Present Value Relations (Ch 7 in Campbell
et al.)
Consider asset prices instead of returns.
Predictability of stock returns at long hori-
zons: There is weak evidence of predictabil-
ity when the return history is used. The goal
is now whether the predictability can be im-
proved, when other variables, such as divi-
dend price ratio or the level of interest rate,
are brought into the analysis.
An empirical nding is that expected returns
are at long horizon roughly consistent with
persistent AR(1). Why this is so is yet an
open question. We consider the basic method-
ology to investigate this.
1
Relations between Prices, Dividends and Re-
turns
The simple return:
R
t+1
=
P
t+1
+D
t+1
P
t
1: (1)
Continuously compounded return:
r
t+1
= log(1 +R
t+1
) (2)
Simple Case: Constant expected return
E
t
[R
t+1
] = R (a constant);
where E
t
denotes the conditional expectation
given information up to time point t. Then
P
t
= E
t
"
P
t+1
+D
t+1
1 +R
#
: (3)
A simple application of the law of iterated
expectations, E
t
[E
t+1
[X]] = E
t
[X] gives then
2
P
t
= E
t
"
K
X
i=1
1
1 +R
i
D
t+i
#
+E
t
1
1 +R
K
P
t+K
: (4)
(Just for fun|verify!)
Assume that
lim
K!1
E
t
"
1
1 +R
K
P
t+K
#
= 0: (5)
Thus we obtain a stock price model, called a
constant return present value model, referred
to as P
Dt
P
t
= P
Dt
= E
t
2
4
1
X
i=1
1
1 +R
i
D
t+i
3
5
: (6)
3
A further simplied (but more unrealistic)
special case is when
E
t
[D
t+i
] = (1 +G)E
t
[D
t+i1
] = (1 +G)
i
D
t
;
i.e., a constant growth model with growth
rate G < R. Then
P
t
=
E
t
[D
t+1
]
R G
=
1 +G
RG
D
t
:
The Gordon growth model.
4
Implications of the dividend discount model
P
t
is not a martingale
, because E
t
[P
t+1
] 6=
P
t
.
However, if all the dividends are reinvested
in the stock, giving a portfolio with N
t
number of shares at time t, where
N
t+1
= N
t
1 +
D
t+1
P
t+1
!
:
The value of the portfolio is
M
t
=
N
t
P
t
(1 +R)
t
;
which is a martingale.
A stochastic process, Y
t
, is a martingale if E
t
[Y
t+1
] =
Y
t
.
5
P
t
follows a linear process with unit root,
if D
t
follows a linear process with unit
root.
If D
t
(and P
t
) are I(1)) then they are
under the model (6) (Prove it!) cointe-
grated
y
Above prices and dividends are related by a
linear modeled. A more appropriate approach
may be by using log prices instead. This will
be done in what follows.
Y
t
has a unit root, or is integrated of order one,
denoted as Y
t
I(1) if Y
t
= Y
t
Y
t1
is stationary
[is integrated of order 0, denoted as Y
t
I(0)]
y
There exist a real number such that P
t
+aD
t
is sta-
tionary
6
Rational Bubbles
Relax the assumption in the model (4) that
discounted end value (5) of the stock con-
verges to zero. Then P
t
is no more unique.
Any solution can be written in the form
P
t
= P
Dt
+B
t
; (7)
where
B
t
= E
t
B
t+1
1 +R
: (8)
7
The term P
Dt
is sometimes called fundamen-
tal value, and B
t
a rational bubble.
Consider the following example by Blanchard
and Watson (1982)
B
t+1
=
(
1+R
B
t
+
t+1
; with probability
t+1
with probability 1 :
where E
t
t+1
= 0. This example bubble obeys
the restriction (8). The bubble bursts at
any period with probability 1 . If it does
not burst it grows at the rate (1 +R)= 1,
faster than R to compensate the probability
of bursting.
Several other examples of bubbles can be
given.
Can rational bubbles exist?
The word "bubble" recalls some of the famous
episodes in nancial history in which asset prices rose
far higher than could be explained by fundamentals,
and in which investors appeared to be betting that
other investors would drive prices even higher. "Ra-
tionale" is used because B
t
in the price equation is
fully consistent with rational expectations and con-
stant expected returns.
8
Approximate Present-Value Relation
with Time Varying Expected Returns
Empirical evidence that stock returns are pre-
dictable at least to some extends implies that
expected returns are time-varying rather than
constant. As a consequence working with
present value relations become much more
dicult, for the relation between prices and
returns become nonlinear
A popular approach is a log linear approxi-
mation for the nonlinearity. Rationale in ac-
counting framework: High prices must be fol-
lowed eventually by
high future dividends,
low future returns, or
or some combination of the two above.
9
The log linear approximation:
r
t+1
= log(P
t+1
+D
t+1
) log(P
t
)
= p
t+1
p
t
+log
1 +exp(d
t+1
p
t+1
)
;
where in the last form it is necessary to as-
sume that D
t
> 0. The last term is a nonlin-
ear function of the dividend price ratio.
Taylor approximation
gives
r
t+1
k +p
t+1
+(1 )d
t+1
p
t
; (9)
where k and are linearization parameters
dened by
=
1
1 +exp(d p)
;
with d p the average of log dividend-price
ratio, and
k =log (1 ) log(
1
1):
(Exercise: Verify).
If f(x) is a dierentiable function then the rst order
Taylor approximation of f at c is
f(x) f(c) +f
0
(c)(x c)
10
Thus the approximation replaces the log of
the sum of the price and dividend by the
weighted average.
Accuracy of the approximation:
If the dividend-price is constant the approxi-
mation relation holds exactly (Exercise: prove
it).
Generally it has proven that the approxima-
tion is reasonably accurate, especially for in-
vestigating the asset price dynamics.
Implication for Prices:
Imposing the (terminal) condition
lim
j!1
j
p
t+j
= 0
we obtain
p
t
=
k
1
+
1
X
j=0
j
[(1 )d
t+1+j
r
t+1+j
]:
(Exercise: Verify)
11
The relation holds also for ex ante, for E
t
[p
t
] =
p
t
, consequently
p
t
=
k
1
+E
t
2
4
1
X
j=0
j
[(1 )d
t+1+j
r
t+1+j
]
3
5
:
Interpretation of the model:
A high stock price implies expected high fu-
ture dividends and low future return values,
and vice versa. Campbell and Shiller
call the
model as dynamic Gordon growth model or
the dividend-ratio model.
Once the dividend growth rates and the dis-
count rates are constants the DGGM reduces
to GGM (Again verify.)
Campbell, J. and R. Chiller (1988). The dividend-
price ratio and expectations on future dividends and
discount factors. Review of Financial Studies, 1,
195{227.
Campbell, J. and R. Chiller (1988). Stock prices,
earnings and expected dividends. Journal of Finance,
43, 661{676.
12
Dening
p
dt
= (1 )E
t
2
4
1
X
j=0
j
d
t+1+j
3
5
and
p
rt
= E
t
2
4
1
X
j=0
j
r
t+1+j
3
5
we can rewrite the model
p
t
=
k
1
+p
dt
p
rt
;
The log dividend-price ratio can be written
in terms of the model
d
t
p
t
=
k
1
+E
t
2
4
1
X
j=0
j
[d
t+1+j
+r
t+1+j
]
3
5
:
Hence, the dividend-price ratio is high if the
dividends are expected to grow slowly or the
stock returns are expected to be high.
Note again that if d
t
I(1) and p
t
I(1),
then d
t
, and p
t
are cointegrated with a coin-
tegration coecient equal to one.
13
Consider next the return series. Using the
above relations we can write (work this out
yourself)
r
t+1
E
t
[r
t+1
] = E
t+1
P
1
j=0
j
d
t+1+j
E
t
P
1
j=0
j
d
t+1+j
E
t+1
P
1
j=0
j
r
t+1+j
E
t
P
1
j=0
j
r
t+1+j
:
or
r
t+1
E
t
[r
t+1
] =
t+1
=
d;t+1
r;t+1
; (10)
where
t+1
is the unexpected stock return,
d;t+1
is the change in expectations of future
dividends, and
r;t+1
is the change in expec-
tations of future returns.
Hence, returns must be associated with changes
in expectations of future dividends or real re-
turns.
14
A Simplied Example
Suppose
E
t
[r
t+1
] = r +x
t
(11)
where
x
t+1
= x
t
+
t+1
; 1 < < 1 (12)
a stationary zero-mean AR(1) process. If
is close to one the process is highly persistent
(approaches to the random walk). Denoting
2
x
= Var(x
t
) and
2
= Var(
t
), we obtain
2
x
=
2
1
:
Then
p
rt
= E
t
2
4
1
X
j=0
j
r
t+1+j
3
5
=
r
1
+
x
t
1
: (13)
(Exercise: Verify!)
15
Consequently the more persistent the expected
return is the greater is the eect on stock
price. Since is close to one, a 1% in-
crease in return today decreases the stock
price by about 2% if = 0:5, by about 4% if
= 0:75, and by about 10% if = 0:9.
Note that the variability of expected stock
returns are measured by the standard devi-
ation of x
t
. Consequently it is tempting to
think that if
x
is small, then changing ex-
pected returns have little inuence on stock
prices, i.e. the variability in p
rt
.
However, this may not be so, because the
variability in p
rt
comes through x
t
=(1 ),
and if is close to one, then (1) is close
to zero, and hence already small changes in
x
t
may impose large changes in p
rt
.
16
In terms of the above model the one-period
return becomes
r
t+1
= r +x
t
+
d;t+1
t+1
1
: (14)
Consequently
Var[r
t+1
] =
2
d
+
2
x
"
1 +
2
2
(1 )
2
#
2
d
+
2
2
x
1
;
where the approximate equality holds when
, and is close to one. Variability is the
larger the higher is the persistence (provided
> 0).
Equations (12) and (14) can be used to show
that realized stock returns follow an ARMA(1,1)
process. (Exercise: Show this and calculate
the autocorrelation function. What are the
implications? Assume < .)
17
