Professional Documents
Culture Documents
analysis. A key point of departure from and applied work, as we discuss later. 4
real business cycle theory (as we later There are, however, important strands
make clear) is the explicit incorporation of the literature that either reject the
of frictions such as nominal price rigidi- idea of nominal price rigidities (e.g.,
ties that are needed to make the frame- real business cycle theory) or focus on
work suitable for evaluation of monetary other types of nominal rigidities, such
policy. as frictions in money demand. 5 For this
This paper summarizes what we have reason, we append “New Keynesian
learned from this recent research on Perspective” to the title. In particular,
monetary policy. We review the prog- we wish to make clear that we adopt the
ress that has been made and also iden- Keynesian approach of stressing nomi-
tify the central questions that remain. nal price rigidities, but at the same time
To organize the discussion, we exposit base our analysis on frameworks that in-
the monetary policy design problem in a corporate the recent methodological ad-
simple theoretical model. We start with vances in macroeconomic modeling
a stripped-down baseline model in or- (hence the term “New”).
der to characterize a number of broad Section 2 lays out the formal policy
principles that underlie optimal policy problem. We describe the baseline
management. We then consider the im- theoretical model and the objectives of
plications of adding various real world policy. Because we are interested in
complications. Finally, we assess how characterizing policy rules in terms of
the predictions from theory square with primitive factors, the model we use
policy-making in practice. evolves from first principles. Though it
Throughout, we concentrate on ex- is quite simple, it nonetheless contains
positing results that are robust across a the main ingredients of descriptively
wide variety of macroeconomic frame- richer frameworks that are used for pol-
works. As Ben McCallum (1997b) em- icy analysis. Within the model, as in
phasizes, the key stumbling block for practice (we argue), the instrument of
policy formation is limited knowledge monetary policy is a short-term interest
of the way the macroeconomy works. rate. The policy design problem then is
Results that are highly model-specific to characterize how the interest rate
are of limited use. This literature, how- should adjust to the current state of the
ever, contains a number of useful prin- economy.
ciples about optimal policy that are rea- An important complication is that pri-
sonably general in applicability. In this vate sector behavior depends on the ex-
respect there is a “science of monetary pected course of monetary policy, as
policy,” as Alan Blinder suggests in the well as on current policy. The credibil-
quote above. We provide support for ity of monetary policy thus becomes
this contention in the pages that follow. relevant, as a considerable contemporary
At the same time, we should make literature has emphasized. 6 At issue is
clear that the approach we take is based
on the idea that temporary nominal 4 See, for example, the survey by Goodfriend
whether there may be gains from en- gain from commitment then is to elimi-
hancing credibility either by formal nate this inflationary bias. How realistic
commitment to a policy rule or by intro- it is to presume that a perceptive cen-
ducing some kind of institutional ar- tral bank will try to inadvisedly reap
rangement that achieves roughly the short-term gains from pushing output
same end. We address the issue by ex- above its natural level is a matter of re-
amining optimal policy for both cases: cent controversy (e.g., Blinder 1997;
with and without commitment. Along McCallum 1997a). We demonstrate,
with expositing traditional results, we however, that there may be gains from
also exposit some new results regarding commitment simply if current price set-
the gains from commitment. ting depends on expectations of the fu-
Section 3 derives the optimal policy ture. In this instance, a credible com-
rule in the absence of commitment. If mitment to fight inflation in the future
for no other reason, this case is of inter- can improve the current output/infla-
est because it captures reality: No ma- tion trade-off that a central bank faces.
jor central bank makes any type of bind- Specifically, it can reduce the effective
ing commitment over the future course cost in terms of current output loss that
of its monetary policy. A number of is required to lower current inflation.
broad implications emerge from this This result, we believe, is new in the
baseline case. Among these: The opti- literature.
mal policy embeds inflation targeting in In practice, however, a binding com-
the sense that it calls for gradual adjust- mitment to a rule may not be feasible
ment to the optimal inflation rate. The simply because not enough is known
implication for the policy rule is that about the structure of the economy or
the central bank should adjust the the disturbances that buffet it. Under
nominal short rate more than one-for- certain circumstances, however, a pol-
one with expected future inflation. That icy rule that yields welfare gains rela-
is, it should adjust the nominal rate suf- tive to the optimum under discretion
ficiently to alter the real rate (and thus may be well approximated by an opti-
aggregate demand) in the direction that mal policy under discretion that is ob-
is offsetting to any movement in ex- tained by assigning a higher relative
pected inflation. Finally, how the cen- cost to inflation than the true social
tral bank should adjust the interest rate cost. A way to pursue this policy opera-
in response to output disturbances de- tionally is simply to appoint a central bank
pends critically on the nature of the dis- chair with a greater distaste for infla-
turbances: It should offset demand tion than society as a whole, as Kenneth
shocks but accommodate supply shocks, Rogoff (1985) originally emphasized.
as we discuss. Section 5 considers a number of prac-
Section 4 turns to the case with com- tical problems that complicate policy-
mitment. Much of the literature has making. These include: imperfect infor-
emphasized that an inefficiently high mation and lags, model uncertainty and
steady state inflation rate may arise in non-smooth preferences over inflation
the absence of commitment, if the cen- and output. A number of pragmatic is-
tral bank’s target for real output ex- sues emerge, such as: whether and how
ceeds the market clearing level. 7 The to make use of intermediate targets, the
7 The potential inflationary bias under discre-
choice of a monetary policy instrument,
tion was originally emphasized by Kydland and and why central banks appear to smooth
Prescott (1977) and Barro and Gordon (1983). interest rate changes. Among other
1664 Journal of Economic Literature, Vol. XXXVII (December 1999)
things, the analysis makes clear why that U.S. monetary policy in the fifteen
modern central banks (especially the years or so prior to Paul Volcker did not
Federal Reserve Board) have greatly always follow the principles we have de-
downgraded the role of monetary aggre- scribed. Simply put, interest rate man-
gates in the implementation of policy. agement during this era tended to ac-
The section also shows how the recently commodate inflation. Under Volcker
advocated “opportunistic” approach to and Greenspan, however, U.S. mone-
fighting inflation may emerge under a tary policy adopted the kind of implicit
non-smooth policy objective function. inflation targeting that we argue is
The opportunistic approach boils down consistent with good policy management.
to trying to keep inflation from rising The section also considers some pol-
but allowing it to ratchet down in the icy proposals that focus on target vari-
event of favorable supply shocks. ables, including introducing formal
As we illustrate throughout, the opti- inflation or price-level targets and
mal policy depends on the degree of nominal GDP targeting. There is in ad-
persistence in both inflation and out- dition a brief discussion of the issue of
put. The degree of inflation persistence whether indeterminacy may cause prac-
is critical since this factor governs the tical problems for the implementation of
output/inflation trade-off that the pol- simple interest rate rules. Finally, there
icy-maker faces. In our baseline model, are concluding remarks in section 8.
persistence in inflation and output is
due entirely to serially correlated ex- 2. A Baseline Framework for Analysis
ogenous shocks. In section 6 we con- of Monetary Policy
sider a hybrid model that allows for en-
dogenous persistence in both inflation This section characterizes the formal
and output. The model nests as special monetary policy design problem. It first
cases our forward-looking baseline presents a simple baseline macro-
model and, also, a more traditional economic framework, and then de-
backward-looking Keynesian frame- scribes the policy objective function.
work, similar to the one used by Lars The issue of credibility is taken up next.
Svensson (1997a) and others. In this regard, we describe the distinc-
Section 7 moves from theory to prac- tion between optimal policies with and
tice by considering a number of pro- without credible commitment—what
posed simple rules for monetary policy, the literature refers to as the cases of
including the Taylor rule, and a forward- “rules versus discretion.”
looking variant considered by Clarida, 2.1 A Simple Macroeconomic
Galí, and Gertler (1998; forthcoming). Framework
Attention has centered around simple
rules because of the need for robust- Our baseline framework is a dynamic
ness. A policy rule is robust if it pro- general equilibrium model with money
duces desirable results in a variety of and temporary nominal price rigidities.
competing macroeconomic frameworks. In recent years this paradigm has be-
This is tantamount to having the rule come widely used for theoretical analy-
satisfy the criteria for good policy man- sis of monetary policy. 8 It has much of
agement that sections 2 through 6 es- the empirical appeal of the traditional
tablish. Further, U.S. monetary policy may 8 See, e.g, Goodfriend and King (1997), McCal-
be judged according to this same met- lum and Nelson (1997), Walsh (1998), and the ref-
ric. In particular, the evidence suggests erences therein.
Clarida, Galí, Gertler: The Science of Monetary Policy 1665
IS/LM model, yet is grounded in dy- In addition, let πt be the period t infla-
namic general equilibrium theory, in tion rate, defined as the percent change
keeping with the methodological ad- in the price level from t–1 to t; and let i t
vances in modern macroeconomics. be the nominal interest rate. Each vari-
Within the model, monetary policy able is similarly expressed as a deviation
affects the real economy in the short from its long-run level.
run, much as in the traditional Keynes- It is then possible to represent the
ian IS/LM framework. A key difference, baseline model in terms of two equa-
however, is that the aggregate behav- tions: an “IS” curve that relates the out-
ioral equations evolve explicitly from put gap inversely to the real interest
optimization by households and firms. rate; and a Phillips curve that relates
One important implication is that cur- inflation positively to the output gap.
rent economic behavior depends criti- xt = − ϕ[it − Etπt + 1] + Etxt + 1 + gt (2.1)
cally on expectations of the future
course of monetary policy, as well as on πt = λxt + βEtπt + 1 + ut (2.2)
current policy. In addition, the model where gt and ut are disturbances terms
accommodates differing views about that obey, respectively:
how the macroeconomy behaves. In the gt = µgt − 1 + ^
gt (2.3)
limiting case of perfect price flexibility,
for example, the cyclical dynamics re- ut = ρut − 1 + ^
ut (2.4)
semble those of a real business cycle where 0 ≤ µ,ρ ≤ 1 and where both gt and ^
model, with monetary policy affecting ^t are i.i.d. random variables with zero mean
u
only nominal variables. and variances σ2g and σ2u, respectively.
Rather than work through the details Equation (2.1) is obtained by log-
of the derivation, which are readily linearizing the consumption euler equa-
available elsewhere, we instead directly tion that arises from the household’s
introduce the key aggregate relation- optimal saving decision, after imposing
ships. 9 For convenience, we abstract the equilibrium condition that con-
from investment and capital accumula- sumption equals output minus govern-
tion. This abstraction, however, does ment spending. 11 The resulting expres-
not affect any qualitative conclusions, as sion differs from the traditional IS
we discuss. The model is as follows: curve mainly because current output
Let y t and z t be the stochastic compo- depends on expected future output as
nents of output and the natural level of well as the interest rate. Higher ex-
output, respectively, both in logs.10 The pected future output raises current out-
latter is the level of output that would put: Because individuals prefer to
arise if wages and prices were perfectly
11 Using the market clearing condition Y = C + E ,
flexible. The difference between actual t t t
where E t is government consumption, we can re-
and potential output is an important vari- write the log-linearized consumption Euler equa-
able in the model. It is thus convenient tion as:
to define the “output gap” xt: yt − et = − ϕ[it − E tπt + 1 ] + E t{yt + 1 − et + 1 }
xt ≡ yt − zt Et
where et ≡ − log(1 − ) is taken to evolve ex-
Yt
9 See, for example, Yun (1996), Kimball (1995), ogenously. Using xt ≡ yt − zt, it is then possible to
King and Wolman (1995), Woodford (1996), and
derive the demand for output as
Bernanke, Gertler, and Gilchrist (1998) for step-
by-step derivations. xt = − ϕ[it − E tπt + 1 ] + E txt + 1 + g t
10 By stochastic component, we mean the devia-
tion from a deterministic long-run trend. where gt = E t{∆zt + 1 − ∆et + 1}.
1666 Journal of Economic Literature, Vol. XXXVII (December 1999)
1997). A key difference with the stan- We allow for the cost push shock to en-
dard Phillips curve is that expected fu- able the model to generate variation in
ture inflation, Etπt + 1, enters additively, inflation that arises independently of
as opposed to expected current infla- movement in excess demand, as appears
tion, Et − 1πt. 14 The implications of this present in the data (see, e.g., Fuhrer
distinction are critical: To see, iterate and Moore 1995).
(2.2) forward to obtain To close the model, we take the
∞ nominal interest rate as the instrument
πt = Et∑ βi[λxt + i + ut + i] (2.6) of monetary policy, as opposed to a
i=0 money supply aggregate. As Bernanke
and Ilian Mihov (1998) show, this as-
In contrast to the traditional Phillips
sumption provides a reasonable descrip-
curve, there is no arbitrary inertia or
tion of Federal Reserve operating pro-
lagged dependence in inflation. Rather,
cedures since 1965, except for the brief
inflation depends entirely on current
period of non-borrowed reserves target-
and expected future economic condi-
ing (1980–82) under Paul Volcker. 16
tions. Roughly speaking, firms set nomi-
With the nominal rate as the policy in-
nal price based on the expectations of
strument, it is not necessary to specify a
future marginal costs. The variable xt + i
money market equilibrium condition
captures movements in marginal costs
(i.e., an LM curve). 17 In section 5, we
associated with variation in excess de-
discuss the implications of using instead
mand. The shock ut + i, which we refer to
a narrow monetary aggregate as the
as “cost push,” captures anything else that
policy instrument.
might affect expected marginal costs. 15
Though simple, the model has the
same qualitative core features as more
14 Another key difference is that the explicit
derivation restricts the coefficient λ on the output On this latter point, see Erceg, Henderson, and
gap. In particular, λ is decreasing in θ, which mea- Levin (1998). Another interpretation of the ut shock
sures the degree of price rigidity. Thus, the longer (suggested by Mike Woodford) is that it could re-
prices are fixed on average, the less sensitive is flect a shock to the gap between the natural and
inflation to movements in the output gap. potential levels of output (e.g., a markup shock).
15 The relation for inflation that evolves from 16 Roughly speaking, Bernanke and Mihov
the Calvo model takes the form (1998) present formal evidence showing that the
πt = βE t{πt + 1} + δ mct Federal Reserve intervenes in the market for non-
borrowed bank reserves to support its choice for
where mc t denotes the deviation of (real) marginal the level of the Federal Funds rate, the overnight
cost from its steady state value. To then relate in- market for bank reserves. (Christiano, Eichen-
flation to the output gap, the literature typically baum, and Evans 1998, though, take issue with the
makes assumptions on technology, preferences, identifying assumptions in the Bernanke-Mihov
and the structure of labor markets to justify a pro- test). Informally, Federal Reserve policy actions in
portionate relation between real marginal cost and recent years routinely take the form of announcing
the output gap, so that mc t = κ xt holds, where κ is a target for the Federal funds rate (see, e.g, Rude-
the output elasticity of real marginal cost. In this busch 1995). Policy discussions, further, focus on
instance, one can rewrite the relation for inflation whether to adjust that target, and by how much.
in terms of the output gap, as follows: In this context, the view that the Funds rate is the
πt = βE t{πt + 1} + λ xt (see Galí and Gertler (1998) policy instrument is widely held by both practitio-
for details). In this context, the disturbance u t in ners of monetary policy and academic researchers
(2.2) is interpretable as reflecting deviations from (see, e.g., Goodfriend 1991, Taylor 1993, and
the condition mc t = κ x t. (Indeed the evidence in Walsh 1998).
Galí and Gertler 1998 suggests that mc t does not 17 With the interest rate as the policy instru-
vary proportionately with x t). Deviations from this ment, the central bank adjusts the money supply
proportionality condition could be caused, for ex- to hit the interest rate target. In this instance, the
ample, by movements in nominal wages that push condition that money demand equal money supply
real wages away from their “equilibrium” values simply determines the value of the money supply
due to frictions in the wage contracting process. that meets this criteria.
1668 Journal of Economic Literature, Vol. XXXVII (December 1999)
where the parameter α is the relative 19 Put differently, under the optimal policy, the
weight on output deviations. Since target inflation rate pins down the trend inflation
xt ≡ yt − zt, the loss function takes poten- rate. The loss function thus penalizes deviations
tial output z t as the target. It also implic- from this trend.
20 Some examples of this approach include Aiya-
itly takes zero as the target inflation, but gari and Braun (1997), King and Wolman (1995),
there is no cost in terms of generality Ireland (1996a), Carlstrom and Fuerst (1995), and
Rotemberg and Woodford (1997).
18 Some prominent examples include the re- 21 Underlying this kind of cost is the observation
cently renovated large scale model used by the that contracts are typically written in nominal
Federal Reserve Board, the FRB-US model (see terms and, for reasons that are difficult to explain,
Brayton, Levin, Tyron, and Williams 1997), and not perfectly indexed to the price level. On this
the medium scale models of Taylor (1979, 1993b) issue, see the discussion in Shiller (1997) and the
and Fuhrer and Moore (1995a,b). associated comment by Hall (1997).
Clarida, Galí, Gertler: The Science of Monetary Policy 1669
literature takes a pragmatic approach to But they define price stability as the in-
this issue by simply assuming that the flation rate at which inflation is no
objective of monetary policy is to mini- longer a public concern. In practice, it
mize the squared deviations of output is argued that an inflation rate between
and inflation from their respective tar- one and three percent seems to meet
get levels. However, Julio Rotemberg this definition (e.g., Bernanke and
and Michael Woodford (1999) and Mishkin 1997). A further justification
Woodford (1998) provide a formal justi- for this criteria is that the official price
fication for this approach. These indices may be overstating the true in-
authors show that an objective function flation rate by a percent or two, as ar-
looking something like equation (2.7) gued recently by the Boskin Commis-
may be obtained as a quadratic approxi- sion. In this regard, interestingly, the
mation of the utility-based welfare Bundesbank has had for a long time an
function. In this instance, the relative official inflation target of two percent. 23
weight, α, is a function of the primitive They similarly argue that this positive
parameters of the model. rate of inflation is consistent with price
In what follows, we simply adopt the stability, and cite measurement error as
quadratic objective given by (2.7), ap- one of the reasons (Clarida and Gertler
pealing loosely to the justification of- 1997).
fered in Rotemberg and Woodford It is clear that the experience of the
(1999). Judging by the number of pa- 1970s awakened policy-makers to the
pers written by Federal Reserve econo- costs of high inflation (DeLong 1997).
mists that follow this lead, this formula- Otherwise, there is no directly observ-
tion does not seem out of sync with the able indicator of the relative weights as-
way monetary policy operates in prac- signed to output and inflation objec-
tice (at least implicitly). 22 The target tives. Nor, argues Blinder (1997), is
level of output is typically taken to be there any obvious consensus among pol-
the natural level of output, based on the icy-makers about what these weights re-
idea that this is the level of output that ally are in practice. It is true that there
would obtain absent any wage and price has been a growing consensus that the
frictions. Yet, if distortions exist in the primary aim of monetary policy should
economy (e.g., imperfect competition be to control inflation (see, e.g., Ber-
or taxes), a case can be made that the nanke and Mishkin 1997). But this dis-
welfare maximizing level of output may cussion in many respects is about what
exceed its natural level. This issue be- kind of policy rule may be best, as op-
comes important in the context of posed to what the underlying welfare
policy credibility, but we defer it for function looks like.
now. For our purposes, however, it is rea-
What should be the target rate of in- sonable to take the inflation target and
flation is perhaps an even more ephem- preference parameters as given and
eral question, as is the issue of what simply explore the implications for
should be the relative weight assigned optimal policy rules.
to output and inflation losses. In the 23 Two percent is also the upper bound of the
U.S., policy-makers argue that “price inflation target range established by the European
stability” should be the ultimate goal. Central Bank. On the other hand, Feldstein (1997)
argues that the tax distortions that arise because
corporate and personal income taxes are not in-
22 See, for example, Williams (1997) and refer- dexed to inflation justify moving from three per-
ences therein. cent to zero inflation.
1670 Journal of Economic Literature, Vol. XXXVII (December 1999)
2.3 The Policy Problem and Discretion From the standpoint of policy design,
versus Rules the issue is to identify whether some
type of credibility-enhancing commit-
The policy problem is to choose a ment may be desirable. Answering this
time path for the instrument i t to engi- question boils down to comparing opti-
neer time paths of the target variables mal policy under discretion versus rules
x t and π t that maximize the objective (using the terminology of the litera-
function (2.7), subject to the constraints ture). In our context, a central bank op-
on behavior implied by (2.1) and (2.2). erating under discretion chooses the
This formulation is in many ways in the current interest rate by reoptimizing
tradition of the classic Jan Tinbergen every period. Any promises made in the
(1952)/Henri Theil (1961) (TT) targets past do not constrain current policy.
and instruments problem. As with TT, Under a rule, it chooses a plan for the
the combination of quadratic loss and path of the interest rates that it sticks to
linear constraints yields a certainty forever. The plan may call for adjusting
equivalent decision rule for the path of the interest rate in response to the state
the instrument. The optimal feedback of the economy, but both the nature
rule, in general, relates the instrument and size of the response are etched in
to the state of the economy. stone.
There is, however, an important dif- Two points need to be emphasized.
ference from the classic problem: The First, the key distinction between dis-
target variables depend not only on the cretion and rules is whether current
current policy but also on expectations commitments constrain the future
about future policy: The output gap de- course of policy in any credible way. In
pends on the future path of the interest each instance, the optimal outcome is a
rate (equation 2.5); and, in turn, inflation feedback policy that relates the policy
depends on the current and expected instrument to the current state of the
future behavior of the output gap economy in a very specific way. The two
(equation 2.6). As Finn Kydland and approaches differ, however, in their im-
Edward Prescott (1977) originally em- plications for the link between policy
phasized, in this kind of environment, intentions and private sector beliefs.
credibility of future policy intentions Under discretion, a perceptive private
becomes a critical issue. For example, a sector forms its expectations taking into
central bank that can credibly signal its account how the central bank adjusts
intent to maintain inflation low in the policy, given that the central bank is
future may be able to reduce current free to reoptimize every period. The ra-
inflation with less cost in terms of out- tional expectations equilibrium thus has
put reduction than might otherwise be the property that the central bank has
required. 24 In section 4, we illustrate no incentive to change its plans in an
this point explicitly. unexpected way, even though it has the
24 In this regard, we stress further that, in discretion to do so. (For this reason, the
contrast to conventional wisdom, the issue of policy that emerges in equilibrium under
credibility in monetary policy is not tied to central discretion is termed “time consistent.”)
bank objectives over output. In the classic,
Barro/Gordon (1983) formulation (and countless In contrast, under a rule, it is simply
papers thereafter), the central bank’s desire to
push output above potential output gives rise to
the credibility problem. However, as we make depends on beliefs about the future, even if cen-
clear in section 4, gains from commitment poten- tral bank objectives over output are perfectly
tially emerge whenever private sector behavior aligned.
Clarida, Galí, Gertler: The Science of Monetary Policy 1671
the binding commitment that makes the possible gains from commitment to a
policy believable in equilibrium. policy rule and other institutional de-
Second, (it should almost go without vices that might enhance credibility, it
saying that) the models we use are no- is necessary to understand what the
where near the point where it is possi- benchmark case of discretion yields.
ble to obtain a tightly specified policy Under discretion, each period the
rule that could be recommended for central bank chooses the triplet {xt,πt,it},
practical use with great confidence. consisting of the two target variables
Nonetheless, it is useful to work and the policy instrument, to maximize
through the cases of discretion and the objective (2.7) subject to the aggre-
rules in order to develop a set of norma- gate supply curve (2.2) and the IS
tive guidelines for policy behavior. As curve, (2.1). It is convenient to divide
Taylor (1993a) argues, common sense the problem into two stages: First, the
application of these guidelines may im- central bank chooses x t and π t to maxi-
prove the performance of monetary pol- mize the objective (2.7), given the infla-
icy. We expand on this point later. In tion equation (2.2). 25 Then, conditional
addition, understanding the qualitative on the optimal values of x t and π t, it de-
differences between outcomes under termines the value of i t implied by the
discretion versus rules can provide les- IS curve (2.1) (i.e., the interest rate
sons for the institutional design of that will support x t and π t).
monetary policy. For example, as we Since it cannot credibly manipulate
discuss, Rogoff’s (1985) insightful beliefs in the absence of commitment,
analysis of the benefits of a conservative the central bank takes private sector
central bank chair is a product of this expectations as given in solving the
type of analysis. Finally, simply under- optimization problem. 26 (Then, condi-
standing the qualitative aspects of opti- tional on the central bank’s optimal
mal policy management under discre- rule, the private sector forms beliefs ra-
tion can provide useful normative tionally.) Because there are no en-
insights, as we show shortly. dogenous state variables, the first stage
We proceed in the next section to de- of the policy problem reduces to the fol-
rive the optimal policy under discretion. lowing sequence of static optimization
In a subsequent section we then evaluate
the implications of commitment. 25 Since all the qualitative results we derive
stem mainly from the first stage problem, what is
critical is the nature of the short run Phillips
3. Optimal Monetary Policy without curve. For our baseline analysis, we use the Phil-
Commitment lips curve implied the New Keynesian model. In
section 6 we consider a very general Phillips curve
that is a hybrid of different approaches and show
We begin with the case without com- that the qualitative results remain intact. It is in
mitment (“discretion”) for two reasons. this sense that our analysis is quite robust.
26 We are ignoring the possibility of reputational
First, at a basic level this scenario ac-
equilibria that could support a more efficient out-
cords best with reality. In practice, no come. That is, in the language of game theory, we
major central bank makes any kind of restrict attention to Markov perfect equilibria.
binding commitment over the course of One issue that arises with reputational equilibria is
that there are multiplicity of possible equilibria.
its future monetary policy. In this re- Rogoff (1987) argues that the fragility of the re-
spect, it seems paramount to under- sulting equilibria is an unsatisfactory feature of
stand the nature of optimal policy in this approach. See also, Ireland (1996b). On the
other hand, Chari, Christiano, and Eichenbaum
this environment. Second, as we have (1998) argue that this indeterminacy could provide
just discussed, to fully comprehend the a source of business fluctuations.
1672 Journal of Economic Literature, Vol. XXXVII (December 1999)
baseline model. 29 In (σx,σπ) space the pends only on current and future de-
locus is downward sloping and convex mand. By adjusting interest rates to set
to the origin. Points to the right of the xt = 0, ∀ t, the central bank is able to hit
frontier are inefficient. Points to the its inflation and output targets simulta-
left are infeasible. Along the frontier neously, all the time. If cost push fac-
there is a trade-off: As α rises (indicat- tors drive inflation, however, it is only
ing relatively greater preference for possible to reduce inflation in the near
output stability), the optimal policy en- term by contracting demand. This
gineers a lower standard deviation of consideration leads to the next result:
output, but at the expense of higher in- Result 2: The optimal policy incorpo-
flation volatility. The limiting cases are rates inflation targeting in the sense
instructive: that it requires to aim for convergence
σu of inflation to its target over time. Ex-
As α → 0: σx = ; σπ = 0 (3.7) treme inflation targeting, however, i.e.,
λ
adjusting policy to immediately reach
σu an inflation target, is optimal under
As α → ∞: σx = 0; σπ = (3.8)
1 − βρ only one of two circumstances: (1) cost
where σu is the standard deviation of the push inflation is absent; or (2) there is
cost push innovation. no concern for output deviations (i.e.,
It is important to emphasize that the α = 0).
trade-off emerges only if cost push in- In the general case, with α > 0 and
flation is present. In the absence of cost σu > 0, there is gradual convergence of
inflation (i.e., with σu = 0), there is no inflation back to target. From equations
trade-off. In this instance, inflation de- (3.5) and (2.4), under the optimal policy
29 Equations (3.4) and (3.5) define the frontier lim Et{πt + i} = lim αqρi ut = 0
for the baseline model. i→∞ i→∞
1674 Journal of Economic Literature, Vol. XXXVII (December 1999)
In this formal sense, the optimal pol- several results regarding the behavior of
icy embeds inflation targeting. 30 With the policy instrument, i t.
exogenous cost push inflation, policy af- Result 3: Under the optimal policy,
fects the gap between inflation and its in response to a rise in expected infla-
target along the convergent path, but tion, nominal rates should rise suffi-
not the rate of convergence. In con- ciently to increase real rates. Put differ-
trast, in the presence of endogenous in- ently, in the optimal rule for the
flation persistence, policy will generally nominal rate, the coefficient on expected
affect the rate of convergence as well, inflation should exceed unity.
as we discuss later. Result 3 is transparent from equation
The conditions for extreme inflation (3.6). It simply reflects the implicit tar-
targeting can be seen immediately from geting feature of optimal policy de-
inspection of equations (3.7) and (3.8). scribed in Result 2. Whenever inflation
When σ u = 0 (no cost push inflation), is above target, the optimal policy re-
adjusting policy to immediately hit the quires raising real rates to contract de-
inflation target is optimal, regardless of mand. Though this principle may seem
preferences. Since there is no trade-off obvious, it provides a very simple crite-
in this case, it is never costly to try to ria for evaluating monetary policy. For
minimize inflation variability. Inflation example, Clarida, Galí, and Gertler (forth-
being the only concern of policy pro- coming) find that U.S. monetary policy
vides the other rationale for extreme in- in the pre-Volcker era of 1960–79 vio-
flation targeting. As equation (3.7) indi- lated this strategy. Federal Reserve pol-
cates, it is optimal to minimize inflation icy tended to accommodate rather than
variance if α = 0, even with cost push fight increases in expected inflation.
inflation present. Nominal rates adjusted, but not suffi-
Result 2 illustrates why some con- ciently to raise real rates. The persis-
flicting views about the optimal transi- tent high inflation during this era may
tion path to the inflation target have have been the end product of the fail-
emerged in the literature. Marvin ure to raise real rates under these cir-
Goodfriend and Robert King (1997), for cumstances. Since 1979, however, the
example, argue in favor of extreme in- Federal Reserve appears to have adopted
flation targeting. Svensson (1997a,b) the kind of implicit inflation targeting
and Laurence Ball (1997) suggest that, strategy that equation (3.6) suggests.
in general, gradual convergence of in- Over this period, the Fed has systemati-
flation is optimal. The difference stems cally raised real rates in response to an-
from the treatment of cost push infla- ticipated increases in inflationary ex-
tion: It is absent in the Goodfriend- pectations. We return to this issue later.
King paradigm, but very much a factor Result 4: The optimal policy calls for
in the Svensson and Ball frameworks. adjusting the interest rate to perfectly off-
Results 1 and 2 pertain to the behav- set demand shocks, g t, but perfectly ac-
ior of the target variables. We now state commodate shocks to potential output, zt,
by keeping the nominal rate constant.
30 Note here that our definition is somewhat
That policy should offset demand
different from Svensson (1997a), who defines shocks is transparent from the policy
inflation targeting in terms of the weights on the rule (3.6). Here the simple idea is that
objective function, i.e., he defines the case with countering demand shocks pushes both
α = 0 as corresponding to strict inflation targeting
and α > 0 as corresponding to flexible inflation output and inflation in the right direc-
targeting. tion. Demand shocks do not force a
Clarida, Galí, Gertler: The Science of Monetary Policy 1675
short run trade-off between output and David Gordon (1983), and Rogoff
inflation. (1985), a voluminous literature has de-
Shocks to potential output also do not veloped on the issue of credibility of
force a short run trade-off. But they re- monetary policy. 34 From the standpoint
quire a quite different policy response. of obtaining practical insights for pol-
Thus, e.g., a permanent rise in produc- icy, we find it useful to divide the pa-
tivity raises potential output, but it also pers into two strands. The first follows
raises output demand in a perfectly off- directly from the seminal papers and
setting manner, due to the impact on has received by far the most attention
permanent income. 31 As a consequence, in academic circles. It emphasizes the
the output gap does not change. In problem of persistent inflationary bias
turn, there is no change in inflation. under discretion. 35 The ultimate source
Thus, there is no reason to raise inter- of this inflationary bias is a central bank
est rates, despite the rise in output. 32 that desires to push output above its
Indeed, this kind of scenario seems to natural level. The second is emphasized
describe well the current behavior of more in applied discussions of policy. It
monetary policy. Output growth was focuses on the idea that disinflating an
substantially above trend in recent times, economy may be more painful than nec-
but with no apparent accompanying in- essary, if monetary policy is perceived
flation. 33 Based on the view that the rise as not devoted to fighting inflation.
in output may mainly reflect productivity Here the source of the problem is
movements, the Federal Reserve has simply that wage and price setting
resisted large interest rate increases. today may depend upon beliefs about
The central message of Result 4 is where prices are headed in the future,
that an important task of monetary pol- which in turn depends on the course of
icy is to distinguish the sources of monetary policy.
business cycle shocks. In the simple These two issues are similar in a
environment here with perfect ob- sense: They both suggest that a central
servability, this task is easy. Later we bank that can establish credibility one
explore some implications of relaxing way or another may be able to reduce
this assumption. inflation at lower cost. But the source
of the problem in each case is different
4. Credibility and the Gains in subtle but important ways. As a
from Commitment consequence the potential empirical
Since the pioneering work of Kydland relevance may differ, as we discuss
and Prescott (1977), Robert Barro and below.
31 In this experiment we are holding constant
We first use our model to exposit the
the IS shock gt. Since gt = [(et − zt) − E t(et + 1 − zt + 1)], famous inflationary bias result. We then
(see footnote 9), this boils down to assuming illustrate formally how credibility can
either that the shock to zt is permanent (so that reduce the cost of maintaining low in-
E tzt + 1 − zt = 0) or that et adjusts in a way to offset
movements in gt. flation, and also discuss mechanisms in
32 That monetary policy should accommodate
movements in potential GDP is a theme of the 34 For recent surveys of the literature, see Fis-
recent literature (e.g., Aiyagari and Braun 1997; cher (1995), McCallum (1997) and Persson and
Carlstrom and Fuerst 1995; Ireland 1996a; and Tabellini (1997).
Rotemberg and Woodford 1997). This view was 35 While the inflationary bias result is best
also stressed in much earlier literature. See Fried- known example, there may also be other costs of
man and Kuttner (1996) for a review. discretion. Svennson (1997c), for example, argues
33 See Lown and Rich (1997) for a discussion of also that discretion may lead to too much inflation
the recent “inflation puzzle.” variability and too little output variability.
1676 Journal of Economic Literature, Vol. XXXVII (December 1999)
to acting as if k were zero. There would description of how incentive contracts for central
bankers may reduce the inflation bias; also, Faust
be no change in the path of output, but and Svensson (1998) for a recent discussion of
inflation would decline. reputational mechanisms.
1678 Journal of Economic Literature, Vol. XXXVII (December 1999)
1 − λω 1
= ut (4.9) ←→ max − [α (xct )2 + (πct )2]Lt
1 − βρ 2 (4.11)
The problem for the central bank is ∞
to choose the optimal value of the feed- with Lt ≡ Et∑ βi(ut + i/ut)2 > 0. The prob-
i = 0
back parameter ω. Relative to the case
of discretion, the ability to commit to lem then is to choose ω to maximize
a feedback policy provides the central (4.11), subject to (4.10). In this instance,
bank with an improved short-run out- the optimality condition is given by:
put/inflation trade-off. To this end, note λ c
that it is possible to express equation xct = − πt
αc (4.12)
(4.9) as
where
λ 1
πct = xct + ut αc ≡ α(1 − βρ) < α (4.13)
1 − βρ 1 − βρ (4.10)
In this case, a one percent contraction in Since αc < α, relative to the case of dis-
xct reduces πct by the factor 1 −λβρ . Under cretion, commitment to the rule implies
that it is optimal for the central bank to
41 The policy rule only depends on u because
t engineer a greater contraction in output
the central bank can adjust i t to offset any impact in response to inflationary pressures.
of movements in g t on aggregate demand. See
equation (4.16). Intuitively, the more aggressive response
1680 Journal of Economic Literature, Vol. XXXVII (December 1999)
and output. This gain from commitment xt + i and πt + i to maximize the objective
arises even if the central bank does not (2.7) given that the aggregate supply
prefer to have output above potential curve (2.2) holds in every period
(i.e., even when k = 0). The solution t + i, i ≥ 0. We no longer restrict the
under commitment in this case perfectly choice of xt to depend on the contempo-
resembles the solution that would ob- raneous value of the shock (i.e., ut ), but
tain for a central bank with discretion allow instead for rules that are a func-
that assigned to inflation a higher cost tion of the entire history of shocks. To
than the true social cost. find the globally optimal solution to the
One additional interesting feature of linear quadratic policy problem under
this case with commitment involves the commitment, we follow David Currie
behavior of interest rates. This can be and Paul Levine (1993) and Woodford
seen formally by simply replacing α (1998), and form the Lagrangian: 45
with α c in the interest rate rule under
discretion (given by equation 3.6) to 1 ∞ i
2 i∑
obtain max − Et β [αxt2 + i + πt2 + i
=0
(4.17)
1
it = γπc Etπt + 1 + gt (4.16)
ϕ
+ φt + i(πt + i − λxt + i − βπt + i + 1 − ut + i)]
with
(1 − ρ)λ (1 − ρ)λ 1
γπc ≡ 1 + >1+ ≡ γπ where φt + i is the (state-contingent)
ρϕα c ρϕα 2
multiplier associated with the constraint at
In particular, relative to the case of dis-
t+i. It is straightforward to show that the
cretion, the central bank increases the
first order conditions yield the following
nominal interest rate by a larger amount
optimality conditions
in response to a rise in expected
inflation. λ
xt + i − xt + i − 1 = − π t + i,
4.2.2 Monetary Policy under Commit- α
ment: The Unconstrained Optimum for i = 1,2,3,... (4.18)
We now provide a brief description of and
the general solution for the optimal pol- λ
icy under commitment. 44 Because the xt = − πt (4.19)
α
derivation is more cumbersome than for
the restricted case just described, we Recall that under discretion the opti-
defer most of the details to an appen- mal policy has the central bank adjust
dix. As with the simple fundamental the level of the output gap in response
based policy, however, the general solu- to inflation. The optimal policy under
tion exploits the ability that commit- commitment requires instead adjusting
ment affords to manipulate private the change in the output gap in re-
sector expectations of the future. sponse to inflation. In other words,
The first stage problem remains to commitment changes the level rule for
choose a state-contingent sequence for x t under discretion into a difference
rule for x t, as a comparison of equations
44 We thank Chris Sims and Albert Marcet for
calling to our attention that the globally optimal 45 See also King and Wolman, who analyze the
rule under commitment would likely not fall optimal monetary policy under commitment in a
within the restricted family of rules considered in version of Taylor’s (1980) staggered contracts
the previous sub-section. model.
1682 Journal of Economic Literature, Vol. XXXVII (December 1999)
(3.3) and (4.8) indicates. 46 The one ca- tion (given the dependency of π t on fu-
veat is that in the initial period the pol- ture values of x t). Relative to the case of
icy is implemented (i.e., period t) the discretion, accordingly, the cost push
central bank should simply adjust the shock has a smaller impact in current
level of the output gap x t is response to inflation. 48
π t, as if it were following the optimal As with the constrained policy, the
policy under discretion, but for that globally optimal policy under commit-
period only. ment exploits the ability of the central
Because xt + i depends in general on bank to influence π t with expected fu-
xt + i − 1, the (unconstrained) optimal pol- ture values of xt + i as well as current x t.
icy under commitment is in general not It is also easy to see that, as was the
simply a function of the contemporane- case with the more restrictive rule, the
ous state variable ut + i. As Woodford policy is not time consistent. Clearly, if
(1998) emphasizes in a related context, it could reoptimize at t + i, the central
the lagged dependence in the policy bank would choose the same policy it
rule arises as a product of the central implemented at t, the one which mimics
bank’s ability under commitment to di- the rule under discretion for the first
rectly manipulate private sector expec- period only.
tations. 47 To see this for our framework, A disadvantage of the unconstrained
keep in mind that π t depends not only optimal policy under commitment is
on current x t but also on the expected that it appears more complex to imple-
future path of xt + i. Then suppose, for ment than the constrained one (de-
example, that there is a cost push shock scribed by equation 4.12). As we have
that raises inflation above target at time seen, the constrained rule resembles in
t. The optimal response under discre- every dimension the optimal policy un-
tion, as we have seen, is to reduce xt, der discretion, but with relatively more
but then let xt + i revert back to trend weight placed on fighting inflation. Ac-
over time as πt + i falls back to target. cordingly, as we discussed, it is possible
The optimal policy under commitment, to approximate this policy under discre-
however, is to continue to reduce xt + i as tion with an appropriately chosen cen-
long as πt + i remains above target. The tral banker. The same is not true, how-
(credible) threat to continue to contract ever, for the unconstrained optimal
x t in the future, in turn, has the imme- policy. A conservative central banker
diate effect of dampening current infla- operating with discretion has no obvi-
ous incentive to stick to the difference
46 Woodford (1998) makes the connection be- rule for the output gap implied by
tween the lagged dependence in the optimal rule equation (4.18).
under commitment and the lagged dependence
that appears to arise in interest rate behavior un- A further complication, discussed at
der practice (see section 5.2). Roughly speaking, length in Woodford (1998), is that the
since the interest rate affects the output gap, interest rate rule that implements the
lagged dependence in the latter translates into
lagged dependence in the former. optimal policy might have undesirable
47 Woodford (1998) considers a closely related
environment. The difference is that in his frame-
work the policy-maker confronts a trade-off be- 48 On the surface it appears that the difference
tween inflation and the output gap ultimately be- rule for x t might be unstable. However, πt adjusts
cause his objective function includes a target for to ensure that this is not the case. In particular,
the nominal interest rate (along with targets for the optimal response to a positive cost push shock
the output gap and inflation), whereas in our is to contract x t sufficiently to push π t below tar-
framework the trade-off arises due to the cost get. x t then adjusts back up to target over time.
push shock. The appendix provides the details.
Clarida, Galí, Gertler: The Science of Monetary Policy 1683
bank uses the short-term interest rate perfect information) and because the er-
or a monetary aggregate as the policy rors in forecasting the target variables
instrument, so long as the money de- are additive.
mand function yields a monotonic rela- For ease of exposition, assume that
tion between the two variables. 50 With there is no serial correlation in the
imperfect information, the ex post vola- cost push shock; that is, ρ = 0, so that
tility of a variety of key variables hinges ut = ^
ut. The implied equilibrium values
on the instrument choice, as originally of the target variables under imperfect
argued by William Poole (1970). We information, xIt and πIt , are given by
illustrate each of these issues below. 51 λ ^ ^ ^
xIt = xt + u t + g t = g t (5.2)
λ + α
2
5.1.1 Forecasts as Targets and
Intermediate Targets
λ2
πIt = 1 + πt + λg
^t = ^
ut + λg^t (5.3)
We now return to the baseline model α
with no commitment, and modify it as
where xt and πt are the optimal values of
follows. Suppose that the central bank
the target variables that emerge in case
cannot observe the contemporaneous
of perfect information (when ut is serially
values of output, inflation, or any of the
uncorrelated),53 and where ^ ut and ^
gt are
random shocks. Then let Ω t be the cen-
the unexpected movements in the cost
tral bank’s information set at the time it
push and demand shocks, respectively. Im-
fixes the interest rate that prevails at
perfect information clearly implies greater
time t. 52 The optimality condition for
volatility of inflation, since the central
policy now is expressed in terms of the
bank cannot immediately act to offset
expected as opposed to realized target
the impact of the shocks. The net effect
variables.
on the volatility of the output gap is un-
λ clear: the inability to offset the demand
E xt | Ω t = − E π t | Ω t
(5.1)
α
shock clearly raises output volatility. On
Equation (5.1) is the certainty equivalent the other hand, the central bank cannot
version of the condition for the case of offset the inflationary impact of the cost
perfect information, given by equation push shock, which works to reduce the
(3.3). Certainty equivalence applies here volatility of the output. There is, however,
because of the linear quadratic setup an unambiguous reduction in welfare. 54
(that gives linear decision rules under One additional result is worth noting.
Since demand shocks now affect the be-
50 To clarify, a money aggregate can serve as an
havior of output, a positive short-run
instrument only if it is directly controllable. A can-
didate aggregate then would be bank reserves. A co-movement between inflation and
broad aggregate such as M3 would not qualify. output can emerge if ^ gt has a variance
51 For a broad survey of the literature on mone-
sufficiently large relative to that of u ^t.
tary policy targets and instruments, see Friedman
(1991). It is straightforward to generalize the
52 Thus, Ω is similarly the private sector’s infor-
t analysis to a setting where the imper-
mation set. Specifically, we let firms observe the fect observability stems from lags in the
current values of their marginal costs, but neither −λ
53
firms nor households can observe contemporane- When u t is serially uncorrelated, xt = 2
ut
ous aggregate variables. In this instance, the IS λ +α
and Phillips curve equations are respectively given α
by and πt = u t.
λ2 + α
54 To prove that imperfect information leads to a
xt = − ϕ[(it | Ω t) − E t − 1 πt+1 ] + E t − 1 xt + 1 + g t
reduction in welfare, evaluate the welfare function
πt = λxt + βE t − 1 πt + 1 + u t with xIt and πIt versus xt and πt.
Clarida, Galí, Gertler: The Science of Monetary Policy 1685
inflation is likely to vary with changes where pt is the price level and v t is a
in the policy rule. In the end, there is random disturbance to money demand.
no simple substitute for employing a If v t is perfectly observable then it does
structural model. not matter whether i t or m t is employed
To summarize, we have as the policy instrument. Given the time
Result 9: With imperfect informa- path of i t implied by the optimal policy,
tion, stemming either from data prob- it is possible to back out a time path
lems or lags in the effect of policy, the for m t that supports this policy from
optimal policy rules are the certainty equation (5.4).
equivalent versions of the perfect infor- Matters change if v t is not observ-
mation case. Policy rules must be ex- able. With the interest rate as the in-
pressed in terms of the forecasts of tar- strument, the central bank lets the
get variables as opposed to the ex post money stock adjust to the money de-
behavior. Using observable intermediate mand shock. There is no impact of
targets, such as broad money aggregates is money demand shocks on output or in-
a possibility, but experience suggests that flation because the central bank per-
these indirect indicators are generally fectly accommodates them. With money
too unstable to be used in practice. targeting, the reverse is true: the inter-
est rate and (possibly) output adjust to
5.1.2 The Instrument Choice Problem: clear the money market. Assume for
The Interest Rate versus a Narrow simplicity that demand and cost push
Monetary Aggregate shocks are absent (i.e., g t = 0, u t = 0), so
that the only shock is the innovation to
We now turn to the issue of instru- money demand. Then the interest rate
ment choice. In practice, the interest implied by a money supply instrument
rate that major central banks adjust is im
t , is given by
an overnight rate on interbank lending 1
t = it +
of funds to meet reserve require- im ^
vt (5.5)
η + ϕ(κ + λ)
ments. 58 They control this rate by ma-
nipulating the supply of bank reserves, where it is the rate that would arise un-
i.e., the quantity of high-powered der interest rate targeting and ^ vt is the
money available for meeting bank re- unexpected movement in money demand.
serve requirements. The issue that The key point is that money demand
arises is whether, from an operational shocks can induce volatile behavior of
standpoint, policy should prescribe interest rates. This is particularly true if
paths (or rules) for bank reserves or for money demand is relatively interest in-
interest rates. Suppose that the demand elastic in the short run, as is the case
for bank reserves m t is given by 59 for bank reserves. This short run volatil-
ity in interest rates will then feed into
m t − p t = κ y t − η it + v t (5.4)
output volatility, via the aggregate de-
58 See Bernanke and Mihov (1997a) for a discus- mand relation, equation (2.1). It is for
sion of Federal Reserve operating procedures and this reason that in practice central banks
how they have changed over time. use interbank lending rates as the pol-
59 In the optimizing IS/LM framework of sec-
tion 2, it is possible to motivate this specification icy instrument, an insight due originally
of the money demand function from first princi- to Poole (1970). 60 Recent empirical
ples, assuming that utility is separable in consump-
tion and real money balances and that consump- 60 Poole also argued that if unobservable de-
tion is the only type of good (see, e.g., Woodford mand shocks were large relative to money demand
1996). shocks, then it may be preferable to use a money
Clarida, Galí, Gertler: The Science of Monetary Policy 1687
work by Bernanke and Mihov (1997a) This finding is not uncommon. The
confirms that except for the brief pe- FRB-US model also generates high in-
riod of non-borrowed reserve targeting terest rate volatility under an optimal
under Volcker (1979:10–1982:10), the rule. Because this degree of volatility
Federal Reserve Board has indeed seems greater than monetary policy
treated the Funds rate as the policy in- makers seem willing to tolerate in prac-
strument. In summary, we have tice, optimal rules are also computed
Result 10: Large unobservable with constraints on the volatility of
shocks to money demand produce high interest rate changes (see, e.g., John
volatility of interest rates when a mone- Williams 1997). 61
tary aggregate is used as the policy in- The tendency of the Federal Reserve
strument. It is largely for this reason to adjust rates cautiously is generally re-
that an interest rate instrument may be ferred to as “interest rate smoothing.”
preferable. To be precise, as a number of authors
The analysis thus makes clear why the have shown, a monetary policy rule of
new Federal Reserve Board model does the following form captures the last
not even bother to include a money ag- twenty or so years of data fairly well:
gregate of any form (see Flint Brayton
et al. 1997). Narrow aggregates are not it = (1 − ρ)[α + β πt + γ xt] + ρ it − 1 + εt (5.6)
good policy instruments due to the im- where α is a constant interpretable as
plied interest rate volatility. Broad ag- the steady state nominal interest rate62
gregates are not good intermediate tar- and where ρ ∈[0,1] is a parameter that
gets because of their unstable relation reflects the degree of lagged depen-
with aggregate activity. dence in the interest rate.63 Interest rate
smoothing is present in distinct respects.
5.2 Policy Conservatism: Model First, the estimated slope coefficients on
Uncertainty vs. Exploitation of inflation and the output gap, β and γ, are
Forward-Looking Behavior typically smaller than what the optimal
In practice, central banks adjust in- rule would suggest. Second, there is typi-
terest rates more cautiously than stan- cally partial adjustment to movements in
dard models predict. Put differently, πt and xt, reflected by the presence of the
optimal policies derived in a certainty lagged interest in the fitted rule. That is,
equivalent environment generally pre- i t is a weighted average of some desired
dict a much more variable path of inter- value that depends on the state economy
est rates than is observed in practice. (given by the term [ α + βπt + γ xt] ) and the
An interesting illustration of this point lagged interest rate, where the relative
is Rotemberg and Woodford (1997) who weights depend on the smoothing pa-
estimate a model very similar to our rameter ρ. Estimates of ρ for quarterly
baseline model, and then compute an data are typically on the order of 0.8 or
optimal interest rate policy. The histori- 0.9, which suggests very slow adjustment
cal interest rate displays much less vola- 61 An alternative is to penalize large changes in
tility than the optimal interest rate. the nominal interest rate by including the squared
deviations of the change in the interest rate (i.e,
( it − it − 1)2) in the function, as in Rudebusch and
supply instrument. With a money supply instru- Svensson (1998).
ment, interest rates will naturally move in an off- 62 Recall that π represents deviations of infla-
t
setting direction in response to unobserved de- tion from its average (target) level.
mand shocks (see Result 4). In practice, the high 63 See Rudebusch (1995), for example, for a dis-
variability of money demand shocks seems to cussion of the persistence in short term interest
dominate the instrument choice, however. rates.
1688 Journal of Economic Literature, Vol. XXXVII (December 1999)
in practice. The existing theory, by and To be concrete, suppose that the two
large, does not readily account for why parameters of the model, the interest
the central bank should adjust rates in elasticity in the IS equation and the
such a sluggish fashion. slope coefficient on the output gap are
~
Indeed, understanding why central random variables, now given by ϕt = ϕ + εt
~
banks choose a smooth path of interest and by λt = λ + ηt. 65 Assume further that
rates than theory would predict is an εt and ηt are i.i.d random variables with
important unresolved issue. One impli- zero means. The optimality condition
cation is that the standard certainty for policy then becomes:
equivalence models may not adequately
capture the constraints policy-makers λ
E{xt | Ω t} = E{πt | Ω t}
face in practice. A natural possibility is α + λ2σ2η
that policy-makers know far less about σ2ε
+ (α + λ2) rt (5.7)
the way the world works than is ϕ
presumed in simple policy experiments. where rt ≡ it − E{πt + 1 | Ω t} is the ex ante
In general, model uncertainty is a real interest rate. This condition leads to
formidable problem. Ideally, one would the following result:
like to take into account that the central Result 11: Parameter uncertainty
bank is continually learning about the may reduce the response of the policy
economy as it adjusts its policy. Per- instrument to disturbances in the econ-
forming this exercise in a clean way is omy. It can thus motivate a smoother
beyond the frontier of current knowl- path of the interest rate than the cer-
edge. Though, advances in computa- tainty equivalent policy implies.
tional methodology have allowed some Comparing equations (5.1) and (5.7)
progress to be made with relatively reveals how parameter uncertainty re-
simple frameworks. 64 duces policy activism. Under certainty
It is possible to illustrate how model equivalence, a rise in inflation above
uncertainty could in principle introduce target requires the central bank to raise
at least some degree of policy caution. interest rates to contract demand. 66
Suppose the values of several parame- With an uncertain slope coefficient
ters in the model are random. The cen- on the output gap in the AS curve, how-
tral bank knows the distribution of ever, contraction of output below po-
these parameters but not the realiza- tential raises the variability of inflation.
tion. When it adjusts policy, accord- This induces the central bank to moder-
ingly, it cannot be sure of the impact on ate the contraction in demand, as re-
the economy. As originally demon- flected by the presence of the term λ2σ2η
strated by William Brainard (1969), this in the coefficient on E{πt | Ω t}. Similarly,
kind of uncertainty can introduce cau- 65 We are assuming that the policy-maker knows
tion in policy responses. In contrast to the first two moments of the random parameters.
the case of certainty-equivalence, policy It may be more plausible to argue that the policy-
maker in fact has little idea what the true distri-
actions now affect the conditional vari- bution looks like. See Onatski and Stock (1999)
ance of inflation and output, as well as who analyze the policy problem in this kind of en-
the conditional mean. vironment using robust control methods.
66 It should also be clear from equation (5.7)
that with parameter uncertainty the interest rate
64 Wieland (1997) analyzes policy in a frame- no longer adjusts to perfectly offset demand
work where the central bank has to learn the value shocks. Suppose, for example, that there is a posi-
of the natural rate of unemployment (which, in tive demand shock. The interest rate goes up, but
our analysis, corresponds to having to learn about the parameter uncertainty moderates the extent of
potential GDP.) the rise, relative to the certainty equivalence case.
Clarida, Galí, Gertler: The Science of Monetary Policy 1689
uncertainty about the impact of an in- may care about avoiding excessive vola-
crease in the interest rate on the output tility in the short term interest rate in
gap moderates the extent of adjustment pursuing its stabilization goals.
in it. The second term on the right side To illustrate, consider the special
of equation (5.7) captures this latter case of equation (5.6) with ρ = 1. In this
dampening effect. instance, the difference in the interest
This simple form of model uncertainty rate (it − it − 1), as opposed to the level, is
thus may help explain the relatively low a linear function of πt and xt. Under the
variability of interest rates in the data. difference rule, the expected future
One feature of interest rate smoothing short rate at t + i, Et{it + i}, is given by
it does not appear to capture, however,
k
is the strong lagged dependence in the
Et{it + k} = Et∑ (it + j − it + j − 1) + it
interest rate. Put differently, the kind j=1
of parameter uncertainty we have dis- (5.8)
cussed may explain why the slope coef- k
= Et∑ [α + βπt + j + γxt + j] + it
ficients on inflation and the output gap,
j=1
α and β, are small relative to the case of
certainty equivalence. But it does not Assume that the long-term rate depends
explain the partial adjustment, given by on the sum of expected short rates over
the dependence of it on it − 1.67 the same horizon, in keeping with the ex-
Rotemberg and Woodford (1997) of- pectations hypothesis of the term struc-
fer a novel explanation for the lagged ture. Then, in comparison with the level
dependence that is based on the lever- rule, the difference rule increases the re-
age that this kind of adjustment rule sponsiveness of the long term rate under
may provide the central bank over the the feedback policy. Suppose for example
long term interest rate. The idea is that that, in reaction to a rise in inflation above
lagged dependence in it permits the target at time t, the central bank raises it
central bank to manipulate long term above its steady state value. Under the
rates, and hence aggregate demand, difference rule the increase in the inter-
with more modest movements in the est rate has a persistent effect on the path
short term rate than would be otherwise of the expected short rate, since Et it + i
be required. This kind of rule is thus depends additively on it. Further, if changes
desirable to the extent the central bank in inflation and output are persistent,
67 Sack (1997a,b) argues, nonetheless, that pa- then the path of expected short rates will
rameter uncertainty can explain this phenomenon actually be rising, as equation (5.8) makes
if the uncertainty of the impact of the interest rate clear. 68 The difference rule thus en-
on the economy is based on the change in the in-
terest rate (it − it − 1) as opposed to the deviation hances the countercyclical movement of
from trend it. In the former instance, changes in it the long rate relative to the movement of
raise the conditional variability of output, which short rate. Given that aggregate demand
induces the central bank to keep it close to it − 1.
On the other hand, it is not well understood how depends on the long rate, this kind of
the link between model uncertainty and policy rule thus enables the central bank to
conservatism is affected when there is active
learning about the economy. Some results suggest
that learning should induce active adjustments of 68 On the surface it appears that the interest
the policy instrument to facilitate estimating the rate might explode under the difference rule,
true model. See the discussion in Wieland (1997), since it will continue to increase so long as infla-
for example. Also, it is possible to construct exam- tion is above target. However, the rise in the inter-
ples where parameter uncertainty leads to in- est rate will dampen demand and inflation. In the
creased activism. See, for example, Thomas Sar- context of our model, it does so sufficiently to pre-
gent (1998). clude explosive behavior.
1690 Journal of Economic Literature, Vol. XXXVII (December 1999)
also depends negatively on the relative Most of the qualitative results ob-
cost of inflation, measured by 1/α. As in tained in the baseline case extend to
the baseline case, if the distaste for in- this more general setting. As in the
flation is high (α is low), the optimal baseline case, the policy-maker faces a
policy aggressively contracts demand short-run trade-off between output and
whenever inflation is above target: With inflation (Result 1). The effect of infla-
endogenous persistence, this contrac- tion inertia is to make this trade-off less
tion not only reduces inflation but also favorable. Equation (6.3) shows that
increases the speed of convergence to relative to the baseline case of φ = 0, the
target. Figure 2 illustrates the relation optimal policy requires a more aggres-
between aπ and α for three different sive response to any burst of inflation.
values of φ: φ = 0.01 (low inertia), φ = 0.5 The problem is that any inflation not
(medium) and φ = 0.99 (high). eliminated today persists into the fu-
Combining (6.3) with (6.1) yields the ture, potentially requiring more output
implied optimal interest rate rule: contraction. Figure 3 illustrates how the
1 trade-off becomes less favorable in this
it = γ π E t π t + 1 + γ x x t − 1 + g t (6.6) case by plotting the efficient policy
ϕ
frontier for the three benchmark values
with of φ. In addition, since 0 ≤ aπ < 1, the op-
λ(1 − aπ) timal policy calls for gradual adjustment
γπ = 1 +
ϕαaπ(1 − βaπ) of inflation to target (Result 2). With
θ φ > 0, further, extreme inflation target-
γx = ing is only optimal if α = 0, as equation
ϕ
(6.3) and Figure 2 suggest.
E tπ t + 1 = a π π t From the interest rate rule given by
1694 Journal of Economic Literature, Vol. XXXVII (December 1999)
and the right side reflects the two-period installing a conservative central bank
delay on inflation. chair.
Due to the delayed impact of policy,
the central bank takes both the output 7. Simple Rules for Monetary Policy
gap at t, xt, and the forecast of inflation
We next discuss some normative and
at t + 1, Et{πt + 1}, as predetermined from
positive aspects of simple feedback
the vantage of time t. The rest of the
rules for the interest rate that have
solution may thus be expressed in terms
been discussed in the literature. We then
of these predetermined variables:
discuss how these instrument-based
Et{πt + 2} = alπ Et{πt + 1} (6.8) rules are related to simple rules for tar-
it = γ π E t π t + 1 + γ x x t
l
(6.9) gets that have been recently proposed,
with including inflation targeting and nomi-
nal GDP targeting. Finally, we conclude
(γπ − 1)β
γπl = 1 + >1 with a brief discussion of the issue of
alπ possible indeterminacy of interest rate
The solution closely resembles the rules.
case without delay. Any differences just
7.1 Simple Interest Rate Rules
reflect the lagged influence of policy in
this environment. The nominal rate still Taylor (1993a) ignited the discussion
adjusts more than one-for-one with ex- of simple interest rate rules. 78 He pro-
pected inflation. Due to the lag struc- posed a feedback policy of the following
ture, though, it adjusts to the current form:
output gap, as opposed to one from the –) + γ x
i∗ = α + γ (π − π
t π t (7.1)
x t
previous period.
We conclude this section with brief with
discussion of the gains from commit- α = –r + π –
ment. It is possible to show that, as in γπ > 1, γx > 0
the baseline model, the policy rule un- ∗
der commitment resembles the rule un- where it is the target interest rate the
feedback rule defines, π – is the target in-
der discretion that would obtain if the –
policy-maker assigned a higher relative flation rate, and r is the long-run equi-
cost to inflation (lower value of α ) than librium real interest rate.79 Also, we now
the true social cost. Because inflation express all variables in levels, as opposed
inertia is endogenous in this case, the to deviations from trend.
optimal policy with commitment im- A number of other researchers have
plies a faster transition of inflation to considered rules like (7.1) (see, e.g.,
the optimum relative to what occurs un- Henderson and Mckibbon 1993). Tay-
der discretion. This can be seen by not- lor’s contribution is to spell out the nor-
ing that the parameter which governs mative and positive implications. On
the speed of convergence of inflation,
78 McCallum (1988) proposed a simple rule for
aπ, is decreasing in the relative cost of
the monetary base. The rule is less popular in pol-
inflation 1/α (see Figure 4). 77 Simply icy circles due to the implied interest rate volatil-
put, disinflations will be swifter than ity (see Result 9). McCallum (1997) argues, how-
otherwise if credible commitment is ever, that the concern about interest rate volatility
is not well understood, a point with which we
possible either directly or indirectly by agree.
79 The inflation rate Taylor uses is actually the
77 Note that the speed of convergence of infla- rate over the previous year (as opposed to the pre-
tion is decreasing in aπ. vious quarter).
1696 Journal of Economic Literature, Vol. XXXVII (December 1999)
the normative side, the rule is consis- rule provides a reasonably good descrip-
tent with the main principles for opti- tion of policy over the period 1987–92.
mal policy that we have described. It These are: γπ = 1.5, γx = 0.5, π– = 2, and
calls for gradual adjustment of inflation –r = 2. Taylor used informal judgement
to its target (see Result 2). Specifically, to pick them. An interesting question is
it has the nominal rate adjust more than whether a formal methodology would
one-for-one with the inflation rate. To yield something different.
the extent lagged inflation is a good In this spirit, Clarida, Galí, and Gertler
predictor of future inflation, the rule (forthcoming) estimate a simple rule for
thus has real rates adjusting to engineer U.S. monetary policy, and consider how
inflation back to target (see Result 3). this rule has evolved over time. The
Finally, note that the interest rate re- specific formulation is a “forward look-
sponds to the output gap as opposed to ing” version of the simple Taylor rule:
the level of output. Thus, in at least an –) + γ x
approximate sense, the rule calls for a ti∗ = α + γ (E π
π t t+1 −π x t (7.2)
countercyclical response to demand Under this rule, policy responds to
shocks and accommodation of shocks to expected inflation as opposed to lagged
potential GDP that do not affect the inflation. In this respect, the formula-
output gap (see Result 4). tion is consistent with the optimal rules
On the positive side, Taylor showed derived for both the baseline and hy-
that with certain parameter values, the brid models (see equations 3.6 and 6.6).
Clarida, Galí, Gertler: The Science of Monetary Policy 1697
TABLE 1
ESTIMATES OF POLICY REACTION FUNCTION
γπ γx ρ
Pre-Volcker 0.83 0.27 0.68
(0.07) (0.08) (0.05)
Volcker–Greenspan 2.15 0.93 0.79
(0.40) (0.42) (0.04)
Another virtue is that this formulation with standard errors are given by Table
nests the simple Taylor rule as a special 1. 80
case. If either inflation or a linear com- The key lesson involves the parame-
bination of lagged inflation and the out- ter γπ, the coefficient on the inflation
put gap is a sufficient statistic for future gap. The estimate for the pre-Volcker
inflation, then the specification col- rule is significantly less than unity. This
lapses to the Taylor rule. suggests that monetary policy over this
Because of the Federal Reserve’s ten- period was accommodating increases in
dency to smooth interest rate adjust- expected inflation, in clear violation of
ments (see the discussion in section 5), the guidelines suggested by Results 2
a static relation like equation (7.2) can- and 3. For the post-1979 rule the esti-
not capture the serial correlation pres- mate is significantly above unity. It thus
ent in the data. We thus allow for the incorporates the implicit inflation tar-
possibility of partial adjustment to the geting feature that we have argued is a
target rate, according to: critical feature of good monetary policy
management. It is also true that in the
it = ρ it − 1 + (1 − ρ)i∗t (7.3) Volcker–Greenspan era the Federal Re-
where ρ is a parameter that measures the serve was only responding to the output
degree of interest rate smoothing. gap to the extent it had predictive
We estimate different rules for the power for inflation: 81 The estimated co-
pre-Volcker (1960:1–79:2) and Volcker– efficient on the output gap, γx, is not
Greenspan (1979:3–96:4). We do so be- significantly different from zero. Pre
cause it is widely believed that U.S. 1979:4 it is positive and significant. This
monetary policy took an important turn outcome is consistent with the conven-
for the better with the appointment of tional view that pre-1979, the Federal
Paul Volcker as Fed Chairman (see
80 The estimates of the parameters in equation
Friedman and Kuttner 1996 and Gertler
(7.2) are obtained by using an instrumental vari-
1996). Among other things, this period ables procedure based on Generalized Methods of
marks the beginning of an apparently Moments (GMM). See Clarida, Galí, and Gertler
successful and long-lasting disinflation. (forthcoming) for details. The specific numbers
reported here are based on a version of this policy
We find that the simple rule given by reaction function that has the Funds rate respond
equation (7.2) does a good job of char- to expected inflation a year ahead and the current
acterizing policy in the Volcker–Green- output gap (reported in Table 2 of that paper).
The results, however, are robust to reasonable
span era. Further, it adheres to the variations in the horizons for the gap variables.
guidelines for good policy that we have 81 In particular, the output gap enters the in-
established. The estimated pre-Volcker strument set for expected inflation. Thus, the
coefficent γx reflects the influence of the output
rule violates these guidelines. Specifi- gap on the interest rate that is independent of its
cally, the parameter estimates along predictive power for inflation.
1698 Journal of Economic Literature, Vol. XXXVII (December 1999)
not much besides lagged inflation is Mishkin 1997 for a recent survey). In-
useful for predicting future inflation. deed a number of central banks, most
Finally, it is interesting to observe notably the Bank of England, have re-
that the other major central banks, the cently adopted formal inflation targets
Bundesbank and the Bank of Japan, (see, e.g., Andrew Haldane 1996).
have behaved very similarly in the post- In one sense, inflation targeting in-
1979 era. In Clarida, Galí, and Gertler volves nothing more than pursuing the
(1998), we estimate our specification kind of gradualist policy that our opti-
for these central banks. The estimated mal policy calculation implies (see Re-
parameters in each case are quite close sult 2). Indeed, all the leading real-
to those obtained for the Federal Re- world proposals call for gradual
serve during the Volcker–Greenspan convergence of inflation to target. None
period. Thus, good policy management recommend trying to hit the inflation
appears to have been a global phenome- target continuously, which is consistent
non. Perhaps this is not surprising since with our analysis. In this respect, the
the successful disinflation has also been rule we estimate for the period is per-
a world-wide event. fectly consistent with inflation targeting.
The rationale for inflation targeting,
7.2 Simple Target Rules
we think, is twofold. The first is simply
There have also been proposed sim- to guarantee that monetary policy
ple rules for targets, as opposed to in- avoids the mistakes of the pre-Volcker
struments. Of these proposed policies, era by identifying a clear nominal an-
inflation targeting has received by far chor for policy. (After all, Alan Green-
the most attention (see Bernanke and span will not be around forever). The
1700 Journal of Economic Literature, Vol. XXXVII (December 1999)
inflation target is in effect the nominal der an inflation targeting policy. For all
anchor. Since the anchor is directly in these reasons, it is perhaps not surpris-
terms of inflation, it avoids the poten- ing that no major central bank has
tially instability problems associated adopted a price level target.
with alternatives such as money growth Another candidate variable for target-
that are only indirectly linked to infla- ing is nominal GDP. This approach has
tion. For example, if there are large also received less attention in the re-
shocks to money demand, then a money cent literature, however. One problem
growth target may fail precisely to pin is that if there are shifts in the trend
down the equilibrium inflation rate. growth of real GDP, the rule does not
The second rationale has to do with provide a precise nominal anchor. An-
credibility and commitment. We have other problem, emphasized by Ball
seen that it is in general optimal for (1997), is that the policy may be overly
policy-makers to place a higher weight restrictive. In the hybrid model of sec-
on the costs of inflation than the true tion 5, for example, the optimal policy
social loss function suggests (see Re- in general has the interest rate adjust to
sults 6 and 7). The focus on inflation some linear combination of expected in-
targets may be viewed as a way to instill flation, the output gap and demand dis-
a higher effective weight on inflation in turbances. The weights depend upon
the policy choice. the underlying structural parameters of
Price level targeting is another type the model. Under nominal GDP target-
of simple rule that has been discussed ing, the central bank adjusts the inter-
in the literature. This policy, which may est rate to the sum of inflation and real
be thought of as a more extreme version GDP growth. It thus arbitrarily applies
of inflation targeting, has not received an equal weight to each component of
much support among policy-makers and nominal GDP. High nominal GDP
applied economists. There are several growth, further, could occur when the
problems: First, if the price level over- economy is recovering from a recession
shoots its target, the central bank may and is still well below full capacity. A
have to contract economic activity in or- rule that calls for raising interest rates
der to return the price level to its goal. in response to above-target nominal
That is, inflation above the amount im- GDP growth in these circumstances
plied by the price level target must be could stifle the recovery. 85
followed by inflation below this desired
7.3 Indeterminacy under Interest
amount in order to return the target.
Rate Rules
Under inflation targeting, bygones are
bygones: overshooting of inflation in One criticism of simple interest rate
one year does not require forcing infla- rules is that, under certain circum-
tion below target in the following year. stances, they may induce instability.
Second, the source of positive drift in That is, in many models there may not
the price level maybe measurement er- be a determinate equilibrium under
ror (see the discussion in section 2.) It particular parametrizations of the policy
would be unfortunate to have measure- 85 See Ball (1997) and Svensson (1997b) for ex-
ment error induce tightening of mone- plicit examples of how nominal GDP targeting
tary policy. Third, as McCallum (1997b) could produce adverse outcomes. McCallum
shows, the net reduction in price uncer- (1997c), however, argues that these results are
sensitive to the use of a backward-looking Phillips
tainty under a price level target rule, curve. For the case in favor of nominal GDP tar-
may be small relative that obtained un- geting, see Hall and Mankiw (1994).
Clarida, Galí, Gertler: The Science of Monetary Policy 1701
rule. In a classic paper, Thomas Sargent terminacy are possible. First, if in re-
and Neil Wallace (1975), illustrated sponse to a rise in expected inflation,
how nominal indeterminacy may arise if the nominal rate does not increase suf-
prices are perfectly flexible. Under an ficiently to raise the real rate, then self-
interest rate rule the equilibrium pins fulfilling bursts of inflation and output
down the level of real money balances. are possible. A rise in expected infla-
However, there are an infinite number tion, leads to a fall in real rates that, in
of combinations of the nominal money turn, fuels the boom. Indeed, the mone-
stock and the price level that satisfy this tary policy rule that Clarida, Galí, and
equilibrium condition. 86 In this respect, Gertler (forthcoming) estimate for the
the interest rate rule produces nominal pre-Volcker period permits exactly this
indeterminacy. 87 kind of sunspot behavior. The lesson
When there is sluggish price adjust- here is simply that a good monetary pol-
ment, the problem of nominal indeter- icy rule should not accommodate rise in
minacy vanishes. Last period’s price expected inflation. It should instead
level effectively serves a nominal an- pursue the implicit kind of inflation
chor. Simple interest rate rules thus do targeting that we have been emphasiz-
not produce price level indeterminacy ing. This boils down to raising nominal
in the frameworks we have analyzed. rates sufficiently to increase real rates
More generally, since there is little rea- whenever expected inflation goes up.
son to believe that prices are perfectly As Bernanke and Woodford (1998)
flexible, the issue of nominal indetermi- emphasize, indeterminacy is also possi-
nacy does not seem important in prac- ble if the rule calls for an overly aggres-
tice. On the other hand, there is poten- sive response of interest rates to move-
tially a problem of real indeterminacy ments in expected inflation. In this
in the case of price stickiness, as Wil- instance, there is a “policy overkill” ef-
liam Kerr and Robert King (1996), Ber- fect that emerges that may result in an
nanke and Woodford (1997) and Clarida, oscillating equilibrium. Clarida, Galí
Galí and Gertler (forthcoming) have re- and Gertler (forthcoming) show, how-
cently emphasized. 88 Two types of inde- ever, the magnitude of the policy re-
sponse required to generate indetermi-
86 McCallum (1997), however, argues that the nacy of this type greatly exceeds the
price level is in fact determined in this kind of estimates obtained in practice. This po-
environment. tential indeterminacy however does
87 A recent literature shows that the govern-
ment’s intertemporal budget constraint may re- suggest another reason why a gradual
store uniqueness under an interest rule, even in approach to meeting an inflation target
an environment with flexible prices. What is criti- may be desirable.
cal is whether the interest on the debt is financed
by taxes or money creation. See, for example,
Woodford (1994), Sims (1994), and Leeper (1991). 8. Concluding Remarks
88 These papers focus on local indeterminacy.
See Jess Benhabib, Stephanie Schmidt-Grohe, and
Martin Uribe (1998) for a discussion of global in- We conclude by describing several
determinacy. To avoid global indeterminacy, the areas where future research would be
central bank may have to commit to deviate from a quite useful:
simple interest rate rule if the economy were to
get sufficiently off track. This threat to deviate can (1) It is always the case that more
be stabilizing, much the way off the equilibrium knowledge of the way the macro-
path threats induce uniqueness in game theory. economy works can improve the perfor-
Because the threat is sufficient to preclude inde-
terminate behavior, further, it may never have to mance of monetary policy. Particularly
be implemented in practice. critical, however, is a better understanding
1702 Journal of Economic Literature, Vol. XXXVII (December 1999)