. 1
( 4)


Interest and Prices

Michael Woodford
Princeton University

Revised December 2002

Preliminary and Incomplete

c Michael Woodford 2002

3 Optimizing Models with Nominal Rigidities 1
1 A Basic Sticky-Price Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1 Price-Setting and Endogenous Output . . . . . . . . . . . . . . . . . 6
1.2 Consequences of Prices Fixed in Advance . . . . . . . . . . . . . . . . 17
1.3 A “New Classical” Phillips Curve . . . . . . . . . . . . . . . . . . . . 20
1.4 Sources of Strategic Complementarity . . . . . . . . . . . . . . . . . . 26
2 In¬‚ation Dynamics with Staggered Price-Setting . . . . . . . . . . . . . . . . 38
2.1 The Calvo Model of Price-Setting . . . . . . . . . . . . . . . . . . . . 42
2.2 A “New Keynesian” Phillips Curve . . . . . . . . . . . . . . . . . . . 51
2.3 Persistent Real E¬ects of Nominal Disturbances . . . . . . . . . . . . 52
2.4 Consequences of Persistence in the Growth of Nominal Spending . . . 62
2.5 Consequences of Sectoral Asymmetries . . . . . . . . . . . . . . . . . 65
3 Delayed E¬ects of Nominal Disturbances on In¬‚ation . . . . . . . . . . . . . 71
3.1 Staggered Pricing with Delayed Price Changes . . . . . . . . . . . . . 74
3.2 Consequences of Indexation to Past In¬‚ation . . . . . . . . . . . . . . 80
4 Consequences of Nominal Wage Stickiness . . . . . . . . . . . . . . . . . . . 87
4.1 A Model of Staggered Wage-Setting . . . . . . . . . . . . . . . . . . . 89
4.2 Sticky Wages and the Real E¬ects of Nominal Disturbances . . . . . 94
Chapter 3

Optimizing Models with Nominal

We turn now to the analysis of models in which monetary policy a¬ects the level of real
economic activity, and not just the level of money prices of goods and services. This requires
us to go beyond the analysis of endowment economies, as in the previous chapter, and allow
instead for endogenous supply decisions. But as is well-known, even when we allow for en-
dogenous supply, monetary policy can have only small e¬ects on the equilibrium allocation
of resources in an environment with perfect wage and price ¬‚exibility (and identical informa-
tion on the part of all decision-makers). Hence we also extend our analytical framework to
allow for delays in the adjustment of prices and/or wages to changing aggregate conditions.
In this way, we allow for non-trivial real e¬ects of monetary policy.
This extension of our framework is also important for a more realistic discussion of
central-bank interest-rate policy. In the basic model of chapter 2, the equilibrium real rate
of return is completely independent of monetary policy. This means that a central bank can
have no e¬ect on nominal interest rates except insofar as it can shift in¬‚ation expectations.
In our analysis above, we have assumed that it is able to do so, as long as the change in
expectations that is called for involves no violation of the postulate of rational expectations
on the part of the private sector; but our analysis may give the appearance of assuming precise
central-bank control of something upon which banks have little direct in¬‚uence in reality.
The introduction of price stickiness will make our assumption that the central bank can set


a short-term nominal interest rate less paradoxical. If private-sector in¬‚ation expectations
do not change when the central bank seeks to adjust the nominal interest rate, this does
not prevent the central bank from achieving its operating target; it simply means that the
private sector will perceive the real interest rate to have changed, which will a¬ect desired
expenditure and hence the degree of utilization of existing productive capacity. (We re-
examine the topic of in¬‚ation determination under an interest-rate rule in the context of
models with nominal rigidities in chapter 4.)

We shall give primary emphasis to models with sticky prices, though we also discuss the
consequences of wage stickiness as well. In this we follow most of the literature of the past
twenty years, but the choice perhaps deserves brief comment. One reason for emphasizing
price stickiness, at least for pedagogical purposes, is simply that models with only sticky
prices provide a simpler framework for the consideration of basic issues regarding the nature
of in¬‚ation determination. If we are to talk about the determinants of in¬‚ation (which in
the context of current policy debate, almost invariably means the rate of increase in goods
prices rather than wages), we must model the goods market; and if we are to consider such
central issues as the relation between interest rates and expenditure decisions, we need to
adjust our nominal interest rate for the expected rate of price in¬‚ation as well. On the other
hand, it is not equally essential to explicitly consider wage determination. It is possible to
model endogenous supply decisions without any reference to a labor market at all, as in
the familiar “yeoman farmer” models; and so the simplest models developed here will be of
this form, or will be equivalent in their implications to such a model, even if a (completely
frictionless) labor market is represented.

It is also often argued that there is more reason to believe that the stickiness of prices
matters for the allocation of resources. A well-known criticism of the models of nominal
wage rigidity popular in the 1970s was that the mere observation of infrequent changes in
individual nominal wages did not in itself prove the existence of a nominal rigidity with
any allocative consequences (Barro, 1977; Hall, 1980). Because employment is an ongoing
relationship rather than a spot-market transaction, the e¬ective cost to a ¬rm of increased

employment of labor inputs at a point in time need not equal the wage paid per hour of work
at that time; and under an e¬cient implicit contract, wages might well be smoother than the
e¬ective cost of labor, owing to a preference of workers for a smoother income stream. On
the other hand, it is less plausible that the observed rigidity of consumer goods prices should
not have allocative consequences, given the absence of a similar kind of ongoing relationship
between the suppliers of consumer goods and their customers (Rotemberg, 1987). However,
no convincing evidence has ever been o¬ered that the stickiness of nominal wages does not
result in stickiness of the e¬ective nominal cost of labor inputs; and evidence described below
” indicating that the evolution of U.S. in¬‚ation can be well explained by the evolution of
unit labor costs ” suggests that a model of supply costs that treats the reported wage as the
true marginal cost of additional hours of labor is not too inaccurate. Thus a more empirically
realistic model is likely to involve both wage and price stickiness. We treat models of that
kind in section xx below.

We shall give particular attention to the derivation of models with sticky prices and/or
wages in which prices and wages are nonetheless set optimally on the occasions when they
are adjusted. Allowing for optimal price- and wage-setting is important for several reasons.
One is that it allows us to highlight the importance of expectations for wage and price
dynamics. As we shall see in later chapters, forward-looking private sector behavior, in this
and other respects, has profound consequences for the optimal conduct of monetary policy.
It would thus be a serious mistake to simply assume mechanical wage- and price-adjustment
equations (perhaps drawn from the econometric literature), and treat these as structural for
purposes of an analysis of optimal monetary policy, as Lucas (1976) so forcefully argued.

Another reason for modeling optimal price- and wage-setting is that we are interested
in the welfare evaluation of alternative monetary policies. An especially appealing basis for
such evaluation is to ask how alternative possible equilibria compare from the point of view
of the private-sector objectives that underlie the behavior assumed in one™s model of the
e¬ects of alternative policies; but this is only possible insofar as the structural equations of
our model of the monetary transmission mechanism are derived from optimizing foundations.

As we shall see in chapter 6, alternative assumptions about the nature of price and wage
stickiness imply that alternative stabilization objectives for monetary policy are appropriate.

While we shall give detailed attention to the consequences of assumed delays in the ad-
justment of prices and/or wages, we shall not attempt here to say anything new about the
underlying reasons for these delays. Our assumptions about the frequency with which ¬rms
adjust their prices, or the time lag that may be involved between the decision about a price
and the time that the new price takes e¬ect, are treated as structural features of the environ-
ment in which ¬rms sell their products, with the same status as their production functions.
How reasonable this is depends on the question that one intends to ask of one™s model.
The “endogenous growth” has emphasized that, when thinking about the determinants of
economies™ long-run growth prospects, it is probably a mistake to ignore the endogeneity of
production functions ” for changes in economic conditions can change the incentives that
private parties have to devote resources to research and development, to introduce new prod-
ucts, and so on. On the other hand, for purposes of a comparison of alternative monetary
policy rules, it may not be a bad approximation to assume given production functions; the
real e¬ects of alternative monetary policies are relatively short-lived, and over this short
horizon production possibilities are unlikely to be much a¬ected by the temporary alteration
of incentives to innovate that may have occurred.

Similarly, if we wished to analyze the consequences of highly in¬‚ationary policies, we
would surely not want to treat as given the frequency of price adjustment, the degree of
indexation of wage contracts, or even the currency in terms of which prices are quoted;
we know that practices adjust in all of these respects (and for reasons that are easy to
understand) in economies that su¬er from sustained high in¬‚ation. But our interest in
the present study is in the identi¬cation of better monetary policies within the class of
policies under which in¬‚ation is never very great; in fact, we shall make extensive use of
approximations that are expected to be accurate only for the analysis of policies of that
kind. (It will perhaps not be giving away too much to divulge at this point that, according
to our analysis, optimal policy will indeed involve low and stable in¬‚ation!) For this purpose,

treating the delays involved in price and wage adjustment as structural may not be a bad
approximation. The sizes of wage and price increases do clearly vary from year to year, in
response to changes in perceived market conditions; practices with regard to the times at
which prices or wages are reconsidered, or the units in which they are announced, occur
much less often and only in response to more drastic changes in the economic environment.

Finally, we freely grant that the simple models presented here should be viewed only as
crude approximations to the actual monetary transmission mechanism. A realistic quan-
titative model would need to incorporate a large number of complications from which we
abstract here, in order to clarify basic concepts. One may wish to add endogenization of the
timing of price and wage adjustments to the list of re¬nements that one would like to incor-
porate into an eventual, truly accurate model. It is not clear, however, that this particular
re¬nement should be placed too high on the list of re¬nements when ranked in terms of their
likely quantitative importance for the analysis of monetary policy.

Nor is it even clear that any of the models with endogenous timing of price changes
that currently exist should be regarded as more realistic than the models presented below,
quite apart from the question of complexity. Some feel that models of “state-dependent
pricing”, such as those of Caplin and Leahy (1991) or Dotsey et al. (1999), have “better
microfoundations” than do the sorts of models presented here, that assume a given timing
for price and wage changes. But this is not obvious. These models assume that ¬rms are
constantly re-evaluating the price that they would adopt if they were to change their price,
and the expected bene¬ts from the change, and then weighing these bene¬ts against the
current “menu cost” of a price change to decide whether to actually change their price or
not. Yet in reality, the main bene¬t of infrequent price changes is not lower “menu costs”, but
reduction of the costs associated with information collection and decision-making (Zbaracki
et al., 1999). Obtaining this bene¬t necessarily means that the timing of the occasions upon
which prices are reconsidered will be largely independent of current market conditions; for
example, ¬rms often reconsider pricing policy at a particular time of year.

We begin in section 1 with a basic model of monopolistic competition, in which the prices

of some goods must be determined a period in advance. This very simple example of price
stickiness is useful for introducing a number of basic concepts. It also provides optimizing
foundations for a familiar aggregate-supply speci¬cation, the “New Classical Phillips curve”
used in many well-known analyses of optimal monetary policy, such as those of Kydland and
Prescott (1977) and Barro and Gordon (1983). While the literature using this speci¬cation
has produced a number of insights of more general importance, this relation is quite inade-
quate as a realistic account of the co-movement of real and nominal variables; it allows, for
example, no persistent e¬ects of monetary policy on real activity, and no e¬ects of antici-
pated policy. These strong conclusions are not general consequences of optimal price-setting,
as we show in section 2 through the analysis of a slightly more complex speci¬cation, the
Calvo (1983) model of staggered price-setting. While still very simple, this model implies
an aggregate-supply relation, sometimes called the “New Keynesian Phillips curve”, that
has proven capable of explaining at least some of the more gross features of in¬‚ation dy-
namics in the U.S. and elsewhere. Section 3 discusses still more complex speci¬cations with
increased empirical realism, that introduce delays in the e¬ects of monetary policy changes
on in¬‚ation. Finally, section 4 discusses models in which nominal wages are sticky as well
as prices.

1 A Basic Sticky-Price Model
We begin by displaying the structure of a very basic model, in which monetary policy has
real e¬ects as a result of some goods prices being ¬xed in advance. A number of issues
that are easy to analyze in this simple context will turn out also to be relevant to the more
realistic models to be developed in later sections.

1.1 Price-Setting and Endogenous Output

In order to be able to model price-setting, we must ¬rst extend the representative-household
model introduced in the previous chapter in certain respects that are quite distinct from the
issue of whether prices are assumed to be sticky. In particular, we must allow for endogenous

goods supply, rather than simply assuming a given endowment of goods. This requires that
we introduce a production technology and at least one variable factor of production (which
is labor, in this basic model). We shall be concerned to understand the determinants of
the costs of supplying goods, as supply costs are a prime determinant of optimal pricing.
We shall furthermore introduce di¬erentiated goods, and monopolistic competition among
the suppliers of these goods, as in the “New Keynesian” literature originated by Rotemberg
(1982), Mankiw (1985), Svensson (1986) and Blanchard and Kiyotaki (1987), rather than
assuming a single good in competitive supply. This last device, which is now quite com-
monplace, allows individual suppliers a degree of market power, and hence a decision about
how to set their prices. It also implies that a supplier that fails to immediately adjust its
price in response to a change in demand conditions does not su¬er an unboundedly large
(percentage) change in its sales, so that it becomes more plausible that prices should not be
constantly adjusted.1
We thus now assume that the representative household seeks to maximize a discounted
sum of utilities of the form
∞ 1
E0 β u(Ct , Mt /Pt ; ξt ) ’ v(ht (i); ξt )di . (1.1)

Here Ct is now an index of the household™s consumption of each of the individual goods
that are supplied, and Pt is a corresponding index of the prices of these goods, while ht (i)
is the quantity of labor of type i supplied. We assume that each of the di¬erentiated goods
(indexed by i over the unit interval) uses a specialized labor input in its production (and
in this chapter, this will be the only variable input); labor of type i is used to produce
di¬erentiated good i.
The introduction of di¬erentiated labor inputs is not necessary in order to allow us to
The size of the “menu costs” required to rationalize the failure of suppliers to adjust their prices imme-
diately has been the subject of an extensive literature. We shall not pursue this issue here, though we note
that the same sorts of “strategic complementarities” in price-setting that increase the degree of stickiness of
the general price level (as discussed below) when price adjustment is asynchronous also tend to reduce the
size of the costs of price changes to individual suppliers that are required to rationalize a failure to adjust
prices in response to an aggregate demand disturbance. On this latter point, see in particular Ball and
Romer (19xx).

analyze monopolistically competitive goods supply. But as we shall see, it is convenient to
do so. One reason is that we are able, in this case, to derive a model with factor markets
that is equivalent to the frequently-used “yeoman farmer” model, in which households are
assumed to directly supply goods. A more important reason is that it turns out that the
“strategic complementarity” between di¬erent suppliers™ pricing decisions is greater when we
assume that they do not hire labor from a single homogeneous (competitive) labor market.
Because we regard the conclusion obtained in the case of di¬erentiated labor inputs as the
more realistic one, we choose this speci¬cation as our baseline model. (The case of a single
homogeneous labor market is discussed in section 1.4 below.)
The term v(ht (i); ξt ) represents the disutility of supplying labor of type i; we assume that
for each possible value of ξ, v(·; ξ) is an increasing, convex function.2 We have written (1.1)
as if the representative household simultaneously supplies all of the types of labor. However,
we might equally well assume that each household specializes in the supply of only one type
of labor, but that there are an equal number of households supplying each type. In this case,
a household that supplies labor of type i seeks to maximize

β t [u(Ct , Mt /Pt ; ξt ) ’ v(ht (i); ξt )] .

When not all goods prices are set at the same time, households™ wage incomes will be
di¬erent, depending upon the type of labor they supply. But we may assume that there
exist competitive ¬nancial markets in which these risks are e¬ciently shared.
In this case, and if all households start with initial ¬nancial assets that give them the same
initial intertemporal budget constraints,3 then since households value consumption streams
(and money balances) identically and face the same prices, all households will choose identical
consumption and real balances in all states. (Note that while we have allowed for preference
Once again, ξt is a vector, so that the use of the same notation for exogenous disturbances to the functions
u and v involves no assumption about statistical dependence between the shifts in these two functions.
This means that if any households face di¬erent present values of their expected wage incomes as of date
zero, they hold initial ¬nancial claims W0 that di¬er in exactly the way necessary to o¬set the di¬erence in
their expected wages. Note that if this condition has ever held, then optimization in the presence of complete
¬nancial markets implies that it will hold forever after, regardless of which values may be realized for the
exogenous disturbances.

shocks ξt in (1.1), we assume that these are the same for all households “ we here contemplate
only aggregate shocks.) They will also choose portfolios of ¬nancial assets that insure that
they continue to have identical intertemporal budget constraints at all subsequent dates.4
The common intertemporal budget constraint in each state will in turn be exactly that of a
household that supplies all of the types of labor, and pools the wage income received.
Because each household chooses exactly the same state-contingent consumption plan,
the ¬rst-order conditions for optimal supply of each type of labor are exactly the same as
when a single household type supplies all types of labor so as to maximize (1.1). Thus the
conditions that determine equilibrium prices and quantities are the same in the two models.
Furthermore, if our welfare criterion in the specialized-labor model is the average level of
utility of all households, the level of social welfare associated with a given equilibrium will be
measured by the value of (1.1). Thus it makes no di¬erence to our conclusions which version
of the model we assume. The ¬ction that each household directly supplies all types of labor,
and so receives its pro rata share of the aggregate wage bill of the entire economy, simpli¬es
the exposition in that it allows us to dispense with explicit discussion of the risk-sharing
arrangements just referred to.
Following Dixit and Stiglitz (1977), we shall assume that the index Ct is a constant-
elasticity-of-substitution aggregator
1 θ’1
Ct ≡ ct (i) di (1.2)

with θ > 1, and that Pt is the corresponding price index
1 1’θ
pt (i)1’θ di
Pt ≡ . (1.3)

Note that (1.3) de¬nes the minimum cost of a unit of the aggregate de¬ned by (1.2), given
the individual goods prices {pt (i)}; since a household cares only about the number of units
Here our argument relies upon the assumption in (1.1) that the disutility of labor supply is additively
separable from the other terms. This implies that even if households expect to work di¬erent amounts in
particular states, they equalize their marginal utility of income in each state by holding assets that allow
them to a¬ord to consume exactly the same amount and hold exactly the same money balances as one
another. The more complicated case of non-separable preferences is treated in section xx below.

of this aggregate that it can purchase, de¬‚ation by Pt is an appropriate measure of the
purchasing power of nominal money balances Mt .
The household™s budget constraints are then as in chapter 2, except that the term pt ct
for nominal consumption expenditure must now be replaced by pt (i)ct (i)di, and the term

pt yt for income from the sale of goods must now be replaced by
1 1
wt (i)ht (i)di + Πt (i)di, (1.4)
0 0

where wt (i) is the nominal wage of labor of type i in period t, and Πt (i) represents the
nominal pro¬ts from sales of good i. In writing this last expression, we assume that each
household owns an equal share of all of the ¬rms that produce the various goods. Again,
given our assumption of complete ¬nancial markets, this assumption of distributed ownership
is irrelevant. We could also introduce trading in the shares of the ¬rms, without any change
in the conditions for a rational expectations equilibrium, except that then equilibrium share
prices would also be determined. As these extensions of the framework have no consequences
for the equilibrium evolution of goods prices or the quantities of goods supplied, we omit
further discussion of them.
As in chapter 2, each household then faces a single intertemporal budget constraint. Op-
timal (price-taking) household behavior is then described by the conjunction of three sets
of requirements. First, the household™s consumption spending must be optimally allocated
across di¬erentiated goods at each point in time, taking as given the overall level of expen-
diture. Thus the relative expenditures on di¬erent goods must be such as to maximize the
index (1.2) given the level of total expenditure. As in other applications of the Dixit-Stiglitz
model, this requires that purchases of each good i satisfy
pt (i)
ct (i) = Ct . (1.5)
This rule for distributing expenditure is easily seen to imply that total expenditure will
equal Pt Ct . Using this substitution, we can write both the household™s utility and its budget
constraints solely in terms of Pt and Ct , without any reference to quantities or prices of the
individual goods that are purchased.

Second, taking as given the optimal allocation of consumption expenditure at each date
(just described) and the amount of labor supplied (considered below), the household must
choose optimal levels of total consumption expenditure at each date, an optimal level of
money balances to hold at each date, an optimal amount of ¬nancial wealth to accumulate,
and an optimal portfolio allocation across the various types of state-contingent bonds that
are available. Necessary and su¬cient conditions for optimization in this respect are given by
exactly the same conditions as in chapter 2 “ namely, conditions (1.2), (1.12), (1.13), (1.15)
and (1.16) of that chapter must again hold at all times, where however Pt now refers to the
price index (1.3), ct is replaced by the index Ct de¬ned in (1.2), and yt is replaced by Yt ,
a similarly de¬ned aggregate of the quantities supplied of the various di¬erentiated goods.5
This is because both preferences over alternative streams of the consumption aggregate and
budget constraints written in terms of a¬ordable paths for the consumption aggregate are
exactly the same as in chapter 2.6
And ¬nally, the household must choose an optimal quantity of each kind of labor to
supply, given the wages that it faces and the value to it of additional income (determined by
the consumption-allocation problem just described). The ¬rst-order condition for optimal
supply of labor of type i at date t is given by
vh (ht (i); ξt ) wt (i)
= . (1.6)
uc (Ct , mt ; ξt ) Pt
These conditions, together with those listed earlier, comprise a complete set of necessary and
su¬cient conditions for household optimization.
We turn next to the speci¬cation of production possibilities. We assume that each good
i has a production function
yt (i) = At f (ht (i)), (1.7)
The relative quantities supplied of the various goods must be distributed in the same way as the relative
demands implied by (1.5). (When we allow below for government purchases, we assume that the government
seeks to maximize a similar aggregate of its purchases, and so distributes its purchases in the same manner.)
It then follows that total non-¬nancial income (1.4), which must equal total sales revenues of all of the ¬rms,
can be written as Pt Yt .
Here the additive separability of the disutility-of-labor terms in (1.1) is again crucial, allowing us to
obtain preferences over paths for the consumption aggregate and real balances that are the same as in our
earlier model with no labor supply decision.

where At > 0 is a time-varying exogenous technology factor, and f is an increasing, concave
function. Here labor is represented as the only factor of production (with one speci¬c type of
labor being used in the production of each good). We may think of capital as being allocated
to each ¬rm in a ¬xed amount, with capital goods never depreciating, never being produced,
and (because they are speci¬c to the ¬rm that uses them) never being reallocated among
¬rms; in this case, the additional argument of the production function may be suppressed.
(An extension of the model to allow for endogenous capital accumulation is presented in the
next chapter.)
It follows that the variable cost of supplying a quantity yt (i) of good i is given by

wt (i)f ’1 (yt (i)/At ).

Di¬erentiating this, we ¬nd that the (nominal) marginal cost of supplying good i is equal
wt (i)
St (i) = Ψ(yt (i)/At ),
Ψ(y) ≡ (1.8)
f (f ’1 (y))
is an increasing positive function. Substituting the labor supply function (1.6) for the wage,
we obtain a relation between the real marginal supply cost and the quantity supplied:

st (i) ≡ St (i)/Pt = s(yt (i), Yt ; ξt ),

where the real marginal cost function is de¬ned by
˜ = vh (f (y/A); ξ) Ψ(y/A).
s(y, Y ; ξ) (1.9)
uc (Y ; ξ)A
Here we assume that the producer is a wage-taker, even though we have supposed each good uses a
di¬erent type of labor with its own market. But our assumption of di¬erentiated labor inputs need not
imply that each producer is a monopsonist in its labor market. The only assumption that is important for
our results below is that producers that change their prices at di¬erent times also hire labor inputs from
distinct markets. We might, for example, assume a double continuum of di¬erentiated goods, indexed by
(i, j), with an elasticity of substitution of θ between any two goods, as above. We might then assume that
all goods with the same index i change their prices at the same time (and so always charge the same price),
and are also all produced using the same type of labor (type i labor). The degree of market power of each
producer in its product market would then be as assumed here, but the fact that a continuum of producers
all bid for type i labor would eliminate any market power in their labor market.

In this last expression, ξt represents the complete vector of exogenous disturbances, in which
the preference shocks ξt have been augmented by the technology factor At , we have substi-
tuted into the labor supply function the sectoral labor requirement as a function of sectoral
output, and we have used the fact that in equilibrium, the index of aggregate consumption
Ct must at all times equal the index of output Yt .8
Note also that we have suppressed real balances as an argument of uc (and hence as an
argument of the real marginal cost function) in the denominator. Abstracting from such
“real balance e¬ects” can be justi¬ed along any of several grounds discussed in chapter 2.
It is simplest to suppose that the economy considered in this chapter is a “cashless” one, in
which monetary policy is implemented in the way considered in section 1 of chapter 2.9 As in
chapter 2, we shall here assume a cashless economy as our baseline model; the consequences
of real balance e¬ects are considered in section 3.2 of chapter 4.
This model of production costs might alternatively be derived from a “yeoman farmer”
model, in which households directly supply goods, seeking to maximize
∞ 1
E0 β u(Ct , Mt /Pt ; ξt ) ’ v (yt (i); ξt )di
˜ . (1.10)

If we convert the marginal disutility of supply of good i into units of an equivalent quantity
of the consumption aggregate, we obtain a “real marginal cost” of good i equal to
vy (yt (i); ξt )
uc (Yt ; ξt )
When we allow for exogenous variation in government purchases, we can still apply this equation, if
we understand u(Y ; ξ) to mean the function u(Yt ; ξt ) introduced in section xx of chapter 2, measuring
household utility ¬‚ow as a function of aggregate demand rather than consumption expenditure. Under this
interpretation, the level of government purchases is just one element of the vector of exogenous disturbances
ξ that shifts this relation. When we allow for endogenous variations in investment spending, matters are
more complex; in such a case, it is important to remember that it is really uc (C; ξ) rather than uc (Y ; ξ) that
belongs in the denominator of (1.9).
Alternatively, our results apply to a Sidrauski-Brock model in which utility is separable, as discussed
in section xx of chapter 2; to a cash-in-advance model of the special type discussed in section xx of the
appendix to chapter 2; or to a “cashless limiting economy” of the sort discussed in section xx of chapter 2.
They also apply to a much broader class of models with transactions frictions, in the case that monetary
policy is implemented through a procedure under which the interest-rate di¬erential ∆t is held constant,
under a suitable reinterpretation of the parameter σ and the disturbance gt , so that ’σ ’1 (Yt ’ gt ) is the
deviation of the log marginal utility of real income from its steady-state level, in the case of a constant
interest-rate di¬erential (rather than a constant level of real balances). For discussion of this last case, see
section 3.2 of chapter 4, especially footnote xx.

This is in fact identical to (1.9) if the disutility of output supply is given by

v (y; ξ) ≡ v(f ’1 (y/A); ξ).

This concept of “real marginal cost” plays exactly the same role in optimal pricing in the
yeoman farmer model as does the more conventional concept in the case of supply by ¬rms
that purchase inputs, and the results that we obtain below are identical to those that one
would obtain from a yeoman farmer model. As noted above, this is one reason for interest
in the model with di¬erentiated labor inputs assumed here. However, explicitly modeling
the labor market has the advantage of allowing us to derive additional implications of the
model. It will also make the extension, below, to a model with sticky wages as well as prices
more straightforward.
With our theory of marginal supply costs in place, we now turn to the question of optimal
pricing. We shall ¬rst examine the case of perfectly ¬‚exible prices; that is, we shall assume
that the supplier of each good chooses a price for it each period, not constrained in any way
by the price that has been charged for the good in the past, and with full information about
current demand and cost conditions. As usual in a model of monopolistic competition, we
assume that each supplier understands that his sales depend upon the price charged for his
good, according to the demand function
pt (i)
yt (i) = Yt . (1.11)
(The form of the demand curve assumed here follows from (1.5); when all purchases are for
private consumption, the index of aggregate demand Yt corresponds simply to the represen-
tative household™s choice of the index Ct .) Because good i accounts for only an in¬nitesimal
contribution to households™ budgets and their utility from consumption, the supplier of an
individual good does not believe that his pricing decision can a¬ect the evolution of either
the index of aggregate demand Yt or the price index Pt ; thus pt (i) is chosen taking the latter
two quantities as given. Optimization by the supplier of good i then involves setting a price
pt (i) = µSt (i), where µ ≡ θ/(θ ’ 1) > 1 is the seller™s desired markup, determined by the
usual Lerner formula.

It follows that each supplier will wish to charge a relative price satisfying
pt (i) ˜
= µs(yt (i), Yt ; ξt ). (1.12)
It then follows from (1.11) that the relative supply of good i must satisfy
yt (i) ˜
= µs(yt (i), Yt ; ξt ).
Because s is increasing in its ¬rst argument, this equation must have a unique solution for
yt (i) given Yt . It follows that in equilibrium, the same quantity must be supplied of each
good, and that common quantity must equal Yt . Equilibrium output must then be given by
Yt = Y n (ξt ), where the latter function indicates the solution to the equation

s(Ytn , Ytn ; ξt ) = µ’1 . (1.13)

Because s is also increasing in its second argument, this equation as well must have a unique
solution for each speci¬cation of the exogenous shocks ξt .
We thus ¬nd that in the case of fully ¬‚exible prices, equilibrium output is completely
independent of monetary policy. Given this solution for aggregate output as a function of
the exogenous shocks, our model of price-level determination then reduces to exactly the
model analyzed in chapter 2 (where an exogenous supply of goods was simply assumed).
Thus neither our introduction of endogenous supply nor our assumption of monopolistic
competition has any necessary consequences for the e¬ects of monetary policy. But they
now make it possible for us to consider other assumptions about pricing behavior, and we
shall see that in the case of sticky prices our conclusions are di¬erent.
The solution to equation (1.13) “ which we call the natural rate of output following
Friedman (1968) “ continues to be a useful construct even in the case of sticky prices (though
it no longer need equal the equilibrium10 level of output at all times). This is because a log-
linear approximation to the real marginal cost function (1.9) is given by

ˆ ˆ
st (i) = ω yt (i) + σ ’1 Yt ’ (ω + σ ’1 )Ytn ,
ˆ ˆ (1.14)
Some authors would say that it is the “equilibrium” level of output, but that the level of output actually
observed as a result of sticky prices is instead a “disequilibrium” level of output. We shall avoid this
terminology in this study; for us, “equilibrium” always refers to the prediction of our model, whether it
involves fully ¬‚exible prices or not.

where ω > 0 represents the elasticity of s with respect to its ¬rst argument, and σ > 0 is
the intertemporal elasticity of substitution of private expenditure, as in chapter 2. (Here
we log-linearize around the steady-state equilibrium in the case of ¬‚exible prices and ξt = 0
at all times.11 Letting Y be the constant level of output in this steady state, we de¬ne
ˆ ¯ˆ ¯
Yt as in chapter 2, and correspondingly de¬ne yt (i) ≡ log(yt (i)/Y ), Ytn ≡ log(Ytn /Y ), and
st (i) ≡ log(µst (i)).) Thus the natural rate of output provides a useful summary of the way
in which disturbances shift the real marginal cost function, whether prices are constantly
adjusted or not.
For later purposes it is useful to note that in (1.14), the elasticity ω can be decomposed
ω = ωw + ωp , (1.15)

where ωw > 0 is the elasticity of the marginal disutility of work with respect to output
increases, and ωp > 0 is the elasticity of the function Ψ de¬ned in (1.8). Thus ωw indicates
the elasticity of real wage demands with respect to the level of output, holding ¬xed the
marginal utility of income, while ωp indicates the negative of the elasticity of the marginal
product of labor with respect to the level of output.12
Of course, our result here that monetary policy is completely irrelevant to the determi-
nation of real activity is rather special. If, for example, we allow for real balance e¬ects,
we shall ¬nd that monetary policy can a¬ect equilibrium output even under ¬‚exible prices,
owing to the e¬ects of expected in¬‚ation upon equilibrium real balances. If we furthermore
allow for endogenous capital accumulation, we shall ¬nd that the natural rate of output de-
pends upon the capital stock, and insofar as real balance e¬ects are able to a¬ect equilibrium
capital accumulation, they may have a further e¬ect upon equilibrium output under ¬‚exible
prices through this channel as well. However, these e¬ects are not plausibly very large in
quantitative terms, as studies such as that of Cooley and Hansen (1989) have shown. Thus
The element of ξt that measures aggregate technology is here taken to be at ≡ log At rather than At
itself. Thus the steady-state value of the technology factor At is normalized as one.
In a model with wage and price-setting, ωw indicates the degree to which higher economic activity
increases workers™ desired wages given prices, while ωp indicates the degree to which higher economic activity
increases producers™ desired prices given wages.

the conclusion from our simple model remains essentially correct.

1.2 Consequences of Prices Fixed in Advance

We now contrast these results to those that we obtain under a simple form of price stickiness.
Let us suppose that all prices pt (i) must be ¬xed a period in advance; that is, when pt (i) is
chosen, the exogenous disturbances (including possible random variation in monetary policy)
realized in periods t ’ 1 or earlier are known, but not any of the disturbances that are to
be realized only in period t. (Whether the stickiness of prices results because the price that
applies in period t has to be announced at an earlier time, or simply because price-setters
make their decision on the basis of old information, does not matter for our conclusions.) We
shall suppose that the supplier of good i is committed to supply whatever quantity buyers
may wish to purchase at the predetermined price pt (i), and hence to purchase whatever
quantity of inputs may turn out to be necessary to ¬ll orders.
When the price pt (i) is chosen, in period t ’ 1, the consequences for sales and pro¬ts in
period t are not yet known with certainty. Hence we assume that the ¬rm seeks to maximize
the present value of period t pro¬ts, given by

Et’1 [Qt’1,t Πt (i)], (1.16)

where Qt’1,t is the stochastic discount factor introduced in chapter 2, and pro¬ts from the
sale of good i are given by

Πt (i) = pt (i)yt (i) ’ wt (i)ht (i).

Using (1.7) and (1.11), we can write this objective as a function of the choice variable pt (i),
Et’1 Qt’1,t [Yt Ptθ pt (i)1’θ ’ wt (i)f ’1 (Yt Ptθ pt (i)’θ /At )] . (1.17)

The supplier of good i then chooses pt (i) on the basis of information available at date
t ’ 1 so as to maximize this expression, given the expected state-contingent values of the
random variables Qt’1,t , Yt , Pt , wt (i) and At .

The expression inside the square brackets in (1.17) is easily seen to be a concave function
of pt (i)’θ (which we might alternatively choose as the choice variable), so that the entire
expression is similarly concave. It follows that expected pro¬ts have a unique maximum
at the price that satis¬es the ¬rst-order condition obtained by di¬erentiating (1.17) with
respect to pt (i). This may be written

Et’1 Qt’1,t Yt Ptθ [pt (i) ’ µSt (i)] = 0. (1.18)

This way of writing the ¬rst-order condition indicates that the price is set to equal µ times
a weighted average of the values of the nominal marginal cost of supplying good i that are
expected to occur in di¬erent possible states at date t.
Substituting the demand function (1.11) into the real marginal cost function (1.9), one
sees that the marginal cost of supplying good i is a decreasing function of the price pt (i)
(given the values of the variables that the producer cannot a¬ect) in each state. Hence
(1.18) has a unique solution for the optimal price pt (i), that is the same for each good i. In
equilibrium, each producer sets an identical price, and this common price will be Pt . Using
the result that pt (i) = Pt , and substituting the solution for the stochastic discount factor
from equation (xx) of chapter 2, we ¬nd that (1.18) requires that

Et’1 uc (Yt ; ξt )Yt [µ’1 ’ s(Yt , Yt ; ξt )] = 0. (1.19)

This is a restriction that must be satis¬ed by the joint distribution of Yt and the exogenous
disturbances ξt , conditional upon information at date t ’ 1; note that it involves no nominal
variables. It is a weaker version of our result in the case of ¬‚exible prices, that Yt = Ytn at all
times. Output equal to the natural rate is equivalent to requiring that s(Yt , Yt ; ξt ) = µ’1 at
all times; instead, (1.19) requires only that this hold “on average” (where the “average” in
question does not involve weights exactly equal to the probability of each state™s occurrence).
Together with the stipulation that Pt is predetermined, (1.19) represents an aggregate
supply relation for this model. We can examine its implications, without needing to specify
the rest of the model, if we assume that monetary policy is used to achieve an exogenous

target path for nominal GDP, Yt = Pt Yt . This sort of aggregate demand speci¬cation is
very commonly assumed in the literature on sticky-price models, usually by stipulating
that nominal GDP is proportional to the money supply, and then that monetary policy is
speci¬ed by an exogenous target path for the money supply. It is not attractive, for our
purposes, to assume either a constant velocity of money or monetary targeting. But we may
nonetheless examine an equivalent aggregate demand speci¬cation, by assuming that policy
is speci¬ed in terms of a target path for nominal GDP, that is then achieved by adjusting
the interest-rate instrument as necessary. In addition to allowing comparisons with familiar
literature, this assumption allows us to examine the consequences of alternative aggregate
supply speci¬cations without needing to specify the way in which monetary policy a¬ects
aggregate spending. Accordingly, we specify aggregate demand in terms of an exogenous
process {Yt } throughout the present chapter. (The interest-rate adjustments required in
order to control aggregate spending are then taken up in chapter 4.)
Substituting Yt = Yt /Pt into (1.19), we observe that the equilibrium price level is given
by the solution to the equation

Et’1 uc (Yt /Pt ; ξt )Yt [µ’1 ’ s(Yt /Pt , Yt /Pt ; ξt )] = 0. (1.20)

This implies that Pt is a function solely of the joint distribution of {Yt , ξt }, conditional
upon information at date t ’ 1, and that this function is homogeneous of degree one in the
distribution of values anticipated for Yt . Given the value of Pt determined by this ex ante
distribution, the level of output Yt is then determined by the ex post realization of Yt . The
homogeneity property just referred to implies that Yt depends only upon the level of Yt
relative to the distribution of levels of nominal spending that were regarded as possible at
date t ’ 1.
This result can be stated more simply if we make use of a log-linear approximation to the
solution (1.20). We log-linearize around the steady-state equilibrium in which ξt = 0 and
Yt /Yt’1 = 1 at all times, and obtain an expression that approximates the exact solution as
long as ξt and Yt /Yt’1 are always su¬ciently close to these values. Making use of the log-

linear approximation to the real marginal cost function (1.14), we ¬nd that the equilibrium
price level is approximately given by

log Pt = Et’1 log Yt ’ Et’1 log Ytn ,

from which it follows that

log Yt = Et’1 log Ytn + [log Yt ’ Et’1 log Yt ].

We then observe that the component of output that can be forecasted a period in advance
is always equal to the forecast of the natural rate,

Et’1 log Yt = Et’1 log Ytn ,

and hence is independent of monetary policy. The unexpected component of output ¬‚uctu-
ations, by contrast, is equal to the unexpected component of nominal GDP (or of nominal
GDP growth):
log Yt ’ Et’1 log Yt = log Yt ’ Et’1 log Yt .

Thus monetary policy a¬ects real activity in this model only insofar as it causes unexpected
variation in nominal spending, and the resulting variations in output must themselves be
purely unexpected.

1.3 A “New Classical” Phillips Curve

The above model can be generalized by allowing some prices to be ¬‚exible, though others
are ¬xed in advance. This allows us to consider the robustness of our previous conclusions
to allowing some prices to be ¬‚exible, even in the very short run (as we do in fact observe).
It also allows us to derive a “Phillips curve” relation between price movements and output
movements, of a kind familiar from the “New Classical” literature of the 1970s.
Suppose now that a fraction 0 < ι < 1 of the goods prices are fully ¬‚exible “ which is to
say, set each period on the basis of full information about current demand and cost conditions
“ while the remaining 1 ’ ι are set a period in advance, as in the previous subsection. The

supplier of each ¬‚exible-price good will then set its price each period according to (1.12),
while the supplier of each sticky-price good will set its price in advance according to (1.18).
The marginal costs of supplying di¬erent goods i will di¬er only insofar as the quantities
supplied di¬er, and these in turn will di¬er only insofar as the prices of the goods di¬er. It
follows that all ¬‚exible-price goods will have a common price p1t , and all sticky-price goods
will similarly have a common price p2t . We similarly let y1t denote the common equilibrium
output of all ¬‚exible-price goods, and y2t the output of sticky-price goods.
Taking a log-linear approximation to the two pricing equations, we obtain

ˆ ˆ
log p1t = log Pt + ω y1t + σ ’1 Yt ’ (ω + σ ’1 )Ytn ,

ˆ ˆ
log p2t = Et’1 [log Pt + ω y2t + σ ’1 Yt ’ (ω + σ ’1 )Ytn ],
de¬ning yit ≡ log(yit /Y ) for i = 1, 2, and again using (1.14) to approximate the real marginal
cost function. These approximations apply as long as the ¬‚uctuations in pit /Pt , Yt , and Ytn
¯ ¯
around the values (1, Y , and Y respectively) near which we log-linearize are small enough.
Note that up to this log-linear approximation, (log) predetermined prices are set at a value
equal to a constant markup over the conditional expectation, at the time that the price is
set, of (log) marginal cost. Substituting the demand function (1.11) to eliminate the yit
variables, we obtain more simply

ˆ ˆ
log p1t = log Pt + ζ(Yt ’ Ytn ), (1.21)

ˆ ˆ
log p2t = Et’1 log Pt + ζ(Yt ’ Ytn ) ,

ω + σ ’1
ζ≡ > 0. (1.22)
1 + ωθ
These relations imply that up to our log-linear approximation,

log p2t = Et’1 log p1t .

But a corresponding log-linear approximation to the aggregate price index (1.3) yields

log Pt = ι log p1t + (1 ’ ι) log p2t .

It follows that

πt ’ Et’1 πt = log Pt ’ Et’1 log Pt = (log p1t ’ log Pt ).

Then using (1.21), we obtain the “New Classical” aggregate supply relation

ˆ ˆ
πt = κ(Yt ’ Ytn ) + Et’1 πt , (1.23)

κ≡ ζ.
Note that this relation has the form of an “expectations-augmented Phillips curve” of the
kind hypothesized by Phelps (1967) and Friedman (1968), in which the speci¬c in¬‚ation
expectation that is relevant is the expectation at the time at which current predetermined
prices were ¬xed regarding in¬‚ation over the interval until the present. This particular form
of dependence of aggregate supply upon in¬‚ation expectations was stressed in the “New
Classical” literature of the 1970s (e.g., Sargent and Wallace, 1975).
This aggregate supply relation implies once again that

Et’1 log Yt = Et’1 log Ytn ,

so that the component of output that can be forecasted a period in advance is still indepen-
dent of monetary policy. However, unexpected variations in aggregate demand now give rise
to in¬‚ation variation as well, rather than a¬ecting only output. Again taking the stochastic
process for aggregate nominal spending Yt as given, (1.23) implies that aggregate output
will equal
log Yt = log Ytn + (1 + κ)’1 (log Yt ’ Et’1 log Y t ), (1.24)

so that the aggregate price level will equal

log Pt = (Et’1 log Y t ’ log Ytn ) + (log Yt ’ Et’1 log Yt ).

How does the degree of price ¬‚exibility a¬ect the impact of ¬‚uctuations in nominal
spending upon real activity? Not surprisingly, a larger number of ¬‚exible prices (larger ι)

implies a higher value of κ (steeper short-run Phillips curve), and hence a smaller value of
the elasticity of output with respect to unexpected variations in nominal spending in (1.24).
In the limit as ι approaches one (all prices ¬‚exible), κ becomes unboundedly large, and
the output e¬ects of variations in nominal spending approach zero, as found earlier. More
interesting is the question whether intermediate values of ι result in e¬ects more like those
of the fully-¬‚exible limit or the fully-sticky limit. This turns out to depend upon the size of

Figure 3.1 plots the elasticity 1/(1 + κ) occurring in (1.24) as a function of ι, for each
of several values of ζ. In each case, the function is monotonically decreasing, as one would
expect, and takes the same values in the two limiting cases of ι = 0 and ι = 1. However, when
ζ > 1, the function is convex (and more so the larger is ζ), while when ζ < 1, the function
is concave (and more so the smaller is ζ). Thus if ζ is large, that even some goods have
¬‚exible prices is enough to prevent variations in nominal spending from having much e¬ect
on aggregate output. In such a case, the ¬‚exible-price model would provide a reasonable
approximation to the evolution of aggregate prices and quantities. But if ζ is small, that
even some prices are sticky is enough to result in a substantial e¬ect of variations in nominal
spending upon output. Indeed, in such a case, the overall price index responds very little to
unexpected variations in nominal spending, and the simple model in which all prices were
assumed to be predetermined is a reasonably good guide to aggregate outcomes.

There is a simple intuition for the signi¬cance of the parameter ζ: it describes the degree
of strategic complementarity between the price-setting decisions of the suppliers of di¬erent
goods. Let us consider the following simple game among price-setters: aggregate nominal
spending Yt is given, and the suppliers of individual goods i are to simultaneously choose the
prices pt (i) for their goods. In what way does a given supplier™s optimal price depend upon
the level of the prices chosen by the other suppliers? We shall say that the pricing decisions
are strategic complements if an increase in the prices charged for other goods increases the
price that it is optimal to charge for one™s own good. Correspondingly, one may speak of
strategic substitutes if an increase in the other prices makes it optimal for one to reduce the



ζ = 0.1


ζ = 0.5





ζ = 10


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 3.1: Real e¬ects of variation in nominal spending, for alternative degrees of strategic
complementarity in price-setting.

price of one™s own good.13
To determine the answer to this question, we must ¬rst consider what one may call the
notional short-run aggregate supply (SRAS) curve,

p— (i) ˜
= „¦(Yt ; ξt ),

which indicates how each supplier™s desired relative price varies with the level of aggregate
activity. Here the “desired” price p— (i) is the price that the supplier of good i would prefer if

the price of good i could be freely set on the basis of demand and cost conditions at date t,
independent of any prices charged or announced in the past, and with no consequences for
the prices that may be charged at any later dates as well. This is a purely notional supply
See Bulow, Geanakoplos and Klemperer (1985) for the general notion of strategic complementarity, and
Haltiwanger and Waldman (1989) for discussion of it in the context of adjustment to aggregate shocks more

curve in that it need not indicate how the prices of any goods are actually set, in an economy
where prices are sticky; nonetheless, we shall see that the concept is a useful one, even when
no prices are perfectly ¬‚exible. In the context of our present framework, the notional SRAS
curve is given by (1.12). When log-linearized, it takes the form (1.21); thus the coe¬cient ζ
is just the elasticity of the function „¦ with respect to Y . Ball and Romer (1990) refer to an
economy in which this elasticity is small as being characterized by real rigidities.14
Best-response curves for individual price-setters can then be derived by substituting the
identity Yt = Yt /Pt into the notional SRAS curve, to obtain

p— (i) = Pt „¦(Yt /Pt ; ξt ).

The degree of strategic complementarity is then indicated by the partial derivative of the
right-hand side with respect to Pt . We thus ¬nd that strategic complementarity exists15 (a
positive partial derivative) if and only if ζ < 1, while the pricing decisions of the separate
¬rms are strategic substitutes if ζ > 1.
Thus it is the existence or not of strategic complementarity that determines whether
the fraction of the suppliers with sticky prices exert a disproportionate e¬ect upon the
degree of adjustment of the aggregate price index. The intuition for this is simple: if prices
are strategic complements, then the fraction of prices that do not adjust in response to a
disturbance to nominal spending lead even the ¬‚exible-price suppliers to adjust their prices
by less than they otherwise would. If the strategic complementarity is strong enough, there
will be little aggregate price adjustment (and so a large output e¬ect) even if the fraction
The analogy is with “nominal rigidity”, a situation in which the nominal price charged by a given supplier
(rather than its relative price) is insensitive to the quantity that it must supply. Because the nominal SRAS
as de¬ned here is independent of any speci¬cation of the speed at which prices may be adjusted, the degree
to which an economy is characterized by “real rigidities” depends solely upon real factors, the structure of
production costs and of demand. The term, though widely used, is somewhat unfortunate, as it suggests
that there are costs of adjusting real quantities, whereas in fact it means a situation in which (notional)
supply is very elastic. We shall thus mainly speak here of “elastic supply” or “strategic complementarity in
price-setting” instead.
Note that the question of whether di¬erent ¬rms™ pricing decisions are strategic complements is only
de¬ned relative to a particular pricing game, in which aggregate nominal spending is taken as given, in-
dependent of the prices chosen by any of the ¬rms. This is a useful question to consider, however, when
analyzing the dynamics of prices and output under an arbitrarily speci¬ed stochastic process for nominal
spending, as we shall repeatedly do in this chapter.

of ¬‚exible-price suppliers is large (though less than one). On the other hand, if prices are
strategic substitutes, then the aggregate price level may adjust nearly in proportion to the
unexpected change in nominal spending, even when many prices are sticky. For in this case,
the fact that some prices do not adjust causes the ¬‚exible-price suppliers to compensate for
this by adjusting their prices even more than they would in a ¬‚exible-price equilibrium. In
the case of a high degree of strategic substitutability (large ζ), this can result in the aggregate
price index adjusting nearly in proportion to the change in nominal spending; but it will not
adjust by quite that amount, else there would be no change in real aggregate demand, and
so no desire on the part of ¬‚exible-price suppliers to change their relative prices.
The possibility of strategic substitutability explains why some authors, such Ohanian et
al. (1996) and Christiano and Eichenbaum (1996), ¬nd in models with sticky prices for only
some goods that the partial price stickiness has little e¬ect upon the aggregate e¬ects of
monetary shocks. These papers assume parameter values that imply an elasticity ζ greater
than one, the case of strategic substitutes. At the same time, the “New Keynesian” literature
of the 1980™s routinely assumed an elasticity ζ < 1, and hence found that pricing decisions
should be strategic complements. This result favored the conclusion that it was plausible to
suppose that nominal rigidities mattered a great deal for the character of short-run responses
to shocks. It is thus worth considering in further detail possible determinants of the degree
of strategic complementarity in pricing, and the plausible size of the elasticity ζ.

1.4 Sources of Strategic Complementarity

In the simple model introduced above, the elasticity of the notional SRAS curve is given by
(1.22). Most early “New Keynesian” literature16 assumes linear utility of consumption. This
corresponds in our notation to the limiting case σ ’1 = 0, in which case (1.22) reduces to
ζ= .
1 + ωθ
Given that we must have θ > 1, this expression implies that ζ < 1, so that the pricing
decisions of di¬erent suppliers are necessarily strategic complements “ the conclusion relied
See, for example, the presentation in Blanchard and Fischer (1989, sec. 8.1),

upon in that literature. However, this conclusion is less obviously true if, more realistically,
we assume a ¬nite value for the intertemporal elasticity of substitution σ.
Indeed, many of the early authors in the 1990™s seeking to incorporate sticky prices
into otherwise standard real business cycle models adopted numerical calibrations (derived
from the RBC literature) that implied strategic substitutability. One reason for this was
the widespread assumption, unlike what we have assumed above, of common economy-wide
factor markets, so that the marginal cost of supply would be equal for all goods i at any
point in time.17 In this case the marginal cost of supplying any good is given by

St = ,
At f (ht )

where wt is the wage paid by all producers for the homogeneous labor input, and ht is the
common labor-capital ratio chosen by all producers. Capital is reallocated among ¬rms
so as to allow all ¬rms to use the same e¬cient labor-capital ratio, even if they produce
di¬erent quantities of their respective goods; this common labor-capital ratio will be given
by ht = f ’1 (Xt /At ), where the common output-capital ratio Xt satis¬es
Xt = yt (i)di.

Finally, the economy-wide wage wt satis¬es (1.6), where the representative household™s labor
supply must equal ht . Marginal supply cost is therefore given by

St (i) = Pt s(Xt , Yt ; ξt ),

independent of the quantity produced of the individual good i, where s is again the function
de¬ned in (1.9).
Using this alternative expression for marginal cost in the derivation of our pricing equa-
tions, we again obtain equations such as (1.21), but in which we now de¬ne

ζ ≡ ω + σ ’1 > 0.
The consequences of this issue for the degree of strategic complementarity in pricing are stressed by
Kimball (1995).

(Note that up to a log-linear approximation, percentage ¬‚uctuations in Xt equal percentage
¬‚uctuations in Yt .) In the real business cycle literature, it is standard to assume that σ = 1.
(See, e.g., Cooley and Prescott, 1995.) Under this assumption, ζ > 1, and pricing decisions
are strategic substitutes, regardless of the (necessarily positive) value assigned to ω.

On the other hand, even if one were to assume linear utility from consumption (the most
extreme possible value for σ), one might easily conclude that pricing decisions should be
strategic substitutes, under the assumption of common factor markets. For example, the
baseline calibration of Chari et al. (2000) implies that ω = 1.25, while σ = 1. As they
assume common factor markets, their parameterization implies a value of ζ = 2.25, and thus
a considerable degree of strategic substitutability. But even if they were to assume linear
utility of consumption (σ ’1 = 0), despite the implausibility of such an extreme value, the
assumed degree of diminishing returns to labor in the production function and the assumed
degree of increasing marginal disutility of work (which together imply their value of ω) would
still imply ζ = 1.25, so that there would still be strategic substitutability. On the other hand,
even with their assumed values for ω and σ, if these authors were to assume speci¬c factor
markets (as in our analysis above), then (given that they also assign the value θ = 10) they
would have found that ζ = .17, using (1.22). This would imply a great degree of strategic
complementarity. Thus it is the assumption of common factor markets that makes the most
crucial di¬erence for the ¬nding of these and other authors that pricing decisions are strategic

The assumption of common factor markets is thus far from innocuous. It is also far from
realistic. Of course, it makes sense that high wages in one part of the economy eventually raise
wages in the rest, as workers migrate from low-wage to high-wage labor markets. Similarly,
it makes sense that high returns to capital in one part of the economy eventually raise the
rental rate for capital services throughout the economy, as capital is shifted to higher-return
uses (if only through the allocation of new investment). A failure to allow for these factor-
price equalization mechanisms makes our model of speci¬c factor markets unrealistic as a
model of the long-run e¬ects upon sectoral marginal costs of a permanent failure of prices

in one sector to adjust.
Nonetheless, the opposite extreme assumption ” that all factor prices are instantaneously
equalized across the suppliers of di¬erent goods ” is also unrealistic. In the short run, it
is not easy for workers to migrate to regions, specialties, or even ¬rms that happen to have
temporarily higher labor demand; and it is even less easy for capital goods, once installed,
to be reassigned to ¬rms with a temporarily high rate of utilization of their capital stock.
This “quasi-¬xed” character of factor inputs allows equilibrium factor prices to vary across
suppliers for some period following a shock that a¬ects them asymmetrically; and it is
mainly these short-run dynamics of factor prices that matter for determining the short-run
dynamics of price adjustment in response to shocks, which are what matter for comparing
the real allocation of resources under alternative monetary policies.18
Factor speci¬city need not be an all-or-nothing assumption. For example, one might
assume (as in Sbordone, 1998) that all suppliers hire the same kind of labor inputs in a
single economy-wide labor market, but that each ¬rm has a production function of the form
yt (i) = At f (ht (i)), where f is strictly concave. This amounts to assuming that each ¬rm™s
allocation of capital goods remains ¬xed, rather than letting capital be reallocated to equalize
its rental rate. In this case, real marginal costs are given by

ˆ ˆ
st (i) = ωp yt (i) + (ωw + σ ’1 )Yt ’ (ω + σ ’1 )Ytn ,
ˆ ˆ

where ωp , ωw > 0 are the two components of ω introduced in (1.15). It then follows that
ω + σ ’1
ζ= > 0. (1.25)
1 + ωp θ
This value of ζ is higher than that implied by (1.22), so that there is less strategic comple-
mentarity than in the case of speci¬c labor markets as well (given a ¬nite Frisch elasticity
of labor supply, so that ωw > 0). But it remains a lower value than is obtained under the
assumption of common factor markets, i.e., instantaneous equalization of the rental price of
In chapter 4, we show how endogenous reallocation of capital across sectors can be added to the model,
so that the rental price of capital must be equalized across sectors in the long run. We nonetheless ¬nd that
in the presence of adjustment costs for investment, the short-run dynamics of marginal supply costs are not
too di¬erent from those predicted by our simple model with a ¬xed allocation of capital to ¬rms.

capital services. For example, Chari et al. (2000) assume a Cobb-Douglas production func-
tion with a labor share of .67, which implies ωp = .5. Thus under their baseline calibration,
if they were to assume a common labor market but speci¬c capital inputs, they would have
obtained ζ = .38, and so a substantial degree of strategic complementarity in price-setting.
There are other reasons as well to believe that the simple model with common fac-
tor markets set out above underestimates the likely degree of strategic complementarity in
price-setting. One that has received some attention in the literature is the possibility that
preferences over di¬erentiated goods need not be of the constant-elasticity (Dixit-Stiglitz)
form (1.2). The result is that the elasticity of demand, and hence the desired markup of
price over marginal cost, need not be independent of the relative quantities produced of
di¬erent goods. In particular, Kimball (1995) shows that if the elasticity of demand is lower
for the products of suppliers who sell more (because they have relatively low prices), this
will increase strategic complementarity, and hence the real e¬ects of variations in nominal
spending. For in the event of a decline in nominal spending, price cuts by ¬‚exible-price
suppliers will lead to an increase in their relative sales (though a reduction in the absolute
quantity sold), because the sticky-price suppliers fail to cut their prices. If this increase in
relative sales leads to less elastic demand for their products, their desired markups will rise,
mitigating the e¬ect upon their desired relative price of the decline in their sales (and hence
reduction in their marginal supply cost); this will cause them to cut their prices less, so that
the decline in their output is greater.
Following Kimball, let us generalize (1.2) by assuming that the consumption aggregate
Ct is implicitly de¬ned by a relation of the form

ψ(ct (i)/Ct )di = 1, (1.26)

where ψ(x) is an increasing, strictly concave function satisfying ψ(1) = 1. (Note that this
reduces to (1.2) if ψ(x) = xθ’1/θ .) The demand curve for good i is then implicitly de¬ned by

yt (i) pt (i)
ψ = ψ (1) , (1.27)
Yt Pt

where the price index Pt for aggregate (1.26) is implicitly de¬ned by
pt (i) ’1 pt (i)
ψ ψ (1) di = 1. (1.28)
Pt Pt

The elasticity of the demand curve (1.27) faced by supplier i is then equal to θ(yt (i)/Yt ),
ψ (x)
θ(x) ≡ ’ . (1.29)
xψ (x)
It follows that the desired markup of price over marginal cost for a ¬‚exible-price supplier is
equal to µ(yt (i)/Yt ), where
µ(x) ≡ . (1.30)
θ(x) ’ 1
Note that in general, neither θ nor µ is a constant as assumed before; but we continue to
assume a function ψ(x) such that θ(x) > 1 for all x in a neighborhood of 1, so that µ(x) > 1
is well-de¬ned, at least in that neighborhood.
The desired relative price of a ¬‚exible-price supplier is then given by
pt (i) ˜
= µ(yt (i)/Yt )s(yt (i), Yt ; ξt ), (1.31)
generalizing (1.12). Log-linearization of this relation in turn yields

ˆ ˆ ˆ ˆ
’ Yt ) + sy (ˆt (i) ’ Ytn ) + sY (Yt ’ Ytn ),
log pt (i) = log Pt + µ (ˆt (i)
y y

where is the elasticity of µ(x) at the value x = 1, and sy and sY are the elasticities of

the real marginal cost function with respect to its ¬rst two arguments respectively. Using a
log-linear approximation to (1.27),

yt (i) = Yt ’ θ(log pt (i) ’ log Pt ),

to eliminate yt (i), and then solving for log pt (i), we obtain a log-linear notional SRAS curve,
and observe that now
sy + sY
ζ= . (1.32)
1 + θ( µ + sy )
(Here the coe¬cient θ refers to θ(1).) For given values of the other parameters, we observe
that a positive ” which means θ(x) decreasing in x, as discussed above ” lowers the

value of ζ, thus increasing the degree of strategic complementarity.

In principle, as Kimball shows, could be an arbitrarily large quantity, and so this factor

could result in ζ being arbitrarily small, regardless of the size of the coe¬cients sy , sY and θ.
(Note that ζ ’ 0 as ’ ∞.) Bergin and Feenstra (2000), however, argue that a plausible

calibration would assign a value of approximately θ = 1, so that is positive but not
µ µ

extremely high. If the desired markup depends upon relative demand for di¬erent goods as
above, one can show that the desired price of supplier i satis¬es
1 θµ
log pt (i) = log St (i) + log Pt ,
1+θ 1+θ
µ µ

regardless of the determinants of nominal marginal cost St (i). Bergin and Feenstra note
that the literature on exchange-rate pass-through typically ¬nds that a devaluation raises
the domestic-currency price of imported goods by only about 0.5 of the percentage of the
devaluation. If one interprets this is as measuring the elasticity of pt (i) with respect to
changes in St (i) (on the assumption that the devaluation does not a¬ect the foreign-currency
marginal cost of supplying the imported goods, or the domestic-currency prices of most goods
equals approximately 1.19
in the domestic price index), then one may conclude that θ µ

In the case of homogeneous factor markets (sy = 0, sY = ω + σ ’1 ), assumed by Bergin
and Feenstra, (1.32) reduces to
ω + σ ’1
ζ= .
1+θ µ
In this case, their suggested value of reduces ζ by a factor of 2 relative to what one would

obtain under the assumption of Dixit-Stiglitz preferences. This is not enough to radically
change one™s views about the real e¬ects of monetary policy (as they conclude), but it is
nonetheless a non-trivial correction. However, in the case of speci¬c factor markets (sy = ω,
sY = σ ’1 ), the correction matters much less. In this case, (1.32) reduces to
ω + σ ’1
ζ= .
1 + θ( µ + ω)
Here setting θ equal to 1 instead of zero does not make such a great di¬erence, since the

term θω in the denominator is now likely to be much greater than one. For example, in the
Bergin and Feenstra also note that this size of markup elasticity would be implied by the translog
speci¬cation of preferences that has been popular in econometric studies of demand.

case of the Chari et al. values for ω, σ and θ discussed above, allowing for the variation in
desired markups only reduces the predicted ζ from 0.17 to 0.16.
Another reason for greater strategic complementarity than is indicated by the baseline
analysis above is the economy™s input-output structure, stressed by Basu (1995). Our above
analysis of marginal supply costs assumes that labor is the only variable factor of production,
ignoring the role of intermediate inputs. While this is a familiar assumption in equilibrium
business-cycle models (it is, for example, routine in the real business cycle literature), it
is far from being literally correct. The production function for “output” as a function of
capital and labor inputs that one typically encounters in such models must be interpreted
as a functional relation between the value added in production (GDP) and primary factor
inputs, rather than a relation between gross output and all factors of production (including
those that are themselves produced). Under certain conditions, and for some purposes, it
su¬ces to model the economy as if this “value-added production function” were the actual
production function of individual producers ” for example, for purposes of predicting the
evolution of real GDP in the context of a ¬‚exible-price, perfectly competitive model, as in
RBC theory. But as Rotemberg and Woodford (1995) note, it is important not to con¬‚ate
gross-output and value-added production functions in the case of a model with imperfect
competition and prices that do not co-move perfectly with marginal cost.
Rotemberg and Woodford (1995) propose a simple gross-output production function of
the form20
At f (ht (i)) mt (i)
yt (i) = min , , (1.33)
1 ’ sm sm
where At f (ht (i)) is the value-added production function (as above), mt (i) is the quantity of
materials inputs used by ¬rm i, and 0 ¤ sm < 1 is a parameter of the production technology
(that can be identi¬ed, for purposes of calibration, with the share of materials costs in the
Basu (1995) and Bergin and Feenstra (2000) instead assume that gross output is a Cobb-Douglas function
of labor and materials inputs. The simple analysis here, however, which has the advantage of nesting our
previous speci¬cation as a limiting case, allows us to reach quite similar conclusions about the consequences
of the input-output structure for the degree of strategic complementarity in pricing in a highly transparent
way. There seems in any event to be no argument other than tractability o¬ered for the Cobb-Douglas
speci¬cation used in these other papers.

value of gross output). The argument mt (i) refers to the number of units of the aggregate
de¬ned in (1.2) ” or more generally in (1.26) ” that are purchased for use in producing good
i; it is thus a composite of all of the goods produced in the economy, which are all assumed
equally to be both ¬nal goods and intermediate inputs. If ¬rms allocate their intermediate
input purchases across goods j in a cost-minimizing way, the quantity purchased of good j
for use in producing good i will equal mt (i)(pt (j)/Pt )’θ (or the corresponding generalization
in the case of a non-Dixit-Stiglitz aggregator), and the cost per unit of materials inputs will
be the same price index Pt as for consumption purposes. Total demand for ¬rm i™s output,
given by the sum of ¬nal demand and intermediate-input demand from other producers, will
still be given by (1.5) ” or more generally by (1.27) ” where now aggregate demand Yt is
equal to
Yt = Ct + mt (i)di. (1.34)

It follows from the production function (1.33) that real marginal cost for ¬rm i will equal

st (i) = (1 ’ sm )sV A (i) + sm , (1.35)

where sV A (i) is the real marginal cost of an additional unit of theoretical value-added (the

function of primary input use given by At f (ht (i))), and we have used the fact that the price
of the materials aggregate is Pt . The marginal cost of supplying value-added is given by the
same real marginal cost function derived earlier; for example, in the case of speci¬c factor
markets, it is given by
sV A (i) = s((1 ’ sm )yt (i), Yt ’ sm yt (i)di; ξt ),

where s(y, Y ; ξ) is again de¬ned as in (1.9). The existence of a symmetric steady state then
requires that sm < µ’1 , where the latter quantity represents the steady-state level of real
marginal cost, given ¬rms™ desired markup of µ > 1.21 Such a steady state involves a constant
level of ¬nal-goods demand (or real value added) Y implicitly de¬ned by22
1 ’ µsm
s(Y , Y ; 0) = . (1.36)
µ(1 ’ sm )
In the case that the desired markup is variable, we continue to use the coe¬cient µ to refer to the
steady-state value, µ(1).

Regardless of the factor market structure, we ¬nd that a log-linear approximation to the
real marginal cost function for value added is given by

sV A (i) = sy yt (i) + sY Yt ,
ˆt ˆ

where the elasticities sy and sY are the same as earlier. Log-linearization of (1.35) then
implies that
st (i) = (1 ’ µsm )(sy yt (i) + sY Yt ).
ˆ ˆ

It is the appearance of the multiplicative factor 1 ’ µsm < 1 here that explains how the
economy™s input-output structure gives rise to “real rigidity” of the sort that increases the
strategic complementarity among pricing decisions. Substitution of the demand curve for
yt (i) as before then allows us to solve once again for ¬rm i™s desired relative price; we now
¬nd that
(1 ’ µsm )(sy + sY )
ζ= , (1.37)
1 + θ[ µ + (1 ’ µsm )sy ]
generalizing (1.32).
We observe that allowing for a positive materials share lowers ζ, for given values of
the other parameters; and once again, this e¬ect could in principle result in an arbitrarily
small value for ζ, regardless of the values of the other parameters. (Note that ζ ’ 0 as
sm ’ µ’1 < 1.) However, the share of materials costs in total costs for U.S. manufacturing
sectors is typically on the order of 50 or 60 percent; this suggests that a reasonable calibration
would be on the order of µsm = 0.6.23 In the case of homogeneous factor markets (sy = 0),

. 1
( 4)