Graduate Macro Theory II:

The Basic New Keynesian Model

Eric Sims
University of Notre Dame

Spring 2024

1 Introduction
This set of notes lays and out and analyzes the canonical New Keynesian (NK) model. The NK
model takes a real business cycle model as its backbone and adds to that sticky prices, a form
of nominal rigidity that allows purely nominal shocks to have real effects, and which alters the
response of the economy to real shocks in a way that gives rise to a non-trivial role for active
stabilization policy.
To get price stickiness in the model, we have to have firms as price-setters, which means we
need to move away from the perfectly competitive benchmark. To do so we introduce monopo-
listic competition. We split production into two sectors, where the final goods sector is perfectly
competitive and aggregates intermediates into a final good for consumption. This generates a
downward-sloping demand for intermediate goods. There are a continuum of intermediate goods
producers who can set their own prices, but take all other prices as given. All the action in the
model is at the level of the intermediate producers. We assume that they are not freely able to
adjust their prices each period. In particular, the Calvo (1983) assumption posits that each period
firms face a fixed probability of being allowed to change their price. This seems a little ridiculous
in terms of its realism, but this assumption facilitates aggregation, and this is why it is so popular.
With any price rigidity, any firm’s price becomes a state variable. With a continuum of interme-
diate goods firms, we’d have a continuum of state variables. The Calvo (1983) assumption allows
us to aggregate out this heterogeneity. Even though it seems somewhat bizarre on its surface, it
has some normative implications that seem pretty reasonable (in particular, price stability ends up
being an important policy goal).
The basic New Keynesian model that I’ll lay out below (and which is laid out in Woodford (2003)
and Gali (2007) textbook treatments) has no investment or capital. This simplifies the analysis
quite a bit and permits us to get better intuition. It is not a completely innocuous omission, and
we’ll later look at how the inclusion of capital in the model affects things.

1
2 Households
There is a representative household that consumes, supplies labor, accumulates bonds, owns, and
accumulates money. Its problem is:

∞ !
Ct1−σ N 1+χ

X Mt
max E0 βt −θ t + ψ ln
Ct ,Nt ,Bt+1 ,Mt 1−σ 1+χ Pt
t=0

Here I have gone ahead and assumed that utility from real balances is logarithmic. As long as
real balances are additively separable from consumption and labor, money in the utility function
doesn’t do much interesting here.1 The nominal flow budget constraint is:

Pt Ct + Bt+1 + Mt − Mt−1 ≤ Wt Nt + Pt Dt − Pt Tt + (1 + it−1 )Bt

Here money is the numeraire, and Pt is the price of goods in terms of money. Bt is the stock of
nominal bonds a household enters the period with, and they pay out (known as of t − 1) nominal
interest rate it−1 . The household also enters the period with a stock of money, Mt−1 . Note that
I’m not being super consistent with timing notation here: Mt−1 and Bt are both known at t − 1.
The reason I write it this way is because the aggregate supply of money in period t, Mt , is not
going to be predetermined but rather set by a central bank. Wt is the nominal wage (denominated
in units of money, not goods). Dt denotes (nominal) profits remitted by firms, and Tt is a lump
sum tax paid to a government (the government will have the role of setting the money supply and
remitting any seignorage revenue back to the household lump sum).
A Lagrangian for the household is:

∞
Ct1−σ N 1+χ
   
X Mt
L = E0 βt −θ t + ψ ln + λt (Wt Nt + Pt Dt − Pt Tt + (1 + it−1 )Bt − Pt Ct − Bt+1 − Mt + Mt−1 )
t=0
1−σ 1+χ Pt

The FOC are:

∂L
= 0 ⇔ Ct−σ = λt Pt
∂Ct
∂L
= 0 ⇔ θNtχ = λt Wt
∂Nt
∂L
= 0 ⇔ λt = βEt λt+1 (1 + it )
∂Bt+1
∂L 1
=0⇔θ = λt − βEt λt+1
∂Mt Mt

We can eliminate the multiplier and re-write these conditions as:

1
If we were to assume that central bank policy focuses on an interest rate rather than a monetary aggregate, as we
will do below, then we could ignore money altogether so long as utility from money is separable. This is sometimes
referred to as a “cashless” economy.

2
 θNtχ = Ct−σ wt (1)
"  #
Ct+1 −σ

Pt
1 = Et β (1 + it ) (2)
Ct Pt+1
 −1
Mt it
ψ = C −σ (3)
Pt 1 + it t
(1) is a standard labor supply condition, where I have defined wt ≡ Wt /Pt as the real wage. (2)
 −σ
is the standard Euler equation for bonds, with β CCt+1t
= Λt,t+1 the stochastic discount factor.
(3) implicitly defines a demand function for real balances that is increasing in consumption and
decreasing in the nominal interest rate.

3 Production
For the production side of things we split into two. There is a representative competitive final goods
firm that aggregates intermediate inputs according to a CES technology. To the extent to which the
intermediates are imperfect substitutes in the CES aggregator, this generates a downward-sloping
demand for each intermediate, which gives these intermediate producers pricing power. There
are a continuum (large number) of intermediates, so these producers behave as monopolistically
competitive (they treat all prices but their own as given). These firms produce output using labor
and are subject to an aggregate productivity shock. They are not freely able to adjust prices each
period, in a way that we will discuss in more depth below.

3.1 Final Good Producer

The final output good is a CES aggregate of a continuum of intermediates:
ϵ
Z 1 ϵ−1
 ϵ−1
Yt = Yt (j) ϵ dj (4)
0

Here ϵ > 1. This parameter measures how substitutable different varieties are. The profit
maximization problem of the final good firm is:
ϵ
Z 1 ϵ−1
 ϵ−1 Z 1
max Pt Yt (j) ϵ dj − Pt (j)Yt (j)dj
Yt (j) 0 0

The FOC for a typical variety of intermediate j is:

ϵ
Z 1  ϵ−1 −1
ϵ ϵ−1 ϵ−1 ϵ−1
Pt Yt (j) ϵ dj Yt (j) ϵ −1 = Pt (j)
ϵ−1 0 ϵ
This can be written:

3
 1
Z 1  ϵ−1
ϵ−1 1 Pt (j)
Yt (j) ϵ dj Yt (j)− ϵ =
0 Pt
Or:
ϵ
− ϵ−1 −ϵ
Z 1 
ϵ−1 Pt (j)
Yt (j) ϵ dj Yt (j) =
0 Pt
Making note of the definition of the aggregate final good, we have::
 −ϵ
Pt (j)
Yt (j) = Yt (5)
Pt
(5) says that the demand for each intermediate is (i) proportional to total output, Yt , and, (ii),
decreasing in the relative price of the intermediate. ϵ has the interpretation of the price elasticity
of demand.
We can derive an aggregate price index by defining nominal output as the sum of prices times
quantities:
Z 1
Pt Yt = Pt (j)Yt (j)dj
0

Plugging in the demand for each variety, we have:

Z 1
Pt Yt = Pt (j)1−ϵ Ptϵ Yt dj
0

Pulling out of the integral things which don’t depend on j:

Z 1
P t Yt = Ptϵ Yt Pt (j)1−ϵ dj
0

Simplifying, we get an expression for the aggregate price level:

1
Z 1  1−ϵ
Pt = Pt (j)1−ϵ dj (6)
0

3.2 Aside on the (Constant) Elasticity of Substitution

The elasticity of substitution for the production function is defined as the percentage change in
relative demand for two intermediates (call them k and j) with respect to their relative prices:

∂ (Yt (k)/Yt (j)) Pt (j)/Pt (k)

×
∂ (Pt (j)/Pt (k)) Yt (k)/Yt (j)
Take careful note of the indices here – this is measuring how the relative demand for intermediate
k changes when the relative price of good j goes up.
From the demand function for each intermediate, (5), we have:

4
  −ϵ  ϵ
Yt (k) Pt (k) Pt (j)
= =
Yt (j) Pt (j) Pt (k)
The derivative part of the elasticity is then:
 ϵ−1
∂ (Yt (k)/Yt (j)) Pt (j)
=ϵ
∂ (Pt (j)/Pt (k)) Pt (k)
This just says (naturally) that the relative demand for intermediate k will rise (ϵ > 0) when
the relative price of intermediate j increases.
The ratio of relative prices to relative demand is:
 −ϵ  1−ϵ
Pt (j)/Pt (k) Pt (j) Pt (j) Pt (j)
= =
Yt (k)/Yt (j) Pt (k) Pt (k) Pt (k)
Therefore, the elasticity of substitution is:
 ϵ−1  1−ϵ
∂ (Yt (k)/Yt (i)) Pt (k)/Pt (j) Pt (j) Pt (j)
× =ϵ =ϵ
∂ (Pt (k)/Pt (i)) Yt (k)/Yt (j) Pt (k) Pt (k)
The elasticity of substitution is therefore a constant at ϵ. As ϵ → ∞, intermediates are perfect
substitutes – the CES aggregator is just linear in varieties then.

3.3 Aside: More on CES Functions

The CES production function shown above looks kind of gnarly. There’s an integral and complicated
exponents.
It’s useful to take a step back and note that we’ve actually been repeatedly using a special case
of a CES function the entire course. That is the Cobb-Douglas production function. In particular,
suppose that output is given by:

1
Yt = (αKtν + (1 − α)Ntν ) ν

This is a special case (two inputs, not an infinite number of inputs) with share parameters (α
and 1 − α) of what we did above (relabeling ν = ϵ−1 ϵ . The function has constant returns to scale –
if you scale both capital and labor by a factor ϑ, you scale output by the same factor.
Suppose that ν = 1. This corresponds to a case of perfect substitutes – the production function
is just linear in capital and labor. The Cobb-Douglas case occurs when ν = 0 (which would
correspond to ϵ = 1 as written above). To see why ν = 0 corresponds to Cobb-Douglas, take logs:

1
ln Yt = ln (αKtν + (1 − α)Ntν )
ν
I took logs because it allows me to invoke L’Hopital’s rule. When ν = 0, we get ln 1/0 = 00 . To
make L’Hopital’s rule easier to use, rewrite the term in parentheses using exponents and logs:

5
 1
ln Yt = ln (α exp(ν ln Kt ) + (1 − α) exp(ν ln Nt ))
ν
L’Hopital’s rule says (when both numerator and denominator go to 0):

f (ν) f ′ (ν)
lim = ′
ν→0 g(ν) g (ν)
g(ν) = ν, so g(0) = 0 and g ′ (ν) = 1 regardless of what value ν takes on. For the numerator, we
have:

f (ν) = ln (α exp(ν ln Kt ) + (1 − α) exp(ν ln Nt ))

And:

1
f ′ (ν) = [α ln Kt exp(ν ln Kt ) + (1 − α) ln Nt exp(ν ln Nt )]
α exp(ν ln Kt ) + (1 − α) exp(ν ln Nt )

When ν = 0, we get:

f (0) = ln 1 = 0

Hence, we can use L’Hopital’s rule. Evaluating the derivative at ν = 0, we have:

1
f ′ (0) = [α ln Kt + (1 − α) ln Nt ]
1
This is because exp(0) = 1. We therefore have:

ln Yt = α ln Kt + (1 − α) ln Nt

Or, taking the exponential function to undo the logs:

Yt = Ktα Nt1−α

So, the Cobb-Douglas production function is a special case of a CES production function. In
this case, the elasticity of substitution (defined above) is one – a one percent increase in the relative
price of two factors results in a one percent increase in the relative use of those factors.
In the linear case, when the elasticity of substitution goes to ∞, goods are perfect substitutes –
a one percent increase in the relative price of two factors results in an infinite shift in the relative
use of those factors. In contrast, when the elasticity of substitution is less than one, we say that
two factors are complements. In this case, a one percent increase in the relative price of two factors
results in a less than one percent increase in the relative use of the cheaper factor. In the limiting
case in which the elasticity of substitution goes to 0, we have a Leontief function in which goods
are perfect complements.

6
 The reason we assume ϵ > 1 (but less than ∞) above is that in monopolistic competition goods
are substitutes, but imperfectly so.
Note that sometimes CES functions are referred to as Armington aggregators. They are very
common in international macro.

3.4 Aside: Putting the CES Aggregate in Preferences

As written, I have done this with a two-sector production model: the final good that is sold to the
household, and intermediate goods. We could alternatively put the CES aggregator used by the
final good in preferences. In particular, the household problem would be:

∞ !
Ct1−σ N 1+χ

X
t Mt
max E0 β −θ t + ψ ln
Ct ,Nt ,Bt+1 ,Mt 1−σ 1+χ Pt
t=0

s.t.

Pt Ct + Bt+1 + Mt − Mt−1 ≤ Wt Nt + Pt Dt − Pt Tt + (1 + it−1 )Bt

Z 1 ϵ
 ϵ−1
ϵ−1
Ct = Ct (j) ϵ
0
Z 1
P t Ct = Pt (j)Ct (j)dj
0

This just says that the household gets utility from a composite consumption basket, that is a
CES aggregate of the individual consumption varieties. ϵ > 1 still has the interpretation as the
elasticity of substitution. The household’s FOC for Ct is:

Ct−σ = λt Pt

If we think about the choice of an individual variety, we would have the FOC:

Z 1  (1−σ)ϵ −1
∂L ϵ ϵ−1 ϵ−1
ϵ−1 ϵ−1
=0⇔ Ct (j) ϵ Ct (j) ϵ −1 = λt Pt (j)
∂Ct (j) ϵ−1 0 ϵ
Simplifying a bit, and substituting out λt = Ct−σ Pt−1 , we have:

Z 1  1−σϵ
ϵ−1 ϵ−1 1 Pt (j)
Ct (j) ϵ Ct (j)− ϵ = Ct−σ
0 Pt
Which can be written:

Z 1  ϵ(σϵ−1)  −ϵ
ϵ−1 ϵ−1
Pt (j)
Ct (j) ϵ Ct (j) = Ctσϵ
0 Pt
But note that:

7
 σϵ2
Z 1 ϵ−1
 ϵ−1
Ctσϵ = Ct (j) ϵ
0

So, we have:

 ϵ(σϵ−1) σϵ2
Z 1 − ϵ−1  −ϵ
ϵ−1 ϵ−1
Pt (j)
Ct (j) ϵ Ct (j) =
0 Pt
Which is:
ϵ
− ϵ−1 −ϵ
Z 1 
ϵ−1 Pt (j)
Ct (j) ϵ Ct (j)
0 Pt
But the first part of the left hand side is just Ct−1 !. So we have:
 −ϵ
Pt (j)
Ct (j) = Ct (7)
Pt
This is the same interpretation as the demand function for each intermediate from the final
good producers, just written in terms of consumption instead of output.

3.5 Intermediate Producers

There are a continuum of intermediate goods producers, j ∈ [0, 1]. A typical intermediate producer
produces output according to a constant returns to scale technology in labor, with a common
productivity shock, At :

Yt (j) = At Nt (j) (8)

Nominal profit for an intermediate producer is:

Pt Dt (j) = Pt (j)Yt (j) − Wt Nt (j)

We can sub out for Nt (j) using the production function:

Wt
Pt Dt (j) = Pt (j)Yt (j) − Yt (j)
At
Wt /At is the ratio of the nominal wage (wt is the real wage) to the marginal product of labor,
which is just At . This is just (nominal) marginal cost – to produce an extra unit of output, a firm
needs 1/At extra units of labor, which costs Wt . wt /At is real marginal cost.
Plug in the demand function for each intermediate:
 −ϵ  −ϵ
Pt (j) Wt Pt (j)
Pt Dt (j) = Pt (j) Yt − Yt
Pt At Pt
Which can be written:

8
 Wt
Pt Dt (j) = Pt (j)1−ϵ Ptϵ Yt − Pt (j)−ϵ Ptϵ Yt
At
We can also write this in real terms by dividing through by Pt :

wt
Dt (j) = Pt (j)1−ϵ Ptϵ−1 Yt − Pt (j)−ϵ Ptϵ Yt
At

3.5.1 No Price Rigidity

If there were nothing more to the problem, we could solve for the Pt (j) that would maximize profit.
Take the derivative of real dividends with respect to Pt (j):

∂Dt (j) wt
= (1 − ϵ)Pt (j)−ϵ Pt1−ϵ Yt + ϵ Pt (j)−ϵ−1 Ptϵ Yt
∂Pt (j) At
Setting this equal to zero and simplifying a bit:

wt
(ϵ − 1)Pt (j) = ϵ Pt
At
Or, since Wt /At is nominal marginal cost, M Ct :

ϵ
Pt# = M Ct
ϵ−1
Pt# is the optimal price, which does not vary across j – it is the same for all intermediates. The
optimal price is just a markup over marginal cost (this follows because all firms face the same wage
and have the same productivity). Since we are assuming that ϵ > 1, the term ϵ/(ϵ − 1) > 1. This
is the desired markup of price over marginal cost.
Since all firms would choose the same price, Pt∗ = Pt (the aggregate price index). We could
then write this as:

ϵ−1
wt = At (9)
ϵ
(9) is a labor demand condition in the aggregate. In an efficient allocation (i.e. what a planner
would choose), we would have wt = At (i.e firms would hire labor up until the point at which
the real wage equals the marginal product of labor, which is just At with a linear production
technology). The term ϵ−1 ϵ , which is the inverse markup, is less than one. This says that the
equilibrium allocation is going to be distorted relative to an efficient allocation. This distortion
arises because of the monopoly power of intermediate firms, where the strength of their monopoly
power depends on the size of ϵ. The more substitutable intermediates are (i.e. the bigger is ϵ), the
smaller will be the distortion.

9
3.5.2 Staggered Price Setting

Let us instead assume that prices are “sticky.” In particular, each period there is a fixed probability
of 1 − ϕ that a firm can adjust its price. This means that the probability a firm will be stuck with
a price one period is ϕ, for two periods is ϕ2 , and so on. Consider the pricing problem of a firm
given the opportunity to adjust its price in a given period. Since there is a chance that the firm
will get stuck with its price for multiple periods, the pricing problem becomes dynamic.
Because the pricing problem is dynamic, we need to discount future profit flows. Intermediate
βj λ
firms will use the stochastic discount factor of the household to do this – Λt,t+j = λtt+j . In
addition, when setting a price today, the firm will take into account the probability that a price
chosen today will be in effect in the future – this will be given by ϕj .
The dynamic problem of an updating firm can therefore be written:

∞  −ϵ  −ϵ !
X Pt (j) Pt (j) wt+s Pt (j)
max Et ϕs Λt,t+s Yt+s − Yt+s
Pt (j) Pt+s Pt+s At+s Pt+s
s=0

Multiplying out, and for simplicity defining mct = wt /At , we have:

∞
X
ϵ−1
ϕs Λt,t+s Pt (j)1−ϵ Pt+s Yt+s − mct+s Pt (j)−ϵ Pt+s
ϵ


max Et Yt+s
Pt (j)
s=0

The first order condition can be written:

∞
X ∞
X
(1 − ϵ)Pt (j)−ϵ Et ϵ−1
ϕs Λt,t+s Pt+s Yt+s + ϵPt (j)−ϵ−1 Et ϕs Λt,t+s mct+s Pt+s
ϵ
Yt+s = 0
s=0 s=0

Simplifying:
∞
X
Et ϕs Λt,t+s mct+s Pt+s
ϵ
Yt+s
ϵ s=0
Pt (j) = ∞
ϵ−1 X
ϵ−1
Et ϕs Λt,t+s Pt+s Yt+s
s=0

First, note that since nothing on the right hand side depends on j, all updating firms will update
to the same reset price, call it Pt# . We can write the expression more compactly as:

ϵ X1,t
Pt# = (10)
ϵ − 1 X2,t
Here:

X1,t = mct Ptϵ Yt + ϕEt Λt,t+1 X1,t+1 (11)

X2,t = Ptϵ−1 Yt + ϕEt Λt,t+1 X2,t+1 (12)

10
 If ϕ = 0, then the right hand side would reduce to mct Pt = M Ct . In this case, the optimal
ϵ
price would be a fixed markup, ϵ−1 , over nominal marginal cost.

3.6 Alternative Setup: Retailers and Wholesalers

Above, we assumed that production firms were potentially heterogeneous. There is an alternative
setup this is sometimes easier to use.
In this setup, the final output good is still a CES aggregate, but it is an aggregate of retail
output. Retailers face the same downward-sloping demand curve, but don’t hire labor. Instead,
they purchase wholesale output, YW,t (j) at price PW,t . There is a representative wholesale firm,
indexed by W , that produces wholesale output according to:

YW,t = At Nt

The wholesaler just hires all the labor supplied by the household. It is competitive and takes
its price, PW,t , as given. It nominal profit is:

Pt DW,t = PW,t At Nt − Wt Nt

The first order condition is:

Wt
PW,t =
At
You will note: the right hand side is the same as nominal marginal cost above. We can define
pW,t = PW,t /Pt as the real price of wholesale output, and then have:

wt
pW,t =
At
Where wt = Wt /Pt is the real wage. This is real marginal cost.
There are a continuum of retailers indexed by j ∈ [0, 1]. Their production function is simple:
they purchase some wholesale output, YW,t (j), and repackage this into retail output, Yt (j) = YW,t (j).
Given that Yt (j) = YW,t (j), their nominal flow profit is:

Pt Dt (j) = Pt (j)Yt (j) − PW,t Yt (j)

Plugging in the demand function for retail output that comes from the CES aggregator, this
is the same problem we encountered before given what PW,t is equal to. We get exactly the
same solution as above. This retailer-wholesaler distinction can be useful if there are features of
production that might not aggregate well (which is not an issue with a linear production function).

11
4 Policy, Equilibrium, and Aggregation
Now we need to close the model. First, we need to say something about money supply. Suppose
that the nominal money supply is set by the central bank and follows an AR(1) process in the
growth rate, where gtM = ln Mt − ln Mt−1 :

gtM = (1 − ρM )g M + ρM gt−1
M
+ sM εM,t (13)

g M is the steady state growth rate, εM,t is drawn from a standard normal distribution, and sM
is the standard deviation of the shock.
The government’s nominal budget constraint is:

it−1 BtG ≤ Pt Tt + Mt − Mt−1 + Bt+1

G
− BtG (14)

I am assuming that the government does no spending; this is easy to add. It will turn out to
be irrelevant whether the government issues any debt or not. But the government earns revenue
(in nominal terms) from “printing” money.
Bond market clearing requires that Bt = BtG in all periods. Combining (14) with the household’s
budget constraint at equality, we get:

P t Ct = W t N t + P t D t

There are two sources of dividends: dividends from ownership in the final good firm and div-
idends from ownership in the intermediate goods firms. The final good firm nominal dividend is
simple:
Z 1
Pt DtF = Pt Yt − Pt (j)Yt (j)dj
0

Define the dividends from ownership in the intermediate firms as:

Z 1 Z 1 Z 1
Pt DtI = [Pt (j)Yt (j) − Wt Nt (j)] dj = Pt (j)Yt (j)dj − Wt Nt (j)dj
0 0 0
R1
Market clearing for labor requires that 0 Nt (j)dj = Nt . We can therefore write this as:
Z 1
Pt DtI = Pt (j)Yt (j)dj − Wt Nt
0

Total dividends received by the household are the sum of dividends from the two types of firms:

Z 1 Z 1
Pt Dt = Pt DtF + Pt DtI = Pt Yt − Pt (j)Yt (j)dj + Pt (j)Yt (j)dj − Wt Nt = Pt Yt − Wt Nt
0 0

Plugging this into the household’s budget constraint, after making use of the market-clearing

12
condition for bonds and imposing the government’s budget constraint, gives the aggregate resource
constraint:

Yt = Ct (15)

To get an aggregate production function, note that individual intermediate production functions
are:
 −ϵ
Pt (j)
Yt (j) = Yt
Pt
The left hand side is:
 −ϵ
Pt (j)
At Nt (j) = Yt
Pt
Now integrate both sides:
Z 1 Z 1 −ϵ
Pt (j)
At Nt (j)dj = Yt dj
0 0 Pt
Which is:
Z 1 Z 1 −ϵ
Pt (j)
At Nt (j)dj = Yt dj
0 0 Pt
Using the market clearing condition for labor, this can be written:

At Nt = Yt vtP (16)

Where:
Z 1 −ϵ
Pt (j)
vtP = dj (17)
0 Pt
vtP is a measure of price dispersion – if there were no price-setting friction, firms would all
charge the same price, and this would be one. If prices are different, this expression will be bound
from below by unity.
An equilibrium in this economy is a set of prices (wt , it , Pt , Pt# , and Πt ) and allocations (Ct ,
Nt , Yt ) such that the households and firms behave optimally and markets clear. We need to close
the model with a description of the exogenous productivity variable, which we assume follows a
stationary AR(1):

ln At = ρA ln At−1 + sA εA,t (18)

Our full set of conditions can be written:

13
 1 = Et Λt,t+1 (1 + it )Π−1
t+1 (19)

θNtχ = wt Ct−σ (20)

 
Mt σ 1 + it
= ψCt (21)
Pt it
Ct −σ
 
Λt−1,t = β (22)
Ct−1
Z 1 
1−ϵ 1−ϵ
Pt = Pt (j) dj (23)
0

ϵ X1,t
Pt# = (24)
ϵ − 1 X2,t
X1,t = mct Ptϵ Yt + ϕEt Λt,t+1 X1,t+1 (25)

X2,t = Ptϵ−1 Yt + ϕEt Λt,t+1 X2,t+1 (26)

wt
mct = (27)
At
Yt = Ct (28)

At Nt = Yt vtP (29)
Pt (j) −ϵ
Z 1 
P
vt = dj (30)
0 Pt
ln At = ρA ln At−1 + sA εA,t (31)

gtM = (1 − ρM )g M + ρM gt−1
M
+ sM εM,t (32)
Pt
Πt = (33)
Pt−1
gtM = ln Mt − ln Mt−1 (34)

As written, this is 16 equations and 16 variables (Ct , Nt , Yt , Λt−1,t , it , Πt , Pt , Mt , Pt# , X1,t ,

X2,t , mct , wt , At , vtP , gtM ). I could reduce this down if I wanted to.
There are a couple of issues with how I wrote this model. First, I have the price level showing
up. To the extent to which the price level is growing over time, I want to write the model without
reference to that, and so need to re-write things in terms of the inflation rate. Second, I still have
j subscripts showing up in two places – first, in the definition of the aggregate price level, and
second, in the price dispersion term. I need to get rid of the j subscripts if possible. The Calvo
assumption allows us to do that.
Start with the definition of the price index. Note that exactly 1 − ϕ of firms will choose the
same price, Pt# . Since these firms are random, we can order them however we want along the unit
interval. We can therefore break up the integral as:

14
 Z 1−ϕ  1−ϵ Z 1
Pt1−ϵ = Pt# dj + Pt (j)1−ϵ dj
0 1−ϕ

Which becomes:

 1−ϵ Z 1
Pt1−ϵ = (1 − ϕ) Pt# + Pt (j)1−ϵ dj
1−ϕ

This follows because the reset price doesn’t depend on j. For the firms who cannot change their
price, their prices will just be whatever was charged in the previous period. So:

 1−ϵ Z 1
Pt1−ϵ = (1 − ϕ) Pt# + Pt−1 (j)1−ϵ dj
1−ϕ

Because price adjusters and non-adjusters are randomly chosen, taking the sum of a subset of
old prices is just proportional to summing over all old prices. Put another way:
Z 1 Z 1
Pt−1 (j)1−ϵ dj = ϕ Pt−1 (j)1−ϵ dj
1−ϕ 0
1−ϵ
But the right hand side is just Pt−1 !. Hence, we have:
 1−ϵ
Pt1−ϵ = (1 − ϕ) Pt# 1−ϵ
+ ϕPt−1

In effect, the current price level (raised to a power) is a convex combination of the reset price
(common across all firms) and the old price level. The Calvo assumption means we don’t need to
keep track of individual firm prices, just the aggregate price level. To write this in terms of inflation
rates, divide both sides by Pt1−ϵ . We get:
!1−ϵ
Pt#
 1−ϵ
Pt−1
1 = (1 − ϕ) +ϕ
Pt Pt

Define p# #
t = Pt /Pt as the optimal relative reset price. The last term is just the inverse of gross
price inflation. So we have:
 1−ϵ
1 = (1 − ϕ) p#
t + ϕΠϵ−1
t (35)

(35) describes the evolution of aggregate inflation dynamics, rather than the price level, and
does so in terms of stationary variables. Now go to the price dispersion term. Using similar logic,
we can break up the integral:
!−ϵ −ϵ
1−ϕ
Pt# 1
Z Z 
Pt−1 (j)
vtP = dj + dj
0 Pt 1−ϕ Pt

Here, I have used the fact that non-updating firms charge their previous period’s price. We can
−ϵ
multiply and divide the right-hand term by Pt−1 to get (since this doesn’t vary with j, it can go

15
inside or outside the integral):

−ϵ Z 1  −ϵ  −ϵ

 Pt−1 (j) Pt−1
vtP = (1 − ϕ) p#
t + dj
1−ϕ Pt−1 Pt
But this can be written:

−ϵ Z 1  −ϵ
 Pt−1 (j)
vtP = (1 − ϕ) p#
t + Πϵt dj
1−ϕ Pt−1
But now use the same trick as before. The integral is:
Z 1  −ϵ Z 1 −ϵ
Pt−1 (j) Pt−1 (j) P
dj = ϕ dj = ϕvt−1
1−ϕ Pt−1 0 Pt−1
So we can write the price dispersion term just in terms of its own lag:
 −ϵ
vtP = (1 − ϕ) p#
t + ϕΠϵt vt−1
P
(36)

b2,t = X2,t /P ϵ−1 .

b1,t = X1,t /P ϵ and X
Now, we need to play with the price-setting terms. Define X t t
Playing around:

X1,t X1,t+1
ϵ = mct Yt + ϕEt Λt,t+1 (37)
Pt Ptϵ
ϵ :
Multiply and divide the last term by Pt+1
 ϵ
b1,t = mct Yt + ϕEt Λt,t+1 X1,t+1
X
Pt+1
(38)
ϵ
Pt+1 Pt
So:

b1,t = mct Yt + ϕEt Λt,t+1 Πϵ X

X (39)
t+1 1,t+1
b

Do the same with X2,t+1 :

X2,t X2,t+1
ϵ−1 = Yt + ϕEt Λt,t+1
Pt Ptϵ−1
So:
 ϵ−1
b2,t+1 = Yt + ϕEt Λt,t+1 X2,t+1
X
Pt+1
ϵ−1 Pt
Pt+1
And finally:

b2,t+1 = Yt + ϕEt Λt,t+1 Πϵ−1 X

X (40)
t+1 2,t+1
b

Now go to the reset price condition. Re-write it so it is in terms of X

b1,t and X
b2,t :

16
 ϵ X1,t /Ptϵ Ptϵ
Pt# =
ϵ − 1 X2,t /Ptϵ−1 Ptϵ−1
But this is:

ϵ X
b1,t
Pt# = P
b2,t t
ϵ−1X

Which can be written in terms of p#

t :

ϵ X
b1,t
p#
t = (41)
ϵ−1X
b2,t

Now, let’s re-write things in terms of real balances, mt = Mt /Pt . For the money demand
specification, this is straightforward:
 
1 + it
mt = ψCtσ (42)
it
For the definition of nominal money growth, to write this in terms of real balance growth, add
and subtract logs of the price level as necessary:

gtM = (ln Mt − ln Pt ) + ln Pt − (ln Mt−1 − ln Pt−1 ) − ln Pt−1

Which is then:

gtM = ln mt − ln mt−1 + ln Πt (43)

Okay, now I’m done. The equilibrium is now characterized by:

1 = Et Λt,t+1 (1 + it )Π−1
t+1 (44)

θNtχ = wt Ct−σ (45)

 
σ 1 + it
mt = ψCt (46)
it
Ct −σ
 
Λt−1,t = β (47)
Ct−1
 1−ϵ
1 = (1 − ϕ) p#t + ϕΠϵ−1
t (48)

ϵ X
b1,t
p#
t = (49)
ϵ−1X
b2,t

b1,t = mct Yt + ϕEt Λt,t+1 Πϵt+1 X

X b1,t+1 (50)
b2,t = Yt + ϕEt Λt,t+1 Πϵ−1 X
X (51)
t+1 2,t+1
b

17
 wt
mct = (52)
At
Yt = Ct (53)

At Nt = Yt vtP (54)
 −ϵ
vtP = (1 − ϕ) p#t + ϕΠϵt vt−1
P
(55)

ln At = ρA ln At−1 + sA εA,t (56)

gtM = (1 − ρM )g M + ρM gt−1
M
+ sM εM,t (57)

gtM = ln mt − ln mt−1 + ln Πt (58)

This is the same system of equations, but it is now 15 variables and 15 equations (I have
dropped Pt and re-written things in terms of mt instead of Mt ). Importantly, nothing here inherits
any nominal trend, and there are no j subscripts – this is just aggregate variables!

5 The Steady State

In the non-stochastic steady state, A = 1, real variables are constant, and g M is a constant
parameter (the average growth rate of the nominal money supply). To solve for the steady state,
note that this requires:

ln Π = g M

Hence:

Π = exp(g M )

If g M = 0, for example, then steady state gross inflation will be 1 (so net inflation will be zero).
The steady state stochastic discount factor is:

Λ=β

Therefore, the steady state nominal interest rate is:

1 + i = Π/β

Given Π, we then have steady state reset price inflation as:

 1−ϵ
1 − ϕΠϵ−1 = (1 − ϕ) p#

So:

18
 1
1 − ϕΠϵ−1
  1−ϵ
#
p =
1−ϕ
Next, in steady state, note that:

mcY
X
b1 =
1 − ϕβΠϵ
Y
X
b2 =
1 − ϕβΠϵ−1
Hence:

X
b1 1 − ϕβΠϵ−1
= mc
X
b2 1 − ϕβΠϵ
Therefore, we can solve for mc (which is in turn equal to the real wage):

ϵ − 1 # 1 − ϕβΠϵ
mc = p
ϵ 1 − ϕβΠϵ−1
Now we can solve for steady state price dispersion:

−ϵ
(1 − ϕ) p#
vP =
1 − ϕΠϵ
Now we can solve for steady state N from the first order condition for labor supply. Since
C = Y = N/v P , we have:
 −σ
χ N
θN = mc
vP
So:

N σ+χ = (v p )σ mc

So:
1
(v p )σ mc
 
σ+χ
N=
θ
Then:

Y = N vP = C

Once we have Y and C, we can solve for steady state m from the money demand relationship:
 
σ 1+i
m = ψC
i

19
5.1 Long-Run Distortions
There are two distortionary features of this model in the long run: monopolistic competition and
inflation. To focus on the monopolistic competition distortion, it is easiest to think about a world
in which g M = 0 (since Π = 1, so there is no net inflation in the long run). It is easy to see that
mc = w = ϵ−1 ϵ < 1. The steady state wage without monopolistic competition would be w = mc = 1
(which coincides with ϵ → ∞). Monopolistic competition will have the effect of giving lower hours
and output in the long run.
The second distortion is inflation. When g M = 0 so that Π = 1, v P = 1. But for any g M ̸= 0,
v P > 1. This is like having lower productivity – for a given N , you get less Y . This arises because of
staggered price-setting. If trend inflation is non-zero (Π ̸= 1), then relative prices of intermediates
will not all be the same. But this means that some output is “lost” when aggregating up in the
CES aggregator.
To get a sense of these factors, I solve for the steady state using some different values of these
parameters. First, I fix β = 0.99, θ = 1, χ = 1, ϕ = 3/4, and ψ = 1. Then, fixing g M = 0 (so no
trend inflation), I solve for steady state output as a function of ϵ, for values of ϵ ranging from 2 to
20. This is shown below.

Figure 1: Steady-State Y and ϵ

1

0.95

0.9
Y

0.85

0.8

0.75

0.7
2 4 6 8 10 12 14 16 18 20

We can see that output is everywhere increasing in ϵ. The bigger ϵ is, the weaker is monopoly
power, and the less distorted is the economy.
Next, to get a sense of the role of trend inflation, I fix ϵ = 11, and consider values of g M
ranging from -0.005 up to 0.01. I plot steady state values of the optimal reset price, inflation, price
dispersion, output, and real money balances as a function of trend growth in money.

20
 Figure 2: Steady-State Inflation and Other Variables

1.04 1.015
#
1.03

1.02 1.01

vP
1.01
p#,

1 1.005

0.99

0.98 1
-5 0 5 10 -5 0 5 10
gM 10-3 gM 10-3

0.954 200

0.952

0.95 150
m
Y

0.948

0.946 100

0.944

0.942 50
-5 0 5 10 -5 0 5 10
gM 10-3 gM 10-3

Since steady state gross inflation is the exponential of steady state money growth, inflation
plotted against money growth is not exactly a 45-degree line (it has some curvature), though for
some levels of trend growth in money (which I am considering), net inflation would be approximately
equal to money growth. When g M = 0, we get p# = Π = 1. For negative values of money growth,
reset price inflation lies below inflation; for positive values, it lies above. Here is what is going on.
If trend inflation is positive, when given a chance to reset its price, a firm has to overadjust to
factor in trend inflation during the period in which it will be (in expectation) stuck with the price
it chooses today (and vice-versa on the downside). Essentially, firms want a constant markup of
price over marginal cost of ϵ/(ϵ − 1). If the aggregate price level is trending up due to inflation,
they have to over-adjust to get that desired markup on average over the duration of the time when
their price will be in effect.
The right plan plots price dispersion. This bottoms out when Π = 1 at v P = 1. It lies
everywhere else above unity. If there is trend inflation (or deflation), then prices are going to
be disperse. This effectively acts like a negative productivity shifter in the production function,
lowering output. Indeed, steady state output is essentially the mirror image of steady state price

21
dispersion. Steady state output peaks at g M = 0 so Π = 1 and v P = 1.
The mechanisms above call for Π = 1 (i.e. no steady state inflation). Any steady state inflation
ends up distorting relative prices of intermediate goods, resulting in lost output, which is bad. But,
as we see in the lower right panel, there is still the Friedman rule logic at play – deflation results
in higher steady state real money balances, which increases welfare. Any discussion of optimal
trend inflation must balance the Friedman rule logic with money in the utility function (lower-right
panel) against the bad effects of price dispersion (lower-left panel).

6 Impulse Responses
Next, I solve the model in Dynare and produce impulse responses. I need to parameterize the
model. I pick β = 0.99, σ = 3, χ = 1, and θ = 1. I set ϕ = 0.75 (as we will see below, this implies
that the average duration between price changes is four quarters). I set ϵ = 11. This implies that
average markups of price over marginal cost are 10 percent. I set ψ = 1 (this parameter really has
no influence on the model dynamics). I set g M = 0, so there is zero steady state inflation (Π = 1,
or π = 0). I set ρM = 0.5 and ρA = 0.9. I set the standard deviations of both the monetary and
productivity shocks to one percent (0.01).

6.1 Aside: Calibrating the Calvo Parameter

How does one come up with a reasonable value for ϕ? This is a really important parameter in the
model – the bigger it is (the stickier are prices), the bigger will be the effects of nominal shocks
and the more distorted will be the response of variables to real shocks.
It turns out that there exists a close mapping between ϕ and the expected duration of a price
change. Consider a firm that gets to update its price in a period. In expectation, how long will
it be stuck with that price? The probability of getting to adjust its price one period from now
is 1 − ϕ. The probability of adjusting in two periods is ϕ(1 − ϕ): ϕ is the probability it doesn’t
adjust after one period, and 1 − ϕ is the probability it can adjust in two periods. The probability
of adjusting in three periods is ϕ2 (1 − ϕ): ϕ2 is the probability it gets to the third period with its
initial price, and 1 − ϕ is the probability it can adjust in that period. And so on. So the expected
duration of a price chosen today is:
∞
X
Expected Duration = (1 − ϕ) ϕj−1 j
j=1

The summation part on the inside can be written:

22
 S = 1 + 2ϕ + 3ϕ2 + 4ϕ3 + . . .
Sϕ = ϕ + 2ϕ2 + 3ϕ3 + . . .
(1 − ϕ)S = 1 + ϕ + ϕ2 + ϕ3 + . . .
1
(1 − ϕ)S =
1−ϕ
1
S=
(1 − ϕ)2

1
The second to last line above uses the fact that 1 + ϕ + ϕ2 + · · · = 1−ϕ as long as ϕ < 1.
Combining this with the 1 − ϕ outside the summation term above, we have:

1
Expected Duration =
1−ϕ
Bils and Klenow (2004, JPE ) analyze micro data on pricing and computing the average length of
time between prices changes. Though there is substantial heterogeneity across types of goods (e.g.
the price of newspapers rarely changes, while gasoline changes daily), for most goods, prices change
on average once every six months, which would suggest that ϕ ≈ 1/2 at a quarterly frequency (av-
erage duration of two quarters). For these models to produce realistic responses to monetary policy
shocks you need ϕ much higher (more like the 0.75 value I’ve been using). So an important area of
research essentially involves ways to “flatten” the Phillips Curve without assuming counterfactually
large levels of price rigidity. I will typically assume something like ϕ = 3/4 or ϕ = 2/3.

6.2 IRFs to Monetary and Productivity Shocks

Consider first impulse responses to a monetary shock. I’m plotting the log level of the money supply
(the cumulated response of gtM . There is a permanent effect on the money supply that grows over
time (given that I assumed ρM > 0). Output increases, inflation increases, the price level increases,
real marginal cost (wt /At ) increases, and the nominal interest rate decreases (although not quite
on impact, though it does subsequent to that). These look kind of “standard” from the perspective
of most people’s basic intuition. The increase in real marginal cost is because firms have to pay
workers more to get them to work more, which is necessary for output to increase.
Effectively, price stickiness (ϕ > 0) is limiting how much the price level can react to the increase
in Mt in the short run. This means that mt goes up in the short run (but ultimately the long-run
increase in Mt is met by a proportional increase in Pt ). It is easiest to think about the real effects
of a nominal shock by focusing on the money demand relationship, and assuming that the nominal
rate is approximately constant. Then, in log-linear terms, we would have:

m
e t = σyt

For ease of notation (see discussion below), I am just going to use lowercase variables to denote

23
 Figure 3: IRFs to a Monetary Shock

M 10-3 Y 10-3
0.02 5 6

5
0.018 4

4
0.016 3
3
0.014 2
2

0.012 1
1

0.01 0 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

P 10-5 i mc
0.02 1 0.02

0 0.015
0.015

-1 0.01

0.01
-2 0.005

0.005 -3 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

24
 Figure 4: IRFs to a Productivity Shock

A 10-3 Y 10-3
0.012 2.5 1

0.01
0
2
0.008
-1
0.006 1.5
-2
0.004
1
-3
0.002

0 0.5 -4
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

10-3 P 10-5 i 10-3 mc

-2 -1 0

-3
-2
-4 -5

-5 -3

-6 -10
-4
-7

-8 -5 -15
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

log deviations of variables that are already uppercase. This says that output and real balances
must move one-to-one. This is effectively quantity theory logic – since Mt Vt = Pt Yt , we have
mt = Vt−1 Yt . If velocity is constant, as assumed in quantity theory (and as would be the case with
a constant nominal interest rate), then real balances and output move one-to-one. Price stickiness
allows the increase in the nominal money supply to cause real balances to rise in the short run,
which necessitates a temporary increase in output until price adjustment has taken place.
Next, consider a positive productivity shock. The impulse responses are shown below. Output
increases, but by substantially less than the increase in At . Mechanically, this means that hours,
Nt are going down (which is not shown). Inflation falls, as does the price level. For this parame-
terization, the nominal interest rate declines. Real marginal cost falls – this is saying that the real
wage increases less than the increase in productivity.
In terms of intuition, it is easiest to think about what is going in with the quantity theory logic.
When productivity goes up, output wants to increase. But it can only increase if mt increases.
Because prices are sticky, and there is no change in the money supply, the price level falls by “too

25
 Figure 5: IRFs to a Monetary Shock, Different ϕ

M 10-3 Y
0.02 6 0.02

0.018 5
0.015
4
0.016
3 0.01
0.014
2
0.005
0.012
1

0.01 0 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

P 10-4 i mc
0.02 1 0.025
= 0.75
0.02 =0
0.015 = 0.9

0.015
0.01 0
0.01

0.005
0.005

0 -1 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

little” relative to what it would do without price stickiness, which limits the increase in Yt .

6.3 The Role of Price Stickiness

We can solve the model for different values of ϕ. Recall that ϕ → 0 means that prices are flexible,
whereas ϕ → 1 means that prices are very sticky.
Below, I show IRFs to a monetary and productivity shock for three different values of ϕ: 0.75
(what I showed above), ϕ = 0 (prices flexible), and ϕ = 0.9 (prices more sticky).
If prices are flexible (dashed line), then the monetary shock has no effect on output or real
marginal cost. The price level jumps up almost all the way to its new higher steady state value.
This means that, on impact, real money balances decline (the price level jumps more than the
nominal money supply initially). To accommodate real money balances declining, the nominal
interest rate actually increases. Inflation jumps up a lot. There is no impact on output or real
marginal cost (nor on the real interest rate or hours). In contrast, when prices are stickier (bigger
ϕ), output reacts more, and the price level reacts less. Thus, naturally, the real effects of monetary

26
 Figure 6: IRFs to a Productivity Shock, Different ϕ

A 10-3 Y 10-3
0.012 6 5

0.01 5
0
0.008 4

0.006 3 -5

0.004 2
-10
0.002 1

0 0 -15
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

P 10-4 i 10-3 mc
0 0 0

-0.2
-5
-0.005
-0.4 = 0.75
-10 =0
= 0.9
-0.6
-0.01
-15
-0.8

-0.015 -1 -20
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

expansion are bigger the stickier are prices.

Next, consider a productivity shock.

6.4 The Relative Unimportance of Steady State Inflation

Above, I solved the model with g M = 0, which means Π = 1 (gross inflation, which equivalently
means π, net inflation, is zero in the steady state). Obviously, we live in a world where inflation is,
on average, positive.
It turns out that for low values of g M (equivalently, values of Π near one) the model dynamics
are not much affected. And it is much easier to linearize the model about a zero inflation steady
state (as we will see in a bit). Below, I show impulse responses to monetary and productivity shocks
for different parameterizations of steady state inflation. I’m using a quarterly calibration. Inflation
is typically quote at an annual frequency (so approximately four times the quarterly number).
Hence, g M = −0.005 corresponds to π A = −2 percent and g M = 0.005 corresponds to π A = 2
percent, where the A superscript stands for annualized.

27
 Figure 7: IRFs to a Monetary Shock, Different g M

M 10-3 Y 10-3
0.02 5 6

4 5
0.018

3 4
0.016
2 3
0.014
1 2
0.012
0 1

0.01 -1 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

P 10-5 i mc
0.02 1 0.02
gM = 0
0
g M = -0.005
0.015
g M = 0.005
0.015 -1

-2 0.01

0.01 -3
0.005
-4

0.005 -5 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

Qualitatively, the impulse responses for to both shocks are the same regardless of the level of
trend inflation. There are some (small) quantitative differences, particularly in response to the pro-
ductivity shock. But, ignoring trend inflation is typically not much of a big deal for understanding
equilibrium dynamics, at least in a small-scale model. This does not mean that trend inflation is
irrelevant, particularly for welfare – as noted above, higher trend inflation leads to price dispersion,
which works like a reduction in aggregate productivity. It also does not mean that substantially
higher levels of inflation (like 4 percent or more at an annualized frequency) will not have substan-
tive impacts (indeed, if trend inflation is high or low enough, it will not be possible to solve the
model).

28
 Figure 8: IRFs to a Productivity Shock, Different g M

A 10-3 Y 10-3
0.012 3 1

0.01
2.5 0

0.008
2 -1
0.006
1.5 -2
0.004

1 -3
0.002

0 0.5 -4
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

10-3 P 10-5 i 10-3 mc

-2 0 0

-3 -1

-4 -2 -5

-5 -3 gM = 0
g M = -0.005
-6 -4 -10 g M = 0.005

-7 -5

-8 -6 -15
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

6.5 A Taylor Rule Formulation

I have written the model with an exogenous rule for the money supply. This is not the way that the
model is typically written down. Modern central banks pay little attention to the money supply,
and instead focus on targeting short-term interest rates.
We can replace the money growth rule with an interest rate rule of the sort proposed by Taylor
(1993). In particular:

it = i + ϕπ (ln Πt − ln Π) + ϕy (ln Yt − ln Y ) + ut (59)

ut = ρu ut−1 + su εu,t (60)

One can alternatively write this in levels; taking logs gives the expression above.
 ϕπ  ϕy
Πt Yt
1 + it = (1 + i) exp(ut ) (61)
Π Y

29
 In this formulation, the (nominal) interest rate reacts to deviations of inflation from its long-run
level and to output from its steady state. For a determinate equilibrium (more on this below), we
need ϕπ and ϕy to be sufficiently big (more on this later). I will always assume that ϕπ > 1 – this
is typically referred to as the “Taylor principle.” The condition needed for determinacy is actually
slightly different than that, but if ϕy = 0, then ϕπ > 1 is necessary for a unique equilibrium. i
is the steady state interest rate (which is a function of the steady state real interest, β1 − 1, and
the steady state inflation rate, Π, which we can treat as an exogenous policy parameter). ut is
an autocorrelated shock – given that there are no endogenous state variables in the model, for
persistent effects of a monetary shock, we need an autocorrelated shock. Alternatively, we could
get persistence by assuming some form of interest rate smoothing (more on this later).
To solve the model with a Taylor rule, we simply replace the money growth rule with the rule
for it . This makes the nominal money supply an endogenous variable – it reacts to meet money
demand given the interest rate target from the policy rule. Impulse responses to a monetary shock
(ut ) and a productivity shock with a Taylor rule formulation are shown below. In generating these,
I assume ϕπ = 1.5 and ϕy = 0.5.

Figure 9: IRFs to a Monetary Shock, Taylor Rule

M 10-3 Y 10-3
0 0 0

-0.5
-0.1
-1
-1
-0.2
-2 -1.5
-0.3
-2
-3
-0.4
-2.5

-0.5 -4 -3
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

10-3 P 10-3 i mc
-2.5 5 0

-3
4

-3.5 -0.005
3
-4
2
-4.5 -0.01

1
-5

-5.5 0 -0.015
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

30
 Figure 10: IRFs to a Productivity Shock, Taylor Rule

A 10-3 Y 10-3
0.012 4 0

0.01
-1
3
0.008
-2
0.006 2
-3
0.004
1
-4
0.002

0 0 -5
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

P 10-3 i 10-3 mc
0 0 0

-1 -1
-0.01

-2 -2
-0.02
-3 -3
-0.03
-4 -4

-0.04 -5
-5

-0.05 -6 -6
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20

The responses to a monetary shock are qualitatively similar to what we get with a money growth
rule, except that the nominal interest rate response is more conventional. Output, inflation, the
price level, and the money supply all fall. Output and inflation eventually revert to where they
started, and the price level ends up permanently lower. In response to the productivity shock, we
also get a similar pattern – output rises, but undershoots the increase in productivity. Inflation
and the price level fall. One difference relative to the money growth rule – the price level ends
up permanently lower in response to the productivity shock (it was mean-reverting in the money
growth specification).
As noted, a Taylor rule makes the money supply endogenous – it adjusts to meet money demand
given the interest rate target. Below, I show the impulse response of the nominal money supply
to a productivity shock with a Taylor rule. The money supply increases when there is a positive
productivity shock. This means that the central bank is (partially) accommodating the productivity
shock when it follows a Taylor rule. This is what allows output to respond more (relative to the
money growth rule) in response to the productivity shock.
As long as utility from real money balances is additively separable, the evolution of the money

31
 Figure 11: Endogenous Money Response to Productivity Shock, Taylor Rule
M
0.6

0.5

0.4

0.3

0.2

0.1

0
0 2 4 6 8 10 12 14 16 18 20

supply with a Taylor rule is actually irrelevant for the equilibrium dynamics of output, inflation,
and the interest rate to any kind of shock. We could solve the model without reference to money
demand altogether and get the same exact impulse responses we have above. For this reason,
it is quite common to consider the economy to be “cashless” in New Keynesian models. In the
background, there is a demand for money and an endogenous reaction of the money supply to meet
that demand at the desired interest rate, but the exact specification of money demand is irrelevant
for the equilibrium dynamics of other variables. So, many people ignore it altogether.

7 Log Linearization
The basic New Keynesian model is most often presented in log-linear form. I’m going to use the
notation that xt = ln Xt − ln X = dX X for variables that are already capitalized, and x
t
e notation
for variables that are lowercase (with the exception that I’m doing to treat it = dit and πt d ln Πt ).
This just eases things up a bit.
Take logs of the Euler equation on bonds. We get:

ln 1 = ln Λt,t+1 + ln(1 + it ) − ln Πt+1

Note that we can write the stochastic discount factor in terms of consumption:

ln 1 = ln β − σ ln Ct+1 + σ ln Ct + ln(1 + it ) − ln Πt+1

Totally differentiating:

32
 0 = −σEt ct+1 + σct + it − Et πt+1

Here, I am using the notation that ct = dCt /C, it ≈ ln(1 + it ), and πt ≈ ln Πt . Imposing the
resource constraint (i.e. ct = yt ), we have:

1
yt = Et yt+1 − (it − Et πt+1 ) (62)
σ
(62) is often referred to as the “IS” equation. This sounds a little odd, because there is no
investment (nor saving) in the model, but it’s the same idea of imposing asset market-clearing
and taking the optimal condition (Euler equation) and imposing the aggregate resource constraint.
Since rt = it − Et πt+1 , this is showing a negative relationship between current output and the real
interest rate, taking the expected future level of output as given.
Next, note that the labor supply condition is already log-linear:

et − σct
χnt = w

Again, just to be clear nt = dNt N and ct = Ct /C, but w et = dwt /w since I’m already using a
lowercase variable. But we can substitute out ct = yt from the resource constraint, leaving:

et − σyt
χnt = w (63)

Next, log-linearize the money demand relationship:

ln mt = ln ψ + σ ln Ct + ln(1 + it ) − ln it

Totally differentiating, we have:

it
e t = σct + it −
m
i
Again, keep the notation in mind: it = dit . We can write this as:

i−1
m
e t = σyt + it (64)
i
Since i < 1, the coefficient on it is negative. This just says that money demand is increasing in
output and decreasing in the nominal interest rate.
I’m going to proceed a bit out of order. The marginal cost condition and production function
are already log-linear:

mc et − at
ft = w (65)

at + nt = yt + vetP (66)
dAt
Again, take note of my notation: for example, at = d ln At = A . Given this, exogenous

33
processes are straightforward:

at = ρA at−1 + sA εA,t (67)

getM = ρM get−1
M
+ sM εM,t (68)

getM = m
et − m
e t−1 + πt (69)

The nasty part involves linearizing the pricing conditions. To do this, we’re going to assume
we are linearizing about a zero inflation steady state: Π = 1 (so g M = 0). For small steady state
money growth rates, this isn’t going to be a big deal (see above), and it makes the math much
easier.
Start with the evolution of inflation condition. Take logs:
  1−ϵ 
ln 1 = ln (1 − ϕ) p#
t + ϕΠϵ−1
t

Now, totally differentiate. This is pretty easy because, in steady state, the term in brackets is
one. So we have:
 −ϵ
0 = (1 − ϕ)(1 − ϵ) p# dp#
t + (ϵ − 1)ϕΠ
ϵ−1
dΠt

Or:

1−ϵ dp#

#
0 = (1 − ϕ)(1 − ϵ) p t
+ (ϵ − 1)ϕΠϵ−1 dΠt
p#
Now, since p# = 1 if steady state inflation is zero, and because Π = 1 and dΠt = πt , we have:

p#
0 = (1 − ϕ)(1 − ϵ)et + ϕ(ϵ − 1)πt

Hence, we have:

1−ϕ #
πt = pe (70)
ϕ t
Next, go to the price dispersion term. Take logs:
  −ϵ 
ln vtP = ln (1 − ϕ) p#
t + ϕΠ ϵ P
v
t t−1

With zero trend inflation, we know that vtP = 1 in steady state. This makes the total differen-
tiation easier. We have:

dvtP
  −ϵ−1 
1 # # ϵ−1 P ϵ P
= P −ϵ(1 − ϕ) p dpt + ϵϕΠ v dΠt + ϕΠ dvt−1
vP v
Now, again using the fact that p# = Π = v P = 1, we can write this as:

34
 p#
vetP = −ϵ(1 − ϕ)et + ϵϕπt + ϕe
P
vt−1

Or:
 
p#
vetP = −ϵ (1 − ϕ)et − ϕπ t + ϕe
P
vt−1

But, using (70), the terms involving inflation and the relative reset price drop out, so we are
left with:

vetP = ϕeP
vt−1 (71)

But since there is nothing to shock vetP out of steady state (the point of approximation), we
have vetP = 0 – it drops out of the linearization (including in the production function). This would
not be true if trend inflation were non-zero, but around a zero inflation steady state, we can ignore
price dispersion.
Now we need to log-linearize the reset price expression. Since this is multiplicative, it is already
log-linear:

pe# b1,t − x
t =x b2,t

Where x
b1,t = dX
b1,t /X
b1 . Now we need to linearize those two things. We have:

h i
b1,t = ln mct Yt + ϕEt Λt,t+1 Πϵt+1 X
ln X b1,t+1

Totally differentiate:

dX
b1,t 1 h i
= Y dmct + mcdYt + ϕΛΠϵ dX
b1,t+1 + ϵϕΛΠϵ−1 X
b1 dΠt+1 + ϕΠϵ X
b1 dΛt,t+1
X
bt X
b1,t

Given that Λ = β and Π = 1, we can write this as:

1 h i
x
b1,t = mcY mc
f t + mcY yt + ϕβ X
b1 x
b1,t+1 + ϵϕβ X
b1 πt+1 + ϕβ X
b1 λt,t+1
X
b1
dΛt,t+1
Where λt,t+1 = Λ . We can write this further:

mcY mcY
x
b1,t = mc
ft + y + ϕβEt x
b1,t+1 + ϵϕβEt πt+1 + ϕβEt λt,t+1
Xb1 Xb1 t
mcY
Note, that, in steady state, X
b1 =
1−ϕβ assuming Π = 1, so this reduces further to:

b1,t = (1 − ϕβ)mc
x f t + (1 − ϕβ)yt + ϕβEt x
b1,t+1 + ϵϕβEt πt+1 + ϕβEt λt,t+1 (72)

Now do log-linearization of X
bt :

35
 h i
b2,t = ln Yt + ϕEt Λt,t+1 Πϵ−1 X
ln X b2,t+1
t+1

Totally differentiate:

dX
b2,t 1 h i
= dYt + ϕΛΠϵ−1 dX
b2,t + (ϵ − 1)ϕΛΠϵ−2 X
b2 dΠt+1 + ϕΛΠϵ−1 dλt,t+1
X
b2 X
b2
Which can be written:

1 h i
x
b2,t+1 = Y yt + ϕβ X b2,t+1 + (ϵ − 1)ϕβ X
b2 x b2 πt+1 + ϕβ X
bt λt,t+1
X
b2
Where I have imposed steady state values where relevant: Λ = β and Π = 1. Note that
Y
X2 = 1−ϕβ
b . Using this fact, we can write the above as:

b2,t+1 = (1 − ϕβ)yt + ϕβEt x

x b2,t+1 + (ϵ − 1)ϕβEt πt+1 + ϕβEt λt,t+1 (73)

Now subtract (73) from (72). We have:

b1,t − x
x b2,t = (1 − ϕβ)mc
c t + ϕβEt πt+1 + ϕβ (Et x1,t+1 − Et x2,t+1 )

Now, since x b2,t = pe#

b1,t − x t , we can write this:

pe# c t + ϕβEt πt+1 + ϕβEt pe#

t = (1 − ϕβ)mc t+1

But from (70), we know that:

ϕ
pe#
t = πt
1−ϕ
Plug this in, and we have:

ϕ ϕ2 β
πt = (1 − ϕβ)mc
f t + ϕβEt πt+1 + Et πt+1
1−ϕ 1−ϕ
But this is:

(1 − ϕ)(1 − ϕβ)
πt = f t + (1 − ϕ)βEt πt+1 + ϕβEt πt+1
mc
ϕ
But this is just:

(1 − ϕ)(1 − ϕβ)
πt = mc
f t + βEt πt+1 (74)
ϕ
(74) is called the New Keynesian Phillips Curve and is one of the most important equations in
modern macro. With this equation in hand, we have eliminated (and no longer need to keep track
of) “intermediate” variables like pe#
t , x
b1,t , or x
b2,t ¿

36
 To gain some intuition for this expression, solve it forward. We can express inflation as propor-
tional to a present discounted value of real marginal cost:
∞
(1 − ϕ)(1 − ϕβ) X s
πt = Et β mc
f t+s (75)
ϕ
s=0

With monopolistic competition, intermediate good firms want their price to be a markup (ϵ/(ϵ−
1)) over marginal cost. If current (or expected future) marginal cost is supposed to be high, firms
who can need to raise prices today to hit their desired markup. If they don’t adjust today, they will
have lower than desired markups in the future. So marginal cost pressures put upward pressure
on inflation. How much upward pressure depends on ϕ – if ϕ is small, the coefficient (1−ϕ)(1−ϕβ)
ϕ is
large, and lots of firms will adjust, so inflation will react to marginal cost (current or future) a lot
(and vice-versa if ϕ is low). In the limiting case where prices are flexible, (1−ϕ)(1−ϕβ)
ϕ → ∞, which
necessitates mc
c t+s = 0, which means all firms will have their desired markup each period.

7.1 Linearized Conditions in One Place

Below are the remaining conditions in one place, where I have substituted out pe# t , x b2,t , vetp ,
b1,t , x
ct , and λt,t+1 . As noted early, price dispersion is constant to a first-order approximation about a
zero inflation steady state, and pe# t , x
b1,t , x
b2,t are subsumed in the New Keynesian Phillips Curve.
I no longer need to keep track of the stochastic discount factor either, and I have imposed the
market-clearing condition that ct = yt . This leaves me with nine equations and nine variables (yt ,
it , πt , nt , w
et , mc e t , at , and getM ) as follows:
f t, m

1
yt = Et yt+1 − (it − Et πt+1 ) (76)
σ
et − σyt
χnt = w (77)
i−1
m
e t = σyt + it (78)
i
mc et − at
ft = w (79)

at + nt = yt (80)
(1 − ϕ)(1 − ϕβ)
πt = mc
f t + βEt πt+1 (81)
ϕ
getM = ρM get−1
M
+ sM εM,t (82)

at = ρA at−1 + sA εA,t (83)

getM = m
et − m
e t−1 + πt (84)

In terms of the economics of what these equations are saying, (76) is the IS equation, which
is essentially an aggregate demand condition (note that rt = it − Et πt+1 ). (77) is a labor supply
condition. (78) is a money demand condition. The money supply equation is (82), and (84) is

37
just a relationship between growth in the nominal money supply, real balances, and inflation. (79)
defines real marginal cost; as discussed earlier, this is essentially a labor demand condition. (80)
is the production function. (81) is the Phillips curve, which can be interpreted as an aggregate
supply relationship (more below). Essentially, we have three markets – for goods, for labor, and for
money – with demand and supply relationships going into each. In equilibrium, prices must adjust
so that all three hold. (83) is just an exogenous process for productivity.
Note that, if instead of an exogenous money supply rule, we wanted to model monetary policy
with an interest rate rule, we would just replace (82) with a Taylor rule. The money supply would
then be exogenous, given a target for the interest rate. The linearized Taylor rule (which takes out
the constants) would be:

it = ϕπ πt + ϕy yt + ut (85)

7.2 System Reduction, Flexible Price Output, the Output Gap, and the Natural
Rate
We can actually reduce the system quite a bit further. Combining the production function with
the labor supply condition to substitute out nt , we have:

χ(yt − at ) = w
et − σyt

From the labor demand condition, we can eliminate w

et from this expression:

χ(yt − at ) = mc
f t + at − σyt

But this can be written:

(σ + χ)yt = mc
f t + (1 + χ)e
at

Now, let us introduce an important concept, called the “flexible price level of output.” Some-
times this is called “potential output.”2 The flexible price level of output is defined as the equi-
librium level of output that would emerge in equilibrium in the hypothetical world in which prices
are completely flexible (i.e. ϕ = 0). We could get this using the conditions in levels, but it is
straightforward to do so in the linearized model. When prices are flexible, mc f t = 0 – real marginal
ϵ−1
cost will be constant (in levels, equal to ϵ ). This would just mean that all firms charge a constant
markup over marginal cost. Let’s call yetf the flexible price level of output. From above, we have:

(σ + χ)ytf = (1 + χ)at

Or:
2
Though one wants to be careful here. The flexible price level of output is not in general going to be efficient since
there is a monopoly distortion. Even without price stickiness, labor would be paid less than its marginal product.

38
 1+χ
ytf = at
σ+χ
In other words, since productivity is the only (real) exogenous state variable, equilibrium output
in the hypothetical world with flexible prices is just a linear function of it. But using this definition,
we can actually sub out real marginal cost. We have:

f t = (σ + χ)yt − (1 + χ)at
mc
σ+χ f
But since at = 1+χ yt , we have:
 
f t = (σ + χ) yt − ytf
mc

In other words, real marginal cost is just proportional to the output gap, yt − ytf . Define
xt ≡ yt − ytf as the gap. If xt > 0, then mct > 0, which means that the average markup is lower
than desired. This puts upward pressure on inflation (and vice-versa). With this, we can write the
Phillips Curve more compactly as:

πt = γxt + βEt πt+1 (86)

Where γ = (1−ϕ)(1−ϕβ)
ϕ (σ + χ). This is in-line with empirical Phillips Curves, which show some
relationship between economic activity and inflation (typically in terms of an unemployment rate,
which would naturally be negatively related to an output gap). With this, we can reduce the system
further:

1
yt = Et yt+1 − (it − Et πt+1 ) (87)
σ
i−1
m
e t = σyt + it (88)
i
πt = γxt + βEt πt+1 (89)

getM = ρM get−1
M
+ sM εM,t (90)

at = ρA at−1 + sA εA,t (91)

getM = m
et − m
e t−1 + πt (92)
1+χ
ytf = at (93)
σ+χ
xt = yt − ytf (94)

In this formulation, (89) summarizes the supply-side of the economy: production, labor demand,
labor supply, and price-setting. (87) summarizes demand for goods, and (88) the demand for money.
We can actually go even further. Add and subtract ytf and Et yt+1f
from the (87):

39
 1
yt − ytf + ytf = Et yt+1 − Et yt+1
f f
+ Et yt+1 − (it − Et πt+1 )
σ
Which can be written more compactly:

f 1
xt = Et xt+1 + Et yt+1 − ytf − (it − Et πt+1 )
σ
Now, one might wonder where this is going. Define the natural rate of interest, rtf (or sometimes
rt∗ , or “r-star”) as the hypothetical real interest rate that would clear the goods market with no
price rigidity. This would satisfy:

1 f f
r = Et yt+1 − ytf
σ t
But then we can write the IS equation:

1 
xt = Et xt+1 − it − Et πt+1 − rtf
σ
The term it −Et πt+1 −rtf = rt −rtf is a (real) interest rate gap – the gap between the equilibrium
real interest rate and the natural rate. We can actually write the entire system as a function of rtf
and drop ytf and at . How? Note that:
 
rtf = σ Et yt+1
f
− ytf

But, from above, this may be written:

 
1+χ 1+χ σ(1 + χ)
rtf =σ Et at+1 − at = (Et at+1 − at )
σ+χ σ+χ σ+χ
But since Et at+1 = ρA at , this is:

σ(1 + χ)(ρA − 1)
rtf = at
σ+χ
But solving backwards, we can write this as:

σ(1 + χ)(ρA − 1) σ(1 + χ)(ρA − 1)

rtf = ρA at−1 + sA εA,t
σ+χ σ+χ
σ+χ f
But since at−1 = σ(1+χ)(ρA −1) rt−1 , we have:

rtf = ρA rt−1
f
+ sr εA,t (95)

Where:

σ(1 + χ)(ρA − 1)
sr = sA
σ+χ
Then the full system is:

40
 1 
xt = Et xt+1 − it − Et πt+1 − rtf (96)
σ
πt = γxt + βEt πt+1 (97)
i−1
m
e t = σyt + it (98)
i
getM = ρM get−1
M
+ sM εM,t (99)

getM = m
et − m
e t−1 + πt (100)

rtf = ρA rt−1
f
+ sr εA,t (101)

(96) is an aggregate demand relationship (IS equation) and (97) is an aggregate supply rela-
tionship. (98)-(100) describe the money market, and (101) is an exogenous process for the natural
rate of interest.

7.3 The Three-Equation NK Model

Although I have done it with money, it is uncommon to even bother modeling money in most
applications. Rather, most of the time researchers assume some kind of interest rate rule. A
modified version of the Taylor rule (that reacts to the output gap, rather than the log deviation of
output from steady state) is:

it = ϕπ πt + ϕx xt + ut (102)

Where ut obeys a potentially persistent process. One also often sees people build in interest
rate persistence via a lagged interest rate term:

it = ρi it−1 + (1 − ρi )ϕπ πt + (1 − ρi )ϕx xt + ut (103)

Here, 0 ≤ ρi < 1, and I have just scaled the coefficients on inflation and the output gap by
1 − ρi (more on why later). In (103), we would typically assume that ut is an iid shock, but one
could entertain persistence. Either way, if one includes (102) or (103), one gets a three-equation
model:

1 
xt = Et xt+1 − it − Et πt+1 − rtf (104)
σ
πt = γxt + βEt πt+1 (105)

it = ϕπ πt + ϕx xt + ut (106)

This is three equations in three endogenous variables – inflation, the output gap, and the
interest rate. We could include the money demand equation, (99), and the relationship between
nominal money growth, real balances, and inflation, (100), if we wanted to (note we would drop
the exogenous money supply rule in favor of the interest rate rule), but we actually don’t need to

41
in order to determine xt and it . We also need the exogenous process for rtf (and potentially ut ),
but (104)-(106) summarize the endogenous variables of the model concisely in three equations.
This three equation model is nice. It’s nice for several reasons. First, it is easy to work with
and build intuition for. Second, it has an AD-AS flavor from an undergraduate textbook. Third,
it focuses in on exactly the variables central banks care about. In the US, the Federal Reserve
has a “dual mandate” for “price stability” and “full employment.” We can interpret πt (inflation)
as measuring price stability and xt (the output gap) as a measure of how close we are to full
employment. The Fed’s principal policy tool is the short-term interest rate, it . So these three
equations describe the two targets modern central banks focus on (inflation and the output gap /
unemployment gap) and its principal instrument (a nominal interest rate). Kind of cool!

7.4 Cost-Push Shocks

It is common to include a “cost-push” shock into the NKPC:

πt = γxt + βEt πt+1 + upt (107)

Where upt follows a stationary AR(1) process:

uPt = ρp upt−1 + sp εp,t (108)

A shock like this can be motivated as time-variation in desired markups (i.e. a shock to ϵt in
the full, non-linear model). People also sometimes think about a shock like this as representing an
oil price shock. In any event, as we will see, this shock is “nice” in that it introduce a non-trivial
tradeoff for a central bank that wants to stabilize both inflation and the output gap.

8 Method of Undetermined Coefficients

The method of undetermined coefficients is a way to solve for linearized policy functions by hand.
In principle, it could be used in a more complicated model, but it is especially useful in a small-scale
model with no endogenous state variables. It allows us to build better intuition for how the model
economy works than simply plugging equations into a computer.
The basic gist of the methodology is to guess that forward-looking jump variables are linear
functions of state variable(s). In principle, we could do this with many exogenous states, but it’s
most straightforward with just one. Consider the basic NK model from above, (104)-(106). Suppose
that inflation and output are linear function of the exogenous state variable, rtf :

πt = θπ rtf
xt = θx rtf

42
 Given that we have assumed rtf (which can be thought of as exogenous) is stationary, we do not
need to worry about imposing that inflation and the gap do not explode. These linear functions
will be time-invariant (i.e. they hold at all times). Plug them into the NKPC:

θπ rtf = γθx rtf + βθπ Et rt+1

f

f f
The last line follows because Et πt+1 = θπ Et rt+1 . But Et rt+1 = ρA rtf . So we have:

θπ rtf = γθx rtf + βθπ ρA rtf

This is one equation in two unknowns. rtf drops out. Solving for θπ as a function of θx , we
have:

γ
θπ = θx
1 − βρA
Now go to the IS equation and do the same. Plug in the Taylor rule to write it in terms of πt
and xt (and assume ut = 0, which is not without loss of generality given the linearity of the model
– we will get the same policy functions in response to rtf whether we allow for the monetary policy
shock or not):

1  1
θx rtf = θx Et rt+1
f
− f
ϕπ πt + ϕx xt − θπ Et rt+1 + rtf
σ σ
f
Note again that Et rt+1 = ρA rtf , and plug in inside the Taylor rule to eliminate πt and xt . We
have:

σθx rtf = σθx ρA rtf − ϕπ θπ rtf − ϕx θx rtf + θπ ρA rtf + rtf

We can now drop the rtf . We have:

θx (σ(1 − ρA ) + ϕx ) + θπ (ϕπ − ρA ) = 1

We now have two equations in two unknowns. Plug in for θπ using the expression we found
from the NKPC:

γ(ϕπ − ρA )
θx (σ(1 − ρA ) + ϕx ) + θx =1
1 − βρA
Which is:

θx [(1 − βρA ) (σ(1 − ρA ) + ϕx ) + γ(ϕπ − ρA )] = 1 − βρA

Or:

1 − βρA
θx = (109)
(1 − βρA ) (σ(1 − ρA ) + ϕx ) + γ(ϕπ − ρA )

43
 But then:

γ
θπ = (110)
(1 − βρA ) (σ(1 − ρA ) + ϕx ) + γ(ϕπ − ρA )
These are pretty intuitive. First, suppose that γ → ∞ (prices are completed flexible). Then
θx = 0 (but θπ will not be, since γ is in both the numerator and the denominator). In other
words, if prices are flexible, then the output gap is always zero (i.e. yt = ytf ). Second, suppose
that ϕπ → ∞. Then both θx and θπ go to 0. Third, suppose that ϕx → ∞. Again, both θπ and
θx go to zero. This is a manifestation of something called the “Divine Coincidence” – in the basic
NK model, even though there is on instrument (the policy rate) and two targets (inflation and the
output gap), it is possible to stabilize both targets at the same time. To see this better, note that
if ϕπ → ∞, then πt = 0 for the Taylor rule to feature a non-explosive policy rate. But, from the
Phillips Curve, if inflation is always zero, then xt = 0 as well.
We can use these coefficients to recover an expression for the nominal interest rate (which is
effectively a static variable, given that it only appears in the equilibrium conditions at date t. We
have:

it = (ϕπ θπ + ϕx θx ) rtf

Which is:

ϕπ γ + ϕx (1 − βρA )
it = rf (111)
(1 − βρA ) (σ(1 − ρA ) + ϕx ) + γ(ϕπ − ρA ) t

9 Alternative Price Stickiness Model: Rotemberg (1982) Pricing

We have thus far modeled price stickiness via the Calvo (1983) assumption, wherein each period
firms face a constant hazard of being able to adjust their price. This means that in equilibrium
firms are heterogeneous, but the model is rigged in such a way that aggregation works out nicely.
An alternative pricing assumption is based on Rotemberg (1982). In the Rotemberg model, firms
face a quadratic cost of price adjustment. In equilibrium they all end up behaving identically. To a
first order approximation about a zero inflation steady state, the Rotemberg and Calvo models can
be parameterized to be identical. They are not identical to order higher than one, and they have
different implications for micro data (e.g. in the Calvo model only a fraction of firms will adjust
their price in a given period, but in the Rotemberg model all firms will be changing their prices
each period).
In the Rotemberg model, there are many intermediate goods firms. Firms face the same demand
curve from the final good firm as earlier. They produce output according to Yt (j) = At Nt (j), just
as in the Calvo model. Nominal flow profit for producer j is given by:
 −ϵ  2
Pt (j) ψp Pt (j)
Πt (j) = Pt (j) Yt − Wt Nt (j) − − 1 Pt Yt (112)
Pt 2 Pt−1 (j)

44
 The parameter ψp ≥ 0 measures the cost of price adjustment, and it is measured in units of the
final good. We can write the profit function in real dollars as:
 −ϵ  2
Pt (j) Yt ψp Pt (j)
Πt (j) = Pt (j) − wt Nt (j) − − 1 Yt (113)
Pt Pt 2 Pt−1 (j)
Since Nt (j) = Yt (j)/At , we can write this as:
 −ϵ  2
Pt (j) Yt wt ψp Pt (j)
Πt (j) = Pt (j) − Yt (j) − −1 Yt (114)
Pt P t At 2 Pt−1 (j)
But since mct = wt /At , this is just:
 −ϵ  2
Pt (j) Yt ψp Pt (j)
Πt (j) = Pt (j) − mct Yt (j) − −1 Yt (115)
Pt Pt 2 Pt−1 (j)
Plugging in the demand function for variety j, we get:
 −ϵ  2
Pt (j) Yt ψp Pt (j)
Πt (j) = Pt (j) − mct Pt (j)−ϵ Ptϵ Yt − − 1 Yt (116)
Pt Pt 2 Pt−1 (j)
Each period, firms choose price to maximize the expected present discounted value of flow
profit, where discounting is by the household’s stochastic discount factor. The problem is dynamic
because the adjustment cost function means that current prices are relevant for future profits. The
optimality condition for price-setting can be written:

 −ϵ
Pt (j) Yt
(ϵ − 1) =
Pt Pt
 −ϵ−1  
Pt (j) Yt Pt (j) Yt
ϵmct − ψp −1 +
Pt Pt Pt−1 (j) Pt−1 (j)
    
Pt+1 (j) Pt+1 (j) Yt+1
ψp Et Λt,t+1 −1 (117)
Pt (j) Pt (j) Pt (j)

This first order condition holds for all firms. We will therefore converge to a symmetric equi-
librium where all firms behave identically, so Pt (j) = Pt for all j. We can therefore write the FOC
as:
 
Yt Yt Yt Yt+1
(ϵ − 1) = ϵmct − ψp (Πt − 1) + ψp Et Λt,t+1 (Πt+1 − 1) Πt+1
Pt Pt Pt Pt
Which can be written:

Yt+1
ϵ − 1 = ϵmct − ψp πt Πt+1 + ψp Et Λt,t+1 πt+1 Πt+1
Yt
Here I am using Πt = PPt−1t
and πt = Πt − 1. To make life especially straightforward, suppose
that we have log utility over consumption, so Λt,t+1 = β CCt+1
t
. Since Ct = Yt will be the aggregate

45
resource constraint, all terms involving Yt and Yt+1 drop out, leaving:

ϵ − 1 = ϵmct − ψp πt Πt+1 + ψp βEt πt+1 Πt+1 (118)

If there is no price rigidity (ψp = 0), then this reduces to mct = ϵ−1
ϵ , which again just says that
price is a constant markup over marginal cost (equivalently, labor is paid a markdown relative to
its marginal product).
Let’s now log-linearize (118) about a zero inflation steady state. Take logs of both sides:

ln(ϵ − 1) = ln [ϵmct − ψp πt Πt+1 + ψp βEt πt+1 Πt+1 ]

Totally differentiate:

1
0= [ϵdmct − ψp πdΠt+1 − ψp Πdπt + ψp πdΠt+1 + ψp βΠdπt+1 ]
ϵ−1
We are linearizing about a zero (net) inflation steady state: this means π = 0, Π = 1, dπt = πt ,
and dΠt = πt . So we have:
 
1 dmct
0= mcϵ − ψp πt + ψp βEt πt+1
ϵ−1 mc
Which can be written:

f t − ψp πt + ψp βEt πt+1
0 = mcϵmc
ϵ−1
In steady state, mc = ϵ , so we can write this further as:

ψp πt = (ϵ − 1)mc
f t + ψp βEt πt+1

Or:

ϵ−1
πt = mc
f t + βEt πt+1 (119)
ψp
Although the coefficient on real marginal cost is different, this is exactly the same expression as
(74)! We can pick ψp to generate exactly the same slope of the NKPC if we want. We again have
that, if prices are quite flexible, then ψp is low, and the coefficient on marginal cost is big (and
vice-versa).
The rest of the model works out similarly without adjustment. The one area where there is
some potential adjustment is in the aggregate resource constraint. Because the adjustment cost
for changing prices comes out of profit, this will show up in the aggregate resource constraint. In
particular, the aggregate resource constraint will be:

ψp 2
Yt = Ct + π Yt (120)
2 t

46
 In other words, there is some resource loss due to inflation moving around. But if you linearize
this about a zero inflation steady state, it drops out:
 
ψp 2
ln Yt = ln Ct + πt Yt
2
dCt ψp 2 e
Yet = + π Yt + 2ψp πe
πt
Y 2
If we are linearizing about the point π = 0, then the latter terms drop out, and C = Y , leaving:

Yet = C
et (121)

This is analogous to how, in the Calvo model, price dispersion potentially lowers productivity,
but this term disappears to first order about a zero inflation steady state.

47