Consumption-based asset pricing theory

David Schroder
Birkbeck College
11-14 April 2011
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 1 / 35
Contents: Continuous-Time Asset Pricing
1
Asset prices
2
Utility and utility maximization
3
The stochastic discount factor in continuous time
4
The risk-free rate
5
Risky assets
6
Example
7
Sharpe ratio
8
Habit formation models
Required reading: Cochrane, Chapter 1.5
Note: These lecture notes are incomplete without having attended the
lectures.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 2 / 35
III Continuous-Time Asset Pricing
In this part of the lecture we extend our asset pricing theory of part I
to the continuous-time framework.
How can we price assets in continuous-time?
Combination of part I (asset pricing) and part II (stochastic calculus)
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 3 / 35
III1. Modeling asset prices
Revision: Assets (mainly shares) in the two-period model:
price: p
t
payo: x
t+1
= dividend D
t+1
+ future price p
t+1
In continuous time:
price: p
t
at any moment of time (stochastic process)
payo: D
t
continuous ow (or rate) of dividends or other payments
(again: stochastic process)
Over the small time interval dt, an asset pays dividends D
t
dt
Note that dividends are denoted by D
t
and not by d
t
to allow for a better distinction
with the dierence operator d.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 4 / 35
Asset returns
An investor holding some asset has two sources of income:
Capital gains. Over a short time interval: dp
t
Dividend payments. Over a short time interval: D
t
dt
The instantaneous total net return of an asset is:
dr
t
= r
t
dt =
dp
t
p
t
+
D
t
p
t
dt =
dp
t
+ D
t
dt
p
t
i.e., total return = price return (capital gain) + dividend yield
The total gross return:
dR
t
= R
t
dt = 1 + dr
t
= 1 +
dp
t
+ D
t
dt
p
t
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 5 / 35
Share prices process
Shares (and other risky assets) are usually assumed to follow a
diusion (Ito) process:
dp
t
= a(p
t
, t)dt + b(p
t
, t)dz
t
Shares are often assumed to follow a GBM:
dp
t
p
t
= dt + dz
t
Advantages:
A GBM can never fall below zero.
Returns are normally distributed.
It comes close to empirical evidence.
This setting is also called Merton-Black-Scholes model.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 6 / 35
Risk-free assets in continuous time
There are two possibilities to model risk-free assets (bonds):
1
Risk-free asset has
a price of 1 at all times p = 1, and
pays the risk-free rate as dividend: D
t
= r
f
t
2
Risk-free asset pays no dividend (D = 0), but the price grows at the
risk-free rate:
dp
t
p
t
= r
f
t
dt
Both formulas are equivalent, so it depends on the model/exercise
which description is preferred.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 7 / 35
Interest rates and bonds in continuous time
In continuous time, r
f
t
is the (annualized) continuously compounded
interest rate.
Depending on the context, we assume that it is either
constant r
f
t
= r
f
stochastic. Then it is also called instantaneous interest rate.
The instantaneous return of a (risk-free) bond equals the
instantaneous interest rate.
Some money B
0
invested in the risk-free rate becomes at time t:
B
t
= B
0
e

t
0
r
f
s
ds
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 8 / 35
II2. Utility and utility function
In continuous time, an investor has a consumption stream
{c
t
}t [0, T]. Sometimes this consumption stream is innite, then
t [0, ].
Accordingly, instead of a period utility function u(c
t
), we now have a
instantaneous utility function u(c
t
).
Usually, future consumption is not predictable, such that the
consumption stream is stochastic.
Note that utility can also depend on other variables than
consumption, such as leisure.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 9 / 35
Lifetime utility
The investors lifetime utility is given over his expected consumption
stream {c
t
}:
U({c
t
}) = E
0
__
T
t=0
e
t
u(c
t
)dt
_
where is the continuously compounded time preference rate
(subjective discount factor). Note that we use instead of .
Remember the discrete-time version:
U(c
t
, c
t+1
) = u(c
t
) + E
t
[u(c
t+1
)]
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 10 / 35
Utility maximization
Again, the basic element of continuous-time asset pricing is the utility
maximization of the investor.
The investor has, as before, the possibility to invest into an asset x,
that pays a stream of payos (i.e., dividends) D
t
.
The investor maximizes his expected lifetime utility:
max

U({c
t
}) = max

E
0
__

t=0
e
t
u(c
t
)dt
_
where denotes the amount invested into a given asset x.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 11 / 35
Budget constraint and solution
The budget constraint over a small interval of time dt is given by:
c
0
= e
0
p
0
c
t
= e
t
+ D
t
t > 0
Interpretation: consumption can be nanced by the endowment, the
dividend income and selling / buying assets.
The solution of this seemingly simple problem is not trivial, using the
Hamilton-Jacobi-Bellman partial dierential equation (a special case
of the Feynman-Kac theorem).
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 12 / 35
The rst order condition
The FOC is given by:
p
0
u

(c
0
) = E
0
_

t=0
e
t
u

(c
t
)D
t
dt
The loss from buying one unit of the asset is
u

(c
0
)(e
0
c
0
)dt = u

(c
0
)p
0
The gain from buying one unit of the asset is the right-hand side of
the equation above times :
For one given time t, it is equal to E
0
[e
t
u

(c
t
)D
t
]
The FOC is the continuous-time analogue to
p
0
= E
0

t=0

t
u

(c
t
)
u

(c
0
)
D
t
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 13 / 35
II3. The stochastic discount factor
In the two-period model, the SDF was given by:
m
t+1
=
u

(c
t+1
)
u

(c
t
)
However, in continuous time, dividing by u

(c
t
) is not possible, since
u

(c
t+t
)
u

(c
t
)
is not well behaved for small t.
(c
t
is a stochastic process)
Rather, we use directly the level of marginal utility u

(c
t
). Hence we
can dene the stochastic discount factor as:
m
t
= e
t
u

(c
t
)
We always take the derivative with respect to consumption! Note that
we use m
t
instead of
t
in the Cochrane textbook.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 14 / 35
Pricing function
Using
m
t
= e
t
u

(c
t
)
we can rewrite the pricing function
p
0
u

(c
0
) = E
0
_

t=0
e
t
u

(c
t
)D
t
dt
to
p
t
m
t
= E
t
_

s=0
m
t+s
D
t+s
ds
This is the continuous-time counterpart of
p
t
= E
t

j =1

j
d
t+j
u

(c
t+j
)
u

(c
t
)
= E
t

j =1
d
t+j
m
t,t+j
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 15 / 35
Derivation of the one-period formula
p
t
m
t
= E
t
_

s=0
m
t+s
D
t+s
ds (I)
E
t
[p
t+t
m
t+t
] = E
t
_

s=t
m
t+s
D
t+s
ds (II)
p
t
m
t
E
t
[p
t+t
m
t+t
] = E
t
_
t
s=0
m
t+s
D
t+s
ds (I-II)
p
t
m
t
= E
t
_
t
s=0
m
t+s
D
t+s
ds + E
t
[p
t+t
m
t+t
]
for t small, we have
p
t
m
t
m
t
D
t
t + E
t
[p
t+t
m
t+t
]
0 m
t
D
t
t + E
t
[p
t+t
m
t+t
p
t
m
t
]
Taking the limit t dt
0 = m
t
D
t
dt + E
t
[d(p
t
m
t
)]
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 16 / 35
Interpretation
0 = m
t
D
t
dt + E
t
[d(m
t
p
t
)]
is the continuous-time counterpart of p = E[mx].
The continuous-time counterpart looks quite dierent than the
discrete-time version, where we relate prices to expected discounted
payos - or expected prices.
The continuous-time is an expression in dierences in prices over
small time intervals dt.
In equilibrium,
marginal utility from (dividend) payments of an asset (m
t
D
t
dt) and
expected change in marginal utility of the assets price (E
t
[d(m
t
p
t
)])
should oset each other. Example: Dividend payment.
Its really the same!
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 17 / 35
Interpretation
With no dividends and a constant (non-stochastic) SDF
0 = m
t
D
t
dt + E
t
[d(m
t
p
t
)] (1)
reduces to
0 = E
t
[dp
t
] = E
t
[p
t+dt
p
t
]
At the latest, in three slides this will become clear.
Interpretation: prices follow a martingale.
Hence, 0 = E
t
[d(m
t
p
t
)] means that marginal utility-weighted prices
should follow a martingale.
Equation (1) adjusts in addition for dividends.
It is hence the same as the equivalent discrete time expression:
p
t
= E
t
[m
t+1
(p
t+1
+ d
t+1
)]
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 18 / 35
Martingale
A stochastic process x
t
is a martingale, if
For all t
E[|x
t
)|] <
For all s and t with s t, we have
E
s
[x
t
] = x
s
Note that a martingale is not automatically a Markov process, and a
Markov process is not automatically a martingale.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 19 / 35
Stochastic discount factor
The stochastic discount factor m
t
itself is a stochastic process. It is
the marginal-rate-of-substitution process.
It is convenient to break up d(m
t
p
t
) using Itos lemma:
d(m
t
p
t
) = p
t
dm
t
+ m
t
dp
t
+ dp
t
dm
t
We can insert this expanded version into the pricing equation:
0 = m
t
D
t
dt + E[d(m
t
p
t
)]
= m
t
D
t
dt + E
t
[p
t
dm
t
+ m
t
dp
t
+ dp
t
dm
t
]
=
D
t
p
t
dt + E
t
_
dm
t
m
t
+
dp
t
p
t
+
dp
t
p
t
dm
t
m
t
_
where we divided both sides by p
t
m
t
.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 20 / 35
Interpretation
0 =
D
t
p
t
dt + E
t
_
dm
t
m
t
+
dp
t
p
t
+
dp
t
p
t
dm
t
m
t
_
=
D
t
p
t
dt + E
t
_
dm
t
m
t
_
+ E
t
_
dp
t
p
t
_
+ E
t
_
dp
t
p
t
dm
t
m
t
_
D
t
p
t
dt: dividend yield over time interval dt
E
t
_
dp
t
p
t
_
: expected capital gain = expected percentage change in
prices
E
t
_
dm
t
m
t
_
: expected percentage change in the SDF. Interpretation?
E
t
_
dp
t
p
t
dm
t
m
t
_
: covariance between SDF and price dynamics... Why?
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 21 / 35
Covariance
Usually, the covariance is given by:
cov(x, y) = E[xy] E[x]E[y]
Here, we have:
cov
_
dp
t
p
t
,
dm
t
m
t
_
= E
_
dp
t
p
t
dm
t
m
t
_
E
_
dp
t
p
t
_
E
_
dm
t
m
t
_
Note that in the Merton-Black-Scholes model, we have
E
_
dp
t
p
t
_
= dt
and
E
_
dm
t
m
t
_
= r
f
t
dt
as we will see very soon. Hence,
E
_
dp
t
p
t
_
E
_
dm
t
m
t
_
= r
r
t
dt
2
= 0
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 22 / 35
Covariance
The covariance between SDF dynamics and price dynamics
E
t
_
dp
t
p
t
dm
t
m
t
_
This way of displaying the covariance has a nice consequence in the
continuous-time set up:
Means have no impact on the covariance!
To see this, consider the two processes:
dm
t
m
t
= a()dt + b()dz
dp
t
p
t
= c()dt + d()dz
when multiplying, only the dz terms remain:
dp
t
p
t
dm
t
m
t
= b()d()dt
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 23 / 35
III4. Risk-free rate
Using the pricing equation
0 =
D
t
p
t
dt + E
t
_
dm
t
m
t
+
dp
t
p
t
+
dp
t
p
t
dm
t
m
t
_
and the denition of the risk-free rate
dp
t
p
t
= r
f
t
dt
we can obtain the following expression:
0 = 0 + E
t
_
dm
t
m
t
+ r
f
t
dt + 0
_
r
f
t
dt = E
t
_
dm
t
m
t
_
Why is the covariance 0?
This is the continuous-time counterpart of
R
f
t
=
1
E
t
[m
t+1
]
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 24 / 35
III5. Risky assets
We can also rearrange the pricing equation for the expected total net return
of any risky asset:
0 =
D
t
p
t
dt + E
t
_
dm
t
m
t
_
+ E
t
_
dp
t
p
t
_
+ E
t
_
dp
t
p
t
dm
t
m
t
_
E[dr
t
] = E
t
_
dp
t
p
t
_
+
D
t
p
t
dt = r
f
t
dt E
t
_
dp
t
p
t
dm
t
m
t
_
where the last term is again the covariance of the instantaneous price return
and the marginal utility dynamics.
Total returns are given by the risk free rate plus a risk-correction
(risk-premium).
Assets that co-vary negatively with the SDF have higher expected returns
and thus lower prices.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 25 / 35
Risky assets
Expected excess returns are given by the risk premium, i.e., the
covariance only.
E
t
_
dp
t
p
t
_
+
D
t
p
t
dt r
f
t
dt = E
t
_
dp
t
p
t
dm
t
m
t
_
The discrete-time counterpart is given by:
E
t
[R
i
t+1
] R
f
= cov(m
t+1
, R
i
t+1
)
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 26 / 35
III6. Example: SDF
Assume the following:
Utility function: u(C
t
) = ln C
t
Dynamics of c
t
= ln C
t
: dc
t
=
c
dt +
c
dz
t
The stochastic discount factor m
t
:
Given u(C
t
) = ln C
t
, we have u

(C
t
) =
1
C
t
, u

(C
t
) =
1
C
2
t
,
u

(C
t
) =
2
C
3
t
Hence m
t
= e
t 1
C
t
From that, we can calculate the dynamics of m
t
:
dm
t
= d
_
e
t
1
C
t
_
= d
_
e
t
_
1
C
t
+ e
t
d
_
1
C
t
_
d
_
1
C
t
_
=
1
C
2
t
dC +
1
C
3
t
(dC)
2
d
_
e
t
_
= e
t
dt
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 27 / 35
Example: SDF
Using Itos lemma, we can calculate dC
t
, using dc
t
=
c
dt +
c
dz
t
:
C
t
= e
c
t
dC =
_

c
e
c
t
+
1
2

2
c
e
c
t
_
dt + e
c
t

c
dz
t
= C
t
_

c
+
1
2

2
c
_
dt + C
t

c
dz
t
dC
C
t
=
_

c
+
1
2

2
c
_
dt +
c
dz
t
dC
t
= C
t
__

c
+
1
2

2
c
_
dt +
c
dz
t
_
Furthermore we can calculate:
(dC
t
)
2
= C
2
t
__

c
+
1
2

2
c
_
dt +
c
dz
t
_
2
= C
2
t

2
c
dt
Note that all terms higher than dt are 0 in the limit!
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 28 / 35
Example: SDF
Using this solution, we have:
d
_
1
C
t
_
=
1
C
2
t
dC +
1
C
3
t
(dC)
2
=
1
C
2
t
C
t
__

c
+
1
2

2
c
_
dt +
c
dz
t
_
+
1
C
3
t
C
2
t

2
c
dt
=
1
C
t
__

c

1
2

2
c
+
2
c
_
dt
c
dz
t
_
=
1
C
t
__
+
1
2

2
c

c
_
dt
c
dz
t
_
Which gives:
dm
t
= d
_
e
t
1
C
t
_
= d
_
e
t
_
1
C
t
+ e
t
d
_
1
C
t
_
= e
t
dt
1
C
t
+ e
t
1
C
t
__
+
1
2

2
c

c
_
dt
c
dz
t
_
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 29 / 35
Example: Risk-free rate
Hence, since m
t
= e
t 1
C
t
:
dm
t
m
t
= dt +
_
1
2

2
c

c
_
dt
c
dz
t
=
_
+
1
2

2
c

c
_
dt
c
dz
t
Risk-free rate:
r
f
t
dt = E
t
_
dm
t
m
t
_
=
_
+
c

1
2

2
c
_
dt
or
r
f
t
= +
c

1
2

2
c
which is the same as we had in the discrete-time case:
ln(R
f
t
) = r
f
t
=

+ E
t
[ln(c
t+1
)]

2
2

2
[ln(c
t+1
)]
where

= ln(). Remember that log-utility is when = 1.

2
c

c
_
dt
c
dz
t
__
= E
t
[
p
dz
p
t

c
dz
t
] =
p

p
dt
Such that:
E
t
_
dp
t
p
t
_
+
D
t
p
t
dt =
_
+
c

1
2

2
c
+
p

p
_
dt
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 31 / 35
III7. The Sharpe ratio in continuous time
Using the continuous-time set-up, we can reexamine the relationship
between consumption growth and asset prices.
Since m
t
= e
t
u

(c
t
), we have
dm
t
= e
t
u

(c
t
)dt + e
t
u

(c
t
)dc
t
+
1
2
e
t
u

(c
t
)(dc
t
)
2
dm
t
m
t
= dt +
c
t
u

(c
t
)
u

(c
t
)
dc
t
c
t
+
1
2
c
2
t
u

(c
t
)
u

(c
t
)
(dc
t
)
2
c
2
t
We use power utility:
u(c
t
) =
1
1
c
1
t
u

(c
t
) = c

t
u

(c
t
) = c
(+1)
t
u

(c
t
) = ( + 1)c
(+2)
t
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 32 / 35
Asset correlation with consumption
Such that
dm
t
m
t
= dt +
c
t
u

(c
t
)
u

(c
t
)
dc
t
c
t
+
1
2
c
2
t
u

(c
t
)
u

(c
t
)
(dc
t
)
2
c
2
t
dm
t
m
t
= dt
dc
t
c
t
+
1
2
( + 1)
(dc
t
)
2
c
2
t
We have:
E
t
_
dp
t
p
t
_
+
D
t
p
t
dt r
f
t
dt = E
t
_
dp
t
p
t
dm
t
m
t
_
= E
t
_
dp
t
p
t
dc
t
c
t
_
Note that this simplication is due to the fact that all terms higher
than dt are equal to zero (only the dc
t
term is of order

dt before
the multiplication).
Assets that covary more strongly with consumption levels have higher
mean excess returns - times the risk aversion parameter.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 33 / 35
The Sharpe ratio
Using

p
dt E
t
_
dp
t
p
t
_

2
p
dt E
t
_
_
dp
t
p
t
_
2
_

2
c
dt E
t
_
_
dc
t
c
t
_
2
_
we have for the expected excess returns risk aversion parameter times the
covariance between prices and consumption:
E
t
[dr
e
t
] = E
t
_
dp
t
p
t
_
+
D
t
p
t
dt r
f
t
dt = E
t
_
dp
t
p
t
dc
t
c
t
_

p
dt +
D
t
p
t
dt r
f
t
dt =
c

cp
dt
Since correlation is less than 1:
E
t
[dr
e
t
]

p
=

p
+
D
t
p
t
r
f
t

p

c
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 34 / 35
III8. Solving the equity premium puzzle: Habit formation
models
Idea: The concept of simple utility from consumption might be too
simplistic. As many studies and surveys show, the utility / happiness
of consumption does not change with the consumption level.
However, we get utility (and dis-utility) from changes in consumption.
This might look as an extreme step, however.
Models in-between are so called habit models:
We tend to compare our consumption to some habit level of
consumption.
David Schroder (Birkbeck College) Consumption-based asset pricing theory 11-14 April 2011 35 / 35