}

2.056
(0.433)
[0.000]
{0.064}

0.344
(0.074)
[0.000]
{0.114}

0.272
(0.746)
[0.719]
{0.003}

0.607
(0.154)
[0.000]
{0.094}

5.459
(1.929)
[0.004]
{0.035}

0.129
(0.096)
[0.182]
{0.008}

4.595
(0.873)
[0.000]
{0.087}

12

0.241
(0.100)
[0.021]
{0.035}

0.327
(1.118)
[0.775]
{0.003}

0.946
(0.165)
[0.000]
{0.167}

5.788
(2.417)
[0.015]
{0.027}

0.033
(0.152)
[0.825]
{0.003}

7.545
(1.299)
[0.000]
{0.113}

16

0.157
(0.117)
[0.178]
{0.009}

1.134
(1.414)
[0.426]
{0.001}

0.924
(0.181)
[0.000]
{0.140}

5.465
(2.643)
[0.040]
{0.020}

0.061
(0.201)
[0.761]
{0.002}

10.748
(1.692)
[0.000]
{0.140}

20

0.069
(0.115)
[0.540]
{0.001}

1.965
(1.607)
[0.227]
{0.003}

0.810
(0.174)
[0.000]
{0.104}

4.774
(2.631)
[0.070]
{0.014}

0.119
(0.242)
[0.610]
{0.001}

14.073
(2.053)
[0.000]
{0.165}

553

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564


Table 7

Multivariate forecast of quarterly fundamentals: Decomposition regressions (1926:Q12008:Q4)


This table reports estimates from regressions of fundamentals on the lagged decomposed financial ratios.
The dependent variable is the sum of H-period log dividend growth (DD), log earnings growth (EE), and
log accounting return (EB). Financial ratios are log dividend-price ratio (DP), log earnings-price ratio
(EP), and log book-to-market (BP), respectively. They all are from the S&P 500 index. C and dG are the
cyclical and stochastic growth rate (the first difference of trend component) components of financial ratios.
The table reports OLS estimates of regressors, Newey-West corrected standard errors (in parentheses),
and bootstrapping p-value (in brackets). Adjusted R2 also is reported. The sample period is from 1926:Q1
to 2008:Q4.
Dependent variable: DD

Dependent variable: EE
CEP

dGEP

R2

Dependent variable: EB
CBP

dGBP

R2

CDP

dGCP

R2

0.098
(0.014)
[0.000]

0.178
(0.100)
[0.077]

0.247

0.051
(0.023)
[0.031]

1.300
(0.492)
[0.008]

0.031

0.035
(0.009)
[0.000]

0.470
(0.100)
[0.000]

0.094

0.198
(0.027)
[0.000]

0.313
(0.182)
[0.083]

0.322

0.047
(0.043)
[0.267]

2.218
(0.730)
[0.002]

0.032

0.068
(0.018)
[0.000]

0.963
(0.201)
[0.000]

0.096

0.282
(0.038)
[0.000]

0.410
(0.266)
[0.126]

0.329

0.004
(0.062)
[0.954]

3.066
(0.975)
[0.001]

0.031

0.097
(0.027)
[0.001]

1.487
(0.305)
[0.000]

0.098

0.341
(0.047)
[0.000]

0.458
(0.356)
[0.196]

0.303

0.097
(0.082)
[0.220]

3.808
(1.187)
[0.001]

0.039

0.119
(0.036)
[0.001]

2.044
(0.412)
[0.000]

0.097

0.345
(0.074)
[0.000]

0.340
(0.758)
[0.655]

0.112

0.616
(0.136)
[0.000]

5.671
(1.751)
[0.002]

0.132

0.122
(0.078)
[0.116]

4.565
(0.856)
[0.000]

0.094

12

0.241
(0.099)
[0.019]

0.267
(1.148)
[0.816]

0.032

0.958
(0.143)
[0.000]

6.184
(1.965)
[0.001]

0.199

0.019
(0.120)
[0.879]

7.539
(1.291)
[0.000]

0.110

16

0.156
(0.116)
[0.175]

1.094
(1.445)
[0.449]

0.009

0.935
(0.164)
[0.000]

5.869
(2.127)
[0.006]

0.164

0.082
(0.156)
[0.601]

10.775
(1.682)
[0.000]

0.139

20

0.067
(0.113)
[0.551]

1.949
(1.627)
[0.241]

0.002

0.820
(0.163)
[0.000]

5.104
(2.186)
[0.020]

0.121

0.145
(0.182)
[0.411]

14.118
(2.037)
[0.000]

0.165

accounting returns negatively in short horizons while positively in long horizons.


The stochastic trend component of book-to-market predicts accounting returns with
a positive coefficient, especially in long horizons beyond 16 quarters. In Table 9 with
multivariate forecast regression, we find that cyclical component tends to forecast
decreasing dividend growth (DD) and accounting returns (EB), particularly in short
horizons. The stochastic trend component tends to forecast increasing fundamental

dGDP

0.233
(0.072)
[0.001]
{0.041}

0.494
(0.122)
[0.000]
{0.064}

0.758
(0.164)
[0.000]
{0.079}

1.031
(0.202)
[0.000]
{0.093}

CDP

0.023
(0.009)
[0.010]
{0.024}

0.037
(0.016)
[0.018]
{0.022}

0.048
(0.023)
[0.034]
{0.019}

0.050
(0.029)
[0.080]
{0.012}

0.312
(0.186)
[0.094]
{0.006}

0.269
(0.149)
[0.073]
{0.008}

0.168
(0.109)
[0.123]
{0.005}

0.112
(0.062)
[0.072]
{0.008}

CAY

Dependent variable: DD
RTB

4.805
(0.740)
[0.000]
{0.095}

4.183
(0.571)
[0.000]
{0.116}

3.016
(0.407)
[0.000]
{0.118}

1.644
(0.232)
[0.000]
{0.103}

CEP

0.213
(0.163)
[0.196]
{0.009}

0.043
(0.115)
[0.697]
{0.004}

0.014
(0.073)
[0.842]
{0.004}

0.061
(0.048)
[0.201]
{0.001}

0.978
(1.329)
[0.464]
{0.001}

1.024
(1.141)
[0.382]
{0.001}

0.946
(0.906)
[0.303]
{0.003}

0.777
(0.684)
[0.276]
{0.005}

dGEP

1.038
(0.996)
[0.309]
{0.001}

0.200
(0.691)
[0.772]
{0.004}

0.102
(0.573)
[0.858]
{0.004}

0.437
(0.346)
[0.209]
{0.000}

CAY

Dependent variable: EE

2.248
(4.134)
[0.590]
{0.004}

6.803
(5.420)
[0.209]
{0.008}

10.185
(4.567)
[0.030]
{0.044}

7.392
(2.402)
[0.004]
{0.046}

RTB

CBP

0.120
(0.034)
[0.001]
{0.036}

0.104
(0.027)
[0.000]
{0.047}

0.074
(0.019)
[0.000]
{0.050}

0.040
(0.010)
[0.000]
{0.056}

0.226
(0.366)
[0.547]
{0.003}

0.133
(0.285)
[0.642]
{0.003}

0.060
(0.197)
[0.760]
{0.004}

0.009
(0.105)
[0.933]
{0.004}

dGBP

0.677
(0.329)
[0.043]
{0.018}

0.586
(0.248)
[0.020]
{0.023}

0.396
(0.169)
[0.019]
{0.022}

0.227
(0.085)
[0.008]
{0.028}

CAY

Dependent variable: EB

(Continued)

12.370
(1.512)
[0.000]
{0.273}

10.215
(1.172)
[0.000]
{0.318}

7.223
(0.808)
[0.000]
{0.345}

3.664
(0.417)
[0.000]
{0.329}

RTB

This table reports estimates from regressions of fundamentals on the lagged decomposed financial ratios. The dependent variable is the sum of H-period log
dividend growth (DD), log earnings growth (EE), and log accounting returns (EB). Financial ratios are log dividend-price ratio (DP), log earnings-price ratio (EP),
and log book-to-market ratio (BP), respectively. They all are from the S&P 500 index. C and dG are the cyclical and stochastic growth rate (the first difference
of trend component) components of financial ratios. The table reports OLS estimates of regressors, Newey-West corrected standard errors (in parentheses),
bootstrapping p-value (in brackets) and adjusted R2 (in curly brackets). The sample period is from 1952:Q1 to 2008:Q4.

Univariate forecast of quarterly fundamentals: Decomposition regressions (1952:Q12008:Q4)

Table 8

554
X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

4.489
(0.545)
[0.000]
{0.227}

5.436
(0.602)
[0.000]
{0.262}

0.090
(0.070)
[0.205]
{0.006}

0.055
(0.078)
[0.480]
{0.002}

0.046
(0.086)
[0.593]
{0.003}

12

16

20

3.345
(0.458)
[0.000]
{0.179}

2.175
(0.346)
[0.000]
{0.136}

0.018
(0.052)
[0.725]
{0.004}

dGDP

CDP

0.084
(1.679)
[0.963]
{0.005}
2.250
(2.065)
[0.271]
{0.002}

0.410
(0.449)
[0.361]
{0.001}

0.730
(0.529)
[0.173]
{0.003}
3.818
(2.569)
[0.142]
{0.002}

3.220
(1.304)
[0.014]
{0.010}

0.332
(0.332)
[0.312]
{0.001}

1.103
(0.586)
[0.061]
{0.009}

RTB

CAY

Dependent variable: DD

0.535
(0.158)
[0.002]
{0.036}

0.857
(0.179)
[0.000]
{0.093}

0.936
(0.189)
[0.000]
{0.115}

0.761
(0.209)
[0.000]
{0.087}

CEP

5.378
(1.937)
[0.005]
{0.038}

3.914
(1.982)
[0.046]
{0.017}

1.976
(1.882)
[0.302]
{0.001}

0.191
(1.743)
[0.909]
{0.005}

dGEP

3.739
(1.653)
[0.026]
{0.024}

4.540
(1.564)
[0.004]
{0.037}

3.997
(1.633)
[0.015]
{0.030}

2.763
(1.656)
[0.101]
{0.016}

CAY

Dependent variable: EE

31.684
(7.277)
[0.000]
{0.086}

38.343
(7.442)
[0.000]
{0.120}

39.601
(7.544)
[0.000]
{0.134}

31.323
(5.488)
[0.000]
{0.097}

RTB

Univariate forecast of quarterly fundamentals: Decomposition regressions (1952:Q12008:Q4)

Table 8 (continued)

0.461
(0.101)
[0.000]
{0.066}

0.338
(0.096)
[0.001]
{0.040}

0.128
(0.086)
[0.135]
{0.004}

0.080
(0.062)
[0.196]
{0.001}

CBP

3.619
(1.167)
[0.002]
{0.044}

2.264
(1.035)
[0.027]
{0.019}

1.323
(0.861)
[0.127]
{0.006}

0.720
(0.643)
[0.266]
{0.001}

dGBP

2.743
(1.132)
[0.015]
{0.031}

1.240
(1.063)
[0.249]
{0.004}

0.163
(0.888)
[0.856]
{0.004}

0.844
(0.624)
[0.176]
{0.006}

CAY

Dependent variable: EB
RTB

4.046
(4.110)
[0.320]
{0.001}

6.490
(3.469)
[0.062]
{0.006}

11.164
(3.119)
[0.000]
{0.037}

14.657
(2.657)
[0.000]
{0.121}

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

555

CDP

0.025
(0.010)
[0.012]
2 0.047
(0.017)
[0.006]
3 0.061
(0.023)
[0.009]
4 0.071
(0.029)
[0.015]
8 0.039
(0.053)
[0.447]
12
0.002
(0.072)
[0.979]
16 0.063
(0.072)
[0.376]
20 0.183
(0.070)
[0.010]

0.285
(0.107)
[0.007]
0.659
(0.170)
[0.000]
0.962
(0.230)
[0.000]
1.346
(0.284)
[0.000]
2.951
(0.455)
[0.000]
4.487
(0.597)
[0.000]
6.012
(0.608)
[0.000]
7.047
(0.639)
[0.000]

dGDP

RTB

0.167
1.569
(0.102) (0.236)
[0.106] [0.000]
2.891
0.421
(0.163) (0.428)
[0.011] [0.000]
0.556
3.927
(0.213) (0.610)
[0.008] [0.000]
0.764
4.387
(0.259) (0.776)
[0.004] [0.000]
1.393
2.431
(0.436) (1.244)
[0.002] [0.052]
1.768 0.832
(0.610) (1.465)
[0.005] [0.576]
2.193 3.424
(0.640) (1.777)
[0.000] [0.058]
2.519 5.593
(0.657) (2.100)
[0.000] [0.008]

CAY

Dependent variable: DD

0.311

0.274

0.212

0.161

0.192

0.200

0.196

0.151

R2
0.044
(0.058)
[0.442]
0.065
(0.095)
[0.487]
0.125
(0.147)
[0.386]
0.222
(0.195)
[0.252]
0.570
(0.217)
[0.010]
0.529
(0.199)
[0.009]
0.263
(0.191)
[0.165]
0.126
(0.168)
[0.447]

CEP
2.221
(1.568)
[0.165]
1.797
(1.255)
[0.157]
2.096
(2.047)
[0.316]
1.141
(2.003)
[0.574]
2.172
(1.948)
[0.267]
7.426
(1.811)
[0.000]
12.233
(1.819)
[0.000]
14.721
(1.634)
[0.000]

dGEP
1.589
(1.164)
[0.175]
0.776
(0.833)
[0.362]
1.094
(1.434)
[0.442]
0.250
(1.480)
[0.867]
1.490
(2.005)
[0.463]
5.584
(1.887)
[0.003]
9.619
(1.941)
[0.000]
11.321
(1.980)
[0.000]

CAY
7.089
(2.169)
[0.002]
10.598
(4.875)
[0.032]
7.546
(5.645)
[0.179]
0.457
(4.682)
[0.924]
27.164
(5.181)
[0.000]
34.879
(7.119)
[0.000]
34.568
(6.621)
[0.000]
30.379
(5.991)
[0.000]

RTB

Dependent variable: EE

0.267

0.287

0.248

0.152

0.002

0.007

0.045

0.072

R2
0.019
(0.010)
[0.070]
0.038
(0.020)
[0.062]
0.052
(0.028)
[0.062]
0.060
(0.037)
[0.114]
0.007
(0.072)
[0.930]
0.152
(0.102)
[0.137]
0.232
(0.112)
[0.033]
0.191
(0.105)
[0.071]

CBP
0.168
(0.119)
[0.159]
0.230
(0.190)
[0.235]
0.318
(0.293)
[0.277]
0.323
(0.399)
[0.414]
0.029
(0.791)
[0.971]
1.277
(1.066)
[0.237]
4.006
(1.213)
[0.001]
7.514
(1.259)
[0.000]

dGBP
0.155
(0.108)
[0.156]
0.201
(0.186)
[0.284]
0.292
(0.282)
[0.306]
0.292
(0.378)
[0.437]
0.314
(0.776)
[0.689]
0.433
(1.106)
[0.699]
2.809
(1.318)
[0.033]
6.050
(1.334)
[0.000]

CAY

3.472
(0.415)
[0.000]
6.916
(0.829)
[0.000]
9.774
(1.203)
[0.000]
11.895
(1.572)
[0.000]
14.324
(2.753)
[0.000]
11.950
(3.197)
[0.000]
8.593
(3.501)
[0.016]
6.937
(3.910)
[0.082]

RTB

Dependent variable: EB

0.200

0.088

0.046

0.110

0.280

0.333

0.362

0.355

R2

This table reports estimates from regressions of fundamentals on the lagged decomposed financial ratios. The dependent variable is the sum of H-period log
dividend growth (DD), log earnings growth (EE), and log accounting returns (EB). Financial ratios are log dividend-price ratio (DP), log earnings-price ratio (EP),
and log book-to-market ratio (BP), respectively. They all are from the S&P 500 index. C and dG are the cyclical and stochastic growth rate (the first difference
of trend component) components of financial ratios. The table reports OLS estimates of regressors, Newey-West corrected standard errors (in parentheses), and
bootstrapping p-value (in brackets). Adjusted R2 is reported separately. The sample period is from 1952:Q1 to 2008:Q4.

Multivariate forecast of quarterly fundamentals: Decomposition regressions (1952:Q12008:Q4)

Table 9

556
X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

557

growth, particularly in long horizons. CAY tends to forecast fundamental growth in


long horizons, too. RTB tends to forecast DD and EE with a positive sign in short
horizons and a negative sign in long horizons.23
In sum, cyclical components tend to predict decreasing fundamental growths in
short horizons while stochastic trend components tend to predict increasing fundamental growths in long horizons.

4.3. Behavior of financial ratios


So far, we observe that the decomposed financial ratios (DP, EP, and BP) predict
stock excess returns. This prompts us to ask the following: How do financial ratios
predict excess returns? What does this say about the channels of the prediction?
Summers (1986), Fama and French (1988), and Campbell and Shiller (1988b, 2005)
suggest a simple theory of slow mean reversion to explain the predictability; that
is, stock prices cannot drift too far from their fundamentals (dividend, earnings, or
book value). The theory of slow mean reversion requires that financial ratios have to
be stationary in the long run. The predictive power comes from the long-run mean
reversion in financial ratios. Previous findings also support the finding that financial
ratios have more predictive power in long horizons.
We observe that the cyclical components of the financial ratios predict increases
in stock returns and decreases in fundamentals, particularly in short horizons, while
the stochastic trend components of the financial ratios predict decreases in stock
returns, particularly in long horizons. Based on the adjusted R2 s in univariate forecast
regressions and the relative importance of each component in multivariate forecast regressions, we show that the cyclical components tend to capture more predictability in
relatively shorter horizons while the stochastic trend components tend to capture the
predictability in relatively longer horizons. By construction, the cyclical components
are less persistent than the stochastic trend components. Therefore, the cyclical component reflects a local mean reversion effect, while the stochastic trend component
reflects a slow mean reversion effect.
Our conjecture helps explain previous findings that raw financial ratios tend
to show some predictive power for stock returns in long horizons. In addition, our
finding that the two components predict stock returns in the opposite directions
suggests that, in short horizons, the two components of financial ratios tend to offset
much of each others prediction. This helps explain the failure of previous studies to
find strong predictive power of raw financial ratios in short horizons in the absence
of decomposition.

23

In a further robust check, we implement Lewellens (2004) bias-corrected method, since the first
differenced stochastic trend component is still quite persistent. In an unreported table, we confirm the
predictive power of the decomposed financial ratios in Tables 25.

558

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

5. Out-of-sample prediction
Our in-sample prediction results are striking. However, since both components
are estimated using the full sample, there is a concern that the results are subject
to a look-ahead bias. We address this concern by implementing OOS predictions.
The Hodrick-Prescott (1997) decomposition is based on the relative long-run equilibrium relation. A meaningful and consistent estimation requires a large number of
observations. On the other hand, the OOS prediction has to rely on the real-time
data that may induce sampling errors in forming the cyclical and stochastic trend
components. Therefore, we consider two types of OOS prediction. First, we predict
OOS using the ex post decomposition (components are estimated by using the full
sample). Second, we predict OOS using alternative ex ante decomposition. In this
approach, we estimate the cyclical and stochastic trend components using real-time
data (available at the time of prediction). We call the first approach pseudo OOS, and
the second pure OOS.24
We choose the random walk model as a benchmark model (restricted model).
We implement nested prediction comparisons. We run the prediction regression using
data from 1926:Q1 to 1989:Q4, and thus the first period for prediction is 1990:Q1.
We report the ratio of the root mean squared errors (MSE) for the unrestricted
and restricted model forecasts, Theils U, MSE-F (H0 : restricted and unrestricted
predictions are equal) by McCracken (2004), ENC-F (H0 : restricted prediction encompasses the unrestricted prediction) by Clark and McCracken (2001), and pseudo
R2 measured as one minus the ratio of mean square error of unrestricted to restricted
models.
Table 10 reports the results for the pseudo OSS prediction with bootstrapping
errors. When the unrestricted models include raw financial ratios (DP, EP, and BP),
we find that the prediction errors in the restricted model are smaller than those in the
unrestricted models. Theils U is greater than one, and pseudo R2 are negative. Both
MSE-F and ENC-F show that financial ratios do not perform well in OOS predictions.
Our evidence is consistent with evidence provided by Goyal and Welch (2008) and
Lettau and Van Nieuwerburgh (2008) in that the random walk model outperforms
the models with raw financial ratios. We find a similar pattern for the model with
RTB as a predictor. We find that CAY with fixed cointegrating parameters for the
full sample outperforms the random walk model in the Clark and McCracken (2001)
test.
We are particularly interested in the OOS prediction with the cyclical or stochastic trend component of financial ratios as a predictor. When we compare the performance of the random walk model with that of the model with the ex post cyclical
component of financial ratios, we find that the Theils U is less than one, and pseudo
R2 is positive for CDP, CEP, and CBP . Both MSE-F and ENC-F are statistically

24

We use the terms following Lettau and Van Nieuwerburgh (2008).

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

559

Table 10
Pseudo out-of-sample (OOS) forecast
This table reports the results of one-quarter-ahead nested prediction comparison of excess returns on the
S&P 500 index over a three-month T-bill rate and fundamentals. Financial ratios are log dividend-price
ratio (DP), log earnings-price ratio (EP), and log book-to-market (BP), respectively. They all are from
the S&P 500 index. C and dG are the cyclical and stochastic growth rate (the first difference of trend
component) components of the financial ratios. The table reports pseudo OOS prediction for CAY, RTB,
C and dG. The benchmark (restricted) model is the random walk model. We report the Theils U (the
ratio of the root mean squared errors for the unrestricted and restricted model forecasts), MSE-F (H0 :
restricted and unrestricted predictions are equal) by McCracken (2004), ENC-F (H0 : restricted prediction
encompasses the unrestricted prediction) by Clark and McCracken (2001), and pseudo R2 measured as
one minus the ratio of mean square error of unrestricted to restricted models.
Return regression
Variable Theils U

MSE-F

DP

2.117
1.218
[1.456] [0.660]
7.016
1.198
[1.399] [0.637]
3.895
1.059
[1.595] [0.694]
4.211
2.550
[1.079] [0.571]
8.118 2.670
[0.357] [0.178]
3.618
2.631
[1.052] [0.545]
1.969
4.477
[1.047] [0.526]
1.148
4.372
[1.012] [0.517]
4.669
6.217
[1.596] [0.953]
5.765
6.119
[1.353] [0.736]
6.643
6.301
[1.396] [0.813]

EP
BP
CAY
RTB
CDP
CEP
CBP
dGDP
dGEP
dGBP

1.014
[0.010]
1.050
[0.010]
1.027
[0.011]
1.029
[0.007]
1.058
[0.002]
0.977
[0.007]
0.987
[0.007]
0.993
[0.007]
0.971
[0.011]
0.964
[0.009]
0.959
[0.009]

ENC-F

Fundamentals regression
Pseudo

R2

0.029
[0.020]
0.102
[0.020]
0.054
[0.023]
0.059
[0.014]
0.120
[0.005]
0.045
[0.014]
0.025
[0.014]
0.015
[0.014]
0.058
[0.022]
0.071
[0.018]
0.080
[0.019]

Theils U

MSE-F

ENC-F

Pseudo R2

1.958
[0.010]
1.016
[0.008]
0.867
[0.008]

47.309 2.545
[1.230] [0.639]
1.984
0.183
[1.031] [0.542]
21.212
51.344
[1.032] [0.573]

2.834
[0.020]
0.032
[0.017]
0.249
[0.016]

1.319
[0.007]
1.005
[0.009]
0.962
[0.006]
0.968
[0.013]
0.990
[0.013]
0.977
[0.009]

32.348 2.918
[1.112] [0.587]
0.695 0.257
[1.292] [0.641]
6.062
3.675
[0.976] [0.524]
5.165
3.335
[1.881] [1.063]
1.472
1.027
[1.873] [1.043]
3.554
4.610
[1.398] [0.753]

0.741
[0.015]
0.009
[0.018]
0.074
[0.013]
0.064
[0.026]
0.019
[0.027]
0.045
[0.019]

significant with bootstrap. When we compare the performance of the random walk
model with that of the model with the ex post stochastic trend component of financial ratios, we find a similar pattern. Our evidence shows that the models including
the cyclical or stochastic trend components of financial ratios (estimated in the full
sample) in the restricted model have superior OOS predictions.
Table 11 reports the results for the pure OOS prediction. Here, we use the 25
years moving average of financial ratio as a measure of stochastic trend component
while using the difference between the financial ratio and the stochastic trend as a

560

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Table 11
Pure out-of-sample (OOS) forecast
This table reports the results of one-quarter-ahead nested prediction comparison of excess returns on the
S&P 500 index over a three-month T-bill rate and fundamentals. Financial ratios are log dividend-price
ratio (DP), log earnings-price ratio (EP), and log book-to-market (BP), respectively. They all are from
the S&P 500 index. C and dG are the cyclical and stochastic growth rate (the first difference of trend
component) components of the financial ratios. Here, to avoid the look-ahead bias, we simply use 25 years
moving average as a proxy for the long-run growth component, the difference of the financial ratio series
and the moving average as a proxy for the cyclical component. The table reports pure OOS prediction
for C and dG. The benchmark (restricted) model is the random walk model. We report the Theils U (the
ratio of the root mean squared errors for the unrestricted and restricted model forecasts), MSE-F (H0 :
restricted and unrestricted predictions are equal) by McCracken (2004), ENC-F (H0 : restricted prediction
encompasses the unrestricted prediction) by Clark and McCracken (2001), and pseudo R2 measured as
one minus the ratio of mean square error of unrestricted to restricted models.
Return regression
Variable Theils U

MSE-F

CDP

2.249
2.056
[2.116] [1.393]
3.190
2.488
[0.955] [0.477]
1.638
1.740
[2.002] [1.444]
1.442
1.704
[1.038] [0.596]
0.095 0.046
[0.957] [0.494]
0.068 0.033
[0.994] [0.543]

CEP
CBP
dGDP
dGEP
dGBP

0.986
[0.015]
0.980
[0.006]
0.989
[0.014]
0.991
[0.007]
1.001
[0.006]
1.000
[0.007]

ENC-F

Fundamentals regression
Pseudo R2

Theils U

0.029
[0.031]
0.040
[0.013]
0.021
[0.027]
0.019
[0.014]
0.001
[0.013]
0.001
[0.013]

1.495
[0.007]
0.999
[0.008]
0.845
[0.007]
1.255
[0.007]
1.006
[0.008]
0.843
[0.007]

MSE-F

ENC-F

42.017 3.481
[1.060] [0.631]
0.086
0.190
[1.249] [0.749]
30.500
19.782
[1.103] [0.591]
27.709
3.650
[1.005] [0.564]
0.937 0.262
[1.164] [0.579]
30.924
22.768
[0.999] [0.555]

Pseudo R2
1.236
[0.014]
0.001
[0.017]
0.286
[0.015]
0.574
[0.014]
0.012
[0.016]
0.289
[0.013]

measure of the cyclical component. In this way, we use the cyclical and stochastic
trend components estimated using real-time data in OOS prediction tests. We find
that the Theils U is less than one for all cyclical components of three financial
ratios, suggesting that the MSE for the unrestricted model is less than the MSE for
the restricted model. MSE-F rejects the null of equal forecast accuracy between the
constant expected returns benchmark model and the cyclical component-augmented
model. ENC-F also rejects the null hypothesis that cyclical components have no
predictive power for excess returns. However, bootstrap errors show that only the
MSE-F and ENC-F for the cyclical component in earnings-price ratio are significant.
For the stochastic trend components, the Theils U in dividend-price ratio is less
than one while the others are greater than one. The bootstrap errors show that only
the ENC-F for the stochastic component of dividend-price ratio is significant. In
sum, the result shows some of the cyclical components of financial ratios display
statistically significant OOS predictive power for excess returns. The stochastic trend

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

561

components display little OOS predictive power for excess returns. For the growth in
fundamentals, both the cyclical and stochastic trend components of book-to-market
ratio show statistically significant predictive power for accounting returns. The Theils
U of the cyclical and stochastic trend components are 0.845 and 0.843, respectively.
MES-F and ENC-F are 30.500 and 19.782 (30.924 and 22.768), respectively, for
the cyclical (stochastic trend) component. Bootstrap errors also suggest the OOS
predictability.

6. Conclusions
Cochrane (2008, p. 1535) points out If both returns and dividend growth are
unforecastable, then present value logic implies that the price/dividend ratio is constant, which is obvious not. Some explanation should be explored. Motivated by the
finding that financial ratios are highly persistent and that the mean of financial ratios
is time-varying, we have examined the prediction of excess returns and fundamentals
by financial ratios in-sample and OOS. We examine the prediction by decomposing
the financial ratios into a cyclical component and a stochastic trend component using
Hodrick and Prescotts (1997) Kalman filter procedure. Previous studies find that
financial ratios predict long-horizon returns but scant short-horizon returns and fundamentals. In contrast, we find that decomposed financial ratios significantly predict
excess returns and fundamentals in both long and short horizons. The OOS forecast
result shows that the cyclical component of earnings-price ratio is superior in forecasting market return, and the stochastic trend component of book-to-market ratio is
superior in forecasting fundamentals.
We also find that the cyclical components of financial ratios predict increases
in returns, while the stochastic trend components predict decreases in returns. We
interpret these findings as evidence that the cyclical components reflect a local mean
reversion effect while the stochastic trend components reflect a long-run persistence
effect. This helps explain previous findings that, in the absence of decomposition,
financial ratios find little predictive power in short horizons due to offsetting effects.
It also helps explain the failure of dividend-price ratio to predict dividend growth.
Our result shows that the decomposed financial ratios based on the Hodrick and
Prescott (1997) method predict stock returns and fundamentals better than financial
ratios alone.

References
Ang, A. and G. Bekaert, 2007. Stock return predictability: Is it there? Review of Financial Studies 20,
651707.
Ang, A., R. Hodrick, Y. Xing, and X. Zhang, 2006. The cross-section of volatility and expected returns,
Journal of Finance 51, 259299.
Blanchard, O. and M. Watson, 1982. Bubbles, rational expectations and financial markets, in: P. Wachtel,
ed., Crises in the Economic and Financial Structure (Lexington Books, Lexington, MA).

562

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Campbell, J.Y., 1991. A variance decomposition for stock returns, Economic Journal 101, 157179.
Campbell, J.Y. and R.J. Shiller, 1987. Cointegration and tests of present value models, Journal of Political
Economy 95, 10621088.
Campbell, J.Y. and R.J. Shiller, 1988a. The dividend-price ratio and expectations of future dividends and
discount factors, Review of Financial Studies 1, 195228.
Campbell, J.Y. and R.J. Shiller, 1988b. Stock price, earnings and expected dividends, Journal of Finance
43, 661676.
Campbell, J.Y. and R.J. Shiller, 2005. Valuation ratios and the long-run stock market outlook: An update,
in: R.H. Thaler, ed., Advances in Behavioral Finance, Volume II (Princeton University Press, Princeton,
NJ).
Campbell, J.Y. and T. Vuolteenaho, 2004. Inflation illusion and stock prices, American Economic Review
Papers and Proceedings 94, 1923.
Chen L., R. Petkova, and L. Zhang, 2008. The expected value premium, Journal of Financial Economics
87, 269280.
Clark, P.K., 1987. The cycle component of the U.S. economic activity, Quarterly Journal of Economics
102, 797814.
Clark, T. and M. McCracken, 2001. Tests of equal forecast accuracy and encompassing for nested models,
Journal of Econometrics 105, 85110.
Cochrane, J.H., 1992. Explaining the variance of price-dividend ratios, Review of Financial Studies 5,
243280.
Cochrane, J.H., 2001. Asset Pricing, (Princeton University Press, Princeton, NJ).
Cochrane, J.H., 2008. The dog that did not bark: A defense of return predictability, Review of Financial
Studies 21, 15331575.
Fama, E.F., 2006. The behavior of interest rates, Review of Financial Studies 19(2), 359379.
Fama, E.F. and K.R. French, 1988. Dividend yields and expected stock returns, Journal of Financial
Economics 22, 327.
Fama, E.F. and K.R. French, 2002. The equity premium, Journal of Finance 57, 637659.
Fama, E.F. and K.R. French, 2007. The anatomy of value and growth stock returns. Working paper,
University of Chicago.
Ferson, W.E., S. Sarkissian, and T.T. Simin, 2003. Spurious regressions in financial economics? Journal
of Finance 58(4), 13931413.
Ferson, W.E. and B. Xie, 2009. The out-of-sample performance of long-run risk models. Working paper,
University of Southern California.
Froot, K.A. and M. Obstfeld, 1991. Intrinsic bubbles: The case of stock prices, American Economic Review
81, 11891214.
Goetzman, W.N. and P. Jorion, 1993. Testing the predictive power of dividend yields, Journal of Finance
48, 663679.
Goncalves, S. and L. Kilian, 2004. Bootstrapping autoregressions with conditional heteroskedasticity of
unknown form, Journal of Econometrics 123, 89120.
Goyal, A. and P. Santa-Clara, 2003. Idiosyncratic risk matters! Journal of Finance 58, 9751007.
Goyal, A. and I. Welch, 2003. Predicting the equity premium with dividend ratios, Management Science
49(5), 639654.
Goyal, A. and I. Welch, 2008. A comprehensive look at the empirical performance of the equity premium
prediction, Review of Financial Studies 21, 14551508.
Graham, B. and D. Dodd, 1934. Security Analysis (McGraw-Hill, New York).
Harvey, A.C., 1985. Trends, cycles in macroeconomic time series, Journal of Business and Economics
Statistics 3, 216227.
Hecht, P., T. Vuolteenaho, 2006. Explaining returns with cash-flows proxies, Review of Financial Studies
19, 159194.
Hodrick, R., 1992. Dividend yields and expected stock returns: Alternative procedures for inference and
measurement, Review of Financial Studies 5, 357386.

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

563

Hodrick, R. and E.C. Prescott, 1997. Postwar U.S. business cycles: An empirical investigation, Journal of
Money, Credit, and Banking 29, 116.
Jiang, X.Q. and B.S. Lee, 2006. The dynamic relation between returns and idiosyncratic volatility, Financial Management 35, 4366.
Jiang, X.Q. and B.S. Lee, 2007. Stock returns, dividend yield, and book-to-market ratio, Journal of Banking
and Finance 31, 455475.
Jiang, X.Q. and B.S. Lee, 2009, The intertemporal risk-return relation in the stock market, The Financial
Review 44(4), 541558.
Kendall, M.G., 1954, Note on bias in the estimation of auto-correlation, Biometrika 41, 403404.
King, R.G., C.I. Plosser, J.H. Stock, and M.W. Watson, 1991. Stochastic trends and economic fluctuations,
American Economic Review 81, 819840.
Lee, B.S., 1996a. Time-series implications of aggregate dividend behavior, Review of Financial Studies
9(2), 589618.
Lee, B.S., 1996b. Comovements of earnings, dividends, and stock prices, Journal of Empirical Finance
3(4), 327346.
Lee, B.S., 1998. Permanent, temporary, and nonfundamental components of stock prices, Journal of
Financial and Quantitative Analysis 33(1), 132.
Lettau, M. and S.C. Ludvigson, 2001. Consumption, aggregate wealth and expected stock returns, Journal
of Finance 56(3), 815849.
Lettau, M. and S.C. Ludvigson, 2005. Expected returns and expected dividend growth, Journal of Financial
Economics 76, 583626.
Lettau, M. and S. Van Nieuwerburgh, 2008. Reconciling the return predictability evidence, Review of
Financial Studies, 21, 16071652.
Lewellen, J.W., 2004. Predicting returns with financial ratios, Journal of Financial Economics 74(2),
209235.
Lintner, J., 1956. Distribution of income of corporations among dividends, retained earnings and taxes,
American Economic Review 46, 97113.
Lucas, R.E. Jr., 1980. Methods and problems in business cycle theory, Journal of Money, Credit, and
Banking 12, 696715.
Lucas, R.E. Jr., 1981. Studies in Business Cycle Theory (MIT Press, Massachusetts, MA).
Marsh, T.A. and R.C. Merton, 1986. Dividend variability and variance bounds tests for the rationality of
stock market prices, American Economic Review 76, 483498.
McCracken, M., 2004. Asymptotics for out-of-sample tests of causality. Working paper, University of
Missouri-Columbia.
Nelson, C. and M. Kim, 1993. Predictable stock returns: The role of small-sample bias, Journal of Finance
48, 641661.
Park, S.C., 2010. The moving average ratio and momentum, The Financial Review 45, 415447.
Paye, B.S. and A. Timmermann, 2006. Instability of return prediction models, Journal of Empirical
Finance 13, 274315.
Pindyck, R. and J. Rotemberg, 1993. The comovement of stock prices, Quarterly Journal of Economics
108(4), 10731104.
Pontiff, J. and L.D. Schall, 1998. Book-to-market ratios as predictors of market returns, Journal of
Financial Economics 49, 141160.
Rozeff, M.S., 1984. Dividend yields are equity risk premia, Journal of Portfolio Management 49, 141160.
Stambaugh, R.F., 1986. Bias in regressions with lagged stochastic regressors. Working paper, University
of Chicago.
Stambaugh, R.F., 1999. Predictive regressions, Journal of Financial Economics 54, 375421.
Summers, L., 1986. Does the stock market rationally reflect fundamental values? Journal of Finance 41,
591602.
Valkanov, R., 2003. Long-horizon regressions: Theoretical results and applications, Journal of Financial
Economics 68, 201232.

564

X. Jiang and B. S. Lee/The Financial Review 47 (2012) 531564

Viceira, L., 1996. Testing for structural change in the predictability of asset returns. Unpublished
manuscript, Harvard University.
Vuolteenaho, T., 2000. Understanding the aggregate book-market ratio and its implications to current
equity-premium expectations. Unpublished paper, Harvard University.
Vuolteenaho, T., 2002. What drives firm-level stock returns? Journal of Finance 57, 233264.
Wu, Y., 1997. Rational bubbles in the stock market: Accounting for the U.S. stock-price volatility,
Economic Inquiry 35, 309319.

Copyright of Financial Review is the property of Wiley-Blackwell and its content may not be copied or emailed
to multiple sites or posted to a listserv without the copyright holder's express written permission. However,
users may print, download, or email articles for individual use.

Вам также может понравиться