Sticky Prices Versus Monetary Frictions: An Estimation of Policy Trade-Offs

Sticky Prices Versus Monetary Frictions:
An Estimation of Policy Trade-os

By S. Bora gan Aruoba and Frank Schorfheide
We develop a two-sector monetary model with a centralized and decen-

tralized market. Activities in the centralized market resemble those in
a standard New Keynesian economy with price rigidities. In the decen-
tralized market agents engage in bilateral exchanges for which money is
essential. This paper is the rst to formally estimate such a model, eval-
uate its t based on postwar U.S. data, and assess its money demand
properties. Steady state welfare calculations reveal that the distortions
created by the monetary friction may be of similar magnitude as the dis-
tortions created by the New Keynesian friction.
JEL: C5, E4, E5.
Key words: Bayesian Estimation, Long-run Ination Target, Monetary
Models, Welfare Costs of Ination.
This paper develops a dynamic stochastic general equilibrium (DSGE) model that
bridges the gap between the search-based monetary theory initiated by (Nobuhiro Kiy-
otaki and Randall Wright 1989) and the literature on estimable New Keynesian DSGE
models laid out in the textbook by (Michael Woodford 2003). Contrary to popular
belief, it is fairly straightforward to combine interesting elements of the monetary micro-
foundations literature with New Keynesian models and create empirical models that can
be confronted with the data and used to study important substantive questions. In our
model, following the basic structure of (Ricardo Lagos and Randall Wright 2005), hence-
forth LW, and (S. Boragan Aruoba, Christopher J. Waller and Randall Wright 2008),
henceforth AWW, in every period economic activity takes place in two markets. In
a decentralized market (DM), households engage in bilateral trade, with a fraction of
households producing and a fraction of households consuming. The centralized market
(CM) resembles a standard DSGE model with admittedly reduced-form nominal rigidi-
ties, where production is carried out by rms. Demand for money arises because the
particular frictions in the decentralized market necessitate the facilitation of transactions
by a medium of exchange. We represent monetary policy by an interest rate feedback rule,
and introduce stochastic disturbances to technology, preferences, government spending,
Aruoba: Department of Economics, University of Maryland, College Park, MD 20742,

[email protected]. Schorfheide: Department of Economics, University of Pennsylvania and NBER,
Philadelphia, PA 19104. [email protected]. Previous versions of this paper circulated as Insights
from an Estimated Search-Based Monetary Model with Nominal Rigidities. We thank Randy Wright for
valuable discussions and two anonymous referees, Sanjay Chugh, Kai Christoel, Peter Ireland, Ricardo
Lagos, Chris Otrok, Giorgio Primiceri, Chris Sims, Alex Wolman, as well as seminar participants at
various institutions and conferences for helpful comments. Maxym Kryshko provided excellent research
assistance. Schorfheide gratefully acknowledges nancial support from the National Science Founda-
tion under Grant SES 0617803. Gauss Programs that implement the empirical analysis are available at
http://www.ssc.upenn.edu/schorf.
1
2 AMERICAN ECONOMIC JOURNAL
and monetary policy to make the model amenable to econometric estimation methods.
While the structure of our model to a large extent resembles that of a canonical New
Keynesian model with capital, the presence of the decentralized market provides a micro-
founded motive for holding money and creates a non-separability between consumption
and the value of real money balances.
The proposed model is estimated and evaluated using post-war U.S. data on output,
ination, interest rates, and inverse M1 velocity. While most of the work on search-
based monetary models has been theoretical, we use the Bayesian techniques surveyed by
(Sungbae An and Frank Schorfheide 2007) to conduct a full-edged econometric analysis.
A novel feature of our estimation is that we construct a measure of the target ination
rate from low frequency dynamics of ination as well as ination expectations and then
use this series along with output, ination, interest rates, and velocity to estimate the
DSGE model. To assess the t of the search-based DSGE model we also estimate a
vector autoregression (VAR) and a standard New Keynesian model in which real money
balances enter the households utility function (MIU) in a separable fashion and conduct
a detailed comparison.
Overall time series t in a Bayesian framework is summarized by the so-called marginal
likelihood, which approximately takes the form of an in-sample goodness-of-t measure
penalized for the number of estimated parameters. In terms of marginal likelihoods both
the search-based DSGE model and the MIU model are clearly dominated by the VAR.
This outcome is not uncommon in the comparison of DSGE models and VARs. The
MIU model fairs somewhat better than the search-based DSGE model. The separable
utility specication of the MIU model implies that to the extent that money balances are
mostly driven by money demand shocks, money can evolve largely independent of output,
ination, and interest rates over the business cycle, which happens to be a feature of our
data. In contrast, the search-based DSGE model that we build features an inherently non-
separable structure and creates a tighter link between the dynamics of money balances
and the other macroeconomic aggregates. As such, while this tighter link is conceptually
appealing, the joint dynamics predicted by the particular model considered in this paper
are not fully consistent with post-war U.S. data.
As explained in (Frank Schorfheide 2000), in situations where the structural models
under consideration are dominated by a more richly parameterized VAR in the sense that
the posterior probabilities of the former are essentially zero, it is sensible to evaluate the
DSGE models by comparing some of their specic implications, e.g. population moments
or impulse responses, to those derived from the less restrictive VAR. Under the assump-
tion that agents forecast the target ination rate with a random walk model, we are
able to identify impulse responses to a target ination rate shock in the VAR. Matching
the responses to this shock is desirable since we use the DSGE model subsequently to
examine the eect of target ination changes on welfare. Note that the VAR itself, while
being the better empirical model from a Bayesian perspective, is not suitable to conduct
such a welfare analysis. A comparison of DSGE and VAR impulse responses shows that
the Bayesian estimates of dierent versions of the search-based DSGE model as well as
the MIU model are successful in matching the responses of output, interest rates and
VOL. NO. STICKY PRICES VERSUS MONETARY FRICTIONS 3
ination. However, all of the models can only capture the (small) short-run response but
not the (large) long-run response of velocity. This nding also translates into the failure
of the models to match the short- and long-run elasticities of velocity with respect to
interest rates simultaneously.
Finally, we study the steady state welfare implications of the estimated DSGE model to
determine whether the monetary friction is quantitatively important for policy analysis.
Specically, we compute measures of welfare gain from changing the target ination from
our end-of-sample value of 2.5% to a new value
. Our model incorporates two key

channels through which ination can aect welfare. First, non-zero ination rates lead
to relative price distortions and inecient use of intermediate goods because it is costly
for rms to adjust nominal prices. We label this channel the New Keynesian channel. Its
strength is determined by the probability with which rms are unable to re-optimize their
prices the nominal rigidity in the centralized market is based on the mechanism proposed
by (Guillermo A. Calvo 1983) and the degree to which non-optimizing rms index their
past price to lagged ination. Second, non-zero nominal interest rates constitute a tax
on money holdings and hence depress activity in the decentralized market. We label
this channel the Friedman channel. It is to a large extent controlled by the probability
with which households engage in bilateral exchange in the decentralized market, which in
turn determines the interest elasticity of money demand. Since both the New Keynesian
and the Friedman channels agree that positive target ination rates are undesirable,
we focus on the range of
[2.5%, 0%], where there is a trade-o between the two

channels. According to our parametrization, at
= 2.5% the nominal interest rate is

zero, which is the celebrated Friedman rule. Unlike a cashless New Keynesian model,
which favors a target ination rate of zero, our estimated search-based models imply that
the optimal ination rate lies in the range of 2.5% to 1%. We interpret this nding as
evidence that the distortions created by monetary frictions may be of similar magnitude
as the distortions created by the New Keynesian friction.
Our paper is related to several strands of the literature in monetary economics and the
estimation of DSGE models. While the literature on estimated DSGE models with New
Keynesian features is large see (Frank Schorfheide 2008) for an extensive survey only
very few papers use a measure of money as observable and hence implicitly or explicitly
estimate a DSGE model-implied money demand function. The search-theoretic literature
for the most part has only recently started to conduct quantitative exercises. Our paper
is the rst in applying formal estimation methods to a model based on the LW framework
and examining its money demand properties.
Since we are also analyzing the steady-state welfare properties of our search-based
DSGE model, a few remarks about the literature on optimal monetary policy are in
order. First, most of the policy analysis in the search-theoretic literature focuses on
the optimal long-run monetary policy. In general the Friedman rule is found to be the
optimal policy unless the model features some other frictions, e.g., endogenous partici-
pation in (Guillaume Rocheteau and Randall Wright 2005), credit rationing by banks in
(Aleksander Berentsen, Gabriele Camera and Christopher Waller 2007), or government-
nancing in (S. Boragan Aruoba and Sanjay K. Chugh 2010). The New Keynesian chan-
nel and its potential inuence on monetary policy has not been analyzed in this class of
models before.
Second, there is a large literature on monetary policy analysis in New Keynesian mod-
els. Much of it, as summarized in (Woodford 2003), focuses on stabilization policies,
assuming the absence of steady state distortions. Among the few papers that study the
optimal long-run target ination and consider both New Keynesian and monetary fric-
tions, the following three articles are most closely related to our work. (Robert G. King
and Alexander L. Wolman 1996) nd that once monetary frictions that generate money
demand, shopping-time in their case, are added to the most stripped-down New Key-
nesian model, the Friedman rule is approximately optimal. (Aubhik Khan, Robert G.
King and Alexander L. Wolman 2003) use a framework where the probability of a price
change for a rm depends on the time since last change. The optimal long-run ination
target in their benchmark calibration is 0.75%. (Stephanie Schmitt-Grohe and Martin
Uribe 2007) show that in a medium-scale New Keynesian model, one with more frictions
than ours, and with a transaction cost of consumption to motivate money demand, the
optimal policy is a long-run ination target of 0.5%, but it is very sensitive to changing
the degree of price stickiness. By and large, normative prescriptions derived from our
estimated search-based DSGE model, while obtained under very dierent assumptions
about the demand for money, are consistent with this earlier work which gets a range of
results between the Friedman rule and full price stability.
The remainder of the paper is organized as follows. We provide a detailed derivation
and discussion of the search-based DSGE model in Section I. The Bayesian estimation
results are presented in Section II and the welfare analysis is summarized in Section III.
Finally, Section IV concludes. Detailed derivations as well as additional estimation results
are provided in the online appendix.
I. The Model
The model is an extension of the two-sector model developed in AWW, consisting of a
decentralized and centralized market. To generate price stickiness we replace the perfectly
competitive CM rms by monopolistically competitive rms that are constrained in their
ability to change nominal prices. The centralized market is essentially identical to the
goods market in a standard New Keynesian DSGE model described, for instance, in
(Woodford 2003) with a (Calvo 1983)-style nominal rigidity.
A. Households
There is a continuum of ex-ante identical households in the economy. In every period,
households rst trade in the DM. According to an idiosyncratic taste shock that is realized
at the beginning of the period, households become buyers with probability , sellers with
probability , or non-participants with probability 12. These shocks are independent
across time and across households. Given that there are equal measures of buyers and
sellers, we assume there is an ecient matching technology that matches exactly one buyer
with one seller. The taste shocks create a double-coincidence problem where frictionless
barter cannot occur. AWW show that a search-based setup in which households meet at
random leads to the same mathematical construct. All households are anonymous in this
market which means IOUs will not be accepted in trade. (Kiyotaki and Wright 1989)
show that a double-coincidence problem and the anonymity of households, will make
money essential in the decentralized market, since trade can only happen with a quid
pro quo. The terms of trade in such a match are determined via one of two alternative
schemes: generalized Nash bargaining (B) or price-taking (PT).
1
Our model features two
other durable assets, bonds and capital claims, that in principle can serve as a medium
of exchange. In our benchmark model we follow AWW and assume this possibility away.
In Section II.G we relax this restriction and allow that in addition to money a fraction
of the claims to the capital stock are liquid and may be used to purchase DM goods.
Once the households leave the DM, they proceed to the CM where neither of the two
frictions that create a role for money in the DM is present: the households are identical
in their preferences and abilities and they are not anonymous. Using labor and capital
income, the households acquire the nal goods produced in the CM and use them for
consumption and to accumulate capital. Households also adjust their asset holdings.
We assume that households have access to a set of claims contingent on all possible
realizations of the aggregate states. To characterize the households behavior in this
economy, we start from the households CM problem.
1. Household Activity in the Centralized Market. The households take
as given the aggregate price level P
t
in the CM, the gross nominal interest rate R
t
on
one-period bonds, the wage W
t
, the rental rate of capital R
k
t
, and the set of aggre-
gate shocks S
t
, along with their laws of motion. We use V
CM
t
( m
t
, k
t
, i
t1
, b
t
, S
t
) and
V
DM
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) to denote the period t value functions in the CM and DM, where
m
t
(m
t
) is the money balance of a household entering the CM (the DM), k
t
is its capital
stock, i
t1
is lagged investment, and b
t
denotes its bond holdings. As it is clear from the
notation, S
t
, the only source of uncertainty for the CM, is realized at the beginning of
the period and the households are able to compute the outcomes in the CM when they
are in the DM. The CM problem takes the form
V
CM
t
( m
t
, k
t
, i
t1
, b
t
, S
t
) (1)
= max
xt,ht,mt+1,it,kt+1,bt+1
_
U(x
t
) Ah
t
+E
t
[V
DM
t+1
(m
t+1
, k
t+1
, i
t
, b
t+1
, S
t+1
)]
_
1
From an a priori sense, there is no reason to prefer one pricing mechanism over the other. Historically
Nash bargaining has been used in the search-based literature and only recently - by (Rocheteau and
Wright 2005) - competitive pricing (price-taking) has been introduced. To add to the accumulating
literature on the positive properties of these models and because their normative implications are quite
dierent we chose to include both pricing mechanisms.
subject to the constraints
P
t
x
t
+P
t
i
t
+b
t+1
+m
t+1
P
t
W
t
h
t
+P
t
R
k
t
k
t
+
t
+R
t1
b
t
+ m
t
T
t
+
t
(2)
k
t+1
= (1 )k
t
+
_
1 S
_
i
t
i
t1
__
i
t
. (3)
Here U(x
t
) is the instantaneous utility from consuming x
t
units of the nal good, A is
the disutility of one unit of labor, h
t
is hours worked, T
t
is a nominal lump-sum tax,
t
denotes the total prots the household receives from intermediate good producers,
and
t
is the households net cash-in-ow from trading state-contingent securities. The
assumption of quasi-linear preferences is crucial and leads to a degenerate distribution
of asset holdings at the end of each period. This assumption can be motivated by the
indivisible labor setup of (Richard Rogerson 1988) and it is used in the monetary model of
(Thomas F. Cooley and Gary D. Hansen 1995) as well as in many of the New Keynesian
models discussed in (Woodford 2003).
Equation (3) determines the capital accumulation. The adjustment cost function S(.)
satises properties S(1) = 0, S
(1) = 0 and S
(1) > 0. We adopt the timing convention

that k
t+1
and m
t+1
denote capital and money holdings at the end of period t and do not
depend on period t +1 shocks. The individual state variables ( m
t
, k
t
, b
t
) do not appear in
the households optimality conditions, and thus for any distribution of assets ( m
t
, k
t
, b
t
)
across agents entering the CM, the distribution of (m
t+1
, k
t+1
, b
t+1
) is degenerate.
2
It
can also be shown that V
CM
t
(.) is linear in m
t
which will be important in the DM problem
below.
2. Household Activity in the Decentralized Market. The value of starting
the DM for a household whose taste shock has not been realized yet is given by
V
DM
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) = V
b
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) +V
s
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) (4)
+(1 2)V
CM
t
(m
t
, k
t
, i
t1
, b
t
, S
t
),
where the values of being a buyer and a seller are
V
b
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) =
t
u(q
t
) +V
CM
t
(m
t
d
m
t
, k
t
, i
t1
, b
t
, S
t
) (5)
V
s
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) = c(q
t
, k
t
, Z
t
) +V
CM
t
(m
t
+d
m
t
, k
t
, i
t1
, b
t
, S
t
) . (6)
In this transaction q
t
units of the consumption good are exchanged against d
m
t
units of
currency. A household that consumes q
t
receives utility
t
u(q
t
). The disturbance
t
is
a preference shock for goods produced in the DM. Since money is essential to purchase
DM goods,
t
can also be interpreted as a money demand shock. A seller household in
2
In the absence of investment adjustment costs, this statement is exactly correct. In the presence
of investment adjustment costs, since i
t1
appears in the optimality conditions, we also need i
t1
to
be identical across households. This can be achieved, for example, by assuming i
0
is identical across
households.
the DM experiences disutility e
t
, where e
t
denotes the eort the household exerts to
produce q
t
units of the DM good according to the technology q
t
= Z
t
f(e
t
, k
t
). The total
factor productivity process Z
t
is assumed to be exogenous and common across DM and
CM. We invert the production function to express the level of eort as e = c(q, k, Z),
which appears in (6). The terms of trade are determined via bilateral generalized Nash
bargaining, which is one of the most common schemes in the search literature, or price
taking, which was rst considered by (Rocheteau and Wright 2005).
Bargaining. Exploiting the linearity of the CM value function, and using threat points
that have the agents continuing to the CM, our bargaining problem is
max
q,d
m
_
u(q) U
(X)
d
m
P
_
_
U
(X)
d
m
P
c(q, k
s
, Z)
_
1
s.t. d
m
m
b
,
where U
(X) is the marginal utility of CM consumption and is the bargaining power of

the buyer. The rst term captures the buyers surplus and the second term is the sellers
surplus. We dropped the time subscripts since the bargaining problem is static. Using
the insights of LW and AWW, in any monetary equilibrium d
m
= m
b
, that is the buyer
spends all his money in exchange for some q that the seller produces using his capital
and eort.
Price-Taking. Let p be the DM price level that is taken as given by buyers and sellers.
To ensure the quid pro quo nature of the trade, the value of the goods purchased has
to equal the value of the money transferred from buyer to seller: pq = d
m
. Using this
condition, we write the value functions as
V
b
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) = max
qt
_
t
u(q
t
) +V
CM
t
(m
t
p
t
q
t
, k
t
, i
t1
, b
t
, S
t
)
_
(7)
s.t. p
t
q
t
m
t
V
s
t
(m
t
, k
t
, i
t1
, b
t
, S
t
) = max
qt
_
c(q
t
, k
t
, Z
t
) +V
CM
t
(m
t
+ p
t
q
t
, k
t
, i
t1
, b
t
, S
t
)
_
, (8)
It can be shown that in any monetary equilibrium buyers spend all of their money so
that q = m
b
/ p holds.
B. Firms in the Centralized Market
The setup of the centralized market resembles that of a New Keynesian DSGE model.
Production is carried out by two types of rms in the CM: nal good producers combine
dierentiated intermediate goods. Intermediate goods producing rms are subject to a
(Calvo 1983)-style friction. They hire labor and capital services from the households to
produce the inputs for the nal good producers.
1. Final and Intermediate Goods Producers. The nal good Y
t
in the CM
is a composite made of a continuum of intermediate goods Y
t
(i):
(9) Y
t
=
__
1
0
Y
t
(i)
1
1+
di
_1+
with elasticity of substitution (1 + )/. We constrain [0, ). The nal good
producers buy the intermediate goods on the market, package them into Y
t
units of the
composite good, and resell them to consumers. These rms maximize prots in a perfectly
competitive environment taking P
t
(i) as given, which yields the demand for good i
(10) Y
t
(i) =
_
P
t
(i)
P
t
_
1+
Y
t
.
Combining this demand function with the zero prot condition one obtains the following
expression for the price of the composite good
(11) P
t
=
__
1
0
P
t
(i)
di
_
.
Ination in the CM is dened as
t
= P
t
/P
t1
.
Intermediate goods producers, indexed by i, face the demand function (10) and use a
Cobb-Douglas technology with xed costs F:
(12) Y
t
(i) = max
_
Z
t
K
t
(i)
H
t
(i)
1
F, 0
_
.
The technology shock Z
t
is identical to the one that appears in the DM production
function. Following (Calvo 1983), we assume that rms are only able with probability
1 to re-optimize their price in the current period. A random fraction of the rms
that are not allowed to choose P
t
(i) optimally update their price P
t1
(i) according to last
periods ination rate
t1
, whereas the remaining 1 rms keep their price constant.
3
We treat , the degree of dynamic indexation, as a parameter to be estimated.
For a rm that is allowed to re-optimize its price, the problem is to choose a price
level P
o
t
(i) that maximizes the expected present discounted value of prots in all future
states in which the rm is unable to re-optimize its price. This rm uses the time t value
of a dollar in period t + s for the consumers, to discount future prots. Here we are
considering only the symmetric equilibrium in which all rms that can re-adjust prices
will choose the same P
o
t
(i). The solution of this problem leads to a dynamic relationship
between the optimal price p
o
t
= P
o
t
/P
t
and marginal costs MC
t
(New Keynesian Phillips
3
In most estimated DSGE models it is assumed that the fraction of rms 1 index their past price
by the steady state ination rate. However, in order to preserve the steady state eects of the New
Keynesian distortion if < 1, we do not make such an assumption. The sensitivity of policy analysis to
this assumption has recently been emphasized by (Guido Ascari and Tiziano Ropele 2007).
Curve).
C. Government Spending and National Accounts
In period t, the government collects a nominal lump-sum tax T
t
, spends G
t
on goods
from the centralized market, issues one-period nominal bonds B
t+1
that pay gross interest
R
t
tomorrow and supplies the money to maintain the interest rate rule. The government
satises the following budget constraint every period
(13) P
t
G
t
+R
t1
B
t
+M
t
= T
t
+B
t+1
+M
t+1
.
We assume that government spending G
t
evolves exogenously as specied below.
Adding the households CM budget constraints, the government budget constraint and
the prots of intermediate goods producers we obtain
(14) X
t
+I
t
+G
t
= Y
t
,
which is the resource constraint in the CM. Since there is no savings in the DM, there is
a trivial resource constraint that sets consumption equal to output. The quantity of nal
goods in the CM is related to the total output of the intermediate goods rms according
to
(15) Y
t
=
1
D
t
_
Z
t
K
t
H
1
t
F
, D
t
=
_ _
P
t
(i)
P
t
_
1+
di,
where D
t
measures the extent of price dispersion. Unless P
t
(i) = P
t
for all rms, D
t
will be greater than unity, which in turn implies the economy will produce inside its
production-possibilities frontier. Since we have a model with two sectors, we aggregate
DM and CM output and ination using a Fisher index to obtain a measure of GDP,
Y
GDP
t
, as well as a GDP deator ination,
GDP
t
. Moreover, we use Y
t
to denote total
output across the two sectors measured in terms of the CM good.
D. Monetary Policy
Following authors like (Thomas J. Sargent 1999) and (Robert E. Lucas 2000) we assume
that low frequency movements of ination, such as the rise of ination in the 1970s and the
subsequent disination episode in the early 1980s, can be attributed to monetary policy
changes. Unlike in the learning models considered by (Thomas J. Sargent, Tao Zha and
Noah Williams 2006) or (Giorgio Primiceri 2006), our DSGE models oer no explanation
why monetary policy shifts occur over time and simply assumes a time-varying target
ination rate
,t
. The central bank supplies money to control the nominal interest rate.
Following the setup in (Frank Schorfheide 2005), we assume that it systematically reacts
to ination and output growth according to the rule
(16) R
t
= R
1
R
,t
R
R
t1
exp{
R
R,t
}, R
,t
= (r
,t
)
_
GDP
t
,t
_
1
_
Y
GDP
t
Y
GDP
t1
_
2
,
where r
is the steady state real interest rate, is the gross steady state growth rate of
the economy, and
R,t
is a monetary policy shock. With the exception of the time-varying
ination target
,t
the specication (16) is widely used in the literature on estimable
monetary DSGE models. The parameter
R
captures interest rate smoothing, that is,
within the period the central bank does not fully adjust the nominal rate to the desired
level R
,t
. The coecients
1
and
2
determine how strongly the central bank reacts to
deviations of ination and output growth from their respective target values. Finally, the
monetary policy shock
R,t
reects short-run deviations from the systematic part of the
interest rate feedback rule that are unanticipated from the perspective of the public.
E. Closing the Model
We consider ve aggregate disturbances in our model economy. Z
t
is the random
productivity term that aects production in both CM and DM. g
t
is a shock that shifts
government spending according to
(17) G
t
= (1 1/g
t
) Y
t
.
Government consumption goods are purchased in the centralized market. The money
demand shock
t
shifts preferences for goods produced in the DM. Finally, our model
has two monetary policy shocks:
R,t
is assumed to be serially uncorrelated and captures
short-run shifts in monetary policy, whereas the time-varying ination target
,t
captures
long-run policy changes. We dene

Z
t
= ln (Z
t
/Z
),
t
= ln (
t
/
) and g
t
= ln (g
t
/g
),
where Z
and g
are steady state values of the respective exogenous disturbances.

We assume that these exogenous disturbances evolve according to stationary AR(1)
processes

Z
t
=
z

Z
t1
+
z
z,t
,
t
=

t1
+
,t
and g
t
=
g
g
t1
+
g
g,t
. We
also dene
,t
= ln (
,t
/
), where
is a constant discussed below and

,t
evolves
as a random walk
,t
=
,t1
+
,t
. The innovations are stacked in the vector
t
= [
z,t
,
,t
,
g,t
,
,t
,
R,t
] and are assumed to be independently and identically dis-
tributed according to a vector of standard normal random variables.
The law of motion for the exogenous processes completes the specication of our DSGE
model. The equilibrium conditions are summarized in the online appendix. To solve the
model, we compute the steady state conditional on
,t
= 0 and
= 1.01, corresponding
to an annual ination rate of 4% which is the mean in our sample. Next, we use a log-
linear approximation around this steady state to form a state-space representation that
is used for the Bayesian estimation. While our model implies that the ination target
can move arbitrarily far away from
, in our sample
,t
and
t
are never greater than
10% in absolute value. It should be noted that these deviations are commensurable to
the deviations in a model with a xed target ination rate that is equal to the sample
mean.
4
II. Empirical Analysis
We now turn to the DSGE model estimation. We use a Bayesian approach discussed
in detail in (An and Schorfheide 2007). Our data set and the construction of the target
ination series is described in Section A. Functional forms are specied in Section B and
a description of the prior distribution is provided. Parameter estimates as well as implied
steady states are presented in Section C and the implied model dynamics are analyzed
via variance decompositions and impulse response functions in Section D. We assess the
t of the search-based DSGE model in Section E and discuss the properties of money
demand in Section F. Finally, in Section G we study the sensitivity of key parameter
estimates to some of our modeling choices.
A. Data
Our empirical analysis is based on quarterly U.S. postwar data on aggregate output,
ination, ination expectations, interest rates, and (inverse) velocity of money.
5
Our
estimation sample ranges from 1965:I to 2005:I and we use likelihood functions conditional
on data from 1964:I to 1964:IV to estimate our DSGE model and vector autoregressions
(VARs). As explained in Section I, we assume that the target-ination rate
,t
is time
varying. One could simply treat
,t
as a latent variable in the likelihood-based estimation
of the DSGE model and use the Kalman smoother to obtain ex-post estimates of
,t
based on the observations that are included in the construction of the likelihood function.
We shall deviate from this commonly used approach for two reasons. First, we will assess
the time series t of the DSGE model and the propagation of unanticipated changes in
the target ination rate through a comparison with a VAR. To facilitate this comparison,
it is helpful to treat the target ination rate as observable. Second, from the perspective
4
AWW use a nonlinear solution scheme (projection method with Chebyshev polynomials) with no
shocks and nd that around a reasonable neighborhood of the steady state the decision rules are well-
approximated linearly. (Aruoba and Chugh 2010) use a version of the AWW model with shocks and report
that rst- and second-order linear approximations and non-linear approximations lead to very similar
results. In both of these implementations, the uctuations of ination are comparable to what we have
in this paper. Accumulated evidence from estimating New Keynesian DSGE models, see for example,
(Sungbae An 2007), also suggests that log-linear solution techniques work well for the approximation of
equilibrium dynamics.
5
Unless otherwise noted, the data are obtained from the FRED2 database maintained by the Federal
Reserve Bank of St. Louis. Per capita output is dened as real GDP (GDPC96) divided by civilian
non-institutionalized population (CNP16OV). We take the natural log of this measure and extract a
linear trend and link the deviations from this trend to the stationary uctuations around the deter-
ministic steady state, that our model produces. Ination is dened as the log dierence of the GDP
deator (GDPDEF) and our measure of nominal interest rates corresponds to the federal funds rate
(FEDFUNDS). Money is incorporated as an observable by using inverse M1 velocity. We use the sweep-
adjusted M1S series provided by (Barry Cynamon, Donald Dutkowsky and Barry Jones 2006). The
M1S series is divided by quarterly nominal output to obtain inverse velocity and we relate the natural
logarithm of the resulting series to the log deviations from 100 ln(M
/Y
).
of the agents
,t
can be interpreted as a long-run ination expectation. Hence, we will
incorporate survey expectations in the construction of the
,t
series.
In order to obtain a measure of the ination target we combine three ination expec-
tation measures which are plotted in the top panel of Figure 1: GDP deator ltered
through a one-sided band-pass lter, 1-year and 10-year-ahead ination expectations.
6
Prior to 1986 these three measures of target ination move together very closely. Between
1987 and 1992 the bandpass-ltered ination series is about 1% lower than the ination
expectations. After 1992 the bandpass-ltered ination essentially tracks one-year ahead
survey forecasts, which tend to be slightly lower than the 10-year expectations. To com-
bine the three series we use a small state-space model and extract the common factor
using the Kalman lter. The ltered target ination series
,t
is displayed in the second
panel of Figure 1 together with the GDP deator ination.
7
The dynamics of
,t
are
well approximated by the random walk that the DSGE model agents use to forecast the
target ination rate. Finally, the bottom panel of Figure 1 overlays the federal funds
rate and M1 inverse velocity. According to our theoretical framework, the rise and fall
of the nominal interest rate is to a large extent generated by exogenously changing pref-
erences of monetary policy makers, as reected in
,t
. The postwar U.S. data exhibit
a strong negative correlation between inverse velocity and nominal interest rates that at
least qualitatively resembles a money-demand relationship.
B. Functional Forms, Restricted Parameters, and Priors
We use the following functional forms in our estimation:
u(q) = ln (q +) ln(), U (x) = Bln(x), f (e, k) = e
1
k
,
where is set equal to 1E-4 to make sure the threat point of a buyer in the DM, which
involves q
t
= 0, is well-dened. The parameter B determines the relative weight of the
utility from consuming the CM and DM goods. We use a natural logarithm for both
utility functions and use the same Cobb-Douglas production function as the function
used by the intermediate good producers in the CM. As (Christopher J. Waller 2010)
shows, these are necessary conditions for balanced-growth in this model.
One goal of our empirical analysis is to compare the propagation of shocks and the
steady state welfare implications for various specications of our model. Hence, it is
desirable to normalize and restrict a subset of the model parameters prior to estimation.
The steady states of real GDP, Y
, and the DM preference shock process,
, are nor-
malized to one. The steady state log inverse velocity is xed at the sample mean 0.38.
We xed H
/Y
at 0.03. To obtain this value we computed the sample average of quar-

terly hours worked per capita divided by quarterly real per capita GDP (in 1996$). To
6
Ination expectations are from the Survey of Professional Forecasters with the following exception:
the 10-year ination expectations for the period 1979-1991 are from the Livingston Survey and the Blue
Chip Economic Indicators. All series are provided by the Federal Reserve Bank of Philadelphia.
7
If one regresses the ltered series ,t on the three observed measures, the coecients are 0.57
(
BP
t
), 0.22 (
1y
t
), and 0.23 (
10y
t
).
0
2
4
6
8
10
12
65 70 75 80 85 90 95 00 05
Filtered Target
Survey Expectations (1 year)
Survey Expectations (10 years)
0
2
4
6
8
10
12
65 70 75 80 85 90 95 00 05
Inflation Filtered Target
-.7
-.6
-.5
-.4
-.3
-.2
-.1
.0
0
4
8
12
16
20
65 70 75 80 85 90 95 00 05
M1 Inverse Velocity
Federal Funds Rate
Figure 1. Inflation, Target Inflation, M1-Velocity and Interest Rates
a good approximation, the preference parameters A and B and the steady state level of
technology Z
are determined by the steady state hours, velocity and labor productivity,
respectively.
The DSGE model is log-linearized around the average ination rate in our sample,
which is approximately 4%. We let r
A
be equal to the dierence of the average federal
funds rate and the average ination rate between 1965 and 2005 and set = 1/(1 +
r
A
/400). We set g
= 1.2, which is computed from the average ratio of government

consumption plus investment and GDP. We x the depreciation rate at 0.014. This
value is obtained as the average ratio of xed asset depreciation and the stock of xed
assets between 1959 and 2005.
8
It is well known that the central banks reaction to ination deviations,
1
, is dicult
to identify. Since the primary focus of the paper is not to estimate monetary policy rules,
we set
1
= 1.7. This value is taken from (Schorfheide 2005), who estimated a model
with a regime-switching target ination rate over a similar time period. The steady state
gross growth rate of GDP, which is parameterized by in the interest rate feedback rule,
is set equal to one since we model deviations from the steady state. Finally, we let F = 0
(no xed costs).
Suppose we stack the remaining DSGE model parameters in the vector with elements
i
, i = 1, . . . , k. Our prior distribution for takes the form p() f()
k
i=1
p
i
(
i
).
The marginal densities p
i
(
i
) capture prior information for individual parameters and
are summarized in the rst four columns of Table 1. Following (Marco Del Negro and
Frank Schorfheide 2008), the function f() is used to incorporate beliefs about the steady
state that are functions of multiple parameters. In particular,
f() = exp
_
1
2
_
(I
()/Y
() 0.16)
2
0.005
2
+
(lsh() 0.60)
2
0.01
2
+
(mu
DM
() 0.15)
2
0.01
2
+
(mu() 0.15)
2
0.01
2
__
.
Thus, f() down-weighs the overall prior density at parameter combinations for which
the investment output ratio, the labor share, and the mark-ups in the DM and the overall
economy deviate from 0.16, 0.60, and 0.15, respectively. For the price-taking version of
the search-based DSGE model the mark-up in the decentralized market is zero and we
drop the corresponding term from the function f().
The two remaining preference parameters are related to the search and matching fric-
tions that generate a role for money demand. The probability of a single coincidence
in the DM, is bounded between zero and 0.5 and we use an almost uniform prior on
this interval. As we demonstrate further below, this parameter aects the steady state
velocity and the responsiveness of money demand to changes in the interest rate. In the
bargaining version of our model the parameter measures the bargaining power of the
buyer and aects the mark-up in the decentralized market. Our prior for is indirectly
8
We use NIPA-FAT11 (current cost net stock) and NIPA-FAT13 (current cost depreciation) for xed
assets and consumer durables).
Table 1Prior and Posterior Distributions
Prior Distributions Posterior Distributions
SBM(B) SBM(PT)
Name Domain Density Para (1) Para (2) Mean 90% Intv Mean 90% Intv
Households
[0, 1) Uniform 0.00 1.00 0.95 [0.95, 0.96]
2 [0, 1) Beta 0.40 0.20 0.63 [0.56, 0.70] 0.59 [0.52, 0.66]
Firms
[0, 1) Beta 0.30 .025 0.32 [0.31, 0.34] 0.27 [0.26, 0.28]
IR
+
Gamma 0.15 0.05 0.14 [0.12, 0.16] 0.19 [0.18, 0.21]
[0, 1) Beta 0.60 0.15 0.83 [0.79, 0.87] 0.84 [0.80, 0.88]
[0, 1) Beta 0.50 0.25 0.72 [0.54, 0.91] 0.57 [0.31, 0.82]
S
IR
+
Gamma 5.00 2.50 4.89 [2.50, 7.36] 5.08 [2.42, 7.71]
Central Bank
2
IR
+
Gamma 0.20 0.10 0.86 [0.64, 1.06] 0.83 [0.64, 1.02]
R
[0, 1) Beta 0.50 0.20 0.61 [0.56, 0.66] 0.60 [0.55, 0.65]
R
IR
+
InvGamma 0.50 4.00 0.36 [0.31, 0.41] 0.37 [0.31, 0.42]
R,2
IR
+
InvGamma 1.00 4.00 0.85 [0.63, 1.07] 0.85 [0.62, 1.08]

A,0
IR Normal 0.00 2.00 0.05 [-3.21, 3.26] 0.02 [-3.22, 3.28]
IR
+
InvGamma 0.05 4.00 0.05 [0.04, 0.05] 0.05 [0.04, 0.05]
Shocks
g
[0, 1) Beta 0.80 0.10 0.84 [0.81, 0.88] 0.87 [0.83, 0.90]
g
IR
+
InvGamma 1.00 4.00 1.01 [0.90, 1.11] 1.06 [0.94, 1.16]
[0, 1) Beta 0.80 0.10 0.97 [0.97, 0.98] 0.96 [0.95, 0.97]
IR
+
InvGamma 1.00 4.00 1.80 [1.63, 1.97] 1.88 [1.70, 2.05]
Z
[0, 1) Beta 0.80 0.10 0.83 [0.76, 0.90] 0.83 [0.77, 0.89]
Z
IR
+
InvGamma 1.00 4.00 1.04 [0.90, 1.17] 1.06 [0.91, 1.21]
Notes: Para (1) and Para (2) list the means and the standard deviations for Beta,
Gamma, and Normal distributions; the upper and lower bound of the support for the
Uniform distribution; s and for the Inverse Gamma distribution, where p
IG
(|, s)
1
e
s
2
/2
2
.
determined by f(). Turning to the rms, we use a uniform prior on the indexation
parameter . Our prior for is chosen to be broadly consistent with micro-evidence on
the frequency of price changes. The parameter corresponds to the markup in the cen-
tralized market and is centered at 15%. The prior distributions for
g
,
z
, and
reect
the belief that the government spending (demand) disturbance, the technology shock,
and the DM preference shock are fairly persistent. The priors for the shock standard
deviations were loosely chosen such that the implied distribution of the variability of the
endogenous variables is broadly in line with the variability of the observed series over a
pre-sample from 1959 to 1964.
C. Parameter and Steady State Estimates
Posterior means and 90% credible intervals for the estimated DSGE model parame-
ters are reported in Table 1. The bargaining model is abbreviated as SBM(B) and the
price-taking model as SBM(PT). The estimated single-coincidence probability is around
0.3. We will document in Section E that this estimate captures the fairly low short-run
elasticity of money demand with respect to interest rates in the data. The estimate of
= 0.95 in SBM(B) is strongly inuenced by the prior distribution that favors parameter
values consistent with a mark-up of about 15% throughout the sectors of the economy.
This leads to a DM mark-up of 17% and this along with the 14% mark-up in the CM and
the DM share of 20% matches our target of aggregate markup. To provide a comparison,
in AWW, was calibrated to be around 0.90 using a DM markup of 30% as the target.
Turning to the rms, we observe a number of departures from standard parameter
estimates due to both the two-sector structure of our model and the dierences in pricing
mechanisms in the two sectors. The posterior mean of the CM mark-up is higher in
the price taking model, because the DM mark-up is zero and we are using a fairly tight
prior that implies an economy-wide mark-up of about 15%. Much of the information
about stems from the prior distribution which utilizes information about long-run
averages not included in the likelihood function. The capital share is signicantly
larger in SBM(B) than in SBM(PT). This is due to the holdup problems in SBM(B)
which, everything equal, reduces the steady state capital stock. Since we are using priors
that restrict the investment-output ratio to be approximately 16% in both models, the
hold-up problem present in the bargaining model requires a larger capital share parameter
in the production function.
The estimates of the price-stickiness parameter and the degree of indexation are rel-
atively high in both models, implying an average duration between price re-optimizations
in the CM of about 6 quarters and a dynamic indexation of 60%-70%. Our coecient
estimates would roughly translate into a Phillips curve slope of 0.02 (with respect to
marginal costs) and the coecient on lagged ination would be about 0.42. Compared
to the slope estimates surveyed in (Schorfheide 2008), which range from 1E-3 to about
4, our estimate is fairly small but not unreasonable. Since the degree of indexation is
inherently dicult to identify, the estimates of the coecient on lagged ination reported
in the literature are essentially uniformly distributed over the range 0 to 0.5 and are very
sensitive to auxiliary assumptions about the law of motion of exogenous shocks. Note
that and in our model only aect CM ination dynamics, whereas aggregate ination
is a weighted average of CM and DM ination. We will document subsequently that
ination in the DM lacks persistence and prices in the DM are essentially exible. Thus,
in order to match observed ination dynamics, CM prices need to be more rigid than in
the one-sector model.
9
Since CM rms generate 80% of total production, the probability
9
Our CM specication abstracts from real rigidities that are often introduced in New Keynesian
that a given price cannot be changed is 0.8 0.83 = 0.66. This implies an average du-
ration of 8.8 months between price changes in the aggregate, perfectly in line with other
empirical studies.
The estimates of the parameters that describe the central bank behavior and the evo-
lution of the exogenous shocks are very similar across SBM(B) and SBM(PT). The esti-
mated reaction coecient to output growth is about 0.85 and the interest rate smoothing
parameter is 0.6. The preference shock for DM goods is the most persistent among the
shocks with an autocorrelation of about 0.97. We treat the initial value of the target
ination rate as a parameter that appears as
0,A
in Table 1. Since it is well-known that
interest rate feedback rules tend to t poorly over the period 1979:I to 1982:IV, we allow
the standard deviation of the monetary policy shock over this period (
R,2
) to dier from
the standard deviation for the remainder of the sample (
R
). From an econometric per-
spective this parametrization generates a heteroskedasticity correction for the monetary
policy rule.
D. Dynamics
Variance decompositions for output, ination, and interest rates are reported in Ta-
ble 2. The decompositions are computed for business cycle frequencies ranging from 6 to
32 quarters per cycle. Since the decompositions for SBM(B) and SBM(PT) are very sim-
ilar, we will focus on the bargaining version. Our model was built upon the assumption
that the target ination shock only aects low frequency movements and we nd indeed
that its contribution to business cycle uctuations is essentially zero. Technology shocks
cause about 30% of the output uctuations and the demand or government spending
shocks explain roughly 50%. Technology shocks are also the most important source of in-
ation dynamics and generate 50% of its business cycle movements through marginal cost
uctuations. A key feature of the search-based models is their non-separable structure,
meaning that even under an interest-rate feedback rule, the economy is not insulated
from money demand shocks. These money demand shocks arise from time-varying taste
for the goods produced in the decentralized market and explain around 5% of output
uctuations and about 70% of the cyclical uctuations of real money balances.
Impulse response functions to a technology shock
10
for SBM(B) are depicted in Fig-
ure 2. A positive technology shock decreases current and future expected marginal costs.
As a result the increase in technology on impact creates an immediate decrease in prices
in the DM, which is reected in the response of DM/CM relative price and DM ination.
Due to the rise in productivity, CM and DM production increase on impact. According to
the estimated monetary policy rule, the central bank responds to negative ination and
positive GDP growth by lowering the nominal interest rate. The drop in interest rates
reduces the opportunity costs of holding money and raises the demand for DM goods
and, hence, real money balances. Recall that according to our timing convention time
t real money balances reect end-of-period holdings. As a result of the this increased
models to generate ination persistence.
10
Impulse response functions to other shocks are available in the NBER Working Paper 14870.
Table 2Posterior Variance Decomposition (Business Cycle Freq)
Shock SBM(B) SBM(PT)
Mean 90% Intv Mean 90% Intv
Output
Gov Spending 0.51 [0.43, 0.61] 0.53 [0.42, 0.60]
Money Demand 0.05 [0.03, 0.07] 0.06 [0.03, 0.09]
Monetary Policy 0.12 [0.07, 0.17] 0.12 [0.06, 0.18]
Technology 0.32 [0.23, 0.40] 0.29 [0.21, 0.38]
Target Ination 0.01 [0.00, 0.01] 0.01 [0.00, 0.01]
Ination
Gov Spending 0.18 [0.14, 0.23] 0.17 [0.13, 0.21]
Money Demand 0.01 [0.00, 0.01] 0.01 [0.00, 0.02]
Monetary Policy 0.23 [0.17, 0.28] 0.21 [0.15, 0.25]
Technology 0.50 [0.45, 0.58] 0.51 [0.45, 0.58]
Target Ination 0.08 [0.05, 0.12] 0.10 [0.06, 0.13]
Inverse Velocity
Gov Spending 0.44 [0.38, 0.49] 0.46 [0.40, 0.52]
Money Demand 0.52 [0.46, 0.57] 0.50 [0.44, 0.55]
Monetary Policy 0.02 [0.02, 0.03] 0.03 [0.02, 0.03]
Technology 0.02 [0.01, 0.03] 0.01 [0.00, 0.02]
Target Ination 0.00 [0.00, 0.00] 0.00 [0.00, 0.00]
Real Money Balances
Gov Spending 0.11 [0.07, 0.14] 0.12 [0.08, 0.16]
Money Demand 0.70 [0.65, 0.74] 0.69 [0.63, 0.73]
Monetary Policy 0.13 [0.09, 0.17] 0.13 [0.09, 0.16]
Technology 0.07 [0.05, 0.11] 0.06 [0.03, 0.10]
Target Ination 0.00 [0.00, 0.00] 0.00 [0.00, 0.00]
Notes: Real money balances are measured in terms of the CM good.
demand for DM goods, after period 1, DM ination increases as does CM ination. CM
ination (not shown) has a typical negative hump-shaped response since price adjust-
ments in the CM are subject to the Calvo friction. The DM ination, on the other hand,
reacts instantly to shocks and mimics very closely the changes in the interest rate. We
consider this to be evidence that DM prices are less sticky than CM prices.
11
Output and
consumption in both markets show a hump-shaped response after the shock prolonged
by the expansionary policy of the central bank. Since the technology shock is transitory,
CM and DM output eventually return to their steady state levels.
11
Using a simulation of our model, we nd that the aggregate ination has an autocorrelation of
between 0.34 0.53, which is broadly in line with the data, as it should be. This can be decomposed
into CM ination persistence of between 0.74 0.91 and DM ination persistence of around 0.10.
Figure 2. Impulse Responses to Technology Shock
Notes: The gure depicts pointwise posterior means and 90% credible intervals of impulse
responses for SBM(B) model. Responses of ination and fed funds rate are measured in
percentage points and responses of real output, real money balances, and relative prices
are measured in percentage deviations from the steady state.
E. Model Fit
In order to assess the t of the estimated search-based DSGE model, we will consider
two reference models. The rst reference model is a standard New Keynesian DSGE
model in which real money balances directly enter the utility function (MIU) in an addi-
tively separable manner. The second reference model is a restricted vector autoregression,
in which the target ination rate evolves exogenously. We consider various measures of
relative t, including marginal log likelihood values, in-sample root-mean-squared errors
(RMSE), and discrepancies between the DSGE model and the VAR impulse response
functions.
MIU Model: We construct the MIU model by shutting down the decentralized market
( = 0) in the search-theoretic models described in Section I and adding a real-money
balance term to the households instantaneous utility function:
(18) U
t
= U(x
t
) Ah
t
+

t
1
_
m
t
P
t
A
Z
1/1
_
1
.
The shock
t
captures time-varying preferences for money and the parameter controls
the interest-rate elasticity of money demand. The scaling by A/Z
1/(1)
can be inter-
preted as a re-parameterization of
t
, which has the eect that steady state velocity stays
constant as we change A and Z. To mimic the timing conventions in the search-based
models, we assume that m
t
is the (pre-determined) money stock at the beginning of the
period, and P
t
is the price at which the nal good is sold in period t. A detailed de-
scription of the model and its approximate solution can be found in the online appendix.
For the common parameters, we impose the same restrictions and use the same prior
distributions as in the estimation of SBM(B) and SBM(PT). In addition, we assume that
the parameter is a priori distributed according to a gamma distribution with mean 20
and variance 5. The posterior estimate of is 31.75.
VAR: We collect output, ination, interest rates, and inverse velocity in the 4 1 vector
y
1,t
and the target ination rate in the scalar y
2,t
. Moreover, we let y
t
= [y
1,t
, y
2,t
]
.
We assume that y
t
follows a Gaussian vector autoregressive law of motion subject to the
restrictions that the target ination rate evolves according to a random walk process and
that the innovations to the target ination rate are orthogonal to the remaining shocks.
These restrictions are consistent with the assumptions that underlie our DSGE model
and identify the propagation of unanticipated changes in the target ination. The VAR
takes the form
y
1,t
=
0
+
1
y
t1
+. . . +
p
y
tp
+y
2,t
+u
1,t
(19)
y
2,t
= y
2,t1
+
,t
, (20)
where u
1,t
N(0,
11
) and is independent of
,t
. We estimate the VAR composed
of (19) and (20) with p = 4 using the version of the Minnesota prior described in
(Thomas Lubik and Frank Schorfheide 2006).
12
According to the log marginal likelihoods reported in Table 3, the bargaining version of
the SBM is slightly preferred over the price taking version. A comparison of the RMSEs
suggests that the ranking is mainly due to dierences in the RMSE for inverse veloc-
ity. However, by and large the estimated models produce very similar impulse-response
dynamics which makes it dicult to identify the pricing mechanism for the bilateral
exchange from the aggregate data. The MIU model attains an even larger marginal like-
lihood value than SBM(B). While the MIUs in-sample output predictions are slightly
less precise, the ination, interest rate, and velocity forecasts are more accurate than
those of the search-based models.
12
The Minnesota prior tilts the estimates of the VAR coecients toward univariate unit root repre-
sentations. The hyperparameters are = 0.1, d = 3.1, w = 5, = 1, = 1. Our prior assumes that the
elements of are independently distributed according to N(0,
2
).
Table 3Marginal Data Densities and RMSEs
In-Sample RMSE
Model ln p(Y
T
) Output Ination Interest Inv. Velocity
SBM(B) = 0.32 -998.43 0.81 1.18 1.41 2.17
SBM(PT) = 0.30 -1,007.26 0.83 1.18 1.42 2.32
MIU = 31.8 -949.14 0.86 1.08 1.06 1.43
VAR(4) -924.14 0.85 0.96 0.87 1.31
SBM(B) = 0.06 -1,126.00 0.83 1.08 1.15 3.22
SBM(PT) = 0.06 -1,126.59 0.83 1.08 1.15 3.20
MIU = 5.15 -1,092.52 0.86 1.09 1.07 2.39
Notes: The marginal data densities for all models are computed conditional on the four
observations from 1964:I to 1964:IV that are used to initialize the lags of the VAR.
The RMSEs are computed at the posterior mode and measured as follows: output is in
percentage deviations from the linear trend, inverse velocity is in percentage deviations
from the sample mean, ination and interest rates are in annualized percentages.
The two main dierences between the MIU model and the SBMs are that, rst, the
MIU model only has one sticky-price sector whereas the SBMs are composed of a sticky
price and a exible price sector that are aggregated into GDP. Second, the MIU model
has a separable structure that insulates the economy from money demand shocks. We
will focus on the latter aspect. The estimated value of
in the MIU model is 0.98 and

most of the variation in real money is explained by the highly persistent money demand
shock
t
, which has no eect on output, ination, and interest rates. Thus, the weak
correlation between real money and the other variables in the data allows the money
demand shock in the MIU model to capture real money balance uctuations without
compromising the t for any other variable. In contrast, aggregate output and ination
in the search-based models are not insulated from money demand shocks.
13
(Peter N. Ireland 2004), using likelihood-based methods, nds that when money is
included as observable, U.S. data tend to prefer a separable MIU model to a more general
MIU model where real money balances and consumption enter utility in a non-separable
way. The relative ranking of the MIU model and the search-based models we present
in this paper is consistent with (Ireland 2004)s nding. On the other hand, (Giovanni
Favara and Paolo Giordani 2009) provide VAR evidence that money demand shocks
generate output and ination uctuations. This evidence cannot be reconciled with the
separable MIU model, in which the economy is insulated from money demand shocks.
Our interpretation of these ndings is that non-separable structure in the SBM could
provide a useful starting point for understanding monetary aggregates, but it does suer
13
When we produce the ltered signals shutting down all shocks except for the money demand shocks
(not shown), in the MIU model output, ination and the interest rate are completely at while for our
model there are small but non-trivial uctuations.
from some misspecication. This interpretation is consistent with the marginal likelihood
dierentials of the estimated DSGE models relative to the VAR. The VAR relaxes the
restrictions that the DSGE models place on the joint dynamics of output, ination,
interest rates, and velocity and thereby attains a substantially higher marginal likelihood
than all three estimated DSGE models.
F. Money Demand
As discussed in detail in (Schorfheide 2000), since the VAR attains a better time series
t than the DSGE models, we can use its implications with respect to the propagation of
shocks as a benchmark to assess the search-based models. In a typical VAR, identication
of structural shocks requires restrictions which may not necessarily be in line with the
restrictions imposed by the DSGE model. In our setup, since we use the same law
of motion for the ination target in both the VAR and the DSGE model and treat
it as observable in our DSGE model, the VAR identication scheme ts squarely with
our assumptions in the DSGE model. Figure 3 depicts impulse responses to a target
ination shock that raises ination by about 20 basis points in the long run computed
from the VAR (short-dashed bands show 90% credible sets) and SBM(B)(solid lines show
the posterior mean). The DSGE model restrictions imply that the long-run responses
of ination, nominal interest rates, and target ination are identical. While the long-
run responses of the VAR are unconstrained, the impulse response bands for ination
and interest rates are approximately consistent with a 20-basis-point increase over long
horizons.
The DSGE model is able to reproduce the VAR estimates of the real GDP, ination,
and interest rate response functions. The only striking discrepancy between DSGE model
and VAR responses arises for inverse velocity. According to the VAR, the initial response
of inverse velocity is sluggish, but after 20 periods it has fallen 100 basis points below
its mean value, whereas the model based responses are a magnitude smaller. The VAR
estimates indicate that the interest elasticity of money demand is much lower in the
short-run than in the long-run. According to the posterior mean VAR responses interest
rates increase 5 basis points in the rst period and velocity drops roughly 10 basis points
below its steady state value. After ve periods, the interest rate is up 26 basis points
and velocity is down 58 basis points. In the long-run, after 200 periods (not shown), the
interest rate has increased by 29 basis points, whereas velocity has fallen about 210 basis
points.
The discrepancy between the DSGE model and the VAR can be explained by the
inability of our DSGE model to match both short- and long-run interest elasticities of
money demand. In our model the nominal return on a bond is R
t
, whereas the expected
nominal return on holding money is
IE
t
_

t+1
u
(q
t+1
)
c
q
(q
t+1
, K
t+1
, Z
t+1
)
+ (1 )
_
,
which depends on the realization of the idiosyncratic taste shock as well as the money
Figure 3. Impulse Responses to Inflation Target (,t) Shock
Notes: Figure depicts pointwise posterior 90% credible interval of impulse responses
for VAR (short dashes) and posterior mean responses for SBM(B): estimated (solid)
= 0.06 (long dashes). Responses of ination and fed funds rate are measured in
annualized percentages and responses of real output and inverse velocity are measured in
percentage deviations from the steady state.
demand shock. The term
t+1
u
(q
t+1
) captures the marginal utility of consuming q
t+1
units of the DM good, and c
q
(.) captures the marginal disutility of producing it. Thus,
the smaller the probability of participating in the DM, the more interest rate sensitive
the return to holding money conditional on participating in the DM has to be to equate
the expected returns on bond and money holdings. Since in equilibrium the return to
holding money is (inversely) proportional to money balances, the interest rate elasticity
of money demand has to be decreasing in . It can be shown through a log-linear
approximation that in SBM(PT) both the short- and long-run interest semi-elasticity is
given by R
/(R
1 +).
14
14
Note that for small interest rates, the elasticity with respect to the gross interest rate is equal to
According to our posterior estimates, the interest semi-elasticity is about 3 in SBM(PT).
(Stephen M. Goldfeld and Daniel E. Sichel 1990) estimate the short-run interest semi-
elasticity to be around one. Estimates of the long-run semi-elasticity reported in (Lucas
2000), (James H. Stock and Mark W. Watson 1993), and (Laurence Ball 2001) range
from 5 to 11. Thus, the likelihood-based estimation picks up the low short-run elasticity.
In Figure 3, this is reected in DSGE-model-based velocity responses that are small at all
horizons. Since the interest rate elasticity is important for the strength of the Friedman
channel in our subsequent welfare calculation, we consider a second set of DSGE model
parameter estimates in which we constrain to be approximately 0.06, which raises the
interest semi-elasticity from 3 to 13 in SBM(PT).
Only the estimates of parameters that govern the dynamics of the money demand shock
and the price rigidity in the CM are signicantly aected by restricting . The estimated
persistence of drops slightly and the standard deviation
increases dramatically
because the velocity forecasts are deteriorating. The implied size of the decentralized
market shrinks from 20% to 4% of GDP which yields smaller estimates for and . Less
price rigidity in the CM is needed to capture the same aggregate ination dynamics.
Due to the parameter restriction, the log marginal data density for the two search-based
models drops by more than 100 points (see Table 3) and the in-sample RMSE of inverse
velocity rises from 2.17 to 3.22 for SBM(B) and from 2.32 to 3.20 for SBM(PT). The
RMSEs for output, ination, and interest rates do not change by the same order of
magnitude. Thus, imposing a low value of in the search-based models leads to an
unambiguous deterioration of time series t.
Our estimated MIU model suers from the same problem. Just as the search-based
models, it is unable to match the long-run elasticity of money demand and capture the
VAR implied long-run response of inverse velocity to a target ination rate shock (MIU
responses are not shown in Figure 3). When we re-estimated the MIU model subject to
the restriction = 5.15, which implies an increase of the interest semi-elasticity from 2
to 12, we observe a similar deterioration in t as for SBM(B) and SBM(PT).
15
Figure 3 also depicts the posterior mean impulse response to an ination target shock
from the restricted version of SBM(B) using long-dashed lines. For = 0.06 the initial
response of inverse velocity is almost 100 basis points, which lies outside the VAR credible
interval, while the unrestricted model captures the small short-run response of inverse
velocity. After 20 periods, inverse velocity is about 10 basis points below its steady
state level in the restricted model, which is still small but closer to the VAR credible
interval. The long-run response (not shown in the gure) is about 50 basis points,
whereas the 90% VAR credible interval ranges from 60 to 390 basis points. Thus,
given the restrictions generated by the search-based DSGE models, we can either match
the short-run or the long-run interest rate elasticity of money demand, but not both.
the semi-elasticity with respect to the net interest rate.
15
(Pablo A. Guerron-Quintana 2009) points out the inability of a standard monetary model to match
both elasticities and considers a model where in every period only a fraction of households are able to
re-optimize their money balances to successfully match them.
G. Sensitivity Analysis
Table 4Sensitivity Analysis for Key Parameters, SBM(PT)
2
Specication Sample Mean 90% Intv Mean 90% Intv Mean 90% Intv
Benchmark 1965:I to 2005:I 0.84 [0.80, 0.88] 0.57 [0.31, 0.82] 0.59 [0.52, 0.66]
Estimated
1
1965:I to 2005:I 0.86 [0.82, 0.90] 0.83 [0.71, 0.98] 0.69 [0.61, 0.78]
Latent
,t
1965:I to 2005:I 0.89 [0.87, 0.92] 0.84 [0.71, 0.97] 0.68 [0.60, 0.76]
Constant
1965:I to 2005:I 0.91 [0.87, 0.95] 0.72 [0.46, 0.97] 0.35 [0.29, 0.43]
Constant
1965:I to 1979:IV 0.89 [0.87, 0.91] 0.85 [0.71, 0.99] 0.57 [0.50, 0.63]
Constant
1984:I to 2005:I 0.84 [0.78, 0.90] 0.61 [0.17, 0.99] 0.64 [0.52, 0.79]
Liquid Capital 1965:I to 2005:I 0.83 [0.79, 0.88] 0.54 [0.28, 0.77] 0.59 [0.52, 0.66]
Notes: For convenience, we reproduce the key parameter estimates for the Benchmark
specication reported in Table 1.
In order to examine the robustness of our key parameter estimates that control the
magnitudes of the New Keynesian and the monetary distortion, we re-estimated the
SBM(PT) based on various assumptions about the target ination rate as well as for
dierent subsamples. Moreover, we consider a version of the price taking model in which
a fraction of the capital holdings is liquid and can be used for DM purchases. Since ,
, and are the most important parameters for the magnitude of the two distortions of
interest, we summarize their estimates in Table 4.
As discussed previously, we decided to x the central banks response to ination
deviations at
1
= 1.7. If we estimate this parameter instead, the Markov-Chain seems to
become less stable and the parameter drifts to a value close to one. Given the accumulated
evidence about monetary policy rule coecients, we decided to x the coecient at 1.7 for
our benchmark empirical analysis, which spans the period from 1965 to 2005. The second
row of Table 4 indicates that if
1
is estimated despite the aforementioned problems,
= 0.86 stays roughly the same, and both = 0.83 and = 0.35 increase compared to
the benchmark estimation.
Similar results emerge, if we treat the target ination rate as a latent variable rather
than an observable. If we assume that the target ination rate had been constant between
1965 and 2005, the estimate of increases to about 0.91 and = 0.18 drops, implying
a slightly larger interest elasticity of money demand. Subsample estimates under the as-
sumption that the target ination rate is constant are similar to the full sample estimates
obtained if the target ination rate is treated as latent variable.
As we will discuss in more detail in Section III, larger values of weaken the New
Keynesian distortion and create a greater incentive for the policy maker to choose a target
ination rate that keeps the nominal interest rate strictly below zero. Vice versa, large
values for and the implied lower interest elasticity of money demand, tend to reduce the
monetary distortion and hence the welfare costs associated with positive nominal interest
rates. We veried that on balance the parameter estimates obtained from this sensitivity
analysis tend to push the optimal target ination rate closer to the lower bound of -
2.5%. Thus, the results reported in Section III, to the extent that they are sensitive to
the assumptions made in the benchmark estimation, tend to overestimate the optimal
ination rate.
Finally, since the time period in the model is a quarter, it is likely that agents can
liquidate some of their assets to make purchases in the DM during a period. To investigate
this possibility, we consider an extension of SBM(PT) in which buyers can use a fraction
a of their capital stock holdings k
b
t
to acquire goods in the decentralized market.
16
The
extended model is similar to the one studied by (Ricardo Lagos and Guillaume Rocheteau
2008), with the main exception that in our version capital is used as a factor of production
in the DM. Let d
m
(d
k
) denote the amount of money (capital) transferred from buyer to
seller. The constraint of the buyer is now given by
(21) pq = d
m
+Pd
k
, d
m
m
b
, d
k
ak
b
and the value functions (7) and (8) have to be modied accordingly. The equality in (21)
implies that the value of the purchased goods has to equal the value of the transferred
assets. Here P is the price of the CM good in terms of the currency and is the price of a
unit of capital in this transaction. In equilibrium, is set such that the seller is indierent
between accepting money or capital. This price reects that the seller can only re-balance
her asset portfolio at the end of the decentralized market, which implies that she will earn
the rental rate of capital for d
k
while the CM is open. The two inequalities in (21) imply
that the money and capital used in the transactions cannot exceed the buyers holdings
m
b
and ak
b
, where ak
b
is the fraction of capital that is liquid. The remainder of the model
is identical to the price-taking model described in Section I. A detailed description of the
equilibrium conditions is provided in (S. Boragan Aruoba and Frank Schorfheide 2010).
If the fraction a of liquid capital is small, then there exists an equilibrium in which
money and liquid capital co-exist as a medium of exchange in the DM. In this equilibrium
the buyer spends all her money and liquid capital in the bilateral meeting. In addition
there always exists an equilibrium in which money is not valued. However, since we
are using the model to explain observations from an economy in which money is valued,
we restrict our attention to the monetary equilibrium. Bayesian inference for the liquid
capital model is based on the same prior distribution that we used for SBM(PT). In
addition we have to specify a prior distribution for a. We use a Gamma distribution
centered at 0.05 with standard deviation 0.03. The right tail of this prior distribution
16
We thank John Leahy (co-editor) and Ricardo Lagos for suggesting this extension.
contains values for which the monetary equilibrium does not exist. Hence, we truncate
the joint prior for all model parameters to ensure the existence of a unique rational
expectations equilibrium in which money is valued. The resulting (truncated) marginal
prior for a has a mean of 0.033 and a standard deviation of 0.02.
The liquid capital model leads to a more general money demand function that also in-
cludes the capital stock and the return of holding capital while the DM is open. While this
generalized money demand function can in principle improve the t of the search-based
model, it turned out that our liquid capital specication was empirically not success-
ful. The posterior distribution of a concentrates near zero and its marginal likelihood is
lower than that of the SBM(PT) specication. The remaining parameter estimates are
essentially identical to the ones reported in the last two columns of Table 1.
III. Steady State Welfare Implications
To illustrate the relative magnitude of the monetary distortion and the New Keyne-
sian distortion in the estimated search-based DSGE model, we compute the steady state
welfare eects of changes in the long-run ination target
. The only sources of uncer-

tainty are the realization of the Calvo shock on the rm side and households opportunity
to engage in a bilateral exchange in the DM. All aggregate shocks are set to zero and
hence aggregate outcomes are non-stochastic. We hereby expect to capture the most
important rst-order eects. The social welfare function which places equal weights on
all households, up to a constant is given by
(22) V (
) = [u(q
) c(q
, k
, Z
)] +U(x
) Ah
.
We solve for the percentage change required in x
and consumption in the DM (the q
in u(q
)) to make the households indierent between two economies with dierent steady
state ination rates. We use an annual ination rate of 2.5% as a benchmark, which is
the average ination rate at the end of our sample.
17
The monetary distortion and the New Keynesian distortion constitute opposing chan-
nels through which changes in the long-run ination target aect welfare in our search-
based DSGE model. First, an increase in ination raises the opportunity cost of holding
money, reduces real money balances, and reduces the equilibrium consumption in the
DM, which will directly reduce welfare. Since capital is used as an input to DM pro-
duction, the return to holding capital falls due to the drop in DM consumption, leading
to reduced investment in the CM. This will further depress real activity in the CM,
including consumption. A version of this channel, which we label Friedman channel,
is present in virtually all monetary models and it underlies Friedmans prescription of
a zero percent net nominal interest rate. In traditional models of money demand, the
opportunity cost of holding money can be measured by the area under the money de-
mand curve as rst discussed by (Martin J. Bailey 1956) and subsequently, for instance,
17
We replace (1 g
1
)Y with simply a constant G obtained from the estimations to prevent any

welfare eects coming through this term.
by (Lucas 2000). (Ben Craig and Guillaume Rocheteau 2008) show that this result ex-
tends to search-based models under certain conditions: in the basic LW model without
holdup problems, the area under the money demand curve very closely approximates the
consumption-equivalent welfare measure. Thus, the strength of the Friedman channel in
our model is very sensitive to the interest rate elasticity of money demand, which in turn
depends on the parameter .
Second, our model has a nominal rigidity that prevents a fraction of rms in each
period from choosing their prices optimally. This relative price distortion is captured
by the deviation of D
t
in (15) from unity and moves the economy inside the production
possibility frontier. The welfare loss associated with this distortion becomes more severe
as the steady state ination rate moves away from 0% (in both directions) because it
becomes more costly for rms not to adjust their prices. The magnitude of the distortion
is an increasing function of the probability that rms are unable to re-optimize their
price and it is decreasing in the degree of dynamic indexation . Moreover, monopolistic
competition among intermediate good producers leads to a positive markup in the CM,
given by , and generates an additional distortion by moving the real wage rate away
from the marginal product of labor.
18
We will label this link between long-run ination
and welfare as the New Keynesian channel, discussed more extensively in (Alexander L.
Wolman 2001).
Our discussion makes it clear that the Friedman channel and the New Keynesian chan-
nel have opposing implications for welfare. The welfare loss of ination from the Friedman
channel is eliminated in the steady state if the central banks ination target equals minus
the real interest rate, which is determined in our model by the rate of time preference.
On the other hand, the loss due to the New Keynesian channel is minimized around a
zero percent ination target. When both channels are present, the ination rate that
minimizes the overall distortions may be at either of the two extremes or somewhere
in between. Much of the recent literature that uses cashless New Keynesian models as
tool for policy analysis operates under the premise that the distortion resulting from the
Friedman channel is negligible. We will subsequently show that this is not the case for
our estimated search-based DSGE model.
Figure 4 plots the welfare cost of deviating from the benchmark target of 2.5% ination
for four versions of the model using the posterior mean parameter estimates discussed
in the previous section. In particular, we distinguish between the bargaining (B) and
price-taking (PT) specications and consider versions in which was either estimated
to capture the short-run interest elasticity of money demand (SR) or xed to capture
the long-run interest elasticity (LR). All versions show that some target in the interval
between the Friedman rule and price stability is strictly better than the benchmark
target, with a gain between 0.2%-0.6% of consumption. This is not surprising since all
the channels we identied above agree that positive ination is not desirable. While
welfare costs are fairly steep to the right of 0%, they are quite at to the left of 0%.
18
It is common to use a labor income subsidy to oset the eect of the positive markup. As (Schmitt-
Grohe and Uribe 2007) also note, we nd this arbitrary and refrain from doing so, especially given our
objective of nding the net welfare eect of all the distortions in our model.
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Inflation Target (%)
W
e
l
f
a
r
e

C
o
s
t

(
%

o
f

C
o
n
s
u
m
p
t
i
o
n
)
SBM(B-SR)
SBM(B-LR)
SBM(PT-SR)
SBM(PT-LR)
Figure 4. Steady State Welfare Costs
Notes: Welfare costs of deviating from a 2.5% target in terms of consumption. They
are calculated at the posterior mean parameter estimates of the four models. Negative
numbers correspond to welfare gains.
For some versions, the welfare dierence between the Friedman rule and price stability is
less than 0.05%. Thus, in this target ination region the Friedman channel and the New
Keynesian channel work in opposite directions and their strengths are similar.
As we compare the four DSGE model versions, several simultaneous changes need to
be accounted for. First, the Friedman channel is amplied in the bargaining version of
the search-based model through two hold-up problems that are explained in detail in
AWW: a buyer tries to take advantage of the fact that the sellers capital only yields
a return if it is used for DM production. Vice versa, a seller tries to exploit that the
buyers money only yields utility if it is used to purchase DM goods. As a consequence,
buyers (sellers) bring too little money (capital) to the DM relative to what is socially
optimal. These holdup problems, especially the money holdup problem, become more
severe as ination increases, since higher ination further reduces the benet of holding
money and accumulating capital. Thus, as we replace bargaining by price-taking the
hold-up problems disappear, which reduces the welfare gain of moving to the Friedman
rule. Second, as we move from SR to LR versions, two eects are present. The money
demand curve becomes steeper, which increases the area underneath in the region to the
left of 0% ination. This rotation strengthens the Friedman channel. However, after the
model is re-estimated, the estimates of are only slightly lower, while the estimates of
drop signicantly. The reduced dynamic indexation strengthens the New Keynesian
channel substantially. As a result, the welfare cost curve for the LR versions are lower
(reecting the increased Friedman channel) and more convex (reecting the increased
New Keynesian channel).
To isolate the eects of the holdup channel and the slope of the money demand curve
from that of the New Keynesian channel, Figure 5 plots welfare costs, holding the New
Keynesian channel xed at = 0.81 and = 0.09. We choose these parameters because
among the four sets of estimates reported in Section II.C, these values maximize the
strength of the New Keynesian channel. Figure 5 highlights that the Friedman channel is
larger in the bargaining version than in the price-taking version, and it becomes stronger
as one increases the interest elasticity of money demand. Therefore, returning to Figure 4,
it seems that going from SR to LR, the New Keynesian channel increases in strength
relative to the Friedman channel. We deduce that except for the bargaining (SR) version,
which has a very sharp prediction about welfare in favor of the Friedman rule, the welfare
gains of reducing the target ination rate below 0% are fairly insensitive to the actual
value that is chosen.
We also plot the welfare costs implied by our estimated MIU model in Figure 5 for
= 0.81 and = 0.09. For the MIU model the monetary distortion is clearly dominated by
the New Keynesian distortion and welfare costs are almost symmetric around 0% ination.
Thus, we conclude that it matters how the monetary friction is modelled: based on our
estimated search-based DSGE models there is evidence that for ination rates between
2.5% and 0% the distortion caused by the Friedman channel is not negligible relative
to the New Keynesian distortion.
IV. Conclusion
As an alternative to the commonly used cashless New Keynesian model, or its cash-
lled MIU counterpart, we have developed an estimable DSGE model in which the
presence of a decentralized market creates an incentive for households to hold money and
moneys role as a medium of exchange emerges endogenously. Using postwar U.S. data
on output, ination, interest rates, and (inverse) velocity, we estimate several versions
of our search-based DSGE model and document its empirical t. While the money de-
mand relationships derived from our particular model specications had some diculties
capturing the relationship between monetary aggregates on the one hand and output, in-
ation, and interest rates on the other hand, we view this paper as a promising rst step
in bridging the gap between the theoretical literature on micro-founded models of money
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Inflation Target (%)
W
e
l
f
a
r
e

C
o
s
t

(
%

o
f

C
o
n
s
u
m
p
t
i
o
n
)
SBM(B-SR)
SBM(B-LR)
SBM(PT-SR)
SBM(PT-LR)
MIU
Figure 5. Steady State Welfare with Fixed New Keynesian Channel
Notes: Figure depicts welfare costs xing the parameters of the New Keynesian channel
at = 0.81 and = 0.09.
and the empirical literature on estimable DSGE models and providing empirical insights
that help improving the theoretical models. A steady state analysis of the welfare eects
of ination suggests that empirical versions of micro-founded models of money demand
may pose a challenge to proponents of cashless New Keynesian models: in the vicinity
of a zero-ination steady state, the distortions from monetary frictions may be of similar
magnitude as the distortions created by the New Keynesian friction and hence not be
negligible after all.
REFERENCES
An, Sungbae. 2007. Bayesian Estimation of DSGE Models: Lessons from Second-Order
Approximations. Available at http://www.mysmu.edu/faculty/sungbae/.
An, Sungbae, and Frank Schorfheide. 2007. Bayesian Analysis of DSGE Models.
Econometric Reviews, 26(2-4): 113172.
Aruoba, S. Boragan, and Frank Schorfheide. 2010. An Estimated
Search-Based Monetary DSGE Model with Liquid Capital. Available at
http://www.boraganaruoba.com.
Aruoba, S. Boragan, and Sanjay K. Chugh. 2010. Optimal Fiscal and Monetary
Policy When Money is Essential. Journal of Economic Theory, forthcoming.
Aruoba, S. Boragan, Christopher J. Waller, and Randall Wright. 2008. Money
and Capital: A Quantitative Analysis. Available at http://www.boraganaruoba.com.
Ascari, Guido, and Tiziano Ropele. 2007. Optimal Monetary Policy Under Low
Trend Ination. Journal of Monetary Economics, 54(8): 25682583.
Bailey, Martin J. 1956. The Welfare Cost of Inationary Finance. The Journal of
Political Economy, 64(2): 93110.
Ball, Laurence. 2001. Another Look at Long-Run Money Demand. Journal of Mon-
etary Economics, 47(1): 3144.
Berentsen, Aleksander, Gabriele Camera, and Christopher Waller. 2007.
Money, Credit and Banking. Journal of Economic Theory, 135(1): 171195.
Calvo, Guillermo A. 1983. Staggered Prices in a Utility-Maximizing Framework.
Journal of Monetary Economics, 12(3): 383398.
Cooley, Thomas F., and Gary D. Hansen. 1995. Money and the Business Cycle.
In Frontiers of Business Cycle Research. , ed. T.F. Cooley. Princeton University Press.
Craig, Ben, and Guillaume Rocheteau. 2008. Ination and Welfare: A Search
Approach. Journal of Money, Credit and Banking, 40(1): 89119.
Cynamon, Barry, Donald Dutkowsky, and Barry Jones. 2006. Redening the
Monetary Aggregates: A Clean Sweep. Eastern Economic Journal, 32: 661672.
Del Negro, Marco, and Frank Schorfheide. 2008. Forming Priors for DSGE Mod-
els (and How it Aects the Assessment of Nominal Rigidities). Journal of Monetary
Economics, 55(7): 11911208.
Favara, Giovanni, and Paolo Giordani. 2009. Reconsidering the Role of Money for
Output, Prices and Interest Rates. Journal of Monetary Economics, 56(3): 419430.
Goldfeld, Stephen M., and Daniel E. Sichel. 1990. The Demand for Money. In
Handbook of Monetary Economics. Vol. 1, , ed. Benjamin M. Friedman and Frank H.
Hahn, 299356. Elsevier.
Guerron-Quintana, Pablo A. 2009. Money Demand Heterogeneity and the Great
Moderation. Journal of Monetary Economics, 56(2): 255266.
Ireland, Peter N. 2004. Moneys Role in the Monetary Business Cycle. Journal of
Money, Credit and Banking, 36(6): 969983.
Khan, Aubhik, Robert G. King, and Alexander L. Wolman. 2003. Optimal
Monetary Policy. The Review of Economic Studies, 70(4): 825860.
King, Robert G., and Alexander L. Wolman. 1996. Ination Targeting in a St.
Louis Model of the 21st Century. Federal Reserve Bank of St. Louis Review, , (May): 83
107.
Kiyotaki, Nobuhiro, and Randall Wright. 1989. On Money as a Medium of Ex-
change. The Journal of Political Economy, 97(4): 927954.
Lagos, Ricardo, and Guillaume Rocheteau. 2008. Money and Capital as Compet-
ing Media of Exchange. Journal of Economic Theory, 142(1): 247258. Monetary and
Macro Economics.
Lagos, Ricardo, and Randall Wright. 2005. A Unied Framework for Monetary
Theory and Policy Analysis. The Journal of Political Economy, 113(3): 463484.
Lubik, Thomas, and Frank Schorfheide. 2006. A Bayesian Look at Open Economy
Macroeconomics. In NBER Macroeconomics Annual 2005. Vol. 20, , ed. Mark Gertler
and Kenneth Rogo, 313366. MIT Press, Cambridge.
Lucas, Robert E. 2000. Ination and Welfare. Econometrica, 68(2): 247274.
Primiceri, Giorgio. 2006. Why Ination Rose and Fell: Policymakers Beliefs and US
Postwar Stabilization Policy. Quarterly Journal of Economics, 121: 867901.
Rocheteau, Guillaume, and Randall Wright. 2005. Money in Search Equilibrium,
in Competitive Equilibrium, and in Competitive Search Equilibrium. Econometrica,
73(1): 175202.
Rogerson, Richard. 1988. Indivisible Labor, Lotteries and Equilibrium. Journal of
Monetary Economics, 21(1): 316.
Sargent, Thomas J. 1999. Conquest of American Ination. New Jersey: Princeton
University Press.
Sargent, Thomas J., Tao Zha, and Noah Williams. 2006. Shocks and Govern-
ment Beliefs: The Rise and Fall of American Ination. American Economic Review,
94(4): 11931224.
Schmitt-Grohe, Stephanie, and Martin Uribe. 2007. Optimal Ination Stabiliza-
tion in a Medium-Scale Macroeconomic Model. In Monetary Policy Under Ination
Targeting. , ed. K. Schmidt-Hebbel and F. Mishkin, 125186. Central Bank of Chile.
Schorfheide, Frank. 2000. Loss Function-Based Evaluation of DSGE Model. Journal
of Applied Econometrics, 15(6): 645670.
Schorfheide, Frank. 2005. Learning and Monetary Policy Shifts. Review of Economic
Dynamics, 8(2): 392419.
Schorfheide, Frank. 2008. DSGE Model-Based Estimation of the New Keynesian
Phillips Curve. Federal Reserve Bank of Richmond Economic Quarterly, Fall Issue: 397
433.
Stock, James H., and Mark W. Watson. 1993. A Simple Estimator of Cointegrating
Vectors in Higher Order Integrated Systems. Econometrica, 61(4): 783820.
Waller, Christopher J. 2010. Random Matching and Money in the Neoclassical
Growth Model: Some Analytical Results. Macroeconomic Dynamics.
Wolman, Alexander L. 2001. A Primer on Optimal Monetary Policy with Staggered
Price-Setting. Federal Reserve Bank of Richmond Economic Quarterly, Fall Issue: 27
52.
Woodford, Michael. 2003. Interest and Prices. Princeton University Press.

Sticky Prices Versus Monetary Frictions: An Estimation of Policy Trade-Offs

Uploaded by

Copyright:

Available Formats

Sticky Prices Versus Monetary Frictions: An Estimation of Policy Trade-Offs

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sticky Prices Versus Monetary Frictions: An Estimation of Policy Trade-Offs

Uploaded by

Copyright:

Available Formats

Sticky Prices Versus Monetary Frictions:

An Estimation of Policy Trade-os

We develop a two-sector monetary model with a centralized and decen-

Aruoba: Department of Economics, University of Maryland, College Park, MD 20742,

. Our model incorporates two key

[2.5%, 0%], where there is a trade-o between the two

= 2.5% the nominal interest rate is

(1) > 0. We adopt the timing convention

(X) is the marginal utility of CM consumption and is the bargaining power of

are steady state values of the respective exogenous disturbances.

is a constant discussed below and

, and the DM preference shock process,

at 0.03. To obtain this value we computed the sample average of quar-

= 1.2, which is computed from the average ratio of government

in the MIU model is 0.98 and

. The only sources of uncer-

and consumption in the DM (the q

)Y with simply a constant G obtained from the estimations to prevent any

You might also like