Econ 202A, 10/30/2012 Guest Lecture

Professor Romer covered already consumption theory under these assump-

tions:
certainty, constant :
uncertainty about labor income, constant :
Today we do a third case: certainty and variable :.
Adding uncertainty takes us into asset pricing, which Professor Romer
will cover later.
Unlike the book, I will look at the case of continuous time.
I will use two solution methods, one analogous to section 8.1 and one
based on optimal control theory, and show they give the same answer.
Notation
1: time horizon
1 (t): labor income
C(t): consumption
(t): nancial (non-human capital) wealth
:(t): instantaneous rate of interest
The individual problem is to maximize
_
T
0
e
t
n[C(t)]dt.
where n
0
(C) 0. n
00
(C) < 0. lim
t!0
n
0
(C) = , subject to
_
(t) = :(t)(t) + 1 (t) C(t).
where the individual takes (0) and the paths :(t). 1 (t) as given.
1
Parallel to what was done in the (discrete-time) setting of the government
budget constraint earlier in the course, we work with the constraint that
the PDV of consumption _ PDV of lifetime resources. Dene the market
discount factor as
1(t) =
_
t
0
:(t)dt
(so that e
R(t)
is the present value at time 0 of a unit of output available at
time t). Then the lifetime budget constraint is
_
T
0
e
R(t)
C(t)dt _ (0)+
_
T
0
e
R(t)
1 (t)dt. (+)
Approach I: Calculus
We know it is optimal for the budget constraint to hold with equality.
Set up the Lagrangean
/ =
_
T
0
e
t
n[c(t)]dt + `
_
(0) +
_
T
0
e
R(t)
1 (t)dt
_
T
0
e
R(t)
C(t)dt
_
.
The rst-order condition for C(t) is
e
t
n
0
[C(t)] = `e
R(t)
where I have canceled the dt that multiplies both sides.
Note that ` is :ot a function of time here because there is a single con-
straint, which depends on the PDV of lifetime resources and not its particular
time-path. And notice that only if 1(t) = ot for every t is the path of con-
sumption going to be at.
Since
n
0
[C(t)] = `e
tR(t)
.
2
we can write
ln n
0
[C(t)] = ln ` + ot 1(t) = ln ` + ot
_
t
0
:(t)dt
and so we may dierentiate using the chain rule and the fundamental theorem
of calculus to get
n"[C(t)]
n
0
[C(t)]
_
C(t) = o :(t) =
_
C(t)
C(t)
=
_

C(t)n"[C(t)]
n
0
[C(t)]
_
1
[:(t) o] . (++)
We saw this same relationship in the Cass-Koopmans-Ramsey model.
Recall that

C(t)n
00
[C(t)]
n
0
[C(t)]
= coecient of relative risk aversion
= inverse of intertemporal substitution elasticity.
When this coecient is big, the utility function is more sharply curved, mean-
ing that the marginal utility of consumption drops o quickly as consumption
rises.
So consumption is:
rising whenever : o
falling whenever : < o
stationary whenever : = o (this last being our old result).
The response of consumption growth to a deviation between : and o
depends on the curvature of the utility function. It is smaller when the utility
function is more highly curved (low intertemporal substitution elasticity).
Of course, at an optimum, lifetime budget constraint (+) has to hold as
an equality. Using that PDV budget constraint and the equation for
_
C just
derived, one could solve explicitly for C(0). In general the solution is messy,
but I will derive it later in a simplied special (innite-horizon) case.
3
It is worth making explicit something that may or may not be obvious
to you already, just in case it is not. The way we derive (+) from the ow
constraint
_
(t) = :(t)(t) + 1 (t) C(t) is by integrating it forward in time
to solve for (t) (in analogy to the iterative forward substitution procedure
we followed in discrete time). The solution is
(t) =
_
t
0
[1 (:) C(:)] e
R(ts)
d: + e
R(t)
(0).
where 1(t :) =
_
t
s
:(t)dt. as you can check by dierentiating.Multiply this
through by e
R(t)
, rearrange, and conclude that
_
t
0
C(:)e
R(s)
d: + e
R(t)
(t) = (0) +
_
t
0
1 (:)e
R(s)
d:.
Setting t = 1, we see that the constraint that (1) _ 0 is the same as con-
straint (+). Furthermore, requiring that constraint (+) holds as an equality
is exactly the same as requiring that (1) = 0.
Aside on the innite-horizon case: As you saw earlier, applying the no-
Ponzi game condition that
lim
t!1
e
R(t)
(t) _ 0
leads to equation (+) above with 1 = . which states that the PDV of
C cannot exceed the PDV of 1 plus initial nancial assets (0). (Recall
Chapter 2.2 of Romer, Advanced Macroeconomics.)
Approach II: Maximum Principle
You have seen this before. The optimal control problem is to nd
max
fC(t)g
_
T
0
e
t
n[C(t)]dt.
subject to the same "givens" as before and
_
(t) = :(t)(t) + 1 (t) C(t).
(1) _ 0.
4
where the second inequality constraint states that you cannot plan to die
owing money.
We write the Hamiltonian as:
H[C(t). (t). j(t)] = e
t
n[C(t)] + j(t) [:(t)(t) + 1 (t) C(t)] .
As we saw earlier, the costate variable j(t) can be interpreted as the mar-
ginal contribution of the state variable (nancial wealth) to lifetime utility,
discounted to date t, whereas e
t
j(t) is the same value discounted to the
initial time 0.
Necessary conditions for an optimum are:
optimality of the control C:
JH
JC
= 0. \t =n
0
[C(t)] = j(t). \t
equation of motion for the costate:
d
dt
_
e
t
j(t)

=
JH
J
. \t =
_ j(t)
j(t)
=
o :(t). \t
terminal necessary condition: j(1)(1) = 0.
The rst of these three equations says that if I raise consumption by a
little bit at time t and reap n/[C(t)], the opportunity cost is the marginal
value of wealth at that same time t, j(t). The second equation is actually the
intertemporal Euler equation (in continuous time). If I reduce consumption
and raise saving on date 0 by dC, thereby forgoing n
0
[C(0)]dC. then if I am
optimizing, it does not matter at what time t I choose to consume the pro-
ceeds of my saving (including interest), thereby reaping e
t
n
0
[C(t)]e
R(t)
dC.
In particular, then, e
t
n
0
[C(t)]e
R(t)
is constant for all t and equal to n
0
[C(0)].
As a result,
d
dt
_
e
R(t)t
n
0
[C(t)]
_
= 0.
and if we denote n
0
[C(t)] = j(t). the result is
_ j(t)
j(t)
= o :(t).
5
as derived above. As for the third necessary condition, we saw how to derive
it in discrete time earlier using the Kuhn-Tucker theorem, and unsurprisingly,
it holds in continuous time as well. Unless the terminal marginal value of
wealth, j(1), is zero, it can never be optimal to die holding positive wealth
one should plan to drive terminal wealth (1) down precisely to zero. As
I pointed out above, this means that constraint (+) must hold as an equality.
In the innite-horizon setting, the analogous terminal condition to j(1)(1) =
0 (recall the Cass-Koopmans-Ramsey model) is the transversality condition
that lim
t!1
e
t
j(t)(t) = 0. The intertemporal Euler equation, however,
tells us that n
0
[C(0)] = e
R(r)t
n
0
[C(t)] = e
R(t)t
j(t), \t. so we may express
the transversality condition equivalently as
n
0
[C(0)] lim
t!1
e
R(t)
(t) = 0 = lim
t!1
e
R(t)
(t) = 0.
We can conclude that whereas the no-Ponzi game condition imposes the
constraint that lim
t!1
e
R(t)
(t) _ 0, the transversality condition tells us
that this constraint will have to bind at an optimum. Thus, also for the case
1 , constraint (+) must hold as an equality at an optimum.
Observe nally that
_ j(t)
j(t)
=
n
00
[C(t)]
n
0
[C(t)]
_
C(t). Thus, our two approaches yield
precisely the same result.
An Innite-Horizon Example
Let us take
n(C) =
C
1
1

1
1

and also assume that : is xed, but not necessarily equal to o. We also let
1 . Then the lifetime budget constraint (which we can assume will hold
as an equality at an optimum) is:
_
1
0
e
rt
C(t)dt = (0) +
_
1
0
e
rt
1 (t)dt.
6
Optimal consumption, we know follows
_
C(t)
C(t)
= o (: o) .
How can we calculate C(0) in this case? (C wont be constant unless
: = o.) Notice that the preceding dierential equation has the solution
C(t) = C(0)e
(r)t
.
Substitution into the preceding intertemporal budget constraint gives
_
1
0
e
rt
C(0)e
(r)t
dt = C(0)
_
1
0
e
[(1)r]t
dt
= (0) +
_
1
0
e
rt
1 (t)dt.
or, integrating the expression
_
1
0
e
[(1)r]t
dt. assuming that (o 1) :oo <
0.
C(0) = [oo (o 1):]
_
(0) +
_
1
0
e
rt
1 (t)dt
_
.
Here we see very clearly the substitution, income, and wealth eects on
consumption (and saving) of a change in the interest rate :.
But you are probably wondering: why did we have to assume that (o 1) :
oo < 0? This inequality will always hold if o < 1, but if o 1 and : is su-
ciently higher than o, it may not, in which case we cannot compute a nite
value for the integral
_
1
0
e
[(1)r]t
dt. The problem is that consumption
optimally is growing at the proportional rate o(: o) and if this exceeds
:, so that (o 1) : oo _ 0, then the present value of consumption is :ot
,i:itc: the budget constraint is therefore violated.
By investigating a little further this case in which (o 1) : oo _ 0, we
can gain an independent verication of our optimal consumption function, as
well as some other insights. To simplify notation, let 1 (t) = 1 , a constant,
for all t, so that the PDV of income is simply 1,:.
7
Lets assume that consumption grows at the arbitrary proportional rate
q < :, so that the PDV of consumption is well dened.
1
I would like to
gure out what level of lifetime utility this policy yields. In this case, date t
consumption will be C(t) = C(0)e
gt
and therefore fromthe budget constraint,
C(0) = (: q)
_
(0) +
1
:
_
.
On the other hand, lifetime utility is
l(0) =
C(0)
1
1

1
1

_
1
0
e
t
_
e
gt
_
1
1

dt =
_
1
0
e
[(1
1

)g]
dt.
If also
_
1
1

_
q o < 0. this implies that lifetime utility is
l(0) =
(: q)
1
1

_
1
1

_ _
o
_
1
1

_
q

_
(0) +
1
:
_
1
1

.
Take natural logs and dierentiate with respect to q to get
d lnl(0)
dq
=
_
1
1
o
_
1
: q
+
_
1
1
o
_
1
o
_
1
1

_
q
.
so that in the case o 1 that is of interest here,
d lnl(0)
dq

(: q)
_
o
_
1
1

_
q

(: q)
_
o
_
1
1

_
q
=
o(: o) q
o (: q)
_
o
_
1
1

_
q
.
where the symbol signies "is proportional to." Notice rst that setting
d lnl(0)
dq
=
0 yields the implication that if it so happens that o(: o) < :, then the
optimum growth rate of consumption the one that maximizes l(0) is
q = o(: o), which we derived above. This is yet another way of deriving
our earlier results.
1
You should recall from equation (++) that this will be true only if the utility function
is in the CRRA class (equivalently, constant intertemporal substitution elasticity).
8
But if o(: o) _ :. we have a problem. In that case, setting q = o(: o)
would violate the intertemporal budget constraint, so we cannot plan to have
that high a growth rate of consumption. So what should the consumer do?
To see the answer, notice that when o
_
1
1

_
q 0 as we have assumed,
the preceding proportionality shows that lifetime utility is increasing in q for
any q < : _ o(: o). So in this case, no optimum value of q exists. In other
words, for any value of q that is consistent with the intertemporal budget
constraint, we can always increase utility by choosing q to be slightly higher.
2
The lesson of this example is that when addressing a maximization prob-
lem, one cannot always take it for granted that a maximum actually exists!
In the case o 1. the utility function n(C) is not bounded above, so it is
plausible there might be conditions in which the innite-horizon maximiza-
tion problem cannot be solved. In contrast, n(C) is bounded above (by 0)
when o < 1, and in that case there is a well-dened maximum.
2
Notice that if g = (r) < r, then
_
1
1

_
g = (1)(r) = (1)r < 0,
so the utility integral necessarily converges.
9