Problem Set 1 - Econ 202A (Second Half )

Prof. David Romer

GSI: Victoria Vanasco

Exercise 1. Consumption under Uncertainty

The basic model of consumption under uncertainty (with quadratic utility, and uncertainty only about labor income)
predicts that:
A. The change in income will not be predictable on the basis of past changes in consumption.
B. The change in consumption will not be predictable on the basis of past changes in income.
C. The change in consumption will not be correlated with the current change in income.
D. (A) and (B).
E. (A) and (C).
F. (B) and (C)

Answer.

Note that option (A) is false. Our basic model of consumption under uncertainty predicts that changes in con-
sumption occur when news about changes in income are received. In this sense, changes in consumption due to an
expected future increase in income would help us predict the change in income. In other words, if we see consump-
tion jumping, with no change in current income, we should expect an increase in future income. Thus, changes in
income can be predicted on the bases of past changes in consumption. This rules out answers (A), (D), and (E).
Option (B) is true. When β (1 + r) = 1, our basic model of consumption with quadratic utility tells us that
consumption follows a random walk, which means that the expected change in consumption is zero. Therefore,
any changes in consumption should be unpredictable, and thus we should not be able to predict them on the basis
of past changes in income. If β(1 + r) 6= 1, we might be able to predict that consumption is growing or decreasing,
but note that this expected changes in consumption are unrelated to changes in income and have more to do with
the relative prices of consumption today vs. tomorrow.
Option (C) is tricky. On the one hand, if the changes in current income were unpredictable, we should observe
a positive correlation between the changes in consumption today and the change in current income. On the other
hand, if the changes in current income were predictable, consumption already adjusted when news about this change
arrived, and thus should not react to the actual change in current income; in this case, the correlation should be
zero. Since both cases are possible, I would say we should observe some correlation between changes in consumption
and current income, and that this correlation reects the unanticipated shocks to income. So I would say (C) is
false.
Finally, (B) is the correct answer.

Exercise 2. Consumption with state-contingent good

Consumer maximizes the expected present value utility:

∞ 
" t #
X 1
E U (cst ) , U ′ > 0 and U ′′ < 0
t=0
1 + δ

1
 Consumer can purchase state-contingent goods, where the price of one consumption good at time t state s is
given by pst . Let S be the space of all possible states. Given that the probability of state s at time t is given by
πst , and assuming agents' income is also contingent on the state, we can write the agent's problem as:

∞ X  t
X 1
max πst U (cst )
{cst }∞
t=0,s∈S
t=0 s∈S
1+δ
s.t.
X∞ X ∞ X
X
pst cst ≤ pst Yst
t=0 s∈S t=0 s∈S

a. The consumer's maximization problem is:

∞ X ∞ X ∞ X
 t " #
X 1 X X
L= max πst U (cst ) − λ pst cst − pst Yst
{cst }∞
t=0,s∈S
t=0 s∈S
1+δ t=0 s∈S t=0 s∈S

Taking FOC wrt (cst ) we get:  t

1
πst U ′ (cst ) = pst λ, ∀t, ∀s
1+δ
b. Consumption at time t will be the same for states s and s′ if:
U ′ (cst ) = U ′ (cs′ t )
 t  t
1 1
πst 1+δ πs′ t 1+δ
=
λpst λps′ t
πst πs ′ t
=
pst p s′ t
The ratio between the probability of the state occurring and the price of consumption in that state is equalized
across states.
c. In this case, let t′ > t WLOG:
U ′ (cst ) = U ′ (cs′ t′ )
 t  t ′
1 1
πst 1+δ πs′ t′ 1+δ
=
λpst λps′ t
 t′ −t
πst πs ′ t 1
=
pst p s′ t 1 + δ

In this case we need to also consider the discount rates, but the condition is the same as before (after controlling
for discount).
d. In consumers ONLY dier in the Yst′ s, the statement of the question is true. The proof goes as follows:
Since consumers do not dier in anything else, their FOC with respect to cst are the same, but they might have
dierent λ:

2
  t
1
U ′ c1st pst λ1


πst =
1+δ
 t
1
U ′ c2st pst λ2


πst =
1+δ

Dividing one FOC by the other we get (and note that it is key for this that prices, probabilities and discount
rates are the same for both agents):
′ 1 ′ 2



U cst U cst
= , ∀t, ∀s
λ1 λ2
So, if in one period c1st > c2st ⇒ U ′ c1st < U ′ c2st (using the fact that U ′′ < 0). Using our last condition, we

can see that this implies:

λ 1 < λ2

Since this condition holds ∀s, ∀t, we can then conclude that if c1st > c2st for one particular s and at some point
in time t, then:
U ′ c1st U ′ c2st , ∀t, ∀s ⇒



<
c1st > c2st , ∀t, ∀s

QED.
The intuition for this result is the following. Agents sell all claims to their future income at the beginning of
their lives and proceed to buy consumption for all the future periods and states with this income. So they start
their lives with lifetime income or wealth:
∞ X ∞ X
pst Yst1 and
X X
pst Yst2
t=0 s∈S t=0 s∈S

The fact that their Yst dier simply implies that one might be wealthier than the other. Since in everything else
they are identical, they would both behave the same way if their wealth was the same. If in any period we observe
agent 1 consuming more than the other, this must mean that agent 1 had higher wealth initially (and thus for this
consumer the shadow value of wealth is less λ1 < λ2 ) and was able to buy more consumption for all periods. In
other words, both consumers substitute intertemporally and intratemporally following the same rules, so we should
expect dierences only to come from dierences in the budget constraint (lifetime wealth).
e. Let's take logs of FOC for agent 1:
log (πst ) − t log (1 + δ) + log U ′ c1st = log pst + log λ1

And see how this changes with time (no derivative since we're working with discrete time, so let ∆xt = xt+1 −xt ):

U ′′ c1st c1st ∆c1st

1
log (1 + δ) + = ∆pst
U ′ (c1st ) c1st pst
∆c1 ∆pst
RRA. 1st = δ−
cst pst

3
 Using the fact that since δ < 1 ⇒ log (1 + δ) ≈ δ.
Idem for agent 2:
∆c2st ∆pst
RRA. =δ−
c2st pst
Dividing one by the other we get:
∆c1st ∆c2
1 = 2st
cst cst
This tells us that if consumers have the same constant relative risk aversion, consumption growth rates should
be perfectly correlated between them. This is, if consumer 1 increases her consumption between period st and
s′ t′ , then so will consumer 2, and by the same proportion. This implies that the ratio of consumption between
1
consumer 1 and consumer 2 across time should be constant and given by cc20 (since increments will be the same
0
across consumers, what determines this ratio are initial consumptions).
f. In practice, we often see consumption reversals, one consumer might initially have consumption higher than
another, but later lower consumption. How do we explain this with this model? We can't. Let's see which
assumptions of the model we can relax to explain this fact:
1. Complete markets: agents can buy contingent claims to consumption in the future.
This is one of the strongest assumptions of this model. Let's change this assumption, and go to the other
extreme: not only consumers cannot buy contingent goods, but they cannot participate in nancial markets of any
kind. This is, consumers cannot save nor borrow. In this case, all consumers can do is consume their income:
cst = Yst . Clearly in this situations, consumption would uctuate as income does, and we could observe reversals
whenever there is a reversal of their incomes.
2. Again complete markets! But now let's focus on the uncertainty part. By having complete markets,
we're making uncertainty irrelevant in the model, since by trading consumption goods, agents are not exposed to
idiosyncratic risk. However, we do think uncertainty plays a role in observed consumption patterns. So let's analyze
a dierent way in which this assumption fails to explain reality. Let's assume agents do have access to nancial
markets, they can save and borrow. However, they face uncertainty about their future income. To simplify even
further the analysis, certainty equivalence holds in this model (we have quadratic utility for example). In this case,
agents' decisions to lend and borrow will be based on expected future incomes. Now, if in one period one of the
consumers receives a very high positive shock to income (receives much more income than expected), then she will
save and adjust her consumption upwards. If previously to the shock this consumer was consuming less than the
other (because she expected lower income than the other), then we would observe a reversal. This is, in a world
without complete markets, but with access to nancial markets like the ones we know, unexpected income changes
can explain reversals as the one previously described.
3. I would continue talking about the problems of assuming complete markets...but let's think about something
else: unique state distribution. In the model we assumed that the probability of a state happening at a
given time is given by πst , which is known by everyone, everyone agrees. Maybe this is not realistic, we don't
see people agreeing about the probabilities of future outcomes, if anything, we see people discussing about future
outcomes a lot! So you can see that if we remove this assumption, and consider πst 1 2
6= πst , nothing guarantees that
consumptions grow at the same rate, and we might observe reversals in periods and in states where there was a
large initial disagreement.
g. This will have no impact on the consumers' consumption in later periods (or any period). Remember
that consumers sold the claims to their own output in the rst period. With this money, they decided how much
to consume in every future period, at any given state, and bought the respective contingent goods to do so. So
consumption cst is pinned down initially for all s, t. Consumers decided in t = 0 how much they wanted consume at

4
time t if the good state happened, so when it actually happens, they will receive the cst they had bought initially
and consume, but this will not aect the consumption they decided initially for the periods ahead. The intuition
is that when we have complete markets, uncertainty is killed, so shocks to income do not have any impact on
consumption decisions.

Exercise 4. Consumption Choice and Present Value Hamiltonian

The Household solves the following problem:

ˆ T
max e−D(t) u (c (t)) dt
{ct }T
0 0
s.t. A (0) given
Ȧ (t) = r (t) A (t) + Y (t) − c (t)
A (T ) ≥ 0

where D (t) = τt=0 ρ (τ ) dτ .

´

a) Setup the Present Value Hamiltonian (PVH):

Ht (c (t) , A (t)) = e−D(t) [u (c (t)) + µ (t) [r (t) A (t) + Y (t) − c (t)]]

b) Conditions for optimality:

1. Optimality of Control
∂Ht
= 0, ∀t
∂c (t)
u′ (c (t)) − µ (t) = 0

2. Motion for the costate

∂ h −D(t) i ∂Ht
e µ (t) = −
∂t ∂A (t)
e−D(t) [µ̇ (t) − δ (t) µ (t)] = −e−D(t) µ (t) r (t)
µ̇ (t)
= δ (t) − r (t)
µ (t)

3. Combining 1+2, we get an expression for the growth rate of consumption:

ċ (t) r (t) − δ (t)
=
c (t) RRA

where remember RRA = − u u(c(t))c(t) , and this comes from showing that c(t) .
′′ ′′
′ (c(t))
µ̇(t)
µ(t) = − u u(c(t))c(t)
′ (c(t))
ċ(t)

5
Extra-Exercises. Permanent Income Hypothesis.

In response to an unexpected, permanent fall in her labor income, the consumer will:
Answer B. The consumer received an unexpected negative shock to it's present labor income. To make things
easier, assume that 1+δ
1
(1 + r) = 1. In this case, the optimal response is to adjust consumption by the amount of
the fall in labor income, since the change is permanent.
Remember that in the case with certainty and innite horizon (and the intuition is the same if we have T nite):
" T  s−t #
r X 1
ct = At + Ys
1+r s=t
1+r

If Ỹs = Ys + ε for all s ≥ t and the consumer nds this out at time t, she will adjust her consumption after the
news to c̃t , given by:

" T  s−t #
r X 1
c̃t = At + Ỹs
1+r s=t
1+r
" T  s−t #
r X 1
c̃t = At + (Ys + ε)
1+r s=t
1+r
" T  s−t X T  s−t #
r X 1 1
c̃t = At + ε + Ys
1+r s=t
1+r s=t
1+r
" T  s−t #
r 1+r X 1
c̃t = At + ε + Ys
1+r r s=t
1+r
" T  s−t #
r X 1
c̃t = At + Ys + ε = c t + ε
1+r s=t
1+r

We can see that answer A is not possible. As a response to a fall in permanent income, the consumer should
not be able to borrow against the future since it will violate the no-ponzi condition (it will never receive enough
labor income to repay the increased debt). This also tells us that the consumer will have to somehow reduce its
consumption as a response to a negative income shock.
As for C. and D., they both have the word gradually, this should hint that these cannot be right answers.
Remember that under our assumption of concave utility function and 1+r 1+δ this consumer prefers to smooth con-
sumption over time (this does not mean gradual, it means constant over time!). So, gradual adjustment would
imply decreasing consumption over time until some point. Given the concavity of the utility function, it is strictly
preferred to do the cut in consumption at once.

6
SOLUTIONS FROM ROMER’S BOOK

Problem 8.5

(a) Consider the usual experiment of a decrease in consumption by a small (formally, infinitesimal)
amount dC in period t. With the CRRA utility function given by
(1) u(Ct ) = Ct1-θ /(1 - θ),
the marginal utility of consumption in period t is Ct-θ. Thus the change has a utility cost of
(2) utility cost = Ct-θ dC.
The marginal utility of consumption in period t + 1 is Ct+1-θ . With a real interest rate of r, the individual
gets to consume an additional (1 + r)dC in period t + 1. This has a discounted expected utility benefit of
1
(3) expected utility benefit =
1+ ρ
[ ]
E t C t +1−θ (1 + r ) dC .
If the individual is optimizing, a marginal change of this type does not affect expected utility. This means
that the utility cost must equal the expected utility benefit or
1+ r
(4) C t −θ =
1+ ρ
[
E t C t +1−θ , ]
where we have (rather informally) canceled the dC's. Equation (4) is the Euler equation.

(b) For any variable x, elnx = x, and so we can write

[ ]
(5) E t [ C t +1−θ ] = E t e −θln C t +1 .
Using the hint in the question -- if x ~ N(µ, V) then E[ex ] = eµ eV/2 -- then since the log of consumption is
distributed normally, we have
 −θE t ln C t +1 θ2σ2 / 2 
[ ]
E t C t +1−θ = E t  e

e 
(6)
2 2
−θE ln C
t +1 e θ σ / 2 .
=e t

In the first line, we have used the fact that conditional on time t information, the variance of log
consumption is σ2. In addition, we have written the mean of log consumption in period t + 1, conditional
2 2
−θE t ln C t +1
on time t information, as Et lnCt+1 . Finally, in the last line we have used the fact that e eθ σ /2
is simply a constant.

Substituting equation (6) back into equation (2) and taking the log of both sides yields
(7) - θlnCt = ln(1 + r) - ln(1 + ρ) - θEt lnCt+1 + θ2 σ2/2.
Dividing both sides of equation (7) by (- θ) leaves us with
(8) lnCt = Et lnCt+1 + [ln(1 + ρ) - ln(1 + r)]/θ - θσ2/2.

(c) Rearranging equation (8) to solve for Et lnCt+1 gives us

(9) Et lnCt+1 = lnCt + [ln(1 + r) - ln(1 + ρ)]/θ + θσ2/2.
Equation (9) implies that consumption is expected to change by the constant amount
[ln(1 + r) - ln(1 + ρ)]/θ + θσ2/2 from one period to the next. Changes in consumption other than this
deterministic amount are unpredictable. By the definition of expectations we can write
(10) Et lnCt+1 = lnCt + [ln(1 + r) - ln(1 + ρ)]/θ + θσ2/2 + ut+1 ,
where the u's have mean zero and are serially uncorrelated. Thus log consumption follows a random walk
with drift where [ln(1 + r) - ln(1 + ρ)]/θ + θσ2/2 is the drift parameter.

(d) From equation (9), expected consumption growth is

(11) Et [lnCt+1 - lnCt ] = [ln(1 + r) - ln(1 + ρ)]/θ + θσ2/2.
Clearly, a rise in r raises expected consumption growth. We have
∂ E t [ ln C t +1 − ln C t ] 1 1
(12) = > 0.
∂r θ (1 + r )
Note that the smaller is θ the – the bigger is the elasticity of substitution, 1/θ – the more that consumption
growth increases due to a given increase in the real interest rate.

An increase in σ2 also increases consumption growth since

∂ E t [ ln C t+1 − ln C t ] θ
(13) = > 0.
∂ σ2 2
It is straightforward to verify that the CRRA utility function has a positive third derivative. From
equation (1), u '(Ct ) = Ct-θ and u ''(Ct ) = -θCt-θ-1.. Thus
(14) u '''(Ct ) = -θ(-θ - 1)Ct-θ-2 = (θ2 + θ)Ct-θ-2 > 0.
So an individual with a CRRA utility function exhibits the precautionary saving behavior explained in
Section 7.6. A rise in uncertainty (as measured by σ2, the variance of log consumption) increases saving
and thus expected consumption growth.

EXTRA EXERCISES

Exercise 8.3
(a) We need to find an expression for [(Ct+2 + Ct+3 )/2] - [(Ct + Ct+1 )/2]. We can write Ct+1 , Ct+2 and Ct+3
in terms of Ct and the e's. Specifically, we can write
(1) Ct+1 = Ct + et+1 ,
(2) Ct+2 = Ct+1 + et+2 = Ct + et+1 + et+2 , and
(3) Ct+3 = Ct+2 + et+3 = Ct + et+1 + et+2 + et+3 ,
where we have used equation (1) in deriving (2) and equation (2) in deriving (3). Using equations (1)
through (3), the change in measured consumption from one two-period interval to the next is
C t +2 + C t + 3 C t + C t +1 ( C t + e t +1 + e t +2 ) + ( C t + e t +1 + e t +2 + e t +3 ) C t + ( C t + e t +1 )
(4) − = − ,
2 2 2 2
which simplifies to
C t +2 + C t + 3 C t + C t +1 e t + 3 + 2e t +2 + e t +1
(5) − = .
2 2 2

(b) Through similar manipulations as in part (a), the previous value of the change in measured
consumption would be
C t + C t +1 C t − 2 + C t −1 e t +1 + 2e t + e t −1
(6) − = .
2 2 2
Using equations (5) and (6), the covariance between successive changes in measured consumption is
 C t + 2 + C t + 3 C t + C t +1   C t + C t +1 C t − 2 + C t −1 
(7) cov  − ,  −  =
 2 2  2 2 
 e t + 3 + 2e t + 2 + e t +1   e t +1 + 2e t + e t −1 
cov  ,   .
 2  2 
Since the e's are uncorrelated with each other and since et+1 is the only value of e that appears in both
expressions, the covariance reduces to
 C t + 2 + C t + 3 C t + C t +1   C t + C t +1 C t −2 + C t −1  σe 2
(8) cov  − ,  −  = ,
 2 2  2 2  4
where σe2 denotes the variance of the e's. So the change in measured consumption is correlated with its
previous value. Since the covariance is positive, this means that if measured consumption in the two-
period interval (t, t + 1) is greater than measured consumption in the two-period interval (t - 2, t - 1),
measured consumption in (t + 2, t + 3) will tend to be greater than measured consumption in (t, t + 1).
When a variable follows a random walk, successive changes in the variable are uncorrelated. For
example, with actual consumption in this model, we have Ct - Ct-1 = et and Ct+1 - Ct = et+1 . Since et and
et+1 are uncorrelated, successive changes in actual consumption are uncorrelated. Thus if Ct were bigger
than Ct-1 , it would not mean that Ct+1 would tend to be higher than Ct . Since successive changes in
measured consumption are correlated, measured consumption is not a random walk. The change in
measured consumption today does provide us with some information as to what the change in measured
consumption is likely to be tomorrow.

(c) From equation (5), the change in measured consumption from (t, t + 1) to (t + 2, t + 3) depends on et+1
, the innovation to consumption in period t + 1. But this is known as of t + 1, which is part of the first
two-period interval. Thus the change in consumption from one two-period interval to the next is not
uncorrelated with everything known as of the first two-period interval. However, it is uncorrelated with
everything known in the two-period interval immediately preceding (t, t + 1). From equation (5), et+3 ,et+2
and et+1 are all unknown as of the two-period interval (t - 2, t - 1).

(d) We can write Ct+3 as a function of Ct+1 and the e's. Specifically, we can write
(9) Ct+3 = Ct+2 + et+3 = Ct+1 + et+2 + et+3 .
Thus the change in measured consumption from one two-period interval to the next is
(10) Ct+3 - Ct+1 = Ct+1 + et+2 + et+3 - Ct+1 = et+2 + et+3 .
The same calculations would yield the previous value of the change in measured consumption,
(11) Ct+1 - Ct-1 = et + et+1 .
And so the covariance between successive changes in measured consumption is
(12) cov[(Ct+3 - Ct+1 ), (Ct+1 - Ct-1 )] = cov[(et+2 + et+3 ), (et + et+1 )].
Since the e's are uncorrelated with each other, the covariance is zero. Thus measured consumption is a
random walk in this case. The amount that Ct+1 differs from Ct-1 does not provide any information about
what the difference between Ct+1 and Ct+3 will be.

Exercise 8.6
(a) Substituting the expression for consumption in period t, which is
r  ∞ E [Y ] 
t t +s
(1) C t = A t + ∑ ,
1+ r  s= 0 (1 + r ) 
s

into the expression for wealth in period t + 1, which is

(2) At+1 = (1 + r)[At + Yt - Ct ],
gives us
 r r  E t Yt +1 E t Yt + 2 
(3) A t +1 = (1 + r )  A t + Yt − At −  Yt + + +Κ  .
 1+ r 1+ r  1+ r (1 + r ) 2 
Obtaining a common denominator of (1 + r) and then canceling the (1 + r)'s gives us
E Y E t Yt +2 
t t +1
(4) A t +1 = A t + Yt − r  + +Κ .
 1+ r (1 + r ) 2 
Since equation (1) holds in all periods, we can write consumption in period t + 1 as
r  ∞ E
t +1 [ Yt +1+ s ]

(5) C t +1 =  A t +1 + ∑ .
1+ r  s= 0 (1 + r ) s 
Substituting equation (4) into equation (5) yields
r  E Y E Y   E Y 
(6) C t +1 =  A t + Yt − r  t t +1 + t t + 2 +Κ  +  E t +1Yt +1 + t +1 t + 2 +Κ  .
1 + r   1+ r (1 + r ) 2   1+ r 
Taking the expectation, conditional on time t information, of both sides of equation (6) gives us
r  E Y E Y   E Y 
(7) E t C t +1 =  A t + Yt − r  t t +1 + t t + 2 +Κ  +  E t Yt +1 + t t +2 +Κ  ,
1 + r   1+ r (1 + r ) 2   1+ r 
where we have used the law of iterated projections so that for any variable x, Et Et+1 xt+2 = Et xt+2 . If this
did not hold, individuals would be expecting to revise their estimate either upward or downward and thus
their original expectation could not have been rational. Collecting terms in equation (7) gives us
r   r   1 r  
(8) E t C t +1 =  
A t + Yt − 1 −  E Y + −  E Y +Κ  ,
1 + r   1 + r  t t +1  1 + r (1 + r ) 2  t t + 2 
which simplifies to
r  E t Yt +1 E t Yt + 2 
(9) E t C t +1 =  A t + Yt + + + Κ .
1+ r  1+ r (1 + r ) 2 
Using summation notation, and noting that Et Yt = Yt , we have
r  ∞ E Y
t t +s

(10) E t C t +1 = A t + ∑ .
1+ r  s
s= 0 (1 + r ) 
The right-hand sides of equations (1) and (10) are equal and thus
(11) Et Ct+1 = Ct .
Consumption follows a random walk; changes in consumption are unpredictable.

Since consumption follows a random walk, the best estimate of consumption in any future period is
simply the value of consumption in this period. That is, for any s ≥ 0, we can write
(12) Et Ct+s = Ct .
Using equation (12), we can write the present value of the expected path of consumption as
∞ E [C ∞ ∞
t t +s ] Ct 1
(13) ∑ s
= ∑
s
= C t ∑
s
.
s= 0 (1 + r ) s= 0 (1 + r ) s= 0 (1 + r )
Since 1/(1 + r) < 1, the infinite sum on the right-hand side of (13) converges to 1/[1 - 1/(1+r)] = (1 + r)/r
and thus
∞ E [C
t t +s ] 1 + r
(14) ∑ s
= Ct.
s= 0 (1 + r ) r
Substituting equation (1) for Ct into the right-hand side of equation (14) yields
∞ E [C
t +s ] r 1 + r   ∞ E [Y ] 
t t +s
∞ E [Y ]
t t +s
(15) ∑ ∑ ∑
t
=    A t +  = A t + .
s= 0 (1 + r )
s 1 + r  r   s= 0 (1 + r ) 
s
s= 0 (1 + r )
s

Equation (15) states that the present value of the expected path of consumption equals initial wealth plus
the present value of the expected path of income.

(b) Taking the expected value, as of time t - 1, of both sides of equation (1) yields
r  ∞ E
t −1 [ Yt + s ]

(16) E t −1C t = A t + ∑ ,
1+ r  s= 0 (1 + r )
s

where we have used the fact that At = (1 + r)[At-1 + Yt-1 - Ct-1 ] is not uncertain as of t - 1. In addition, we
have used the law of iterated projections so that Et-1 Et [Yt+s ] = Et-1 [Yt+s ]. Subtracting equation (16) from
equation (1) gives us the innovation in consumption:
 r  ∞ E t [ Yt + s ] ∞ E t −1 [ Yt + s ]  r  ∞ E t [ Yt +s ] − E t −1 [ Yt + s ] 
(17) C t − E t −1C t = ∑ − ∑  = ∑ .
1 + r  s= 0 (1 + r ) s s= 0 (1 + r )
s
 1 + r  s= 0 (1 + r ) s 
The innovation in consumption will be fraction r/(1 + r) of the present value of the change in expected
lifetime income.

The next step is to determine the present value of the change in expected lifetime income. That is, we
need to determine
∞ E t [ Yt + s ] − E t −1 [ Yt + s ]  E t Yt +1 − E t −1Yt +1   E t Yt +2 − E t −1Yt +2 
(18) ∑ = [ Yt − E Y
t −1 t ] + 
1+ r
 +  +Κ .
s= 0 (1 + r ) s  (1 + r ) 2 
In what follows, "expected to be higher" means "expected, as of period t, to be higher than it was, as of
period t - 1". We are told that ut = 1 and thus
(19) Yt - Et-1 Yt = 1.

In period t + 1, since ∆Yt+1 = φ∆Yt + ut+1 , the change in Yt+1 is expected to be φ∆Yt = φ higher. Thus the
level of Yt+1 is expected to be higher by 1 + φ. Thus
E t Yt +1 − E t −1Yt +1 1 + φ
(20) = .
1+ r 1+ r

In period t + 2, since ∆Yt+2 = φ∆Yt+1 + ut+2 , the change in Yt+2 is expected to be higher by φ∆Yt+1 = φ2.
Thus the level of Yt+2 is expected to be higher by 1 + φ + φ2. Therefore, we have
E t Yt + 2 − E t −1Yt + 2 1 + φ + φ 2
(21) = .
(1 + r ) 2 (1 + r ) 2

The pattern should be clear. We have

∞ E t [ Yt + s ] − E t −1 [ Yt + s ] 1+ φ 1+ φ + φ2 1+ φ + φ2 + φ3
(22) ∑ = 1 + + + +Κ .
s= 0 (1 + r ) s 1+ r (1 + r ) 2 (1 + r ) 3
Note that this infinite series can be rewritten as
∞ E t [ Yt +s ] − E t −1 [ Yt + s ]  1 1   φ φ φ 
(23) ∑ = 1 + + + Κ  +  + + +
(1 + r ) s  1 + r (1 + r )   1 + r (1 + r )
2 2
s= 0 (1 + r ) 3 
 φ2 φ2 
 + + Κ +Κ .
 (1 + r ) 2 (1 + r ) 3 
For ease of notation, define γ ≡ 1/(1 + r). Then the first sum on the right-hand side of (23) converges to
1/(1 - γ). The second sum converges to φγ/(1 - γ). The third sum converges to φ2 γ2 /(1 - γ). And so on.
Thus equation (23) can be rewritten as
∞ E t [ Yt + s ] − E t −1 [ Yt + s ]
(24) ∑
s= 0 (1 + r ) s
= [1
1− γ
1 + φγ + φ ]
2 2
γ + Κ =
1 1
(1 − γ ) (1 − φγ )
.

Using the definition of γ to rewrite equation (24) yields

∞ E t [ Yt + s ] − E t −1 [ Yt + s ] 1 1 (1 + r ) (1 + r )
(25) ∑ = = .
s= 0 (1 + r ) s 1 − [1 (1 + r )] 1 − [φ (1 + r )] r (1 + r − φ )
Substituting equation (25) into equation (24) gives us the following change in consumption:
r  (1 + r ) (1 + r )  (1 + r )
(26) C t − E t −1C t =  = .
(1 + r )  r (1 + r − φ )  (1 + r − φ )
(c) The variance of the innovation in consumption is
 (1 + r )   (1 + r )  2
(27) var ( C t − E t −1C t ) = var  ut  =   var( u t ) > var( u t )
 (1 + r − φ )   (1 + r − φ ) 
Since (1 + r)/(1 + r - φ) > 1, the variance of the innovation in consumption is greater than the variance of
the innovation in income. Intuitively, an innovation to income means that, on average, the consumer will
experience further changes in income in the same direction in future periods.

It is not clear whether consumers use saving and borrowing to smooth consumption relative to income.
Income is not stationary, so it is not obvious what it means to smooth it.

Exercise 8.12
Let Et denote Xt from the book.

(a) The change in purchases in period t, dEt , must leave the present value of spending unchanged, so that
(1) dEt + dEt+1 + dEt+2 = 0.
In addition, it must leave consumption in period t + 2 unchanged, or
(2) (1 - δ)2 dEt + (1 - δ)dEt+1 + dEt+2 = 0.
To see why equation (2) must hold, note that we can write the change in Ct as dCt = dEt . The change in
Ct+1 is dCt+1 = (1 - ρ)dCt + dEt+1 or substituting for dCt , we have dCt+1 = (1 - ρ)dEt + dEt+1 . The change
in Ct+2 is dCt+2 = (1 - δ)dCt+1 + dEt+2 or substituting for dCt+1 , we have dCt+2 = (1 - δ)2 dEt + (1 - δ)dEt+1 +
dEt+2 . Thus if Ct+2 is not to change, equation (2) must hold.

Thus we have two equations in two unknowns. Solving equation (1) for dEt+2 gives us
(3) dEt+2 = -dEt - dEt+1 .
Substituting equation (3) into equation (2) yields
(4) (1 - δ)2 dEt + (1 - δ)dEt+1 - dEt - dEt+1 = 0.
Expanding and collecting terms gives us
(5) dEt [1 - 2δ + δ2 - 1] + [1 - δ - 1]dEt+1 = 0,
and thus
(6) dEt+1 = (δ - 2)dEt .
Substituting equation (6) into equation (3) yields
(7) dEt+2 = -dEt - (δ - 2)dEt ,
and thus
(8) dEt+2 = (1 - δ)dEt .

(b) Since Ct = (1 - δ)Ct-1 + Et , then

(9) dCt = dEt .
Since Ct+1 = (1 - δ)Ct + Et+1 , then
(10) dCt+1 = (1 - δ)dCt + dEt+1 .
Substituting equations (9) and (6) into equation (10) gives us
(11) dCt+1 = (1 - δ)dEt + (δ - 2)dEt = -dEt .

Since only Ct and Ct+1 are changed – Ct+2 is unchanged by construction – we only need to look at expected
utility in periods t and t + 1. Since instantaneous utility is quadratic, the marginal utility of consumption
in period t is 1 - aCt . Thus the change in utility in period t is (1 - aCt )(dEt ). The marginal utility of
consumption in period t + 1 is given by 1 - aCt+1 . Since dCt+1 = -dEt , the change in expected utility in
period t + 1 is the expected value of (1 - aCt+1 )(-dEt ).
(c) For this change in expected utility to be zero – as it must be, if the individual is optimizing – we
require
(12) (1 - aCt )(dEt ) + expected value of [(1 - aCt+1 )(-dEt )] = 0.
Canceling the dEt 's (which is somewhat informal), subtracting one from both sides and then dividing both
sides by (- a) yields
(13) expected value of Ct+1 = Ct .
Thus consumption follows a random walk since changes in consumption are unpredictable. The best
estimate of consumption in period t + 1 is simply what consumption equals this period.

(d) Rearranging Ct = (1 - δ)Ct-1 + Et to solve for Et gives us

(14) Et = Ct - (1 - δ)Ct-1 .
Equation (14) holds for all periods and so we can write
(15) Et-1 = Ct-1 - (1 - δ)Ct-2 .
Subtracting equation (15) from equation (14) gives us
(16) Et - Et -1 = Ct - (1 - δ)Ct-1 - Ct-1 + (1 - δ)Ct-2 ,
which implies
(17) Et - Et-1 = (Ct - Ct-1 ) - (1 - δ)(Ct-1 - Ct-2 ).
Since consumption is a random walk, we can write
(18) Ct = Ct-1 + ut ,
where ut is a variable whose expectation as of t - 1 is zero. Using equation (18), and the fact that (18)
holds in all periods, equation (17) can be rewritten as
(19) Et - Et-1 = ut - (1 - δ)ut-1 .

Equation (19) states that the change in purchases from t - 1 to t has a predictable component – a
component that is known as of t - 1 – which is ut-1 , the innovation to consumption in period t - 1. Thus
purchases of durable goods will not follow a random walk.

As explained in Section 7.2, any change in expected lifetime resources is spread out equally among
consumption in each remaining period of the individual's life. Although we are simplifying by using a
discount rate of zero, the basic ideas are general.

Now suppose that in period t - 1, the individual's estimate of lifetime resources changes in such a way that
Ct-1 is one unit higher than Ct-2 , that is, ut-1 = 1. This also means that expected consumption in all future
periods is one unit higher than it used to be. In order to get Ct-1 up by one, purchases in period t - 1 must
be one higher than they were expected to be. But now look at the change in purchases from t - 1 to t.
From equation (19), the expectation (as of period t - 1) of the change in purchases from t - 1 to t is
-(1 - δ), since ut-1 is assumed to equal one.

Intuitively, some of the new goods purchased in period t - 1 will still be around in period t. Thus to keep
expected consumption in period t at the new higher path – one higher than it was before – it is not
expected to be necessary to buy one unit of goods all over again. The individual only has to purchase
enough to replace the fraction of the extra t - 1 purchases that depreciated, which is fraction δ. Thus
purchases in period t are expected to be less than purchases in t - 1. Specifically, they are expected to be
lower by the amount that does not depreciate, which is (1 - δ). Thus, as of period t - 1, part of the change
in purchases between t - 1 and t is predictable and thus purchases do not follow a random walk.

Now consider what happens if δ = 0, the case of no depreciation. Then from equation (19), the
expectation (as of period t - 1) of the change in purchases from t - 1 to t is -1. Now all of the new goods
purchased in period t - 1 will still be around in period t. Thus to keep expected consumption at its new
higher path – one higher than it was before – it is not expected to be necessary to purchase anything new
in period t. Thus purchases are expected to fall by the whole amount of the innovation in purchases the
previous period.