Beruflich Dokumente
Kultur Dokumente
Matthew Basilico
Spring, 2013
Course Outline:
1.
2.
Lecture 1
Sequence Problem
What is the sequence problem?
Find
v(x0 )
such that
v (x0 ) =
subject to
sup
{xt+1 }
t=0 t=0
xt+1 (x)
Variable denitions
xt
F (xt , xt+1 )
function
is stationary
t F (xt , xt+1 )
with
x0
given
ln =
so
d t
dt
t
e =
Bellman Equation
What is the Bellman Equation? What are its components?
Bellman equation expresses the value function as a combination of a ow payo and a discounted
continuation payo
v(x) =
sup
x+1 (x)
Components:
Flow payo is
v()
F (x, x+1 )
v(x).
v(x+1 )
Policy Function
What is a policy function? What is an optimal policy?
A policy is a mapping from
v(x)
Notation with Sequence Problem and Bellman example: Optimal Growth with CobbDouglas Technology
How do we write optimal growth using log utility and Cobb-Douglas technology?
sup
{ct }
t=0
subject to the constraints
t ln(ct )
t=0
c, k 0, k = c + k+1 ,
and
k0
given
v(k0 ) =
such that
sup
{kt+1 }
t=0
t ln(kt kt+1 )
t=0
and
k0 given
v(k) =
sup
k+1 [0,k ]
1. Guess a solution
2. Iterate a functional operator analytically (This is really just for illustration)
3. Iterate a functional operator numerically (This is the way iterative methods are used in most
cases)
0=
F (x, x+1 )
+ v 0 (x+1 )
x+1
Envelope Theorem:
F (x, x+1 )
x
v 0 (x) =
v()
Expanded:
Accept
Reject
x :
if
if
x x
x < x
v(x) =
4. Rene Value Function using continuity of
x
v
if
if
x x
x < x
v()
v = x
x x
x < x
v(x) =
x
x
if
if
such that
v(x ) = x = Ev(x+1 )
so
x=x
x =
x=1
1
1
xf (x)dx = (x )2 +
2
2
p
x = 1 1 1 2
x f (x)dx +
x=x
x=0
Giving solution
Lecture 2
Bellman Operator
What is the Bellman operator? What are its properties as a functional operator?
Bellman operator
B,
operating on a function
(Bw) (x)
sup
w, is dened
{F (x, x+1 ) + w(x+1 )} x
x+1 (x)
Note: The Bellman operator is a contraction mapping that can be used to iterate until convergence to
a xed point, a function that is a solution to the Bellman Equation.
Hence if
Notation: When you iterate, RHS value function lags (one period). When you are not iterating, it
doesn't lag.
Operator
maps function
Hence operator
to new function
Bw
If
Function
is a xed point of
B (B
maps
to
Bv = v
(that is,
(Bv)(x) = B(x) x)
v)
Contraction Mapping
Why does
Bnw
converge as
n ?
theorem?
The reason
Bnw
Denition : Let
converges as
(S, d)
is that
is a contraction mapping.
is a
Intuition:
Bg
and
g)
A metric (distance function) is just a way of representing the distance between two functions
(e.g. maximum pointwise gap between two functions)
(S, d)
1.
2. for any
3.
B n v0
B: SS
vS
v0 S , lim B n v0 = v
ln
Blackwell's Theorem
What is Blackwell's Theorem? How is it proved?
be a contraction mapping
Let
Let
1.
(Monotonicity) if
2.
f, g C(X)
and
f (x) g(x)x X ,
(0, 1)
Then
is a constant, and
f +a
f C(X), a 0, x X
Proof:
For any
f, g C(X)
we have
such that
then
f g + d(f, g)
Bf Bg d(f, g)
Bg Bf d(f, g)
|(Bf )(x) (Bg)(x)| d(f, g) x
sup |(Bf )(x) (Bg)(x)| d(f, g)
x
f]
n
o
+1 (x c) + y+1
u(c) + Ef R
x
c[0,x]
1. Monotonicity:
f (x) g(x)x. Suppose cf is the optimal policy when the continuation value function is f .
n
o
+1 (x c) + y+1
+1 (x c ) + y+1
(Bf )(x) = sup u(c) + Ef R
= u(cf ) + Ef R
f
Assume
c[0,x]
n
o
+1 (x c) + y+1
+1 (x c ) + y+1 sup u(c) + Eg R
= (Bg)(x)
u(cf ) + Eg R
f
c[0,x]
[Note (in top line) elimination of sup by using optimal policy
cf .
to add in sup]
2. Discounting
Adding a constant
()
n
h
io
+1 (x c) + y+1 +
u(c) + E f R
c[0,x]
n
o
+1 (x c) + y+1
+ = (Bf )(x) +
= sup u(c) + Ef R
c[0,x]
v0 (x) = 0
B:
vn (x) = (B n v0 )(x) = B(B n1 v0 )(x) = max x, E(B n1 v0 )(x)
Let
xn
xn
xn E(B n1 v0 )(x+1 ),
vn (x) = (B n v0 )(x)
vn (x).
vn (x) = (B n v0 )(x)
v0 (x) = 0:
v1 (x) = x:
v2 (x) = (B 2 v0 )(x) = (Bv1 )(x) = max {x, Ev1 (x+1 )} = max {x, Ex+1 }
x2 = Ex+1 =
v2 =
x2
x
x x2
x x2
if
if
Proof at limit:
We have:
(B n1 v0 )(x) =
"
xn = E(B
n1
xn1
x
x xn1
x xn1
if
if
x=xn1
x=0
We set
xn = xn1
x=1
xn1 f (x)dx +
v0 )(x) =
xf (x)dx =
x=xn1
2
xn1 + 1
2
p
lim xn = 1 1 1 2
Lecture 3
Classical Consumption Model
1. Consumption; 2. Linearization of the Euler Equation; 3. Empirical tests without precautionary
savings eects
Classical Consumption
What is the classical consumption problem, in Sequence Problem and Bellman Equation notation?
v(x)
such that
v(x0 ) = sup E0
{ct }
0
(
subject to
Variables:
t u(ct )
t=0
is vector of assets,
is consumption,
xt+1
C
ct (x)
t+1 , yt+1 , . . .
X xt , ct , R
is vector of
labor income
Common Example:
x:
t+1 , yt+1 , . . . R
t+1 (xt ct ) + yt+1 ; x0 = y0
xt , ct , R
The only asset is cash on hand, and consumption is constrained to lie between 0 and
ct C (xt ) [0, xt ];
Assumptions:
y is
xt+1 X
is concave;
limc0 u0 (c) =
(so
c>0
as long as
x > 0)
v(xt ) =
xt+1
is chosen at time
x,
is stochastic, so
t)]
v(xt ) =
sup
n
o
t+1 (xt ct ) + yt+1
u(ct ) + Et v R
x
ct [0,xt ]
t+1 v 0 (xt+1 )
u0 (ct ) = Et R
0
t+1 )
F OCct : 0 = u0 (ct ) + Et v 0 (xt+1 ) (R
if
if
0 < ct < xt
ct = xt
i
(interior)
(boundary)
2. Envelope Theorem
v 0 (xt ) = u0 (ct )
h
h
ct
= u0 (ct )
xt : v 0 (x) = u0 (ct ) x
t
xt+1 +
yt+1
t+1
R
Putting the FOC and Envelope Condition together, and moving the Envelope Condition forward one
period, we get:
IEuler
Equation:
t+1 u0 (ct+1 )
u0 (ct ) = Et R
0
t+1 u0 (ct+1 )
u (ct ) Et R
if
if
0 < ct < xt
ct = xt
(interior)
(boundary)
1. General Intuition
= u0 (ct )
t+1
R
more
dollars tomorrow:
t+1 u0 (ct+1 )
= Et R
2. Perturbation
**Idea: at an optimum, there cannot be a perturbation that improves welfare of the agent.
If there is such a feasible pertubation, then we've proven that whatever we wrote down is not an
optimum.
Proof by contradiction: [Assume inequalities, and show they produce contradictions so cannot hold]
1. Suppose
[If
u (ct )
t+1 u0 (ct+1 )
u0 (ct ) < Et R
is less than the discounted value of a dollar saved, then what should I do? Save more,
consume less.]
(a) Then we can reduce
ct
by
and raise
ct+1
by
t+1
R
h
i
t+1 u0 (ct+1 ) u0 (ct )
0 < Et R
= this
u0 (ct )
(a) If this perturbation is possible that raises welfare, then can't be true that we started this
analysis at an optimum.
i. This perturbation is always possible along the equilibrium path.
ii. Hence, if this cannot be an optimum, we've shown that at an optimum, we must have:
t+1 u0 (ct+1 )
u0 (ct ) Et R
2. Suppose
t+1 u0 (ct+1 )
u0 (ct ) > Et R
u (ct )
ct
by
and reduce
ct+1
by
t+1
R
is greater than the value of a dollar saved, then what should I do? Save less,
consume more.]
h
i
t+1 u0 (ct+1 ) > 0
u0 (ct ) Et R
= this
(b) If this perturbation is possible that raises welfare, then can't be true that we started this
analysis at an optimum.
i. This perturbation is always possible along the equilibrium path, as long as
ct < x t
t+1 u0 (ct+1 )
u0 (ct ) Et R
10
as long as
ct < xt
It follows that:
t+1 u0 (xt+1 )
u0 (ct ) = Et R
0
t+1 u0 (xt+1 )
u (ct ) Et R
if
if
0 < ct < xt
ct = xt
(interior)
(boundary)
to do
u(c) =
Note: lim1
2. Assume
ln ct+1
Rt+1
c1 1
=ln c.
1
c1 1
1
is known at time
c
= Et Rt+1 c
t
t+1
4. [Algebra] Divide by
c
t
1 = Et Rt+1 c
t+1 ct
1 = Et exp ln Rt+1 c
t+1 ct
6. [Algebra] Distribute ln:
ln ct+1
1 2
1 = exp Et rt+1 ln ct+1 + Vt ln ct+1
2
h
i
a
E
a+ 21 V ar
a
Note: Ee = e
where a
is a random variable
2
Here we have: ln ct+1 N ( ln ct+1 , Vt ln ct+1 )
11
1
0 = Et rt+1 ln ct+1 + 2 Vt ln ct+1
2
10. [Algebra] Divide by
rearrange:
() ln ct+1 =
Terms of linearized Euler Equation
1
1
(Et rt+1 ) + Vt ln ct+1
():
t.
(i.e. conditional on being in year 49, what is the variance in consumption growth between 49 and 50)
It is also known as the precautionary savings term as seen in the regression analysis
ln ct+1
Why is ln ct+1
consumption growth?
ct+1 ct
ct+1
ct+1
ct+1 ct
ln 1 +
= ln 1 +
1 = ln
ln ct+1
ct
ct
ct
ct
[the second step uses the log approximation,
x ln (1 + x)]
Life Cycle Hypothesis (Modigliani & Brumberg, 1954; Friedman) = Eat the Pie
Problem
(a) Assumptions
i.
ii.
t = R
R
R = 1 [ < 1; R > 1]
iii. Perfect capital markets, no moral hazard, so future labor income can be exchanged for
current capital
(b) Bellman Equation
x+1 = R(x c)
x0 = E
X
t=0
12
Rt yt
[At date 0, sells all his labor income for a lump sum, discounted at rate
r.
r]
R ct = E0
t=0
Rt yt
t=0
[(Discounted sum of his consumption) must be equal to [the expectation at date 0 of (the
discounted sum of his labor income)]]
(e) Subsitute Euler Equation
R c0 = E0
t=0
Rt yt
t=0
c0 =
1
1
R
E0
Rt yt
Note:
t=0
Rt =
1
1
1 R
t=0
1
2.
1
R
is the annuity scaling term; approximately equal to the real net interest rate
t = R
R
R = 1 [ < 1; R > 1]
13
limc0 u0 (c) =
ct = Et ct+1 = Et ct+n
t+1 ; u0 (ct+1 ) = Et u0 (ct+1 ) [Since R = 1]
u0 (c) = c. u0 (ct ) = Et R
= ct = Et [ ct+1 ] = Et ct+1 = ct = Et ct+1
Et ct+1
So ct = Et ct+1 = Et ct+n implies consumption is a random walk :
i. Euler becomes:
ii.
ct =
ct+1 = ct + t+1
iii.
ct ,
ct+1 ;
any
Rt ct
t=0
[Sum from t to
Rt yt
t=0
of discounted consumption is
the
income]
i. Budget Constraint at time
t:
ct+s = xt +
s=0
Rs yt+s
s=1
[Discounted sum of remaining consumption = cash on hand (at date t) and the discounted sum of remaining labor income]
ii. Now applying expectation: [if it's true on every path, it's true in expectation ]
Et
Rs ct+s = xt + Et
s=0
(e) Subsitute Euler Equation
Rs yt+s
s=1
[ct = Et ct+s ]
Rs ct = xt + Et
s=0
Rs yt+s
s=1
!
X
1
xt + Et
Rs yt+s
ct = 1
R
s=1
t
=
t=0 R
1
1
1 R
Intuition: Just like in Modigliani, consumption is equal to interest rate scaling my total net
worth at date
1 R1 is the annuity scaling term (approximately equal
P
(xt + Et s=1 Rs yt+s ) is total net worth at date t
14
and
t+1
ln ct+1 =
[where
t+1
1
1
(Et rt+1 ) + Vt ln ct+1 + t+1
t.]
[Note: when we replace precautionary term with a constant, we are eectively ignoring its eect
(since it is no longer separately identied from the other constant term
)]
1
is EIS, the elasticity of intertemporal substitution [for this model, EIS is the inverse of
1
the CRRA; in other models EIS not
]
ln ct+1
Et rt+1 , with equation
t t+1
(a) This means, test the restriction that information available at time
sumption growth in the following regression [new
ln ct+1 = constant +
(a) For example, does the date
Xt
1
Et rt+1 + Xt + t+1
Et ln Yt+1
predict date
t+1
ln ct+1 = constant +
should be 0.]
1
Et rt+1 + Et ln Yt+1 + t+1
C. Summary of Results:
1.
[0, 0.2]
(Hall, 1988)
(a) EIS is very smallpeople are fairly unresponsive to expected movements in the interest rate;
this doesn't seem to be a big driver of consumption growth
i. the parameter that is scaling the expected value of the interest rate is estimated to
be close to 0
15
(b) But estimate may be messed up once you add in liquidity contraints. Someone who is very
liquidity constrained
EIS of 0.
(c) So what exactly are we learning when we see American consumers are not responsive to
the interest rate? [Is it just about liquidity constraints, or does it mean that even in their
absence, something deep about responsiveness to interest rate]
2.
[0.1, 0.8]
(a)
(b)
=
t+1
=
t)
the assumptions (1) the Euler Equation is true, (2) the utility function is CRRA, (3)
Vt ln ct+1
cit = Yit
Lecture 4
Precautionary Savings and Liquidity Contraints
1. Precautionary Savings Motives; 2. Liquidity Constraints; 3. Application: Numerical solution of
problem with liquidity constraints; 4. Comparison to eat-the-pie problem
Precautionary Motives
Why do people save more in response to increased uncertainty?
1
(Et rt+1 ) + Vt ln ct+1
2
= Et [ ln ct+1 Et ln ct+1 ]2
Et ln ct+1 =
where
Vt ct+1
Vt ct+1 ,
raising
Et ln ct+1
but rather
u0 (ct ) = Et u0 (ct+1 )
Et (ct+1 )
ct
> 1)
Liquidity Constraints
Buer Stock Models
What are the two key assumptions of the buer stock models? Qualitatively, what are some of their
predictions
Since the 1990s, consumption models have emphasized the role of liquidity constraints (Zeldes, Caroll,
Deaton).
1. Consumers face a borrowing limit: e.g.
ct x t
(a) This matters whether or not it actually binds in equilibrium (since desire to preserve buer
stock will aect marginal utilty as you get close to full-borrowing level)
2. Consumers are impatient:
>r
Predictions
17
W0 = E0
Rt yt
t=0
(b) Budget constraint
W+1 = R(W c)
2. Bellman Equation
(
v(W ) =
W1
+ ln W
[0, ], 6= 1
=1
if
if
c = W
1
= 1 (R1 )
Lecture 5
Non-stationary dynamic programming
18
So far we have assumed problem is stationary - that is, value function does not depend on time,
only on the state variable
E.g. Born at
t = 1,
terminate at
t = T = 40
backwards induction
We index the value function, since each value function is now date-specic
vt (x)
E.g.,
vt (x) = Et
PT
s=t
st u(cs )
vT (x),
where
In most cases, the last period is easy to calculate, (i.e. end of life with no bequest
vT (x) = u(x))
motive:
Backward Induction
How does backward induction proceed?
Since you begin in the nal period, the RHS equation is one period ahead of the LHS side
equation (opposite of previous notation with operator)
Generally we have:
vT n (x) = (B n vT )(x)
19
Lecture 6
Hyperbolic Discounting
Context:
For
< 1, Ut = u(ct ) + u(ct+1 ) + 2 u(ct+2 ) + 3 u(ct+3 ) + . . .
2.
3.
Envelope Theorem
W 0 (x) = U 0 (C(x))
FOC
equation?
and
to subsitute for
V0
dCt+1
= REt W 0 (xt+1 ) (1 )u0 (ct+1 )
dxt+1
3. Use envelope theorem
20
(a) [substitute
u0 (xt+1 )
for
W 0 (ct+1 )]
dCt+1
0
0
= REt u (xt+1 ) (1 )u (ct+1 )
dxt+1
4. Distribute
dCt+1
dCt+1
= REt
+ 1
u0 (ct+1 )
dxt+1
dxt+1
()
()
dCt+1
dxt+1 is the marginal propensity to consume (MPC).
We then see that the main dierence in this Euler equation is that it is a
discount factors
weighted average of
dCt+1
dCt+1
u0 (ct+1 )
+
1
= REt
dx
dx
t+1
t+1
| {z }
{z
}
|
short run
long run
More cash on hand will shift to to the long-run term (less hyperbolic discounting)
Sophisticated Agents
This is reected in the specication of the consumption policy function (as above)
Naive Agents
between all
periods.
(Another way of saying: he wrongly believes that he will be willing to commit to the
path that he currently sets)
He will decide today, believing that he will choose from tomorrow forward by the exponential discount function. (
reoptimizes:
21
1. Dene
(a) Continuation value function, as given by standard exponential discounting Bellman
equation:
e
Vt+1
(x) =
max
e
ct+1 [0,x]
e
u cet+1 + Et+1 Vt+2
R(x cet+1 ) + y
ct [0,x]
e
u (cn ) + Et Vt+1
(R(x cet ) + y)
2. Since he is not able to commit to the exponential continuation function, realized consumption path will be given:
(a) Realized consumption path:
(b) Solution at each time
{cn (x )} =t
is characterized by
Lecture 7
Asset Pricing
1. Equity Premium Puzzle; 2. Calibration of Risk Aversion; 3. Resolutions to the Equity Premium
Puzzle
Intuition
Classical economics says that you should take an arbitrarily small bet with any positive re-
as long as the payos of the bet are uncorrelated with your consumption path
turn...
(punchline)
Don't want to make a bet where you will give up resources in a low state of consumption (when
the marginal utility of consumption is higher) for resources in a high state of consumption (where
marginal utility of consumption is lower) [all because of curvature of the utility function]
ij = ic jc
22
where
ij = ri rj
iand j )
ic = Cov( i i , ln c)
Risk-Free Asset:
Denition: we denote the risk free return:
When
i =equities
and
j =risk
Rtf .
t 1, Rtf
is known.
equity,f = equity,c
Intuition:
equity,f ,
equity,c ,
1
2
i
I
Rt+1
, Rt+1
, . . . , Rt+1
, . . . , Rt+1
i
Rt+1
=e
2
i
rt+1
+ i it+1 21 [ i ]
where
it+1 has
unit variance
x N (, 2 ) Ax + B N (A + B, A2 2 )
E [exp (Ax + B)] = exp A + B + 21 A2 2
1
2
[Since x N (, ) E(exp(x)) = exp(E(x) + V ar(x));
2
1 2
= exp( + 2 )]
it+1 N (0, 1)
so if standard normal:
i
i
i
i it+1 + rt+1
12 [ i ]2 N ( i 0 + rt+1
12 [ i ]2 , i2 12 ) = N (rt+1
12 [ i ]2 , i2 )
i
E [Rit ] = E exp(rt+1
+ i it+1 21 [ i ]2 ) which as we've seen is now exp(a random variable)
Hence:
so
we can use
i
i
= exp(mean + 12 V ar) = exp(rt+1
12 [ i ]2 + 21 i2 ) = exp(rt+1
)
Finally we have that since for small
x: ln(1 + x) x
= 1 + x ex
i
exp(rti ) 1 + rt+1
.
ij = ic jc
i
u0 (ct ) = Et Rt+1
u0 (ct+1 )
2. Assume
is isoelastic (CRRA):
u(c) =
23
c1 1
1
= exp()
(b)
i
Rt+1
=e
(c)
2
i
rt+1
+ i it+1 12 [ i ]
u (c) = c
where
it+1 has
unit variance
(given CRRA)
i
c = Et Rt+1
c
t+1
1 i 2
i
i i
ct+1
= Et exp + rt+1 + t+1
2
"
1 = Et
1 i 2
i
exp + rt+1
+ i it+1
2
ct+1
ct
#
1 i 2
ct+1
i
1 = Et exp + rt+1
+ i it+1
exp ln
2
ct
1 i 2
i
ln (ct+1 )
1 = Et exp + rt+1
+ i it+1
2
(a) Rearrange, and note
1 i 2
i
i i
1 = Et
t+1 ln (ct+1 )
exp + rt+1 2 +
{z
}
{z
} |
|
non-stochastic
random variable
5. Take Expectation, applying the formula for the expectation of a Random Variable
Sx N (S, S 2 2 ) E [exp (Sx)] = exp S + 21 S 2 2
i i
For the random variable t+1 ln (ct+1 )
i i
i i
1
That is, E t+1 ln (ct+1 ) = exp Et [ ln (ct+1 )] + V t+1 ln (ct+1 )
2
1 i 2
1 i i
i
1 = exp + rt+1
Et [ ln (ct+1 )] + V t+1 ln (ct+1 )
2
2
6. Take Ln
i
0 = + rt+1
1 i 2
1
Et [ ln (ct+1 )] + V i it+1 ln (ct+1 )
2
2
j
i
0 = rt+1
rt+1
and
j,
and Re-write:
h
ii
2 i 1 h i i
1 h i 2
j
+
V t+1 ln (ct+1 ) V j jt+1 ln (ct+1 )
2
2
24
, Et [ ln (ct+1 )]
j
i
=
rt+1
rt+1
ii
h
2 i 1 h i i
1 h i 2
V (A + B)
i. Also note
Note, in our case:
= i
2
+ 2 V ln (ct+1 ) 2ic
j
i
rt+1
rt+1
=
i h 2
ii
2 h 2
1 h i 2
j i + 2 V ln (ct+1 ) 2ic + j + 2 V ln (ct+1 ) 2jc
2
rti rtj =
1
2
2
2
2
2
( i )\
( i ) + ( j )\
( j ) + 2 V ln (ct+1\
) 2 V ln (ct+1 ) + 2ic 2jc
j
i
ij = rt+1
rt+1
= ic jc
equity,f
equity,c
equity,f .06
equity,c .0003
=
.06
= 200
.0003
25
equity,f
= equity,c
to isolate
Lecture 8
Summary Equations
x x(t + t) x(t) =
+h
h
p
q =1p
with prob
with prob
t
t (p
Calibration
h = t
h
i
p = 12 1 +
t , (p q) =
E [x(t) x(0)] =
V [x(t) x(0)] =
t
t (p
t
t
t
q)h =
t =
t
|{z}
linear drift
2 t
|{z}
linear variance
Weiner
1.
z = t
where
N (0, 1),
Vertical movements
are independent
Ito:
1. Random Walk:
dx = dt + dz
z N (0, t)]
[also stated
Ito's Lemma:
b(x, t)dz
| {z }
standard deviation
[with drift
and variance
dx = xdt + xdz
2 ]
and proportional
2 ]
z(t)
is Wiener,
x(t)
is Ito with
Let
V = V (x, t),
then
1 V
2
dV = V
dt + V
x dx + 2 x2 b(x, t) dt
h t
i
V
1 2V
2
= V
dt +
t + x a(x, t) + 2 x2 b(x, t)
V
x
b(x, t)dz
V
V
1 2V
V
2
dV = a
(x, t)dt + b(x, t)dz =
+
a(x, t) +
b(x, t) dt +
b(x, t) dz
t
x
2 x2
|x {z }
{z
}
|
dV =
V
t
dt +
1 2V
2 t2
(dt)2 +
V
x
dx +
1 2V
2 x2
(dx)2 +
2V
x2
dxdt + h.o.t.`
26
in
continuous time
t.
t 0.
At every
intervals, a process
x(t)
+h
h
x x(t + t) x(t) =
with prob
with prob
p
q =1p
V (x) = E [x Ex]
2
2
2
We have: V (x) = E [x Ex] = E (x)
[Ex] = 4pqh2
2
2
(Since)E (x)
= ph2 + q (h) = h2
Denition of Variance of a random variable:
Analysis of
in this process
x(t) x(0)
= x(t) x(0)
implies
is
n
k
x(t) x(0)
t
t steps in
n=
x(t) x(0)
is a certain combination of
and
h:
k nk
p q
n
k
(where
n!
k!(nk)! )
t
t (p
q)h (*)
x.
times each
V [x(t) x(0)] = n 4pqh2 =
t
2
t 4pqh (*)
when we look at sum of random variables, where random
variables are all independent, the variance of the sum is the sum of the variances)
We have binomial random variable. As we chop into smaller and smaller time steps, will converge to
normal density.
So far: we've taken the discrete time process, discussed the individual steps
and then we've aggregated across many time steps to an interval of length
are
n steps n =
t
t , and we've observed that this random variable,
27
t
t
2
t (p q)h and variance t 4pqh . Our next job is to move from discrete number
of steps to smaller and smaller steps.
variable with mean
Calibrate h and p to give us properites we want as we take t 0: [Linear drift and variance]
Let:
h = t (*)
h
i
p = 12 1 +
t
(*)
These Follow:
h
i
q = 1 p = 12 1
t
(p q) =
t
t
Consider the variance (of the aggregate time random variable): t
4pqh2 .
t is going to zero; p and q are numbers that are around .5 so can think of them as constants
4 is constant, t is the thing we'd like everything to be linear with respect to
in the limit,
No other way to calibrate this for it to make sense as a continuous time random walk
E [x(t) x(0)] =
t
t
t t =
(p q)h =
t
t
t
|{z}
linear drift
V [x(t) x(0)] =
t
t
4pqh2 =
4
t
t
2
2
1
1
t 2 t = t 2 1
t
4
2 t
|{z}
(as t 0)
linear variance
Intuition as
t 0
As take
At a point
to
0,
t,we
2 t.
[These are the four graphs simulating the process, which appear closer and closer to Brownian
motion as
gets smaller]
28
Hence this Random Walk, a continuous time stochastic process, is the limit case of a family of
stochastic processes
t (not t)
Binomial D N ormal
1
1
period > nh =
t t = t
since
(a) (Since length will be greater than the number of step sizes, time the length of each step
size)
(b) Hence length of curve is innity over any nite period of time
4.
x
t
t
t
dx
dt , doesn't exist
5.
E(x)
t
(pq)h
t
(a) So we write
t)( t)
t
E (dx) = dt
6.
V (x)
t
2
2
4( 14 ) 1(
) t t
t
(a) So we write
t,
V (dx) = 2 dt
When we let
converge to zero, the limiting process is called a continuous time random walk with
(instantaneous) drift
2 .
Wiener Process
Weiner processes: Family of processes described before with zero drift, and unit variance per unit time
Denition
If a continuous time stoachastic process
sponding to a time interval
1.
z = t
where
t,
z(t)
N (0, 1)
29
z, z ,
corre-
z N (0, t)]
piece, and
t is
is the gausian
t1 t2 t3 t4
are independent
then
A process tting these two properties is a Weiner process. It is the same process converged to in the
rst part of the lecture above.
Properties
A Wiener process is a continuous time random walk with zero drift and unit variance
z(t)
has the Markov property: the current value of the process is a sucient statistic for the
now.
z(t) N (z(0), t)
so the
variance of
Note on terminology :
Wiener processes are a subset of Brownian motion (processes), which are a subset of Ito Processes.
Ito Process
Now we generalize. Let drift (which we'll think of as
t and state x.
limt0 Ex
t equal
with arguments
and
t, a(x, t))
depend on
time
[Let
to a function
z(t)
x(t)
limt0 Ex
t = a (x, t)
2
x
limt0 Vt
= b (x, t)
i.e.
i.e.
E (dx) = a (x, t) dt
2
V (dx) = b (x, t) dt
Note that the i.e. equations are formed by simply bringing the
We summarize these properties by writing the expression for an
dt
"drift"
"variance"
to the RHS.
Ito Process :
dx = a(x, t)dt +
| {z }
drift
30
b(x, t)dz
| {z }
standard deviation
We think of
dz
x.
dz .
changes, and what you want to know is how does a little change in
z this
background
causes a
b.
dx =
a(x, t)dt
| {z }
deterministic
+ b(x, t)dz
| {z }
stochastic
Random Walk
dx = dt + dz
Random Walk with drift
and variance
dx = xdt + xdz
Geometric Random Walk with proportional drift
x
dx
dx
Can also think of as
x
= dt + dz ,
so percent change in
is a random walk
**Note on Geometric Random Walk: If it starts above zero, if will never become negative
(or even become zero): even with strongly negative drift, given its proportionality to
the
Ito's Lemma
Motivation is to work with functions that take Ito Process as an argument
Ito's Lemma
Basically just a taylor expansion
z(t) be
V = V (x, t), then
Theorem: Let
Let
dV =
x(t)
V
1 2V
V
dt +
dx +
b(x, t)2 dt
t
x
2 x2
31
=
V
V
V
1 2V
2
b(x,
t)
dt +
+
a(x, t) +
b(x, t)dz
t
x
2 x2
x
Motivation: Work with Value Functions (which will be Ito Processes) that take Ito
Processes as Arguments
We are going to want to work with functions (value functionsthe solutions of Bellman equations)
that are going to take as an argument, an Ito Process
as an argument, where
is an
Ito Process
More specically, we'd like to know what the Ito Process looks like that characterizes the
value function
Value function
V.
V
is
2nd
x,
Since we are usually looking for a solution to a question: i.e. Oil Well
x,
oil well)
V (x, t)
dx = xdt + xdz
and time
V (x, t),
where
follows an Ito
Process
and
b on V
dV = a
(V, x, t)dt+b(V, x, t)dz
will drop
V:
dV = a
(x, t)dt + b(x, t)dz
terms
Note that the hat
(
a, b)
(a, b)
Ito's Lemma gives the solution to this problem in the following form:
dV =
V
V
1 2V
V
1 2V
V
V
2
2
dt+
dx+
b(x,
t)
dt
=
+
a(x,
t)
+
b(x,
t)
dt+
b(x, t)dz
2
2
t
x
2 x
t
x
2 x
x
V
V
1 2V
V
2
dV =
+
a(x, t) +
b(x, t) dt +
b(x, t) dz
t
x
2 x2
|x {z }
|
{z
}
32
where:
dV =
V
V
2V
1 2V
1 2V
2
2
(dt)
+
(dx)
+
dxdt + h.o.t.
dt +
dx
+
t
2 t2
x
2 x2
x2
(dt) 2
[This is because
tant than
Note that
(t)2
(t) 2 ,
dt.
(t)]
2
(dz) = dt.
Since
t,
recalling
z h
making this true, this is math PhD land and not something Laibson knows or expects us to
know]
(dt) = h.o.t.
dxdt = a(x, t)(dt)2 + b(x, t)dzdt = h.o.t.
2
(dx) = b(x, t)2 (dz)2 + h.o.t. = b(x, t)2 dt + h.o.t.
Combining these results we have:
V
1 2V
V
dt +
dx +
b(x, t)2 dt
t
x
2 x2
dz is a Wiener process that depends only on t through random motion
dx is a Ito process that depends on x and t through random motion
V (x, t) is a value function (and an Ito Process) that depends on a Ito process dx
and directly on
Even if we assume
doesn't depend on
We still have
If
a(x, t) = 0
t]
E(dV ) =
is concave,
1 2V
2 x2
V (x) = ln x, hence V =
dx = xdt
+ xdz
h
Ex:
=
dt
[holding
xed,
1
x and
00
[convex
rise]
x12
V
x
b(x, t)dz
=0
b(x, t)2 dt 6= 0
i
V
1 2V
2
dt +
dV = V
t + x a(x, t) + 2 x2 b(x, t)
h
i
2
= 0 + x1 x 2x1 2 (x) dt + x1 xdz
= 12 2 dt + dz
V
t
(dx)
behaves like
Because of Jensen's inequality, if you are moving on independent variable (x) axis randomly,
expected value will go down
33
Lecture Summary
Talked about a continuous time process (Brownian motion, Weiner process or its generalization as
an Ito Process) that has these wonderful properties that is kind of like the continuous time analog of
things in discrete time are thought of as random walks. But it is even more general that this:
Has perverse property that it moves innitely far over unit time, but when you look at how it changes
over any discrete interval of time, it looks very natural.
Also learned how to study functions that take Ito Processes as arguments: key tool in these processes
is Ito's Lemma. It uses a 2nd order taylor expansion and throws out higher order terms. So we can
Lecture 9
Continuous Time Bellman Equation & Boundary Conditions
V
1 2V
V
2
+
a+
b(x, t) dt
V (x, t)dt = max w(u, x, t)dt +
u
t
x
2 x2
stopping problem.
Intuition
V (x, t)dt
| {z }
= max
u
w(u, x, t)dt
| {z }
instantaneous dividend
E [dV ]
| {z }
Derivation
Let
Let
w(x, u, t) =
Arguments:
is state variable,
x = x + x and t = t + t .
is control variable,
is time
34
t 0
n
o
1
V (x, t) = max w(x, u, t)t + (1 + t) EV (x0 , t0 )
u
V (x, t) = max
u
w(x, u, t)t
|
{z
}
payo time
ow
(1 + t)
+
step
(1 + t) EV (x0 , t0 )
|
{z
}
term:
n
o
2
tV (x, t) = max w(x, u, t)t + w(x, u, t) (t) + EV (x0 , t0 ) V (x, t)
u
Let
t 0
(dt) = 0
. Terms of order
(*)
Substitute for
E [dV ]
where
V
1 2V
V
V
2
+
a(x, u, t) +
b(x, u, t) dt +
b(x, u, t)dz
dV =
t
x
2 x2
x
V
since E
bdz = 0
x
V
V
1 2V 2
E [dV ] =
+
a+
b
dt
t
x
2 x2
V
1 2V
V
2
V (x, t)dt = max w(u, x, t)dt +
+
a+
b(x,
t)
dt
u
t
x
2 x2
and
t)
Interpretation
Sequence Problem:
et w (x(t), u(t), t) dt
t
V (x, t)dt
| {z }
= max
u
w(x, u, t)
| {z }
instantaneous dividend
35
E [dV ]
| {z }
Instantaneous capital gain
V
V
1 2V
b(x, t)2
+
a+
t
x
2 x2
Merton's Consumption
Equity: return
Invests share
r
r+
x,
dx = [rx + x c] dt + xdz
Intuition
assets.
equity premium
a = [rx + x c]
aka
xdz .
b = x
and
Distinguishing Variables
He can easily consume at a rate that is above his total stock of wealth
is bounded between
is getting an additional
Not between
and
and
Bellman Equation
V
1 2V
2
V (x, t)dt = max w(u, x, t)dt +
[rx + x c] +
(x) dt
c,
x
2 x2
Note the
cdt.
Standard Deviation term: cost to holding equities is volatility. Volatility scales with
the amount of assets in risky category:
So,
A fraction
36
(In principle, in every single instant, could be picking a new consumption level
and asset allocation level
It will turn out that in CRRA world, he will prefer a constant asset allocation
t,
c1
1 [Lecture]
It turns out that in this problem, the value function inherits similar properties to the utilty
function:
u(c) =
V (x) = x1
u(c) =
c1
1 , then it can be proved that
V (x) = x1
Assume
V (x) = x1
We need to know
The restriction
Two FOCs,
allows us to do this
u(c) =
c1 V 2 V
1 , x , x2
n
h
i o
x1
x1
= max
c,
1
V (x) = x1
V (x) = x1
Taking
V (x) = x1
and
h
i
c1
+ x [(r + )x c] x1 (x)2
1
2
c:
F OC : x x 2
F OCc : u0 (c) = c = x
x1 (x)2 = 0
Simplifying
1
c = x
=
Interpretation:
Nice result: Put more into equities the higher the equity premium
with higher RRA
Why is variance of equities replacing familiar covariance term from Asset Pricing lecture?
In Merton's world, all the stocasticity is coming from equity returns.
37
r+
+ 1 1
Special case:
So
=1
2
2 2
implies
M P C ' .05
c = x
0.06
1(0.16)2
= 2.34
Saying to borrow money so have assets of 2.34 in equity [and therefore 1.34 in liabilities]
We don't know what's wrong here. This model doesn't match up with what wise souls think you
should do.
Can we x by increasing
With
0.06
5(0.16)2
= 5, =
= 0.47
Typical retiree:
No liabilities; $600,000 of social security (bonds), $200,000 in house, less than $200,000 in 401(k)
401(k) usually 60% bonds and 40% stocks
Equity allocation is
8%
of total assets
V (x) = ln(x) +
n
h
V (x, t)dt = maxc, w(u, x, t)dt + V
x [rx + x c] +
V (x)
and
1
(x)2 x2
2
FOCs:
F OCc : c =
F OC : =
1
x
+ ln(x) = r ln() 1 +
i o
(x)2 dt
u(c):
1 2V
2 x2
2
2 2
+ ln(x)
Given that the RHS is constant, this will only hold if
be increasing in
Hence,
x,
= 1
and
0 = r ln() 1 +
= 1
and
= 1.
i
= r ln() 1 +
2
2 2
38
2
2 2
This means:
Giving us
2
2 2
+ ln() 1 +
Optimal Policy
c = x
and
Note
dx
V (x, t) = w(x, u , t) +
u = u(x, t) =optimal
V
t
V
x
a(x, u , t) +
1 2V
2 x2
b(x, u , t)2
and
Need structure from economics to nd the right solution to this equationthe solution the
makes economic sensefrom the innite number of solutions that make mathematical sense.
These restrictions are called boundary conditions, which we need to solve PDE
Name of the game in these problems is to use economic logic to nd right solution, and
eliminate all other solutions
(Any policy generates a solution, we're not looking for any solution that corresponds to any
policy, but for the solution that corresponds to the optimal policy)
[In Merton's problem, we exploited 'known' functional form of value function, which imposes
boundary conditions]
T
39
Where
(x, t)
V (x, T ) = (x, T ) x
T]
This restriction gives enough structure to pin down solution to the PDE
2. Stationary
horizon
problem
t = becomes
an
ODE
1
V (x) = w(x, u ) + a(x, u )V 0 + b(x, u )2 V 00
2
V0
V,
notation
V (x, t) = max w(x, t)t + (1 + t)1 EV (x0 , t0 ), (x, t)
where
(x, t)
if
x > x (t)
continue; if
x x (t)
stop
Assume that
V
t
V
x
a(x, t) +
1 2V
2 x2
i
b(x, t)2 dt
Value Matching
Value Matching:
V (x, t) = (x, t)
for all
x, t
such that
x(t) x (t)
40
A discontinuity is not permitted, otherwise one could get a better payo from not following
the policy
below
is above
or to have
on y axis,
on x-axis]
Smooth Pasting
Derivatives with respect to
Smooth Pasting:
Vx (x (t), t) = x (x (t), t)
Intuition
and
Proof intuition:
if slopes are not equal at convex kink, then can show that the
x (t),
then stopping at
x (t)
can't be optimal:
If there is a (convex) kink at the boundary, then the gain from waiting is in
and the cost from waiting is in
t.
41
on y axis,
on x-axis]
Lecture 10
Third Boundary Condition, Stopping Problem and ODE Strategies
x(t)
is an Ito Process,
dx = adt + bdz
i.e.
w(x) = x
(Or
w(x) = x c
Assume rm can costlessly (permanently) exit the industry and realize termination payo
=0
if
x > x
continue
if
x x
stop (exit)
=0
in below graph)
42
V (x) = max xt + (1 + t)1 EV (x0 ), 0
Where
and
is interest rate
In stopping region
x0 = x(t + t),
V (x) = 0
In continuation region
General
t 0
Substitute for
E(dV )
(dt)2 = 0.
V = x + aV 0 +
b2 00
V
2
Interpretation
Value function must satisfy this dierential equation in the continuation region
We've derived a 2nd order ODE that restricts the form of the value function
in
2nd order because highest order derivative is 2nd order, ordinary because only
depends on one variable (x),
drift from jensen's inequality
drift in argument
V
|{z}
required return
x
|{z}
dividend
z}|{
aV 0
+
|
z }| {
b2 00
V
2
{z
V is
instantly drifting
b2 00
2 V is drift resulting from brownian motion'jiggling and jaggling': it is negative
if concave, and positive if convex (as in this case).
Term
Intuition:
V (x) = = 0. This
x ]
Why? Option value love the ability take the chance at future prots. Even if drift is
negative, will still stay in, since through Brownian motion may end up protable again.
And now, we have continuum of solutions to ODE, so need restrictions to get to single solution
43
*Three variables we need to pin down: two constants in 2nd order ODE, and free boundary (x )
Value Matching:
V (x ) = 0
Smooth Pasting:
V 0 (x ) = 0
It is so unlikely that prots would become negative (and if it happened, would likely
be very negatively discounted) that value of being able to shut down is close to zero
As
gets very large, value function approaches that of the rm that does not have the
option to shut down (which would mean it would theoretically need to keep running at
major losses)
Hence
converges to the value associated with the policy of never exiting the industry
lim
x 1
V (x)
=1
x + a
limx V (x) =
x+
lim V 0 (x) =
x,
Intuition: at large
how does value function change with one more unit of price
x?
1
et x(t)dt et (x0 )(t)dt =
et 1 dt =
0
{z
} |
{z
}
|
Complete Equation
44
Reduced Equation
C(x)
is replaced by
0
F 00 (x) + A(x)F 0 (x) + B(x)F (x) = 0
Theorem 4.1
Any solution
F (x),
F1
and
F2
F1 , F2 linearly
provided
independent
Note that two solutions are linearly independent if there do not exist constants
such that
A1 F1 (x) + A2 F2 (x) = 0
Theorem 4.2
Complete Equation
V = x + aV 0 +
b2 00
V
2
Reduced equation
0 = V + aV 0 +
b2 00
V
2
45
A1
and
A2
Consider class
erx
0 = erx + arerx +
b2 2 rx
r e
2
0 = + ar +
b2 2
r
2
r=
r+
Let
p
a2 + 2b2
b2
C + er + C er
V = x + aV 0 +
b2 00
2 V
Consider specic example: payo function of policy never leave the industry
et x(t)dt
E
0
In solving for value of the policy without the expectation, we'll also need to know
t
E [x(t)] = E x(0) +
dx(t)
0
t
t
= E x(0) +
[adt + bdz(t)] = E x(0) + at +
bdz(t)
0
bdz(t) = 0
E [x(t)] = x(0) + at
(x) =
46
E [x(t)]:
udv = uv
Taking
et x(t)dt:
et [x(0) + at] dt
0
et [x(0)] dt +
=E
0
et [at] dt =
0
x(0)
+E
et [at] dt
0
So if
dv = e
and u = at
v = 1 et and du = adt
1 t
udv = 0 et (at)dt then uv vdu = [at 1 et ]
adt
0 0 e
Note
1 et ]
0 =
Think of
Then
= [at 1 et ]
0
[ 12 et a]
0
= 0 + a 12
1
=
a
x(0) +
1
V (x) =
x(t)dt = E
vdu
a
x+
V = x + aV 0 +
b2 00
2 V
1
a
a
1
b2
x+
=x+ =x+a
+
0
+ r+ x
C e
r x
+C e
with roots
r ,r =
p
a2 + 2b2
b2
a
x+
[Sum of particular solution to complete equation and general solution to reduced equation]
1
V (x) =
a
x+
47
+ C + er
+ C er
3 Conditions
Value Matching:
V (x ) = 0
Smooth Pasting:
V 0 (x ) = 0
limx V 0 (x) =
C+ = 0
[from
limx V 0 (x) =
1
]
V (x ) =
V 0 (x ) =
x +
+ C er
+ r C er
=0
=0
(1) implies
C er
= 1 x + a
r 1 x + a = 0
=
Simplifying: 1 = r x + a
Gives us
x =
Plugging in for
1
r
r =
= x +
1
r
a2 +2b2
b2
x
x =
b2
a+
a2
2b2
a
<0
()
the sum of a particular solution to the complete dierential equation and the general solution to
the reduced equation
Note: this is usually done in search problems by looking for the particular solution of the
policy never leave. Then you can get a general solution, and then use value matching and
smooth pasting to come up with the threshold value.
48
Lecture 12
Discrete Adjustment - Lumpy Investment
Plant level data suggests individual establishments do not smooth adjustment of their capital
stocks.
Instead, investment is lumpy. Ex rms build new plant all at once, not a little bit each year
1
this ratio would be
18 .
Instead average value was 14 . Hence, on average rms did 25% of investment in 1 year over
18 year period
smooth convex cost functions assume small adjustments are costlessly reversible
use ane functions, where small adjustments are costly to reverse
Lumpy Investment
Capital Adjustment Costs Notation
CU
is xed cost,
cU
CU + cu I
is variable cost
CD + cD (I).
we usually assume
cD > 0
and
I < 0,
Firm's Problem
Firm loses prots if the actual capital stock deviates from target capital stock
Deviations
(x x )
Functional form:
b
b
2
(x x ) = X 2
2
2
where
X = (x x ).
49
so we get
dX = dt + dz
where
dz
e( t)
V (X) = max E
=t
A (n) =
)
X
b 2
X d
e( (n)t) A (n)
2
n=1
CU + cU In
CD + cD |In |
if
if
In > 0
In < 0
A (n) =
nth
adjustment;
Notation:
(n) =
date of
nth
adjustment;
cost of
In =
adjustment
x :
Above x :
Below
Adjust capital at
Adjust capital at
X = U . Adjust to X = u.
X = D. Adjust to X = d.
Bellman Equation
Using Ito's
1
b
V (X) = X 2 + V 0 (X) + 2 V 00 (X)
2
2
h
i
V
V
1 2V 2
Lemma: E [dV ] =
dt
t + x a + 2 x2 b
[If
X U]
V (X) = V (u) [CU + cU (u X)]
This implies
V 0 (X) = cU
X U
[If
X D]
V (X) = V (d) [CD + cD (X d)]
This implies
V 0 (X) = cD
50
X D
investment in
nth
Boundary Conditions
1. Value Matching
XU
XU
XD
XD
2. Smooth Pasting
XU
XU
XD
XD
Intuition:
=
When making capital adjustment, willing to move until the marginal value of moving
marginal cost
V 0 (u) = cU
V 0 (d) = cD
51
PS6 2.d
2
+2X
X2
+ 2
is a solution to the continuation region equa +
2
3
b
1 2 00
2
0
tion: V (X) = X + V (X) + V (X).
2
2
Show that this is the expected present value of the rm's payo stream assuming that adjustment
V (X) = 2b
This is guess and check, where we've been provided the guess:
b
V (X) =
2
Calculate
V 0 (X)
and
X2
+ 2X
22
+
+
2
3
V 00 (X):
b
V (X) =
2
0
2X
22
+ 2
, V 00 (X) =
1 2 00
2
0
V (X) = 1 b
2 X + V (X) + 2 V (X) :
1 b 2
b 2X
22
1 2b
b
2X
2
2 2
2
V (X) =
X
+ 2
=
X +
+
+ 2
2
2
2
2
52
2
b X2
2 + 2X
+
+
2
2
3
u =
and
d = ,
X R.
|x| < 1:
xk =
k=0
1
1x
L'Hopital's Rule
If
xc
xc
and
f 0 (x)
xc g 0 (x)
lim
and
then
or
exists
g 0 (x) 6= 0
f (x)
f 0 (x)
= lim 0
xc g(x)
xc g (x)
lim
Probability
Expectation of RV:
E(exp(
a)) = exp[E[
a] + 12 V ar[
a]].
When a
is a random variable, or anything
x N (, 2 ) Ax + B N (A + B, A2 2 )
E [exp (Ax + B)] = exp A + B + 12 A2 2
1
2
[Since x N (, ) E(exp(x)) = exp(E(x) + V ar(x));
2
so if standard normal:
= exp( + 21 2 )]
Rit
Rit = exp(rit + i it 12 [ i ]2 )
i
i
i
where t has unit variance. Hence t is a normally distributed random variable: t N (0, 1)
1 i 2
1 i 2
i
t
i2
i i
t
From above we see that this implies: t + ri [ ] N ( 0 + ri [ ] ,
12 ) =
2
2
1 i 2
i2
2 [ ] , )
We have:
53
N (rit
Then:
Covariance:
If A and B are random variables, then
x:
ln(1 + x) x
1 + x ex
For small
54