How to: Monetary Policy

1 Introduction
In general the setup is described somewhat like the following:

πt = mt + vt (1.1)
xt = θt + πt − πet − t (1.2)
∞
X
L= βt Lt (πt , xt ) (1.3)
t=1
1 2

Lt = (πt − π̄)2 + λ (xt − x̄) (1.4)
2
Equation (1.1) describes the demand side of the economy. For simplicity, we assume that inflation is
determined by the monetary policy and some stochastic demand shock vt . Equation (1.2) describes
a Phillips-curve relationship, describing that output/employment, xt , is determined by two types of
stochastic shocks as well as surprise inflation, πt − πet . Lastly, equation (1.3)-(1.4) describes how we can
measure social loss as fluctuations from a target level of inflation and employment.1

We can extend our assumptions/setup in a number of ways, some of which being:

• Assume shocks are persistent, for instance that t follows an AR(1) process:

t = ρt−1 + ut , 0<ρ<1 (1.5)

• Assume that inflation evolves according to the New Keynesian Phillips Curve (NKPC) instead of
(1.2):

xt = θt + κ1 πt − κ2 E[πt+1 ] − t (1.6)

• Assume that private agents has a certain trigger-strategy, such as tit-for-tat:

 
e E[πC C
t ], for πt−1 = πt−1
πt = (1.7)
E[πD C
t ], for πt−1 6= πt−1

• Include foreign currency zone and allow for a pegged currency, for instance the euro zone. As-
suming they have commitment policy this would imply:

λEU
πEU
t = EU , Corr(t , EU
t )=α (1.8)
1 + λEU t

• Assume different political parties have different preferences, λj .

1 For a microeconomic foundation for the loss function, you can check out appendix 8.6.2 in Walsh (2010), http://www2.um.
edu.uy/dtrupkin/walsh.pdf.

1
Macro III, E2016 Department of Economics, University of Copenhagen

• Assume imperfect control over inflation, such that monetary policy affects output through a DIS
curve:

xt = Et [xt+1 ] − σ−1 (it − Et [πt+1 ]) + ηt (1.9)

In the case of (1.5)-(1.6), see notes from assignment 4. In the case of the dynamic setup of (1.7) and
trigger-strategies, see PS15. For different political parties and cooperation see PS16, and for the pegged
currency case see PS17.2 In the following sections, we will go through some standard results/cookbook
approaches in the simple static cases and leave the more advanced dynamic problems for problem sets.
2 Forthe case of imperfect control over inflation, forward guidance and other extensions, one source is Henrik Jensen’s
Monetary Policy course site, http://www.econ.ku.dk/personal/Henrikj/monpol2016/slides.asp.

2
Macro III, E2016 Department of Economics, University of Copenhagen

2 A collection of cookbook approaches for finding equilibria

Beside the system of equations (1.1)-(1.4) we need to specify the timing in the model. In the following
’standard’ sections, we will generally assume the timing:

1. The monetary authority announces their policy rule (only commitment case)

2. The demand shock, vt , and the θt shock are realized.

3. Private agents form expectations of inflation, knowing both vt and θt .

4. t is realized

5. Monetary authority chooses their actual monetary policy, mt .

In the last section we will briefly show how a different assumption of timing can affect our solution
approach.

2.1 Commitment policy in equilibrium

The cookbook approach for commitment policy is the following:

• Postulate a monetary policy rule: It can be shown that optimal monetary policy with a quadratic
loss function follows a linear rule. Thus in our case, we announce a rule of the type:

mt = a1 + a2 vt + a3 θt + a4 t (2.1)

where the parameters a1 through a4 are yet to be determined.

• Impose credibility of the rule: If we are in an equlibrium, private agents trust that we will in
fact carry out monetary policy according to (2.1). With our assumed timing, this implies that
expected inflation follows:

πet = E[mt + vt |θt , vt ] = a1 + vt (a2 + 1) + a3 θt (2.2)

• Minimize expected social loss: Use the result in (2.1), (2.2) and the Phillips-curve to describe the
expected loss
=xt
 
=πt
1  z }| { z }| {
E[Lt ] = E (mt + vt −π̄)2 + λ(θt + πt − πet − t −x̄)2 
2
 
Using (2.1) Using (2.1) and (2.2)
1  z }| { z }| {
= E (a1 + vt (a2 + 1) + a3 θt + a4 t −π̄)2 + λ(θt + t (a4 − 1) −x̄)2  (2.3)

2

Next we need to square the terms in (2.3) and take expectations, before minimizing the expres-
sion. We will use that (almost) all cross-products in expectation are zero.3 We finally obtain the
expected loss

1 2
a1 + (a2 + 1)2 σ2v + a23 σ2θ + a24 σ2 + π̄2 − 2a1 π̄ + λ(σ2θ + (a4 − 1)2 σ2 + x̄2


E[Lt ] = (2.4)
2
3 See appendix for an elaboration on this part.

3
Macro III, E2016 Department of Economics, University of Copenhagen

The expression in (2.4) is straightforwardly differentiable. Thus we will have the 4 first order
conditions

min E[Lt ] (2.5)

a1 ,a2 ,a3 ,a4
∂E[Lt ]
= a1 − π̄ =0 (2.6)
∂a1
∂E[Lt ]
= (a2 + 1)σ2v =0 (2.7)
∂a2
∂E[Lt ]
= a3 σ2θ =0 (2.8)
∂a3
∂E[Lt ]
= a4 σ2 + λ(a4 − 1)σ2 =0 (2.9)
∂a4
This solves for the optimal monetary policy, inflation and output. Firstly, monetary policy is
given by plugging in optimal policy parameters from (2.6)-(2.9) into the policy rule in (2.1):
λ
mC
t = π̄ − vt + t (2.10)
1+λ
Next, plug in monetary policy in (2.10) into equation (1.1) to obtain equilibrium inflation:
λ
πC
t = π̄ + t (2.11)
1+λ
Lastly, xC C e
t is derived by using πt − πt in the phillips-curve equation (1.2):

πt −πe
z }| {
t

C λ
xt = θt + t −t
1+λ
1
=θ− t (2.12)
1+λ
The key take-aways of commitment policy (with this particular timing at least) is:
• Inflation is stabilized around target level π̄ (often assumed zero) perfectly offsetting demand
shocks. Letting our policy parameter a2 = −1 the demand shock vt will not affect inflation or
output.

• We cannot stabilize output around the target level x̄, as this would be anticipated and only create
an inflation bias.

• We cannot stabilize θt shocks, as they are anticipated by private agents. This is due to our
assumption of timing however.

• We can stabilize t shocks partially (not fully as with demand shocks, vt ). The optimal trade-off
between inflation and output is given by the coefficients, λ/(1 + λ) and −1/(1 + λ).

2.2 Discretionary Policy

When solving for discretionary policy, we’ll just start by simplifying the setup, and note that by picking
mt after the demand shock, vt , we have perfect control over inflation. Thus we will use πt as control
variable.
• Minimize actual social loss.4 Plug in all equations in our actual loss function and minimize wrt.
πt directly:
1
(πt − π̄)2 + λ(θt + πt − πet − t − x̄)2


min Lt = (2.13)
πt 2
4 Note that the crucial assumption for this first step, is that the monetary authority chooses their monetary policy after all

other shocks.

4
Macro III, E2016 Department of Economics, University of Copenhagen

From this we have the first order condition that

∂Lt λ
= πt − π̄ + λ(θt + πt − πet − t − x̄) = 0 ⇒ πt = π̄λ−1 + πet + x̄ + t − θt (2.14)


∂πt 1+λ

• Compute rational expectations of inflation: Use that private agents know that inflation will
follow the rule (2.14), such that
π̄ + λ(πet + x̄ − θt )
 
e λ −1 e
πt = E (π̄λ + πt + x̄ + t − θt )|θt = (2.15)
1+λ 1+λ
using that E[t |θt ] = 0. Solving for πet in (2.15) we have the rational expectation of inflation given
by

πet = π̄ + λ(x̄ − θt ) (2.16)

• Combine results in equilibrium: Substitute (2.16) into (2.14) to obtain equilibrium inflation
λ
πD
t = π̄ + λ(x̄ − θt ) + t (2.17)
1+λ
To arrive at the equilibrium for xD
t , use (2.16)-(2.17) in the Phillips-curve to obtain

xD D e
t = θt + πt − πt − t
1
= θt − t (2.18)
1+λ
The main result compared to commitment policy is the inflation bias: Discretionary policy entails an
inflation bias compared to commitment policy: The monetary authority has an incentive to stabilize
output around the level x̄ − θt . As private agents anticipate this, only inflation will increase, without
having an effect on output.

2.3 Pegging the currency

The cookbook approach for finding the equilibrium outcome with a pegged currency is:

• Assume pegged currency: In our setup, this translates to adopting a monetary policy. Per con-
struction the inflation level is thus given by foreign inflation rule:

πPt = πEU
t , (2.19)

which may be further specified as following some sort of policy rule as in equation (1.8).

• Impose credibility: Assuming that we are in an equilibrium, private agents expect the rule in
(2.19), thus we have

t |θt ]
πet = E πEU

(2.20)

• Combine results in equilibrium: Use (2.19) and (2.20) in the Phillips-curve to obtain the output
level

xPt = θt + πt − πet − t = θt − t + πEU − E πEUt |θt

 
t (2.21)

Main results of pegged currency equilibrium:

• The main upside of the pegged currency is that compared to discretionary policy, we eliminate
the inflation bias.

• The downside of the approach depends on how the adopted inflation policy correlates with
domestic shocks. If domestic and foreign shocks are uncorrelated, the pegged currency will not
be able to stabilize domestic shocks t .

5
Macro III, E2016 Department of Economics, University of Copenhagen

2.4 Optimal central banker solution

When choosing an optimal central banker, we assume that we have discretionary policy, but that we
can hire a central banker with a different λ than our own true preference for output-stabilization. This
means that we can choose the policy outcome

λB
 
B
!
B
πt π̄ + λ (x̄ − θt ) + 1 + λB t 
= (2.22)
B
xt 1 
θt −  t
1 + λB
where λB represents the relative preference for output-stabilization of our hired central banker. We
solve the problem as follows:

• When choosing the optimal CB, we assume that we have to hire him before knowing the shocks.
Thus we need to evaluate the expected loss of hiring λB :

1
E (πB 2
+ λE (xB 2
   

E[Lt ] = t − π̄) t − x̄)
2 " 2 # " 2 #!
1 B λB 1
= E λ (x̄ − θt ) + t + λE θt − t − x̄ (2.23)
2 1 + λB 1 + λB

Before differentiating we start squaring the terms and take expectations. As in the case of com-
mitment policy, we will use that all the cross-terms are zero in expectations.
 B
2  !
1 λ 1
E[Lt ] = (λB )2 (x̄2 + σ2θ ) + σ2 + λ σ2θ + x̄2 + σ2 (2.24)
2 1 + λB (1 + λB )2 

• Ideally we would minmize the expected loss in (2.24) by differentiating with respect to λB setting
it equal to zero and solving for λB . That is, solve the equation:

∂E[Lt ] λB λ
= λB (x̄2 + σ2θ ) + σ2 − σ2 = 0 (2.25)
∂λ B B
(1 + λ ) 3 (1 + λB )3 

However, as we cannot solve this, we will instead charactize the optimal solution by evaluating
derivatives in certain values of λB .

• Is it optimal to choose λB = 0?

∂E[Lt ] B
(λ = 0) = −λσ2 < 0, λ>0 (2.26)
∂λB
The derivative in (2.26) states that if we consider choosing a central banker with λB = 0, marginally
increasing λB would lower the expected loss. Thus the optimal central banker is characterized by
(λB )∗ > 0.

• Is it optimal to choose λB = λ?

∂E[Lt ] B
(λ = λ) = λ(x̄2 + σ2θ ) > 0, λ>0 (2.27)
∂λB
The derivative in (2.27) states that if we consider hiring a central banker with the same preferences
for output-stabilization as our own, λB = λ, we could lower the expected loss by decreasing λB .
Thus the optimal central banker must have preferences such that (λB )∗ < λ. Thus we know that
0 < (λB )∗ < λ.

The intuition behind choosing a central banker where λB < λ comes from balancing the two effects:

• Choosing λB < λ means that the loss from the inflation bias is lowered: λ(x̄2 + σ2θ ).

6
Macro III, E2016 Department of Economics, University of Copenhagen

• Choosing λB < λ means that the stabilization of the t shocks is not optimal anymore, as a
suboptimal large part of the shock affects output.

• The optimal monetary policy weighs the two effects in optimum, thus 0 < (λB )∗ < λ. Lastly,
note that choosing λB = λ is equivalent to choosing our own discretionary policy, while λB = 0 is
equivalent to choosing a pegged currency where foreign inflation is always zero. Thus the central
banker solution is preferred to discretionary policy and pegged currency.

7
Macro III, E2016 Department of Economics, University of Copenhagen

3 Solutions without the equilibrium assumption

When we deal with the dynamic setups we will often consider scenarios, where private agents intially
expect the monetary authority to perform discretionary policy and we have to create credibility of an
alternative policy such as commitment or pegged currency.

3.1 Obtaining credibility of pegged currency

Assume that private agents expect discretionary policy. In order to obtain credibility of our pegged
currency policy, we need to perform the pegged currency policy even though it is not expected. Thus
we find the outcome for creating credibility by:

• Assume pegged currency: The same first step as in the equilibrium approach. Thus we adopt
the foreign inflation level:

πt = πEU
t (3.1)

• Impose no credibility: Assuming that we are not in equilibrium implies that private agents
expect the outcome of second step in discretionary policy, equation (2.16):

πet = π̄ + λ(x̄ − θt ) (3.2)

• Combine results in Phillips-curve:

xt = θt + πEU
t − (π̄ + λ(x̄ − θt )) − t
= (1 + λ)θt + πEU
t − π̄ − t − λx̄ (3.3)

In order to create credibility we have to live with a relatively large shock to output/employment,
as we play πEU
t , but agents expect the inflation bias.

3.2 Obtaining credibility of commitment rule

The idea is basically the same. Monetary authority uses commitment rule, but private agents expect
discretionary policy:

• Use commitment policy rule from equilibrium approach:

λ
πt = π̄ + t (3.4)
1+λ

• Private agents expect discretionary policy:

πet = π̄ + λ(x̄ − θt ) (3.5)

• Combine results in Phillips-curve:

λ
xt = θt + π̄ + t − (π̄ + λ(x̄ − θt )) − t
1+λ
1
= (1 + λ)θt − t − λx̄ (3.6)
1+λ
In order to obtain credibility of the policy rule, we have to live with a relatively large shock to
employment. The intuition is that we will not have any inflation bias, but agents will expect it.

8
Macro III, E2016 Department of Economics, University of Copenhagen

4 Comparing policy solutions

In general the upsides and downsides of the different policy approaches are summarized by:

Monetary policy Upsides Downsides Other comments

1) No inflation bias. 1) No stabilization of the In a dynamic setup with

2) Optimal stabilization of x̄ − θt part. credibility, we can deviate in
Commitment
t shocks one period and improve by
stabilizing the x̄ − θt part.

1) Optimal stabilization of 1) No stabilization of the Nash-equilibrium solution

Discretion t shocks. x̄ − θt part. in static game.
2) Inflation bias: λ(x̄ − θt ). Never preferred over CB.

1) Eliminates part of 1) No stabilization of the CB chooses optimal

inflation bias x̄ − θt part. trade-off between inflation
2) Part of t shocks 2) Suboptimal stabilization bias and t shocks.
Central banker
stabilized of t shocks. Discretionary policy and
3) Part of inflation bias not pegged currency solution is
eliminated. nested in CB solution.

1) No inflation bias. 1) No stabilization of the Never preferred over CB.

Pegged currency x̄ − θt part.
(with πEU = 0) 2) No stabilization of t
shocks.

1) No inflation bias. 1) No stabilization of x̄ − θt If α = 1 solution can be

Pegged currency 2) When α > 0 partial shocks. equivalent to commitment
(corr(t , EU
t )= α > 0) stabilization of t shocks. 2) Suboptimal stabilization policy → optimal.
of t shocks.

5 Robustness of solution approaches, timing and specification

The solution approaches outlined in section 2 are relatively robust towards changes in specification and
timing. Some changes in model specification, such as including a DIS curve (equation (1.9)), changes
the fundamental approach.5

However, other changes such as those explained in equation (1.5)-(1.8) in the introduction does not
change the fundamental steps of our solution approach, but alters how we perform each step. For
instance, changing our model setup as in assignment 4 such that

t = ρt−1 + ut
xt = θt + πt − Et [πt+1 ] − t ,

we would still perform discretionary policy by the steps 1) Minimize actual social loss, 2) compute
rational expectations and 3) combine results in equilibrium. However, the way we have changed
our model implies that the second step, computing rational expectations, is a lot more involved than
outlined in section 2.2. The same general comment can be tied to the assumption of timing: The general
steps for solving the problem will often be the same, but often change how we carry out the specific
steps.
5 For the interested, this involves solving a system of expectational difference equations. This can be done by using a Bellman

equation/value-function approach. Once again, you could check out the course site for Monetary Policy for inspiration. (or ask
me)

9
Macro III, E2016 Department of Economics, University of Copenhagen

5.1 An example of an alternative assumption of timing

As an example on how our solution changes with the timing of the model, assume that we have the
following setup:

π t = mt (5.1)
xt = θt + πt − πet − t (5.2)
∞
X
L= βt Lt (πt , xt ) (5.3)
t=1
1 2
πt + λ(xt − x̄)2 ,


Lt = (5.4)
2
where we assume that both θt and t are IID errors with zero mean and constant standard deviations
σθ , σ . Furthermore, we assume that the timing of the model follows from:

1. Monetary policy rule is announced.

2. Private agents form expectations of inflation.

3. The supply shocks are realized, t , θt .

4. Monetary policy is carried out.

Commitment policy with alternative timing:

We can carry out our 3 steps of commitment policy:

• Postulate a policy rule (recall that a linear rule is sufficient):

πt = a1 + a2 θt + a3 t (5.5)

• Impose credibility of the rule: Assuming that we are in an equilibrium, private agents expect
monetary policy to follow the rule (5.5). Using the timing stated above, they form expectations
not knowing t and θt is, as they are realized after expectations are formed:

πet = E [a1 + a2 θt + a3 t ] = a1 , (5.6)

implying that surprise inflation will follow πt − πet = a2 θt + a3 t .

• Minimize expected social loss: Use the results in (5.5) and (5.6) and the Phillips-curve in our
expected loss function:
=πt =xt
 
1 z }| { z }| {
E[Lt ] = E (a1 + a2 θt + a3 t )2 + λ(θt + πt − πet − t −x̄)2 
2
1
E (a1 + a2 θt + a3 t )2 + λ(θt (1 + a2 ) + t (a3 − 1) − x̄)2


= (5.7)
2
In order for us to minimize this expression, we need to start by evaluate expectations first. Once
again using that all cross-terms in expectations are zero, we only have squared terms left after
squaring and taking expectations:

1 2
a + a22 σ2θ + a23 σ2 + λ(σ2θ (1 + a2 )2 + σ2 (a3 − 1)2 + x̄2 )


E[Lt ] = (5.8)
2 1

10
Macro III, E2016 Department of Economics, University of Copenhagen

Mimizing this expression we differentiate with respect to our 3 policy parameters, a1 , a2 , a3 and
solves:
∂E[Lt ]
=a1 = 0 (5.9)
∂a1
∂E[Lt ]
=a2 σ2θ + λ(1 + a2 )σ2θ = 0 (5.10)
∂a2
∂E[Lt ]
=a3 σ2 + λ(a3 − 1)σ2θ = 0 (5.11)
∂a3
Solving this we get the inflation policy with commitment
λ λ
πC
t = |{z}
0 − θt + t . (5.12)
1 + λ
| {z } | {z λ}
1 +
=a1
=−a2 =a3

Lastly, using the Phillips-curve we get the equilibrium employment:

1
xC
t = (θt − t ) . (5.13)
1+λ

In general, the solution approach is not changed with the new timing. However, performing the steps
of imposing credibility of rule, and minimizing expected loss involves some new information as we
no longer condition inflation expectations on θt .

Discretionary policy with alternative timing:

We follow the same general solution approach as described in section 2.2:

• Minimize actual social loss: As monetary policy is carried out after observing all shocks and
we do not need to announce a policy rule, as we do not have commitment, the objective of the
monetary authority is to solve the problem:
1 2
πt + λ(θt + πt − πet − t − x̄)2 .


min Lt = (5.14)
πt 2
This yields the first order condition and solution that
∂Lt λ
= πt + λ(θt + πt − πet − t − x̄) = 0 ⇒ πt = (x̄ − θt + t + πet ) (5.15)
∂πt 1+λ

• Compute rational expectations of inflation: Assuming that we are in a discretionary equilibrium,

private agents form expectations based on equation (5.15). With our assumed timing, private
agents do not know θt or t shocks, implying that they will expect inflation according to
 
e λ e λ
πt = E (x̄ − θt + t + πt ) = (x̄ + πet ),
1+λ 1+λ
as they expect both θt and t to be zero. Solving for πet this implies

πet = λx̄ (5.16)

• Combine results in equilibrium: Lastly, we combine the result in (5.15) − (5.16) as well as the
Phillips-curve to obtain equilibrium outcome:
λ
πD
t =λx̄ + (t − θt ) (5.17)
1+λ
1
xD
t =(θt − t ) (5.18)
1+λ

Thus once again our solution approach is the same, but the derivations, in particular how to compute
rational expectations of inflation change slightly.

11
Macro III, E2016 Department of Economics, University of Copenhagen

Appendices
A Expectations, squared terms and derivatives
When we solve for the optimal policy under commitment or central banker, we have to take into
account that we are using expected losses. In section 2.1 we encountered the expected loss function:
 K1 K2

z }| { z }| {
1 
E[Lt ] = E (a1 + vt (a2 + 1) + a3 θt + a4 t − π̄)2 +λ (θt + t (a4 − 1) − x̄)2  (A.1)

2

In general, it turns out that it matters whether we start by taking derivatives or expectations.6 Thus
we would prefer to write out (A.1) before differentiating. In this case, we have assumed that vt , θt , t
are all mean zero stochastic shocks with zero covariance. If we were to write out the K1 part we would
obtain:

K1 =a21 + (a2 + 1)2 v2t + a23 θ2t + a24 2t + (−π̄)2 (A.2)
+2a1 ((1 + a2 )vt + a3 θt + a4 t − π̄)
+2(1 + a2 )vt (a3 θt + a4 t − π̄)
+2a3 θt (a4 t − π̄)
+2a4 t (−π̄),

where the first line contains all the squared terms, and the last 4 contains all the cross-products. Taking
expectations now, we use that we assume that all shocks are mean zero and have zero covariance:

E[vt ] = E[θt ] = E[t ] = 0 (A.3)

E[vt θt ] = E[vt t ] = E[θt t ] = 0. (A.4)

Note that assumptions (A.3) − (A.4) imply that almost all the cross-products in (A.2) will be zero in
expectation:

E[K1 ] =a21 + (a2 + 1)2 E v2t + a23 E θ2t + a24 E 2t + π̄2
     
(A.5)
−2a1 π̄

Lastly, note that with the assumption that the 3 shocks are mean zero variables implies that
2
σ2x ≡ E x2 − = E x2 .
   
E [x] (A.6)
| {z }
=0, when mean zero

Using this we conclude that

E[K1 ] = a21 + (a2 + 1)2 σ2v + a23 σ2θ + a24 σ + π̄2 − 2a1 π̄. (A.7)

You can verify by arguments (A.3) − (A.4) and (A.6) that the last part of (A.1) is given by

E[K2 ] = σ2θ + σ2 (a4 − 1)2 + x̄2 (A.8)

6 In our simple monetary policy setup, we generally do not need to take expectations first, as the Lebesgue Dominated

Convergence Theorem holds. However, taking expectations first does not make our problem harder to solve, but it makes our
approach more robust towards more ’funky’ setups. Thus we prefer this approach.

12