Econ 202a Spring 2000

Marc Muendler (TA)

Answer Keys for Problem Set 3

1 Delegated control over monetary policy
Output is given by the Lucas supply function

y = ȳ + b(π − π e), (1)

social welfare by
a
S = γ̃y − π 2, (2)
2
and the central banker’s objective function by
a
S CB = cγ̃y − π 2, (3)
2
where c ∈ R. The coefficient γ̃ is a random variable with mean E[γ̃] = γ̄ and
variance Var(γ̃) = σ 2 .
We want to solve the model so that all decisions are ‘time-consistent’,
as macroeconomists like to say. If these macroeconomists also were game
theorists, they would probably prefer to call the equilibrium of the game
‘subgame perfect’ and not ‘time-consistent.’ The concepts are the same,
but let’s keep the notion of ‘time consistency’ for the purpose of this macro
exercise. In order to find a time-consistent equilibrium, we need to start with
the choice of the last agent in the chain of decisions. In our model, there are
three stages of decisions. First, the ‘public’ or the ‘private sector’ chooses its
inflation expectations given its expectations of γ̃. Then, on the second stage,
nature draws a general preference parameter γ̃ and reveals it to the central
banker. Thus, the central banker has superior information about the general
public’s preferences. Finally, on the third stage, the central banker chooses
the inflation rate.

1
1.1 [1a] Central banker’s choice
Let’s start at the end of the chain. (Game theorists would say: Let’s apply
backward induction.) The central banker maximizes
 a 
maxS CB = max cγ ȳ + cγb(π − π e ) − π 2 , (4)
π π 2
where we have used (1) to express output in terms of the inflation rate π.
Note that the central banker knows the realization of γ at the time of her
decision. The first-order condition to this problem is

cγb − aπ ∗ = 0.

Thus, the central banker will optimally choose1

γb
π∗ = c . (5)
a
The central banker will choose the higher an inflation rate the more respon-
sive output is (b) and the less inflation-averse the general public is (a).

1.2 [1b] Expectation of inflation π e

The general public has to make up its mind about expected inflation before
nature reveals what realization γ̃ takes this time. Their expectations about
the impact of γ̃ are rational. Therefore,
γ̄b
π e = E [π ∗] = c . (6)
a
So, again, output will only respond to unexpected and unsystematic devia-
tions of the central banker from her ‘rule’ π ∗ = cγb/a. This can be seen from
(1):

cb2
y = ȳ + (γ̃ − γ̄) . (7)
a
h 2
i
So, E [y] = E ȳ + cba (γ − γ̄) = ȳ. Note also that inflation is positive as long
as c is, but output does not systematically rise above its reference level ȳ.
1
We need not worry about second-order conditions. This problem is concave in π
because π 2 is a convex function of π.

2
The problem here is not even one of possible time-inconsistency. Irrespective
of what π e the general public has chosen on the first stage, the central banker
will always set π ∗ = cγb/a on the third stage of the game (see (5)). But does
the general public want a central banker to choose a positive inflation rate,
π ∗ > 0? After all, output will not systematically change for π ∗ > 0, but
inflation is considered bad since a > 0.

1.3 [1c] Expected Social Welfare

To see what kind of central banker the general public would love to employ,
let’s maximize expected social welfare. Using (7) in (2) and taking expecta-
tions of both sides, we find expected social welfare
"  2 #
cb2 a cb
E [S] = E γ̃ ȳ + γ̃ (γ̃ − γ̄) − γ̃
a 2 a
2 n   2 o
cb c 2
= γ̄ ȳ + 1− E γ̃ − γ̄
a 2
cb2 n c  2 c 2o
= γ̄ ȳ + 1− σ − γ̄
a 2 2
cb2 2 c2 b2 2

= γ̄ ȳ + σ − σ + γ̄ 2 . (8)
a 2a

1.4 [1d] The best central banker

What is the best central banker? Or, what value of c maximizes expected
social welfare? To see that, maximize (8) with respect to c. The first order
condition of this problem is2

b 2 2 c∗ b 2 2

σ − σ + γ̄ 2 = 0.
a a
Hence,

σ2
c∗ = . (9)
σ 2 + γ̄ 2
2
Again, we need not concern ourselves with second-order conditions. This problem is
concave in c because c2 is a convex function of c.

3
 So, the general public should specify a contract with the central banker.
That contract should state: We will pay you exactly c∗ γ̃y − a2 π 2 each period.
Now you choose π. The optimal contract induces the central banker to be
more conservative than the general public. Since c∗ ∈ (0, 1], the central
banker will always put less weight on output stabilization and more weight
on low inflation than the general public. The optimal contract makes the
central banker more conservative than the public sector is.
We can say even more about the optimal contract. Whenever the variance
of preferences is extremely high, that is whenever σ 2 is large, c∗ is close to
one. Thus, the public wants a very responsive central banker whenever it
knows that there is a lot of change in fundamentals. However, the more
responsive the central banker gets, the higher equilibrium inflation will be,
too. Hence, there is a trade-off. How strong the trade-off becomes depends
on the expected value of γ. Why? The higher the expected value of γ, the
higher expected equilibrium inflation will be for any given c: π e = γ̄cb/a.
Thus, the expected welfare loss from high inflation will weigh more if γ̄ is
high. Therefore, c∗ is falling in γ̄. In fact, it is close to zero when γ̄ far
exceeds σ. The optimal contract for the central banker incorporates this
trade-off. A responsive central banker is nice, but a too responsive central
banker causes too high an equilibrium inflation rate.

2 A simple model of overshooting

Our simple Dornbusch model of overshooting in discrete time consists of three
equations:

m̄ − pt = φytd − η(et+1 − et ), (10)

1
ytd = δ(et − pt ) δ ∈ (0, ), (11)
φ

pt+1 − pt = π(ytd − ȳ). (12)

All variables are in logs. pt is the aggregate price level, ytd is aggregate
demand, et is the nominal exchange rate (denoted in dollars per foreign cur-
rency), m̄ is fixed money supply, and ȳ the full-employment level of output.

4
Note that this model is Keynesian in style; we accept that output can sys-
tematically deviate from the full-employment level for several periods until
price adjusts in a manner determined by (12). Our Dornbusch model is in
fact a dynamic system in two variables, et and pt , and two equations. To see
this most clearly, plug (11) into (10) and (12) to obtain
 
φδ φδ 1
et+1 − et = et + 1 − pt − m̄
η η η

and

pt+1 − pt = πδ(et − pt ) − π ȳ.

But there are more insightful ways than this method of brute force to find
out about the dynamic properties of our version of the Dornbusch model.

2.1 [2a] The steady state

A very good point of departure usually is the steady state. We can derive it
by setting all endogenous variables equal to hypothesized steady state values.
So, we simply try whether such a steady state exists or not. In steady state,
et+1 = et = ē and pt+1 = pt = p̄. Therefore, et+1 − et = 0 and pt+1 − pt = 0.
Using pt+1 − pt = 0 in (12), we find that ytd = ȳ in a steady state. Using
ytd = ȳ and et+1 − et = 0 in (10), we find that

p̄ = m̄ − φȳ. (13)

Finally, using that very fact in (11), we obtain

1 1 − φδ
ē = ȳ + p̄ = m̄ + ȳ. (14)
δ δ
Thus, a steady state exists. We have just found it by trial. Whether this
steady state is stable or not still has to be investigated. It will turn out that
it is unstable in one dimension, but stable in another. More on this below.
For now, let’s only make one more observation. The requirement that δ < φ1
is obviously crucial to let the steady state output have a positive relationship
with the steady state nominal exchange rate. For δ < φ1 it must be the case
that φδ < 1. The key term 1 − φδ will show up several more times.

5
 pt A
P

Price level E

p
E

0 e et
Nominal exchange rate

Figure 1: Phase Diagram for initial steady state

2.2 [2b] General dynamics of the system

In order to find out more about the dynamic behavior of the system, we
want to know under what conditions the nominal exchange rate is increasing
(depreciating), and under what conditions the price level is increasing. Start
with the nominal exchange rate. Using (10) and (11), the change in the
exchange rate over one period can be related to previous levels and exogenous
variables:
m̄ − pt φ d m̄ − pt φ
et+1 − et = − + yt = − + δ(et − pt )
η η η η
1 − φδ φδ m̄
= pt + et −
η η η
1 − φδ φδ
= (pt − p̄) + (et − ē) , (15)
η η

6
where the last step makes use of the steady state levels of et and pt . Thus,
φδ m̄
et+1 − et ≥ 0 ⇔ pt ≥ − et +
1 − φδ 1 − φδ
φδ
⇔ pt − p̄ ≥ − (et − ē) . (16)
1 − φδ
Using (12) along with (11) and also applying the steady state definitions,
we can derive the change in price levels, too.

pt+1 − pt = π(ytd − ȳ) = πδ (et − ē) − πδ (pt − p̄) . (17)

Hence,

pt+1 − pt ≥ 0 ⇔ pt − p̄ ≤ et − ē (18)
ȳ
⇔ pt ≤ et − .
δ
The second equivalence follows since pt ≤ et − (ē − p̄) = et − ȳδ . Last, we
have a no-arbitrage relationship

pt+1 − p̄ = −θ̂ (et+1 − ē) (19)

to which the economy must always adhere.

We can put this information together in a phase diagram. First, we
want to know the curves where the dynamic forces on either the exchange
rate changes or the price level are absent. That is, we want to draw the
lines where et+1 − et = 0 and where pt+1 − pt = 0. From (16) and (18)
we know that these curves are straight lines with pt − p̄ = − 1−φδ φδ
(et − ē)
and pt − p̄ = et − ē. Hence, both of them pass through the steady state.
And they better do because the steady state is certainly a place where no
dynamic forces are present at all. Figure 1 depicts these two lines as EE
for pt − p̄ = − 1−φδ
φδ
(et − ē) (from (16)) and P P for pt − p̄ = et − ē (from
(18)). Second, how does the system behave off these lines? For that, we can
use (16) and (18) again. From (16) we can infer that the nominal exchange
rate increases (depreciates) whenever we are above the EE line in our phase
diagram; it decreases (appreciates) otherwise. We indicate this fact with
arrows pointing East above the EE line, and pointing West below the EE
line. Similarly, from (18) we can infer that the price level increases whenever
we are below the P P line and decreases otherwise. This gives rise to an

7
interesting pattern. Before we turn to that, let’s draw the so-called ‘no-
arbitrage’ line (19) into the diagram, too. Call it AA. Here we were careful
to make the AA line steeper than the EE. The reason for that will become
clear shortly.

2.3 [2c] Stability of the steady state

Is the steady state stable? Well, as both lawyers and economists usually like to
say: It depends. The arrows—the dynamic forces of the system—possibly take
us back to the steady state whenever we are either to the Northwest or to the
Southeast of the steady state. In this dimension the steady state can be stable
indeed. However, the forces take us away from the steady state whenever our
economy initially finds itself either to the Northeast or to the Southwest of
the steady state. In that dimension the system is unstable. Hence, there
must be a saddle path that prevents the economy from exploding. This
saddle path must run from Northwest to Southeast, and it must go through
the steady state.
In fact, we just drew such a line. We know that the economy must
always obey the ‘no-arbitrage’ relationship. In order to not be explosive, the
economy must also always be on the saddle path which has to go through the
steady state, too. Since both of these relationships always have to be satisfied,
they can’t correspond to two separate lines. If they corresponded to two
separate lines, they would only be simultaneously satisfied in the steady state.
But both the ‘no-arbitrage’ condition and the ‘non-explosiveness’ condition
have to be satisfied in and out of steady state. Thus, the two corresponding
lines must coincide. The no-arbitrage condition is nothing but the saddle
path. If we had happened to draw the no-arbitrage line flatter than the EE
line, after this reasoning we would have to go back to the phase diagram in
figure 1 and draw the AA line steeper than EE. The AA line simultaneously
represents the no-arbitrage condition and the saddle path.
In the Dornbusch model, we have to require that our economy be non-
explosive. Somehow, this assumption comes out of the blue in the Dornbusch
model. And it will have to come out of the blue in any Keynesian style
model because unfortunately we don’t have any optimizing agents here who
could tell us what they would optimally choose to do. Maybe they really
love explosive economies. As it turns out in models with micro-foundations,
however, agents don’t like explosive paths in general. So the requirement of
a non-explosive economy is reasonable, although not rigorously justifiable in

8
 pt
A'
P

Price level E

p 1
E' 2
3 E
p'

E'

A'
0 e0 e' e et
Nominal exchange rate

Figure 2: Phase Diagram for change in money supply

Keynesian models.

2.4 [2d] A monetary contraction at home

We have setup the model, and we have derived its main properties. Now let’s
play with the model. Suppose the steady monetary supply is unexpectedly
reduced from m̄ to m̄0 < m̄ at time t = 0. What happens to our phase
diagram? As (18) shows, the P P curve is unaffected because pt = et −
(ē − p̄) = et − ȳδ . The EE curve is shifted downward, however. Following
(16), the underlying relationship has changed from pt = − 1−φδ φδ m̄
et + 1−φδ
m̄0
to pt = − 1−φδ
φδ
et + 1−φδ . The no-arbitrage condition must hold again, and
therefore the AA curve must now pass through the new steady state.
By assumption, we are in a world of sticky prices. Whereas all variables
are assumed to adjust immediately, pt remains at its given level for one period.
What happens to et right after a reduction in the monetary base from m̄ to
m̄0? Since prices respond too little (not at all), it is not so surprising that the

9
other leading variable in our system, the nominal exchange rate must make
up for the slack in prices. It will overshoot. In which way? Note that at the
time of the unexpected change in money supply, the economy is still in the
old steady state at point 1 in figure 2. Now everything is allowed to change
except for prices. So the economy is restricted to jump out of the old steady
state along a horizontal line through point 1. Where will it jump? The
only possible point to jump to is 2. From anywhere else, the economy would
have to explode subsequently. Some sample paths off the saddle path are
drawn in figure 2. They are all explosive. Even remaining at the old steady
state would make the economy explode. (Note that the dynamic forces, the
arrows, are now all relative to the new steady state 3.) So, at date t = 0,
when the money supply is changed, the economy jumps to 2. From then
on, it simply obeys the dynamics of the new system. These dynamics take
the economy gradually from point 2 to point 3 in the phase diagram. The
economy converges to the new steady state at 3. The nominal exchange rate
et and the price level pt move jointly along the saddle path from t ≥ 1 on.
The nominal exchange rate overshoots. It initially appreciates to a level e0
beyond the future steady state level ē0 . Subsequently, the nominal exchange
rate depreciates to the new steady-state level ē0 despite the initial monetary
tightening at home. Yet, the over-all effect of the monetary contraction will
be an appreciation of the exchange rate from ē to ē0, as we should reasonably
expect after a reduction of the money base at home.

3 Dynamics around the steady state

Let’s make the graphical analysis of our Dornbusch model from section 2 rig-
orous. We will derive the saddle path formally. The first three sub-questions
prepare for the derivation.

3.1 [3a] Output deviations from steady state

Using (11) and the steady states of ē and p̄ from section 2, the output devi-
ation from steady state can be expressed as

ytd − ȳ = δ(et − pt ) − ȳ
= δ(et − ē) + δē − δ(pt − p̄) − δ p̄ − ȳ
= δ(et − ē) − δ(pt − p̄). (20)

10
3.2 [3b] Exchange rate deviations from steady state
Similarly, using (10) along with the results in section 2.1 and section 3.1,
exchange rate deviations from steady state are
1

et+1 − ē = φytd + pt − m̄ + et − ē
η
1
= [φδ(et − ē) − φδ(pt − p̄) + φȳ + pt − m̄] + et − ē
η
η + φδ 1 − φδ
= (et − ē) + (pt − p̄). (21)
η η

3.3 [3c] Price deviations from steady state

Finally, using (11) and (12) along with the results in section 2.1, the price
deviation from steady state becomes a weighted sum of et − ē and pt − p̄:

pt+1 − p̄ = pt − p̄ + πδ(et − pt ) − π ȳ
= πδ(et − ē) + (1 − πδ) (pt − p̄). (22)

3.4 [3d] The eigenvalues of the dynamic system

With results (20), (21), and (22) at hand, we can rewrite the entire Dornbusch
model in terms of deviations from steady state. This is particularly useful
because we obtain a system of two first-order difference equations for which
we know the general solution. Writing (21) and (22) in matrix form, we find
   η+φδ 1−φδ
 
et+1 − ē η η e t − ē
= . (23)
pt+1 − p̄ πδ 1 − πδ pt − p̄

We know the eigenvalues and eigenvectors of such a system. We even have

formulas for them.3 The eigenvalues of a linear difference equation system in
two variables, xt+1 − x̄ = A (xt − x̄) with x ∈ R2 , are
q
tr(A) 1 2
λ1,2 = ± [tr(A)] − 4 det(A).
2 2
3
For a derivation, see section 3.2 of the handout “Linear Difference Equations and
Autoregressive Processes.” (http://socrates.berkeley.edu/˜muendler/teach/diffeqn.pdf)

11
The eigenvectors are
 λi−a22

ei = a21 i = 1, 2,
1
and the general solution is
 λ1 −a22 λ2 −a22
 
a21 a21 c1 (λ1 )t
xt = ,
1 1 c2 (λ2 )t
where the coefficients c1 and c2 have to be determined through boundary
conditions.
For simplicity, we assume that π = η1 and φ = 3. Then (23) becomes
    
et+1 − ē 1 η + 3δ 1 − 3δ et − ē
= . (24)
pt+1 − p̄ η δ η−δ pt − p̄

And, as stated in the hint, tr(A) = 2(δ+η)η

and det(A) = 1− (1−2η)δ
η2
. Plugging
this into the formulas for the eigenvalues, and simplifying, yields
q
tr(A) 1 2
λ1,2 = ± [tr(A)] − 4 det(A)
2 2
r
δ+η 4 (δ 2 + 2δη + η 2) − 4 + 4δ/η 2 − 8δ/η
= ±
η 4
δ 1 p
= 1+ ± δ (1 + δ)
η η
 √
 1 − δ(1+δ)−δ
= √η . (25)
 1 + δ+ δ(1+δ)
η

Clearly, the lower eigenvalue in (25) exceeds one. Thus, the system will be
unstable. The according eigenvectors are
 λ1 −a22 λ2 −a22 
(e1, e2) = a21 a21
1 1
 √ √ 
δ(1+δ)−δ δ+ δ(1+δ)
1− η
−1+ ηδ 1+ η
−1+ ηδ
=  δ
η
δ
η

1 1
q q !
1+δ 1+δ
2− 2+
= δ δ .
1 1

12
With this, we also know the general solution to the Dornbusch model:
  √ t 
  q q ! δ(1+δ)−δ
et − ē 2 − 1+δ 2 + 1+δ  c1 1 − η 
=   √ t 
 . (26)
δ δ
pt − p̄ 
1 1 c2 1 +
δ+ δ(1+δ)
η

3.5 [3e] Stability and the saddle path

We have already seen in (25) that the system must be unstable because at
least one eigenvalue exceeds one in absolute value. Are they both unstable?
We have seen in the graphical derivation in section 2 that the system is stable
in one dimension. Therefore let’s suppose that the other eigenvalue is less
than one in absolute value and try where we get. If things are consistent,
we can infer that one of the two eigenvalues must be below one in absolute
value. The price and exchange rate will only converge to the steady state if
the coefficient of the explosive eigenvalue, c2 here, is set to zero. Thus, the
general solution (26) further simplifies to

  q q !  √ t 
1+δ 1+δ δ(1+δ)−δ
et − ē 2− 2+  c1 1 −  , (27)
= δ δ η
pt − p̄ 1 1 0

so that
r ! p !t
1+δ δ (1 + δ) − δ
et − ē = c1 2 − 1− (28)
δ η

and
p !t
δ (1 + δ) − δ
pt − p̄ = c1 1 − . (29)
η

How can we pin down c1 ? Well, we generally have some boundary conditions.
For example, we know where the undisturbed Dornbsuch model started out
in the very beginning of section 2: at ē and p̄ (figure 1, p. 6). Similarly,
we know that, after the monetary contraction from m̄ to m̄0 at home, the
system had to start out at e0 and p̄ on the new saddle path (figure 2, p. 9).

13
These kinds of boundary conditions are enough to completely and rigorously
describe the dynamic behavior of the Dornbusch model.
However, for the mere purpose of deriving the saddle path we need not
even worry about the proper boundary condition. We can simply plug (29)
into (28) and we get, after all:
r !
1+δ
et − ē = 2 − (pt − p̄) . (30)
δ

What does this line look like in the phase diagram? It certainly passes
through the steady state. What slope does it have? Recall that δ < φ1 = 13 .
q q √
1 1+δ 1
Hence δ > 3 and δ
= δ
+ 1 > 4 = 2. The slope must be negative.
Beautiful. So, we just found the analytic expression
 qfor the
 saddle path AA
in figure 1 (p. 6). If we write (30) as et − ē0 = 2 − 1+δ δ
(pt − p̄0 ) we have
the analytic expression for the saddle path in figure 2 (p. 9) as well.
There are mainly two ways to think about this relationship. One inter-
pretation is, as we saw in lecture, that the economy must obey a no-arbitrage
relationship pt+1 − p̄ = −θ̂ (et+1 − ē). If prices and the exchange rate would
not move in phase at any point in time, an investor could construct a costless
currency portfolio at that time and make a profit one period later. An other
interpretation is that the system must not become explosive. The only sched-
ule along which the economy does not explode is the saddle path, depicted
as AA in figures 1 q (p. 6) and 2 (p. 9). With this interpretation, we have in
fact found θ̂: θ̂ = 1+δδ
− 2. Whenever the economy encountered itself off
the saddle path, the exchange rate would have to adjusts immediately and
to take the whole economy back to a point on the saddle path. Since prices
adjust sloppily, the exchange rate will have to overshoot at times.

14