Name

Midterm Exam
180.604
Spring, 2007
Answers
You are expected to answer all parts of all questions. If you cannot solve part of a
question, do not give up. The exam is written so that you should be able to answer later
parts even if you are stumped by earlier parts.
Write all answers on the exam itself; if you run out of room, use the back of the previous
page.
 2

Part I. Short Discussion Question.

1. Effects of G in the Tractable Buffer Stock Model.
Consider the behavior of an employed consumer in the tractable buffer stock model
discussed in class. Leading up to time t this consumer has been at the steady state
target level of market resources m̄. Before the consumption decision is made in period
t, the consumer learns that G has permanently decreased. Show how the phase
diagram for the problem changes, show how the Et [∆ log Ct+1 ] diagram changes, and
draw a graph showing the time paths of c and m before and after the change in G.
Explain whether the new target m is higher, lower, or the same as before.
Answer:

An increase in G has similar effects on the phase diagram to the decrease in

τ analyzed in TractableBufferStock: For any given m the consumption
function shifts down. This reflects the fact that human wealth has declined,
because expected future income is lower.
The effect in the Et [∆ log Ct+1 ] diagram is to shift down the g locus, and
¯ > m̄.
therefore to increase the target level of m to m̄
The explanation is as follows. The lower expected consumption growth
means that consumers have less ‘impatience’ in the sense of less desire
to borrow against future income (which is lower) to finance current con-
sumption. With less impatience, they consume less at any given m. A
change in expected future G has no effect on the ratio of current M to
current P , but the target level of m is higher. To achieve this higher m the
consumer must cut the level of consumption, and experience higher-than-
before growth rates of c asymptoting back into the new steady-state growth
rate, at which the growth rate of C matches the new lower steady-state
growth rate of income.
 3

Part II. Medium Analytical Question.

Horizon-Invariance of Portfolio Shares (Samuelson (1969))
Personal finance advisors and websites often recommend that the proportion of a per-
son’s wealth invested in risky assets should decline with age. This question examines
whether that advice is explained by the benchmark CRRA Merton-Samuelson portfolio
choice model.
Consider a consumer with CRRA utility function u(c) = −c1−ρ with risk aversion
parameter ρ > 1. This consumer has assets at the end of period t equal to at and is trying
to decide how much to invest in a risky asset that earns stochastic return log Φt+1 = ϕt+1 ∼
N (ϕ, σϕ2 ) compared to a safe asset that earns a log return r = 0 (so that R = 1). (The
notational convention is that when an object like ϕ does not have a subscript, it is the
time-invariant mean of the time-varying realizations of that object, e.g. Et [ϕt+1 ] = ϕ.)

1. Suppose that for a consumer ending period t with assets at , expected value is a
function of the form

vt (at ) = −at1−ρ Et eϕ̂t+1 ς(1−ρ)

 
(1)

where ϕ̂ = ϕ is the equity premium (because r = 0) and the optimal portfolio share
in the risky asset is
ϕ̂
ς = . (2)
(ρ − 1)σϕ2

Use the fact [NormTimes] ([NormTimes]: If z ∼ N (z̄, σz2 ), then γz ∼ N (γ z̄, γ 2 σz2 ))
to show that
h 2)
i 2 2
Et eϕ̂t+1 (−ϕ̂/σϕ
= e−ϕ̂ /2σϕ . (3)

Answer:

ϕ̂t+1 ς(1 − ρ) = −ϕ̂t+1 ϕ̂/σϕ2 (4)

ϕ̂t+1 ∼ N (ϕ̂, σϕ̂2 ) (5)
ϕ̂t+1 (−ϕ̂/σϕ2 ) ∼ N (−ϕ̂(ϕ̂/σϕ2 ), (ϕ̂/σϕ2 )2 σϕ̂2 ) (6)
2
= N (−ϕ̂ /σϕ2 , ϕ̂2 /σϕ2 ) (7)

which implies that

h 2
i 2 2 2 2
Et eϕ̂t+1 (−ϕ̂/σϕ ) = e−ϕ̂ /σϕ +ϕ̂ /2σΦ (8)
2 /2σ 2
= e−ϕ̂ ϕ (9)
 4
2. Show that the value function as of the time when the consumption decision is made
is

vt (mt ) = −mt1−ρ ζ (10)

for some constant

 ϕ̂ ζ. ς(1−ρ)
(Hint:
 Solve for the consumption function using the FOC,
defining α = Et e t+1
, to show that optimal consumption is a constant fraction
γ of mt .)
Answer:
 
Defining α = Et eϕ̂t+1 ς(1−ρ) , decision-period value is

vt (mt ) = max − c1−ρ

t − βat1−ρ α (11)
ct

with FOC

c−ρ
t = (mt − ct )−ρ αβ (12)
ct = (mt − ct )(αβ)−1/ρ (13)
(1 + (αβ)−1/ρ )ct = mt (αβ)−1/ρ (14)
(αβ)−1/ρ
 
ct = mt (15)
(1 + (αβ)−1/ρ )
| {z }
≡γ

so

vt (mt ) = −(γmt )1−ρ − β((1 − γ)mt )1−ρ α (16)

= −m1−ρ
t (γ 1−ρ + (1 − γ)1−ρ αβ) (17)
| {z }
≡ζ
 5

It is a standard result in portfolio theory that if the value function in period t + 1 has
the generic form vt+1 (mt+1 ) = −m1−ρt+1 ζ and the consumer faces a portfolio investment
choice like the one outlined above, and the consumer will receive no income in period
t+1 except the income on his capital (that is, no labor or pension or transfer income),
then the optimal portfolio share to invest in the risky asset at the end of period t is
given by (2). (This is the Merton-Samuelson model).
3. Given this information, does the Merton-Samuelson model support or contradict the
advice provided by financial advisors? Explain your conclusions.
Answer:
What this result means is that the share of your total wealth allocated to
risky assets is always constant and equal to (2), at every age no matter how
far you are from T . So this contradicts the advice from financial advisors,
unless there is some other kind of wealth not encompassed in the model as
specified above.
4. One key feature of reality that is omitted from the Merton-Samuelson model is the
fact that people derive income from sources other than their portfolio of risky and
riskless financial assets. Considering human wealth as a riskless asset (and ignoring
the possibility of liquidity constraints), does the pattern of human wealth over the
lifetime suggest any pattern of portfolio shares in risky versus riskless financial assets?
Answer:
The model as described does not include human wealth. However, if hu-
man wealth were perfectly certain and there were no liquidity constraints,
it would be equivalent to current market resources, and would therefore
operate like an unobserved part of mt . Since the proportion of total wealth
accounted for by human wealth declines as people get older, if we imagine
a component ht that varies with age and is incorporated in a measure of
total wealth, the constant portfolio share in risky assets (including human
wealth) implies that as the perfectly certain human wealth declines, the
share of observed wealth invested in the risky asset may decline.
5. Another possible feature of reality omitted from the usual economic analysis is that
the coefficient of relative risk aversion might vary by age. Suppose that intrinsic
risk aversion increases with age. Would this make the model more consistent or less
consistent with the advice given by financial advisors?
Answer:
This is another potential explanation for the advice of the personal finance
gurus. The equation for the portfolio share invested in risky assets declines
directly as ρ increases (cf. (2)), so if ρ increases with age it would be natural
to expect ς to decline.
 6

Part III. Long Analytical Question.

Productivity Growth and Dynamic Inefficiency in the OLG Model.
Consider a Diamond (1965) OLG economy like the one in the handout OLGModel,
assuming logarithmic utility and a Cobb-Douglas aggregate production function,

Yt = F (Kt , Pt Lt ) (18)

where Pt is a measure of labor productivity that grows by

Pt+1 = GPt (19)

from period to period. Assume that population growth is zero (N = 1) and for convenience
normalize the population at Ls = 1 ∀ s, and assume that productivity growth has occurred
at the rate g = G − 1 forever.
One unit of the quantity P L is called an ‘efficiency unit’ of labor: It reflects a unit of
labor input to the production process.

1. Assume that F (K, P L) is a CRS function, and show how to rewrite the capital
accumulation equation

Kt+1 = A1,t (20)

in per-efficiency-unit terms as

kt+1 = a1,t /G (21)

Answer:

In an OLG economy, aggregate capital in period t+1 is equal to the savings

from period t:

Kt+1 = A1,t (22)

Kt+1 = a1,t Pt (23)
 
Kt+1
= a1,t (24)
Pt
  
Kt+1 Pt+1
= a1,t (25)
Pt+1 Pt
kt+1 = a1,t /Gt+1 (26)

2. Show that under these assumptions, the process for aggregate k dynamics is
 
(1 − ε)β
kt+1 = ktε (27)
Gt+1 (1 + β)
Answer:
 7
a1,t can be found from:
=0
z }| {
W1,t + W2,t+1 /Rt+1
C1,t = (28)
1+β
wt
c1,t = (29)
1+β
a1,t = w1,t − c1,t (30)
= w1,t (1 − 1/(1 + β)) (31)
= w1,t (β/(1 + β)) (32)
 
ε β
= (1 − ε)kt (33)
1+β

Thus (26) and (33) can be combined to yield:

=a1,t /Gt+1  
z}|{ (1 − ε)β
kt+1 = ktε (34)
Gt+1 (1 + β)
as required.
3. Derive the steady-state level of kt that the economy achieves if the rate of productivity
growth is constant at Gt = G ∀ t.
Answer:
The steady-state will be the place where kt+1 = kt = k̄. Substituting into
equation (27):
 
ε (1 − ε)β
k̄ = k̄ (35)
G(1 + β)
 
1−ε (1 − ε)β
k̄ = (36)
G(1 + β)
 1/(1−ε)
(1 − ε)β
k̄ = (37)
G(1 + β)

Now suppose that the economy had been growing at this constant rate G since the
beginning of time, but all of a sudden at the beginning of period t everybody learns
that henceforth and forever more, productivity will grow at a faster rate than before,
Ĝ > G.
¯
4. Define the new steady-state as k̂. Will this be larger or smaller than the original
steady state k̄? Explain your answer.
Answer:
 8
kt#1

45° Line$

!kt#1 !kt "

"

! ! kt
" k
k

Figure 1: Convergence of OLG Economy After Increase in G

Since (1 − ε) > 0, equation (37) implies that a larger value of G implies

a lower steady-state capital stock per efficiency unit. The reason is that
with faster productivity growth, the efficiency units of labor provided by
the young generation are larger relative to the size of the capital stock
saved by the previous generation, so the ratio of capital to efficiency units
of labor is smaller.

5. Next, use a diagram to show how the kt+1 (kt ) curve changes when the new growth
rate takes effect, and show the dynamic adjustment process for the capital stock
toward its new steady-state, assuming that the economy was at its original steady
state leading up to period t.
Answer:

Defining the original steady-state capital stock as k̄ and the new steady-
¯
state capital stock as k̂, the convergence process looks as indicated in fig-
ure 1.

6. Define an index of aggregate consumption per efficiency unit of labor in period t as

χt = c1,t + c2,t /G, and derive a formula for the sustainable level of χ associated with
a given level of k.
Answer:
 9

Kt+1 = Kt + Ktε Pt1−ε − C1,t − C2,t (38)

 
Kt+1 c2,t Pt−1
= kt + ktε − c1,t − (39)
Pt Pt
 
Kt+1 Pt+1
= kt + ktε − c1,t − c2,t /G (40)
Pt+1 Pt
kt+1 G = kt + ktε + χt (41)

The sustainable level of χ is the level χ̄ such that kt+1 = kt = k:

(1 + g)k = k + k ε − χ̄ (42)
χ̄ = k ε − gk. (43)

7. Derive the conditions under which a marginal increase in the productivity growth
rate g will result in an increase in the steady-state level of χ, and explain in words
why this result holds. (You can leave the term ∂ k̄/∂g unevaluated in your answer,
using only what we know about this term from above).
Answer:
There are two effects of an increase in g. First, for a given k, the sustainable
amount of χ(k) declines, because the faster productivity growth means
that to keep capital per efficiency unit constant the economy must save
more (each efficiency unit of labor must be supplied with its own capital;
faster growth of efficiency units therefore requires faster growth of capital).
Second, with a faster g the endogenous saving rate and steady-state capital-
per-capita k̄ will change. Whether steady-state consumption per capita
rises or falls depends on the balance between these two things.
Steady-state χ can be written as χ̄(k̄). We are interested in
     
dχ̄ ∂ χ̄ ∂ χ̄ ∂ k̄
= + (44)
dg ∂g ∂ k̄ ∂g
 
∂ χ̄ ∂ k̄
= −k̄ + (45)
∂ k̄ ∂g

But we know (from above) that ∂ k̄/∂g is negative; since −k̄ < 0, (45) can
possibly be positive only if
 
∂ χ̄
< 0 (46)
∂ k̄
ε−1
εk̄ −g < 0 (47)
r̄ < g, (48)
 10

where r̄ = f 0 (k̄) = εk̄ ε−1 . This is just the dynamic efficiency condition.
In words: We know from above that a higher value of g will decrease the
steady-state capital stock per efficiency unit. We know from our analysis in
class that the only circumstance in which a decrease in capital per efficiency
unit will directly result in an increase in consumption per efficiency unit
is if the dynamic efficiency condition fails to hold. So in order for there to
be any hope of an increase in g increasing χ̄, the economy must start out
as being dynamically inefficient.

However, dynamic inefficiency is not enough - the second term in (45) must
be larger than k̄ in order to offset the negative effect of faster g on χ̄. The
economy must be sufficiently dynamically inefficient that the increase in
the raw marginal product of capital that comes from lower k̄ more than
offsets the capital-dilution effect from the requirement to equip the new
efficiency units of labor with capital. In math, faster growth increases
consumption per efficiency unit when
 
dχ̄
> 0 (49)
dg
 
∂ k̄
−k̄ + (r̄ − g) > 0 (50)
∂g
 
∂ k̄
(r̄ − g) > k̄. (51)
∂g
REFERENCES 11

References
Diamond, Peter A. (1965): “National Debt in a Neoclassical Growth Model,” American
Economic Review, 55, 1126–1150.

Samuelson, Paul A. (1969): “Lifetime Portfolio Selection By Dynamic Stochastic Pro-

gramming,” Review of Economics and Statistics, 51, 239–46.