Working Paper No.

55

Monetary Policy with Nonlinear Phillips Curve

and Endogenous NAIRU

by

Willi Semmler and Wenlang Zhang

University of Bielefeld
Department of Economics
Center for Empirical Macroeconomics
P.O. Box 100 131

33501 Bielefeld, Germany

 Monetary Policy with Nonlinear Phillips Curve
and Endogenous NAIRU
Willi Semmler∗ Wenlang Zhang†
January 18, 2004

Abstract
The recent literature on monetary policy has questioned the shape of
the Phillips curve and the assumption of a constant NAIRU. In this paper
we explore monetary policy considering nonlinear Phillips curves and an
endogenous NAIRU which can be affected by the monetary policy. We
first study monetary policy with different shapes of the Phillips curve:
Linear, convex and convex-concave. We find that the optimal monetary
policy changes with the shape of the Phillips curve, but there exists a
unique equilibrium no matter whether the Phillips curve is linear or non-
linear. We also explore monetary policy with an endogenous NAIRU, since
some researchers, Blanchard (2003) for example, have proposed that the
NAIRU may be influenced by monetary policy. Based on some empirical
evidence and assuming that monetary policy can influence the NAIRU,
we find that there may exist multiple equilibria in the economy, different
from the results of models presuming a constant NAIRU.
JEL: E0, E5
Keywords: Nonlinear Phillips Curve, Endogenous NAIRU, Kalman
Filter

∗ Center for Empirical Macroeconomics, Bielefeld, and New School University, New York.
† Center for Empirical Macroeconomics, Bielefeld University, Germany. The authors would
like to thank Olivier Blanchard for his comments.

1
1 Introduction
Recently an important topic in studies of monetary policy has been the shape
of the Phillips curve. The shape of the Phillips curve plays an important role in
monetary policy and has important implications for policy-makers, because the
IS- and Phillips curves have been the core model for monetary policy studies
from a Keynesian perspective.
In earlier studies the Phillips curve has been proposed to be linear. But
most of the recent literature has casted doubt on the linearity of the Phillips
curve. Dupasquier and Ricketts (1998a), for example, survey several models
that explore why the Phillips curve may be nonlinear. It has been argued that
the Phillips curve may have three shapes. The first group of researchers propose
that the Phillips curve is convex. This possibility has been considered by Clark,
Laxton and Rose (1996), Schaling (1999) and Bean (2000) for example. A
convex Phillips curve appears in an economy subject to capacity constraints.
The second group of researchers propose that the Phillips curve is concave.
Eisner (1997), for example, reports some results concerned with this possibility.
A concave Phillips curve may exist in an economy where firms are not purely
competitive. Besides these two shapes of the Phillips curve, Filardo (1998), on
the basis of some empirical research with U.S. data, proposes that the Phillips
curve is not purely convex or concave, but instead convex-concave. He finds
that the Phillips curve is convex if the output gap is positive and concave if the
output gap is negative. Therefore he points out that the supporters of a convex
or concave Phillips curve have studied only one case and overlooked the other.
The shape of the Phillips curve has crucial implications for central banks, since
the optimal monetary policy may change with the shape of the Phillips curve.
Another important topic of monetary policy is whether the NAIRU is con-
stant. In the 1960s Friedman and Phelps proposed a vertical long-run Phillips
curve. The traditional view is that money can not influence the unemployment
rate in the long run and therefore the unemployment rate returns to the natural
rate or the NAIRU over time, which is presumed to be constant. The recent lit-
erature has put this view into question and proposes that the long-run Phillips
curve may be non-vertical and the NAIRU nonconstant. That is, the NAIRU
can be affected by the inflation rate and monetary policy. Gordon (1997), for
example, estimates the time-varying NAIRU for the U.S. with and without sup-
ply shocks. Some other researchers have also tried to estimate a time-varying
NAIRU. An earlier survey on the estimation of the NAIRU is given by Staiger,
Stock and Watson (1996). Blanchard (2003), moreover, remarks that the natu-
ral rate of unemployment has been affected by the real interest rate in Europe in
the 1970s and 1980s in different directions. He further proposes several mech-
anisms in which the real interest rate may influence the natural rate of the
unemployment. On the other hand, some researchers, Stiglitz (1997) for ex-
ample, maintain that the NAIRU may have feedback effects on policies in the
sense that it might produce a moving target for monetary policy. Therefore, an
economic model with an exogenous NAIRU might not explore monetary policy
properly.

2
 The remainder of this paper is organized as follows. In the second section
we explore monetary policy in a traditional model with a linear Phillips curve
and a constant NAIRU. In Section 3 we study monetary policy with different
shapes of the nonlinear Phillips curve, maintaining the assumption of a constant
NAIRU. In the fourth section, however, we study the monetary policy with an
endogenous NAIRU which can be affected by the monetary policy and the last
section concludes the paper.

2 Monetary Policy in a Traditional Model: Lin-

ear Phillips Curve with and without Expecta-
tion
The traditional IS curve reads as

µ̇(t) = αr(t) + ϑµ(t), α > 0,

where µ(t) denotes the gap between the actual unemployment u(t) and the
NAIRU un (t), namely u(t) − un (t), r(t) is the monetary policy instrument.
Assuming ϑ = 0 for simplicity, we obtain

µ̇(t) = αr(t). (1)

In the traditional model the NAIRU is assumed to be an exogenous variable

which remains constant because of the neutrality of money. “·” denotes the
derivative with respect to time t. The equation above implies

µ̇(t) = u̇(t) − u̇n (t) = u̇(t) = αr(t), (2)

since the NAIRU is constant. Let π denote the deviation of the actual inflation
rate from its target (assumed to be zero here). Following Walsh(1999), Ball
(1999) and Hall (2000) who assume that the inflation rate is affected by the
real interest rate as well as the unemployment gap, we write the Phillips curve
without expectation as1

π(t) = −βr(t) − θµ(t), β, θ > 0. (3)

1 In fact, the Phillips curve above is equivalent to an open economy Phillips curve with
π(t) = −τ e(t) − θµ(t),
with e(t) denoting the real exchange rate and τ the share of imports in domestic spending.
A higher e(t) means appreciation. Such a Phillips curve can be found in Walsh (1999) and
Guender (2001). Ball (1999) employs a similar equation except that the change of e(t) is
included in the open-economy Phillips curve. Moreover, following Ball (1999) we can link the
real exchange rate and the real interest rate by
e(t) = ωr(t) + ν(t),
with ν(t) being a white noise which captures other effects on the exchange rate. The Phillips
curve in eq. (3) is then obtained by substituting the deterministic version of the exchange
rate equation into the above Phillips curve. Hall (2000) also includes the real interest rate in
the Phillips curve. Furthermore, we follow Ball (1999) and take the real interest rate as the

3
Suppose the central bank has the following loss function
L(t) ≡ π 2 (t) + λµ(t)2 , (4)
where λ(> 0) denotes the weight of unemployment stabilization. We will drop
the notation “t” in variables in the remainder of the paper just for simplicity.
Suppose the goal of the monetary policy is to minimize the loss function with
an infinite horizon, the central bank’s problem turns out to be
Z ∞
M in e−ρt Ldt
r 0

subject to
µ̇ = αr, (5)
where ρ (0 < ρ < 1) is the discount factor and L defined by (4). The current-
value Hamiltonian of the above problem reads as
Hc = [βr + θµ]2 + λµ2 + γαr, (6)
where γ is a costate variable. The optimal conditions for this problem turn out
to be
∂Hc
=0, (7)
∂r
∂Hc
γ̇ = − + ργ, (8)
∂µ
with the following transversality condition
lim γe−ρt = 0. (9)
t→∞

From (7) we get the optimal monetary policy

αγ θ
r=− 2
− µ, (10)
2β β
from which we know
2β
γ̇ = −
(β ṙ + θαr). (11)
α
Some rearrangement of (8), (10) and (11) gives us the following dynamic system
of r and µ:
θρ α
ṙ = ρr + [ + 2 (θ2 + λ2 )]µ, (12)
β β
µ̇ = αr. (13)
policy instrument. Although McCallum casts doubt on Ball (1999) for the use of the real rate
as a policy instrument, Ball (1999, footnote 2), however, correctly argues that policy makers
can move the real interest rate to their desired level by setting the nominal rate equal to the
desired real rate plus the inflation rate. Ball (1999) also considers the exchange-rate effect in
the IS equation. Given the exchange rate equation above, eq. (1) can also be considered as
the IS equation in the open economy.

4
Setting ṙ = µ̇ = 0, we obtain the unique equilibrium

r∗ = 0, µ∗ = 0 (u∗ = un ).

The Jacobian matrix of the dynamic system evaluated at the equilibrium is

ρ θρ α 2 2
 
J= β + β 2 (θ + λ ) .
α 0

If x1 and x2 are two characteristic roots of J, it is obvious that

x1 + x2 = ρ > 0,
θρ α
x1 x2 = −α[ + 2 (θ2 + λ2 )] < 0.
β β
We find that x1 and x2 are both real with opposite signs. This indicates that
the unique equilibrium (0, 0) is a saddle point.

Above we have explored monetary policy in a traditional model with a linear

Phillips curve without expectation considered. Next we explore monetary policy
with expectation in the Phillips curve. The expectation-augmented Phillips
curve is written as
π = −βr − θµ + πe , (14)
with πe denoting the expectation of the inflation rate. Following the traditional
literature we assume that the expectation evolves in an adaptive way2

π̇e = κ(π − πe ), κ > 0. (15)

The loss function L now looks like

L = [πe − βr − θµ]2 + λµ2 . (16)

After some rearrangement the problem of a central bank turns out to be

Z ∞
M in e−ρt Ldt
r 0

subject to

π̇e = −κ[βr + θµ], (17)

µ̇ = αr. (18)

The current-value Hamiltonian of this problem reads

Hc = [πe − βr − θµ]2 + λµ2 − γ1 κ[βr + θµ] + γ2 αr, (19)

2 We follow here Turnovsky (1981) and use an adaptive expectation dynamics. This is in

line with Rudebusch and Svensson (1999) in the sense that only past information is used in
the inflation equation.

5
where γ1 and γ2 are costate variables. The optimal conditions turn out to be
∂Hc
= 0, (20)
∂r
∂Hc
γ̇1 = − + ργ1 , (21)
∂πe
∂Hc
γ̇2 = − + ργ2 , (22)
∂µ
lim γ1 e−ρt = 0, (23)
t→∞
lim γ2 e−ρt = 0. (24)
t→∞

From (20) we know

1 κγ1 αγ2
r= [πe − θµ + − ]. (25)
β 2 2β
After substituting (25) into (17), (18), (21) and (22), we obtain the following
dynamic system
κ α
π̇e = − κ(πe + γ1 − γ2 ), (26)
2 2β
α κ α
µ̇ = [πe − θµ + γ1 − γ2 ], (27)
β 2 2β
α
γ̇1 =(ρ + κ)γ1 − γ2 , (28)
β
αθ
γ̇2 =(ρ + )γ2 − 2λ2 µ. (29)
β
Let π̇e = µ̇ = γ̇1 = γ̇2 = 0, we obtain the unique equilibrium:
πe∗ = 0, µ∗ = 0 (u∗ = un ), γ1∗ = 0, γ2∗ = 0,
and therefore r ∗ = 0 and π ∗ = 0. In order to consider the out-of-steady-state
dynamics around the equilibrium, we compute the Jacobian matrix of the system
(26)-(29) evaluated at the equilibrium:
 2 
−κ 0 − κ2 ακ
2β
 α α2 
− αθ ακ
2β − 2β 2 
J = β β .

 0 0 ρ+κ −α β

2 αθ
0 −2λ 0 ρ+ β
The characteristic equation of J can be written as
4(x) ≡ |xI − J| = x4 + b1 x3 + b2 x2 + b3 x + b4 = 0. (30)
The Routh-Hurwitz sufficient and necessary condition for local stability in this
case is3
bi > 0 (i = 1, 2, 3, 4), b1 b2 b3 − b21 b4 − b23 > 0.
3 The reader is referred to some advanced textbooks or papers on dynamic systems, Yoshida

and Asada (2001) for example, for the computation of bi .

6
But it is obvious that
b1 = −2ρ < 0,
therefore we conclude that the equilibrium is unstable. This is consistent with
the results from the case without expectation. For example, if we take α = 0.6,
κ = 0.9, β = 0.6, θ = 0.5, ρ = 0.1 and λ = 0.5, the eigenvalues of J turn out to
be
x1 = −1.060, x2 = −0.421, x3 = 1.160, x4 = 0.521.
This indicates that the unique equilibrium is unstable, since two characteristic
roots are positive.

3 Monetary Policy with Nonlinear Phillips Curve

In the previous section we have explored monetary policy in a traditional model
with a linear Phillips curve. There exists a unique saddle point equilibrium, no
matter whether the expectation is taken into account or not.
Recently most of the literature, however, casts doubt on the linearity of the
Phillips curve. Dupasquier and Ricketts (1998a), for example, survey several
models of the nonlinearity in the Phillips curve. The five models surveyed are
the capacity constraint model, the mis-perception or signal extraction model,
the costly adjustment model, the downward nominal wage rigidity model and
the monopolistically competitive model. These models explain the nonlinearity
of the Phillips curve from various perspectives. As also mentioned by Akerlof
(2002), the nonlinearity of the Phillips curve has been an important issue of
macroeconomics, since it has crucial implications for the monetary authorities.
Many empirical studies have been undertaken to explore the nonlinearity of
the Phillips curve. Dupasquier and Ricketts (1998a), for example, explore the
nonlinearity in the Phillips curve for Canada and the U.S. and conclude that
there is stronger evidence of a nonlinearity for the U.S. than for Canada. Aguiar
and Martins (2002), however, test three kinds of nonlinearities (quadratic, hy-
perbole and exponential) in the Phillips curve and Okun’s law with the aggregate
Euro-area macroeconomic data and find that the Phillips curve turns out to be
linear, but Okun’s law is nonlinear.
Other studies on the nonlinearity of the Phillips curve include Gómez and
Julio (2000), Clements and Sensier (2002), Dupasquier and Ricketts (1998b),
Chadha, Masson and Meredith (1992), Laxton, Meredith and Rose (1995) and
Bean (2000). Monetary policy with a nonlinear Phillips curve has also been
explored, see Schaling (1999), Tambakis (1998) Flaschel, Gong and Semmler
(2001), and Nobay and Peel (2000) for example.
The problem is, if the Phillips curve is nonlinear, what does it look like?
Some researchers, Filardo (1998) for instance, have explored this problem both
theoretically and empirically. One possible shape of the nonlinear Phillips curve
is convex. This possibility has been explored by Clark, Laxton and Rose (1996),
Schaling (1999), Laxton, Rose and Tambakis (1998), Tambakis (1998), Bean
(2000), Zhang and Semmler (2003) and so on. A convex Phillips curve is shown

7
 Inflation Inflation

Output Gap

Output Gap

B
A
Inflation

Convex

Output Gap

Concave

Figure 1: Possible Shapes of Nonlinear Phillips Curve

in Figure 1A. Filardo (1998) describes the property of a convex Phillips curve
as follows:
“...which (the convex Phillips curve) graphically translates into an
upward sloping curve that steepens as output rises relative to trend.
Intuitively, the steepening slope indicates increased sensitivity of in-
flation to the economy’s strength. As the slope of the convex curve
steepens, inflation becomes more sensitive because a given change in
inflation requires a progressively smaller output adjustment. ... The
convex Phillips curve is consistent with an economy subject to capac-
ity constraints. ... As the economy becomes stronger and capacity
constraints increasingly restrict firms’ ability to expand output, an
increase in demand is more likely to show up as higher inflation than
as higher output.” (Filardo , 1998)

Another possible shape of the nonlinear Phillips curve is concave. This

possibility has been proposed, for example, by Stiglitz (1997). A concave Phillips
curve is shown in Figure 1B. The property of a concave Phillips curve can be
described as follows

“... which (the concave Phillips curve) graphically translates into

an upward sloping curve that flattens as output rises relative to
trend. Intuitively, the concave curve’s flattening slope reflects the
declining sensitivity of inflation to the strength of the economy. ...

8
 Theoretically, a concave Phillips curve is consistent with an economy
where firms are not purely competitive. If firms have some pricing
power and thus the ability and desire to influence their market share,
they will be more reluctant to raise prices than to lower them.”
(Filardo, 1998)

Eisner (1997) also reports results that are consistent with a concave Phillips
curve. Filardo (1998) further claims that the implications for the output cost
of fighting inflation may be different for different shapes of the Phillips curve.

“The concave Phillips curve implies that the cost of fighting infla-
tion rises with the strength of the economy because as the economy
strengthens its slope flattens. In contrast, the convex Phillips curve
implies that the costs of fighting inflation falls with the strength of
the economy because its slope steepens.” (Filardo, 1998)

On the basis of these literature, Filardo (1998) explores evidence of a third

shape of the nonlinear Phillips curve and proposes that the nonlinear Phillips
curve is not purely convex or concave, but a combination of both, namely,
convex-concave. He sets up a model as follows

πt = πte + βw yw,t−1 + βb yb,t−1 + βs ys,t−1 + εt , (31)

where πt and πte are actual and expected inflation rates respectively. yt is the
output gap and εt is a supply shock. w, b and s stand for “weak”, “balanced”
and “strong” respectively. The slope coefficients on the output gap (β w , βb and
βs ) measure the sensitivity of inflation to economic activity in the weak, balanced
and overheated times. In the case of a linear Phillips curve, βw = βb = βs . With
the U.S. data from 1959-97, Filardo (1998) finds that βw = 0.2, βb = −0.02 and
βs = 0.49. Therefore, he obtains a convex-concave Phillips curve as shown in
Figure 1C. This nonlinear Phillips curve implies that when the output gap is
positive, the Phillips curve is convex and when the output gap is negative, the
Phillips curve is concave. Filardo (1998) further states that the researchers who
have proposed a convex or concave Phillips curve have considered only one case
and overlooked the other. The policy implication of a convex-concave Phillips
curve is of course different from that of a convex or concave one.
Based on the literature of nonlinear Phillips curves, this section is devoted
to the optimal monetary policy with nonlinear Phillips curves. We will explore
monetary policy in two cases: A convex Phillips curve and a convex-concave
one.4 The expectation in the Phillips curve is not taken into account in this
section, since the results in the previous section are similar no matter whether
the expectation is considered or not. Moreover, the model without expectation
is easier to study, since the dimension of the problem is then lower.
4 Note that we use the unemployment rate instead of the output in the Philips curve.
According to Okun’s law, there exists a negative correlation between the unemployment and
output, therefore, the convexity or concavity mentioned above will not be changed.

9
3.1 A Convex Phillips Curve
Some researchers, Schaling (1999) for example, assume that a convex Phillips
curve can be defined through the following function
φy
f (y) = , f 0 > 0, f 00 > 0, φ > 0, 1 > ϕ ≥ 0, (32)
1 − φϕy

where y is the output gap and the parameter ϕ indexes the curvature of f (·).
In case ϕ = 0, f (·) becomes linear. In line with Okun’s law, we just assume
y = −µ and then obtain the following convex Phillips curve5

π = −βr − θΠ(µ), (33)

with Π(·) defined as

φµ
Π(µ) = . (34)
1 + φϕµ
In the research below we assume θ = 1 in the Phillips curve and α = 1 in (1)
just to simplify the analysis. With the above Phillips curve the current-value
Hamiltonian of the optimal control now looks like
φµ
Hc = [βr + ]2 + λµ2 + γr, (35)
1 + φϕµ
from which we obtain the optimal monetary policy
φµ γ
r=− − 2 (36)
β(1 + φϕµ) 2β

and then by employing (8) and (36), we have the following dynamic system

1 φ2 βφρ
ṙ = [ µ + λµ + µ] + ρr, (37)
β 2 Ω3 Ω
µ̇ = r, (38)

with Ω = 1 + φϕµ. Setting ṙ = µ̇ = 0, we know the equilibrium of r is

r∗ = 0

and the equilibria of µ can then be solved from the equation

Ω3 λµ + φβρΩ2 µ + φ2 µ = 0. (39)

It is obvious that µ = 0, therefore u∗ = un is a solution of (39). As a result,

u∗ = un and r ∗ = 0 is an equilibrium of the system (37)-(38). This is consistent
5 In Okun’s law (µ = −my), the m is usually smaller than one. We take m = 1 just for

simplicity since the result below will not be changed if we assume m < 1.

10
with the unique equilibrium with the traditional linear Phillips curve analyzed
in the previous section. But in case µ 6= 0, (39) becomes

Ω3 λ + φβρΩ2 + φ2 = 0, (40)

which has three solutions. Note that in order for the condition f 00 (·) > 0 in (32)
1
to be satisfied, µ must be larger than − φϕ . We call this the Convex Condition.
The Convex Condition implies Ω > 0. But it is obvious that (40) has no
solutions satisfying this condition and therefore (0, 0) is the only equilibrium
of the above dynamic system. This is consistent with the results of the model
analyzed in the previous section.
Next we explore the out-of-steady-state equilibrium dynamics around the
equilibrium. The Jacobian matrix of the system (37)-(38) evaluated at the
equilibrium is
ρ β12 (φ2 + λ + βφρ)
 
J= .
1 0
It is obvious that the characteristic roots x1 and x2 of J satisfy

x1 + x2 = ρ > 0,
1
x1 x2 = − 2 (φ2 + λ + βφρ) < 0.
β
We find that x1 and x2 are both real with opposite signs, this implies that the
equilibrium is a saddle point.

3.2 A Convex-Concave Phillips Curve

We have explored monetary policy with a convex Phillips curve and find that
there exists a unique equilibrium. As stated before, Filardo (1998) finds that
the Phillips curve is convex-concave, depending on whether the output gap is
positive or negative. A graph of such a curve is shown in Figure 1C. In this
section we will explore monetary policy with such a convex-concave Phillips
curve. The Phillips curve in 1C has breakpoints and moreover, it is assumed
that the derivatives of the curve in the concave and convex zones are constant.
This is in fact not consistent with the cases of the two graphs in 1A and B,
where the derivatives of the curves are dependent on the sizes of the output
gap. Filardo (1998) obtains such a non-smooth graph from the estimation of
a simple model, in which he assumes the parameters βw , βb and βs in (31) to
be constant. But in reality the three parameters can be continuously state-
dependent. If the parameters are continuously state-dependent, the Phillips
curve is then smooth and convex-concave. In the research below we assume
that the Phillips curve is smooth and convex-concave. In order to consider this
possibility we have to design a function which is convex when the output gap
is positive and concave when the output gap is negative, and moreover, the
absolute values of the derivatives of the function with respect to the output gap
are asymmetric around zero. That is, the graph is steeper when the output gap

11
 Inflation

0 Output Gap

Figure 2: A Convex-Concave Phillips Curve

is positive than when the output gap is negative. Fortunately we can design a
function with such properties.
Let y denote the output gap, we define the following function f (y)

f (y) = δ(eay − ay − 1)sgn(y), δ > 0, a > 0, (41)

with (
1, if y ≥ 0;
sgn(y) =
−1, if y < 0.
δ(eay − ay − 1) is the so-called LINEX function proposed by Varian (1975) and
applied, for example, by Nobay and Peel (1998) and Semmler and Zhang (2002).
The LINEX function is nonnegative and asymmetric around zero and, moreover,
the parameter a determines the extent of asymmetry.6 The parameter δ scales
the function.
The graph of f (y) is shown in Figure 2. From this figure we see f (y) has the
properties discussed above. Let y = −µ, we obtain the following convex-concave
Phillips curve

π = −βr − θ[e−aµ + aµ − 1]sgn(µ), a > 0, (42)

where sgn(µ) equals 1 when µ ≥ 0 and −1 when µ < 0. With the Phillips curve
defined above the current-value Hamiltonian of the optimal control now looks
like
Hc = {βr + θ[e−aµ + aµ − 1]sgn(µ)}2 + λµ2 + γr, (43)
from which we obtain the following optimal monetary policy
θ γ
r = − [e−aµ + aµ − 1]sgn(µ) − 2 . (44)
β 2β
6a can be either positive or negative, but in our model it is positive.

12
 We have above derived optimal monetary policies from models with differ-
ent Phillips curves. The optimal monetary policies without expectation are
shown in (10), (36) and (44) respectively. The differences between these policies
are obvious. Exactly speaking, when the Phillips curve is linear, the optimal
monetary policy is also a linear function of the unemployment gap. When the
Phillips curve is convex, the optimal monetary policy is a convex function of
the unemployment gap. And finally, when the Phillips curve is convex-concave,
the optimal monetary policy is a convex-concave function of the unemployment
gap. Therefore, the optimal monetary reaction function of the central bank is
dependent upon the shape of the Phillips curve.

Dynamics

Next we explore the dynamics of the model. By employing (8) and (44) we
obtain the following dynamic system
1
ṙ = {aθ2 (1 − e−aµ )(aµ + e−aµ − 1)
β2
+ βθρ(e−aµ + aµ − 1)sgn(µ) + λµ} + ρr; (45)
µ̇ = r. (46)
Setting µ̇ = 0 we get the equilibrium of r, namely r ∗ = 0. With r ∗ = 0 and
setting ṙ = 0, the equilibrium of µ can be solved from the following equation
aθ2 (1 − e−aµ )(aµ + e−aµ − 1) + βθρ(e−aµ + aµ − 1)sgn(µ) + λµ = 0. (47)
µ = 0, namely u = un is the only solution of the above equation. This implies
that the dynamic system has a unique equilibrium r ∗ = 0 and µ∗ = 0. This is
consistent with the results in the previous sections. The Jacobian matrix of the
system (45)-(46) evaluated at the equilibrium is

ρ βλ2
 
J= .
1 0
It is obvious that the equilibrium is a saddle point since the characteristic roots
of J are both real with oppositive signs.

In this section we have explored monetary policy with two nonlinear Phillips
curves. There exists a unique equilibrium (a saddle point), no matter whether
the Phillips curve is convex or convex-concave. This is consistent with the
results from a model with a linear Phillips curve. But the optimal monetary
policy reaction function changes with the shapes of the Phillips curve.

4 Monetary Policy with Endogenous NAIRU

In the previous sections we have explored monetary policy with linear and non-
linear Phillips curves with a constant NAIRU (like an exogenous variable). The

13
difference of the results lies in the fact that the monetary policy functions change
with the shapes of the Phillips curve. The similarity, however, is that there ex-
ists a unique equilibrium (a saddle point) despite of the shapes of the Phillips
curve.
According to Friedman (1968) and Phelps (1968), monetary policy has no ef-
fects on the unemployment rate in the long run because of the so-called money
neutrality. This implies a vertical long-run Phillips curve. The recent litera-
ture, however, has questioned this assumption. Stiglitz (1997), for example,
surveys some factors that may lead to the movement of the NAIRU. Graham
and Snower (2002) derive a microfounded long-run downward-sloping Phillips
curve and show that a permanent increase in money growth incurs a permanent
increase in the inflation rate and a permanent decrease in the unemployment
level. Empirically he shows that a 1% increase in the money growth rate can
induce a long-run reduction in the unemployment from 15% to 0.5% below its
steady-state level. They further claim that the effects can be large and have a
long half-life in the short and medium run.
Karanassou, Sala and Snower (2002) show that the changes in money growth
can have long-run effects on the unemployment as well as inflation even if there
are no money illusion and money neutrality. Akerlof, Dickens and Perry (2000)
argue that the long-run Phillips curve is not vertical but instead bowed-inward
and then forward-bending. Following Akerlof, Dickens and Perry (2000), Lund-
borg and Sacklén (2001), based on the Swedish data, show that there exists
a negatively sloped long-run Phillips curve. Some researchers further propose
that the NAIRU can be affected by the inflation rate and monetary policy.
Different from the traditional view of Friedman, Blanchard (2003) points
out that monetary policy can and does affect the natural rate of unemploy-
ment. Blanchard and Summers (1988) argue that anything (e.g. a sustained
increase in real interest rates) that increases the actual rate of unemployment
for sufficiently long is likely to raise the natural rate. Blanchard (2003), more-
over, explores several mechanisms in which the real interest rate may affect the
natural rate of unemployment. He points out, for instance, that the capital
accumulation mechanism plays an important role in accounting for the history
of unemployment in Europe over thirty years:
“Low real interest rates in the 1970s probably partly mitigated the
increase in labor costs on profit, limiting the decline in capital ac-
cumulation, and thus limiting the increase in the natural rate of
unemployment in the 1970s. High real interest rates in the 1980s
(an then again, as the result of the German monetary policy re-
sponse to German reunification, in the early 1990s) had the reverse
effect of leading to a larger increase in the natural rate of unem-
ployment during that period. an the decrease in real interest rates
since the mid-1990s is probably contributing to the slow decline in
unemployment in Europe.” (Blanchard, 2003)
Based on a microfounded model, Lengwiler (1998) finds that the NAIRU
may be nonconstant and can be influenced by the expected inflation: It can

14
be downward-sloping or upward-sloping with a vertical long-run Phillips curve
(a constant NAIRU) being only an exception. Perez (2000) explores why the
NAIRU changes over time and finds that the NAIRU may change with the
changes of productivity, minimum wage and so on. Tobin (1998) also points out
that the NAIRU may vary over time because of the change of the relationships
between unemployment, vacancies, and wage changes.
Although it is still questioned whether the NAIRU can be precisely esti-
mated, some economists have tried with different approaches. An earlier survey
on the estimation of the NAIRU is given by Staiger, Stock and Watson (1996).
Gordon (1997), for example, estimates the time-varying NAIRU with the U.S.
data with and without supply shocks. Apel and Jansson (1999) estimate the
potential output and NAIRU of Sweden with a system-based strategy.
Following Blanchard (2003), in this section we will explore optimal monetary
policy with an endogenous NAIRU which can be affected by the monetary policy.
Before exploring the optimal monetary policy with an endogenous NAIRU, we
estimate the time-varying NAIRU for several countries with the model of Gordon
(1997).

4.1 Estimates of Time-Varying NAIRU

Gordon (1997) estimates the time-varying NAIRU with the following state-space
model:

πt = a(L)πt−1 + b(L)(Ut − UtN ) + c(L)zt + et , (48)

UtN = N
Ut−1 + ηt , (49)

where πt is the inflation rate, Ut the actual unemployment rate and UtN the
NAIRU which follows a random walk path indicated by equation (49). zt is a
vector of supply shock variables, L is a polynomial in the lag operator, et is a
serially uncorrelated error term and ηt satisfies the Gaussian distribution with
mean zero and variance ση2 . Obviously, the variance of ηt plays an important
role in the estimation. If it is zero, then the NAIRU is constant and if it is
positive, the NAIRU experiences changes. If no constraints are imposed on σ η2
the NAIRU will jump up and down and soak up all the residual variation in
the inflation variation. This is a standard “stochastic time-varying parameter
regression model” that can be estimated using maximum likelihood methods
with the help of the Kalman filter.7
Gordon (1997) includes a zt to proxy supply shocks such as changes of rel-
ative prices of imports and the change in the relative price of food and energy.
If no supply shocks are taken into account, the NAIRU is referred to as “esti-
mated NAIRU without supply shocks”. Though there are no fixed rules on what
variables should be included as supply shocks, it seems more reasonable to take
supply shocks into account than not, since there are undoubtedly other vari-
ables than the unemployment rate that affect the inflation rate. In this section
7 The reader is referred to Hamilton (1994, Ch. 13) for the Kalman filter.

15
 Figure 3: Unemployment Rates of Germany, France, Italy and the U.K.

the supply shocks considered include mainly price changes of imports (im t ),
food (f oodt ), and fuel, electricity and water (f uelt ). As for which variables
should be adopted as supply shocks for the individual countries, we undertake
an OLS regression for equation (48) before we start the time-varying estimation,
assuming that the NAIRU is constant. In most cases we exclude the variables
whose t-statistics are insignificant. The data source is the International Statis-
tics Yearbook 2000. As mentioned above, the standard deviation of η t plays a
crucial role. Gordon (1997) assumes it to be 0.2 percent for the U.S. for the
period 1955-96. There is little theoretical background on how large ηt should
be, but since the NAIRU is usually supposed to be relatively smooth, we con-
strain the change of the NAIRU within 4 percent, which is also consistent with
Gordon (1997). Therefore we assume different values of ηt for different coun-
tries, depending on how large we expect the change of the NAIRU to be. The
unemployment rates of the four EU countries are presented in Figure 3. The
data used in this section are taken from International Statistics Yearbook.
As for Germany, the variance of ηt is assumed to be 7.5 × 10−6 and the price
changes of foods, imports, and fuel, electricity and water are taken as supply
shocks. The estimates are shown below with t-statistics in parentheses,

πt = 0.004 + 1.052 πt−1 − 0.256 πt−2 + 0.008 πt−3 + 0.013 f uelt−1

(0.056) (10.982) (3.009) (0.153) (1.105)

+ 0.061 f oodt−1 + 0.006 imt−1 − 0.042 (Ut − UtN ) + et ,

(1.994) (0.627) (0.993)

where f uelt indicates the price change of fuel, electricity and water, f oodt the
price change of food and imt the price change of imports. The estimate of the
standard deviation of et is 0.006 with t-statistic being 8.506. The time-varying
NAIRU of Germany is shown in Figure 4A.
As for France, only one lag of the inflation rate is included in the regression,
since the coefficient of the unemployment rate gap tends to zero when more lags

16
Figure 4: Time-Varying NAIRU of Germany, France, the U.K., Italy and the
U.S.

of the inflation are included. The price changes of food and intermediate goods
are taken as supply shocks. Three lags of the price changes of intermediate
goods are included to smooth the NAIRU. The result reads as

πt = 0.004 + 0.989 πt−1 − 0.085 f oodt−1 + 0.132 int

(0.312) (27.967) (2.058) (2.876)

− 0.090 int−1 + 0.029 int−2 − 0.054 (Ut − UtN ) + et ,

(1.313) (0.772) (1.889)

where int denotes the price change of intermediate goods. The estimate of the
standard deviation of et is 0.005 with t-statistic being 8.808, and the variance
of ηt is predetermined as 1.3 × 10−5 . The estimate of the NAIRU of France is
presented in Figure 4B.
For the same reason as for France, one lag of the inflation rate is included
in the regression for the U.K.. The estimation reads

πt = 0.005 + 0.818 πt−1 + 0.130 f oodt−1

(0.066) (11.063) (2.768)

+ 0.017 f uelt−1 − 0.072 (Ut − UtN ) + et .

(0.691) (0.300)

The estimate of the standard deviation of et is 0.013 with t-statistic being 8.764
and the variance of ηt is predetermined as 1.4 × 10−5 . The estimate of the
NAIRU of the U.K. is shown in Figure 4C.

17
 For Italy it seems difficult to get a smooth estimate for the NAIRU if we
include only price changes of food, fuel, electricity and water and imports as
supply shocks. The main reason seems to be that the inflation rate experi-
enced drastic changes and therefore exerts much influence on the estimate of
the NAIRU. Therefore, we try to smooth the estimate of the NAIRU by includ-
ing the short-term nominal interest rate (nrt ) into the regression, which makes
the NAIRU more consistent with the actual unemployment rate. The result is

πt = 0.0035 + 1.594 πt−1 − 0.832 πt−2

(0.001) (6.623) (4.961)

− 0.247 f oodt−1 + 0.322 f oodt−2 − 0.017 f uelt−1

(1.722) (2.255) (1.398)

+ 0.030 f uelt−2 + 0.181 nrt − 0.304 (Ut − UtN ) + et .

(1.599) (1.532) (0.902)

The estimate of the standard deviation of et is 0.010 with t-statistic being 8.982,
and the variance of ηt is assumed to be 2.6 × 10−6 . The time-varying NAIRU
of Italy is shown in Figure 4D.
We also undertake the estimation of the NAIRU for the U.S. with and with-
out “supply shocks” for 1962.3-1999.4. In the estimation without supply shocks,
only four lags of the inflation rate and unemployment gap are included in the
regression and the result is

πt = 0.002 + 1.321 πt−1 − 0.243 πt−2

(0.002) (16.045) (2.199)

− 0.121 πt−3 + 0.015 πt−4 − 0.065 (Ut − UtN ) + et .

(61.971) (0.350) (8.864)

The estimate of the standard deviation of et is 0.004 with t-statistic being 15.651
and the variance of ηt is predetermined as 4.5 × 10−6 . The NAIRU of the U.S.
without supply shocks is presented in Figure 4E, very similar to the result of
Gordon (1997). Considering supply shocks which include price changes in food,
energy and imports, we have the following result for the U.S.:

πt = 0.002 + 0.957 πt−1 − 0.151 πt−2

(0.091) (10.592) (1.350)

− 0.070 πt−3 + 0.120 πt−4 + 0.062 f oodt

(0.647) (1.822) (4.547)

+ 0.007 f uelt−1 + 0.025 imt−1 − 0.060 (Ut − UtN ) + et .

(0.468) (3.808) (3.036)

The estimate of the standard deviation of et is 0.003 with t-statistic being 16.217
and the variance of ηt is predetermined as 4 × 10−6 . The time-varying NAIRU
with supply shocks is shown in Figure 4F.

4.2 Monetary Policy with Endogenous NAIRU

Above we have estimated the time-varying NAIRU with a model proposed by
Gordon (1997). As stated before, Blanchard (2003) argues that monetary policy

18
 Country
Parameter Germany France U.K. U.S.
τ0 0.063 0.051 0.051 0.040
(130.339) (24.263) (57.513) (29.076)
τ1 0.064 0.085 0.128 0.322
(4.801) (1.986) (6.222) (8.809)
2
R 0.250 0.056 0.356 0.526
Correlation 0.437 0.225 0.337 0.387

Table 1: Regression Results of eq. (50) and Correlation Coefficients of r̄ (Com-

puted with the Ex Post Real Rate) and the NAIRU

can and does affect the natural rate of unemployment. Therefore, the problem to
tackle next is how monetary policy affects the NAIRU. Taking Europe as exam-
ple, Blanchard (2003) argues that a tight monetary policy can raise the NAIRU
and an expansionary policy may reduce the NAIRU. There seems to exist a
positive correlation between the NAIRU and the real interest rate. Therefore
below we will analyze the relationship between the NAIRU and the real interest
rate. In Table 1 we show the estimation of the following equation from 1982 to
the end of the 1990s (T-statistics in parentheses):

unt = τ0 + τ1 r̄, (50)

where r̄ denotes the 8-quarter (backward) average of the real interest rate. 8
The real interest rate is defined as the short-term nominal rate minus the actual
inflation rate. The reason that we use the 8-quarter backward average of the
real interest rate for estimation is that some researchers argue that the NAIRU
is usually affected by the lags of the real rate. The reason that the regression
is undertaken only for the period after 1982 is that in the 1970s and at the
beginning of the 1980s these countries experienced large fluctuations in the
inflation and therefore the real rate also experienced large changes. In Table 1
we find that τ1 is significant enough. We also show the correlation coefficients
of the NAIRU and r̄ for the same period in Table 1.
The real rate above is defined as the gap between the nominal rate and the
actual inflation. The real rate defined in this way is usually referred to as the ex
post real rate. According to the Fisher equation, however, the real rate should
be defined as the nominal rate (nrt ) minus the expected inflation, that is

rt = nrt − πt−1|t , (51)

where πt−1|t denotes the inflation rate from t − 1 to t expected by the market
at time t − 1. The real rate defined above is usually called the ex ante real
rate. How to measure πt−1|t is a problem. Blanchard and Summers (1984),
8 The interest rates of Germany, France, the U.K. and the U.S. are the German call money

rate, 3-month interbank rate, 3-month treasury bill rate and the Federal funds rate respec-
tively. Data source: International Statistics Yearbook.

19
 Country
Parameter Germany France U.K. U.S.
τ0 0.063 0.050 0.051 0.040
(162.748) (22.663) (59.251) (31.278)
τ1 0.073 0.097 0.131 0.329
(6.817) (2.117) (6.337) (9.579)
2
R 0.410 0.065 0.371 0.574
Correlation 0.640 0.256 0.609 0.758

Table 2: Regression Results of eq. (50) and Correlation Coefficients of r̄ (Com-

puted with the Ex Ante Real Rate) and the NAIRU

for example, measure the expected inflation by an autoregressive process of the

inflation rate. Below we will measure the expected inflation by assuming that
the economic agents forecast the inflation by learning through the recursive least
squares. Namely, we assume

πt−1|t = c0t−1 + c1t−1 πt−1 + c2t−1 yt−1 , (52)

where yt denotes the output gap. This equation implies that the agents pre-
dict the inflation next period by adjusting the coefficients c0 , c1 and c2 period
by period. Following Sargent (1999), Orphanides and Williams (2002), Evans
and Honkapohja (2001) and Zhang and Semmler (2003) we assume that the
coefficients evolve in the manner of the recursive least squares
0
Ct =Ct−1 + t−1 Vt−1 Xt (πt − Xt Ct−1 )
0
Vt =Vt−1 + t−1 (Xt Xt − Vt−1 )

with Ct = (c0t c1t c2t )0 and Xt = (1 πt−1 yt−1 )0 . Vt is the moment matrix of Xt .9
With the output gap measured by the percentage deviation of the industrial
production index (IPI) from its HP filtered trend and the πt−1|t computed by
the equations above, we show the estimation results for eq. (50) with the ex ante
real rate and the correlation coefficients between the NAIRU and r̄ computed
with the ex ante real rate in Table 2.10 The results in Table 2 are not essentially
different from those in Table 1 except that the correlation coefficients between
the NAIRU and r̄ in Table 2 are larger than those in Table 1.
The ex ante- and ex post real rates of the U.S. from 1981:1 to 1999:4 are
shown in Figure 5. It is obvious that the two rates are not significantly different.
9 The reader can refer to Harvey (1981, ch. 7) and Sargent (1999) for the recursive least

squares.
10 The IPI has also been used by Clarida, Gali and Gertler (1998) to measure the output for

Germany, France, the U.S., the U.K., Japan and Italy. As surveyed by Orphanides and van
Norden (2002), there are many methods to measure the output gap. We find that filtering
the IPI using the Band-Pass filter developed by Baxter and King (1995) leaves the measure
of the output gap essentially unchanged from the measure with the HP-filter. The Band-Pass
filter has also been used by Sargent (1999).

20
 Figure 5: The Ex Ante- and Ex Post Real Rates of the U.S.

The empirical evidence above indicates that there exists a positive relation-
ship between the real rate and the NAIRU. The question then is whether the
path of the NAIRU is affected by r in a linear manner similar to that of the
actual unemployment given by (2) or in a nonlinear way. In the research below
we assume that the the path of the NAIRU can be affected by the monetary
policy in a nonlinear manner because of adjustment costs in the investment.
Exactly speaking, we assume

u̇n = g(r) = a tanh(br). (53)

We know that tanh(x) is a function with upper and lower bounds 1 and −1
respectively. It equals zero when x = 0. The parameter a scales the function
and b determines the slope of g(r) around 0, the larger b is, the steeper the
function is around 0. In case b is relatively small, g(r) tends to be linear. Two
simulations of g(r) with a = 0.6 and different b (1.2 and 0.2) are shown in Figure
6. This model implies that the change of the NAIRU increases with the increase
of r and vice versa, but not proportionally. It increases or decreases faster when
r is close to zero than when r is far from zero. The change of the NAIRU stops
to increase or decrease in case r tends to positive or negative infinity. The reason
for the nonproportional change of the NAIRU to the monetary policy is that
there exist adjustment costs. Since the 1960s economists have been studying the
implication of adjustment costs on the dynamic investment. An earlier study
can be found in Jorgenson (1963). He derives the optimal path of the capital
stock by assuming an exogenously given output. Lucas (1967) suggests that the
adjustment costs can be thought of as a sum of purchase costs and installation
costs which are internal to the firm. Gould (1968) designs a quadratic function
of adjustment costs and explores their effects on investment of the firm. Some
other research on adjustment costs and their effects on the investment can be
found in Eisner and Stroz (1963), Treadway (1969), and Feichtinger, Hartl, Kort
and Wirl (2001). Adjustment costs prevent firms from increasing or decreasing
the investment proportionally with the decrease or increase of the interest rate.

21
 Figure 6: g(r) with Different b

As a result, the aggregate demand does not decrease or increase proportionally

with the increase or decrease of the interest rate and therefore the change of
the NAIRU does not increase or decrease proportionally with the interest rate
either.11
We do not consider the effects of the adjustment costs on the actual unem-
ployment, because there may exist idle capacity and the actual unemployment
is supposed to be more affected by the capacity utilization than the NAIRU,
which can be considered as the trend of the actual unemployment and may be
mainly affected by the investment behavior.
Assuming that the path of the NAIRU can be affected by r as shown in
(53), what is then the optimal monetary policy? Next we explore this problem
with the model of a linear Phillips curve as given in Section 2, since it is more
complicated to analyze the model with the nonlinear Phillips curve. With u̇ n =
g(r) eq.(2) changes to
µ̇ = u̇ − u̇n = αr − g(r), (54)
and the problem of the central bank with expectation taken into account in the
Phillips curve reads Z ∞
M in e−ρt Ldt, (55)
r 0
subject to

π̇e = −κ(βr + θµ), (56)

µ̇ = αr − g(r), (57)

with L given by (16). The current-value Hamiltonian of the problem above

11 Of course g(r) does not have to be symmetric around the origin (0, 0) and we can modify

g(r) by using some parameters so that it becomes asymmetric around (0, 0). We will not do
so below because the results are not essentially changed no matter whether g(r) is symmetric
or not around the origin.

22
reads
Hc = (πe − βr − θµ)2 + λµ2 − γ1 κ(βr + θµ) + γ2 [αr − g(r)]. (58)
From the first-order condition (20) we know the optimal monetary policy r is
the solution of the following equation
−2β(πe − βr − θµ) − γ1 κβ + γ2 [α − g 0 (r)] = 0. (59)
Let ř(Ω) denote the solution of r from the equation above, with Ω denoting the
set of parameters and the variables µ, πe , γ1 and γ2 . Following (21)–(24) and
(59) we obtain the following dynamic system
π̇e = − κ[βř + θµ], (60)
µ̇ =αř − g(ř), (61)
γ̇1 = − 2(πe − βř − θµ) + ργ1 , (62)
γ̇2 =2θ(πe − βř − θµ) − 2λµ + γ1 κθ + ργ2 , (63)
0 = − 2β(πe − βř − θµ) − γ1 κβ + γ2 [α − g 0 (ř)]. (64)
Setting π˙e = µ̇ = γ˙1 = γ˙2 = 0, we can compute the equilibria of the economy
as follows
0 = − κ[βř + θµ], (65)
0 =αř − g(ř), (66)
0 = − 2(πe − βř − θµ) + ργ1 , (67)
0 =2θ(πe − βř − θµ) − 2λµ + γ1 κθ + ργ2 , (68)
0 = − 2β(πe − βř − θµ) − γ1 κβ + γ2 [α − g 0 (ř)]. (69)
We will try some numerical solutions since it is difficult to get analytical solu-
tions of this system. Before trying numerical solutions we explore whether there
exist solutions of this system, starting with (66), since this equation contains
only one variable, ř.
Denote Θ(ř) = αř − g(ř) and let
Θ(ř) = 0. (70)
It is clear that ř = 0 is a solution of the above equation, but with some proper
values assigned to the parameters, eq. (70) can have other solutions, therefore
the dynamic system (60)–(63) may have multiple equilibria.12
In Figure 7 we show two simulations of Θ(ř) with different parameters. It is
clear that with a = 0.6, b = 0.2 and α = 0.5 Θ(ř) cuts the horizontal axis once,
but with a = 0.6, b = 1.2 and α = 0.5 Θ(ř) cuts the horizontal axis three times,
therefore there may exist two other equilibria in the dynamic system (60)–(63)
besides the one with r = 0.
12 As stated before, we do not consider the effects of adjustment costs on the path of the

actual unemployment. But even if r can affect the path of the actual unemployment in a
way similar to (53) instead of (2), the results of multiple equilibria should not be essentially
changed. If we assume, for example, u̇ = %(r) with %(r) defined similarly to g(r) with different
parameters. The two functions g(r) and %(r) can then cut each other once or more times,
corresponding to one or multiple equilibria of the optimal control problem.

23
 Figure 7: Θ(ř) with Different Sets of Parameters

In order to explore whether the multiple equilibria are robust to parameters

chosen, we will try numerical solutions of the system (65)–(69) with a different
set of parameters from those employed in Figure 7. Let a = 0.6, b = 6, α = 0.75,
κ = 0.9, β = 0.6, θ = 0.5, ρ = 0.1 and λ = 0.0075, the three sets of equilibria,
a, b and c are given as follows:13

ra∗ = 0.800, µ∗a = −0.960, πea

∗ ∗
= −0.0012, γ1a ∗
= −0.025, γ2a = −0.0199,
∗
rb∗ = −0.800, µ∗b = 0.960, πeb ∗
= 0.0012, γ1b ∗
= 0.025, γ2b = 0.0199,
and
rc∗ = 0, µ∗c = 0, πec
∗ ∗
= 0, γ1c ∗
= 0, γ2c = 0.
Substituting the πe∗ , µ∗ and r ∗ into eq. (14) we can observe at least three
equilibria values of π, πa∗ = −0.0012, πb∗ = 0.0012 and πc∗ = 0. Therefore we
have at least three steady states for inflation and unemployment, (−0.0012,
−0.960), (0.0012, 0.960) and (0, 0).14

Yet, we can hardly explore the stability of the equilibria because we can not
compute the Jacobian matrix without an explicit expression of ř. Therefore
we will numerically compute the optimal control problem (55)–(57) with the
algorithm developed by Grüne (1997) with the parameters given above. Grüne
(1997) uses the Bellman equation and dynamic programming to compute nu-
merically the optimal control problem with adaptive grids.15
The numerically obtained value function (VF) using the Grüne algorithm is
shown in Figure 8, in which we observe that the value function is not smooth
13 Because the highly nonlinear system (65)-(69) is solved numerically, there might exist
other equilibria which are not detected. Therefore the model may have more than three
equilibria.
14 As above mentioned, given the strong nonlinearities in our functions there might exist

other steady state equilibria as well.

15 For further details of this algorithm see Grüne and Semmler (2003).

24
 VF

Mu

Pi_e

Figure 8: Numerical Computation of the Value Function

at the bottom. The vector field of the state variables with πe on the horizontal
axis and µ on the vertical axis is shown in Figure 9. In Figure 9 we indeed can
observe three equilibria. As mentioned before, there may exist other equilibria
which are difficult to observe in the vector field. In Figure 10 we show the
vector field for the state variables with λ = 0.5, a = 0.3 and b = 2 and other
parameters unchanged and find that there exists a unique equilibrium (0, 0).
In Figure 11 we show the optimal trajectories of πe and µ with different initial
values. The equilibria (-0.0012, -0.960) and (0.0012 0.960) are stable. Note,
however, that in the middle there might exist other fixed points besides (0, 0).
Numerically we can observe that the point (0, 0) is unstable, but there seem to
be two limit cycles close to (0, 0) which can be detected with some initial values
of πe and µ close to (0, 0). In Figure 12 we show the optimal trajectories of πe
and µ with initial values (0.0001, 0.0001) and (−0.0002, −0.0011). We observe
indeed two limit cycles.

Based on some empirical evidence and assuming that the monetary policy
affects the path of the NAIRU, in this section we have explored monetary policy
with an endogenous NAIRU and find that there may exist multiple equilibria
for such type of models. As stated before, Blanchard (2003) remarks that the
monetary policy does and can affect the NAIRU. On the other hand, some re-
searchers, Gordon (1997) and Stiglitz (1997), for example, maintain that the
NAIRU has some feedback effects on the macroeconomic policy (monetary pol-
icy for instance) in the sense that it faces a moving target. Although we know
that there are other forces affecting the NAIRU, a model of monetary policy with
an exogenous NAIRU can probably not be considered a proper device to study
monetary policy effects. Future research is likely to explore more extensively
the way how monetary policy affects the NAIRU.

25
Figure 9: Vector Field with Multiple Equilibria

Figure 10: Vector Field with Unique Equilibrium

26
 Figure 11: Optimal Trajectories of πe and µ

Mu

Pi_e

Figure 12: Limit Cycles with Initial Values Close to (0,0)

27
5 Conclusion
This paper is concerned with the optimal monetary policy with nonlinear Phillips
curves and an endogenous NAIRU. Before exploring the optimal monetary pol-
icy with a nonlinear Phillips curve we have studied the optimal monetary policy
in a simple model with a linear Phillips curve and find that the optimal mone-
tary policy is a linear function of the unemployment gap and that there exists
a unique equilibrium, namely a saddle point. We have then explored the opti-
mal monetary policy with two different shapes of the nonlinear Phillips curve:
Convex and convex-concave. We find that the optimal monetary policy changes
with the shape of the Phillips curve, but there exists a unique equilibrium (a
saddle point) despite of the different nonlinearities of the Phillips curve. Based
on this result, we have then relaxed the traditional assumption of a constant and
exogenous NAIRU. Recent literature has also proposed that the NAIRU can be
influenced by the monetary policy. With some empirical evidence and assuming
that the NAIRU can besides other forces be influenced by the monetary policy
in a certain manner, we find that there may exist multiple steady state equilibria
in the economy, different from the results of models with a constant NAIRU.

28
References
[1] Aguiar, Alvaro, and Manuel M.F. Martins (2002), “Trend, cycle, and non-
linear trade-off in the Euro-area 1970-2001”, CEMPRE.

[2] Akerlof, George A. (2002), “Behavioral macroeconomics and macroeco-

nomic behavior,” American Economic Review, Vol. 92, pp411-33.

[3] Ball, Laurence (1999), “Policy rules for open economies,” in John B. Taylor,
ed., Monetary Policy Rules, The University of Chicago Press.

[4] Baxter, Marianne and Robert G. King (1995), “Measuring business cycles:
Approximate Band-Pass filters for economic time series”, NBER working
paper, No. 5022.

[5] Bean, Charles (2000), “The convex Phillips curve and macroeconomic pol-
icymaking under uncertainty”, manuscript, London School of Economics.

[6] Blanchard, Olivier, and Lawrence H. Summers (1984), “Perspectives on

high world real interest rates”, Brookings Papers on Economic Activity,
2:1984, pp273-324.

[7] Blanchard, Olivier (2003), “Monetary policy and unemployment”, remarks

at the conference “Monetary policy and the labor market”, in honor of
James Tobin, held at the New School University, NY.

[8] Chadha, B., P. Masson, and G. Meredith (1992), “Models of inflation and
the costs of disinflation”, IMF Staff Papers, Vol. 39, No. 2, pp395-431.

[9] Clarida, Richard, Jordi Gali and Mark Gertler (1998), “Monetary policy
rules in practice: Some international evidence”, European Economic Re-
view, Vol. 42, pp1033-1067.

[10] Clark, Peter B., Douglas M. Laxton, and David E. Rose (1996), “Asym-
metry in the U.S. output-inflation nexus: Issues and evidence”, IMF Staff
Papers, Vol. 43, No. 1, pp216-250.

[11] Dupasquier, Chantal, and Nicholas Ricketts (1998a), “Nonlinearities in the

output-inflation relationship”, manuscript.

[12] − − − (1998b), “Nonlinearities in the output-inflation relationship: Some

empirical results for Canada”, Bank of Canada working paper 98-14.
[13] Eisner, Robert (1997), “New view of the NAIRU”, in Paul Davidson and
Jan A. Kregel, eds., Improving the Global Economy: Keynesian and the
Growth in Output and Employment, Edward Elgar Publishing Cheltenham:
Uk and Lyme, U.S.

[14] Eisner, R., and R. Stroz (1963), “Determinants of business investment”,

Impacts of Monetary Policy, Prentice Hall.

29
[15] Evans, George W., and Seppo Honkapohja (2001), Learning and Expecta-
tions in Macroeconomics, The Princeton University Press.

[16] Feichtinger, G, F.H. Hartl, P Kort, and F. Wirl (2001), “The dynamics of
a simple relative adjustment-cost framework”, German Economic Review,
pp255-268.

[17] Filardo, Andrew J. (1998), “New evidence on the output cost of fighting
inflation”, Economic Review, third quarter, pp33-61.

[18] Flaschel, Peter, Gang Gong, and Willi Semmler (2001), “Nonlinear Phillips
curve and monetary policy in a Keynesian macroeconometric model”, work-
ing paper 18, CEM, University of Bielefeld.

[19] Friedman, M. (1968), “The role of monetary policy”, American Economic

Review, 58(1), pp267-291.

[20] Gómez, Javier, and Juan Manuel Julio (2000), “An estimation of the non-
linear Phillips curve in Colombia”, manuscript.

[21] Gordon, Robert J. (1997), “The time-varying NAIRU and its implications
for economic policy”, Journal of Economic Perspectives, Vol. 11, pp11-32.
[22] Gould, J.P. (1968), “Adjustment costs in the theory of investment of the
firm”, Review of Economic Studies, Vol. 35, pp47-55.

[23] Graham, Liam, and Dennis J. Snower (2002), “The return of the long-run
Phillips curve”, Institute for the Study of Labor (IZA), discussion paper,
No. 646.

[24] Grüne, Lars (1997), “An adaptive grid scheme for the discrete Hamilton-
Jacobi-Bellman equation”, Numerische Mathematik, Vol. 75, 319-337.

[25] Grüne, Lars, and Willi Semmler (2003), “Using dynamic programming with
adaptive grid scheme for optimal control problems in economics”, forthcom-
ing in Journal of Economics Dynamics and Control.

[26] Guender, Alfred V. (2001), “Inflation targeting in the open economy: which
rate of inflation to target?”, manuscript, Department of Economics, Uni-
versity of Canterbury.

[27] Hall, Robert E. (2000), “Monetary policy with changing financial and labor-
market fundamentals”, manuscript, Hoover Institution and Department of
Economics, Stanford University, and NBER.
[28] Hamilton, James D. (1994), Time Series Analysis, The Princeton Univer-
sity Press.

[29] Harvey, A.C. (1981), Time Series Models, Philip Allan Publishers Ltd.

30
[30] Jorgenson, S., and P.M. Kort (1963), “Dynamic investment policy with
installation experience effects”, Journal of Optimization Theory and Ap-
plications, Vol. 77, 421-438.
[31] Laxton , D., G. Meredith, and D. Rose (1995), “Asymmetric effects of
economic activity on inflation: Evidence and policy implications”, IMF
Staff Papers, vol. 42, No. 4.

[32] Lengwiler, Yvan (1998), “Certainty equivalence and the non-vertical long
run Phillips curve”, manuscript, Board of Governors of the Federal Reserve
System and Swiss Natioanl Bank.

[33] Lucas, R.E. (1967), “Adjustment costs and the theory of supply”, Journal
of Polotical Economy, Vol. 75, 321-334.

[34] Mankiw, N. Gregory (2001), “Sticky information versus sticky prices: A

prposal to replace the new Keynesian Phillips curve”, NBER working paper
8290.

[35] Nobay A., Robert, and David A. Peel (1998), “Optimal monetary policy
in a model of asymmetric central bank preferences”, Financial Markets
Group, LSE.

[36] − − − (2000), “Optimal monetary policy with a nonlinear Phillips curve”,

Economics Letters, Vol. 67, pp159-164.

[37] Orphanides, Athanasios and John C. Williams (2002), “Imperfect knowl-

edge, inflation expectations and monetary policy”, manuscript, November.

[38] Orphanides, Athanasios and Simon van Norden (2002), “The unreliabil-
ity of output gap estimation in real time”, The Review of Economics and
Statistics, 84(4), pp569-583.

[39] Perez, Stephen J. (2000), “Why does the NAIRU change across time?”,
manuscript, Department of economics, Washington State University.

[40] Phelps, E.S. (1968), “Money-wage dynamics and labor-market equilib-

rium”, Journal of Political Economy, No. 76, Part 2, pp678-711.

[41] Rudebusch, Glenn D. and Lars E. O. Svensson (1999), “Policy rules for
inflation targeting”, in John B. Taylor, ed., Monetary Policy Rules, The
University of Chicago Press.

[42] Sargent, Thomas J. (1999), The Conquest of American Inflation, The

Princeton University Press.

[43] Schaling, Eric (1999), “The nonlinear Phillips curve and inflation forecast
targeting”, Bank of England.

31
[44] Semmler, Willi, and Wenlang Zhang (2002), “Asset price bubbles and mon-
etary policy rules: A dynamic model and empirical evidence”, manuscript,
at www.wiwi.uni-bielefeld.de/∼Semmler/cem.
[45] Staiger, Douglous, James H.Stock, and Mark W. Watson (1996), “How pre-
cise are estimates of the natural rate of unemployment?”, NBER working
paper 5477.

[46] Stiglitz, Joseph (1997), “Reflections on the natural rate hypothesis”, Jour-
nal of Economic Perspectives, Vol. 11, pp3-10.

[47] Svensson, Lars E.O. (1997), “Optimal inflation targets, “conservative” cen-
tral banks, and linear inflation contracts”, American Economic Review, Vol.
87, pp98-114.

[48] Tambakis, Demosthenes N. (1998), “Monetary policy with a convex Phillips

curve and asymmetric loss”, working paper 98/21, IMF.

[49] Tobin, James (1998), “Supply constraints on employment and output:

NAIRU versus natural rate”, International Conference in memory of Fausto
Vicarelli, Rome, Nov. 21–23, 1998.

[50] Treadway, A.B. (1969), “On rational entrepreneurial behaviour and the
demand for investment”, Review of Economic Studies, Vol. 36, pp227-239.

[51] Turnovsky, S. (1981), “The optimal intertemporal choice of inflation and

unemployment: An analysis of the steady state and transitional dynamics”,
Journal of Economics Dynamics and Control, 3, pp357-384.

[52] Varian,H.R. (1975), A bayesian approach to real estate assessment, in Stud-

ies in Bayseian Econometrics and statistics in honour of Leonard J. Savage,
eds S.E. Fienberg and Arnold Zellner, Amsterdam: North-Holland, 1975,
pp195-208.

[53] Walsh, Carl E. (1999), “Monetary policy trade-offs in the open economy”,
University of California, Santa Cruz and Federal Reserve Bank of San Fran-
cisco.

[54] Yoshida, Hiroyuki, and Toichiro Asada (2001), “Dynamic analysis of policy
lag in a Keynes-Goodwin model: Stability, instability cycles and chaos”, at
www.wiwi.uni-bielefeld.de/∼Semmler/cem.

[55] Zhang, Wenlang, and Willi Semmler (2003), “Monetary policy under model
uncertainty: Empirical evidence, adaptive learning and robust control”,
manuscript, at www.wiwi.uni-bielefeld.de/∼Semmler/cem.

32