World Scientific

Publishing Co.
Pte. Ltd.
F a c
t ,
F
a
l
l
a
c
y

a
n
d

F
a
n
t
a
s
y
B
u
s
i
n
e
s
s
C
y
c
l
e
s
This page intentionally left blank This page intentionally left blank
NE W J E RSE Y • L ONDON • SI NGAP ORE • BE I J I NG • SHANGHAI • HONG KONG • TAI P E I • CHE NNAI
World Scientifc
F a
c
t
,

F
a
l
l
a
c
y

a
n
d

F
a
n
t
a
s
y
B
u
s
i
n
e
s
s
C
y
c
l
e
s
S u m r u G A l t u g
Koç University, Turkey &
Centre for Economic Policy Research, UK
Library of Congress Cataloging-in-Publication Data
Altug, Sumru.
Business cycles : fact, fallacy, and fantasy / by Sumru G. Altug.
p. cm.
Includes bibliographical references and index.
ISBN-13: 978-981-283-276-4 (hardcover)
ISBN-10: 981-283-276-9 (hardcover)
1. Business cycles. I. Title.
HB3711.A424 2009
338.5'42--dc22
2009034620
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library.
For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.
Typeset by Stallion Press
Email: enquiries@stallionpress.com
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.
Copyright © 2010 by World Scientific Publishing Co. Pte. Ltd.
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
Printed in Singapore.
Wanda - Business Cycles.pmd 1/13/2010, 1:50 PM 1
October 9, 2009 15:12 9in x 6in b808-fm
Preface
How do intellectual disciplines progress? Undoubtedly, the discipline of
economics — and macroeconomics, in particular — is affected by major
changes in economic conditions. The Great Depression greatly influenced the
perceptions of a generation of economists, beginning with Keynes. The oil
shocks of the 1970s and the 1980s affected economists’ views regarding the
sources of macroeconomic fluctuations.
Sometimes, the development of new techniques or new ways of modeling
can also affect the course that a discipline takes. Large-scale computers in
the post-World War II era played an important role in the development of
simultaneous equation models. In recent years, real business cycle (RBC)
analysis has come to provide a flexible and popular approach for examining
macroeconomic phenomena. In 2004, Finn Kydland and Edward Prescott
received the Nobel Prize in Economics, and The Royal Swedish Academy
of Sciences published a report titled Finn Kydland and Edward Prescott’s
Contribution to Dynamic Macroeconomics: The Time Consistency of Economic
Policy and the Driving Forces Behind Business Cycles [204]. The field of
macroeconomics has changed significantly due to Kydland and Prescott’s
contributions.
This book draws upon Kydland and Prescott’s original contribution. I was a
Ph.D. student at the Graduate School of Industrial Administration at Carnegie
MellonUniversity whenKydlandandPrescott’s “Time-to-BuildandAggregate
Fluctuations” article was published in the early 1980s [141]. My thesis was
on estimating the model in the same article. The model was rejected, much
to the delight of macroeconomists of a more Keynesian bent! Yet many felt
that economic models should be subject to formal econometric and statistical
testing. This debate continues to this day.
v
October 9, 2009 15:12 9in x 6in b808-fm
vi Business Cycles: Fact, Fallacy and Fantasy
Kydland and Prescott’s seminal article initiated the school of RBCanalysis.
This literature evolved in different ways. Talented and creative individuals
extended the initial Kydland–Prescott research in different ways. Not content
with the initial rejection of the model, many researchers also pursued the
econometric analysis of RBC models. In recent years, researchers at central
banks have begun using so-called dynamic stochastic general equilibrium
(DSGE) models for policy analysis.
This book attempts to provide an overview of the burgeoning business
cycle literature that, in many ways, reflects my own interests. There have
been a number of excellent publications that have examined different facets
of this literature. The volume by Thomas Cooley [74] can be considered a
primer of RBC analysis and its applications. James Hartley, Kevin Hoover,
and Kevin Salyer’s [116] collection of articles provides a critique of the
calibration approach. Jordi Gali’s [97] recent text articulates an alternative
New Keynesian framework for describing aggregate fluctuations. This book
takes a more eclectic approach, asking some basic questions about RBCanalysis
and summarizing the ongoing controversies surrounding it.
October 9, 2009 15:12 9in x 6in b808-fm
Contents
Preface v
1. Introduction 1
2. Facts 7
2.1. Defining a Business Cycle . . . . . . . . . . . . . . . . 7
2.2. Stylized Facts . . . . . . . . . . . . . . . . . . . . . . . 15
2.3. The Euro Area Business Cycle . . . . . . . . . . . . . . 19
2.4. Is There a World Business Cycle? . . . . . . . . . . . . . 25
2.5. Historical Business Cycles . . . . . . . . . . . . . . . . 28
3. Models of Business Cycles 33
3.1. An RBC Model . . . . . . . . . . . . . . . . . . . . . . 34
3.2. A Numerical Solution . . . . . . . . . . . . . . . . . . 39
3.3. Initial Criticisms . . . . . . . . . . . . . . . . . . . . . 46
3.4. “Puzzles” . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.1. A Model with Indivisible Labor Supply . . . . . . 49
3.4.2. The Productivity Puzzle . . . . . . . . . . . . . 52
3.4.3. Reverse Causality . . . . . . . . . . . . . . . . . 56
3.5. The Source of the Shocks . . . . . . . . . . . . . . . . . 58
3.5.1. Investment-Specific Technological Shocks . . . . 59
3.5.2. Energy Shocks . . . . . . . . . . . . . . . . . . 63
4. International Business Cycles 65
4.1. Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2. The Role of International Risk Sharing . . . . . . . . . . 68
4.2.1. Pareto Optimal Allocations . . . . . . . . . . . . 69
4.2.2. Complete Contingent Claims . . . . . . . . . . 71
vii
October 9, 2009 15:12 9in x 6in b808-fm
viii Business Cycles: Fact, Fallacy and Fantasy
4.2.3. No Asset Trading . . . . . . . . . . . . . . . . . 75
4.2.4. Nonspecialization in Endowments . . . . . . . . 77
4.2.5. Nontraded Goods . . . . . . . . . . . . . . . . 78
4.2.6. Trade in Equity Shares . . . . . . . . . . . . . . 79
4.2.7. Limited Risk Sharing . . . . . . . . . . . . . . . 81
4.3. Other Extensions . . . . . . . . . . . . . . . . . . . . . 87
4.4. Puzzles Revisited . . . . . . . . . . . . . . . . . . . . . 90
5. New Keynesian Models 93
5.1. The Basic Model . . . . . . . . . . . . . . . . . . . . . 94
5.2. Empirical Evidence . . . . . . . . . . . . . . . . . . . . 99
6. Business Cycles in Emerging Market Economies 101
6.1. A Small Open-Economy Model of Emerging Market
Business Cycles . . . . . . . . . . . . . . . . . . . . . . 102
6.2. Do Shocks to Trend Productivity Explain Business Cycles
in Emerging Market Economies? . . . . . . . . . . . . . 106
7. Matching the Model to the Data 109
7.1. Dynamic Factor Analysis . . . . . . . . . . . . . . . . . 110
7.1.1. Measures of Fit for Calibrated Models . . . . . . 113
7.1.2. Other Applications . . . . . . . . . . . . . . . . 116
7.2. GMM Estimation Approaches . . . . . . . . . . . . . . 117
7.3. The Calibration versus Estimation Debate . . . . . . . . 119
7.3.1. The Dynamics of Output . . . . . . . . . . . . . 120
7.3.2. Calibration as Estimation . . . . . . . . . . . . . 121
7.3.3. Nonlinearity in Macroeconomic Time Series . . . 123
7.3.4. The Debate Reconsidered . . . . . . . . . . . . 127
7.4. DSGE Modeling . . . . . . . . . . . . . . . . . . . . . 129
8. Future Areas for Research 137
Bibliography 139
Index 151
October 9, 2009 15:12 9in x 6in b808-ch01
Chapter 1
Introduction
In the opening page of the book Business Cycles published in 1927, Wesley
Mitchell [163] comments as thus: “As knowledge of business cycles grows,
more effort is required to master it.” Ever since, there have been many
developments in the field of business cycles. This book describes these new
developments.
Historically, the notion of business cycles originated from various types of
panics, depressions, and crises experienced by market economies in the 19th
and early 20th centuries. According to Karl Marx, one of the most prominent
thinkers of the time, “crises” are an endemic feature of capitalist economies.
As Mitchell [163] recounts, much effort was devoted to understanding the
causes of what many viewed as “abnormal” phenomena. However, other
economists observed that the alternating phases of prosperity and depression
seemed to followeach other on a regular basis, when one examined the history
of commercial cycles for the capitalist economies of the time. In the 1920s,
Kondratiev [137] argued that in addition to shorter economic cycles, there
were periodic movements or “long waves” in economic variables. Schumpeter
[187, 188] sought to explain the existence of such long waves as an outcome
of technological innovations. In his framework, both growth and business
cycles could be ascribed to the process of innovation. He identified three long
waves: 1780–1840, corresponding to the Industrial Revolution; 1840–1890,
corresponding to the introduction of steel and steamengines; and 1890–1950,
corresponding to the invention of electricity, chemical processing, and motor
engines.
While some scholars proposed different theories to explain fluctuations
in economic activity, other scholars investigated methods for the systematic
1
October 9, 2009 15:12 9in x 6in b808-ch01
2 Business Cycles: Fact, Fallacy and Fantasy
measurement andidentificationof business cycles. During this time, Burns and
Mitchell [50] and researchers at the National Bureau of Economic Research
(NBER) began to identify the phenomena of business cycles. According
to them, a business cycle is the simultaneous downturns and upturns of a
large number of economic series. Their work involved the dating of business
cycles and the development of leading indicators for the US economy, and
it continues to this day in the business cycle dating methodology employed
by the NBER.
1
We discuss the NBER methodology and the stylized facts of
business cycles in Chapter 2.
Another important channel which affected the study of business cycles
was the development of statistical and time series methods. Writers such as
FrischandSlutsky embeddedthe notionof business cycles into simple dynamic
systems drivenby stochastic shocks. Their influence onthinking about business
cycles has persisted to this day. The Norwegian economist Ragnar Frisch [95]
developed the notions of impulse and propagation mechanisms for describing
business cycles, and modeled business cycles as the response of a second-order
dynamic system to random shocks. Slutsky [193] argued that the sum of a
number of uncorrelated shocks could produce serially correlated or smooth
movements in the generated series. Their ideas were formulated in terms
of linear time series models, which continue to form the main vehicle for
empirically studying business cycles.
The Great Depression and World War II were two major events in the
development of business cycle analysis. The period following World War II
was an era of high and sustained growth in many countries. Nevertheless, the
lessons of the Great Depressionwere vivid inthe minds of many policy-makers.
After the early work of Burns and Mitchell, post WorldWar II, the focus shifted
to stabilization policy. Keynes’ General Theory [127] laid the foundations
for the analysis of short-run economic fluctuations. Post World War II, the
Keynesian framework was interpreted as a model of output determination
at a point in time. The oil shocks of the 1970s and the experience of high
inflation and high unemployment, or stagflation as it is popularly known, led
researchers to account for the observations using new mechanisms for the
effects of money on output. In their seminal contributions, Phelps [170] and
Lucas [148, 149] developed monetary models of the business cycle as a way
1
See also Zarnowitz [211].
October 9, 2009 15:12 9in x 6in b808-ch01
Introduction 3
of providing a consistent theoretical foundation for describing the impact of
changes in money on output. In Phelps’ and Lucas’ framework, the emphasis
was on generating the Phillips-curve type of phenomena between inflation
and unemployment based on informational frictions. Despite providing
great theoretical advances in the analysis of aggregative phenomena, the
specific mechanisms postulated in this literature as leading to business cycles,
namely, unanticipated shocks to money, failed to garner sufficient empirical
support.
During this period, there was a revival of interest more generally in
examining aggregate economic activity as recurrent phenomena characterizing
the functioning of economies with optimizing agents. In his article
“Understanding Business Cycles”, Lucas [151] cataloged the remarkable
conformity in a set of economic series and set forth an agenda for explaining
these facts using an equilibrium approach. Lucas and Rapping [153] argued
that observed fluctuations in aggregate labor supply could be modeled as
the voluntary response of agents based on intertemporal substitution effects.
Long and Plosser [147] developed a simple Robinson Crusoe economy and
generated many of the characteristics of modern macroeconomic time series
through the mechanisms of substitution and wealth effects in response to
technology shocks affecting different sectors of the economy. The literature
on real business cycle (RBC) theory owes its existence to Kydland and Prescott
[141], who presented a model that featured technology shocks as the main
impulse behind cyclical fluctuations and proposed a rich array of propagation
mechanisms for these shocks, namely, the durability of leisure, time-to-
build in investment, and inventories. They also proposed a methodology
for confronting their theory with the data. Widely known as the calibration
approach, this involves matching a small set of moments implied by the model
with those in the data. Though this approach is cited extensively, the merits
of this approach have been a topic of debate. We will discuss RBC models
further in Chapter 3.
Since this book purports to discuss business cycles, we cannot ignore the
calibration approach or, more generally, the debate on the empirical validation
of business cycle models. One of the major issues with Kydland and Prescott’s
contribution, as identified by skeptics, was identifying technology shocks that
could generate cyclical fluctuations of magnitudes observed in the data (see
Summers [202]). In some sense, this feature of Kydland and Prescott’s analysis
October 9, 2009 15:12 9in x 6in b808-ch01
4 Business Cycles: Fact, Fallacy and Fantasy
was viewed as “fantastical” by many. Other skeptics contested the implications
of the RBC approach for the observed behavior of productivity. Productivity,
or the Solow residual, is known to be procyclical. According to the RBC
approach, the observed procyclical movements in productivity should merely
be a response to exogenous technology shocks (see Prescott [173]). In a series
of papers, Hall [110, 111] argued persuasively that there were most likely
endogenous components to the cyclical movement of productivity arising
from imperfect competition at the firm level and internal increasing returns
to scale in production. A deeper problem lay in modeling the movement of
economy-wide averages (possibly fallaciously) in terms of the behavior of a
representative or stand-in household.
Subsequent research that evolved from the original Kydland–Prescott
exercise has proceeded along several different dimensions. On the one hand,
a plethora of papers have presented modifications to the original Kydland–
Prescott framework to reconcile the model with many of the actual features of
the data. For example, the original Kydland–Prescott model could not explain
the relative variability of hours and productivity or real wages. It also failed
to account for the correlation between hours and productivity. The model
lacked money; hence, it faced the problem of reconciling observations on
money–output correlations within a model where the main driving force was
real productivity shocks. Many of these issues have taken on the character
of “puzzles” in the RBC literature, and they have constituted the topic for
much further study. The RBC literature has also generated international
business cycle models to replicate findings on current account dynamics,
international risk sharing, financial diversification, and international capital
flows. More recently, models have been developed that study the role of
market completeness/incompleteness oncyclical fluctuations (see, for example,
Heathcote and Perri [117]). The original RBC framework has also been
extended in recent years to examine the business cycle phenomena in emerging
market economies. We will examine a few of these directions in later chapters.
A more general critique to RBC analysis was mounted by the New
Keynesian challenge. The New Keynesian viewpoint breaks with the RBC
approach by contesting the view that prices adjust frictionlessly to clear
markets. Instead, it introduces alternative mechanisms for generating price
stickiness such as imperfect competition among firms, markups, endogenous
changes in efficiency due to increasing returns to scale, and variable factor
October 9, 2009 15:12 9in x 6in b808-ch01
Introduction 5
utilization. The New Keynesian challenge has proceeded along theoretical
lines (see Rotemberg and Woodford [182]) and empirical considerations (see
Gali [96] or Basu, Fernald, and Kimball [32]). This facilitates the analysis
of the different effects of government versus technology shocks, or monetary
versus technology shocks. We will discuss New Keynesian models in detail in
Chapter 5.
The controversy over the calibration approach also resulted in work on
alternative methods for empirically analyzing business cycle phenomena. One
approach that is popular in the business cycle literature is the method of
unobservable index models or dynamic factor analysis developed by Sargent
and Sims [185] and others. Following Sims [191], vector autoregression
(VAR) and structural VAR (SVAR) have also proven to be popular in
empirical macroeconomic research. While both models allowfor rich dynamic
interrelationships among a set of endogenous variables and an examination
of business cycle dynamics based on impulse response functions, SVAR also
permits an identification of shocks. More recently, dynamic stochastic general
equilibrium (DSGE) models have been developed to identify shocks and
propagation mechanisms of business cycles models (see, for example, Smets
and Wouters [194, 195]).
2
We will discuss the issues involved in matching the
model with the data in Chapter 7.
2
Canova [58] is an excellent reference source on the quantitative and empirical analysis of dynamic
stochastic general equilibrium models.
October 9, 2009 15:12 9in x 6in b808-ch01
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808-ch02
Chapter 2
Facts
The vast literature onbusiness cycles has focusedongenerating the stylizedfacts
regarding cyclical fluctuations. Mitchell [163], Mitchell and Burns [164], and
Burns and Mitchell [50] provide a framework to describe the main features of
business cycles. This framework is based on the principle of identifying turning
points in economic activity and determining which series constitute leading,
coincident, or lagging indicators of the business cycle. Stock and Watson [197]
present a modern methodology to describe business cycles in terms of the
cyclical time series behavior of the main macroeconomic series and their co-
movement with cyclical output. In this chapter, we describe the National
Bureau of Economic Research (NBER) methodology for dating business cycles
and present the basic facts regarding business cycles.
Much of the early work on business cycles was implemented for the US
economy. However, the European or euro area business cycle has also been the
topic of much recent study (see Artis, Kontolemis, and Osborn [20], Artis and
Zhang [18], and Stock andWatson [199], among others). Basu andTaylor [31]
have examined business cycles in an international historical context. We will
also discuss some of the empirical findings in these regards.
2.1. DEFINING A BUSINESS CYCLE
The notion that market economies are subject to recurring fluctuations in a
large set of variables was formalized by Burns and Mitchell [50] in their 1946
work entitled Measuring Business Cycles as follows:
Business cycles are a type of fluctuation found in the aggregate economic
activity of nations that organize their work mainly in business enterprises;
a cycle consists of expansions occurring at about the same time in many
7
October 9, 2009 15:12 9in x 6in b808-ch02
8 Business Cycles: Fact, Fallacy and Fantasy
economic activities, followed by similarly general recessions, contractions,
and revivals which merge into the expansion phase of the next cycle; this
sequence of changes is recurrent but not periodic; in duration business cycles
vary from one year to ten or twelve years; they are not divisible into shorter
cycles of similar cycles with amplitudes approximating their own.
This definition has formed the basis of modern thinking about business
cycles, whether it pertains to the measurement of business cycles or the
construction of models of cyclical fluctuations.
Burns and Mitchell [50] themselves noted that this definition raised as
many questions as it sought to answer. Some of these questions are precisely
the ones that we seek to answer in this book. If one talks about “fluctuations
in the aggregate economic activity of nations”, then should one worry about
differences in business cycle activity across regions? Should business cycles
be considered in an international context? How about the historical nature of
business cycles? Have business cycles moderated over time? Likewise, when one
considers the statement regarding expansions occurring in “many economic
activities”, how broadly should the aggregates that are being considered be
defined? The notion that changes in economic activity occur “at about the
same time” admit the possibility of economic variables that lead or lag
the cycle. In seeking to identify “recurrent changes”, how should we deal
with seasonal changes, random fluctuations, or secular trends? Finally, the
comments regarding the duration and amplitude of business cycles are based
on actual observations of cyclical phenomena, and also lay down rules for
excluding irregular movements and other similar changes.
The NBER approach to identifying business cycles as outlined by
Mitchell [163], Mitchell and Burns [164], and Burns and Mitchell [50] is
comprised of two mutually reinforcing acts: first, find the cyclical peaks and
troughs in a given set of economic variables; and second, determine whether
these turning points are sufficiently common across the series. If the answer
to the latter question is in the affirmative, then an aggregate business cycle or
a reference cycle is identified. Once the reference dates are found, the cyclical
behavior of each series is then examined relative to the reference cycle. As
part of this analysis, the duration, timing, and amplitude of each specific cycle
are compared with that of the reference cycle. Burns and Mitchell [50] stress
that the notion of a reference cycle should not be equated with an observable
construct. In their words (Burns and Mitchell [50], Ch. 2, p. 14):
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 9
When we speak of ‘observing’ business cycles we use figurative language. For,
like other concepts, business cycles can be seen only ‘in the mind’s eye’. What
we literally observe is not a congeries of economic activities rising and falling
in unison, but changes in readings taken from many recording instruments
of varying reliability.
The NBER business cycle methodology identifies a business cycle based
on the (absolute) downturn of the level of output. This is known as a classical
business cycle. There is an alternative approach which considers the decline
in the series measured as a deviation from its long-run trend. Following the
terminology in Zarnowitz [211], such cycles are known as growth cycles. One
advantage of using growth cycles is that they have expansions and contractions
that are approximately of the same duration. By contrast, classical cycles
typically have recessions that are shorter than expansions because of the growth
effect. Figure 2.1 displays the difference between classical and growth cycles.
Point A defines a trough for a classical cycle while point B defines a peak. By
contrast, a trough occurs at point C for a growth cycle while point D defines a
peak. When the economy is moving from a trough to a peak, we say that it is
Fig. 2.1. Classical and Growth Cycles.
October 9, 2009 15:12 9in x 6in b808-ch02
10 Business Cycles: Fact, Fallacy and Fantasy
in an expansion, and a recession is said to occur when the economy is moving
from a peak to a trough. The duration of the business cycle is the length of
time (in months, quarters, or years) that the economy spends between two
troughs or, equivalently, two peaks. The amplitude of a business cycle is the
deviation from trend.
The dating of business cycles for the US is done formally by the NBER
Business Cycle Dating Committee. This committee uses data on real output,
national income, employment, and trade at the sectoral and aggregate levels
to identify and date business cycles. The turning points are determined
judgmentally, although a computer algorithm exists that can approximate the
results (see Bry and Boschan [49]). As an example, this committee recently
announced that the US economy had formally been in a recession since
December 2007. Table 2.1 gives the dates of business cycles or the so-called
“reference dates” for the US economy since 1857. There are 32 complete cycles
in the entire sample period. The average length of a cycle (peak from previous
peak or trough from previous trough) in the post-WWII era is 67 months
or over five years. The shortest cycle is 17 months or nearly six quarters, and
the longest cycle is 128 months or more than ten years. We can also observe
expansionary and contractionary phases of the business cycle from this table.
Post WWII, the average length of a contraction has decreased to ten months
from 17 months or more in the years preceding 1945. Similarly, the average
length of an expansion has increased to 57 months from, at most, 38 months
pre-WWII. One of the longest expansions to have occurred post-WWII is
between March 1991 and November 2001 — a period of 120 months. This
period was dubbed as the period of the “Great Moderation”.
If one chooses to use growth cycles, there is an issue of how to identify the
cyclical component of a given series. As King, Plosser, Stock, and Watson [133]
argue, real business cycle (RBC) models which allow for trends in the
technology shock imply that growthand business cycles are jointly determined.
Nevertheless, the practice of separating the trend and cyclical component using
linear time series methods is well established. There are several approaches to
de-trending economic time series. One approach is to use a linear de-trending
procedure which assumes that the underlying series possesses deterministic
time trends. An alternative approach is to assume a stochastic trend modeled as
a unit root in the series at hand. The contribution of Nelson and Plosser [167]
was to show that economic time series such as real GDP typically possess
unit roots. However, in their survey of empirical business cycles, Stock and
O
c
t
o
b
e
r
9
,
2
0
0
9
1
5
:
1
2
9
i
n
x
6
i
n
b
8
0
8
-
c
h
0
2
F
a
c
t
s
1
1
Table 2.1. US Business Cycle Expansions and Contractions.
Business Cycle Reference Dates Duration in Months
Peak Trough Contraction Expansion Cycle
Peak to Trough Previous Trough Trough from Peak from
to This Peak Previous Trough Previous Peak
December 1854 (IV) — — — —
June 1857(II) December 1858 (IV) 18 30 48 —
October 1860(III) June 1861 (III) 8 22 30 40
April 1865(I) December 1867 (I) 32 46 78 54
June 1869(II) December 1870 (IV) 18 18 36 50
October 1873(III) March 1879 (I) 65 34 99 52
March 1882(I) May 1885 (II) 38 36 74 101
March 1887(II) April 1888 (I) 13 22 35 60
July 1890(III) May 1891 (II) 10 27 37 40
January 1893(I) June 1894 (II) 17 20 37 30
December 1895(IV) June 1897 (II) 18 18 36 35
June 1899(III) December 1900 (IV) 38 24 42 42
September 1902(IV) August 1904 (III) 23 21 44 39
May 1907(II) June 1908 (II) 13 33 46 56
January 1910(I) January 1912 (IV) 24 19 43 32
January 1913(I) December 1914 (IV) 23 12 5 36
August 1918(III) March 1919 (I) 7 44 51 67
January 1920(I) July 1921 (III) 18 10 28 17
May 1923(II) July 1924 (III) 14 22 36 40
October 1926(III) November 1927 (IV) 13 27 40 41
August 1929(III) March 1933 (I) 43 21 64 34
(Continued)
O
c
t
o
b
e
r
9
,
2
0
0
9
1
5
:
1
2
9
i
n
x
6
i
n
b
8
0
8
-
c
h
0
2
1
2
B
u
s
i
n
e
s
s
C
y
c
l
e
s
:
F
a
c
t
,
F
a
l
l
a
c
y
a
n
d
F
a
n
t
a
s
y
Table 2.1. (Continued)
Business Cycle Reference Dates Duration in Months
Peak Trough Contraction Expansion Cycle
Peak to Trough Previous Trough Trough from Peak from
to This Peak Previous Trough Previous Peak
May 1937(II) June 1938 (II) 13 50 63 93
February 1945(I) October 1945 (IV) 8 80 88 93
November 1948(IV) October 1949 (IV) 11 37 48 45
July 1953(II) May 1954 (II) 10 45 55 56
August 1957(III) April 1958 (II) 8 39 47 49
April 1960(II) February 1961 (I) 10 24 34 32
December 1969(IV) November 1970 (IV) 11 106 117 116
November 1973(IV) March 1975 (I) 16 36 52 47
January 1980(I) July 1980 (III) 6 58 64 74
July 1981(III) November 1982 (IV) 16 12 20 18
July 1990(III) March 1991(I) 8 92 100 108
March 2001(I) November 2001 (IV) 8 120 128 128
December 2007 (IV) 73 81
Average, all cycles:
1854–2001 (32 cycles) 17 38 55 56

1854–1919 (16 cycles) 22 27 48 49
∗∗
1919–1945 (6 cycles) 18 35 53 53
1945–2001 (10 cycles) 10 57 67 67

13 cycles;
∗∗
15 cycles.
Source: NBER.
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 13
Watson [197] argue that neither linear time trends nor first-differencing to
eliminate unit roots provides a satisfactory approach to identifying the cyclical
component of a series. The first approach tends to generate spurious business
cycle effects in the de-trended series, whereas the second exacerbates the role
of short-term noise.
Several alternative approaches have been proposed in the recent business
cycle literature to separate the trend versus cyclical component of a time
series. One of these approaches is the so-called Hodrick–Prescott [120] (HP)
filter, which involves minimizing a quadratic form to determine the trend
component in a given series. Specifically, let y
t
denote the series in question
and g
t
denote the unknown trend component. The Hodrick–Prescott filter
defines g
t
to solve
min
g
t

t =−∞
(y
t
− g
t
)
2

t =−∞

(g
t +1
− g
t
) − (g
t
− g
t −1
)

2
.
The cyclical component of the series is defined as y
c
t
= y
t
−g
t
. The parameter
µ controls the smoothness of the trend component. For quarterly data, µ is
typically recorded as 1600. Taking the derivative with respect to g
t
yields

t =−∞

µg
t +2
− 4µg
t +1
+ (1 + 6µ)g
t
− 4µg
t −1
+µg
t −2

=

t =−∞
y
t
.
(1.1)
Using the lag operator notation and simplifying, the cyclical component can
be represented as
y
c
t
= y
t
− g
t
=
µ(1 − L
−1
)
2
(1 − L)
2
1 +µ(1 − L
−1
)
2
(1 − L)
2
y
t
. (1.2)
In practice, the filter is applied using a finite sample. The general equation
characterizing the filter, Eq. (1.1), is modified at the beginning and end
of the sample. The properties of the HP filter have been studied by many
authors, including Singleton [192], King and Rebelo [130], and Cogley and
Nason [69]. Cogley and Nason have argued, in particular, that business
cycle dynamics obtained by using a HP de-trending procedure depend on
the properties of the underlying data. If these are trend-stationary, then the
de-trending procedure has favorable properties; if the underlying data are
October 9, 2009 15:12 9in x 6in b808-ch02
14 Business Cycles: Fact, Fallacy and Fantasy
difference-stationary, however, application of the HP filter induces spurious
business cycle fluctuations.
Asecond approach is based on the spectral analysis of economic time series.
The band-pass filter developed by Baxter and King [37] “filters” out both the
long-run trend and the high-frequency movements in a given time series while
retaining those components associated with periodicities of typical business
cycle durations, namely, periodicities between six quarters and eight years. The
band-pass filter of Baxter and King [37] is obtained by applying a K th-order
moving average to a given time series:
y

t
=
K

k=K
a
k
y
t −k
, (1.3)
where the moving average coefficients are chosen to be symmetric, a
k
= a
−k
for k = 1, . . . , K . They showthat if the sumof the moving average coefficients
is zero,

K
k=K
a
k
= 0, then it has trend elimination properties. In particular,
they show that the lag polynomial describing the K th-order moving average
can be written as
a(L) = (1 − L)(1 − L
−1
)ψ(L), (1.4)
where ψ(L) is a (K −1)-order symmetric moving average polynomial. Hence,
the Baxter–King filter will eliminate deterministic quadratic trends or render
stationary series that are integrated up to order two, i.e, I (2) or less. The filter
is designed to have a number of other properties, including the property that
the results should not depend on the sample size and that it does not alter the
timing relations between series at any frequency. Baxter and King derive the
band-pass filter by considering low-pass and high-pass filters with the required
properties. Let s denote the number of observations in a year. The band-pass
filter retains the movements in a series for the frequencies associated with
the periodicities of 0.5s and 8s. The frequency response function of the ideal
band-pass filter is defined as
β
bp
(ω) = I (2π/(8s) ≤ ω ≤ 2π/(0.5s)),
where I (·) is the indicator function. It turns out that the time-domain
representation of the band-pass filter is an infinite moving average. However,
once the ideal filter’s weights are found in the time domain, the optimal
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 15
7.2
7.6
8.0
8.4
8.8
9.2
9.6
50 55 60 65 70 75 80 85 90 95 00 05
BPTEST USA
-0.05
-0.04
-0.03
-0.02
-0.01
0.00
0.01
0.02
0.03
0.04
50 55 60 65 70 75 80 85 90 95 00 05
BPCYCLEUSA
Fig. 2.2. Trend and Cyclical Components in USA GDP.
approximating filter with maximum lag K is obtained by truncating the
ideal filter’s weights at lag K . Baxter and King [37] employ a finite-order
approximation to obtain the coefficients of this filter with a truncation of
K = 3 years or K = 12 quarters. In what follows, we use this approach to
generate the cyclical components of the different series.
Figure 2.2 illustrates the logarithm of real GDP for the US and its trend
and components estimated according to the Baxter–King filter for the period
1947(I)–2005(III). We observe that real GDP displays a marked positive
trend over the sample period, showing the large gains in productivity and
growth attained over this period. We also observe the cyclical troughs and
peaks of economic activity associated with the business cycle. We can observe
the cyclical downturns and upturns corresponding to the NBERbusiness cycle
reference dates from this graph. The cyclical component tracks very well the
US business cycle as identified by the NBER. Nevertheless, the notion of a
“business cycle” is also concerned with the co-movement of a large number of
economic variables. Section 2.2 discusses these properties of business cycles.
2.2. STYLIZED FACTS
In this section, we provide some stylized facts of business cycles. These facts
constitute important benchmarks by which to judge the performance of
alternative models of business cycles. We consider a measure of the business
cycle as the co-movement of the cyclical behavior of individual series with
October 9, 2009 15:12 9in x 6in b808-ch02
16 Business Cycles: Fact, Fallacy and Fantasy
the cyclical component of real output. According to this approach, variables
that move in the same direction over the cycle as real output are procyclical.
Variables that move in the opposite direction (rise during recessions and
fall in expansions) are countercyclical. Variables that display little correlation
with output over the cycle are called acyclical. We can also examine whether
different time series are out of phase with real GDP. For example, a leading
indicator reaches a peak before real GDP reaches its peak and bottoms out
(reaches a trough) before real GDP. Leading indicators are useful for predicting
subsequent changes inreal GDP. Coincident indicators reacha peak or a trough
at roughly the same time as real GDP. Finally, lagging indicators reach a peak
or trough after real GDP.
The stylized facts of business cycles are typically defined in terms of the
behavior of the main components of GDP, hours, productivity, real wages,
asset returns and prices, and monetary aggregates. These have been described
in a number of economic studies.
1
Information on leading indicators is also
collected by the Conference Board.
2
The salient facts of a business cycle can be described as follows:
1. Real output across virtually all sectors of the economy moves together.
In other words, the contemporaneous correlation of output in different
sectors of the economy is large and positive. Exceptions are production
of agricultural goods and natural resources, which are not especially
procyclical.
2. Consumption, investment, inventories, and imports are all strongly
procyclical. Consumption of durables is much more volatile than
consumption of nondurable goods and services. Consumption of
1
Stock and Watson [197] use data on 71 variables to characterize US business cycle phenomena over
the period 1953–1996. They make use of the Baxter–King band-pass filter to identify the trend versus
cyclical components of each series. Among other measures, they consider the co-movement of the cyclical
component of GDP with the cyclical component of each series, and the cross-correlations of the cyclical
component of GDP with the cyclical component of each series for lags and leads up to six periods. Backus
and Kehoe [23] analyze the properties of historical business cycles for ten developed countries using a
century-long dataset up to the 1980s. They use the HP filter to derive the cyclical components of the
different series, and separate their sample period into the pre-World War I era, the interwar era between
World War I and II, and the post-World War II era.
2
Early work on developing a composite index of leading indicators is due to Mitchell and Burns [164].
Such a composite index was made official by the US Department of Commerce in 1968, and passed over
to the Conference Board in 1995.
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 17
durable goods fluctuates more than GDP, whereas nondurables fluctuate
considerably less.
3. Investment in equipment and nonresidential structures is procyclical with
a lag. Investment in residential structures is procyclical and highly volatile.
4. Government spending tends to be acyclical. The correlation between
government expenditures and output is nearly zero.
5. Net exports are countercyclical. The correlation with output is generally
negative, but weakly so. Since imports are more strongly procyclical than
exports, the trade balance tends to be countercyclical.
6. Total employment, employee hours, and capacity utilization are all
strongly procyclical. The employment series lags the business cycle by
a quarter, while capacity utilization tends to be coincident.
7. Employment fluctuates almost as much as output and total hours of
work, while average weekly hours fluctuate much less. The implication
is that most fluctuations in total hours result from movements in
and out of the work force rather than adjustments in average hours
of work.
8. Real wages are procyclical or acyclical. They have not displayed a steady
pattern in terms of variability to GDP or in terms of leading, coincident,
or lagging indicators.
9. Productivity is slightly procyclical, but both real wages and productivity
vary considerably less than output.
10. Profits are highly volatile.
11. Nominal interest rates tend to be procyclical. The yield curve which shows
the rates of return on bonds of different maturities tends to be upward-
sloping during an expansion and downward-sloping at the onset of a
recession. That is, an expansion is characterized by expectations of higher
interest rates at longer horizons whereas a recession typically signals a
decline in long-term interest rates relative to short-term rates, namely, an
inverted yield curve.
12. Velocity and the money supply are procyclical.
13. The risk premium for holding private debt, or the yield spread between
corporate paper and Treasury bills with six months’ maturity, tends to
shrink during expansions and increase during recessions. The reason for
this countercyclical behavior is likely to be changes in default risk.
October 9, 2009 15:12 9in x 6in b808-ch02
18 Business Cycles: Fact, Fallacy and Fantasy
14. The stock market is positively related to the subsequent growth rate of real
GDP. In this sense, changes in stock prices have been taken as providing
information about the future course of the real economy. Between 1945
and 1980, the stock market fell in the quarter before each of the eight
recessions, although it is important to emphasize that the market has fallen
without a subsequent recession.
15. Money (M2) is procyclical and tends to be a leading indicator of output.
However, the procyclicality of M2 has diminished since the 1980s.
16. The behavior of prices and inflation appears to have changed over time.
In the pre-WWI period and interwar period, inflation was procyclical
with a very low mean. Since the early 1980s, inflation appears to
be countercyclical. A similar change appears to have characterized the
behavior of price-level fluctuations.
17. The standard deviation of inflation is lower than that of real GDP.
18. Inflation is a coincident indicator.
19. There is also a marked increase in the persistence of inflation after WWII.
20. Finally, contemporaneous correlations between output fluctuations in
different countries were highest in the interwar period, reflecting the
common experience of the Great Depression, with the exception of
Germany and Japan. The correlation is typically larger in the post-war
period than in the pre-war period.
This set of facts constitutes a benchmark which a successful business
cycle model should meet. The findings concerning the behavior of hours,
employment, real wages, and productivity have proved among the most
difficult to reconcile by current business cycle theories. The changes in the
cyclical behavior of prices and inflation also have implications for the so-
called impulses and propagation mechanisms of business cycles. In the post-
WWII era, there is more evidence of supply-side-driven fluctuations in output.
Witness the impact of the large oil shocks in the 1970s, a topic to which we
will return; hence, the increased persistence of inflation and the countercyclical
behavior of prices observed in this era. By contrast, prices and inflation were
procyclical in the pre-WWII era. One way to rationalize this phenomenon
may be in terms of monetary policy that is more accommodative — under a
gold standard, for example. Other papers have generated business cycle facts
using data over long samples. Chadha and Nolan [62] identify the stylized facts
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 19
of business cycles for the UK economy over a long sample period. A’Hearna
and Woitek [2] establish the facts of business cycles in the late 19th century
using spectral techniques.
The above findings can also shed light on whether output fluctuations
have moderated in the post-WWII era. On the one hand, Christina Romer
[180, 181] has argued that the finding of lower output variability in the post-
WWII period is due to changes in the measurement of output in the pre-
and post-WWII eras. If the higher-quality data in the post-WWII period are
subjected to a transformation that makes the pre- and post-WWII data of
comparable quality, then one discerns no change in output fluctuations across
the two periods. While her findings have garnered much interest, Backus and
Kehoe [23] argue that it is difficult to reach a firm conclusion on this point
based on US data alone. They suggest that if one considers additional evidence
based on a larger sample of countries over the different periods, the variability
of output appears to have diminished in the post-WWII period. Stock and
Watson [198] seek to avoid the problemof poor data quality in the pre-WWII
period, and compare the volatility of 21 different variables in the 20 years since
the 1980s andthe 40years inthe periodfollowingWWII upto the 1980s. They
find that the standard deviations of a typical variable in their sample declined
by about 20–40% since the 1980s. They attribute around 20–30% of the
reduction in output volatility to better monetary policy, 15%to smaller shocks
to productivity, and another 15% to reduced shocks to food and commodity
prices. Their findings also provide evidence on the factors behind the “Great
Moderation”.
2.3. THE EURO AREA BUSINESS CYCLE
Much of the early work on the European business cycle was concerned with
examining the impact of monetary union arrangements on economic activity.
Artis andZhang [18, 19] investigate the relationshipof the EuropeanExchange
Rate Mechanism (ERM) to the international business cycle in terms of the
linkage and synchronization of cyclical fluctuations between countries. Their
findings suggest the emergence of a group-specific European business cycle
since the formation of the ERM, which is independent of the US cycle. Artis,
Kontolemis, and Osborn [20] propose business cycle turning points for a
number of countries based on industrial production. The countries selected are
October 9, 2009 15:12 9in x 6in b808-ch02
20 Business Cycles: Fact, Fallacy and Fantasy
the G7 countries along with the prominent European countries. They use this
information to examine the international nature of cyclical movements, and to
determine whether cyclical movements are similar across different countries.
They also consider the lead/lag relationships between countries at peaks and
troughs.
Artis, Marcellino, and Proietti [21] discuss alternative approaches to dating
euro area business cycles. As in the Stock and Watson [197] approach, they
distinguish between classical and growth cycles. The euro area experience
makes consideration of both types of business cycles relevant. Whereas in
the post-WWII era the euro area countries exhibited high rates of growth,
making growth cycles the relevant concept, in recent years growth has slowed
and even absolute declines in real GDP have been observed, implying that
classical cycles should also be considered. These authors also articulate a
formal model of turning points based on restrictions imposed on a Markov
process.
3
To describe this algorithm, suppose that the economy can be in one
of two mutually exclusive states, expansion E
t
or recession R
t
. Suppose that
a peak terminates an expansion, and a trough terminates a recession. The
Markov process distinguishes between turning points within the two states by
assuming that
E
t
=

EC
t
expansion continuation
P
t
peak
,
R
t
=

RC
t
recession continuation
T
t
trough
.
Let p
EP
= Pr(P
t +1
|EC
t
) denote the probability of transiting to a peak,
conditional on being in an expansion, which implies that the probability of
continuing an expansion p
EE
= Pr(EC
t +1
|EC
t
) is given by p
EE
= 1 −
p
EP
. Likewise, define p
RT
= Pr(T
t +1
|RC
t
) as the probability of transiting
to a trough, conditional on being in a contraction. Then, the probability of
continuing a contraction p
RR
= Pr(RC
t +1
|RC
t
) is p
RR
= 1 − p
RT
. Let S
t
denote a discrete random variable that follows a first-order Markov process
3
This algorithm follows Harding and Pagan [115], who extended the Bry and Boschan [49] algorithm
to a quarterly setting.
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 21
with four states and transition probability matrix as follows:
EC
t +1
P
t +1
RC
t +1
T
t +1
EC
t
p
EE
p
EP
0 0
P
t
0 0 1 0
RC
t
0 0 p
RR
p
RT
T
t
1 0 0 0
The dating rules impose minimum durations on a phase, expansion or
contraction, and on complete cycles, peak-to-peak or trough-to-trough. The
duration of a phase is required to be two quarters, which is automatically
satisfied since the states (EC
t
, P
t
) both belong to an expansion and (RC
t
, T
t
)
belong to a contraction. The minimum duration of a complete cycle is given
as five quarters. Hence, a peak at date t − 4, P
t −4
, can only be followed by a
peak at t + 1, P
t +1
. A similar condition exists for troughs.
Now consider applying this algorithm to dating classical cycles. Let y
t
denote the underlying series. Then, we can define an expansion termination
sequence, ETS
t
, and a recession termination sequence, RTS
t
, respectively, as
follows:
ETS
t
= {(y
t +1
< 0) ∪ (
2
y
t +2
< 0)}
RTS
t
= {(y
t +1
> 0) ∪ (
2
y
t +2
> 0)}.
Thus, y
t +1
= y
t +1
− y
t
< 0 and
2
y
t +2
= y
t +2
− y
t
< 0 represent
the candidate sequence for terminating an expansion, which defines a peak.
Likewise, y
t +1
= y
t +1
− y
t
> 0 and
2
y
t +2
= y
t +2
− y
t
> 0 represent
the candidate sequence for terminating a contraction, which defines a trough.
The algorithm is completed by finding the probability that the economy will
transit to a peak, conditional on having been in an expansionary phase, p
EP
.
This is the joint event that the history at time t , (S
t −4
, S
t −3
, S
t −2
, S
t −1
, S
t
),
features an expansionary state at time t , S
t
= EC
t
, and that ETS
t
is true.
Else, if ETS
t
is false, the expansion continues. Likewise, the probability
that the economy will transit to a trough, conditional on having been in a
contractionary phase, is p
RT
. This is the joint event that the history at time t ,
(S
t −4
, S
t −3
, S
t −2
, S
t −1
, S
t
), features a contractionary state at time t , S
t
= RC
t
,
October 9, 2009 15:12 9in x 6in b808-ch02
22 Business Cycles: Fact, Fallacy and Fantasy
and that RTS
t
is true. Otherwise, if RTS
t
is false, the contraction continues.
The dating of deviation cycles is achieved by modifying this algorithm to
account for the fact that a peak cannot occur if output is below trend, and a
trough cannot occur if output is above it.
Artis et al. [21] also examine the properties of alternative filters such as the
Hodrick–Prescott and Baxter–King filters for decomposing a time series into
trend and cyclical components. They implement their procedures using data
on euro area GDP and its components for the European Central Bank’s (ECB)
Area-Wide Model for the period 1970–2001.
4
Table 2.2 shows the business
cycle turning points for both classical and deviation or growth cycles obtained
using this approach. The Centre for Economic Policy Research (CEPR) has
recently formed a committee to set the dates of the euro area business cycle. Its
stated mission is to establish the chronology of recessions and expansions of
the 11 original euro area member countries for 1970–1998 and of the current
euro area as a whole since 1999. The turning points determined by the CEPR
Business Cycle Dating Committee are also indicated in this table. We note that
these are similar to the classical cycles identified by Artis et al. [21]. Finally,
Table 2.2. Euro Area Business Cycle Turning Points.
Peak Trough
(1) Classical Cycles

1974 (III) 1975 (I)
1980 (I) 1981 (I)
1982 (II) 1982 (IV)
1992 (I) 1993 (I)
(2) Deviation Cycles

1974 (I) 1975 (III)
1976 (IV) 1977 (IV)
1980 (I) 1982 (IV)
1990 (II) 1993 (II)
1995 (I) 1997 (I)
1998 (I)
(3) CEPR Dating Committee
1974 (III) 1975 (I)
1980 (I) 1982 (IIII)
1992 (I) 1993 (III)

Artis, Marcellino, and Proietti [21].
4
See Fagan, Henry, and Mestre [85].
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 23
13.4
13.6
13.8
14.0
14.2
14.4
1970 1975 1980 1985 1990 1995 2000 2005
BPTRENDEURO EUROAREA
-0.03
-0.02
-0.01
0.00
0.01
0.02
0.03
1970 1975 1980 1985 1990 1995 2000 2005
BPCYCLEEURO
Fig. 2.3. Trend and Cyclical Components in Euro Area GDP.
Fig. 2.3 illustrates the logarithm of real GDP for the euro area and its trend
and cyclical components estimated according to the Baxter–King filter for the
period 1970(I)–2005(III).
Stock and Watson [199] provide a comprehensive analysis of the volatility
and persistence of business cycles in G7 countries (defined to include the
US, UK, France, Germany, Italy, Japan, and Canada) over the period 1960–
2002. They find that the volatility of business cycles has moderated in most
G7 countries over the past 40 years. They also provide evidence on the
synchronization of international business cycles. They base their results on
various measures of correlation of GDP growth across countries. First, they
find no evidence for closer international synchronization over their period
of study. This is similar to the findings of Köse, Prasad, and Terrones [139]
and others. However, in sync with Artis et al. [20], they find evidence on
the emergence of two cyclically coherent groups, the eurozone countries and
English-speaking countries (including Canada, the UK, and the US).
Stock and Watson [199] also seek to provide evidence on the sources of
the changes, namely, do they arise from changes in the magnitudes of the
shocks or the nature of the propagation mechanism? Are the sources of the
changes domestic or international? To answer these questions, they use a so-
called factor-structural vector autoregression (FSVAR), which is specified in
terms of the growth rates of quarterly GDP for the G7 countries. This is a
standard structural vector autoregression (VAR) with an unobserved factor
structure imposed on the VAR innovations. Let Y
t
denote the n ×1 vector of
October 9, 2009 15:12 9in x 6in b808-ch02
24 Business Cycles: Fact, Fallacy and Fantasy
real GDP growth for the G7 countries considered in this study. The standard
VAR model is given by
Y
t
= A(L)Y
t −1

t
, (3.5)
where the vector of reduced-form errors v
t
has the factor structure
ν
t
= f
t
+
t
, (3.6)
where f
t
is a k ×1 vector with k ≤ n. In this representation, the elements of
t
denote the country-specific idiosyncratic shocks, and the elements of f
t
denote
the common international shocks. The covariance structure of the shocks is
given by
E(f
t
f

t
) = diag(σ
f 1
, . . . , σ
fk
)
and
E(ν
t
ν

t
) = diag(σ
ν1
, . . . , σ
νk
),
where the elements of f
t
and
t
are assumed to be mutually independent.
Notice that the common shocks affect the output of multiple countries
contemporaneously, whereas the idiosyncratic shocks affect them with a
lag. This provides the identification scheme to help identify the common
international shocks. The model allows for spillover effects to occur through
the lagged effect of an idiosyncratic shock. These may arise from the role of
international trade, for example.
Stock and Watson [199] find evidence for two common shocks. They
also provide a variance decomposition for the impact of (i) the common
international shocks, (ii) the domestic shocks, and (iii) the spillover effects
of the domestic shocks on the h-step-ahead forecast error for each country.
They consider two sample periods: 1960–1983 and 1984–2002. First, they
find that most of the variance of GDP growth can be attributed to common
and idiosyncratic domestic shocks. However, their relative variance varies by
country and by time period. In the first period, the impact of international
shocks is estimated to be greatest for countries such as Canada, France, and
Germany, and the least for Italy and Japan. In the second period, domestic
shocks explain almost all of the variance for Japan, reflecting the impact of
the ten-year-long deflationary episode for this country. Second, the role of
international sources of fluctuations arising from common shocks or from
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 25
spillovers appears to have increased for the US, Canada, and Italy. However,
they also show that the decline in overall volatility of GDP growth for
countries such as the US, Germany, and the UK is due to the decline in
the variance arising from international shocks. Their overall results indicate
that the magnitudes of common international shocks appear to have become
smaller in the 1980s and 1990s than they were in the 1960s and 1970s.
This is the source of the moderation of individual business cycles, and also of
the failure of business cycles to become more synchronized despite the great
increase in trade flows over this period. Finally, they also find that shocks have
become more persistent in countries such as Canada, France, and the UK.
2.4. IS THERE A WORLD BUSINESS CYCLE?
Köse, Otrok, and Whiteman [138] consider the issue of a world business
cycle, and use data on 60-odd countries covering seven regions of the world
to determine common factors underlying the cyclical fluctuations in the main
macroeconomic aggregates (output, consumption, and investment) across all
countries and regions as well as across countries and aggregates separately.
They argue that many recent studies of international business cycles focus on
a subset of countries — not the world.
5
Their approach assumes that there are K unobservable factors that
are hypothesized to characterize the dynamic interrelationship among a
cross-country set of economic time series. Let N denote the number of
countries, M the number of time series per country, and T the number of
time periods. Let y
i,t
denote the observable variables for i = 1, . . . , N × M,
t = 1, . . . , T. There are three types of factors:
• N country-specific factors ( f
country
n
, one per each country);
• R regional factors ( f
region
r
, where r stands for NorthAmerica, LatinAmerica,
Africa, Developed Asia, Developing Asia, Europe, or Oceania);
• the single world factor ( f
world
).
The behavior of each observable variable is related to the factors as follows:
y
i,t
= a
i
+ b
world
i
f
world
t
+ b
region
i
f
region
r,t
+ b
country
i
f
country
n,t
+
i,t
, (4.7)
5
For recent studies, see, for example, Gregory, Head, and Raynauld [109] or Lumsdaine and
Prasad [154].
October 9, 2009 15:12 9in x 6in b808-ch02
26 Business Cycles: Fact, Fallacy and Fantasy
where E(
i,t

j,t
) = 0 for i = j. The coefficients b
j
i
denote the factor loadings,
and show how much of the variation in y
i,t
can be explained by each factor.
Thus, the model assumes that the behavior of M × N time series can be
explained by N + R + 1 factors, where N + R + 1 < MN. Both the
idiosyncratic errors and the factors are allowed to be serially correlated, and to
follow autoregressions of orders p
i
and q
k
, respectively, as

i,t
= φ
i,1

i,t −1

i,2

i,t −2
+ · · · +φ
i,p
i

i,t −p
i
+ u
i,t
, (4.8)
where E(u
i,t
u
i,t −s
) = σ
2
i
for i = j and s = 0, and zero otherwise. Let
f
k,t
=
f
k
,t
. (4.9)
Then,

f
k
,t
= φ
f
k
,1

f
k
,t −1

f
k
,2

f
k
,t −2
+ · · · +φ
f
k
,q
k

f
k
,t −q
k
+ u
f
k
,t
, (4.10)
where E(u
f
k
,t
u
f
k
,t
) = σ
2
f
k
, E(u
f
k
,t
u
i,t −s
) = 0 for all k, i, s. The innovations
u
i,t
, i = 1, . . . , M × N, and u
f
k
,t
, k = 1, . . . , K , are mean zero,
contemporaneously uncorrelated random variables. Thus, all the covariation
among the observable series y
i,t
is due to the commonfactors, whichthemselves
may be serially correlated.
Since the common factors are unobserved, we cannot identify uniquely
the sign or scale of the factors or the factor loadings. One identification device
is to assume that the factor loading on one of the factors is positive. For the
world factor, the sign of the factor loading for US output is considered positive.
Likewise, for the regional factor, the sign of the factor loading corresponding
to North America is considered positive, and the country factors are identified
by positive factor loadings on the output of each country. Finally, scales
are identified by assuming that each σ
2
f
k
is equal to a constant. There are
several ways to estimate this model. Since the factors are unobservable, one
approachis to use a Kalmanfiltering algorithmtogether withclassical statistical
techniques to estimate the model’s parameters; an alternative is to use Bayesian
estimation techniques. This procedure yields a joint posterior distribution
for the unobserved factors and the unknown parameters, conditional on the
data. In what follows, we report results based on the median of the posterior
distribution for the factors, and defer a discussion of the estimation techniques
to Chapter 7.
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 27
Köse et al.’s paper [138] shows that the world factor identifies many of
the major cyclical events over a 30-year period: the expansionary periods of
the 1960s, the recession of the 1970s associated with the first oil shock, the
recession of the 1980s associated with the debt crisis in developing countries
and the tight monetary policies in the major developed countries, and the
downturn and recession of the early 1990s. However, in contrast to models
estimated using a smaller sample of countries, the world factor based on a large
set of developed and developing countries implies that the recession of the
1980s was, if anything, as severe as the recession of the mid-1970s. Recall that
the world, regional, and country factors are modeled as autoregressive processes
which may display substantial persistence depending on the estimated values
of the parameters. A second important finding from the paper is that most of
the persistent co-movement across countries and aggregates is captured by the
world factor. By contrast, the regional and country factors explain covariation
or co-movement that is less persistent.
The paper also allows for a simple decomposition of variance attributed to
the different factors. Using the representation of the model, we have that
Var(y
i,t
) = (b
world
i
)
2
Var(f
world
t
) + (b
region
i
)
2
Var(f
region
r,t
)
+(b
country
i
)
2
Var(f
country
n,t
) + Var(
i,t
). (4.11)
Thus, the fraction of variance attributed to the world factor can be expressed as
(b
world
i
)
2
Var(f
world
t
)
Var(y
i,t
)
.
They find evidence that, first, the world factor explains nearly 15%of variation
in output growth, 9%of consumption growth, and 7%of investment growth.
They interpret this finding as evidence for a world business cycle. Second, the
world factor is more successful in explaining economic activity in developed
relative to developing countries. Third, country factors account for a much
larger fraction of consumption growth than output growth. This finding has
been documented further in the international business cycle literature, which
is discussed in detail in Chapter 4.
The authors also document some noteworthy findings regarding the
sources of volatility of investment growth. Specifically, they find that country
and idiosyncratic factors account for a much larger share of variability in
investment growth than the world and regional factors. Second, idiosyncratic
October 9, 2009 15:12 9in x 6in b808-ch02
28 Business Cycles: Fact, Fallacy and Fantasy
shocks explain a much larger fraction of the volatility in investment growth in
developing countries relative to developed countries. The authors pose this as
a puzzle. However, much work on investment behavior shows that irreversible
investment decisions are particularly susceptible to risk and uncertainty.
6
Altug, Demers, and Demers [10] argue that an important source of risk
for developing countries is political risk or risk arising from the threat of
expropriation, disruptions to market access, policy reversals, and debt and
currency crises.
7
We conjecture that variability in investment due to such
factors is a key channel for inducing idiosyncratic volatility in developing
economies’ investment behavior.
The last two findings of the paper imply that regional factors play a minor
role in aggregate output fluctuations. Paralleling this finding, there is little
evidence that the volatility of European aggregates can be attributed to the
European regional factor. This last result is interpreted as evidence against
the existence of a European business cycle. Yet this finding could also be due
to model misspecification. As Stock and Watson [199] note, the dynamic
factor model suffers from the shortcoming that all covariation is attributed
to the common factors. Yet, there is also the possibility that some observed
covariation could result from the spillover effects of idiosyncratic or country
shocks, especially in an era of increased trade flows and financial integration.
2.5. HISTORICAL BUSINESS CYCLES
Basu andTaylor [31] examine business cycles within an international historical
perspective. They consider the time series behavior of output, prices, real wages,
exchange rates, total consumption, investment, and the current account for
15 countries including the US, the UK, and other European countries plus
Argentina for the period since 1870. They divide this period into four periods
that also reflect the monetary and capital account regimes prevailing in them.
• 1870–1941: This era corresponded to the classic gold standard. It featured
fixed exchange rates and worldwide capital market integration.
6
For recent reviews, see Caballero [53] or Demers, Demers, and Altug [79].
7
In a related literature, Rodrik [178] emphasizes the importance of political risk (in his case, the risk of
policy reversal) for irreversible investment decisions. Several studies document the importance of political
risk in the unsuccessful recovery of private investment following the adoption of IMF stabilization packages
in various countries. See, for example, Serven and Solimano [189].
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 29
• 1919–1939: During this period, the world economy shifted from a
globalized regime to an autarkic regime. This period also corresponded to
the Great Depression.
• 1945–1971: The third regime is the Bretton Woods era, which
corresponded to the post-World War II era of reconstruction and a
resumption of global trade and capital flows.
• Early 1970s–present: This period features a floating exchange regime and
a period of increasing globalization.
Basu and Taylor [31] argue that considering such a breakdown allows for
the analysis of the impact of different regimes on cyclical phenomena. It also
provides a way to identify the importance of demand- versus supply-side factors
or shocks and the role of alternative propagation mechanisms such as price
rigidity. For example, most explanations of the Great Depression attribute
the source of this massive downturn, which simultaneously occurred in a
number of countries, to monetary phenomena. Consideration of alternative
historical periods, including the era of the Great Depression, thus allows
for an examination of such mechanisms as nominal rigidities in propagating
shocks. Basu and Taylor [31] also argue that their periodization allows for an
analysis of international capital flows and capital mobility in affecting cyclical
fluctuations.
Bordo and Heibling [44] ask whether national business cycles have become
more synchronized. They examine business cycles in 16 countries during
the period 1880–2001, and consider the four exchange rate regimes also
examined by Basu and Taylor [31]. The countries that they examine are
Australia, Canada, Denmark, Finland, France, Germany, Italy, Japan, the
Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, the UK, and
the US. Synchronization refers to the notion that “the timing and magnitudes
of major changes in economic activity appear increasingly similar.” Among
other concepts, they use a statistical measure of synchronization developed
by Harding and Pagan [115], the so-called concordance correlation, which
examines whether the turning points in the different series occur at similar
dates. Instead of using information on NBER-type reference dates, Bordo
and Heibling [44] use real GDP or industrial production series to determine
synchronization. They also use standard output correlations and factor-based
measures.
October 9, 2009 15:12 9in x 6in b808-ch02
30 Business Cycles: Fact, Fallacy and Fantasy
To briefly describe the concordance correlations, let S
it
and S
jt
denote
binary-cycle indicator variables which assume a value of 1 if the economy
is in an expansion, 0 otherwise, for countries i and j. A simple measure of
concordance is defined by the variable
I
ij
=
1
T
T

t =1
[S
it
S
jt
− (1 − S
it
)(1 − S
jt
)]. (5.12)
Let
¯
S
i
= (1/T)

T
t =1
S
it
denote the estimated probability of being in an
expansion, say; then under the assumption that S
it
and S
jt
are independent,
the expected value of the concordance index is 2
¯
S
i
¯
S
j
+1−
¯
S
i

¯
S
j
. Subtracting
this from I
ij
yields the mean-corrected concordance index:
I

ij
= 2
1
T
T

t =1
(S
it

¯
S
i
)(S
jt

¯
S
j
). (5.13)
To derive the concordance correlation coefficient, we divide I

ij
by a consistent
estimate of its standard error under the null hypothesis of independence. Note
that the variance of I

ij
is given by
Var(I

ij
) =
4
T
2
E

T

t =1
(S
it

¯
S
i
)(S
jt

¯
S
j
)

.
If the two cycles are perfectly synchronized (so that S
it
= S
jt
for all t ), then
the standardized index equals 1; if they are in different states in each period
(so that S
it
= 1 −S
jt
), then the standardized index equals −1; and if the two
cycles are unrelated, the standardized index equals 0.
Bordo and Heibling [44] calculate the business cycle indicators S
it
for
each of the four exchange rate regimes. They define a recession as one or
more consecutive years of real GDP growth, while an expansion is defined
as one or more years of positive GDP growth. They argue that S
it
= 1
for most countries during the Bretton Woods era of 1948–1972 and hence
leave this period out. For the gold standard era, they find that the average
of the correlation coefficients is zero, as half of all pairs of business cycles
are negatively related to each other while the other half are positively related.
In the interwar and post-Bretton Woods era, more than half of all national
business cycles become positively related to each other. The key finding is
that during the classical gold standard, cycles were, on average, uncorrelated
October 9, 2009 15:12 9in x 6in b808-ch02
Facts 31
with each other; whereas beginning with the interwar period, they started
becoming synchronized with each other. The authors also examine standard
output correlations, which measure the magnitude as well as the direction of
output changes. They find that there has been a tendency for higher, more
positive bilateral output correlations by era. They also find higher output
correlations for the core European countries (the EEC) and the Continental
Europeancountries. Finally, as inStockandWatson[199], they findanincrease
in correlations for the Anglo-Saxon countries.
The authors also estimate a so-called static approximate factor model for
the growth rates of GDP. In this model, there are no dynamics in the relation
between output growth rates and the factors, and the idiosyncratic shocks
are allowed to be serially correlated and heteroscedastic. They find that the
variability of output due to variability in the common factors has “doubled
from about 20 percent during the Gold Standard era to about 40 percent
during the modern era of flexible exchange rates.” They also estimate restricted
versions of a FSVAR model along the lines of Stock and Watson [199] to
determine the role of common shocks, idiosyncratic shocks, and spillover
effects in generating this result. They consider a so-called center country version
of this model, and a trade linkages version. Inthe former, lagged GDPgrowth of
the center country is included alongside the lagged value of the own country’s
GDP growth in the VAR representation. In the trade linkages version, lagged
GDP growth of the major trading partner is included in place of the center
country’s. They find that both global (common) shocks and transmission
have become more important. However, the importance of transmission for
peripheral countries arises only in the trade model, suggesting that it is not
transmission from the center country that accounts for the increased role of
transmission.
Bordo and Heibling [44] conclude by noting that global shocks are the
dominant influence across all regimes, and that the increasing importance of
global shocks reflects the forces of globalization, especially the integration of
goods and services through international trade and the integration of financial
markets. Eichengreen and Bordo [84] examine the nature of crises in two
periods of globalization, before 1914 and after 1971. They argue that banking
crises were less severe in the period before 1915, but this was not so for financial
or twin crises. Typically, such crises have figured importantly in emerging
market output fluctuations. We discuss emerging market business cycles in
detail in Chapter 6.
October 9, 2009 15:12 9in x 6in b808-ch02
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808-ch03
Chapter 3
Models of Business Cycles
The standard approach to classifying business cycle models follows Ragnar
Frisch’s [95] terminology of impulses and propagation mechanisms. The
shocks or impulses thought to instigate business cycles have typically been
varied and diverse. Technology shocks which alter a society’s production
possibilities frontier, whether they are permanent or transitory such as oil
shocks, have figured prominently in the recent literature. Weather shocks
have always had a place among the impulses thought to trigger fluctuations
in economic activity. Political shocks, wars, and other disruptions in market
activity have also been acknowledged to play a role. More controversially, one
could also assign a role to taste shocks or changes in preferences of consumers.
Equivalently, one could argue, as Keynes [127] did, that investors’ “animal
spirits” or, more precisely, changes in their subjective beliefs could help to
trigger business cycles.
Much of the controversy in the business cycle literature has stemmed from
differences attached to the importance of alternative propagation mechanisms
and the associated shocks. Current real business cycle (RBC) theory argues that
business cycles canarise infrictionless, perfectly competitive, complete markets
in which there are real or technology shocks. This approach emphasizes the
role of intertemporal substitution motives in propagating shocks. By contrast,
Keynesian and New Keynesian models stress the role of frictions such as price
stickiness. In a simple labor market model, if real wages do not adjust downward
when there is a negative shock to demand, the result is unemployment and
greater declines in output relative to a situation with flexible prices. The Great
33
October 9, 2009 15:12 9in x 6in b808-ch03
34 Business Cycles: Fact, Fallacy and Fantasy
Depression and the recent financial crises also indicate the importance of credit
market frictions and financial accelerator-type effects as a potential propagation
mechanism for business cycles. In this case, shocks originating in various
financial markets lead to a widening circle of bankruptcies and bank failures
as well as large and negative effects on real output.
1
In this chapter, we discuss variants of the RBC model. Since the debate
has revolved around the efficacy of technology shocks and intertemporal
substitution effects in replicating business cycles, it seems imperative to discuss
the mainpropagationmechanisms of the RBCmodel. Inthe following chapter,
we discuss an international RBC model to describe how the transmission
mechanisms proposed by this approach translate into an open-economy
setting.
2
3.1. AN RBC MODEL
RBC theory has become popular in recent years because its micro foundations
are fully specified and it links the short run with the neoclassical growth model.
The prototypical RBC model has the structure of a standard neoclassical
growth model with a labor/leisure choice incorporated. As Kydland and
Prescott argue, this is a crucial modification as “more than half of business cycle
fluctuations are accounted for by variations in the labor input” ([143], p. 173).
There is a representative consumer who derives utility from consumption and
leisure, and a constant-returns-to-scale (CRTS) production technology for
producing the single good using labor and capital. All markets are perfectly
competitive, all factors are fully employed, and all prices adjust instantaneously
to clear markets. A number of these assumptions have come under criticism
in the recent literature. The standard RBC model also assumes that there is no
government, that households consume only goods produced in the market,
and that there is no trade with the rest of the world. Some of the extensions
of the RBC approach have modified the basic framework to incorporate the
role of government, home production, and international trade. We will discuss
these extensions in the following sections.
1
See, for example, Bernanke and Gertler [40] or Kiyotaki and Moore [134].
2
Rebelo [177] provides a recent review of alternative models and directions associated with the RBC
agenda, but his discussion is abbreviated to satisfy the constraints imposed by a journal article.
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 35
Turning to the basic model, the representative agent has time-separable
preferences over consumption and leisure choices given by
U = E
0
_

t =0
β
t
u(c
t
, l
t
)
_
, (1.1)
where 0 < β < 1 is the discount factor. For ease of exposition, we will suppose
that the utility function has the form
U(c
t
, l
t
) = c
θ
t
l
1−θ
t
, 0 < θ < 1.
The time constraint requires that the sum of leisure and labor hours
equals 1:
l
t
+ h
t
= 1. (1.2)
Hence, time spent not working is taken as leisure. There is a representative
firm with CRTS production that is affected by a stochastic technology shock
each period:
y
t
= exp (z
t
)k
α
t
h
1−α
t
, 0 < α < 1. (1.3)
Assume that the technology shock follows an AR(1) process:
z
t +1
= µ+ρz
t
+
t +1
, 0 < ρ < 1,
t
∼ i.i.d., and N(0, σ
2

). (1.4)
Capital evolves according to
k
t +1
= (1 − δ)k
t
+ i
t
, (1.5)
where i
t
denotes economy-wide investment and 0 < δ < 1 is the depreciation
rate. The aggregate feasibility constraint is defined as
c
t
+ i
t
= y
t
. (1.6)
Notice that the only shock in this model is the technology shock, which is
assumed to be stationary but persistent as long as ρ = 1. We will discuss the
propagation mechanism when describing the solution for the model.
The problem described above is an infinite-horizon dynamic stochastic
optimization problem. The existence of a solution can be shown using standard
recursive methods for analyzing such problems.
3
Assuming a solution exists,
3
See, for example, Altug and Labadie [6] or Stokey and Lucas with Prescott [201].
October 9, 2009 15:12 9in x 6in b808-ch03
36 Business Cycles: Fact, Fallacy and Fantasy
let the value function associated with the problem be expressed as
V (k
t
, z
t
) = max
c
t
,l
t
,k
t +1
{U(c
t
, l
t
) + βE
t
V (k
t +1
, z
t +1
)} (1.7)
subject to
c
t
+ k
t +1
= exp (z
t
)k
α
t
h
1−α
t
+ (1 − δ)k
t
,
l
t
+ h
t
= 1.
Let λ
t
denote the Lagrange multiplier on the resource constraint. Substituting
for l
t
= 1−h
t
using the time constraint, the first-order conditions with respect
to c
t
, k
t +1
and l
t
and the envelope condition are
U
1
(c
t
, l
t
) = λ
t
, (1.8)
βE
t
V
k
(k
t +1
, z
t +1
) = λ
t
, (1.9)
U
2
(c
t
, l
t
) = λ
t
exp (z
t
)(1 − α)k
α
t
h
−α
t
, (1.10)
V
k
(k
t
, z
t
) = λ
t
_
exp (z
t
)αk
α−1
t
h
1−α
t
+ (1 − δ)
_
, (1.11)
where U
1t
and U
2t
denote the marginal utility of consumption and leisure,
respectively. These conditions can be rewritten as
U
2t
U
1t
= exp (z
t
)(1 − α)(k
t
/h
t
)
α
(1.12)
βE
t
_
U
1,t +1
U
1t
[exp (z
t +1
)α(k
t +1
/h
t +1
)
α−1
+ (1 − δ)]
_
= 1. (1.13)
The first equation shows that the marginal rate of substitution between
consumption and leisure is equal to the marginal product of labor. The second
equation is the intertemporal Euler equation. Observe that the ratio U
2
/U
1
is a function of the capital-labor ratio, k
t
/h
t
. There is no unemployment in
the model as time not spent working is (optimally) taken as leisure. This
means that the capital accumulation process will also be a function of the
capital-labor ratio.
Suppose for the moment that there is no investment in the model, that is,
i
t
= 0, k
t
=
¯
k, and c
t
= y
t
. Notice that this is just a simple Robinson Crusoe
economy in which the Crusoe produces output using capital and labor, which
he then consumes. Thus, his choice is between how much to work and how
much time to spend in leisure. To analyze the optimal consumption-leisure
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 37
choice, we consider equation (1.12), which under our preference specification
becomes
(1 − θ)c
t
θl
t
= exp (z
t
)(1 − α)(k
t
/h
t
)
α
. (1.14)
Substituting for c
t
= y
t
= exp (z
t
)k
α
t
h
1−α
t
yields
(1 − θ) exp (z
t
)k
α
t
h
1−α
t
θ(1 − h
t
)
= exp (z
t
)(1 − α)(k
t
/h
t
)
α
. (1.15)
We can solve this equation for the optimal labor hours h

t
as
h

t
=
θ(1 − α)
1 − θ + θα
.
Thus, for example, if θ = 2/3 and α = 1/2, then Crusoe spends 50% of
his time working and 50% taking leisure. Notice that the technology shock
does not affect optimal hours worked because, for this preference specification,
the income and substitution effects of an increase in productivity on hours
worked cancel each other out. However, we note that if the substitution effect
of an increase in productivity dominates the income effect, then Crusoe will
work less in response to a temporary negative technology shock and will
also consume and produce less. Figure 3.1 describes the representative agent’s
optimum.
Now suppose output can be consumed or invested so that i
t
= 0. Then,
in addition to equation (1.14), we have the intertemporal Euler equation
characterizing the behavior of the optimal capital stock over time. Using our
preference specification, we can rewrite the intertemporal Euler equation by
noting that U
1
(c
t +i
, l
t +i
) = θ(c
t +i
/l
t +i
)
θ−1
for i ≥ 0 and using the result in
equation (1.14). This yields
βE
t
_
_
exp (z
t +1
)
_
k
t +1
h
t +1
_
α
_
θ−1
_
αexp (z
t +1
)
_
k
t +1
h
t +1
_
α−1
+ 1 − δ
__
=
_
exp (z
t
)
_
k
t
h
t
_
α
_
θ−1
. (1.16)
Now a temporary negative technology shock also affects investment or saving.
Crusoe will typically respond to a temporary negative shock by lowering
savings and also by working less. Thus, the impact of these changes is to
induce fluctuations in labor supply and employment as well as investment.
October 9, 2009 15:12 9in x 6in b808-ch03
38 Business Cycles: Fact, Fallacy and Fantasy
Fig. 3.1. Robinson Crusoe’s Optimum.
Furthermore, if the economy experiences a period with lower investment, it
will have a lower capital stock in the future, implying that the effect of the
shock persists.
We can also derive a measure of the Solow residual for this model by using
the form of the production function. To derive the Solow residual, take the
logarithm of both sides of equation (1.3) and then take the first differences:
ln (y
t +1
) = z
t +1
+ αln (k
t +1
) + (1 − α)ln (h
t +1
).
To derive an observable measure of the Solow residual, the parameter α is
typically measured as the share of capital in real output defined as s
kt
=
r
t
k
t
/p
t
y
t
, where r
t
denotes the competitively determined rental rate on capital
and p
t
denotes the product price normalized here as unity. Under constant
returns to scale, s
kt
+ s
ht
= 1, where s
ht
= w
t
h
t
/p
t
y
t
denotes the share of
labor in national income. Under these assumptions, the growth in the total
factor productivity (TFP) is measured as a residual:
z
t +1
= ln (y
t +1
) − s
kt
ln (k
t +1
) − (1 − s
kt
)ln (h
t +1
).
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 39
Solow [196] showed that the residual accounted for about one half of the
US GDP growth between 1909 and 1949. Similarly, according to Denison
[80], 40% of GNP growth in the USA between 1929 and 1957 resulted from
technical development. Notice that in the RBC framework the Solow residual
is typically treated as exogenous. This is an issue that has been contested in
the business cycle literature. Critics of the RBC approach argue that there are
endogenous components to the fluctuations in the Solow residual, rendering
the RBC interpretation invalid.
3.2. A NUMERICAL SOLUTION
We now provide a numerical solution to further illustrate the properties of the
standard RBC model. We implement a quadratic approximation procedure
that was initially suggested by Kydland and Prescott [141], and convert
the original nonlinear dynamic optimization procedure to one which has a
quadratic objective and linear constraints (see also Christiano [64]).
We assume that capital depreciates at the rate 0 < δ < 1andthe technology
shock follows the process in (1.4). Since the utility function is strictly
increasing, consumption plus investment equals output at the optimum.
Using this fact, we can substitute for consumption in the utility function
to obtain u(c
t
) = u( exp (z
t
)k
α
t
h
1−α
t
− i
t
, 1 − h). Letting u(c
t
, 1 − h
t
) =
θ ln (c
t
) + (1 − θ) ln (1 − h
t
), 0 < θ < 1, the problem now becomes
max
{i
t
,h
t
}

t =0
E
0
_

t =0
_
θ ln ( exp (z
t
)k
α
t
h
1−α
t
− i
t
) + (1 − θ) ln (1 − h
t
)
_
_
subject to
k
t +1
= (1 − δ)k
t
+ i
t
, (2.17)
z
t +1
= µ + ρz
t
+
t +1
, (2.18)
c
t
≥ 0, 0 ≤ h
t
≤ 1. (2.19)
The quadratic approximation procedure is implemented as follows.
Step 1: Compute the deterministic steady state for the model. This is obtained
by setting the exogenous technology shock equal to its mean value and
evaluating the conditions (1.14) and (1.16) together with the feasibility
October 9, 2009 15:12 9in x 6in b808-ch03
40 Business Cycles: Fact, Fallacy and Fantasy
conditions at constant values for c
t
, k
t
, and h
t
. Let ¯ z = µ/(1 −ρ) denote the
unconditional mean for the (log of ) technology process. Using (1.16), we can
solve for the steady-state capital-labor ratio as
¯
k
¯
h
=
_
α
exp (¯ z)
r + δ
_
1/(1−α)
. (2.20)
Next, note that in the deterministic steady state, investment is equal to
depreciation:
¯
i = δ
¯
k. (2.21)
Simplifying the result in (1.14), we obtain
1 − θ
θ
¯
h
1 −
¯
h
= (1 − α)
¯ y
¯ c
, (2.22)
where the aggregate feasibility constraint ¯ c + δ
¯
k = ¯ y ≡ exp ¯ z
¯
k
α
¯
h
1−α
is also
assumed to hold. To find an explicit expression for
¯
h in terms of the underlying
parameters, use the relation in (2.22) as
(1 − θ)
¯
h¯ c = (1 − α)θ(1 −
¯
h)¯ y.
Rearranging, substituting for ¯ c first and then dividing through by ¯ y, we obtain
(1 − θ)
¯
h
_
1 − δ
¯
k
¯ y
_
+ θ(1 − α)
¯
h = θ(1 − α).
Now note that
¯
k
¯ y
=
¯
k
exp (¯ z)
¯
k
α ¯
h
1−α
=
(
¯
k/
¯
h)
1−α
exp (¯ z)
=
α
r + δ
.
Substituting back into the equation defining
¯
h and simplifying yields
¯
h
_
(1 − θ) + θ(1 − α) − (1 − θ)δ
α
r + δ
_
= θ(1 − α),
or
¯
h =
θ(1 − α)(r + δ)
(r + δ)(1 − θα) − (1 − θ)δα
. (2.23)
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 41
Step 2: Approximate the original utility function by a quadratic function
around the deterministic steady state.
4
For this purpose, let s ≡ (1, k, z, i, h)
and ¯s ≡ (0, ¯ z,
¯
k,
¯
i,
¯
h). Approximate u(s) near the deterministic steady state ¯s
using a second-order Taylor series approximation as
u(s) = u(¯s) + (s − ¯s)

∂u(s)
∂s
+
1
2
(s − ¯s)


2
u(s)
∂s∂s

(s − ¯s).
Define e as the 5 × 1 vector with a 1 in the first row and zeros elsewhere.
Notice that the utility function can be written as
u(s) = (s − ¯s)

T(s − ¯s),
where
T = e
_
u(¯s) +
1
2
¯s


2
u(s)
∂s∂s

¯s
_
e

+
1
2
_
∂u(s)
∂s
e

+ e
∂u(s)

∂s
_
+
1
2
_

2
u(s)
∂s∂s

_
,
where all partial derivatives are evaluated at ¯s.
Define the vector of state variables as x
t
= (1, k
t

¯
k, z
t
− ¯ z)

and the
vector of control variables as u
t
= (i
t

¯
i, h
t

¯
h)

. We can write the law of
motion for the state variables as
_
_
1
k
t +1
z
t +1
_
_
=
_
_
1 0 0
0 (1 − δ) 0
µ 0 ρ
_
_
_
_
1
k
t
z
t
_
_
+
_
_
0 0 0
0 1 0
0 0 0
_
_
_
_
0
i
t
0
_
_
+
_
_
0
0

t +1
_
_
.
Notice that s − ¯s = (x, u)

. Thus, the quadratic form (s − ¯s)

T(s − ¯s) can be
written as
(s − ¯s)

T(s − ¯s) =
_
x
u
_

_
T
11
T
12
T
21
T
22
_ _
x
u
_
=
_
x
u
_

_
R W
W

Q
_ _
x
u
_
.
4
An alternative approach is to log-linearize the model. See Campbell [55].
October 9, 2009 15:12 9in x 6in b808-ch03
42 Business Cycles: Fact, Fallacy and Fantasy
Step 3: Convert the original dynamic optimization problem into a problem
with linear constraints and a quadratic objective function. Using the definition
of T, x
t
, and u
t
, the dynamic optimization problem can now be written as
max
{u
t
}

t =0
E
0
_

t =0
β
t
[x

t
Rx
t
+ u

t
Qu
t
+ 2x

t
Wu
t
]
_
(2.24)
subject to the linear law of motion
x
t +1
= Ax
t
+ Bu
t
+ ε
t +1
, t ≥ 0, (2.25)
where E(ε
t +1
) = 0 and E(ε
t
ε

t
) = . This is nowan optimal control problem
with a quadratic objective and linear constraints. Such problems can be solved
using the methods for solving dynamic optimization problems that satisfy a
certainty equivalence property; in other words, the solution for the stochastic
optimal control problem is identical to the solution for the deterministic
version of the problem with the shocks ε
t +1
replaced by their expectation
E(ε
t +1
). This class of problems is known as optimal linear regulator problems.
Ljungqvist and Sargent [146] and Anderson, Hansen, McGrattan, and Sargent
[16] provide further discussion of the formulation and estimation of linear
dynamic economic models.
Bellman’s equation for this problem is given by
V (x
t
) = max
u
t
_
x

t
Rx
t
+ u

t
Qu
t
+ 2u

t
Wx
t
+ βE[V (x
t +1
)]
_
subject to x
t +1
= Ax
t
+Bu
t

t +1
. Given the structure of the problem, notice
that the value function will be a quadratic function in the state variables:
V (x) = x

Px + d,
where d and P are quantities to be determined. Substituting for next period’s
state variables and using this expression for the value function, Bellman’s
equation becomes
x

Px + d = max
u
_
x

Rx + u

Qu + 2x

Wu
+βE
_
(Ax + Bu + ε)

P(Ax + Bu + ε) + d
__
. (2.26)
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 43
The first-order conditions with respect to u are
Qu + W

x + β[B

PAx + B

PBu] = 0.
5
(2.27)
Solving for u yields
u = −(Q + βB

PB)
−1
(W

x + βB

PAx) = −Fx, (2.28)
where
F = (Q + βB

PB)
−1
(βB

PA + W

). (2.29)
Notice that the shocks ε
t +1
do not affect the optimal choice of u
t
. Substituting
for u back into the definition of V (x) yields
P = R + βA

PA − (βA

PB + W )(Q + βB

PB)
−1
(βB

PA + W

),
d = (1 − β)
−1
[βE(ε

Pε)].
The equation for P is known as the algebraic matrix Riccati equation for
so-called optimal linear regulator problems.
6
The solution for P can be
obtained by iterating on the matrix Riccati difference equation:
P
n+1
= R + βA

P
n
A − (βA

P
n
B + W )(Q + βB

P
n
B)
−1
(βB

P
n
A + W

),
starting from P
0
= 0. The policy function associated with P
n
is
F
n+1
= (Q + βB

P
n
B)
−1
(βB

P
n
A + W

).
In the RBC literature, the model is then calibrated with the data.
Calibration refers to the practice of determining the parameters of the model
based on its steady state properties and the results of other studies, given
particular functional forms for the productionfunctionandthe utility function
(see Cooley and Prescott [75]). A stochastic process for z
t
is also specified and
a random number generator is used to simulate the TFP time series. The
simulated series are used to generate time series for consumption, output,
5
To derive this result, we have used the rules for differentiating quadratic forms as
∂u

Qu
∂u
= [Q + Q

]u = 2Qu,
∂x

Tu
∂u
= T

x and
∂u

Tx
∂u
= Tx.
6
Notice that the matrices R, Q, A, and B must be further restricted to ensure the existence of a solution
to the matrix difference equation defining P
n
. One condition that suffices is that the eigenvalues of A are
bounded in modulus below unity.
October 9, 2009 15:12 9in x 6in b808-ch03
44 Business Cycles: Fact, Fallacy and Fantasy
investment, and labor. Trends in the data are removed using a given filter. The
means, variances, covariances, and cross-correlations for the simulated time
series are compared to the comparable statistics in the data. The model is then
judged to be close or not close based on this comparison, so the approach
differs significantly from standard econometrics.
We now illustrate this solution method for the standard RBC model with
a labor-leisure choice presented at the beginning of this section. Consider the
parameter values β = 0.95, α = 1/3, θ = 0.36, δ = 0.10, ρ = 0.95, and
σ = 0.028. The steady values for the variables are given by
¯
k = 1.1281,
¯
i = 0.1128, ¯ y = 0.4783, ¯ c = 0.3655, and
¯
h = 0.2952. We first calculate a
set of unconditional moments that have been used in the RBC literature to
match the model with the data. These are obtained by simulating the behavior
of the different series across a given history of the shock sequence. Specifically,
we draw a sequence of shocks {ˆ
t +1
}
2999
t =0
that are normally distributed
with mean zero and standard deviation 0.028, and generate a sequence of
technology shocks beginning from some initial value z
0
as ˆ z
t +1
= ρz
t

t +1
.
We then use the linear decision rules for all variables to simulate for the
endogenous variables based on the same history of shocks. Finally, we calculate
unconditional moments for the different series after dropping the first 1000
observations.
The typical set of unconditional moments used in the RBC literature is
displayed in Table 3.1. We find that consumption fluctuates slightly less than
output and investment fluctuates significantly more. Likewise, productivity is
almost as variable as output. By contrast, we find that the variation in hours is
typically quite low. In terms of the correlation of each series with output, we
find that all series are procyclical, with hours showing the least procyclicality.
The RBC theorists have interpreted these results as implying that, in the
first instance, the model is capable of delivering some of the salient features
Table 3.1. Cyclical Properties of Key Variables.
Standard Deviation (%) Correlation with Output
Output 8.8616 1
Consumption 8.1134 0.9163
Investment 12.9238 0.9807
Hours 1.6684 0.4934
Productivity 7.7496 0.9840
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 45
of the data. Specifically, the model predicts that consumption fluctuates less
than output whereas investment fluctuates more, as we observe in the data.
Consumption, investment, and productivity are all strongly procyclical. In
terms of the variation accounted by the technology shocks, the technology
shock explains much but not all of the variation in output. Thus, the RBC
model has been viewed as being able to generate cycles endogenously and
having economic significance even if the calibration method is controversial.
This is discussed further in the following sections.
Figure 3.2 illustrates the response of all the variables to a 1% shock in
technology starting fromthe steady-state capital stock. We note that hours and
output both increase and then fall back to a lower level. The percentage change
in output exceeds the percentage change in the technology shock because
optimal hours also show a positive response to the technology shock. By
contrast, capital and consumption showa humped-shaped response, rising first
and thendeclining back to a lower level. Consumptionshows a gradual positive
response because investment is initially high. These responses have been widely
0 5 10 15 20
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
Years after shock
%

d
e
v
i
a
t
i
o
n
s

f
o
r
m

s
t
e
a
d
y

s
t
a
t
e
Hours
Output
Consumption
Capital
Technology
Fig. 3.2. Impulse Responses to a Shock in Technology.
October 9, 2009 15:12 9in x 6in b808-ch03
46 Business Cycles: Fact, Fallacy and Fantasy
documented inthe RBCliterature (see, for example, Uhlig [206]). The positive
response of hours toa positive productivity shockhas alsobecome animportant
point of difference between models that assume perfectly flexible prices versus
those that assume nominal price rigidity. We discuss New Keynesian models
with monopolistic competition and prices that are fixed in the short run in
Chapter 5.
3.3. INITIAL CRITICISMS
In perhaps what is the most famous example of an RBC model, Kydland and
Prescott [141] formulated a real business cycle model in which preferences
are not separable over time with respect to leisure and there exists a time-to-
build feature in investment for new capital goods. The production function
displays constant returns to scale with respect to hours worked and a composite
capital good. The only exogenous shock to their model is a randomtechnology
shock whichfollows a stationary first-order autoregressive process. They use the
quadratic approximation procedure to obtain linear decision rules for a set of
aggregate variables, and generate time series for the remaining series by drawing
realizations of the innovation to the technology shock. They calculate a small
set of moments associated with each series to match the model with the data.
When calibrating their model, Kydland and Prescott choose the variance
of the innovation to the technology shock to make the variability of the output
series generated by their model equal to the variability of observed GNP. As
McCallum [156] notes, this feature of their analysis makes it difficult to judge
whether “technology shocks are adequate to generate output, employment, etc.
fluctuations of the magnitude actually observed.” One of the most pervasive
criticisms of the model is that there is no evidence of large, economy-wide
disturbances that can play the role of the technology shocks posited by
Kydland and Prescott. The only exception is oil price shocks, leading one
to question whether the model was developed in reaction to the supply-side
disturbances that characterized the 1970s. The RBC approach in general has
difficulty in answering what some other examples of big shocks are. Summers
[202] argues that “the vast majority of what Prescott labels technology shocks
are in fact observable concomitants of labor hoarding and other behavior
which Prescott does not allow for in his model.” This point was subsequently
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 47
elaborated on by Hall [110, 111], who argued that the procyclical movements
in the Solow residual were, in fact, endogenous changes in efficiency arising
from labor hoarding, increasing returns to scale, or price markups.
Summers [202] also took issue with the parameters of the model that
Kydland and Prescott [141] or Prescott [173] advocated in their calibration
exercise. He argued that the econometric evidence presented by Eichenbaum,
Hansen, and Singleton [83] on the intertemporal substitution of leisure
hypothesis pointed to a share of time allocated to market activities that was
more in the range of one-sixth, not one-third. He also noted that a real interest
rate of 4% was inconsistent with the historical evidence of very low yields
averaging around 1%. Singleton [192] raised the problem of inference and
sampling uncertainty surrounding the various measures of fit proposed in
the RBC literature. He noted that the calibration approach did not provide
a straightforward and transparent way of judging whether a given model
constituted an improvement over another model. Eichenbaum [82] raised
the issue of the model’s fit based on the ratio of variance for GDP implied
by the model and the data. Eichenbaum showed that the standard error of
this ratio was heavily influenced by the parameters of the assumed stochastic
process for the technology shock.
The policy implications of the RBC model have also come under attack.
According to the RBC model, business cycle fluctuations are Pareto optimal
and there is no role for the government to try to smooth or mitigate
the fluctuations. In fact, such policy efforts are inefficient. Prescott [173]
commented as follows:
Economic theory predicts that, given the nature of the shocks to technology
and people’s willingness and ability to intertemporally and intratemporally
substitute, the economy will display fluctuations like those the U.S. economy
displays. . . . Indeed, if the economy did not display the business cycle
phenomena, there would be a puzzle.
Thus, according to this approach, cyclical fluctuations are viewed as the natural
response of the economy to changes that affect individuals’ consumption or
productiondecisions. However, if prices or wages are fixedor rigidor if there are
other types of frictions arising fromcredit markets, thenthe policy implications
of the RBC approach may not be valid and the economy may operate with
high levels of unemployment and excess capacity over long periods. In this
October 9, 2009 15:12 9in x 6in b808-ch03
48 Business Cycles: Fact, Fallacy and Fantasy
case, government policies may be required to move the economy towards a
full employment level.
In the chapters that follow, we will discuss the efficacy of the RBC model
in serving as a model of aggregate fluctuations as well as questions that have
lingered to this day about the adequacy of the calibrationapproachinmatching
the model to the data.
3.4. “PUZZLES”
Many of the initial criticisms which had been voiced by opponents of the RBC
approach took on the character of “puzzles”. These puzzles have constituted
the basis of much further research in the area.
We can list some of these puzzles as follows:
• The variability of hours and productivity: As seen in Table 3.1, one of the
main problems with the standard RBC model is that it cannot capture the
variation in the aggregate labor input (see also Kydland [140]). In the data,
employment is strongly procyclical and almost as variable as output while
real wages are weakly procyclical. In the standard model, a productivity
shock shifts the marginal product of labor so that the observed variations in
employment can only occur if the labor supply curve is relatively elastic. Yet,
micro studies find that wage elasticity of labor supply is quite low. In this
case, the marginal productivity shock should lead to most of the adjustment
in real wages and less in the quantity of labor. Furthermore, in the data,
hours of work per worker adjusts very little over the cycle. About two-thirds
of the variability in total hours worked comes from movements into and
out of the labor force, and the rest is due to adjustment in the number of
hours worked per employee. These findings create a puzzle for the standard
RBC model.
• The productivity puzzle: In the data, the correlation between productivity
and hours worked is near zero or negative, while the correlation between
productivity and output is positive and around 0.5.
7
By contrast, the
RBC model, which is driven entirely by productivity shocks, generates
correlations that are large and positive in both cases. Another problem that
7
See, for example, Christiano and Eichenbaum [65], who measure hours worked and productivity
based on both household and establishment-level surveys conducted by the US Department of Labor.
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 49
arises in matching the model and the data is that in the data, labor’s share
of income moves countercyclically, whereas in the RBC model labor’s share
is fixed.
• Reverse causality: The stylized facts of business cycles state that money,
especially M2, appears to be a leading indicator of aggregate output.
However, this is inconsistent with the notion that TFPshocks are the driving
force behind business cycle activity. The conventional wisdomregarding the
money-output correlation, summarized by Friedman and Schwartz [94] in
their study of monetary history, is that the causality goes from money to
output but with “long and variable lags”.
3.4.1. A Model with Indivisible Labor Supply
The indivisible labor, lottery models studied by Gary Hansen [112] and
Richard Rogerson [179] have been used to explain the stylized fact that
aggregate hours vary more than productivity does. This framework also
provides a way for reconciling large labor supply elasticities at the aggregate
level with low labor supply elasticities at the individual level. This is
accomplished by assuming that individuals can work all the time or not at all.
To account for the nonconvexities introduced by the work/nonwork decision,
it is assumed that individuals choose the probability of working, π
t
. A lottery
then determines whether an individual actually works. This economy is one in
which individuals and a firm trade a contract that commits the household to
work h
0
hours with probability π
t
. Since what is being traded is the contract,
the individual gets paid regardless of whether he works or not. We nowdescribe
a version of the RBC model due to Rogerson [179] and Hansen [112] that
allows for fixed costs and nonconvexities in labor supply.
Suppose that workers are constrained to work either zero or
ˆ
h hours, where
0 <
ˆ
h < 1. (4.30)
The main idea is that there are nonconvexities or fixed costs that make varying
the number of employed workers more efficient than varying hours per worker.
Let π
t
denote the probability that a given agent is employed in period t so
that the number of per capita hours worked is given by
H
t
= π
t
ˆ
h. (4.31)
October 9, 2009 15:12 9in x 6in b808-ch03
50 Business Cycles: Fact, Fallacy and Fantasy
Let c
0,t
denote the consumption of an unemployed worker and c
1,t
denote
the consumption of an employed agent. Then the expected utility of the
representative consumer, taking into account the work versus nonwork
decision, is given by
E[u(c
t
, l
t
)] = π
t
u(c
1,t
, 1 −
ˆ
h) + (1 − π
t
)u(c
0,t
, 1).
Assume that the individual utility function has the form
u(c, l ) = ln (c) + A ln (l ). (4.32)
The social planner solves the problem
max
π
t
,c
0,t
,c
1,t
E[u(c
t
, l
t
)] s.t. π
t
c
1,t
+ (1 − π
t
)c
0,t
= c
t
.
Notice that the social planner chooses the consumption allocations of each
agent plus the probability of their working. When agents do work, they
must supply
ˆ
h hours of work so that there is no choice over hours of work
directly. Let λ
t
denote the Lagrange multiplier on the feasibility constraint
for consumption. Omitting the work/leisure decision for the moment, the
first-order conditions with respect to (c
0,t
, c
1,t
) are
1 − π
t
c
0,t
= (1 − π
t

t
, (4.33)
π
t
c
1,t
= π
t
λ
t
. (4.34)
It follows that c
0,t
= c
1,t
= c
t
so that the agent consumes the same amount
whether or not he is working. Hence, the unemployed worker enjoys higher
utility since working causes disutility. In this model, ex ante all individuals are
alike, but ex post they differ because some work while others enjoy leisure. With
complete insurance and identical preferences that are separable with respect
to consumption and leisure, all individuals have the same consumption but
the unemployed are better off. This is a feature that is counterfactual to the
working of actual labor markets.
Notice that the agent will consume c
t
whether or not he is working. Hence,
expected utility (where the expectation is over whether or not you work) is
ln (c
t
) + π
t
A ln (1 −
ˆ
h) + (1 − π
t
)A ln (1), (4.35)
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 51
where A is a positive constant. Using the definition of H
t
, π
t
= H
t
/
ˆ
h. Now
substitute for π
t
in (4.35) and use ln (1) = 0 to obtain
E[u(c
t
, l
t
)] = ln (c
t
) + π
t
A ln (1 −
ˆ
h)
= ln (c
t
) − BH
t
, (4.36)
where
B =
−A ln (1 −
ˆ
h)
ˆ
h
.
Comparing equation (4.32) with equation (4.36) shows the effect of the
lottery assumption. The former specification for preferences implies a low
intertemporal elasticity of substitution in labor supply, which is consistent with
assumptions about individual behavior. By contrast, the latter specification —
which is linear in total hours H
t
— implies a high intertemporal elasticity at
the aggregate level.
The preferences can now be written as
U = E
0
_

t =0
β
t
u(c
t
, H
t
)
_
,
where u(c
t
, H
t
) = ln (c
t
) −BH
t
. We now solve the model subject to the time
constraint, resource constraint, production function, and law of motion for
the capital stock described earlier. The problem is
max
{c
t
,H
t
,k
t +1
}
E
0
_

t =0
β
t
_
ln (c
t
) − BH
t
+ λ
t
[exp (z
t
)k
θ
t
H
1−θ
t
+(1 − δ)k
t
− c
t
− k
t +1
]
_
_
.
The first-order conditions are
1
c
t
= λ
t
, (4.37)
B = λ
t
exp (z
t
)(1 − θ)k
θ
t
H
−θ
t
, (4.38)
λ
t
= βE
t
λ
t +1
_
exp (z
t +1
)θk
θ−1
t +1
H
1−θ
t +1
+ (1 − δ)
_
. (4.39)
October 9, 2009 15:12 9in x 6in b808-ch03
52 Business Cycles: Fact, Fallacy and Fantasy
Notice that (4.38) can be used to solve for H
t
as
H
t
=
_
Bc
t
exp (z
t
)(1 − θ)
_
−1/θ
k
t
=
_

A ln (1 −
ˆ
h)c
t
exp (z
t
)(1 − θ)
ˆ
h
_
−1/θ
k
t
. (4.40)
Equations (4.37) and (4.39) yield the intertemporal Euler equation as
1 = β
_
c
t
c
t +1
[exp (z
t +1
)θk
θ−1
t +1
H
1−θ
t +1
+ (1 − δ)]
_
, (4.41)
where the term in square brackets shows the rate of return to investing in
the aggregate production technology. With H
t
determined in (4.40), we can
use the resource constraint to solve for c
t
and then substitute for c
t
into the
intertemporal Euler equation to obtain a nonlinear stochastic in k
t +1
with
forcing process {z
t
}.
Hansen [112] calibrates this model by specifying values for the unknown
parameters θ, δ, β, A, and the stochastic process for the technology shock using
the approach in Kydland and Prescott [141]. Hansen argues that the model
with indivisibilities can generate a variability of hours relative to productivity
around 2.7 compared to the model without indivisibilities which implies
a value near unity. The purpose of the framework that Hansen [112] and
Rogerson [179] consider is to generate the stylized fact with respect to the
relative variability of hours versus productivity, and it is not intended to
incorporate the microeconomic foundations of the labor market.
8
3.4.2. The Productivity Puzzle
Several resolutions of the productivity puzzle have been suggested. These
typically specify an extra margin regarding individuals’ choices that helps break
the close link between hours and productivity implied by the standard model.
8
See Browning, Hansen, and Heckman [48] for further discussion regarding the reconciliation of
micro evidence with dynamic general equilibrium models.
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 53
Government Consumption Shocks
Christiano and Eichenbaum [65] note that as long as there is a single shock
that drives the behavior of both hours and productivity, the standard model
cannot deliver the strong procyclical response of hours without procyclical
behavior in productivity. They generate the negative correlation between
hours worked and real wages or productivity by introducing government
consumption shocks.
They consider a prototypical RBC model with a labor/leisure choice,
and utilize the solution of the social planner’s problem to derive the
competitive equilibrium allocations. Let
¯
N denote the time endowment of
the representative household per period. The social planner ranks alternative
consumption/leisure streams according to
E
0
_

t =0
β
t
[ln (c
t
) + γV (
¯
N − n
t
)]
_
, (4.42)
where c
t
denotes consumption and
¯
N −n
t
denotes leisure of the representative
household. Consumption services c
t
are related to private and public
consumption as follows:
c
t
= c
p
t
+ αg
t
, (4.43)
where c
p
t
is private consumption, g
t
is public consumption, and αis a parameter
that governs the impact of government consumption on the marginal utility
of private consumption. They consider two different specifications for the
labor/leisure choice, one which is based on a time-separable logarithmic
specification and a second which incorporates the Hansen indivisible labor
assumption, namely,
V (
¯
N − n
t
) =
_
ln (
¯
N − n
t
)
¯
N − n
t
(4.44)
for all t . Per capita output is produced according to the Cobb–Douglas
production function
y
t
= (z
t
n
t
)
1−θ
k
θ
t
, 0 < θ < 1, (4.45)
where z
t
is a technology shock which evolves according to the process
z
t
= z
t −1
exp (λ
t
). (4.46)
October 9, 2009 15:12 9in x 6in b808-ch03
54 Business Cycles: Fact, Fallacy and Fantasy
In this equation, λ
t
is an independent and identically distributed process with
mean λ and standard deviation σ
λ
. The aggregate resource constraint stipulates
that consumption plus investment cannot exceed output in each period:
c
p
t
+ g
t
+ k
t +1
− (1 − δ)k
t
≤ y
t
. (4.47)
Since the technology shock follows a logarithmic randomwalk, the solutionfor
the model canbe more fruitfully expressedinterms of the transformedvariables
¯
k
t +1
= k
t +1
/z
t
, ¯ y
t
= y
t
/z
t
, ¯ c
t
= c
t
/z
t
, and ¯ g
t
= g
t
/z
t
. The specification of
the model is completed by assuming a stochastic law of motion for ¯ g
t
as
ln (¯ g
t
) = (1 − ρ) ln (¯ g) + ρ ln (¯ g
t −1
) + µ
t
, (4.48)
where ln (¯ g) is the mean of ln (¯ g
t
), |ρ| < 1, and µ
t
is the innovation to ln (¯ g
t
)
with standard deviation σ
µ
.
In this model, government consumption shocks lead to shifts in the labor
supply curve so that, in the absence of technology shocks, hours of work can
increase along a downward-sloping labor demand curve. The key assumptionis
that private and government consumption are not perfect substitutes. Hence,
an increase in government consumption leads to a negative wealth effect
for consumers through the economy-wide resource constraint. If leisure is
a normal good, hours increase and average productivity declines in response
to a positive government consumption shock.
Home Production
Another way of improving the model’s ability to match the data is to introduce
home production. Following Benhabib, Rogerson, and Wright [39], consider
a decision-maker who has preferences

t =0
β
t
u(c
mt
, c
ht
, h
mt
, h
ht
),
where 0 < β < 1. In this expression, c
mt
is the consumption of a market
good; c
ht
is consumption of the home-produced good; h
mt
is labor time spent
in market work; and h
ht
is labor time spent in home work. Assume that
u
1
> 0, u
2
> 0, u
3
< 0, and u
4
< 0. The total amount of time available to
the household is normalized as unity, and leisure is defined as time not spent
working in the market or at home:
l
t
= 1 − h
mt
− h
ht
.
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 55
At each date, the household can purchase market goods c
mt
, market capital
goods k
mt
, and household capital goods k
ht
. Household capital goods are
used in home production, but market capital goods are rented to firms at the
competitive rental rate r
t
. Letting w
t
denote the wage rate and δ
m
and δ
h
denote the depreciation rates on market and home capital, respectively, the
household’s budget constraint is
c
mt
+ k
m,t +1
+ k
h,t +1
≤ w
t
h
mt
+ r
t
k
mt
+ (1 − δ
m
)k
mt
+ (1 − δ
h
)k
ht
.
Home goods are produced according to the home production function:
c
ht
= g(h
ht
, k
ht
, z
ht
),
where z
ht
is a shock to home production and g is increasing and concave in
labor and capital.
Alternative measures put home production to be in the range of 20–50%
of GDP. In the model, home production is used to produce a nontradeable
consumption good. A rise in market productivity may induce households
to substitute away from home production towards market production. This
gives us another margin to substitute market labor and improves the model’s
predictions. Unlike the standard model, the labor supply curve also shifts in
response to a good productivity shock, thereby leading to greater variability
in labor. However, one criticism of the home production theory is that it
suggests that all movements out of the labor force (toward home production)
are voluntary.
Labor Hoarding
A third way to resolve the productivity puzzle is through a labor hoarding
argument. This says that the effective labor input can be altered even though
the total number of workers is fixed. The firm may not alter its work force
every time there is a productivity shock (which would occur through a shift
in the marginal product of labor curve). However, if there are costs to hiring
or laying workers off, then firms may retain workers even though they are
not exerting much effort. Hence, labor effort is likely to be adjusted first in
response to a productivity shock. Eventually more workers may be hired or
fired, but only after longer periods of time.
October 9, 2009 15:12 9in x 6in b808-ch03
56 Business Cycles: Fact, Fallacy and Fantasy
To illustrate the role of labor hoarding, consider a production function for
aggregate output Y
t
which depends on an exogenous technology shock A
t
,
capital K
t
, and a labor input that reflects variations in work effort following
Burnside, Eichenbaum, and Rebelo [52]. Specifically, let f denote a fixed shift
length; N
t
, the total number of workers; and W
t
, the work effort of each
individual. Thus,
Y
t
= A
t
K
α
t
[fN
t
W
t
]
1−α
, 0 < α < 1.
In the standard model, the production function can be written as
Y
t
= S
t
K
α
t
[H
t
]
1−α
,
where H
t
denotes the total hours worked. In the absence of variations in work
effort, H
t
= fN
t
. Notice that the conventionally measured Solow residual S
t
is related to true technology shock A
t
as
ln (S
t
) = ln (A
t
) + αln (K
t
) + (1 − α)[ln (f ) + ln (N
t
) + ln (W
t
)]
−αln (K
t
) − (1 − α)[ln (f ) + ln (N
t
)]
= ln (A
t
) + (1 − α) ln (W
t
).
Decisions to alter effort levels will be the outcome of a maximizing decision, so
that the movements in the Solow residual are not entirely exogenous. If labor
hoarding is included in the model, then the productivity/hours correlation is
reduced and more closely matches the data. The comments made regarding
fluctuations in the effort level of labor also apply to capital. The measured
capital in the Solow residual does not take into account optimal fluctuations
in capital utilization rates. This can lead to variation in the Solow residual that
is not related to changes in productivity or technology.
3.4.3. Reverse Causality
One approach to dealing with this criticism is to introduce money and
banking into the standard RBC model following King and Plosser [129].
These authors observe that transactions services (as provided by money and
the banking sector) can be viewed as an intermediate input that reduces the
cost of producing output and, hence, can be treated as a direct input into the
aggregate production function. To briefly describe their framework, suppose
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 57
the final good is produced according to the production technology
y
t
= f (k
ft
, n
ft
, d
ft

t
, (4.49)
where k
ft
is the amount of capital, n
ft
is the amount of labor services, and
d
ft
is the amount of transactions services used in the final goods industry. In
this expression, φ
t
is a shock to production of the final good at time t . The
financial industry is assumed to provide accounting services that facilitate the
exchange of goods by reducing the amount of time that would be devoted to
market transactions. The production of the intermediate good is given by
d
t
= g(n
dt
, k
dt

t
, (4.50)
where n
dt
and k
dt
denote the amounts of labor and capital allocated to the
financial sector, respectively, and λ
t
captures technological innovations to
the financial services industry. Households maximize the expected discounted
value of utility from consumption c
t
and leisure l
t
as
E
0
_

t =0
β
t
U(c
t
, l
t
)
_
, 0 < β < 1. (4.51)
Households are assumed to own the capital stock and to make investment
decisions i
t
subject to the resource constraint c
t
+ i
t
≤ y
t
+ (1 − δ)k
t
, where
0 < δ < 1 is the depreciation rate on capital. By contrast, firms rent labor,
capital, and transactions services to maximize profits on a period-by-period
basis. Households are also assumed to combine time and transactions services
to accomplish consumption and investment purchases. The time required for
this activity is
n
τt
= τ(d
ht
/(c
t
+ i
t
))(c
t
+ i
t
), (4.52)
where τ

< 0, τ

< 0. The household chooses an amount of transactions
services d
ht
so as to minimize the total transactions costs, w
t
n
τt
+ ρ
t
d
ht
,
where w
t
is the real wage and ρ
t
is the rental price of transactions services.
This implies a demand for transactions services that can be obtained from the
first-order condition
ρ
t
= w
t
τ

(d
ht
/(c
t
+ i
t
)) (4.53)
as d

ht
= h(ρ
t
/w
t
)(c
t
+ i
t
), where h = (τ

)
−1
. Likewise, hours allocated to
producing transactions services is n

τt
= τ

(h(ρ
t
/w
t
))(c
t
+ i
t
). Finally, we
October 9, 2009 15:12 9in x 6in b808-ch03
58 Business Cycles: Fact, Fallacy and Fantasy
require that the household’s time allocated to the different activities sums to
1, n
t
= n
ft
+ n
dt
+ n
τt
.
This framework can be used to rationalize the observations regarding
money and output. Specifically, inside money, or a broad measure of money
that includes commercial credit, is more closely related to output than outside
money. Suppose that a shock to the production of final goods or, equivalently,
a positive shock to productivity occurs. Then, consumption and leisure of the
representative consumer will rise but so will investment demand as consumers
seek to spread the extra wealth over time. If the substitution effect of an increase
in the marginal product of labor outweighs the wealth effect of the productivity
shock, there will be an increase in hours of work. As a consequence, investment
demand rises and output will also increase in response to the additional hours
worked, so firms will wish to finance a greater volume of goods in process.
As a result, commercial credit, or inside money, responds to the positive
productivity shock. The increase in output will also stimulate the demand
for transactions services by households and firms. Hence, the causality runs
from the productivity shock to money even though the increase in money
occurs before the higher productivity shock.
Ahmed and Murthy [3] provide a test of this hypothesis using a small open
economy such as Canada to evaluate the impact of exogenously given terms
of trade and real interest rates versus domestic aggregate demand and supply
disturbances. Their results are derived from a structural vector autoregression
(VAR) with long-run restrictions to identify the alternative structural shocks.
Consistent with the RBC view, they find that an important source of the
money-output correlation is output shocks affecting inside money in the
short run. Another possibility is that the central bank uses the accommodative
monetary policy.
9
During an expansion, if interest rates are rising and firms wish
to invest more in anticipation of higher expected profits, the central bank may
expand the money supply to keep interest rates from rising too rapidly.
3.5. THE SOURCE OF THE SHOCKS
The RBCmodel has come under much criticismfor assuming that technology
or productivity shocks are the sole driving force of cyclical fluctuations. One
9
See Sims [190].
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 59
way of dealing with this criticism is to introduce alternative sources of shocks
into RBCmodels. In the previous section, we discussed the role of government
or fiscal shocks in generating observed co-movements of the data. In this
vein, Christiano and Eichenbaum [65] introduce government consumption
shocks as a way of breaking the strong and positive relationship between hours
worked and productivity implied by the basic RBC model. Other papers that
have employed the RBC framework with fiscal shocks include Braun [46] and
McGrattan [158]. These authors consider the impact of distortionary taxes
on real outcomes. War shocks can also be modeled as a source of cyclical
fluctuations (see Barro [27] and Ohanian [169]).
10
3.5.1. Investment-Specific Technological Shocks
An important source of shocks is the investment-specific technological change.
This has been examined by Greenwood, Hercowitz, and Krusell [106, 107].
Following Gordon [104], these authors motivate their analysis by noting that,
during the period 1950–1990, the relative price of new equipment declined
by 3% at an average annual rate while the equipment-GNP ratio increased
substantially. Furthermore, there was a negative correlation (−0.46) between a
de-trendedmeasure of the relative price of capital andinvestment expenditures.
The notion behind this type of change is that production of capital goods
becomes increasingly efficient over time, thereby stimulating output growth.
11
In Greenwood et al. [107], the role of investment-specific technological
shocks is used to account for cyclical fluctuations. Consider a standard RBC
model with a representative consumer who derives utility from consumption
and leisure as
E
0
_

t =0
β
t
[θ ln (c
t
) + (1 − θ) ln (1 − l
t
)]
_
, 0 < θ < 1,
10
Altug, Demers, and Demers [10] examine the impact of political risk and regime changes due to
changes in the party in power on real investment decisions under irreversibility. Although their analysis is
not general equilibrium, they nevertheless show that political events can have non-negligible effects on real
economic variables.
11
In Greenwood, Hercowitz, and Krusell [106], the authors use a growth accounting approach to
show that investment-specific shocks can account for 60% of postwar US growth of output per man-hour.
October 9, 2009 15:12 9in x 6in b808-ch03
60 Business Cycles: Fact, Fallacy and Fantasy
where c
t
denotes consumption and l
t
denotes labor supply. There is a
production technology that depends on labor and the services from two types
of capital goods, equipment denoted k
e
and structures denoted k
s
. Equipment
is utilized at the time-varying rate h; hence, the service flow is denoted hk
e
.
The production function is given by
y = F(hk
e
, k
s
, l , z) = z(hk
e
)
α
e
k
α
s
s
l
1−α
e
−α
s
, 0 < α
e
, α
s
, α
e
+ α
s
< 1,
where z is a measure of total factor productivity. Structures evolve according
to the standard law of motion:
k

s
= (1 − δ
s
)k
s
+ i
i
, 0 < δ
s
< 1. (5.54)
The law of motion for equipment is given by
k

e
= (1 − δ
e
(h))k
e
+ i
e
q, (5.55)
where the variable q shows the investment-specific technological change in
equipment and
δ
e
(h) =
b
ω
h, ω > 1. (5.56)
Thus, investment in equipment is subject to variable rates of utilization and
depreciation which are convex in the utilization rate. The shocks to technology
and to the efficiency of equipment capital constitute the drivers of the
model. They evolve as z
t +1
= γ
t +1
z
exp (ζ
t +1
) and q
t +1
= γ
t +1
q
exp (η
t +1
),
where ζ
t +1
and η
t +1
are innovations to productivity and investment-
specific technological change, respectively, that follow first-order Markov
processes, and γ
z
and γ
q
denote the average gross growth rates of the two
processes.
Investment in both structures and equipment is also subject to convex costs
of adjustment denoted as a = a
e
+ a
s
, where
a
e
= exp (η)φ
e
(k

e
/q − κ
e
k
e
/q)
2
/(k
e
q), φ
e
, κ
e
> 0, (5.57)
a
s
= φ
s
(k

s
− κ
s
k
s
)
2
, φ
s
, κ
s
> 0. (5.58)
Finally, there is a government which levies taxes on labor and capital income
at the rates τ
l
and τ
k
, respectively. The revenue raised by the government is
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 61
returned to the private sector through lump-sum transfers τ. This yields the
government budget constraint:
τ = τ
k
(r
e
hk
e
+ r
s
k
s
) + τ
l
wl , (5.59)
where r
e
and r
s
denote the rental rate onequipment and structures, respectively,
and w denotes the real wage rate. The resource constraint requires that
consumption plus investment in the two types of capital equal output net
of adjustment costs:
c + i
e
+ i
s
= y − a
e
− a
s
. (5.60)
In this analysis, the investment-specific technological shock plays a key role.
A higher value of q directly affects the quantity of equipment that will be
available to produce output next period. However, it also affects the utilization
cost of existing equipment this period by lowering the replacement cost of old
capital goods, and hence increases its utilization. Hence, investment-specific
technological change increases the services from old equipment at the same
time as it increases the quantity of equipment available for next period.
We note that the economy displays balanced growth as z
t
and q
t
grow at
the rates γ
z
and γ
q
, respectively. In the balanced growth path equilibrium,
the quantity of labor l and the utilization rate h are constant. The resource
constraint and the accumulation equation for structures imply that y, c, i
e
,
i
s
, a
e
, a
s
, and k
s
all have to grow at the same rate; denote this rate by g.
Equipment, however, grows faster, at the rate g
e
= γ
q
g.
12
A key parameter to
be estimated from the data is γ
q
. Since the inverse of q denotes the relative
price of equipment in terms of consumption goods, this quantity is determined
as the ratio of Gordon’s [104] price index of quality-adjusted equipment and
the price index for consumption. The parameters that are determined using
a priori information include the average growth rate of the investment-specific
shock γ
q
, the depreciation rate on structures δ
s
, and the tax rate on wage
income τ
l
. The first parameter is determined using data on the inverse of the
relative price of investment goods following Gordon [104], and it is set at
12
From the production function, we have that g = γ
z

q
g)
α
e
g
α
s
, or
g = γ
1/(1−α
e
−α
s
)
z
γ
α
e
/(1−α
e
−α
s
)
q
,
g
e
= γ
1/(1−α
e
−α
s
)
z
γ
(1−α
s
)/(1−α
e
−α
s
)
q
.
October 9, 2009 15:12 9in x 6in b808-ch03
62 Business Cycles: Fact, Fallacy and Fantasy
1.032. Likewise, δ
s
= 0.056 and τ
l
= 0.40. The second set of parameters is
calculated using information on the long-run behavior of a set of the model’s
variables. This standard calibration exercise yields θ = 0.40, α
e
= 0.18,
α
s
= 0.12, ω = 1.59, τ
k
= 0.53, β = 0.97, g = 1.0124, and bh
ω
= 0.20,
where the parameters b and h cannot be identified separately. The restriction
that adjustment costs are zero along the balanced growth path yields the values
κ
e
= κ
s
= g. Furthermore, the symmetry condition φ
e
= (g/g
e
)
2
φ
s
≡ φ
leaves only the parameter φ to be determined.
The model is simulated to understand the contribution of the investment-
specific technological shocks in generating cyclical fluctuations. Hence, only
the q shock is assumed to be operative. Unlike the standard technology shock
in RBC models, the investment-specific shock affects uses of factors such as
labor and capital through changed investment opportunities. A positive shock
to this variable works by increasing the rate of return to equipment investment,
which tends to raise the stock of equipment next period. Simultaneously, there
is a decline in the equipment’s replacement value, which raises utilization of
equipment today. This leads to increased employment and output. Given
that equipment investment is only 7% of GNP and 18% of the value of
output being derived from the use of equipment, the fraction of output
directly affected by the shock will typically be small. Hence, the impact
of the investment-specific technological shock depends on the quantitative
importance of the transmission mechanism. For this purpose, Greenwood
et al. [107] calibrate the adjustment cost parameter under two alternative
assumptions. According to the first, the parameter φ is set to equalize the
consumption/output correlation implied by the model to that in the data.
This provides a lower bound on the contribution of the q shocks because
a positive shock to q reduces consumption at the same time as it increases
investment. A high value of φ works to reduce investment and to increase the
procyclicality of consumption. By contrast, a low value of the adjustment cost
parameter φ is chosen to make the volatility of investment in the model equal
to that in the data. This constitutes an upper value for the contribution of
the shocks q. They find that under a high value of φ, the investment-specific
technological shocks explain around 28% of business cycle fluctuations as
captured by the standard deviation of output. For a low value of φ, this figure
is 32%. They conclude that though investment-specific shocks contribute to
business cycle fluctuations, they are not the main factor.
October 9, 2009 15:12 9in x 6in b808-ch03
Models of Business Cycles 63
3.5.2. Energy Shocks
From the beginning, the source of large supply-side shocks for the RBC
agenda has been sought in oil shocks. Barsky and Kilian [28] analyze the
relationship between oil shocks and changes in economic activity. Rotemberg
and Woodford [183] provide a rationale for the role of energy price shocks
in a model with imperfect competition. By contrast, Finn [92] argues
that it is possible to rationalize the role of energy shocks in models with
perfect competition. In her model, capital utilization is the channel through
which energy shocks enter the model. We briefly describe her model before
concluding this chapter.
As in the standard RBC framework, there is a representative consumer
with preferences:
E
0
_

t =0
β
t
u(c
t
, l
t
)
_
, 0 < β < 1,
with
u(c
t
, l
t
) =
_
c
α
t
(1 − l
t
)
1−α
_
1−σ
− 1
1 − σ
, 0 < α < 1, σ > 0.
The production function is given by
y
t
= (z
t
l
t
)
θ
(k
t
u
t
)
1−θ
, 0 < θ < 1,
where z
t
is an exogenous technology shock, k
t
is the stock of capital, and u
t
is
the rate of utilization of capital. As in the previous model, the depreciation of
physical capital depends on its utilization rate so that
k
t +1
= (1 − δ(u
t
))k
t
+ i
t
, δ(u
t
) =
ω
0
u
ω
1
t
ω
1
,
where 0 < δ(u
t
) < 1, ω
0
> 0, and ω
1
> 1. The new feature in Finn’s analysis
is that capital utilization requires energy:
e
t
k
t
=
ν
0
u
ν
1
t
ν
1
, ν
0
> 0, ν
1
> 1.
October 9, 2009 15:12 9in x 6in b808-ch03
64 Business Cycles: Fact, Fallacy and Fantasy
We can solve this equation for the utilization rate u
t
and substitute it into the
production function to obtain
y
t
= (z
t
l
t
)
θ
_
k
1−
1
ν
1
t
e
1
ν
1
t
_
ν
1
ν
0
_
1
ν
1
_
1−θ
.
Thus, the production of output depends on labor, capital, and energy. This
shows the direct effect of energy, but there is also an indirect effect which works
through the effect of the utilization rate on the depreciation of capital. Hence,
there are two channels for the transmission of energy shocks in the model.
Finally, the economy-wide resource constraint is given by
c
t
+ i
t
+ p
t
e
t
≤ y
t
,
where p
t
is the price of energy.
Finn [92] describes qualitatively and quantitatively the impact of increases
in the price of energy. First, an increase in p
t
reduces e
t
and u
t
through the
production function. The decrease in e
t
further reduces the marginal product
of labor, and hence the real wage w
t
and work effort l
t
. Thus, a positive energy
price shock operates in a similar way as a negative technology shock. If the price
shock is persistent, then the decrease in the values of e
t
, u
t
, and l
t
continue
into the future and reduce the future marginal productivity of capital. This
tends to reduce the investment i
t
and the future capital stock k
t
. The indirect
effect of the increase in the price of energy on capital’s future marginal energy
cost also tends to reduce investment and the future capital stock. Finn [92]
shows that the quantitative response of value-added and real wages to a shock
in energy prices implied by the model is in line with that in the data.
October 9, 2009 15:12 9in x 6in b808-ch04
Chapter 4
International Business Cycles
The international business cycle literature provides an extension to the
original real business cycle (RBC) approach by allowing for open-economy
considerations. One of the aims of the RBC approach has been to determine
whether two-country versions of the basic model can account for both the
co-movements studied in closed-economy models as well as international
co-movements, including correlations of macroeconomic aggregates across
countries and movements in the balance of trade. The closed-economy model
cannot be used to understand the propagation of shocks across countries or
regions, nor can it be used to discuss notions of international risk sharing.
Unlike the closed-economy model, open-economy business cycle models allow
for the role of technology processes in different countries in generating cyclical
fluctuations. They can be used to model the impact of the ability to trade in
goods and to borrow and lend internationally. In this chapter, we review some
of the facts of international business cycles and discuss models that have been
used to rationalize the findings.
4.1. FACTS
One of the important areas of research has been to derive the so-called
stylized facts of international business cycles. Backus, Kehoe, and Kydland [25]
consider data on12developedcountries over a sample perioddating from1960
to 1990. They consider a two-country RBC model that has the features of the
original Kydland–Prescott model and that assumes a single homogeneous good
andcomplete contingent claims for allocating risk. They study the properties of
a baseline model against the actual features in the data. Backus et al. [26] study
65
October 9, 2009 15:12 9in x 6in b808-ch04
66 Business Cycles: Fact, Fallacy and Fantasy
the properties of the same model for ten developed countries plus a European
aggregate developed by the OECD between 1970 and 1990. The individual
countries they consider are Australia, Austria, Canada, France, Germany, Italy,
Japan, Switzerland, the United Kingdom, and the United States. The data are
filtered using the Hodrick–Prescott filter. Among the important features of
their analysis is the estimation of the correlation properties of international
productivity shocks using data on estimated Solowresiduals. Thus, the exercise
in this literature is to replicate the correlation properties among the different
series given the variability and correlation properties of the shocks. In this
regard, Backus et al. [25, 26] estimate a bivariate AR(1) process for the domestic
and foreign shocks as
_
z
t
z

t
_
= A
_
z
t −1
z

t −1
_
+
_

t


t
_
,
where z
t
, z

t
denote the technology shocks to the domestic and foreign
countries, respectively, and
t
,

t
denote the respective innovations to these
shocks. Based on several estimated specifications for the US and the European
aggregate, these authors use a symmetric specification with
A =
_
0.906 0.088
0.088 0.906
_
,
with Std(
t
) = Std(

t
) = 0.00852 and Corr(
t
,

t
) = 0.258. The positive
off-diagonal entries allow for spillover effects of the shock to productivity in
one country to have a positive effect on the productivity of the other country.
The large autoregressive coefficients displayed on the diagonal capture the
persistence in observed productivity.
The findings can be summarized under two headings:
• The quantity anomaly: In the data, the cross-country correlations of
output are greater than those of productivity, measured as the Solow
residual, and the cross-correlations of productivity are greater than those
of consumption. By contrast, the theoretical model implies the reverse
ranking.
• The price anomaly: The volatility of the terms of trade, defined as the
relative price of imports to exports, is much higher in the data than it is
in the theoretical model.
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 67
Ambler, Cardia, and Zimmermann [14] extend the Backus et al. [26] sample
to 20 industrialized countries over the period 1960–2004. They estimate
the moments of interest using the generalized method of moments (GMM)
estimation, which allows them to obtain standard errors. The findings of
Ambler et al. [14] confirm the findings of Backus et al. [25, 26], though
with lower magnitudes. They also find that the cross-country correlations of
output, investment, and employment are positive and generally high in the
data; whereas in the model, they are negative.
The international RBCmodel produces these results because it implies that
the incentive is to use inputs where they are most productive. This fact leads
to negative cross-correlations between output, investment, and employment
in the theoretical model. In other words, suppose a positive technology shock
occurs in one country, say, the US. This leads to a strong investment response
in the US, which is accompanied by output and employment increases. Since
investment responds strongly to a positive productivity shock, the increase in
investment plus consumption is greater than the increase in output, which
leads to a trade deficit in the domestic country. In a model with perfect capital
mobility, we thus observe a flow of factors from other countries or regions
to the US. By contrast, what is typically observed in the data are positive co-
movements in these series across countries. Furthermore, investment and the
trade balance are more volatile in the model than they are in the data. The
two-country frictionless international business cycle model also implies that
consumers in different countries can share risk perfectly. This feature of the
model leads to the cross-country correlations of consumption being much
higher in the model than in the data. The price anomaly arises because the real
exchange rate in the baseline international RBC model is closely related to the
ratio of consumption across the two countries. With perfect risk sharing, this
ratio displays little volatility, implying that the real exchange rate is also less
volatile in the model than it is in the data.
In their original analysis, Backus et al. [25] also examine a variety of
perturbations of their original setup. First, they consider asymmetric spillovers
between the US and the European aggregate in which the response of US
productivity to shocks to European productivity is smaller than the response of
European productivity to shocks to US productivity. This changes the output-
investment correlation from positive to negative in the domestic country,
but the variability of investment and the trade balance continue to remain
October 9, 2009 15:12 9in x 6in b808-ch04
68 Business Cycles: Fact, Fallacy and Fantasy
counterfactually high, as does the positive correlation of consumption across
countries. Second, they consider

large spillovers across countries by changing
the off-diagonal elements of the A matrix from0.088 to 0.2 and the correlation
of the innovations from 0.258 to 0.5. With this change, they find that
investment and the trade balance become much less volatile, and the cross-
country correlation of output goes from being negative to positive. However,
the cross-country correlation of consumption increases further. They also
examine various sorts of trading frictions. The first of these involves introducing
a small trading friction in the form of a transport cost. They find that this
friction has the effect of reducing the volatility of investment and the trade
balance, but has little effect on the cross-country correlations of consumption
and output. Next, they eliminate all trade in goods and assets by considering
an autarky situation. In this case, it is the correlation between technology
shocks that leads to any connection between countries. They find that the
results are very similar to the case with a small transport cost. Surprisingly,
elimination of trade in state-contingent claims does not help to lower the
cross-country correlations of consumption. Backus et al. [25] argue that this
result is not due to international risk sharing considerations, but stems from
the permanent income hypothesis: when domestic and foreign productivity
shocks are correlated, a rise inproductivity inthe domestic country signals a rise
in productivity in the foreign country through the spillover effects. Hence, the
foreign agent increases consumption and reduces investment in anticipation
of future income increases.
4.2. THE ROLE OF INTERNATIONAL RISK SHARING
Before we continue with other extensions that have been proposed in the
international RBC literature, it is important to understand the role of
alternative financial arrangements in generating the cross-country correlations.
Cole [71] argued that it is important to examine how alternative financial
arrangements interact with the preferences and the production possibilities
set to determine the behavior of the real variables. The standard international
RBC model studied by Backus et al. [25, 26] considers the case with complete
contingent claims and perfect risk sharing across different countries. Cole [71]
and Cole and Obstfeld [72] examine a two-country real version of the Lucas
asset pricing model for this purpose (see Lucas [152]). Agents from both
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 69
countries are identical in terms of preferences and differ only in terms of
endowments. There are two goods, Y
1
and Y
2
. Country 1 has a random
endowment of good Y
1
and country 2 has a random endowment of good
Y
2
. Neither good is storable. We assume that endowments are stationary in
levels.
Let s
t
∈ S ⊆ R
m
+
denote a vector of exogenous shocks that follows a
discrete first-order Markov process with a stationary probability transition
matrix P with (i, j) element p
i,j
, where p
i,j
= p(s
t
= s
j
|s
t −1
= s
i
). Let ˆ π
denote the vector of stationary probabilities which satisfy ˆ π = P

ˆ π. In the
beginning, agents observe the current realization. We assume that endowment
is a time-invariant function of the exogenous shock.
Assumption 2.1. Define Y ≡ [y, ¯ y], where y > 0 and ¯ y < ∞. The functions
y
1
: S → Y and y
2
: S → Y are continuous functions that are bounded away
from zero.
The representative consumer in country j has preferences over random
sequences {c
j
1,t
, c
j
2,t
}

t =0
defined by
E
0
_

t =0
β
t
U(c
j
1,t
, c
j
2,t
)
_
, 0 < β < 1. (2.1)
We have the following assumption.
Assumption 2.2. The utility function U : R
2
+
→ R is continuously
differentiable, strictly increasing, and strictly concave. For all c
1
, c
2
> 0,
lim
c
1
→0
U
1
(c
1
, c
2
)
U
2
(c
1
, c
2
)
= ∞, lim
c
2
→0
U
1
(c
1
, c
2
)
U
2
(c
1
, c
2
)
= 0.
This requirement on the utility function ensures that both goods are consumed
in equilibrium.
4.2.1. Pareto Optimal Allocations
We begin by characterizing the Pareto optimal allocations. To do this, we
can examine the solution to a central planning problem which maximizes the
weighted utility of agents in the domestic and foreign countries subject to the
October 9, 2009 15:12 9in x 6in b808-ch04
70 Business Cycles: Fact, Fallacy and Fantasy
resource constraints. Let φ
j
denote the Pareto weight for country j. Define
π(s
t
) = π(s
t
|s
t −1
)π(s
t −1
|s
t −2
) · · · π(s
1
|s
0
) ˆ π(s
0
),
where s
t
= (s
t
, s
t −1
, . . . , s
0
) denotes the history of the shocks at time t . The
social planner chooses sequences {c
1
i
(s
t
), c
2
i
(s
t
)}

t =0
to maximize the weighted
sum of utilities subject to the sequence of resource constraints

t =0

s
t
π(s
t

t
_
φ
1
U(c
1
1
(s
t
), c
1
2
(s
t
)) + φ
2
U(c
2
1
(s
t
), c
2
2
(s
t
))
_
(2.2)
subject to
c
1
i
(s
t
) + c
2
i
(s
t
) = y
i
(s
t
), i = 1, 2.
Since preferences are time-separable, the endowment is nonstorable, and there
is no investment process, the social planner’s problemat time t involves solving
the static problem
max[φ
1
U(c
1
1,t
, c
1
2,t
) + φ
2
U(c
2
1,t
, c
2
2,t
)] (2.3)
subject to
c
1
i,t
+ c
2
i,t
= y
i,t
, i = 1, 2,
where c
j
i,t
denotes the consumption of good i by residents of country j. Let
µ
i
(s
t
) denote the Lagrange multiplier for the resource constraint for good i.
The first-order conditions imply that
φ
j
U
1
(c
j
1,t
, c
j
2,t
) = µ
1
(s
t
),
φ
j
U
2
(c
j
1,t
, c
j
2,t
) = µ
2
(s
t
)
for j = 1, 2, or
U
1
(c
1
1,t
, c
1
2,t
)
U
1
(c
2
1,t
, c
2
2,t
)
=
φ
2
φ
1
, (2.4)
U
2
(c
1
1,t
, c
1
2,t
)
U
2
(c
2
1,t
, c
2
2,t
)
=
φ
2
φ
1
. (2.5)
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 71
Hence, we find that the ratio of marginal utilities across agents in different
countries is equalized for all states and goods. This is an implication of perfect
risk sharing that arises in a complete contingent claims equilibrium.
It is convenient to assume that preferences are of the Cobb–Douglas variety:
U(c
1,t
, c
2,t
) =
(c
α
1,t
c
1−α
2,t
)
1−ρ
1 − ρ
. (2.6)
Assuming that preferences are of Cobb–Douglas variety, the Pareto optimal
consumption allocations satisfy
c
1
1,t
= δy
1,t
(2.7)
c
1
2,t
= δy
2,t
(2.8)
c
2
1,t
= (1 − δ)y
1,t
(2.9)
c
2
2,t
= (1 − δ)y
2,t
, (2.10)
where
δ =
_
1 +
_
φ
2
φ
1
_
σ
_
and σ =
1
ρ
.
These expressions show that consumers in each country consume a
constant fraction of output in each period. Also note that a shock to the
output of country 1, which also causes the output of country 2 to increase
through positive spillover effects, will cause consumption in both countries
to increase. This implication of the model has been shown to lead to the
counterfactually high cross-country correlations of consumption implied by
the standard international RBC model. In what follows, we will examine the
role of alternative financial arrangements that can be used to rationalize the
observations.
4.2.2. Complete Contingent Claims
Consider first the behavior of allocations in a competitive equilibriumin which
agents can trade claims that pay off contingent on all possible realizations of
the income processes in each country. It is straightforward to demonstrate
October 9, 2009 15:12 9in x 6in b808-ch04
72 Business Cycles: Fact, Fallacy and Fantasy
that there exists a contingent claims equilibrium for this economy, and that
the competitive equilibrium allocations are Pareto optimal.
Consider first the case when agents can trade one-period contingent claims
that pay off in each state next period on the consumption good of each country.
Given the Markov nature of uncertainty, it is without loss of generality to
consider only one-period contingent claims.
1
Agents can also trade in the
goods of the other country. Let p(s
t
) denote the relative price of good y
2
in
terms of the numéraire good y
1
. The sequential budget constraints for residents
of country j for j = 1, 2 are given by
y
1,t
+ z
1
1,t
+ p(s
t
)z
1
2,t
= c
1
1,t
+ p(s
t
)c
1
2,t
+

s
t +1
∈S
q(s
t +1
, s
t
)
_
z
1
1
(s
t +1
) + p(s
t
)z
1
2
(s
t +1
)
_
(2.11)
y
2,t
+
z
2
1,t
p(s
t
)
+ z
2
2,t
=
c
2
1,t
p(s
t
)
+ c
2
2,t
+

s
t +1
∈S
q(s
t +1
, s
t
)
_
z
2
1
(s
t +1
)
p(s
t
)
+ z
2
2
(s
t +1
)
_
, (2.12)
where q(s
t +1
, s
t
) is the price of a contingent claimthat pays off in each possible
state next period conditional on the state today and z
j
i
(s
t +1
) are the holdings
of country i’s assets by consumers from country j for i, j = 1, 2. Thus, these
equations indicate that the output produced in country 1 plus the receipts (or
payments) on assets denominated in the goods of countries 1 and 2 must be
equal to consumption of domestic goods plus imports of foreign goods plus
purchases (sales) of contingent assets denominated in the goods of countries
1 and 2.
The current account for each country, denoted as CA
j
, is given by
CA
1
= y
1,t

_
c
1
1,t
+ p(s
t
)c
1
2,t
_
,
CA
2
= y
2,t

_
c
2
1,t
p(s
t
)
+ c
2
2,t
_
,
1
For further discussion, see Altug and Labadie [6], Ch. 7.
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 73
and the capital account denoted KA
j
is given by
KA
1
= z
1
1,t
+ p(s
t
)z
1
2,t

s
t +1
∈S
q(s
t +1
, s
t
)
_
z
1
1
(s
t +1
) + p(s
t
)z
1
2
(s
t +1
)
_
,
KA
2
=
z
2
1,t
p(s
t
)
+ z
2
2,t

s
t +1
∈S
q(s
t +1
, s
t
)
_
z
2
1
(s
t +1
)
p(s
t
)
+ z
2
2
(s
t +1
)
_
.
We can interpret the budget constraints for each country j by stating that the
sum of the current account plus the capital account must be zero:
CA
j
+ KA
j
= 0, j = 1, 2.
In an economy with trade in international assets, a current account deficit
(CA < 0) must be balanced with a capital account surplus (KA > 0). This
implies that a country which consumes more than it produces must be a net
borrower or, equivalently, it must sell more assets to the rest of the world than
it purchases from the rest of the world.
The market-clearing conditions are given by
c
1
i,t
+ c
2
i,t
= y
i,t
,
z
1
i,t
+ z
2
i,t
= 0,
z
1
i
(s
t +1
) + z
2
i
(s
t +1
) = 0, s
t +1
∈ S
for i = 1, 2.
Given the Markov nature of uncertainty, we can define the value function
for the problem of each consumer in country j as
V
_
s
t
, z
j
1,t
, z
j
2,t
_
= max
{c
j
i,t
,z
j
i,t +1
}
i=1,2
_
_
_
U
_
c
j
1,t
, c
j
2,t
_

s
t +1
∈S
π(s
t +1
|s
t
)V
_
s
t +1
, z
j
1,t +1
, z
j
2,t +1
_
_
_
_
subject to the budget constraint (2.11) (or (2.12)). Let λ
i
(s
t
) denote the
Lagrange multiplier for country i in state s
t
. The first-order conditions with
October 9, 2009 15:12 9in x 6in b808-ch04
74 Business Cycles: Fact, Fallacy and Fantasy
envelope conditions substituted in are given by
U
1
(c
1
1,t
, c
1
2,t
) = λ
1
(s
t
),
U
2
(c
1
1,t
, c
1
2,t
) = p(s
t

1
(s
t
),
U
1
(c
2
1,t
, c
2
2,t
) =
λ
2
(s
t
)
p(s
t
)
,
U
2
(c
2
1,t
, c
2
2,t
) = λ
2
(s
t
),
λ
i
(s
t
)q(s
t +1
, s
t
) = βπ(s
t +1
|s
t

i
(s
t +1
), i = 1, 2.
Simplifying these conditions yields
U
2
(c
i
1,t
, c
i
2,t
)
U
1
(c
i
1,t
, c
i
2,t
)
= p(s
t
), i = 1, 2 (2.13)
and
βπ(s
t +1
|s
t
)U
1
(c
i
1,t +1
, c
i
2,t +1
)
U
1
(c
i
1,t
, c
i
2,t
)
= q(s
t +1
, s
t
), i = 1, 2. (2.14)
Thus, we see that the Pareto optimal allocations can be supported in a complete
contingent claims equilibrium if
λ
i
(s
t
) =
µ
i
(s
t
)
φ
i
, i = 1, 2, where p(s
t
) =
µ
2
(s
t
)
µ
1
(s
t
)
.
In this equilibrium, we find that consumers in each country equate their
marginal rates of substitution for each good across all possible current states.
If preferences satisfy the Cobb–Douglas assumption, then consumers in
each country consume a constant fraction of current output as specified
in equations (2.7)–(2.10). This follows from the fact that the competitive
equilibrium is Pareto optimal. Hence,
p(s
t
) =
(1 − α)c
i
1,t
αc
i
2,t
=
(1 − α)y
1,t
αy
2,t
.
In this case, the contingent claims contracts are considered to be consistent
with the results on consumption for each country. Given the solution for
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 75
the relative price p(s
t
) and the consumption allocations given in equations
(2.7)–(2.10), notice that the solution
z
i
1
(s
t +1
) = z
i
2
(s
t +1
) = 0, z
i
1,t
= z
i
2,t
= 0
with δ = α satisfies the consumers’ budget constraints and the market-clearing
conditions. But this is a contingent claims equilibrium in which no assets are
traded! In such an equilibrium, it is still possible to price the contingent
claims using the expression in (2.14). In the next section, we will demonstrate
explicitly that the Pareto optimal allocations can indeed be attained in a no-
asset-trading equilibrium. Since the price of any financial asset can be obtained
as a function of the contingent claims prices, it follows that any financial asset
can also be priced in such an equilibrium.
4.2.3. No Asset Trading
Now suppose that there is trade in goods but no trade in international assets.
In this case, we are forcing the current account to equal zero in each period.
Country 1 and 2’s budget constraints can be expressed, respectively, as
c
1
1,t
+ p
t
c
1
2,t
= y
1,t
, (2.15)
c
2
1,t
p
t
+ c
2
2,t
= y
2,t
. (2.16)
Once again, this is a static problembecause there are no assets for intertemporal
consumption smoothing and the endowment is nonstorable. Let µ
j
t
denote
the Lagrange multiplier on the budget constraint of country j. If we assume
the Cobb–Douglas functional form for preferences, then the consumption
allocations are
c
1
1,t
= αy
1,t
,
c
1
2,t
=
(1 − α)y
1,t
p
t
,
c
2
1,t
= αp
t
y
2,t
,
c
2
2,t
= (1 − α)y
2,t
.
October 9, 2009 15:12 9in x 6in b808-ch04
76 Business Cycles: Fact, Fallacy and Fantasy
Equilibrium in the two goods markets requires that
c
1
1,t
+ c
2
1,t
= y
1,t
, (2.17)
c
1
2,t
+ c
2
2,t
= y
2,t
. (2.18)
Substitute the consumption functions into the market-clearing conditions and
solve for the relative price to show that
p
t
=
(1 − α)y
1,t
αy
2,t
.
This price can be substituted into the consumption functions above to
show that
c
1
2,t
= αy
2,t
,
c
2
1,t
= (1 − α)y
1,t
.
Notice that the no-asset-trading allocation is identical to the central planning
allocation if
α = δ =
_
1 +
_
φ
2
φ
1
_
σ
_
,
or
φ
1
=
_
1 +
_
1 − α
α
_
ρ
_
−1
and φ
2
= 1 − φ
1
.
This exercise illustrates several points:
• The absence of international capital mobility does not necessarily imply
that the allocation is not Pareto optimal. Efficient risk sharing can occur
despite the lack of financial assets and insurance.
• The international ratio of marginal utilities across countries is identical
across goods and states. A large and positive shock in the amount of good
y
1,t
is positively transmitted to the residents of country 2 by the increase in
demand (and hence the relative price) for good y
2,t
. Alarge negative shock
in the amount of a good is similarly transmitted across borders, despite
the absence of trade in financial assets. Hence, efficient risk sharing occurs
through changes in the relative price of goods.
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 77
• Notice that we can price financial assets in the model, under the
assumption that the current account is zero, and can showthat real interest
rates and real asset returns will be equal for the two countries, despite the
absence of financial capital mobility.
• The total consumption of countries 1 and 2 is
c
1
1,t
+ p
t
c
1
2,t
= αy
1,t
+ αy
2,t
,
c
2
1,t
+ p
t
c
2
2,t
= (1 − α)y
1,t
+ (1 − α)y
2,t
.
Notice that correlation of the total value of consumption of country 1
and country 2 is positive and equal to one.
As Cole and Obstfeld [72] point out, the positive transmission of shocks occurs
because countries specialize in the production of goods.
4.2.4. Nonspecialization in Endowments
To illustrate the impact of nonspecialization, Cole and Obstfeld [72] introduce
a third good. Call this third good w. Assume that both countries receive an
exogenous and stochastic endowment of good w and let w
j
: S → W =
[w, ¯ w] denote the realization of good w in country j. Let α
1
, α
2
, α
w
denote
the expenditure shares under the assumption of Cobb–Douglas preferences
and let w be the numéraire good, so that p
1,t
denotes the relative price of good
y
1,t
in terms of w
t
and p
2,t
denotes the relative price of good y
2,t
in units of
w
t
. Agents in countries 1 and 2 have budget constraints given, respectively, by
p
1,t
c
1
1,t
+ p
2,t
c
1
2,t
+ c
1
w,t
≤ p
1,t
y
1,t
+ w
a
t
,
p
1,t
c
2
1,t
+ p
2,t
c
2
2,t
+ c
2
w,t
≤ p
2,t
y
2,t
+ w
b
t
.
The equilibrium relative prices satisfy
p
1,t
=
α
1
[w
1
t
+ w
2
t
]
α
w
y
1,t
,
p
2,t
=
α
2
[w
1
t
+ w
2
t
]
α
w
y
2,t
.
October 9, 2009 15:12 9in x 6in b808-ch04
78 Business Cycles: Fact, Fallacy and Fantasy
The consumption of good 1 by agents in countries 1 and 2, under the
assumption of Cobb–Douglas preferences, can be shown to satisfy
c
1
1,t
=
_
α
w
_
w
1
t
w
1
t
+ w
2
t
_
+ α
1
_
y
1,t
,
c
2
1,t
=
_
α
w
_
w
2
t
w
1
t
+ w
2
t
_
+ α
2
_
y
1,t
.
Similar expressions can be derived for the consumption of goods y
2
and w.
The ratio of marginal utilities of both countries for each good will be equal
to a constant across all states, a condition for Pareto optimality, only if the
share of the endowment of w
t
, (
w
1
t
w
1
t
+w
2
t
) and (
w
2
t
w
1
t
+w
2
t
), is constant as the total
w
t
varies with s
t
. The shocks to w
1
t
and w
2
t
must be perfectly correlated for
the allocation with no trade in financial assets to be Pareto optimal. If these
shocks are not perfectly correlated, then it is beneficial to trade equity shares
or other forms of financial assets.
This helps to clearly distinguish between country-specific shocks, which
affect all sectors within a country, and industry-specific shocks. Shocks to y
1,t
or y
2,t
are, by definition, country-specific shocks; whereas shocks to w
1
t
, w
2
t
,
where w
1
t
, w
2
t
are not perfectly correlated, are sector-specific shocks. Hence,
when there are sector-specific shocks, there are gains to asset trading that
improve risk sharing and allow diversification. The intuition is that, in the
absence of trade in financial assets, the country with a negative shock to the
endowment of w
t
would like to run a current account deficit by importing
w
t
and borrowing against future endowment. Since the current account must
always be balanced, the country must export more of the good in which it
specializes in production to finance the import of good w
t
. Notice that the
relative price of w
t
may not adjust much if w
1
t
and w
2
t
are negatively correlated
but the sum w
1
t
+ w
2
t
fluctuates very little.
4.2.5. Nontraded Goods
Suppose now that the third good is a nontraded good. Call this good n and
assume that n
j
: S → N = [n, ¯ n] denotes the realization of good n in
country j. Let α
1
, α
2
, α
n
denote the expenditure shares under the assumption
of Cobb–Douglas preferences and let y
1
be the numéraire good. Under the
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 79
assumption of balanced trade and Cobb–Douglas preferences, the demands
for the goods satisfy
c
1
1,t
=
α
1
1 − α
n
y
1,t
,
c
1
2,t
=
α
1
1 − α
n
y
2,t
,
c
2
1,t
=
α
2
1 − α
n
y
1,t
,
c
2
2,t
=
α
2
1 − α
n
y
2,t
.
Each country consumes its endowment of the nontraded good, n
1
t
, n
2
t
. The
ratio of marginal utility across countries for a traded good will now depend
on the ratio
n
1
t
n
2
t
. Unless n
1
t
, n
2
t
are perfectly correlated, then the resulting
allocations will not be Pareto optimal. There are gains from international risk
sharing through asset trade. Notice that the correlation of consumption across
countries will now depend on the proportion of a country’s consumption that
is nontradeable. If this sector constitutes a large fraction of consumption, then
even if consumption of traded goods is perfectly correlated, the correlation
of national consumption levels may be close to zero. Thus, nontraded goods
provide one vehicle for reducing the large cross-country correlations implied
by the baseline model.
4.2.6. Trade in Equity Shares
We nowintroduce trade in financial assets. An agent in country 1 holds equity
shares that are claims to the endowment stream for good y
1
(the domestic
good) and claims to y
2
(the foreign good). We now discuss the impact of asset
trading and relate it to the model without asset trade discussed earlier.
Let z
j
i,t
for j = 1, 2 and i = 1, 2 denote the shares of good i held
by an agent in country j at the beginning of period t . An agent’s budget
constraint is
z
j
1,t
[y
1,t
+ q
1,t
] + z
j
2,t
[p
2,t
y
2,t
+ q
j
2,t
]
≥ c
j
1,t
+ p
2,t
c
j
2,t
+ q
1,t
z
j
1,t +1
+ q
2,t
z
j
2,t +1
. (2.19)
October 9, 2009 15:12 9in x 6in b808-ch04
80 Business Cycles: Fact, Fallacy and Fantasy
Let µ
j
t
denote the Lagrange multiplier. The agent maximizes his objective
function subject to the constraint. The first-order conditions are
U
1
(c
j
1,t
, c
j
2,t
) = U
2
(c
j
1,t
, c
j
2,t
)p
2,t
, (2.20)
U
1
(c
j
1,t
, c
j
2,t
)q
1,t
= βE
t
U
1
(c
j
1,t +1
, c
j
2,t +1
)[q
1,t +1
+ y
1,t +1
], (2.21)
U
1
(c
j
1,t
, c
j
2,t
)q
2,t
= βE
t
U
1
(c
j
1,t +1
, c
j
2,t +1
)[q
2,t +1
+ p
2,t +1
y
2,t +1
]. (2.22)
In equilibrium, all equity shares are held and the endowment of each good
is completely consumed. Lucas [152] assumes that agents hold identical
portfolios, so that z
j
i,t
= 1/2 for j = 1, 2 and i = 1, 2 so that φ
j
= 1/2
and δ = 1/2. In such a world, national wealth is equal across countries
and agents have perfectly diversified portfolios. Agents across countries have
identical consumption in this case, unlike the economy in which there is no
trade in financial assets. In the initial model described earlier, there was no
trade in financial assets and specialization in endowments and the allocation
was Pareto optimal. We commented that we could price financial assets even
if these assets were not traded. If we assume that z
1
1,t
= z
1
2,t
= α and
z
2
1,t
= z
2
2,t
= 1 − α, then the equilibrium allocation with no trade in
financial assets can be achieved. Hence, we can achieve at least two stationary
allocations, depending on the initial distribution of the claims. This simple
example, when combined with our discussion of the equilibriumwith no trade
in financial assets, illustrates an important point. International risk sharing can
be achieved through fluctuations in relative prices in the current account and
by trade in financial assets. In particular, it does not require the existence
of a full set of contingent claims. The presence of nontraded goods or lack
of specialization in production of a good can affect how much consumption
insurance can be achieved through relative price fluctuations. If we introduced
trade in equity shares when there is a third good w that is produced by
both countries, then portfolio diversification may require that an agent hold
equity shares for w
1
and w
2
if the endowment shocks for w are not perfectly
correlated. If these shocks are perfectly correlated, thenportfolio diversification
may be achieved by specializing in the holding of equity shares of one
country only.
There has been substantial literature on the lack of international portfolio
diversification and the degree of international consumption risk sharing. The
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 81
international portfolio diversification puzzle is the notion that investors hold
too little of their wealth in foreign securities to be consistent with the standard
theory of portfolio choice. Baxter and Jermann [36] argue that the failure of
international diversification is substantial. Their model incorporates human
and physical capital and, within the context of their model, optimal behavior
would lead to a short position in domestic assets because of a strong positive
correlation between the returns to human and physical capital. A more recent
paper by Heathcote and Perri [117] extends the Baxter and Jermann model to
include more than one traded good. They find, as we have noted above, that
consumption insurance is available through relative price fluctuations and that
these price fluctuations are capable of achieving efficient risk sharing. Clearly,
the conclusion on whether there is sufficient or insufficient risk sharing is very
sensitive to model specifications.
Empirical evidence on international consumption risk sharing is provided
in Backus et al. [25], who show that the data are inconsistent with the
implications for consumption for the baseline international RBC model. By
contrast, Devereux, Gregory, and Smith [81] showthat preferences that display
a nonseparability between consumption and leisure can help to reduce the
consumption correlations implied by the model. Lewis [145] also documents
that there is insufficient intertemporal risk sharing in consumption. As we have
noted above, the existence of nontraded goods, combined with the assumption
that utility is nonseparable in traded and nontraded goods, makes it more
difficult to determine the optimal degree of consumption risk sharing. We
have also shown above that relative price fluctuations can be a substitute for
trade in financial assets in achieving consumption insurance. Lewis [145]
documents that the nonseparability of utility or the restriction of asset trade
alone is not enough to explain the risk sharing that we observe, but that when
nonseparability and asset trade restrictions are combined, she cannot reject the
hypothesis that there is risk sharing.
4.2.7. Limited Risk Sharing
As described earlier, whether there exists perfect risk sharing in the data appears
to depend on the model specification adopted by the researcher. Nevertheless,
one could ask whether limited risk sharing opportunities among consumers
in different countries could help to better reconcile the data with the model.
October 9, 2009 15:12 9in x 6in b808-ch04
82 Business Cycles: Fact, Fallacy and Fantasy
In this vein, credit market frictions and restrictions on international capital
flows constitute a proposed channel for the propagation of international
shocks. Kollman [136] and Baxter and Crucini [34] examine models where
only non-contingent bonds can be traded internationally. These authors find
that incomplete asset markets help to reduce the cross-country correlation
of consumption, but the cross-country correlations of output, investment,
and hours worked remain counterfactually negative. Moreover, the results for
the cross-country correlations are obtained only if the productivity shocks
are highly persistent and there are very little spillover effects across countries.
Heathcote and Perri [117] examine the implications of a two-country model
under alternative assumptions about the financial structure, given a trade
structure. They show that the model matches the correlations in the data
under a financial autarky assumption. Their analysis involves extending the
analysis in Cole and Obstfeld [72] to incorporate an explicit production side.
Kehoe and Perri [126] consider a model where market incompleteness is
obtained endogenously through the introduction of imperfectly enforceable
international loans. In their framework, a country can incur international
indebtedness only to the extent that such indebtedness can be enforced
through the threat of exclusion fromfuture intertemporal and interstate trade.
This analysis builds on the earlier works of Kehoe and Levine [124, 125],
Kocherlakota [135], Alvarez andJermann[12], andothers ondebt-constrained
asset markets. In these models, the inability of agents to share risk perfectly and
to fully offset the effects of idiosyncratic shocks leads to lower cross-country
correlations of consumption and higher cross-country correlations of output.
Kehoe and Perri [126] consider a standard international RBC model with
production and capital accumulation. Output in each country is produced
using capital and labor according to the constant-returns-to-scale production
function:
F(k
i
(s
t −1
), A
i
(s
t
)l
i
(s
t
)),
where A
i
(s
t
) is a randomshock. The preferences of the representative consumer
in each country are given by

t =0

s
t
π(s
t
)U(c
i
(s
t
), l
i
(s
t
)),
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 83
where c
i
(s
t
) denotes consumption of residents of country i. Under complete
markets, the competitive equilibrium allocations solve the social planner’s
problem defined as

t =0

s
t
π(s
t

t
_
φ
1
U(c
1
(s
t
), l
1
(s
t
)) + φ
2
U(c
2
(s
t
), l
2
(s
t
))
_
, (2.23)
subject to the resource constraints

i=1,2
_
c
i
(s
t
) + k
i
(s
t
)
_
=
2

i=1
_
F(k
i
(s
t −1
), A
i
(s
t
)l
i
(s
t
)) + (1 − δ)k
i
(s
t −1
)
_
. (2.24)
The innovation in Kehoe and Perri [126] is to formulate a version of
this problem that allows for enforcement constraints. These require that each
country prefers the allocation it receives to the one that it could attain if it
were in (financial) autarky from then onwards. The enforcement constraints
are expressed as

r=t

s
r
β
r−t
π(s
r
|s
t
)U(c
i
(s
r
), l
i
(s
r
)) ≥ V
i
(k
i
(s
t −1
), s
t
), (2.25)
where π(s
r
|s
t
) denotes the conditional probability of the history s
r
given
s
t
and V
i
(k
i
(s
t −1
), s
t
) denotes the value of autarky from s
t
onwards, where
V
i
(k
i
(s
t −1
), s
t
) involves choosing k
i
(s
r
), l
i
(s
r
), c
i
(s
r
) for r ≥ t to solve
V
i
(k
i
(s
t −1
), s
t
) = max

r=t

s
r
β
r
π(s
r
|s
t
)U(c
i
(s
r
), l
i
(s
r
))
subject to
c
i
(s
r
) + k
i
(s
r
) ≤ F(k
i
(s
r−1
), A
i
(s
r
)l
i
(s
r
)) + (1 − δ)k
i
(s
r−1
), r ≥ t
given k
i
(s
t −1
).
As Kehoe and Perri [126] explain, the social planner’s problem with
enforcement constraints cannot be formulated recursively as a dynamic
programming problem. The reason is that future decision variables such as
c
i
(s
r
) and l
i
(s
r
) for r > t enter the current enforcement constraints. An
October 9, 2009 15:12 9in x 6in b808-ch04
84 Business Cycles: Fact, Fallacy and Fantasy
alternative approach involves adding a new pseudo state variable to achieve a
recursive formulation of the original problem.
2
This approach is implemented
by defining β
t
π(s
t

i
(s
t
) as the Lagrange multiplier on the enforcement
constraints. The Lagrangian function becomes comprised of three terms: the
weighted sum of utilities defined in (2.23), the standard resource constraints
defined by (2.24), plus the term
β
t
π(s
t

i
(s
t
)
_

r=t

s
r
β
r−t
π(s
r
|s
t
)U(c
i
(s
r
), l
i
(s
r
)) − V
i
(k
i
(s
t −1
), s
t
)
_
.
Since π(s
r
) = π(s
r
|s
t
)π(s
t
), the Lagrangian function is written as

t =0

s
t

i
β
t
π(s
t
)
_
M
i
(s
t −1
)U(c
i
(s
t
), l
i
(s
t
))

i
(s
t
)[U(c
i
(s
t
), l
i
(s
t
)) − V
i
(k
i
(s
t −1
), s
t
)]
_
plus the terms relating to the resource constraints. In this expression, M
i
(s
t −1
)
satisfies
M
i
(s
t
) = M
i
(s
t −1
) + µ
i
(s
t
), t ≥ 0
with M
i
(s
t −1
) = φ
i
. Thus, M
i
(s
t
) are just the original planning weights plus
the sum of the multipliers µ
i
(s
t
) along the path s
t
.
It is beyond the scope of this book to derive a solution for the model
with enforcement constraints. Therefore, we will focus on the results of
simulating this model and compare it with alternatives that have been
obtained in the literature. Kehoe and Perri [126] generate solutions for three
different economies: complete markets, a bondeconomy, andaneconomy with
enforcement constraints. The bond economy stipulates that all intertemporal
trades must be implemented with an uncontingent bond. Thus, the budget
constraints for households in each country are expressed as
c
i
(s
t
) + k
i
(s
t
) + q(s
t
)b
i
(s
t
) ≤ w
i
(s
t
)l
i
(s
t
)
+[r
i
(s
t
) + (1 − δ)]k
i
(s
t −1
) + b
i
(s
t −1
),
2
For other applications, see Marcet and Marimon [155].
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 85
where w
i
(s
t
) and r
i
(s
t
) denote the wage rate and the rental rate on capital in
country i, q(s
t
) is the period t price of an uncontingent that pays off one unit
in period t +1 regardless of the state, and b
i
(s
t
) is the quantity of the bond by
consumers in country i. There is also a borrowing constraint that requires that
borrowing cannot exceed some finite bound, b(s
t
) ≥ −
¯
b, where 0 <
¯
b < ∞.
Preferences in each country are taken to be of the form
U(c, l ) =
_
c
γ
(1 − l )
1−γ
_
1−σ
1 − σ
, 0 < γ < 1, σ ≥ 0, (2.26)
and the production function as
F(k, AL) = k
α
(Al )
1−α
, 0 < α < 1. (2.27)
The parameter values that are considered are standard, and given by β =
0.99, γ = 0.36, σ = 2, α = 0.36, and δ = 0.025. The coefficients of
the A matrix, which governs the persistence properties of the shocks and the
spillovers between them, are parameterized in different ways. According to one
parameterization intended to capture substantial persistence, the autoregressive
coefficients of the shocks for each country captured by the diagonal elements
of the A matrix are set at 0.9 and the off-diagonal elements are set at zero. These
values are consistent with those assumed by Kollman [136] and Baxter and
Crucini [34]. A second parameterization allows for high persistence with the
diagonal elements equal to 0.99 and the off-diagonal elements equal to zero,
and a third one allows for high spillover with the diagonal elements equal to
0.85 and the off-diagonal elements equal to 0.15. Similar to Backus et al. [25],
the variance of the innovations to the productivity shocks in each country is
set at Var(
i,t
) = 0.007 and the correlation of the shocks is set at 0.25.
The findings from the model echo many of the earlier findings. Under
the complete markets specification, the model displays many of the anomalies
reported in the literature. First, the consumption correlations are significantly
higher than the output correlations in the model, whereas these correlations are
the opposite in the data. Second, the cross-country correlations of employment
and investment are negative in the model, whereas they are positive in the data.
Third, net exports and investment are much more volatile in the model than
they are in the data. In the bond economy, the three discrepancies between
the model and the data remain, albeit with some minor improvements in the
various statistics. In the economy with enforcement constraints, the output and
October 9, 2009 15:12 9in x 6in b808-ch04
86 Business Cycles: Fact, Fallacy and Fantasy
consumption correlations are both positive and closer to each other, although
the cross-country correlation of consumption is greater than that of output.
Second, the cross-country correlations of employment and investment have
now switched from being negative to positive; and third, the volatility of
net exports and investment has declined dramatically. The only remaining
discrepancy is that net exports are procyclical in the model whereas they are
countercyclical in the data.
As discussed earlier, the frictionless international RBC model implies that
investment flows to the country with the higher productivity shock, leading
to very volatile investment and net exports. In the literature, exogenous
adjustment costs are typically added to inhibit the flow of investment. Kehoe
and Perri [126] note that the enforcement constraint acts as an endogenous
inhibiting factor. Considering economies with exogenous adjustment costs to
investment, they find that the volatilities of investment and net exports are
diminished, but that the anomalies regarding the cross-correlations of output,
consumption, investment, and employment remain. Finally, the authors
examine the response of the variables in domestic and foreign countries to
a positive shock to productivity in the domestic country. This increases the
productivity of both capital and labor in the domestic country, so resources are
optimally shifted there. The capital stock in the domestic country increases, as
domestic residents save more and also as investment from abroad flows there.
The net flow of investment increases the trade deficit in the domestic country.
In the foreign country, we notice declines in employment and investment.
Thus, we see that the differing responses of the domestic versus the foreign
country lead to the negative cross-country correlations of employment and
investment. Since residents of each country can acquire contingent claims
that allow them to smooth consumption across all possible history of shocks,
risk sharing across countries implies that consumption increases in the foreign
country. Consumption increases substantially more in the domestic country
because consumptionandlabor complement eachother. Hence, we observe the
highcross-country correlations betweendomestic andforeignconsumption. In
the bond economy, the responses are similar to those for the complete markets
economy, but because of restrictions on asset trading, they are somewhat
dampened.
Finally, let us consider the impact of a positive productivity shock in the
enforcement economy. Suppose that the social planner tries to implement
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 87
the complete markets allocation for this economy. The high and persistent
productivity shock in the domestic country increases the value of autarky,
increasing the incentives for default. The increased investment flows to the
domestic country further increase the value of autarky. Hence, the planner
must restrict investment to the domestic country to prevent the default option
from being exercised. Furthermore, the planner also builds up the capital
stock in the foreign country to ensure that the risk-sharing arrangement
between the domestic and foreign countries is not reneged upon. In terms
of the behavior of consumption, the complete markets allocation implies
that an increase in output in the domestic country will lead to increases in
consumption in both the domestic and foreign countries. In the presence of
an enforcement constraint, however, this is not possible. Hence, to ensure that
domestic country residents consume more than foreign country residents, the
social planner increases the relative weight on the utility of domestic country
residents (recall that the social planner’s weight is the inverse of the marginal
utility of consumption). Over time, as the productivity shock dies out, the
value of autarky diminishes and the relative weight on the home country also
falls.
4.3. OTHER EXTENSIONS
The international RBC literature has sought to reconcile the findings in the
data not only in terms of the cross-country consumption correlations, but also
for the behavior of output, investment, and the real exchange rate. As a result,
models that relax assumptions regarding preferences, the production side, and
the presence of additional shocks have been developed. We describe a few of
these extensions in what follows.
Hess and Shin [119] re-explore the two international business cycle
anomalies emphasized by Backus et al. [25] as well as establish the pattern
of productivity growth between industries and countries. They then compare
these findings for the international business cycle to those obtained for data
between regions within a country — the so-called “intranational business
cycle”. They argue that the intranational business cycle is a natural environment
for thinking about the interactions between economies when there are no
trade frictions and when there are not multiple currencies. Ambler et al.
[13] modify the supply side of a two-country model by adding multiple
October 9, 2009 15:12 9in x 6in b808-ch04
88 Business Cycles: Fact, Fallacy and Fantasy
sectors and trade in intermediate goods. The model generates a higher cross-
country correlation of output than standard one-sector models. It also predicts
cross-country correlations of employment and investment that are closer to
the data.
Stockman and Tesar [200] introduce a nontraded goods sector in each
country. Similar to our discussion above, they argue that introducing
nontraded goods may be a way of re-establishing the link between a
nation’s output and its spending. They employ preferences that depend on
consumption of the tradable goods of countries 1 and 2, a nontradable good,
and leisure. Preferences are assumed to be nonseparable with respect to a
composite good, consisting of the tradable goods of countries 1 and 2, and
a nontradable good. In their model, output in the traded and nontraded
goods sector of each country is produced using sector-specific capital and labor
which is mobile between sectors. However, there is no international capital
or labor mobility. The authors find that the model predicts the correlation
between home and foreign output, overstates the cross-country correlation of
aggregate consumption, and greatly overstates the cross-country correlation
of tradable consumption. Hence, they find that introducing nontradable
goods does not suffice to reduce the cross-country correlation of consumption.
Likewise, they find that the model overstates the negative correlation between
the relative price of nontradable to tradable goods and relative consumption of
nontraded to traded goods, and understates the variability of the trade balance.
They conclude that an international business cycle model driven solely by
productivity shocks cannot account for the findings. Instead, they argue that
what is needed is a source of nation-specific shocks that shift demand. They
introduce taste shocks that affect the utility of traded versus nontraded goods
and find that this feature brings the data more in line with the implications of
the model.
Baxter and Farr [35] develop a model with variable capital utilization
as a way of accounting for the international correlations. They argue that
variable capacity utilization has the potential to account for the co-movement
of factor inputs across countries. We already considered the role of variable
factor utilization in reconciling the behavior of procyclical productivity in
closed-economy business cycle models (see also Burnside and Eichenbaum
[51], Basu [29], or Basu and Kimball [30]). The notion is that a positive shock
to productivity in the domestic country will tend to reduce the investment
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 89
flowacross countries by leading to an increase in capital utilization rather than
investment. As we described above, many of the puzzles of the international
business cycle literature can be addressed successfully by devising a mechanism
to limit investment flows across borders. Kehoe and Perri [126] achieve this
through an enforcement constraint which requires that the utility under the
constrained allocation exceed the utility that country could obtain under
autarky after defaulting on its international debt obligations.
Baxter and Farr [35] assume that preferences are given by the specification
in equation (2.26). Output in the domestic country is produced using capital
services S
t
and labor L
t
as
Y
t
= A
t
S
1−α
t
(X
t
L
t
)
α
, 0 < α < 1,
where capital services are the product of the capital stock, K
t
, andthe utilization
rate, Z
t
:
S
t
= Z
t
K
t
.
The capital accumulation process displays both costly utilization of capital and
adjustment costs:
K
t +1
= [1 − δ(Z
t
)] + φ(I
t
/K
t
)K
t
,
where δ

> 0, δ

> 0, φ

> 0, and φ

< 0. Similar functions characterize the
foreign country. International trade in assets is made solely with one-period,
real pure discount bonds which have the price q
t
= [1 + r
t
]
−1
. Assuming
that the household in each country owns the capital stock and makes real
investment decisions, the budget constraint for the representative household
is given by
C
t
+ I
t
+ q
t
B
t +1
≤ Y
t
+ B
t
,
where B
t +1
is the quantity of bonds carried over into the next period.
The parameterization of the capital utilization function and the adjustment
cost function are key aspects of the quantitative analysis. The covariance
properties of the technology processes for the domestic and foreign technology
shocks are parameterized to be near unit roots with zero spillover effects. The
correlation of the innovations to the technology shocks is set at 0.258, which
is consistent with earlier studies. Finally, the variance of the innovations is
set so that the variance of output in the model exactly matches the volatility
October 9, 2009 15:12 9in x 6in b808-ch04
90 Business Cycles: Fact, Fallacy and Fantasy
of the US economy. One of the key findings regarding the role of variable
capital utilization is that, first, the volatility of the shocks necessary to match
output volatility is substantially reduced. Second, introducing variable capital
utilization reduces the volatilities of consumption, capital, and exports, which
is desirable, and also those of hours and employment, which is not. The most
important impact of variable factor utilization is on the cross-correlations of
the factor inputs. Specifically, the cross-correlation of hours and investment
become positive. Intuitively, variable capital utilization takes the place of
investment flows across countries. When a positive productivity shock occurs
in one country, capital utilization rates increase, which in turn increase
employment. Even with a modest cross-country correlation of the innovations,
this is accompanied by increases in investment and employment in the other
country.
4.4. PUZZLES REVISITED
The puzzles in the international business cycle literature have been the topic
of much research (see, for example, the survey by Baxter [33]). In their review,
Obstfeld and Rogoff [168] argue that many of the puzzles in international
macroeconomics may just be specific to the models researchers are using.
In our earlier discussion, we described the puzzles that arose when trying
to match the international RBC model to the data. Obstfeld and Rogoff
[168] provide a broader view of the empirical puzzles in the international
macroeconomics literature, though they also relate their discussion to the
international RBC literature. They enumerate the following discrepancies
between data and existing theory as “puzzles”:
• Home bias in trade: A number of authors have documented that intra-
national trade is typically much greater than international trade. See, for
example, McCallum [157] or Helliwell [118].
• The Feldstein–Horioka puzzle: According to this puzzle, long averages
of saving rates and investment rates tend to be correlated for the OECD
countries (see Feldstein and Horioka [89]). Yet, in a world of integrated
capital markets, we would expect capital to flowto regions where the rates
of return are highest.
October 9, 2009 15:12 9in x 6in b808-ch04
International Business Cycles 91
• Home bias in equity portfolios: This reflects the puzzling preference of
stock market investors for home assets (for a recent elaboration, see Tesar
and Werner [203]). In Section 4.2, we discussed various explanations that
account for this puzzle — the presence of human capital as discussed by
Baxter and Jermann [36] or nontraded goods.
• The international consumption correlation puzzle: We already discussed
the finding that cross-country consumption correlations exceed the cross-
country output correlations ina typical international RBCmodel, whereas
the opposite is true in the data. Backus and Smith [24] show that in an
economy with traded and nontraded goods, perfect risk sharing across
countries implies that countries which experience declines in the relative
price of consumption should receive large transfers of goods.
• The purchasing power parity (PPP) puzzle: The PPP puzzle arises from
the fact that shocks to the real exchange rate are very persistent.
• The exchange rate disconnect puzzle: This puzzle captures the notionthat
there exists a relationship between exchange rates and any macroeconomic
variable. Meese and Rogoff [161] further show that most exchange rate
models forecast exchange rates no better than a naive random walk.
Surprisingly, Obstfeld and Rogoff [168] find that adding transport costs goes
a long way towards accounting for the first five puzzles. They argue that,
first, the international consumption correlation puzzle tends to be specific
to a model’s assumptions, as we described above, and does not have the same
weight as, say, the international portfolio diversificationpuzzle. Second, the last
two puzzles are pricing puzzles, whereas the other four refer to the behavior
of quantities. They also note that the pricing puzzles require such features
as nominal rigidities of the type that we will consider in the next chapter.
In contrast to some of the other contributions that we have discussed, the
analysis of Obstfeld and Rogoff [168] is an attempt to unify the explanations
of a related but disparate set of findings. Clearly, the literature that we have
described has revealed a variety of promising directions for future research.
October 9, 2009 15:12 9in x 6in b808-ch04
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808-ch05
Chapter 5
New Keynesian Models
The New Keynesian approach has gained significance in the modern
macroeconomics literature. Critics of the original real business cycle (RBC)
approach had, from the onset, taken issue with the notion that prices can
adjust costlessly to clear markets. More recently, New Keynesian theories have
revived interest in business cycle models that are capable of producing short-
run economic fluctuations based on the types of forces that Keynes had initially
postulated. Prototypical New Keynesian models such as that by Rotemberg
and Woodford [182] allow for imperfect competition and markups to capture
alternative propagation mechanisms in response to technology shocks or
shocks to government expenditures. Limited participation models also allow
alternative mechanisms for the propagation of real and monetary shocks.
1
The behavior of hours and productivity has been a topic of debate. The
RBC conclusions regarding the response of hours, output, and other variables
to a technology shock have been questioned by empirical results obtained
along several different lines. On the one hand, Gali [96] has argued that in a
suitably restricted vector autoregression (VAR) including measures of hours,
productivity, output, and other variables, the response of hours to productivity
shocks is negative. This is in contrast to the RBC model, which predicts
that hours rise on impact to a positive technology shock. Basu, Fernald, and
Kimball [32] also present evidence that hours worked and other variables fall
in response to technology improvements in the short run. Their approach
involves purging the standard Solow residual of factors that might lead to
procyclicality, suchas variable factor utilization. There alsoexist open-economy
versions of the New Keynesian model that have been used for policy analysis
1
For a review and discussion of these models, see Christiano, Eichenbaum, and Evans [66].
93
October 9, 2009 15:12 9in x 6in b808-ch05
94 Business Cycles: Fact, Fallacy and Fantasy
and for understanding the role of monetary shocks.
2
We first describe a simple
New Keynesian framework that can be used to rationalize the observations,
and then discuss the empirical findings in more detail.
5.1. THE BASIC MODEL
Consider a simple New Keynesian model with monopolistic competition,
price rigidities, and variable labor effort due to Gali [96]. Suppose that a
representative household chooses consumption C
t
, money holdings M
t
, hours
worked N
t
, and effort levels U
t
to maximize
E
0
_

t =0
β
t
_
ln (C
t
) +λ
m
ln
_
M
t
P
t
_
−H(N
t
, U
t
)
_
_
(1.1)
subject to
_
1
0
P
it
C
it
di +M
t
= W
t
N
t
+V
t
U
t
+M
t −1

t
+
t
(1.2)
for t = 0, 1, 2, . . .. In this expression, C
t
is a composite consumption good
defined as
C
t
=
__
1
0
C
(−1)/
it
di
_
/(−1)
, (1.3)
where C
it
is the quantity of good i ∈ [0, 1] consumed in period t , and > 1
is the elasticity of substitution among consumption goods. The price of good
i is given by P
it
, and
P
t
=
__
1
0
P
1−
it
di
_
1/(1−)
(1.4)
is the aggregate price index. The functional form for H(N
t
, U
t
) is given by
H(N
t
, U
t
) =
λ
n
1 +σ
n
N
1+σ
n
t
+
λ
u
1 +σ
u
U
1+σ
u
t
. (1.5)
ϒ
t
and
t
denote monetary transfers and profits, respectively. W
t
and V
t
denote the nominal prices of an hour of work and effort, respectively.
2
For a further discussion, see Bowman and Doyle [45].
October 9, 2009 15:12 9in x 6in b808-ch05
New Keynesian Models 95
The first-order conditions with respect to the household’s problem are
given by
µ
t
P
it
=
1
C
t
__
1
0
C
(−1)/
it
di
_
1/(−1)
C
−1/
it
, i ∈ [0, 1], (1.6)
λ
m
1
M
t
= µ
t
−βE
t

t +1
), (1.7)
λ
n
N
σ
n
t
= µ
t
W
t
, (1.8)
λ
u
U
σ
u
t
= µ
t
V
t
, (1.9)
where µ
t
denotes the Lagrange multiplier on the period-by-period budget
constraint. We can solve for µ
t
from the first-order condition corresponding
to the consumption choice as µ
t
= 1/(P
t
C
t
). Substituting for this variable
and simplifying yields
C
it
=
_
P
it
P
t
_

C
t
, i ∈ [0, 1], (1.10)
1
C
t
= λ
m
P
t
M
t
+βE
t
_
1
C
t +1
P
t
P
t +1
_
, (1.11)
λ
n
N
σ
n
t
C
t
=
W
t
P
t
, (1.12)
λ
u
U
σ
u
t
C
t
=
V
t
P
t
. (1.13)
Inthis economy, goodi is producedby firmi using the productionfunction
Y
it
= Z
t
L
α
it
, (1.14)
where L
it
is the quantity of effective labor used by firm i:
L
it
= N
θ
it
U
1−θ
it
, 0 < θ < 1. (1.15)
Z
t
is an aggregate technology shock whose growth rate follows an i.i.d. process

t
} with η
t
∼ N(0, σ
2
η
):
Z
t
= Z
t −1
exp (η
t
). (1.16)
Consider first the choice of optimal inputs of hours and effort chosen by the
firm to minimize its costs subject to the production technology. Let λ be the
October 9, 2009 15:12 9in x 6in b808-ch05
96 Business Cycles: Fact, Fallacy and Fantasy
Lagrange multiplier on the technology constraint. The first-order conditions
are given by
W
t
= λZ
t
θαN
θα−1
it
U
(1−θ)α
it
, (1.17)
V
t
= λZ
t
(1 −θ)αN
θα
it
U
(1−θ)α−1
it
. (1.18)
The ratio of these conditions yields
θ
1 −θ
U
it
N
it
=
W
t
V
t
. (1.19)
Notice that the firm will be willing to accommodate any changes in demand
at the given price P
it
as long as this price is above marginal cost. Hence, the
firm chooses the output level
Y
it
=
_
P
it
P
t
_

C
t
. (1.20)
Thus, when choosing price, the firm will solve the problem
max
P
it
E
t −1
{(1/C
t
)(P
it
Y
it
−W
t
N
it
−V
t
U
it
)} (1.21)
subject to (1.19) and (1.20). To find the first-order condition for this problem,
we will use the last two constraints to solve for the firm’s cost function as
W
t
N
it
+V
t
U
it
= W
t
N
it
+
1 −θ
θ
W
t
V
t
V
t
N
it
=
W
t
N
it
θ
=
W
t
Y
1/α
it
θZ
1/α
t
((1 −θ)W
t
/θV
t
)
1−θ
=
W
t
(P
it
/P
t
)
−/α
C
1/α
t
θZ
1/α
t
((1 −θ)W
t
/θV
t
)
1−θ
.
Using this result, the first-order condition is
E
t −1
_
(1/C
t
)
_
αθP
it
Y
it


−1
W
t
N
it
__
= 0. (1.22)
Finally, the quantity of money is determined as
M
s
t
= M
s
t −1
exp (ξ
t
+γη
t
), (1.23)
where {ξ
t
} is a white noise that is orthogonal to {η
t
}, with ξ
t
∼ N(0, σ
2
m
).
October 9, 2009 15:12 9in x 6in b808-ch05
New Keynesian Models 97
In a symmetric equilibrium, all firms charge the same price P
t
and choose
the same levels of the inputs and output N
t
, U
t
, and Y
t
. Market clearing in
the goods market requires that C
t
= C
it
= Y
it
= Y
t
. Finally, equilibrium
in the money market requires that the growth rate of the money stock evolve
exogenously as M
t
/M
t −1
= exp (ξ
t
+γη
t
).
Next, assume that consumption is proportional to real balances in
equilibrium, C
t
= M
t
/P
t
. This, together with market-clearing conditions,
implies that P
t
Y
t
= M
t
. Using the first-order condition for consumption in
(1.11) and the money growth rule in (1.23) yields
C
t
= λ
−1
m
M
t
P
t
_
1 −βE
t
_
P
t
P
t +1
Y
t
Y
t +1
__
= λ
−1
m
M
t
P
t
_
1 −β exp (σ
2
m

2
σ
2
η
)/2
_
=
M
t
P
t
, (1.24)
where = λ
−1
m
[1 − β exp (σ
2
m
+ γ
2
σ
2
η
)/2]. Next, use (1.12), (1.13), and
(1.19). Taking the ratio of the first two conditions yields
W
t
V
t
=
λ
n
λ
u
N
σ
n
t
U
σ
u
t
.
Equating this with (1.19) yields
U
t
N
t
=
1 −θ
θ
λ
n
λ
u
N
σ
n
t
U
σ
u
t
.
Solving for U
t
yields
U
t
= A
1/α(1−θ)
N
(1+σ
n
)/(1+σ
u
)
t
, (1.25)
where A = ((1 −θ)/θ(λ
n

u
))
α(1−θ)/(1+σ
u
)
. Substituting this result into the
equation for the production function yields
Y
t
= AZ
t
N
ϕ
t
, (1.26)
where ϕ = αθ + (1 + σ
n
)α(1 − θ)/(1 + σ
u
). Using the price-setting rule in
(1.22) together with (1.12) and the expression for equilibrium consumption
October 9, 2009 15:12 9in x 6in b808-ch05
98 Business Cycles: Fact, Fallacy and Fantasy
and output derived above, we can show that
p
t
= ξ
t −1
−(1 −γ)η
t −1
, (1.27)
y
t
= ξ
t
+γη
t
+(1 −γ)η
t −1
, (1.28)
n
t
=
1
ϕ
ξ
t

1 −γ
ϕ
η
t
, (1.29)
x
t
=
_
1 −
1
ϕ
_
ξ
t
+
_
1 −γ
ϕ

_
η
t
+(1 −γ)
_
1 −
1
ϕ
_
η
t −1
, (1.30)
where x = y −n is the log of labor productivity.
The conditions in (1.27)–(1.30) can be used to describe the impact of
monetary versus technology shocks. A positive monetary shock defined by ξ
t
>
0 has a temporary impact on output, employment, and productivity. This can
be observedby noting that the levels of y
t
, n
t
, andx
t
dependonly onthe current
ξ
t
. Hence, an increase in ξ
t
causes output and employment to go up for one
period and then to revert back to their initial values. The impact of ξ
t
on labor
productivity is also transitory, but the sign depends on whether ϕ < (>)1. We
note that measured labor productivity responds positively whenever ϕ > 1,
which corresponds to the situation of short-run increasing returns to labor.
Finally, the price level responds one-for-one to an increase in ξ
t
, though with
a one-period lag.
A positive technology shock defined by η
t
> 0 has a permanent positive
one-for-one impact on output and productivity and a permanent negative
impact on the price level if γ < 1. More interestingly, a positive technology
shock has a negative short-run impact on employment. This result can be
best understood by considering the case of γ = 0, that is, when there is
no accommodating response in the money supply to real shocks. In such a
case, given a constant money supply and predetermined prices, real balances
remain unchanged in the face of a positive technology shock. Hence, demand
remains unchanged so that firms will be able to meet demand by producing
an unchanged level of output. However, with a positive shock to technology,
producing the same output will require less labor input, and hence a decline
in employment will occur.
October 9, 2009 15:12 9in x 6in b808-ch05
New Keynesian Models 99
5.2. EMPIRICAL EVIDENCE
Gali [96] estimates a structural VAR (SVAR) and identifies technology shocks
as the only shocks that are allowed to have a permanent effect on average
labor productivity. This is similar to the approach in Blanchard and Quah
[43] for identifying demand versus supply shocks using long-run restrictions
on estimated VARs. The SVARmodel interprets the behavior of (log) hours n
t
and (log) productivity x
t
in terms of two types of exogenous disturbances —
technology and non-technology shocks —which are orthogonal to each other
and whose impact is propagated through time based on some unspecified
mechanisms as
_
x
t
n
t
_
=
_
C
11
(L) C
12
(L)
C
21
(L) C
22
(L)
_ _

z
t

m
t
_
= C(L)
t
,
where
z
t
and
m
t
denote the sequences of technology and non-technology
shocks, respectively. The orthogonality assumption, together with a
normalization, implies that E(
t

t
) = I . The identifying assumption is that
only technology shocks have a permanent effect on productivity, which can
be expressed as the restriction C
12
(1) = 0.
3
Gali [96] estimates this model
using postwar US data. Surprisingly, he finds that alternative measures of labor
input decline in response to a positive technology shock while GDP adjusts
only gradually to its long-run level. Furthermore, technology shocks explain
only a small fraction of employment and output fluctuations. By contrast,
Gali [96] finds that variables that have no permanent effects on employment
(and which are referred to as demand shocks) explain a substantial fraction of
the variation in both employment and output. Christiano, Eichenbaum, and
Vigfusson [68] suggest that the standard RBC results hold if per capita hours
are measured in log-levels as opposed to differences.
Chari, Kehoe, and McGrattan [63] have challenged the findings derived
from so-called SVARs. First, they argue that the difference specification for
aggregate hours is a priori misspecified because all RBC models imply that
per capita hours is a stationary variable. Second, they argue that if the
simulated data from a simple RBC model are subjected to a SVAR-based
test, the difference-stationary version of the model implies that the response
3
The value of the matrix C(L) evaluated at L = 1 gives the long-run multipliers for the model, i.e.,
the long-run impact of a given shock.
October 9, 2009 15:12 9in x 6in b808-ch05
100 Business Cycles: Fact, Fallacy and Fantasy
of hours to a shock to productivity is negative as in Gali and Rabanal [98].
However, if hours worked is considered to be stationary in levels, then they
argue that the confidence bands for the impulse response functions from the
SVAR are so wide that the procedure is uninformative about the question
at hand. They trace the source of the difference to the failure to include a
sufficient number of lags inthe estimatedVARspecifications soas toadequately
capture the behavior of hours implied by the underlying theoretical model.
This misspecification, in their view, leads to an erroneous conclusion of the
negative response of hours worked to a productivity shock, even though the
data underlying this test are drawn froma standard RBCmodel which implies
the opposite response!
Despite the controversies surrounding the SVARapproach, Basu et al. [32]
present evidence that support the SVAR findings by generating a modified
Solow residual that accounts for imperfect competition, non-constant returns
to scale, variable factor utilization, and sectoral re-allocation and aggregation
effects. Unlike the SVAR approach, the evidence obtained from this approach
is robust to long-run identifying assumptions or to the inclusion of new
variables in the estimated dynamic system. They find that purging the standard
Solow residual of these effects eliminates the phenomenon of “procyclical
productivity”. They also examine the response of a key set of variables such as
output, hours worked, utilization, employment, non-residential investment,
durables and residential investment, non-durables and services, and various
prices and interest rates to changes in the purified Solow residual. They use
both standard regression analysis and simple bivariate VARs for this purpose.
Their findings corroborate the findings from the SVAR approach regarding
the negative response of hours to technology improvements in the short run.
They also uncover further evidence for the negative response of non-residential
investment to such shocks. Following Gali [96] and Gali and Rabanal [98],
they advance price rigidity as the major reason for these deviations from the
RBC predictions in the short run. These findings have, on the one hand,
generated substantial controversy and, on the other, cast further doubt on the
ability of the RBCmodel driven by technology shocks to provide a convincing
explanation of economic fluctuations for the major developed countries.
October 9, 2009 15:12 9in x 6in b808-ch06
Chapter 6
Business Cycles in Emerging
Market Economies
In recent years, the real business cycle (RBC) agenda has been increasingly
applied in emerging market contexts. On the one hand, researchers have begun
generating business cycle facts for such economies (see, for example, Ratfai and
Benczur [175, 176]). Onthe other hand, they have beenexamining the efficacy
of RBC-type models in accounting for these facts. Business cycles in emerging
markets exhibit different characteristics compared to those in developed
economies. The recent literature on the “Sudden Stop” phenomenon has
emphasized the large reversals in current accounts and the incidence of
capital outflows that have become identified with many recent emerging
market economies’ experience (see, for example, Arellano and Mendoza [17]).
Emerging market business cycles also display a different set of stylized facts
relative to developed economies. For instance, consumption varies more than
output, the trade balance is strongly countercyclical, and income and exports
are typically highly volatile. The question then arises whether a small open
economy-type RBC model can account for both developed and emerging
market business cycles.
Earlier RBCmodels for small open economies were developed by Mendoza
[162] and Correia, Neves, and Rebelo [77]. Kydland and Zaragaza [144] also
study the behavior of an emerging market economy, namely, Argentina, but
their analysis is based on the one-sector optimal growth model. They argue
that the large shortfalls in GDP suffered by Argentina in the 1980s can be
explained in a simple growth model framework using the observed measure
of productivity or the Solow residual. More recently, Aguiar and Gopinath
101
October 9, 2009 15:12 9in x 6in b808-ch06
102 Business Cycles: Fact, Fallacy and Fantasy
[1] and Garcia-Cicco, Pancrazi, and Uribe [101] have examined versions of
a small open-economy business cycle model with permanent and transitory
shocks to account for emerging versus developed economy experiences.
6.1. A SMALL OPEN-ECONOMY MODEL OF EMERGING
MARKET BUSINESS CYCLES
Aguiar and Gopinath [1] present a small open-economy model of business
cycles that allows for permanent and transitory changes to productivity. In
their view, emerging market economies’ experience can be distinguished by the
large number of regime shifts, here modeled as changes in trend productivity
growth. By contrast, developed economies typically face stable political and
economic policy regimes so that changes to productivity are transitory.
To describe the model, let output be produced according to the Cobb–
Douglas production technology as
Y
t
= exp (z
t
)K
1−α
t
(
t
L
t
)
α
, 0 < α < 1, (1.1)
where {z
t
} and {
t
} represent two alternative productivity processes. The
shock z
t
represents the transitory component of productivity, and evolves as a
stationary AR(1) process:
z
t
= ρ
z
z
t −1
+
z
t
, |ρ
z
| < 1, (1.2)
where {
z
t
}

t =0
is distributed as i.i.d. with E(
z
t
) = 0 and Var(
z
t
) = σ
2
z
.
The permanent shock to productivity evolves as

t
= g
t

t −1
=
t

s=0
g
s
, (1.3)
ln (g
t
) = (1 − ρ
g
) ln (µ
g
) + ρ
g
ln (g
t −1
) +
g
t
, |ρ
g
| < 1, (1.4)
where {
g
t
}

t =0
is distributed as i.i.d. with E(
g
t
) = 0 and Var(
g
t
) = σ
2
g
.
Thus, g
t
denotes the shocks to the growth rate of productivity, and µ
g
denotes
average long-run productivity growth. Since productivity and output depend
on the cumulation of the shocks g
t
, a stationarity-inducing transformation is
used to remove the stochastic trend from all variables as
ˆ x
t
=
x
t

t −1
.
October 9, 2009 15:12 9in x 6in b808-ch06
Business Cycles in Emerging Market Economies 103
Aguiar and Gopinath [1] consider two types of preferences over
consumption and leisure. The first is from the class of so-called GHH (after
Greenwood, Hercowitz, and Huffman [105]) and the second are standard
Cobb–Douglas preferences. The GHH preferences are defined as
u
t
=
(C
t
− τ
t −1
L
ν
t
)
1−σ
1 − σ
, ν > 1, τ > 0. (1.5)
Here, the elasticity of labor supply is given by 1/(ν − 1), and the inter-
temporal elasticity of substitution is given by 1/σ. In the Cobb–Douglas case,
preferences are given by
u
t
=
(C
γ
t
(1 − L
t
)
1−γ
)
1−σ
1 − σ
, 0 < γ < 1, σ ≥ 0. (1.6)
The expected discounted utility of the representative consumer can be
written as
1
1 − σ
E
0
{(C
0
− τL
ν
0
)
1−σ
+ β
1−σ
0
(
ˆ
C
1
− τL
ν
1
)
1−σ

2

1−σ
1
(
ˆ
C
2
− τL
ν
2
)
1−σ
. . .}
=
1
1 − σ
E
0
{(C
0
− τL
ν
0
)
1−σ
+ βg
1−σ
0
(
ˆ
C
1
− τL
ν
1
)
1−σ
+E
1

2
g
0
g
1−σ
1
(
ˆ
C
2
− τL
ν
2
)
1−σ
. . .}}
=
1
1 − σ
E
0
{(C
0
− τL
ν
0
)
1−σ
+ βg
1−σ
0
(
ˆ
C
1
− τL
ν
1
)
1−σ
+βg
0
E
1
{βg
1−σ
1
(
ˆ
C
2
− τL
ν
2
)
1−σ
. . .}},
where the last line is obtained by substituting recursively and making use of
an iterated expectations argument. Thus, expected discounted utility remains
bounded provided βE(g
t
) = βµ
g
< 1.
The economy-wide resource constraint is given by
C
t
+ K
t +1
= Y
t
+ (1 − δ)K
t
+
φ
2
_
K
t +1
K
t
− µ
g
_
2
K
t
− B
t
+ q
t
B
t +1
.
(1.7)
This shows that investment is subject to quadratic costs of adjustment, and
that the country can borrow using one-period debt that sells for the price
October 9, 2009 15:12 9in x 6in b808-ch06
104 Business Cycles: Fact, Fallacy and Fantasy
q
t
. The price of debt for the country depends on the quantity of debt
outstanding as
1
q
t
= 1 + r
t
= 1 + r

+ ψ
_
exp
_
B
t +1

t
− b
_
− 1
_
, (1.8)
where r

is the world interest rate, b represents the steady-state level of
debt, and ψ > 0 governs the elasticity of the interest rate to changes in
indebtedness. Furthermore, when choosing how much debt to take on, the
individual country does not internalize the fact that an increase in borrowing
will lead to an increase in the interest rate on those loans.
The model expressed in terms of the transformed or “hatted” variables
is solved by using standard recursive methods. Assuming it exists, the value
function defined in terms of the transformed variables is given by
V (
ˆ
K
t
,
ˆ
B
t
, z
t
, g
t
)
= max
ˆ
C
t
,L
t
,
ˆ
K
t +1
,
ˆ
B
t +1
{u(
ˆ
C
t
, L
t
) + f (β, g
t
)E
t
V (
ˆ
K
t +1
,
ˆ
B
t +1
, z
t +1
, g
t +1
)},
where u(
ˆ
C
t
, L
t
) is defined by (1.5) in the case of GHH preferences and by
(1.6) in the case of Cobb–Douglas preferences. Likewise, f (β, g
t
) is given by
βg
1−σ
t
in the former case and by βg
γ(1−σ)
t
in the latter. The optimization is
subject to the transformed budget constraint
ˆ
C
t
+ g
t
ˆ
K
t +1
=
ˆ
Y
t
+ (1 − δ)
ˆ
K
t

φ
2
_
g
t
ˆ
K
t +1
ˆ
K
t
− µ
g
_
2
ˆ
K
t

ˆ
B
t
+ q
t
g
t
ˆ
B
t +1
.
Substituting for
ˆ
C
t
using the resource constraint, the first-order conditions
with respect to
ˆ
K
t +1
,
ˆ
B
t +1
, and L
t
are
u
c
(
ˆ
C
t
, L
t
)
_
g
t
+ φ
_
g
t
ˆ
K
t +1
ˆ
K
t
− µ
g
_
g
t
_
= f (β, g
t
)E
t
_
∂V

ˆ
K
t +1
_
, (1.9)
u
c
(
ˆ
C
t
, L
t
)g
t
q
t
+ f (β, g
t
)E
t
_
∂V

ˆ
B
t +1
_
= 0, (1.10)
u
L
(
ˆ
C
t
, L
t
) + u
c
(
ˆ
C
t
, L
t
)

ˆ
Y
t
∂L
t
= 0. (1.11)
October 9, 2009 15:12 9in x 6in b808-ch06
Business Cycles in Emerging Market Economies 105
The envelope conditions are given by
∂V (
ˆ
K
t
,
ˆ
B
t
, z
t
, g
t
)

ˆ
K
t
= u
c
(
ˆ
C
t
, L
t
)
_

ˆ
Y
t

ˆ
K
t
+ (1 − δ)

_
g
t
ˆ
K
t +1
ˆ
K
t
− µ
g
_
g
t
ˆ
K
t +1
ˆ
K
t

φ
2
_
g
t
ˆ
K
t +1
ˆ
K
t
− µ
g
_
2
_
_
_
,
∂V (
ˆ
K
t
,
ˆ
B
t
, z
t
, g
t
)

ˆ
B
t
= −u
c
(
ˆ
C
t
, L
t
).
For simplicity, consider only the case with GHH preferences. The
stationary competitive equilibrium satisfies the following conditions:
(
ˆ
C
t
− τL
t
)
−σ
_
1 + φ
_
g
t
ˆ
K
t +1
ˆ
K
t
− µ
g
__
=
β
g
σ
t
E
t
_
(
ˆ
C
t +1
− τL
t +1
)
−σ
_
1 − δ + exp (z
t +1
)(1−α)g
α
t +1
_
L
t +1
ˆ
K
t +1
_
α

_
g
t +1
ˆ
K
t +2
ˆ
K
t +1
− µ
g
_
g
t +1
ˆ
K
t +2
ˆ
K
t +1

φ
2
_
g
t +1
ˆ
K
t +2
ˆ
K
t +1
− µ
g
_
2
_
_
_
_
_
,
(1.12)
(
ˆ
C
t
− τL
ν
t
)
−σ
=
β (1 + r)
g
σ
t
E
t
_
(
ˆ
C
t +1
− τL
ν
t +1
)
σ
_
, (1.13)
τνL
ν−1
t
= exp (z
t
)αg
α
t
_
ˆ
K
t
L
t
_
1−α
, (1.14)
subject to the production function (1.1), the laws of motion for the shocks
(1.2)–(1.4), the resource constraint (1.7), and the equation describing the real
interest rate (1.8).
The solution for the model is obtained by implementing a log-linear
approximation to the equilibrium conditions described above. In the recent
applications, a subset of the parameters is set at values determined a priori. The
remainder of the parameters, including the parameters of the shock processes,
are estimated using a generalized method of moments (GMM) approach that
we describe in detail in the following chapter.
October 9, 2009 15:12 9in x 6in b808-ch06
106 Business Cycles: Fact, Fallacy and Fantasy
6.2. DO SHOCKS TO TREND PRODUCTIVITY
EXPLAIN BUSINESS CYCLES IN EMERGING
MARKET ECONOMIES?
Aguiar and Gopinath [1] consider two versions of their model, an emerging
market economy version estimated using quarterly data for Mexico and a
developed country version estimated for Canada between 1980 and 2003.
The moments that they use for estimation include the standard deviations of
log (filtered) income, investment, consumption, net exports-to-GDP, as well as
the correlations of the latter three withoutput. They also consider the meanand
standard deviationof (unfiltered) income growth as well as the autocorrelations
of (filtered) income and (unfiltered) income growth. This yields 11 moment
conditions. The parameters to be estimated are given by (µ
g
, σ
z
, ρ
z
, σ
g
, ρ
g
, φ).
Since the number of parameters is less than the number of moment conditions,
there are also overidentifying conditions which can be used to test the model’s
implications. The most noteworthy finding that emerges from the estimation
is that the ratio of shock standard deviations, σ
g

z
, is 0.25 or 0.41 for Canada
depending on the specification for preferences, and 2.5 or 5.4 for Mexico. This
ratio captures the importance of shocks to trend productivity. By contrast,
the autocorrelations of transitory shocks are roughly similar, as is the capital
adjustment parameter φ.
The authors argue that differences in the magnitude of shocks to trend
productivity can account for some of the salient features of business cycles in
emerging market economies. In particular, they showthat a positive transitory
shock to productivity reduces consumption in anticipation of lower income
in the future. By contrast, a positive shock to trend productivity causes
consumption to respond more than output in the expectation that income
will be higher in the future. This leads to a large trade deficit which tends to
persist for a considerable period of time (16 quarters). Thus, the shock to trend
productivity can generate consumption booms that tend to appear side-by-
side withcurrent account deficits inemerging market economies. Furthermore,
these results can also explain why consumption is more volatile than income
in economies where shocks to productivity growth are more important than
transitory shocks.
These findings have been challenged by Garcia-Cicco et al. [101], who
argue that the results of Aguiar and Gopinath [1] are due to their use of
October 9, 2009 15:12 9in x 6in b808-ch06
Business Cycles in Emerging Market Economies 107
a short sample to estimate low-frequency movements in productivity. By
contrast, Garcia-Ciccoet al. [101] use annual data for Argentina over the period
1913–2005. As we described earlier, much of the business cycle literature for
developed countries has concentrated on the post-World War II period. Given
the evidence that business cycles have moderated during this period, this choice
of sample period does not seem out of line. Garcia-Cicco et al. [101] argue
that a similar finding is not true for an emerging market economy such as
Argentina. In particular, they show that output fluctuations in the post-World
War II period are as large as those in the pre-World War II period. They
consider a model that is very similar to the one in Aguiar and Gopinath [1],
and estimate the parameters of the stochastic processes for the permanent and
transitory shocks using 16 moment conditions. In particular, they consider
the variances and first- and second-order autocorrelations of output growth,
consumption growth, investment growth, and the trade balance-to-output
ratio; the correlations of output growth with consumption growth, investment
growth, and the trade balance-to-output ratio; and the unconditional mean
of output growth.
First, unlike Aguiar and Gopinath [1], Garcia-Cicco et al. [101] find that
the overidentifying restrictions of the model are rejected. Second, in terms
of the parameter estimates, the permanent shock is estimated to be more
volatile and persistent than the transitory shock. The standard deviations of the
shocks and the autoregressive parameter for the permanent shock are estimated
precisely. However, this is not the case for the autoregressive coefficient for the
transitory shock, which is not significantly different fromzero. As we described
above, consumption growth will be more volatile than output growth if
permanent shocks are more important than transitory shocks, and less volatile
thanoutput growth if the opposite is true. They determine this empirically, and
find that the consumption-smoothing motive in response to a transitory shock
dominates the anticipatory effect of consumption to a permanent productivity
shock, and, in contrast to what is observed in the data, consumption growth
in the estimated model is calculated to be less volatile than output growth.
The authors also find that the trade balance-to-output ratio is estimated to
be around four times as volatile as output growth, whereas in the data these
quantities are roughly equal. Whereas the first result suggests that the estimated
model does not emphasize the role of permanent shocks sufficiently, the second
result suggests that it overemphasizes them. Third, investment growth is found
October 9, 2009 15:12 9in x 6in b808-ch06
108 Business Cycles: Fact, Fallacy and Fantasy
to be insufficiently volatile, suggesting in this case that neither source of shocks
is sufficiently volatile. The authors show that the estimated model also fails
to replicate the autocorrelations of output growth and the trade balance-to-
output ratio. In particular, the trade balance-to-output ratio displays a near
random walk behavior even though the estimated autocorrelation function in
the data is downward-sloping. The authors also show that the random walk
behavior of the trade balance-to-output ratio remains, regardless of whether
shocks toproductivity are permanent or transitory. If the shocks are permanent,
consumption increases in response to an innovation to output in roughly the
same magnitude as output, as households perceive the income increase to be
permanent. Hence, the trade balance is unaffected and the trade balance-to-
output ratio inherits the behavior of output. By contrast, if the shocks are
transitory, the behavior of the trade balance-to-output ratio again follows a
near random walk because in this case, with a constant world interest rate,
consumption in the model follows a near random walk even though output
and investment become stationary variables.
Garcia-Cicco et al. [101] conclude that a pure RBC model driven by
permanent and/or transitory exogenous shifts in productivity does not provide
an adequate explanation of business cycles in emerging markets. In particular,
they argue that some of their findings that we described above point to the
importance of shocks that are different from productivity shocks. However,
they also argue that their results are derived under the joint hypothesis of
business cycles driven by productivity shocks and the propagation mechanisms
of the standard RBC approach. Hence, their findings cannot be used to
disentangle which of these factors is responsible for the failure of the model to
replicate the observations.
October 9, 2009 15:12 9in x 6in b808-ch07
Chapter 7
Matching the Model to the Data
In the previous chapters, we described some criticisms leveled against the
assumptions of the real business cycle (RBC) framework. Amongst the most
prominent of these assumptions is the assumption of perfect price flexibility.
We also discussed variations of the model that could be used to account for
specific correlations in the data. Yet one of the most important aspects of the
debate regarding the RBC approach lies in the use of calibration as a way
of matching the model to the data. In this chapter, we will study alternative
approaches for matching the implications of a theoretical model with the data,
and also provide an overview of the estimation versus calibration debate.
The recent business cycle literature has witnessed the use of a variety of
techniques regarding the empirics of business cycles. One of the most popular
approaches has been based on the dynamic factor model, which seeks to
describe the joint cyclical behavior of a key set of time series in terms of a low-
dimensional vector of unobservable factors and a set of idiosyncratic shocks.
This model was initially proposed by Sargent and Sims [185] for describing
cyclical phenomena. In her original contribution, Altug [5] estimated an
unobservable index model for a key set of aggregate series based on a modified
version of the Kydland and Prescott model [141] using maximum likelihood.
Watson[209] extendedthis approachtoderive measures of fit for anunderlying
economic model. An alternative approach was proposed by Christiano and
Eichenbaum [65], who used the generalized method of moments (GMM)
approach (see Hansen [113]) to match a selected set of unconditional first and
second moments implied by their model. Their approach may be viewed as an
extension of the standard RBC approach, which assesses the adequacy of the
model based on the behavior of the relative variability and co-movement of
109
October 9, 2009 15:12 9in x 6in b808-ch07
110 Business Cycles: Fact, Fallacy and Fantasy
a small set of moments. Canova [56, 57] discusses alternative approaches
for conducting statistical inference in calibrated models. Finally, Bayesian
estimation of the so-called dynamic stochastic general equilibrium (DSGE)
models provides an alternative to incorporating prior beliefs about various
parameters that formalize some of the practices in the calibration approach.
7.1. DYNAMIC FACTOR ANALYSIS
Dynamic factor analysis seeks to describe the joint cyclical behavior of a key set
of time series in terms of a low-dimensional vector of unobservable factors and
a set of idiosyncratic shocks that are mutually uncorrelated and uncorrelated
with the factors. To describe how to formulate unobservable index models, let
{ ˜ w
t
}

t =0
denote an n-dimensional mean zero, covariance stationary stochastic
process used to describe observations on the (possibly de-trended) values of a
set of variables. A k-factor unobservable index model for ˜ w
t
is given by
˜ w
t
=

s=−∞
˜
H(s)
˜
f
t −s
+ ˜ ν
t
, (1.1)
where {
˜
H(s)}

s=−∞
is a sequence of (n ×k)-dimensional matrices,
˜
f
t
is a k ×1
vector of common factors, and ˜ ν
t
is an n × 1 vector of idiosyncratic shocks
that are mutually uncorrelated and uncorrelated with common factors. More
precisely, we require that
E(
˜
f
t
˜ ν
i,t
) = 0 for i = 1, . . . , n (1.2)
E(˜ ν
i,t
˜ ν
j,t
) = 0 for i = j. (1.3)
Both the common factors and the idiosyncratic factors may be serially
correlated, that is, E(
˜
f
t
˜
f
t −s
) = 0 for t = s and E(˜ ν
i,t
˜ ν
i,t −s
) = 0 for all
i, j, t = s. According to this model, covariation among the elements of ˜ w
t
can
arise because they are functions of the same common factor or because they
are functions of different factors which are themselves correlated at different
leads and lags.
Under these assumptions, the variances and autocovariances of the
observed series { ˜ w
t
} can be decomposed in terms of the variances and
autocovariances of a low-dimensional set of unobserved common factors and
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 111
the idiosyncratic shocks. Let
R
w
(r) = E( ˜ w
t
( ˜ w
t +r
)

), r = . . . , −1, 0, 1, . . .
be the autocovariance function of { ˜ w
t
}. Under the assumptions
underlying (1.1),
R
w
(r) = E
__

s=−∞
˜
H(s)
˜
f
t −s
+ ˜ ν
t
__

v=−∞
˜
H(v)
˜
f
t +r−v
+ ˜ ν
t +r
_

_
= E
__
· · · +
˜
H( − 1)
˜
f
t +1
+
˜
H(0)
˜
f
t
+
˜
H(1)
˜
f
t −1
+ · · · + ˜ v
t
_
×
_
· · · +
˜
H( − 1)
˜
f
t +r+1
+
˜
H(0)
˜
f
t +r
+
˜
H(1)
˜
f
t +r−1
+ · · · +˜ v
t +r
_

_
=

s=−∞
˜
H(s)

v=−∞
E(
˜
f
t −s
˜
f
t +r−v
)
˜
H(v) + E(˜ ν
t
˜ ν
t +r
)
=

s=−∞
˜
H(s)

v=−∞
R
f
(r + s − v)
˜
H(v)

+ R
ν
(r).
An alternative representation of the autocovariance function is provided by
the spectral density function
S
w
(ω) =

r=∞
R
w
(r)e
−irω
, |ω| ≤ π.
1
(1.4)
Assuming that S
w
(ω) is non-singular at each frequency ω, notice that off-
diagonal elements of S
w
(ω) are, in general, complex numbers. However, since
R
x
l
x
h
(r) = R
x
l
x
h
(−r), the diagonal elements of S
w
(ω) are real. Substituting
for R
w
(r) into (1.4) yields
S
w
(ω) =

r=∞

s=−∞
˜
H(s)

v=−∞
R
f
(r + s − v)
˜
H(v)

exp (−iωr)
+

r=∞
S
ν
(r) exp (−iωr)
1
This function is well-defined as long as


r=−∞
R
2
x
l
x
h
(r) < ∞ for each l , h = 1, . . . , n.
October 9, 2009 15:12 9in x 6in b808-ch07
112 Business Cycles: Fact, Fallacy and Fantasy
=

s=−∞
˜
H(s) exp (−iωs)

u=−∞
R
w
(u) exp (−iωu)
×

z=−∞
˜
H(z)

exp (−iωz) + S
ν
(ω)
=
˜
H(ω)S
w
(ω)
˜
H(ω)

+ S
ν
(ω),
where
˜
H(ω) denotes the Fourier transformof
˜
H(s). Hence, the dynamic factor
model provides decomposition at each frequency that is analogous to the
decomposition of variance in the conventional factor model. The dynamic
factor model can be estimated and its restrictions tested across alternative
frequencies using a frequency domain approach to time series analysis. The
unrestricted version of the dynamic factor model does not place restrictions on
the matrices
˜
H(s), which describe howthe common factors affect the behavior
of the elements of ˜ w
t
at all leads and lags. Also, it is not possible to identify
the common factors with different types of shocks to the economy.
The use of the dynamic factor model in business cycle analysis dates back to
the work of Sargent and Sims [185]. As these authors observe, dynamic factor
analysis may be linked to the notion of a “reference cycle” underlying the
methodology of Burns and Mitchell [50] and the empirical business cycle
literature they conducted at the National Bureau of Economic Research.
Another well-known application of this approach is due to Altug [5], who
derives an unobservable index model for a key set of aggregate series by
augmenting the approximate linear decision rules for a modified version of
the Kydland and Prescott [141] model with i.i.d. error terms. Sargent [184]
also employs the device of augmenting a singular model with additional
idiosyncratic shocks. He discusses a classical measurement error case, as in
Altug [5], as well as a case with orthogonal prediction errors. Altug also
uses this representation to estimate the model using maximum likelihood
(ML) estimation in the frequency domain. The restricted factor model makes
use of the cross-equation restrictions across the linear decision rules implied
by the original model. The common factor is identified as the innovation
to the technology shock, and the idiosyncratic shocks are interpreted as
i.i.d. measurement errors or idiosyncratic components not captured by the
underlying RBC model. Unlike the unrestricted factor model which can be
estimated frequency by frequency, this model must be estimated jointly across
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 113
all frequencies because the underlying economic model constrains the dynamic
behavior of the different series as well as specifies the nature of the unobserved
factor.
2
Altug [5] initially estimates an unrestricted dynamic factor model for
the level of per capita hours and the differences in per capita values of durable
goods consumption, investment in equipment, investment in structures, and
aggregate output. She finds that the hypothesis of a single unobservable factor
cannot be rejected at conventional significance levels for describing the joint
time series behavior of the variables. However, when the restrictions of the
underlying model are imposed, the model cannot explain the cyclical variation
of the observed variables.
7.1.1. Measures of Fit for Calibrated Models
Watson [209] extended the approach in Altug [5] and Sargent [184] to derive
measures of fit for an underlying economic model. Unlike the approach
adopted by Altug and Sargent, Watson’s analysis does not depend on assuming
that the unobserved factors and the idiosyncratic shocks are uncorrelated.
Instead, his approach involves choosing the correlation properties of the error
process between the actual data and the underlying model such that its variance
is as small as possible. Also, the joint process for the data and the error is
introduced to motivate goodness-of-fit measures, not to describe a statistical
model that can be used to conduct statistical tests.
To describe Watson’s approach, let x
t
denote an n ×1 vector of covariance
stationary random variables. Define the autocovariance generating function
(ACGF) for x
t
by
A
x
(z) =

r=−∞
E(x
t
x
t −r
)z
r
.
Here, {x
t
}

t =0
denotes the stochastic process generated by some underlying
model, and A
x
summarizes the unconditional second-moment properties of
this model. In the data, the vector of variables y
t
corresponds to the empirical
counterpart of x
t
. The ACGF for y
t
is similarly denoted A
y
(z). The question
at hand is whether the data generated by the model are able to reproduce the
behavior of the observed series. For this purpose, define the n × 1 vector of
2
For further discussion of maximum likelihood estimation in the frequency domain, see Hansen and
Sargent [114].
October 9, 2009 15:12 9in x 6in b808-ch07
114 Business Cycles: Fact, Fallacy and Fantasy
errors u
t
that are required to reconcile the ACGF implied by the model with
that of the data.
3
To continue, define u
t
by the relation
u
t
= y
t
− x
t
, (1.5)
which implies that the ACGF for u
t
can be expressed as
A
u
(z) = A
y
(z) + A
x
(z) − A
xy
(z) + A
yx
(z), (1.6)
where A
xy
(z) is the joint ACGF for x
t
and y
t
. To calculate A
u
(z), notice
that A
y
(z) can be determined from the data, and A
x
(z) from the model.
However, since A
xy
(z) is unknown, some additional assumptions are needed
to operationalize this approach. In the standard dynamic factor model, the
assumption regarding A
xy
(z) is that it is zero, A
xy
(z) = 0. This implies
a version of a classical errors-in-variables approach.
4
However, in Watson’s
framework, the error term u
t
is the approximation error in describing the
observed data with the underlying economic model. Hence, the assumption
of pure measurement error which is uncorrelated with the true variable is not
appropriate. Nevertheless, a lower bound may be deduced for the variance of
u
t
without imposing any restrictions on A
xy
(z). This bound is calculated by
choosing A
xy
(z) to minimize the variance of u
t
subject to the constraint that
the implied joint ACGF for x
t
and y
t
is positive definite.
To illustrate this approach, let us consider two cases.
5
In the first case,
x
t
, y
t
, and u
t
are assumed to be serially uncorrelated scalar random variables.
The problem is to choose σ
xy
to minimize the variance of u
t
subject to the
constraint that the covariance matrix of x
t
and y
t
remains positive definite, or

xy
| ≤ σ
x
σ
y
:
min
σ
xy
σ
2
u
= σ
2
y

2
x
− 2σ
xy
s.t. |σ
xy
| ≤ σ
x
σ
y
. (1.7)
3
In a standard goodness-of-fit approach, it is the size of the sampling error that is used to judge whether
a given model fits that data. In this case, let A
y
(z) denote the population autocovariance function and
ˆ
A
y
(z)
denote its estimated counterpart. Then, the difference between A
y
(z) and
ˆ
A
y
(z) is ascribed to sampling
error, and the size of the sampling error can be determined from the data-generating process for y
t
. If, in
addition, A
y
(z) = A
x
(z), then the sampling error in estimating A
y
(z) from actual data can also be used to
determine how different A
y
(z) is from A
x
(z).
4
This is similar to the interpretation provided by Altug [5].
5
We omit the third case that Watson considers because it requires results for the Cramer representation
of stationary time series.
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 115
It is easy to see that the solution for this problem is to set σ
xy
= σ
x
σ
y
. As a
consequence, σ
2
u
= (σ
y
− σ
x
)
2
at the minimum. Furthermore, x
t
and y
t
are
perfectly correlated so that
x
t
= γy
t
, (1.8)
where γ = σ
x

y
.
Now suppose that x
t
and y
t
are serially uncorrelated random vectors with
covariance matrices
x
and
y
. The covariance matrix of u
t
is given by
u
=

y
+
x

xy
+
yx
. In this case, we cannot minimize the variance of u
t
directly. Instead, we consider a transformation that allows us to determine
the size of u
t
. One convenient transformation is the trace of
u
, tr(
u
) =

n
i=1

ij,u
, where
ij,u
denotes the i, j element of . An alternative approach
is to minimize the weighted sum of the variances as tr(W
u
), where W is an
n ×n matrix. The problem in this case is to choose
xy
to minimize tr(W
u
)
subject to the constraint that the covariance matrix of (x

t
, y

t
) is positive semi-
definite. Watson [209] provides a solution for the case in which
x
has rank
k ≤ n so that the number of variables is typically less than the number of
shocks. Proposition 1 in Watson [209] demonstrates that the unique matrix

xy
, which minimizes the weighted sum of the variances of the elements of u
t
subject to the constraint that this matrix is positive semidefinite, is given by

xy
= C
x
VUC

y
. (1.9)
In this expression, C
x
is the n×k matrix square root of
x
as
x
= C
x
C

x
and
C
y
is the n × n matrix square root of
y
. The matrices U and V are defined
fromthe singular value decomposition USV

of C

y
WC
x
such that U is a n×k
orthogonal matrix (with U

U = I
k
), V is a k × k orthonormal matrix (with
VV

= I
k
), and S is a k × k diagonal matrix. An implication of this result is
that, as in the scalar case, the joint covariance matrix of (x

t
, y

t
) is singular. As
a consequence, x
t
can be represented as
x
t
= y
t
, (1.10)
where = C
x
UVC
−1
y
. For k = 1, this result corresponds to the one in the
scalar case, given the properties of the U and V matrices.
Watson [209] uses this methodology to generate goodness-of-fit measures
for a standard RBCmodel with a stochastic trend in technology. Specifically, he
considers the model in King, Plosser, and Rebelo [131, 132]. He log-linearizes
October 9, 2009 15:12 9in x 6in b808-ch07
116 Business Cycles: Fact, Fallacy and Fantasy
the first-order conditions to obtain the solution for log-differences of output,
consumption, investment, and the number of hours worked. His approach
allows for a decomposition by frequency of the variance in each observed
series, the variance explained by the model, and the error in reconciling the
model with the data.
6
He finds that the biggest differences between the spectra
for output, consumption, and investment occur at frequencies corresponding
to the business cycle periodicities of 6–32 quarters. The model also implies
that the number of hours worked is stationary whereas there is considerable
power at the low frequencies for this series in the data, suggesting that the
stochastic trend properties of the model do not hold in the data. Watson’s
analysis shows that focusing only on a small subset of moments implied by
the model can be misleading because the RBC model is unable to reproduce
the typical spectral shape of economic time series.
7.1.2. Other Applications
Forni andReichlin[93] use the dynamic factor model to describe business cycle
dynamics for large cross-sections. Based on a law of large numbers argument,
they show that the number of common factors can be determined using the
method of principal components, the economy-wide shocks can be identified
using structural vector autoregression (SVAR) techniques, and the unobserved
factor model can be estimated by using equation-by-equation ordinary least
squares (OLS). They examine the behavior of four-digit industrial output and
productivity for the US economy for the period 1958–1986 and find evidence
in favor of at least two economy-wide shocks, both having a long-run effect on
sectoral output. However, their results also indicate that sector-specific shocks
are needed to explain the variance of the series.
Giannone, Reichlin, and Sala [103] show how more general classes of
equilibrium business cycle models can be cast in terms of the dynamic factor
representation. They also describe how to derive impulse response functions
for time series models which have reduced rank, that is, models for which the
number of exogenous shocks is less than the number of series.
6
A comparison of the spectra implied by the model and those generated by actual data also figured in
early versions of Altug [5]; see also Altug [4].
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 117
7.2. GMM ESTIMATION APPROACHES
Other papers have employed nonlinear estimation and inference techniques
to match equilibrium business models with the data. Christiano and
Eichenbaum [65] consider the hours-productivity puzzle and use the
generalized method of moments (GMM) approach (see Hansen [113]) to
match a selected set of unconditional first and second moments implied by
their model. Their approach may be viewed as an extension of the standard
RBCapproach, whichassesses the adequacy of the model basedonthe behavior
of the relative variability and co-movement of a small set of time series.
Christiano and Eichenbaum [65] derive a solution for the social planner’s
problemby implementing a quadratic approximation to the original nonlinear
problemaround the deterministic steady states. Since there is a stochastic trend
in this economy arising from the nature of the technology shock process, the
deterministic steady states are derived for the transformed variables. Their
estimation strategy is based on a subset of the first and second moments
implied by their model. To describe how their approach is implemented, let

1
denote a vector of parameters determining preferences, technology, and the
exogenous stochastic processes. Some of the parameters included in
1
may
be the depreciation rate of capital, δ, and the share of capital in the neoclassical
production function, θ. More generally,
1
= (δ, θ, γ, ρ, ¯ g, σ
µ
, λ, σ
λ
)

. As in
the standard RBC approach, the parameters in
1
are estimated using simple
first-moment restrictions implied by the model. For example, the depreciation
rate δ is set to reproduce the average depreciation on capital as
E
_
δ −
_
1 −
i
t
k
t

k
t +1
k
t
__
= 0,
given data on gross investment i
t
and the capital stock k
t +1
. Likewise, the
share of capital satisfies the Euler equation
E
_
β
−1
_
θ
_
y
t +1
k
t +1
_
+ 1 −δ
_
c
t
c
t +1
_
= 0.
Proceeding in this way, the elements of
1
satisfy the unconditional moment
restrictions
E
_
H
1t
(
1
)
_
= 0. (2.11)
October 9, 2009 15:12 9in x 6in b808-ch07
118 Business Cycles: Fact, Fallacy and Fantasy
The elements of
2
consist of the standardRBCsecond-moment restrictions as
E
_
y
2
t

x

y
)
2
− x
2
t
_
= 0, x = c
t
, i
t
, g
t
, (2.12)
E
_
n
2
t
−σ
2
n
_
= 0, (2.13)
E
_
(y/n)
2
t
_
σ
n

y/n
_
2
− n
2
t
_
= 0, (2.14)
E
__
σ
2
n
/(σ
n

y/n
)
_
corr(y/n, n) − (y/n)
t
n
t
_
= 0, (2.15)
where c
t
denotes private consumption; i
t
, private investment; g
t
, public
consumption; n
t
, labor hours; and (y/n)
t
, the average productivity of labor.
The unconditional second moments are obtained through simulating the
model’s solution based on the linear decision rules obtained from the
approximate social planner’s problem. The restrictions of the model can be
summarized as
E
_
H
2t
(
2
)
_
= 0. (2.16)
Christiano and Eichenbaum [65] are interested in testing restrictions for
the correlation between hours and productivity, corr(y/n, n), and the relative
variability of hours versus average productivity, σ
n

y/n
. To do this, they use a
Wald-type test based on the orthogonality conditions implied by the relevant
unconditional moments. For any parameter vector
1
, let
f (
1
) = [f
1
(
1
), f
2
(
1
)]

(2.17)
represent the model’s restrictions for corr(y/n, n) and σ
n

y/n
. Define =
[
1
,
2
]

as the k ×1 vector containing the true values of the parameters and
second moments for the model. Also, let A be a 2×k matrix of zeros and ones
such that
A = [corr(y/n, n), σ
n

y/n
]

(2.18)
and
F() = f (
1
) − A. (2.19)
Under the null hypothesis that the model is correctly specified,
F() = 0. (2.20)
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 119
In practice, there is sampling error in estimating from a finite data set
containing T observations. Letting
ˆ

T
denote the estimated value of , the
test statistic for the second-moment restrictions is based on the distribution of
F(
ˆ

T
) under the null hypothesis. Using this distribution, the authors show
that the statistic
J = F(
ˆ

T
)

Var[F(
ˆ

T
)]
−1
F(
ˆ

T
) (2.21)
is asymptotically distributed as a χ
2
random variable with two degrees of
freedom.
Using data onprivate consumption, government expenditures, investment,
aggregate hours, and average productivity, Christiano and Eichenbaum [65]
estimate the parameters of the model using GMM and examine the various
unconditional second moments implied by the model. As we described earlier,
the GMM approach has been used in other recent applications. For example,
Aguiar and Gopinath [1] and Garcia-Cicco, Pancrazi, and Uribe [101]
employ this approach in their analysis of business cycles in emerging
markets.
7.3. THE CALIBRATION VERSUS ESTIMATION DEBATE
The crux of the recent business cycle debate centers on how a given theoretical
model should be matched with the data. Kydland and Prescott [141] initiated
this debate in the modern business cycle literature. Since then, there have been
ongoing discussions on both sides of the issue.
7
In this section, we will deal
with a variety of criticisms aimed directly at the calibration approach and some
suggested alternatives to it.
In a highly suggestive analysis, Eichenbaum [82] asked whether RBC
analysis, in fact, constituted “wisdom or whimsy”. He took issue with the
calibration approach by showing that the model’s implications for the variance
of output relative to its value in the data are heavily dependent on assumed
values for the underlying parameters for the technology process. In his words:
Indeed, once we quantify the uncertainty in model predictions arising from
uncertainty about model parameter values, calibrated or otherwise, our view
of what the data is [sic] telling us is affected in a first-order way. Even if
7
See Kydland and Prescott [142, 143] for a further discussion and defense of their approach.
October 9, 2009 15:12 9in x 6in b808-ch07
120 Business Cycles: Fact, Fallacy and Fantasy
we do not perturb the standard theory and even if we implement existing
formulations of that theory on the standard postwar sample period and even
if we use the stationary inducing transformation of the data that has become
standard in RBC studies — even then the strong conclusions which mark
this literature are unwarranted. What the data are actually telling us is that,
while technology shocks almost certainly play some role in generating the
business cycle, there is simply an enormous amount of uncertainty of just
what percent of aggregate fluctuations they actually do account for. The
answer could be 70% as Kydland and Prescott [142] claim, but the data
contain almost no evidence against either the view that the answer is really
5% or that the answer is really 200%.
Eichenbaum [82] also presented evidence to show that the performance
of the standard model as well as modifications that allow for labor hoarding,
for example, deteriorate considerably when a break in the sample is allowed
for. Such a break accounts for the slowdown in productivity growth in the US
that occurred in the late 1960s.
7.3.1. The Dynamics of Output
Cogley and Nason [70] examine the dynamics of output as a way of
determining the efficacy of the RBC model in describing the data. As part
of their diagnostics, they consider the autocorrelation function for output, its
power spectrum, and impulse response functions in response to shocks. Since
the single technology shock model is singular, they use the two-shock versionof
the RBCmodel proposed by Christiano and Eichenbaum[65], which includes
shocks to productivity and government consumption as well as the indivisible
labor feature of Hansen [112] and Rogerson [179]. They assume that the
technology shocks are difference-stationary, whereas government shocks evolve
as a persistent AR(1) process. To estimate impulse response functions from
the data, the authors use the SVAR technique developed by Blanchard and
Quah [43]. They consider a two-variable VAR with output and hours worked
(or consumption), and identify the technology shock under the assumption
that it has a permanent effect on output. They compare the autocorrelation
function and the impulse response function in the data with those generated by
the model using generalized Q statistics. The autocorrelation function from
the model is generated by simulating the model 1000 times. The striking
results in Cogley and Nason [70] show that the autocorrelation function
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 121
for output implied by the model is nearly flat, as is the power spectrum.
8
By contrast, the autocorrelations of output at lags 1 and 2 are significant
and positive, and the spectrum for output displays significant power at the
business cycle frequencies of 2.33–7 years per cycle. The test statistic also
shows that the implications of the model are rejected at significance levels
of 1% or less. In terms of the impulse response functions, the model has
some success in matching the estimated response functions in the data for
the permanent component of GDP, but it is unable to match the impulse
response function for the transitory component. In particular, the data display
a hump-shaped response to transitory shocks whereas the model implies a
much smaller monotonic decay. Cogley and Nason [70] also argue that the
model displays weak propagation mechanisms which arise fromintertemporal
substitution, capital accumulation, and adjustment costs.
7.3.2. Calibration as Estimation
Canova [56, 57] proposes an alternative approach for providing statistical
inference in calibrated models. His approach involves constructing prior
distributions for the parameters used in calibrated models based on existing
estimates in the literature or other a priori information available to the
researcher. To describe his approach, let
¯
X
t
denote a vector stochastic process
with a known distribution that describes the evolution of a set of observed
variables, say, GDP, investment, and interest rates. Also, let X
t
= f (Z
t
, β)
denote the process for these variables generated by a specific economic theory
or model as a function of exogenous and predetermined variables Z
t
and the
parameters β. Let G(X
t
|f , β) denote the density of the vector X
t
, conditional
on the function f and the parameters β; let π(β|I , f ) denote the density
for the parameters β, conditional on the information set available to the
researcher I and the function f ; and let H(X
t
, β|f , I ) denote the joint density
for the data and the parameters. Denote the predictive density p(X
t
|I , f ) =
_
H(X
t
, β|I , f )dβ. Define expectations of the functions of the simulated data
8
These results are, in some sense, complementary to those regarding the role of filtering on the cyclical
properties generated from a standard RBC model. See Cogley and Nason [69].
October 9, 2009 15:12 9in x 6in b808-ch07
122 Business Cycles: Fact, Fallacy and Fantasy
(denoted by µ(X
t
)) under the predictive distribution p(X
t
|I , f ) by
E(µ(X
t
)|f , I ) =
_
µ(X
t
)p(X
t
|I , f )dX
t
=
_ _
H(X
t
, β|I , f )dβdX
t
. (3.22)
This approach allows for a probability statement regarding the behavior of
various moments implied by the model. More precisely, suppose that we are
interestedinevaluating Pr(ν(X
t
) ∈ D), where ν(X
t
) is some moment of interest
and D is a bounded set. Then, define µ(X
t
) = 1 if ν(X
t
) ∈ D and zero
otherwise. Notice also that the function f is typically unknown and must be
obtained using some approximation method. Canova [56] describes how to
account for such an approximation error when computing the moments of
the simulated data. While the densities H and p are unknown, the model can
be simulated repeatedly for different values of β and Z
t
to compute sample
paths for X
t
. In general, the approach that Canova advocates for conducting
statistical inference in calibrated models is as follows:
• Select a density π(β|I , f ) for the unknown parameters, given the
information I available to the researcher and a density k(Z
t
) for the
exogenous processes.
• Draw vectors β from π(β|I , f ) and z
t
from k(Z
t
).
• For each drawing of β and z
t
, generate {x
t
}
T
t =1
and compute µ(x
t
) using
the model x
t
= f (z
t
, β) or its approximation.
• Repeat the above two steps N times.
• Construct the frequency distribution of µ(x
t
) and other statistics of interest
that can be used to evaluate the performance of the model.
One of the key aspects of Canova’s approach is the selection of a density
π that summarizes information about existing parameter estimates in an
efficient manner. This information may be derived from alternative data sets,
model specifications, or estimation techniques. For example, one could count
estimates of elements of β obtained from alternative studies and smooth the
resulting histogramto obtain the density π(β|I , f ). If such information is hard
to obtain, then one can use a uniform distribution. The calibration approach
involves putting a point mass on a particular value of β. Simulation exercises
conductedafter a subset of the parameters has beenestimated(using GMM, for
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 123
example) involve putting a point mass onβ

estimatedaccording to a particular
method, say, GMM. One can argue that the approach described above is a
global sensitivity exercise that also allows for an evaluation of the model based
on the probabilities attached to events that the researcher is interested in.
7.3.3. Nonlinearity in Macroeconomic Time Series
Another criticism of the RBC model literature is that it has typically
been concerned with examining the first- and second-moment properties of
aggregate economic variables for the purpose of matching a model to the data.
Yet there is a new literature that shows that macroeconomic time series may
exhibit marked nonlinear behavior. Such nonlinearities may take the form of
conditional heteroscedasticity suchas ARCHor GARCHeffects. Alternatively,
there may exist asymmetries in various economic variables.
Neftci [166] was among the first to demonstrate that unemployment
fluctuations were asymmetrical along the business cycle. Brock and Sayers [47]
tested real macroeconomic variables displayed by deterministic chaotic
dynamics; and while they could not find evidence of chaotic behavior, they
nevertheless showed that postwar employment, unemployment, and industrial
production could be described as nonlinear stochastic processes. Ashley and
Patterson [22] developed a test to test for deviation from linear stochastic
processes, either in the form of nonlinear stochastic dynamics or deterministic
chaos. They found strong evidence of nonlinearity in industrial production,
and argued that any reasonable macroeconomic model should display some
form of nonlinear dynamics (see Potter [172] for a review). In our discussion
in Chapter 3, we discussed the role of alternative factors that could give rise to
endogenous changes in productivity such as labor hoarding, variable capacity
utilization, increasing returns, and time-varying markups. In a novel analysis,
Altug, Ashley, and Patterson [9] examine the implied behavior of output,
the factor inputs, and the underlying productivity shocks using a simple
production function framework. Specifically, they note that observed measures
of output and factor inputs such as the capital stock and hours worked display
marked nonlinear behavior. Using an array of diagnostic tests, they do not
find evidence of nonlinearity in the Solow residuals measured after allowing
for increasing returns to scale, markups, or variable capacity utilization. They
conclude that the nonlinearity must lie in the transmission mechanism.
October 9, 2009 15:12 9in x 6in b808-ch07
124 Business Cycles: Fact, Fallacy and Fantasy
Valderrama [207] examines the statistical behavior of national income and
product account aggregates for the US and a set of OECDcountries including
France, Italy, Japan, Mexico, and the UK; and shows that nonlinearities such as
skewness, kurtosis, and conditional heteroscedasticity are common for many
of these aggregates. He argues that standard general equilibrium models are
able to replicate the first- and second-moment properties of such variables,
but they are unable to reproduce nonlinearities in these time series. He
poses this as a “canonical” challenge to the RBC approach. Valderrama [207]
considers a simple Brock–Mirman-type growth model with GHH preferences
(see Greenwood, Hercowitz, and Huffman [105]) and adjustment costs in
investment. A value iteration approach is used as the solution procedure. This
ensures that the nonlinearities in the underlying model are not eliminated
through a linearization procedure. The approach to matching the model to
the data is through the efficient method of moments (EMM) developed
by Gallant and Tauchen [99], which is a two-step procedure. In the first
step, a seminonparametric (SNP) model is estimated to characterize the
statistical properties of the data. The statistical models are flexible enough
to allow for increasing time dependence in the mean of the process (captured
through a VAR), for conditional heteroscedasticity (ARCH, GARCH), and for
nonnormal disturbances. In the second step, the economic model is simulated
for a given set of parameters. A comparison is made between the statistical
parameter estimates in the SNP step and the statistical parameters obtained
using the simulated data and the same statistical model. Then, the candidate
parameters of the simulated model are adjusted until the simulations of the
economic model have statistical properties similar to those of the data. The
objective function is distributed as a X
2
statistic, as in the GMM approach.
Valderrama [207] selects three statistical models to describe the properties
of the data. The SNP approach nests standard VARs. However, it also
allows for periods of high volatility followed by low volatility (conditional
heteroscedasticity), asymmetric business cycles (i.e., skewness), and excess
volatility (i.e., excess kurtosis). The SNP approach uses a standard VAR to
model the conditional mean and an ARCH-GARCH structure to model the
conditional variance, and it allows for a non-Gaussian error term. The last
feature is achieved by taking a transformation of the normal density using
Hermite polynomials (see Gallant andTauchen[100]). The first model selected
by the SNP is a VAR(3) with conditional heteroscedasticity and a Hermite
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 125
polynomial of degree 4. Two other models are selected for the purpose of
describing the data. The first of these is a linear VAR(3) so that the statistical
procedure is similar to the RBC approach of matching the impulse responses
from a standard VAR with those generated from the underlying economic
model. The second model is a VAR(3) with ARCH(1) errors, which allows
for nonlinearity in terms of the ARCH effects but continues to assume that
the errors are Gaussian. The parameters of the underlying economic model
are chosen based on standard calibration exercises and also to match the three
statistical models using a simulated method of moments approach.
Valderrama [207] finds that the biggest difference among the three
statistical models is in terms of the adjustment cost parameter, φ. When the
statistical model is forced to be a linear VAR(3), this parameter is estimated
to be around 10; whereas in the other two models, its estimates are around 3.
The intuition for this result stems from the fact that a high adjustment cost
parameter helps to smooth the implied behavior of investment and to reduce
conditional volatility or kurtosis in the investment series. By contrast, the other
two statistical models allow for these features and hence do not require such a
high adjustment cost parameter. The linear VAR(3) model also implies a much
lower volatility for the consumption and investment series than the other two
models. Valderrama finds that the RBC model is not successful at generating
the conditional variance of investment; it can capture the nonlinearity in
investment, but not the nonlinearity in consumption. Combined with the
findings of Altug et al. [9] who show that the nonlinearities in the observed
series must lie in the propagation mechanism, these results suggest that such
features as financial frictions or irreversibility in investment are required to
match the nonlinearities of consumption and investment.
Surprisingly, Valderrama [207] finds that the irreversibility constraint does
not bind during the solution of the model, implying that the irreversibility
feature is not important for matching the model to the data. This is similar
to some earlier results in the literature. Veracierto [208] and Thomas [205]
compare model economies with irreversible investment (in the case of
Veracierto) or lumpy investment (in the case of Thomas) with the actual
economy and with a baseline model of flexible investment, and arrive
at results that indicate little or no significant impact of irreversibility or
lumpiness on aggregate investment dynamics. Veracierto [208] argues that the
reason why irreversibility has an impact on aggregate investment in various
October 9, 2009 15:12 9in x 6in b808-ch07
126 Business Cycles: Fact, Fallacy and Fantasy
multi-sector growth models considered in the literature (see, for example,
Coleman [73] or Ramey and Shapiro [174]) is due to their assumption
of unrealistically large sectoral shocks. Similarly, the reason why aggregate
investment displays lower variability in the presence of irreversibility when
firms are subject to idiosyncratic shocks, as in Bertola and Caballero [42], is
due to their assumption of a very large variance for these idiosyncratic shocks.
Veracierto [208] argues that when the size of sectoral shocks is consistent
with that observed in the data, the aggregate impact of irreversible investment
disappears ina general equilibriumframework. More generally, bothVeracierto
and Thomas find that irreversibility or lumpiness at the plant level does not
affect aggregate investment significantly once general equilibrium effects such
as endogenous price adjustments are taken into account. However, this is a
puzzling finding. As Caballero [53] emphasizes: “An important point to note
is that since only aggregate data were used, these microeconomic nonlinearities
must matter at the aggregate level, for otherwise they would not be identified.”
Yet, our analysis of the simple RBC model with flexible prices and wages
suggests that it cannot successfully match many features of the data. Given the
importance of nonlinearity in macroeconomic time series documented in a
number of studies, we suggest that the reasonwhy irreversibility may not matter
in a model with flexible prices is that the introduction of imperfect competition
with optimal price-setting behavior on the part of firms or, alternatively,
changing the way in which the labor market is modeled and introducing
wage contracts in the RBCframework may lead to more pronounced quantity
adjustments. Both Veracierto [208] and Thomas [205] assume that labor is
perfectly mobile across plants. This latter feature may compensate for the
rigidity faced by plants given that capital and labor are substitutable according
to the Cobb–Douglas specification of technology, and may be responsible
for softening the impact of the inflexibility faced by plants in terms of their
capital adjustments. Moreover, as Valderrama [207] states, irreversibility may
play a more important role when financial constraints at the plant level and the
household level are taken into account. At the same time, one may consider the
impact of another type of aggregate shock, namely, monetary shocks. Thus,
for example, financial accelerator models of investment find that financially
constrained firms’ investment responds three times as strongly to a monetary
expansion than that of firms which are not constrained (see Bernanke, Gertler,
and Gilchrist [41]). These arguments suggest that nonlinearities are another
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 127
area where the assumptions of the simple RBC model fail to hold, and a
potential direction for improving the model’s fit.
7.3.4. The Debate Reconsidered
King [128] discusses the arguments on different sides of the debate under the
heading “Quantitative Theory and Econometrics”. King claims to be on the
quantitative theory side of the debate, and argues against the estimation and
comparison of a “heavily restricted linear time series model (for example, an
RBC model with some or all of its parameters estimated) to an unrestricted
time series model. For some time, the outcome of this procedure will be
known in advance of the test: the probability that the model is true is zero for
stochastically singular models and nearly zero for all other models of interest.”
Returning to the inception of the estimation versus calibration debate, we note
that Altug’s [4] results did not lead to support for the model. Many have argued
that this was to be expected (see Hartley, Hoover, and Salyer [116], p. 17). Yet,
Watson’s [209] insights were partly due to Altug’s initial analysis based on the
spectra implied by the model and those inthe data.
9
Inthe absence of the initial
frequency-domain estimation based on the factor representation, it is unlikely
that RBC models would be subjected to the types of diagnostic tests proposed
by Watson [209]. The factor representation underlying Altug’s analysis has
also been resuscitated by Giannone et al. [103] in a VAR framework.
One can also examine the claim that a less restrictive approach to
estimation and model evaluation following the approach in Christiano and
Eichenbaum [65], for example, may be preferable. King [128] advocates such
an approach as an alternative to calibration that does not face the problems of
“testing highly restrictive linear models”. Yet, one could argue that the GMM
approach in Christiano and Eichenbaum [65] involves choosing a subset of
moments of an equally restrictive nonlinear model. If the simple model is
counterfactual, in what sense does adding another simple feature to such a
contrived model “resolve” a very specific empirical puzzle? Adding another
shock may “loosen” the behavior of the model, but in what sense does this
make the model a better interpretation of reality? King also discusses the
shortcomings of this approach, highlighting problems in finding appropriate
9
See Altug [4].
October 9, 2009 15:12 9in x 6in b808-ch07
128 Business Cycles: Fact, Fallacy and Fantasy
instruments and in the appropriate selection of moment conditions for
model evaluation. Gregory and Smith [108] have also argued that the small-
sample properties of the estimators obtained with the approach advocated by
Christiano and Eichenbaum [65] may be far from reasonable if the calibrated
parameters do not consistently estimate the true values.
These arguments show that there is no clear-cut dichotomy between
these approaches. The estimation and testing of a highly restrictive linear
or linearized model may yield many insights regarding the failure of a model
as much as examining the performance of a model based on a small subset of
moments. These developments show that the contribution of the more formal
econometric techniques need not be dismissed as cursorily as is sometimes
the case. Conversely, one could argue that the approach in Christiano and
Eichenbaum [65] is too restrictive. As Geweke [102] notes, examining a small
set of moments does not necessarily lead to a less restrictive approach because
the moments under consideration typically incorporate all the implications
of the underlying theoretical model. In this sense, examining a broader set of
moments may make more sense because this yields more information about
the underlying behavior of the model. One interpretation of quantitative
theorizing is that it helps researchers understand the workings of highly non-
linear dynamic stochastic models and gain some intuition about different
model features. This interpretation precludes the notion that quantitative
theorizing can take the place of formal econometric methods. It is also worth
noting that many of the implications of the standard RBC model have not
withstood the scrutiny conducted under a variety of approaches. The simple
propagation mechanisms inherent in the model have been deemed incapable
of reproducing business cycle dynamics of key variables such as output, and
the New Keynesian challenge has shown that the model fails in generating the
observed negative response of hours to productivity improvements. Another
criticism of the RBC approach is that it fails to capture nonlinearities in key
macroeconomic time series.
Perhaps initially the RBC theorists were concerned that the estimation
versus calibration debate would discredit the use of well-specified dynamic
general equilibrium models for the purpose of capturing the quantitative
behavior of economic variables. In fact, far fromthis occurring, the estimation
versus calibration debate has led to a panoply of research that has extended
the use of these models in a variety of directions. The initial criticisms of
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 129
the RBC approach have led to a wide set of extensions of the original RBC
approach. As we discuss in the next chapter, the original RBC approach has
eveninstigateda newgenerationof Keynesianmodels that offer features suchas
credit accelerators, sunspot equilibria, and animal spirits. Another extension
of the original approach has been the development of applications that are
useful for policy analysis, a topic to which we turn next.
7.4. DSGE MODELING
Another offshoot of the original RBC debate is the development of dynamic
stochastic general equilibrium (DSGE) models that can be used for policy
analysis. The recent class of models developed for this purpose has vastly more
features than any simple RBC model, and a variety of techniques have been
developed for the purpose of matching the model to the data. One can ask
whether this class of models overcomes some of the criticisms leveled against
the original RBC approach.
Consider, for example, the model recently proposed by Kapetanios, Pagan,
andScott [123] for policy analysis of a small openeconomy. According tothem:
[This] model is stark in its assumptions. There are no market frictions and no
locational specificity. For example, there is no banking sector and no specific
role for money and credit in the monetary transmission mechanism. There
are no market frictions and distortions, no fixed costs or discontinuities. The
model assumes a representative household and a symmetric equilibrium for
firms. Above all, markets are assumed to clear at all times, as all agents have
complete knowledge of the economy and complete understanding of shocks
when they hit. In sum, the model contains assumptions that are almost
guaranteed to be violated by the data, especially in the short run.
Yet this model has a core set of 26 equations!
As another example, consider the model proposed by Christiano,
Eichenbaum, and Evans [67] for the analysis of the effects of monetary policy
shocks on real and nominal variables. The aim of this model is to reproduce
the inertial behavior of inflation and persistence in real quantities. To capture
the first feature, the model incorporates wage and price contracts following the
approach inCalvo [54]. The model also incorporates a variety of “real” frictions
such as habit persistence in consumption, adjustment costs in investment, and
variable capital utilization. Finally, firms are assumedto borrowworking capital
to finance their wage bill. There is also an interest-rate-setting rule that defines
October 9, 2009 15:12 9in x 6in b808-ch07
130 Business Cycles: Fact, Fallacy and Fantasy
monetary policy. Christiano et al. [67] pursue a limited information estimation
strategy by estimating a subset of the parameters of the model to match the
impulse responses of eight key macroeconomic variables to a monetary policy
shock with those implied from an identified VAR using the so-called method
of indirect inference.
10
The authors’ analysis is, in many ways, an exploratory
analysis of the role of nominal and real rigidities in propagating shocks. It is
also very much in the spirit of the original Kydland–Prescott approach [141]
in that the goal is to match a set of observed characteristics — in this case, the
impulse responses to a given monetary shock. Christiano et al. provide some
evidence on the role of price and wage rigidities, but in a highly restrictive
environment. For example, the model assumes a continuum of households
which are heterogeneous in their wage and the hours that they work, but
homogeneous in their consumption and asset holdings, with the latter result
derived from the existence of state-contingent securities for consumption.
Christiano et al. [67] derive much of their evidence regarding the types
of real rigidities on the results of calibration-type exercises of the current
generation of macroeconomic models such as habit persistence or adjustment
costs. However, there are shortcomings in their approach. For example,
estimated models of adjustment costs have been shown to imply implausible
degrees of costs of adjustment. Furthermore, the adjustment cost model has
been criticized on the grounds that it implies a constant cost of adjustment
which does not vary with economic conditions. By contrast, the irreversible
investment model studied by Demers [78] and others has been shown to
generate a time-varying adjustment cost that varies in response to changes in
objective and subjective forms of risk and uncertainty.
11
Christiano et al. [67]
assume that monetary policy shocks are drawn from a given and known
distribution. Yet, one could also question the implications of the model
should there be changes in the distribution of the exogenous processes. In
such cases, the assumption of a constant adjustment cost parameter would
fail to hold, implying that the propagation mechanisms would vary with
changes in the distribution of exogenous variables facing agents. This raises
10
For a recent example of the method of indirect inference applied to a model of the EU economy,
see Meenagh, Minford, and Wickens [160].
11
For a review and discussion, see also Demers, Demers, and Altug [79]. For applications, see Altug,
Demers, and Demers [10, 11].
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 131
the issue of whether DSGE models are, in fact, structural or invariant to
interventions.
Following the initial contribution of Altug [5], a variety of other papers
have implemented maximum likelihood estimation of dynamic equilibrium
models. McGrattan, Rogerson, and Wright [159] consider an equilibrium
business cycle model with household production and distortionary taxation.
As in the analysis of Altug [5], they augment the approximate linear laws of
motion with additional shocks. However, they employ time domain methods
by making use of the Kalman filter with a linear law of motion for the state
variables and a linear measurement equation for a key set of observables.
More recent applications include Canova [60] and Ireland [121], amongst
others. The recent research on DSGE models has also employed Bayesian
analysis as a way of incorporating prior uncertainty about the parameters of
interest (see, for example, Schorfhiede [186] and An and Schorfhiede [15]).
Similar to Canova [56], this approach first provides a link with the calibration
approach by allowing the researcher to use information from microeconomic
studies or macroeconometric estimates. Second, the use of Bayesian methods
and, in particular, examining the posterior distribution calculated as the
prior distribution times the likelihood function is often more straightforward
computationally than maximizing the likelihood function, which is typically
a high-dimensional object.
The Bayesian estimation of DSGE models involves several steps. First, the
method presupposes that a solution can be found for the underlying economic
model of interest. There are a variety of approaches for solving nonlinear
dynamic stochastic models. Judd [122] provides a textbook treatment of
many currently used solution methods. Second, the likelihood function for
the observations must be formed. Third, the posterior distribution for the
parameters, which is obtained from the prior distribution and the likelihood
function, must be examined. To illustrate this approach, suppose that the state
space representation for the model’s solution consists of the following:
• a transition equation S
t
= g(S
t −1
, v
t
, θ), where S
t
is a vector of state
variables that describe the evolution of the model over time, v
t
is a vector
of innovations, and θ is a vector of parameters characterizing preferences,
the production technology, and information; and
October 9, 2009 15:12 9in x 6in b808-ch07
132 Business Cycles: Fact, Fallacy and Fantasy
• a measurement equation Y
t
= h(S
t
, w
t
, θ), where Y
t
are the observables
and w
t
are the shocks to the observables (which may take the form of
measurement errors, among others).
This state space representation can be used to derive the likelihood function
of the observables and to obtain the posterior distribution for the parameters
conditional on the observations. Given the probability density functions for
the shocks f
v
( · ) and f
w
( · ), we can also compute the probabilities p(S
t
|S
t −1
; θ)
and p(Y
t
|S
t
; θ). Notice that
Y
t
= h(g(S
t −1
, v
t
, θ), w
t
, θ).
Hence, we can compute p(Y
t
|S
t −1
; θ). Likewise, the state space representation
allows us to compute the likelihood of the observations y
T
= (y
1
, . . . , y
T
) at
the parameter values θ as
p(y
T
; θ) = p(y
1
|θ)
T

t =2
p(y
t
|y
t −1
; θ)
=
_
p(S
1
|S
0
; θ)dS
0
T

t =2
_
p(Y
t
|S
t
; θ)p(S
t
|y
t −1
; θ)dS
t
.
Given a prior distribution π(θ) for the parameters θ, the posterior distribution
can be expressed as
π(θ|y
T
) =
p(y
T
|θ)π(θ)
_
p(y
T
|θ)π(θ)dθ
. (4.23)
The remaining taskis tocompute the sequence {p(S
t
|y
t −1
; θ)}
T
t =1
, whichshows
the conditional distribution of the states given the observations. The excellent
survey by Fernandez-Villaverde [90] describes in detail how to implement
Bayesian estimation of DSGE models. Following his approach, this sequence
is computed by making use of the relation
12
p(S
t +1
|y
t
; θ) =
_
p(S
t +1
|S
t
; θ)p(S
t
|y
t
; θ)dS
t
12
This is known as the Chapman–Kolmogorov equation.
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 133
and an application of Bayes’ theorem as
p(S
t
|y
t
; θ) =
p(y
t
|S
t
; θ)p(S
t
|y
t −1
; θ)
_
p(y
t
|S
t
; θ)p(S
t
|y
t −1
; θ)dS
t
.
Finding the posterior distribution of the parameters displayed in (4.23)
involves a few remaining steps. First, if the transition and measurement
equations are linear andif the shocks are normally distributed, thenthe Kalman
filter is used to obtain the likelihood function. Otherwise, a method known as
the particle filter
13
is used to evaluate the likelihood function. However, even
if we can evaluate the posterior distribution, exploring it for different values of
θ typically involves the use of sampling through Markov chain Monte Carlo
(MCMC) methods.
Smets and Wouters [194, 195] provide two important examples of this
approach. They extend the approach in Christiano et al. [67] to estimate
the parameters of a DSGE model with real and nominal frictions for the
euro area using Bayesian methods. In contrast to the approach in Christiano
et al. [67], however, they consider a full set of structural shocks. These
include a productivity shock, a labor supply shock, a shock to the household’s
discount factor, an investment-specific shock, and a government consumption
shock plus three “cost-push” shocks. They consider the behavior of seven
key macroeconomic time series in the euro area — real GDP, consumption,
investment, the GDP deflator, the real wage, employment, and the nominal
short-term interest rate — by minimizing the posterior distribution model’s
parameters basedona linearizedstate space representationfor the model. Smets
andWouters [194] argue that their approach offers several important advances.
First, they find that, based on Bayesian model evaluation criteria, the estimated
DSGE model performs as well as standard and Bayesian VARs. Second, they
find that the estimates of the structural parameters of the model are plausible
and, on the whole, similar to those for the US economy. Their estimates imply
more price stickiness than the estimates of Christiano et al. [67]. Third, they
argue that the effects of the different shocks on the variables of interest are
consistent with existing evidence that does not rely on a DSGEframework. For
example, a temporary contractionary monetary shock has a temporary negative
effect on output and inflation. Fourth, their findings attribute the largest role
13
See Schorfhiede [186] or Fernandez-Villaverde [90].
October 9, 2009 15:12 9in x 6in b808-ch07
134 Business Cycles: Fact, Fallacy and Fantasy
to labor supply and monetary shocks in accounting for the variation in output
in the euro area. One problem with this approach, as we discussed above, is
that there tends to be a circularity in building a model to match findings, say,
from a standard VAR, which may not be robust themselves.
In his survey written in 1992, Fair [86] expressed his disappointment with
both the RBC and New Keynesian research agendas in the following way:
“The RBC literature is interested in testing in only a very limited way, and the
NewKeynesian literature is not econometric enough to even talk about serious
testing.” In the last 20 years or so, the New Keynesian agenda has progressed
much closer towards gaining an explicitly econometric focus. Yet, it is difficult
to say how much headway has been made in understanding the propagation
mechanisms underlying business cycle fluctuations. Simple models with
features ranging from home production to government consumption shocks
to nonseparable preferences have been examined to understand their inner
workings, as have the roles of nominal rigidities and price-setting behavior.
While these model features have proved useful for accounting for some stylized
facts, they have often continued to retain counterfactual implications. For
example, Cogley and Nason [70] show that the home production framework
produces a negative autocorrelation for output. As discussed earlier, many
of the model features, such as symmetric and constant adjustment costs in
investment, are essentially ad hoc and are at variance with empirical evidence
at the micro level. The recent DSGE models include a large set of shocks that
have little economic meaning. By its very construction, the DSGE framework
cannot easily move into a setup that relaxes the representative consumer
assumption, be this through observed and unobserved forms of heterogeneity
or the presence of uninsurable risk. Yet, even under the complete markets
assumption, Altug and Miller [7, 8] show that exogenous and endogenous
forms of heterogeneity matter for the behavior of individual allocations.
Browning, Hansen, and Heckman [48] provide a comprehensive discussion
regarding the role of heterogeneity in general equilibrium modeling, and raise
cautionary points about linking the current class of dynamic equilibrium
models that dominate macroeconomics with microeconomic models and
evidence.
Could the entire general equilibrium macroeconomic quantitative
theorizing approach — despite the use of much bigger models and more
powerful computational tools — be a detour, as Fair [86] seems to indicate?
October 9, 2009 15:12 9in x 6in b808-ch07
Matching the Model to the Data 135
Should economists build intuition from the workings of well-articulated
economic models at the same time as they use more purely econometric
approaches for quantitative policy analysis? Kapetanios et al. [123] note
that the current class of fully articulated dynamic equilibrium models that
are increasingly being used by policy-makers such as central banks help to
provide restrictions on a reduced form, a practice very much in the spirit of
the Cowles Commission approach. Thus, the exercise is not necessarily to
recover structure in the form of the parameters of preferences and technology,
as argued by Lucas [150] in his famous critique of econometric policy
evaluation. More tellingly, can economists hope to recover “true structure”
using macroeconomic data in the presence of observed and unobserved forms
of heterogeneity? Or, alternatively, when subjective beliefs by economic agents
in a changing stochastic environment constitute “hidden structure”, as Martin
Weitzman [210] claims?
The notion that structural models may not indeed be structural has begun
to be discussed in the literature. Fernandez-Villaverde and Rubio-Ramirez
[91], for example, discuss the stability over time of estimated parameters
in DSGE models, and show that the parameters characterizing the pricing
behavior of households andfirms inCalvo-type pricing functions are correlated
with inflation. Canova and Sala [61] investigate identifiability issues in DSGE
models and their consequences for parameter estimation and model evaluation
when the objective function measures the distance between estimated and
model impulse responses. They show that observational equivalence as well
as partial and weak identification problems are widespread, lead to biased
estimates and unreliable t -statistics, and may induce investigators to select false
models. Canova [59] also examines the issue of identifiability of DSGEmodels,
and argues that researchers should be careful in interpreting the diagnostics
obtained from the estimation of structural models and make use of additional
data in the form of micro data or data from other countries to augment the
information obtained by formal estimation.
October 9, 2009 15:12 9in x 6in b808-ch07
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808-ch08
Chapter 8
Future Areas for Research
The literature onbusiness cycles is vast. Inthis book, I have offereda perspective
based on my own perceptions and mindsets. Nevertheless, even the experience
of reviewing some of this literature shows that it is lively, at times contentious,
but always thought-provoking. There are many other topics of interest that
we could have discussed regarding business cycles. For one, I have omitted a
discussion of multi-sector models in generating business cycle behavior. For
example, it may be that shocks are concentrated in a particular sector and are
propagated to the rest of the economy from that sector.
1
In such cases, one
needs to consider multi-sector models since the one-sector growth model does
not allowfor a considerationof these issues. However, multi-sector models may
also have shortcomings because they typically imply that overall expansions in
the economy are associated with contractions in one sector.
More generally, I have not provided a discussion of models with self-
fulfilling expectations and multiple equilibria. Farmer [87] provides an
eloquent enunciation of this approach, and contrasts it with the standard real
business cycle (RBC) approach. Many have argued that such models are more
inthe spirit of true Keynesianismbecause they stress the role of “animal spirits”.
In a recent analysis, Farmer [88] generates a model of self-fulfilling equilibria
which arises from market failure associated with the labor market. Specifically,
he considers the role of a search externality and a lemons problemin the market
for search inputs. Many recent works assume that there is a search technology
such that firms and workers are randomly matched; the wage is then set by a
1
See, for example, Benhabib and Farmer [38].
137
October 9, 2009 15:12 9in x 6in b808-ch08
138 Business Cycles: Fact, Fallacy and Fantasy
Nash bargaining process.
2
In contrast to this literature, Farmer [88] drops the
assumption that wages are determined as a result of a Nash bargaining process.
Instead, he assumes that firms offer the same wage in advance. This leads to a
model with a continuumof steady-state equilibria. In these equilibria, all firms
earn zero profits but not all equilibria have the same welfare properties. This
feature of the model allows for the role of “confidence”, which is transmitted
to real variables through investors’ beliefs about the stock market. This leads
to a demand-constrained equilibrium. The policy implications of such a model
differ from those of the standard textbook Keynesian model, which seeks to
alleviate demand-induced shortfalls in output through large fiscal stimuli. By
contrast, the model with a search externality emphasizes the role of confidence
in restoring full-employment output. Farmer [88] concludes his analysis by
commenting on the potential efficacy of the Obama fiscal stimulus package(s)
in light of the findings from his model.
This book has also omitted a discussion of models with credit and collateral
constraints. One could argue that this type of model deserves much more
attention given the global financial crisis that originated from the subprime
housing market in the US in 2007.
3
The above considerations show that business cycles will continue to be a
topic of discussion among economists for many years to come. Undoubtedly,
this discussion will be accompanied by the development of new models, new
techniques, and new controversies regarding both. However, irrespective of
what the new models or techniques will look like, we may conclude that the
debate surrounding their development will not cease to arouse interest or to
generate future avenues for research.
2
This approach owes its origins to the analysis in Mortensen and Pissarides [165]. For a recent
application in the context of a standard RBC model, see Cooley and Quadrini [76].
3
For a recent analysis of models with financial frictions in a dynamic general equilibrium setting, see
Pierrard, de Walque, and Rouabah [171].
October 9, 2009 15:13 9in x 6in b808-bib
Bibliography
1. Aguiar, M. and G. Gopinath (2004). “Emerging Market Business Cycles: The Trend Is
the Cycle,” Journal of Political Economy, 115, pp. 69–102.
2. A’Hearna, B. and U. Woitek (2001). “More International Evidence on the Historical
Properties of Business Cycles,” Journal of Monetary Economics, 47, pp. 321–346.
3. Ahmed, S. and R. Murthy (1994). “Money, Output, and Real Business Cycles in a Small
Open Economy,” Canadian Journal of Economics, 27, pp. 982–993.
4. Altug, S. (1985). “Gestation Lags and the Business Cycle,” Unpublished manuscript.
5. Altug, S. (1989). “Time-to-Build and Aggregate Fluctuations: Some New Evidence,”
International Economic Review, 30, pp. 889–920.
6. Altug, S. and P. Labadie (2008). Asset Pricing for Dynamic Economies. Cambridge:
Cambridge University Press.
7. Altug, S. and R. Miller (1990). “Household Choices in Equilibrium,” Econometrica, 58,
pp. 543–570.
8. Altug, S. and R. Miller (1998). “The Effect of Past Experience on Female Wages and
Labor Supply,” Review of Economic Studies, 65, pp. 45–85.
9. Altug, S., R. Ashley, and D. Patterson (1999). “Are Technology Shocks Nonlinear?”
Macroeconomic Dynamics, 3, pp. 506–533.
10. Altug, S., F. Demers, and M. Demers (2007). “Political Risk and Irreversible Investment,”
CESifo Economic Studies, 53, pp. 430–465.
11. Altug, S., F. Demers, and M. Demers (2009). “The Investment Tax Credit and Irreversible
Investment,” Journal of Macroeconomics, forthcoming.
12. Alvarez, F. and U. Jermann (2000). “Efficiency, Equilibrium, and Asset Pricing with Risk
of Default,” Econometrica, 68, pp. 775–797.
13. Ambler, S., E. Cardia, and C. Zimmermann (2002). “International Transmission of the
Business Cycle in a Multi-sector Model,” European Economic Review, 46, pp. 273–300.
14. Ambler, S., E. Cardia, and C. Zimmermann (2004). “International Business Cycles: What
Are the Facts?” Journal of Monetary Economics, 51, pp. 257–276.
139
October 9, 2009 15:13 9in x 6in b808-bib
140 Business Cycles: Fact, Fallacy and Fantasy
15. An, S. and F. Schorfhiede (2007). “Bayesian Analysis of DSGE Models,” Econometric
Reviews, 26, pp. 113–172.
16. Anderson, E., L. Hansen, E. McGrattan, and T. Sargent (1996). “Mechanics of Forming
and Estimating Dynamic Linear Economies,” In H. Amman, D. Kendrick, and J. Rust
(eds.), Handbook of Computational Economics Vol. 1, Handbooks in Economics, Vol. 13,
Amsterdam: Elsevier Science/North-Holland, pp. 171–252.
17. Arellano, C. and E. Mendoza (2003). “Credit Frictions and ‘Sudden Stops’ in Small Open
Economies: An Equilibrium Business Cycle Framework for Emerging Markets Crises,”
In S. Altug, J. Chadha, and C. Nolan (eds.), Dynamic Macroeconomic Analysis: Theory and
Policy in General Equilibrium, Cambridge: Cambridge University Press, pp. 335–405.
18. Artis, M. and W. Zhang (1997). “International Business Cycles and the ERM: Is There a
European Business Cycle?” International Journal of Finance and Economics, 2, pp. 1–16.
19. Artis, M. and W. Zhang (1999). “Further Evidence on the International Business Cycle
and the ERM: Is There a European Business Cycle?” Oxford Economic Papers, 51,
pp. 120–132.
20. Artis, M., Z. Kontolemis, and D. Osborn (1997). “Business Cycles for G7 and European
Countries,” Journal of Business, 70, pp. 249–279.
21. Artis, M., M. Marcellino, and T. Proietti (2003). “Dating the Euro Area Business Cycle,”
Innocenzo Gasparini Institute for Economic Research (IGIER) Working Paper No. 237.
22. Ashley, R. and D. Patterson (1989). “Linear versus Nonlinear Macroeconomics: A
Statistical Test,” International Economic Review, 30, pp. 685–704.
23. Backus, D. and P. Kehoe (1992). “International Evidence on the Historical Perspective
of Business Cycles,” American Economic Review, 82, pp. 864–888.
24. Backus, D. and G. Smith (1993). “Consumption and Real Exchange Rates in Dynamic
Economies withNon-traded Goods,” Journal of International Economics, 35, pp. 297–316.
25. Backus, D., P. Kehoe, andF. Kydland(1992). “International Real Business Cycles,” Journal
of Political Economy, 100, pp. 745–775.
26. Backus, D., P. Kehoe, and F. Kydland (1995). “International Business Cycles: Theory and
Evidence,” In T. Cooley (ed.), Frontiers of Business Cycle Research, Princeton: Princeton
University Press, pp. 331–356.
27. Barro, R. (1987). “Government Spending, Interest Rates, Prices and Budget Deficits in
the United Kingdom, 1701–1918,” Journal of Monetary Economics, 20, pp. 221–247.
28. Barsky, R. and L. Kilian (2004). “Oil and the Macroeconomy Since the 1970s,” Journal
of Economic Perspectives, 18, pp. 115–134.
29. Basu, S. (1996). “Procyclical Productivity: Increasing Returns or Cyclical Utilization?”
Quarterly Journal of Economics, 111, pp. 719–751.
30. Basu, S. and M. Kimball (1997). “Cyclical Productivity with Unobserved Input
Variation,” NBER Working Paper No. 5915.
31. Basu, S. and A. Taylor (1999). “Business Cycles in International Historical Perspective,”
Journal of Economic Perspectives, 13, pp. 45–68.
32. Basu, S., J. Fernald, and M. Kimball (2006). “Are Technology Improvements
Contractionary?” American Economic Review, 96, pp. 1418–1448.
33. Baxter, M. (1995). “International Trade and Business Cycles,” In G. Grossman and
K. Rogoff (eds.), Handbook of International Economics, Amsterdam: North-Holland,
pp. 1801–1864.
October 9, 2009 15:13 9in x 6in b808-bib
Bibliography 141
34. Baxter, M. and M. Crucini (1995). “Business Cycles and the Asset Structure of Foreign
Trade,” International Economic Review, 36, pp. 821–854.
35. Baxter, M. and D. Farr (2005). “Variable Factor Utilization and International Business
Cycles,” Journal of International Economics, 65, pp. 335–347.
36. Baxter, M. and U. Jermann (1997). “The International Diversification Puzzle Is Worse
Than You Think,” American Economic Review, 87(1), pp. 170–180.
37. Baxter, M. and R. King (1999). “Measuring Business Cycles: Approximate Band-
pass Filters for Economic Time Series,” The Review of Economics and Statistics, 81,
pp. 575–593.
38. Benhabib, J. and R. Farmer (1996). “Indeterminacy and Sector-Specific Externalities,”
Journal of Monetary Economics, 37, pp. 421–443.
39. Benhabib, J., R. Rogerson, and R. Wright (1991). “Homework in Macroeconomics:
Household Production and Aggregate Fluctuations,” Journal of Political Economy, 99,
pp. 1166–1187.
40. Bernanke, B. and M. Gertler (1989). “Agency Costs, Net Worth and Business
Fluctuations,” American Economic Review, 79, pp. 14–31.
41. Bernanke, B., M. Gertler, and S. Gilchrist (1996). “The Financial Accelerator and the
Flight to Quality,” Review of Economics and Statistics, 78, pp. 1–15.
42. Bertola, G. and R. Caballero (1994). “Irreversibility and Aggregate Investment,” Review
of Economic Studies, 61, pp. 223–246.
43. Blanchard, O. and D. Quah (1989). “The Dynamic Effects of Demand versus Supply
Disturbances,” American Economic Review, 79, pp. 654–673.
44. Bordo, M. and T. Heibling (2003). “Have National Business Cycles Become More
Synchronized?” NBER Working Paper No. W10130.
45. Bowman, D. and B. Doyle (2003). “New Keynesian, Open-Economy Models and Their
Implications for Monetary Policy,” Federal Reserve Board (FRB) International Finance
Discussion Paper No. 2003-762.
46. Braun, A. (1994). “Tax Disturbances and Real Activity in the Postwar United States,”
Journal of Monetary Economics, 33, pp. 441–462.
47. Brock, W. and C. Sayers (1988). “Is the Business Cycle Characterized by Deterministic
Chaos?” Journal of Monetary Economics, 22, pp. 71–90.
48. Browning, M., L. Hansen, andJ. Heckman(1999). “Micro Data andGeneral Equilibrium
Models,” In J. Taylor and M. Woodford (eds.), Handbook of Macroeconomics, Vol. 1A,
Amsterdam: Elsevier Science, Ch. 8.
49. Bry, G. and C. Boschan (1971). Cyclical Analysis of Time Series: Selected Procedures and
Computer Programs. New York: Columbia University Press for the NBER.
50. Burns, A. and W. Mitchell (1946). Measuring Business Cycles. New York: NBER.
51. Burnside, C. and M. Eichenbaum (1996). “Factor Hoarding and the Propagation of
Business Cycle Shocks,” American Economic Review, 86, pp. 1154–1174.
52. Burnside, C., M. Eichenbaum, and S. Rebelo (1993). “Labor Hoarding and the Business
Cycle,” Journal of Political Economy, 101, pp. 245–274.
53. Caballero, R. (1999). “Aggregate Investment,” In J. Taylor and M. Woodford (eds.),
Handbook of Macroeconomics, Vol. 1B, North Holland: Elsevier Science, Ch. 12.
54. Calvo, G. (1983). “Staggered Prices in a Utility-Maximizing Framework,” Journal of
Monetary Economics, 12, pp. 383–398.
October 9, 2009 15:13 9in x 6in b808-bib
142 Business Cycles: Fact, Fallacy and Fantasy
55. Campbell, J. (1994). “Inspecting the Mechanism: An Analytical Approach to the
Stochastic Growth Model,” Journal of Monetary Economics, 33, pp. 463–506.
56. Canova, F. (1994). “Statistical Inference in Calibrated Models,” Journal of Applied
Econometrics, 9, pp. 123–144.
57. Canova, F. (1995). “Sensitivity Analysis and Model Evaluation in Simulated Dynamic
General Equilibrium Economies,” International Economic Review, 36, pp. 477–501.
58. Canova, F. (2006). Methods for Applied Macroeconomic Analysis. Princeton: Princeton
University Press.
59. Canova, F. (2008). “HowMuch Structure in Empirical Models?” CEPRDiscussion Paper
No. 6791.
60. Canova, F. (2009). “What Explains the Great Moderation in the US? A Structural
Analysis,” Journal of the European Economic Association, 7, pp. 697–721.
61. Canova, F. and L. Sala (2006). “Back to Square One: Identification Issues in DSGE
Models,” ECB Working Paper No. 583.
62. Chadha, J. and C. Nolan (2002). “A Long View of the UK Business Cycle,” National
Institute Economic Review, 182, pp. 72–89.
63. Chari, V., P. Kehoe, and E. McGrattan (2004). “Are Structural VARs Useful Guides for
Developing Business Cycle Theories?” Federal Reserve Bank of Minneapolis Research
Department Working Paper No. 631.
64. Christiano, L. (1988). “Why Does Inventory Investment Fluctuate So Much?” Journal of
Monetary Economics, 21, pp. 247–280.
65. Christiano, L. and M. Eichenbaum (1992). “Current Real Business Cycle Theories and
Aggregate Labor Market Fluctuations,” American Economic Review, 82, pp. 430–450.
66. Christiano, L., M. Eichenbaum, and C. Evans (1997). “Sticky Price and Limited
Participation Models of Money: A Comparison,” European Economic Review, 41,
pp. 1201–1249.
67. Christiano, L., M. Eichenbaum, and C. Evans (2005). “Nominal Rigidities and the
Dynamic Effects of a Shock to Monetary Policy,” Journal of Political Economy, 113,
pp. 1–45.
68. Christiano, L., M. Eichenbaum, and R. Vigfusson (2003). “What Happens After a
Technology Shock?” NBER Working Paper No. 9819.
69. Cogley, T. and J. Nason (1995). “Effects of the Hodrick–Prescott Filter on Trend and
Difference Stationary Time Series: Implications for Business Cycle Research,” Journal of
Economic Dynamics and Control, 19, pp. 253–278.
70. Cogley, T. and J. Nason (1995). “Output Dynamics in Real Business Cycle Models,”
American Economic Review, 85, pp. 492–511.
71. Cole, H. (1988). “Financial Structure and International Trade,” International Economic
Review, 29, pp. 237–259.
72. Cole, H. and M. Obstfeld (1991). “Commodity Trade and International Risk Sharing,”
Journal of Monetary Economics, 28, pp. 3–24.
73. Coleman, W. (1997). “The Behavior of Interest Rates in a General Equilibrium Multi-
sector Model with Irreversible Investment,” Macroeconomic Dynamics, 1, pp. 206–227.
74. Cooley, T. (ed.) (1995). Frontiers of Business Cycle Research. Princeton: PrincetonUniversity
Press.
75. Cooley, T. and E. Prescott (1995). “Economic Growth and Business Cycles,” InT. Cooley
(ed.), Frontiers of Business Cycle Research, Princeton: Princeton University Press, pp. 1–38.
October 9, 2009 15:13 9in x 6in b808-bib
Bibliography 143
76. Cooley, T. and V. Quadrini (1999). “ANeoclassical Model of the Phillips Curve Relation,”
Journal of Monetary Economics, 44, pp. 165–193.
77. Correia, I., J. Neves, and S. Rebelo (1995). “Business Cycles in Small Open Economies,”
European Economic Review, 39, pp. 1089–1113.
78. Demers, M. (1991). “Irreversibility and the Arrival of Information,” Review of Economic
Studies, 58, pp. 333–350.
79. Demers, F., M. Demers, and S. Altug (2003). “Investment Dynamics,” In S. Altug,
J. Chadha, and C. Nolan (eds.), Dynamic Macroeconomic Analysis: Theory and Policy in
General Equilibrium, Cambridge: Cambridge University Press, pp. 34–154.
80. Denison, E. (1985). Trends in American Economic Growth, 1929–1982. Washington, DC:
Brookings.
81. Devereux, M., A. Gregory, and G. Smith (1992). “Realistic Cross-Country Consumption
Correlations in a Two-Country, Equilibrium Business Cycle Model,” Journal of
International Money and Finance, 11, pp. 3–16.
82. Eichenbaum, M. (1991). “Real Business Cycles: Wisdomor Whimsy?” Journal of Economic
Dynamics and Control, 15, pp. 607–626.
83. Eichenbaum, M., L. Hansen, and K. Singleton (1988). “A Time Series Analysis of
Representative Agent Models of Consumption and Leisure Choice Under Uncertainty,”
Quarterly Journal of Economics, 103, pp. 51–78.
84. Eichengreen, B. and M. Bordo (2002). “Crises Now and Then: What Lessons from the
Last Era of Financial Globalization?” NBER Working Paper No. W8716.
85. Fagan, G., J. Henry, and R. Mestre (2001). “An Area-Wide Model (AWM) for the Euro
Area,” ECB Working Paper No. 42.
86. Fair (1992). “The Cowles Commission Approach, Real Business Cycle Theories, and New
Keynesian Economics,” Cowles Foundation Discussion Paper No. 1004.
87. Farmer, R. (1999). The Macroeconomics of Self-Fulfilling Prophecies, 2nd ed. Cambridge,
MA: MIT Press.
88. Farmer, R. (2009). “Confidence, Crashes, and Animal Spirits,” NBER Working Paper
No. 14846.
89. Feldstein, M. andC. Horioka (1980). “Domestic Saving andInternational Capital Flows,”
Economic Journal, 90, pp. 314–329.
90. Fernandez-Villaverde, J. (2009). “The Econometrics of DSGE Models,” NBERWorking
Paper No. 14677.
91. Fernandez-Villaverde, J. and R. Rubio-Ramirez (2007). “How Structural Are Structural
Parameters?” NBER Working Paper No. 13166.
92. Finn, M. (2000). “Perfect Competition and the Effects of Energy Price Increases on
Economic Activity,” Journal of Money, Credit, and Banking, 32, pp. 400–416.
93. Forni, M. and L. Reichlin (1998). “Let’s Get Real: A Factor Analytical Approach to
Disaggregated Business Cycle Dynamics,” Review of Economic Studies, 65, pp. 453–473.
94. Friedman, M. and A. Schwartz (1963). A Monetary History of the United States, 1867–
1960. Princeton: Princeton University Press.
95. Frisch, R. (1933). “PropagationProblems andImpulse Problems inDynamic Economies,”
In Economic Essays in Honor of Gustav Cassel, London: George Allen & Unwin.
96. Gali, J. (1999). “Technology, Employment, and the Business Cycle: Do Technology
Shocks Explain Aggregate Fluctuations?” American Economic Review, 89, pp. 249–271.
October 9, 2009 15:13 9in x 6in b808-bib
144 Business Cycles: Fact, Fallacy and Fantasy
97. Gali, J. (2008). Monetary Policy, Inflation, and the Business Cycle. Princeton: Princeton
University Press.
98. Gali, J. and P. Rabanal (2005). “Technology Shocks and Aggregate Fluctuations: How
Well Does the RBC Model Fit Postwar US Data?” In M. Gertler and K. Rogoff (eds.),
NBER Macroeconomics Annual 2004, Cambridge, MA: MIT Press, pp. 225–318.
99. Gallant, R. and G. Tauchen (1996). “Which Moments to Match?” Econometric Theory,
12, pp. 657–681.
100. Gallant, R. and G. Tauchen (1998). “SNP: A Program for Nonparametric Time Series
Analysis, Version 8.7, User’s Guide,” University of North Carolina Working Paper.
101. Garcia-Cicco, J., R. Pancrazi, and M. Uribe (2006). “Real Business Cycles in Emerging
Countries?” NBER Working Paper No. 12629.
102. Geweke, J. (1999). “Computational Experiments and Reality,” SCE Computing in
Economics and Finance Working Paper No. 401.
103. Giannone, D., L. Reichlin, and L. Sala (2006). “VARs, Common Factors and the
Empirical Validation of Equilibrium Business Cycle Models,” Journal of Econometrics,
132, pp. 257–279.
104. Gordon, R. (1990). The Measurement of Durable Goods Prices. Chicago: University of
Chicago Press.
105. Greenwood, J., Z. Hercowitz, andG. Huffman(1988). “Investment, Capacity Utilization,
and the Real Business Cycle,” American Economic Review, 78, pp. 402–417.
106. Greenwood, J., Z. Hercowitz, and P. Krusell (1997). “Long-Run Implications
of Investment-Specific Technological Change,” American Economic Review, 87,
pp. 342–362.
107. Greenwood, J., Z. Hercowitz, and P. Krusell (2000). “The Role of Investment-Specific
Technological Change in the Business Cycle,” European Economic Review, 44, pp. 91–115.
108. Gregory, A. and A. Smith (1989). “Calibration as Estimation,” Econometric Reviews, 9,
pp. 57–89.
109. Gregory, A., A. Head, and J. Raynauld (1997). “Measuring World Business Cycles,”
International Economic Review, 38, pp. 677–702.
110. Hall, R. (1988). “The Relation Between Price and Marginal Cost in U.S. Industry,”
Journal of Political Economy, 96, pp. 921–947.
111. Hall, R. (1990). “Invariance Properties of Solow’s Productivity Residual,” In P. Diamond
(ed.), Growth/Productivity/Unemployment, Cambridge, MA: MIT Press, pp. 71–112.
112. Hansen, G. (1985). “Indivisible Labor and the Business Cycle,” Journal of Monetary
Economics, 16, pp. 309–327.
113. Hansen, L. (1982). “Large Sample Properties of Generalized Method of Moments
Estimators,” Econometrica, 50, pp. 1029–1054.
114. Hansen, L. andT. Sargent (1980). “Formulating and Estimating Dynamic Linear Rational
Expectations Models,” Journal of Economic Dynamics and Control, 2, pp. 7–46.
115. Harding, D. and A. Pagan (2003). “A Comparison of Business Cycle Dating Methods,”
Journal of Economic Dynamics and Control, 27, pp. 1681–1690.
116. Hartley, J., K. Hoover, and K. Salyer (eds.) (1998). Real Business Cycles: A Reader. London:
Routledge.
117. Heathcote, J. and F. Perri (2002). “Financial Autarky and International Business Cycles,”
Journal of Monetary Economics, 49, pp. 601–627.
October 9, 2009 15:13 9in x 6in b808-bib
Bibliography 145
118. Helliwell, J. (1998). HowMuch Do National Borders Matter? Washington, DC: Brookings
Institution.
119. Hess, G. and K. Shin (1997). “International and Intranational Business Cycles,” Oxford
Review of Economic Policy, 13, pp. 93–109.
120. Hodrick, R. and E. Prescott (1997). “Postwar U.S. Business Cycles,” Journal of Money,
Credit, and Banking, 29, pp. 1–16.
121. Ireland, P. (2004). “Technology Shocks inthe NewKeynesianModel,” Reviewof Economics
and Statistics, 86, pp. 923–936.
122. Judd, K. (1998). Numerical Methods in Economics. Cambridge, MA: MIT Press.
123. Kapetanios, G., A. Pagan, and A. Scott (2007). “Making a Match: CombiningTheory and
Evidence in Policy-Oriented Macroeconomic Modeling,” Journal of Econometrics, 136,
pp. 565–594.
124. Kehoe, T. and D. Levine (1993). “Debt Constrained Asset Markets,” Review of Economic
Studies, 60, pp. 865–888.
125. Kehoe, T. andD. Levine (2001). “Liquidity ConstrainedMarkets versus Debt Constrained
Markets,” Econometrica, 69(3), pp. 575–598.
126. Kehoe, P. and F. Perri (2002). “International Business Cycles with Endogenous Incomplete
Markets,” Econometrica, 70, pp. 907–928.
127. Keynes, J. M. (1936). The General Theory of Employment, Interest, and Money. London:
Macmillan.
128. King, R. (1995). “Quantitative Theory and Econometrics,” Economic Quarterly, 81,
pp. 53–105.
129. King, R. and C. Plosser (1984). “Money, Credit, and Prices in a Real Business Cycle
Model,” American Economic Review, 74, pp. 363–380.
130. King, R. and S. Rebelo (1993). “Low Frequency Filtering and Real Business Cycles,”
Journal of Economic Dynamics and Control, 17, pp. 207–231.
131. King, R., C. Plosser, and S. Rebelo (1988). “Production, Growth, and Business Cycles:
I. The Basic Neoclassical Model,” Journal of Monetary Economics, 21, pp. 195–232.
132. King, R., C. Plosser, and S. Rebelo (1988). “Production, Growth, and Business Cycles:
II. New Directions,” Journal of Monetary Economics, 21, pp. 309–341.
133. King, R., C. Plosser, J. Stock, and M. Watson (1991). “Stochastic Trends and Economic
Fluctuations,” American Economic Review, 81, pp. 819–840.
134. Kiyotaki, N. and J. Moore (1997). “Credit Cycles,” Journal of Political Economy, 105,
pp. 211–248.
135. Kocherlakota, N. (1996). “Implications of Efficient Risk Sharing without Commitment,”
Review of Economic Studies, 63, pp. 595–609.
136. Kollman, R. (1995). “Consumption, Real Exchange Rates, and the Structure of
International Asset Markets,” Journal of International Money and Finance, 14,
pp. 191–211.
137. Kondratiev, N. (1926). “Die Langen Wellen der Konjunktur [The Long Waves of the
Business Cycle],” Archiv für Sozialwissenschaft und Sozialpolitik, 56, pp. 573–606.
138. Köse, A., C. Otrok, and C. Whiteman (2003). “International Business Cycles: World,
Region, and Country-Specific Factors,” American Economic Review, 93, pp. 1216–1239.
139. Köse, A., E. Prasad, and M. Terrones (2003). “How Does Globalization Affect the
Synchronization of Business Cycles?” IZA Discussion Paper No. 702.
October 9, 2009 15:13 9in x 6in b808-bib
146 Business Cycles: Fact, Fallacy and Fantasy
140. Kydland, F. (1995). “Business Cycles and Aggregate Labor Market Fluctuations,” In
T. Cooley (ed.), Frontiers of Business Cycle Research, Princeton: Princeton University Press,
pp. 126–156.
141. Kydland, F. and E. Prescott (1982). “Time-to-Build and Aggregate Fluctuations,”
Econometrica, 50, pp. 1345–1370.
142. Kydland, F. and E. Prescott (1990). “Business Cycles: Real Facts and a Monetary Myth,”
Federal Reserve Bank of Minneapolis Quarterly Review, 14, pp. 3–18.
143. Kydland, F. and E. Prescott (1991). “The Econometrics of the General Equilibrium
Approach to Business Cycles,” Scandinavian Journal of Econometrics, 93, pp. 161–178.
144. Kydland, F. and C. E. J. M. Zaragaza (2002). “Argentina’s Lost Decade,” Review of
Economic Dynamics, 5, pp. 152–165.
145. Lewis, K. (1996). “What Can Explain the Apparent Lack of International Consumption
Risk Sharing?” Journal of Political Economy, 104(2), pp. 267–297.
146. Ljungqvist, L. and T. Sargent (2000). Recursive Macroeconomic Theory. Cambridge, MA:
MIT Press.
147. Long, J. and C. Plosser (1983). “Real Business Cycles,” Journal of Political Economy, 91,
pp. 39–69.
148. Lucas, R. (1972). “Expectations and the Neutrality of Money,” Journal of Economic Theory,
4, pp. 103–123.
149. Lucas, R. (1975). “An Equilibrium Model of the Business Cycle,” Journal of Political
Economy, 83, pp. 1113–1144.
150. Lucas, R. (1976). “Econometric Policy Evaluation: A Critique,” Journal of Monetary
Economics, 1(2) (Supplementary Series), pp. 19–46.
151. Lucas, R. (1977). “Understanding Business Cycles,” In K. Brunner and A. Meltzer (eds.),
Stabilization of the Domestic and International Economy, Amsterdam: North-Holland,
pp. 7–29.
152. Lucas, R. (1982). “Interest Rates and Currency Prices in a Two-Country World,” Journal
of Monetary Economics, 10, pp. 335–359.
153. Lucas, R. and L. Rapping (1969). “Real Wages, Employment and Inflation,” Journal of
Political Economy, 77, pp. 721–754.
154. Lumsdaine, R. and E. Prasad (2003). “Identifying the Common Component in
International Economic Fluctuations,” Economic Journal, 113, pp. 101–127.
155. Marcet, A. and R. Marimon (1999). “Recursive Contracts,” Universitat Pompeu Fabra
(UPF) Economics Working Paper No. 337.
156. McCallum, B. (1986). “On ‘Real’ and ‘Sticky-Price’ Theories of the Business Cycle,”
Journal of Money, Credit, and Banking, 18, pp. 397–414.
157. McCallum, J. (1996). “National Borders Matter: Canada-US Regional Trade Patterns,”
American Economic Review, 85, pp. 615–623.
158. McGrattan, E. (1994). “The Macroeconomic Effects of Distortionary Taxation,” Journal
of Monetary Economics, 33, pp. 573–601.
159. McGrattan, E., R. Rogerson, and R. Wright (1997). “An Equilibrium Model of the
Business Cycle with Household Production and Fiscal Policy,” International Economic
Review, 38, pp. 267–290.
160. Meenagh, D., P. Minford, and M. Wickens (2008). “Testing a DSGE Model of the
EU Using Indirect Inference,” Cardiff Business School Economics Working Paper
No. E2008/11.
October 9, 2009 15:13 9in x 6in b808-bib
Bibliography 147
161. Meese, R. and K. Rogoff (1983). “Empirical Exchange Rate Models of the Seventies: Do
They Fit Out of Sample?” Journal of International Economics, 14, pp. 3–24.
162. Mendoza, E. (1991). “Real Business Cycles in a Small Open Economy,” American
Economic Review, 81, pp. 797–818.
163. Mitchell, W. (1927). Business Cycles: The Problem and Its Setting. New York: National
Bureau of Economic Research.
164. Mitchell, W. and A. Burns (1938). Statistical Indicators of Cyclical Revivals (NBERBulletin
No. 69). New York: National Bureau of Economic Research.
165. Mortensen, D. and C. Pissarides (1994). “Job Creation and Job Destruction in the Theory
of Unemployment,” Review of Economic Studies, 61, pp. 391–415.
166. Neftci, S. (1984). “Are EconomicTime Series Symmetric over the Business Cycle?” Journal
of Political Economy, 92, pp. 307–328.
167. Nelson, C. and C. Plosser (1982). “Trends and Random Walks in Macroeconomic Time
Series,” Journal of Monetary Economics, 10, pp. 139–162.
168. Obstfeld, M. and K. Rogoff (2001). “The Six Major Puzzles in International
Macroeconomics: Is There a Common Cause?” In B. Bernanke and K. Rogoff (eds.),
NBER Macroeconomics Annual 2000, Cambridge, MA: MIT Press, pp. 339–390.
169. Ohanian, L. (1997). “The Macroeconomic Effects of War Finance in the United
States: World War II and the Korean War,” American Economic Review, 87,
pp. 23–40.
170. Phelps, E. (1970). Microeconomic Foundations of Employment and Inflation Theory. New
York: W.W. Norton & Co.
171. Pierrard, O., G. de Walque, and A. Rouabah (2008). “Financial (In)stability, Supervision
and Liquidity Injections: A Dynamic General Equilibrium Approach,” Unpublished
manuscript.
172. Potter, S. (1999). “Nonlinear Time Series Modelling: An Introduction,” Journal of
Economic Surveys, 13, pp. 505–528.
173. Prescott, E. (1986). “Theory Ahead of Business Cycle Measurement,” Carnegie-Rochester
Conference Series on Public Policy, 25, pp. 11–44.
174. Ramey, V. and M. Shapiro (2001). “Displaced Capital: A Study of Aerospace Plant
Closings,” Journal of Political Economy, 109, pp. 958–992.
175. Ratfai, A. and P. Benczur (2005). “Economic Fluctuations in Central and Eastern Europe:
The Facts,” CEPR Discussion Paper No. 4846.
176. Ratfai, A. and P. Benczur (2008). “Facts of Business Cycles Around the Globe,” Paper
presented at the IEA 15th World Congress, Istanbul, Turkey.
177. Rebelo, S. (2005). “Real Business Models: Past, Present, andFuture,” ScandinavianJournal
of Economics, 107, pp. 217–238.
178. Rodrik, D. (1991). “Policy Uncertainty and Private Investment inDeveloping Countries,”
Journal of Development Economics, 36, pp. 229–242.
179. Rogerson, R. (1988). “Indivisible Labor, Lotteries and Equilibrium,” Journal of Monetary
Economics, 21, pp. 3–16.
180. Romer, C. (1986). “Is the Stabilization of the Postwar Economy a Figment of the Data?”
American Economic Review, 76, pp. 316–334.
181. Romer, C. (1989). “The Prewar Business Cycle Reconsidered: New Estimates of GNP,
1869–1908,” Journal of Political Economy, 97, pp. 1–37.
October 9, 2009 15:13 9in x 6in b808-bib
148 Business Cycles: Fact, Fallacy and Fantasy
182. Rotemberg, J. and M. Woodford (1995). “Dynamic General Equilibrium Models of
Imperfectly Competitive Product Markets,” In T. Cooley (ed.), Frontiers of Business Cycle
Research, Princeton: Princeton University Press, pp. 243–293.
183. Rotemberg, J. and M. Woodford (1996). “Imperfect Competition and the Effects of
Energy Price Increases on Economic Activity,” Journal of Money, Credit, and Banking, 28,
pp. 549–577.
184. Sargent, T. (1989). “Two Models of Measurement Error and the Investment Accelerator,”
Journal of Political Economy, 97, pp. 251–287.
185. Sargent, T. and C. Sims (1977). “Business Cycle Modeling Without Pretending to Have
Too Much A Priori Economic Theory,” In New Methods in Business Cycle Research:
Proceedings from a Conference, Minneapolis, MN: Federal Reserve Bank of Minneapolis,
pp. 45–109.
186. Schorfhiede, F. (2000). “Loss Function-based Evaluation of DSGE Models,” Journal of
Applied Econometrics, 15, pp. 645–670.
187. Schumpeter, J. (1934). The Theory of Economic Growth. Cambridge, MA: Harvard
University Press.
188. Schumpeter, J. (1939). Business Cycles. New York: McGraw-Hill.
189. Serven, L. and A. Solimano (1993). “Private Investment and Macroeconomic Adjustment:
A Survey,” In L. Serven and A. Solimano (eds.), Striving for Growth After Adjustment: The
Role of Capital Formation, Washington, DC: The World Bank, pp. 11–30.
190. Sims, C. (1972). “Money, Income and Causality,” American Economic Review, 62,
pp. 540–553.
191. Sims, C. (1980). “Macroeconomics and Reality,” Econometrica, 48, pp. 1–48.
192. Singleton, K. (1988). “Econometric Issues in the Analysis of Equilibrium Business Cycle
Models,” Journal of Monetary Economics, 21, pp. 361–387.
193. Slutsky, E. (1937). “The Summationof RandomCauses as the Source of Cyclic Processes,”
Econometrica, 5, pp. 105–146.
194. Smets, F. and R. Wouters (2003). “An Estimated Dynamic Stochastic General
Equilibrium Model of the Euro Area,” Journal of the European Economic Association, 1,
pp. 1123–1175.
195. Smets, F. and R. Wouters (2005). “Comparing Shocks and Frictions in US and Euro
Area Business Cycles: A Bayesian DSGE Approach,” Journal of Applied Econometrics, 20,
pp. 161–183.
196. Solow, R. (1957). “Technical Change and the Aggregate Production Function,” Review
of Economics and Statistics, 39, pp. 312–320.
197. Stock, J. and M. Watson (1999). “Business Cycle Fluctuations in U.S. Macroeconomic
Time Series,” In J. Taylor and M. Woodford (eds.), Handbook of Macroeconomics, Vol. 1A,
North Holland: Elsevier Science, pp. 3–64.
198. Stock, J. and M. Watson (2003). “Has the Business Cycle Changed and Why?” In
M. Gertler and K. Rogoff (eds.), NBER Macroeconomics Annual 2002, Cambridge, MA:
MIT Press, pp. 159–230.
199. Stock, J. and M. Watson (2005). “Understanding Changes in International Business Cycle
Dynamics,” Journal of the European Economic Association, 5, pp. 968–1006.
200. Stockman, A. and L. Tesar (1995). “Tastes and Technology in a Two-Country Model of
the Business Cycle: Explaining International Comovements,” American Economic Review,
85, pp. 168–185.
October 9, 2009 15:13 9in x 6in b808-bib
Bibliography 149
201. Stokey, N. and R. Lucas, with E. Prescott (1989). Recursive Methods in Economic Dynamics.
Cambridge, MA: Harvard University Press.
202. Summers, L. (1986). “Some Skeptical Observations on Real Business Cycle Theory,”
Federal Reserve Bank of Minneapolis Quarterly Review, 10, pp. 23–27.
203. Tesar, L. and I. Werner (1998). “The Internationalization of Securities Markets Since the
1987 Crash,” In R. Litan and A. Santomero (eds.), Brookings-Wharton Papers on Financial
Services, Vol. 1, Washington, DC: The Brookings Institution, pp. 281–372.
204. The Royal Swedish Academy of Sciences (2004). Finn Kydland and Edward Prescott’s
Contribution to Dynamic Macroeconomics: The Time Consistency of Economic Policy and the
Driving Forces Behind Business Cycles. Advance Report on the Bank of Sweden Prize in
Economic Sciences in Memory of Alfred Nobel.
205. Thomas, J. (2002). “Is Lumpy Investment Relevant for the Business Cycle?” Journal of
Political Economy, 110, pp. 508–534.
206. Uhlig, H. (1997). “AToolkit for Analyzing Nonlinear Dynamic Stochastic Models Easily,”
Available at http://www2.wiwi.hu-berlin.de/institute/wpol/html/toolkit.htm/.
207. Valderrama, D. (2007). “Statistical Nonlinearities in the Business Cycle: A Challenge
for the Canonical RBC Model,” Journal of Economic Dynamics and Control, 31(9),
pp. 2957–2983.
208. Veracierto, M. (2002). “Plant-Level Irreversible Investment and Equilibrium Business
Cycles,” American Economic Review, 92, pp. 181–197.
209. Watson, M. (1993). “Measures of Fit for Calibrated Models,” Journal of Political Economy,
101, pp. 1011–1041.
210. Weitzman, M. (2007). “Subjective Expectations and Asset-Return Puzzles,” American
Economic Review, 97, pp. 1102–1130.
211. Zarnowitz, V. (1992). Business Cycles: Theory, History, Indicators, and Forecasting. Chicago:
The University of Chicago Press.
October 9, 2009 15:13 9in x 6in b808-bib
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:13 9in x 6in b808-Index
Index
accommodative monetary policy, 58
aggregate price index, 94
animal spirits, 33
Bellman’s equation, 42
business cycles
coincident indicator, 16
in emerging markets, 101
international, 19, 65
lagging indicator, 16
leading indicator, 16
calibration, 43
capital account, 73
common factors, 110
complete contingent claims, 71
conditional heteroscedasticity, 124
credit market frictions, 34
cross-country correlations, 66
current account, 73
demand-constrained equilibrium, 138
deterministic steady state, 39
dynamic factor analysis, 110
for large cross-sections, 116
dynamic stochastic general equilibrium
(DSGE) models, 129
identifiability of, 135
dynamics of output, 120
enforcement constraints, 83
generalized method of moments (GMM),
109, 117
goodness-of-fit measures, 113
home production, 54
home-produced good, 54
impulse response functions, 45
indirect inference, 130
indivisible labor, 49
inside money, 58
international portfolio diversification, 80
international risk sharing, 68
investment-specific technological change, 59
irreversible investment, 125
Kalman filter, 131
kurtosis, 124
151
October 9, 2009 15:13 9in x 6in b808-Index
152 Business Cycles: Fact, Fallacy and Fantasy
labor hoarding, 55
limited risk sharing, 81
lottery models, 49
market good, 54
market incompleteness, 82
maximum likelihood estimation
frequency domain, 112
measurement equation, 132
monopolistic competition, 94
multi-sector models, 137
multiple equilibria, 137
New Keynesian model, 33
nonconvexities in labor supply, 49
nonlinear time series models, 123
nonnormal disturbances, 124
nonspecialization in endowments, 77
nontraded good, 78
optimal linear regulator problem, 42
posterior distribution, 132
predictive density, 121
price anomaly, 66
price rigidities, 94
productivity spillovers, 67
puzzles, 48
quantity anomaly, 66
real balances, 97
real business cycle (RBC) theory
numerical solution, 33
reverse causality, 56
Riccati equation, 43
search models, 137
self-fulfilling expectations, 137
seminonparametric (SNP) model, 124
shocks, 33
country-specific, 78
energy, 63
government consumption, 53
industry-specific, 78
monetary, 98
technology, 33
skewness, 124
social planner’s problem, 50
Solow residual, 38
spectral density function, 111
stabilization policy, 2
state space representation, 132
structural models, 135
Sudden Stops, 101
symmetric equilibrium, 97
total factor productivity (TFP), 38
trade balance, 101
transactions services from money and
banking, 56
transition equation, 131
unconditional moments, 44
unobservable index models, 110
variable capacity utilization, 88
vector autoregression (VAR), 58
structural, 99

iness Bus les Cyc
lla Fact, Fa
cy

and

Fa

ta
n

sy

This page intentionally left blank

F World Scientific NEW JERSEY • an F LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TA I P E I ta • sy CHENNAI . Turkey & Centre for Economic Policy Research. UK ess sin Bu les Cyc and y allac Fact.Sumru G Altug Koç University.

. Hackensack. All rights reserved. Sumru. Ltd.com Printed in Singapore.Published by World Scientific Publishing Co. NJ 07601 UK office: 57 Shelton Street. or parts thereof.A424 2009 338. cm. I. This book. fallacy. Singapore 596224 USA office: 27 Warren Street. 222 Rosewood Drive. Business cycles. please pay a copying fee through the Copyright Clearance Center. Ltd. USA. p. For photocopying of material in this volume. including photocopying. recording or any information storage and retrieval system now known or to be invented. may not be reproduced in any form or by any means. In this case permission to photocopy is not required from the publisher. Copyright © 2010 by World Scientific Publishing Co. Typeset by Stallion Press Email: enquiries@stallionpress. HB3711. Includes bibliographical references and index. and fantasy / by Sumru G. 5 Toh Tuck Link. Pte. Altug. Covent Garden. electronic or mechanical. Suite 401-402. without written permission from the Publisher. Inc. MA 01923. . Business cycles : fact. ISBN-13: 978-981-283-276-4 (hardcover) ISBN-10: 981-283-276-9 (hardcover) 1. Title.5'42--dc22 2009034620 British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. London WC2H 9HE Library of Congress Cataloging-in-Publication Data Altug. Pte. Danvers.

the development of new techniques or new ways of modeling can also affect the course that a discipline takes. My thesis was on estimating the model in the same article. the discipline of economics — and macroeconomics. and The Royal Swedish Academy of Sciences published a report titled Finn Kydland and Edward Prescott’s Contribution to Dynamic Macroeconomics: The Time Consistency of Economic Policy and the Driving Forces Behind Business Cycles [204]. In 2004. The model was rejected. The field of macroeconomics has changed significantly due to Kydland and Prescott’s contributions. real business cycle (RBC) analysis has come to provide a flexible and popular approach for examining macroeconomic phenomena. student at the Graduate School of Industrial Administration at Carnegie Mellon University when Kydland and Prescott’s “Time-to-Build and Aggregate Fluctuations” article was published in the early 1980s [141]. Sometimes. The Great Depression greatly influenced the perceptions of a generation of economists. This debate continues to this day. The oil shocks of the 1970s and the 1980s affected economists’ views regarding the sources of macroeconomic fluctuations.Preface How do intellectual disciplines progress? Undoubtedly. beginning with Keynes. This book draws upon Kydland and Prescott’s original contribution. Large-scale computers in the post-World War II era played an important role in the development of simultaneous equation models. Finn Kydland and Edward Prescott received the Nobel Prize in Economics.D. in particular — is affected by major changes in economic conditions. much to the delight of macroeconomists of a more Keynesian bent! Yet many felt that economic models should be subject to formal econometric and statistical testing. In recent years. I was a Ph. v .

. There have been a number of excellent publications that have examined different facets of this literature. researchers at central banks have begun using so-called dynamic stochastic general equilibrium (DSGE) models for policy analysis. Fallacy and Fantasy Kydland and Prescott’s seminal article initiated the school of RBC analysis. Kevin Hoover. many researchers also pursued the econometric analysis of RBC models. James Hartley. This literature evolved in different ways. reflects my own interests. in many ways. This book attempts to provide an overview of the burgeoning business cycle literature that. The volume by Thomas Cooley [74] can be considered a primer of RBC analysis and its applications. This book takes a more eclectic approach.vi Business Cycles: Fact. Jordi Gali’s [97] recent text articulates an alternative New Keynesian framework for describing aggregate fluctuations. asking some basic questions about RBC analysis and summarizing the ongoing controversies surrounding it. Not content with the initial rejection of the model. Talented and creative individuals extended the initial Kydland–Prescott research in different ways. and Kevin Salyer’s [116] collection of articles provides a critique of the calibration approach. In recent years.

. .3. . Pareto Optimal Allocations . Introduction Facts 2. Models of Business Cycles 3. . . The Euro Area Business Cycle . . . . . . . . . . . . . .Contents Preface 1. . 3. . . . .1. 4. . . .2. . . . Investment-Specific Technological Shocks 3.2. . . . . . .2. . A Model with Indivisible Labor Supply . . .4. . . . . . . . . . . 3. . . . . . . . . . . . . . . . . . . 7 15 19 25 28 33 34 39 46 48 49 52 56 58 59 63 65 . .3. . . . . . 3. . . . . Defining a Business Cycle . . . . . . . International Business Cycles 4. The Productivity Puzzle . . Facts . 2. 3. “Puzzles” . . . An RBC Model . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . 2. . . . Historical Business Cycles . . . Stylized Facts . . . . . .1. . . 65 68 69 71 . . .5. . . .2. . . .1. . .5. . . . The Role of International Risk Sharing 4. . . . . . . . 3. .2. . .2. The Source of the Shocks .4. . . . . . . . . . .4. . . . . . .4. . .1. . . . . . . . . . Complete Contingent Claims vii v 1 7 . . 3. . . . . . . . . .4. . . . . . . 3.3. . . . 3. . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . Reverse Causality . . .5. . . . . 4. . . . . . . . . . . . . . . .2.1. . . . . . . . 3. . Initial Criticisms . . . . . . . . .2. . . . .5. . . 2. . A Numerical Solution . . . . Is There a World Business Cycle? 2. . . . . . Energy Shocks .

Nonspecialization in Endowments 4. . Business Cycles in Emerging Market Economies 6. . 7. . . . . . . . . . 7. . 4. . . . . . . .1. . . .7. . . . . . . . .1. . . . . 4. . Future Areas for Research Bibliography Index . . . The Basic Model . . Do Shocks to Trend Productivity Explain Business Cycles in Emerging Market Economies? . . Nontraded Goods . . 7. . Puzzles Revisited . . . . . . . . . The Debate Reconsidered . . . 5. . . . 7.3. 4. . . . . . . . .1. . .2. . . . . . . . . Calibration as Estimation . . . . Limited Risk Sharing . . . . . . . . . . A Small Open-Economy Model of Emerging Market Business Cycles . . . . . . . 8. . . . . 4. . . . .3. . . . . . . . . . . 6. 7. . . . . . No Asset Trading . . . . . . . . . .2. . .2. .4. . . .3. . . .2. 7. . . . . Fallacy and Fantasy 4. Other Applications . . .5. . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . .3. . . . . 7. .1. . . . The Calibration versus Estimation Debate . . . . . . . . 4. . . . . . . . 6. . Trade in Equity Shares . . . .2. . . Empirical Evidence . . . . .4. . . . Dynamic Factor Analysis . . . . . . . .2. GMM Estimation Approaches . Matching the Model to the Data 7. . . . . . . . . . . .2.4. . . .viii Business Cycles: Fact.6. . .4. . . .2. The Dynamics of Output . . .3. . . . . . . . 5. . .1. . . . .3. . . . DSGE Modeling . 75 77 78 79 81 87 90 93 94 99 101 102 106 109 110 113 116 117 119 120 121 123 127 129 137 139 151 New Keynesian Models 5. . . . . .3.3. . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . 7. .2. .1. Other Extensions . . . . . .1. Nonlinearity in Macroeconomic Time Series 7. . . . . . . . . Measures of Fit for Calibrated Models .

Wesley Mitchell [163] comments as thus: “As knowledge of business cycles grows. According to Karl Marx. more effort is required to master it. chemical processing. As Mitchell [163] recounts. other economists observed that the alternating phases of prosperity and depression seemed to follow each other on a regular basis. In his framework. Schumpeter [187. corresponding to the Industrial Revolution. when one examined the history of commercial cycles for the capitalist economies of the time. 188] sought to explain the existence of such long waves as an outcome of technological innovations. and 1890–1950. 1840–1890. corresponding to the invention of electricity. there were periodic movements or “long waves” in economic variables. the notion of business cycles originated from various types of panics. Historically. However. and crises experienced by market economies in the 19th and early 20th centuries. other scholars investigated methods for the systematic 1 . This book describes these new developments. one of the most prominent thinkers of the time. “crises” are an endemic feature of capitalist economies. there have been many developments in the field of business cycles. depressions. much effort was devoted to understanding the causes of what many viewed as “abnormal” phenomena. and motor engines.” Ever since.Chapter 1 Introduction In the opening page of the book Business Cycles published in 1927. While some scholars proposed different theories to explain fluctuations in economic activity. Kondratiev [137] argued that in addition to shorter economic cycles. both growth and business cycles could be ascribed to the process of innovation. In the 1920s. corresponding to the introduction of steel and steam engines. He identified three long waves: 1780–1840.

a business cycle is the simultaneous downturns and upturns of a large number of economic series. 149] developed monetary models of the business cycle as a way 1 See also Zarnowitz [211]. the focus shifted to stabilization policy. the Keynesian framework was interpreted as a model of output determination at a point in time. The Great Depression and World War II were two major events in the development of business cycle analysis. Their work involved the dating of business cycles and the development of leading indicators for the US economy. The Norwegian economist Ragnar Frisch [95] developed the notions of impulse and propagation mechanisms for describing business cycles. Their ideas were formulated in terms of linear time series models. which continue to form the main vehicle for empirically studying business cycles. Post World War II. The oil shocks of the 1970s and the experience of high inflation and high unemployment. and it continues to this day in the business cycle dating methodology employed by the NBER. led researchers to account for the observations using new mechanisms for the effects of money on output.1 We discuss the NBER methodology and the stylized facts of business cycles in Chapter 2. and modeled business cycles as the response of a second-order dynamic system to random shocks. Fallacy and Fantasy measurement and identification of business cycles. During this time. Another important channel which affected the study of business cycles was the development of statistical and time series methods. Keynes’ General Theory [127] laid the foundations for the analysis of short-run economic fluctuations. Nevertheless. . According to them. Their influence on thinking about business cycles has persisted to this day. After the early work of Burns and Mitchell. the lessons of the Great Depression were vivid in the minds of many policy-makers. Phelps [170] and Lucas [148.2 Business Cycles: Fact. In their seminal contributions. The period following World War II was an era of high and sustained growth in many countries. Slutsky [193] argued that the sum of a number of uncorrelated shocks could produce serially correlated or smooth movements in the generated series. or stagflation as it is popularly known. post World War II. Burns and Mitchell [50] and researchers at the National Bureau of Economic Research (NBER) began to identify the phenomena of business cycles. Writers such as Frisch and Slutsky embedded the notion of business cycles into simple dynamic systems driven by stochastic shocks.

They also proposed a methodology for confronting their theory with the data. who presented a model that featured technology shocks as the main impulse behind cyclical fluctuations and proposed a rich array of propagation mechanisms for these shocks. Widely known as the calibration approach. In Phelps’ and Lucas’ framework. Long and Plosser [147] developed a simple Robinson Crusoe economy and generated many of the characteristics of modern macroeconomic time series through the mechanisms of substitution and wealth effects in response to technology shocks affecting different sectors of the economy. the merits of this approach have been a topic of debate. We will discuss RBC models further in Chapter 3. and inventories. the specific mechanisms postulated in this literature as leading to business cycles. unanticipated shocks to money. was identifying technology shocks that could generate cyclical fluctuations of magnitudes observed in the data (see Summers [202]). the debate on the empirical validation of business cycle models. Despite providing great theoretical advances in the analysis of aggregative phenomena. Lucas and Rapping [153] argued that observed fluctuations in aggregate labor supply could be modeled as the voluntary response of agents based on intertemporal substitution effects. the emphasis was on generating the Phillips-curve type of phenomena between inflation and unemployment based on informational frictions. we cannot ignore the calibration approach or. this involves matching a small set of moments implied by the model with those in the data. Since this book purports to discuss business cycles. namely. failed to garner sufficient empirical support. During this period. In some sense. more generally. as identified by skeptics. One of the major issues with Kydland and Prescott’s contribution. time-tobuild in investment. namely. this feature of Kydland and Prescott’s analysis . In his article “Understanding Business Cycles”. there was a revival of interest more generally in examining aggregate economic activity as recurrent phenomena characterizing the functioning of economies with optimizing agents.Introduction 3 of providing a consistent theoretical foundation for describing the impact of changes in money on output. Lucas [151] cataloged the remarkable conformity in a set of economic series and set forth an agenda for explaining these facts using an equilibrium approach. The literature on real business cycle (RBC) theory owes its existence to Kydland and Prescott [141]. Though this approach is cited extensively. the durability of leisure.

a plethora of papers have presented modifications to the original Kydland– Prescott framework to reconcile the model with many of the actual features of the data. endogenous changes in efficiency due to increasing returns to scale. The original RBC framework has also been extended in recent years to examine the business cycle phenomena in emerging market economies. Hall [110. Fallacy and Fantasy was viewed as “fantastical” by many. the original Kydland–Prescott model could not explain the relative variability of hours and productivity or real wages. Many of these issues have taken on the character of “puzzles” in the RBC literature. According to the RBC approach. More recently. and international capital flows. Subsequent research that evolved from the original Kydland–Prescott exercise has proceeded along several different dimensions. for example. For example. it introduces alternative mechanisms for generating price stickiness such as imperfect competition among firms. A more general critique to RBC analysis was mounted by the New Keynesian challenge. The New Keynesian viewpoint breaks with the RBC approach by contesting the view that prices adjust frictionlessly to clear markets. models have been developed that study the role of market completeness/incompleteness on cyclical fluctuations (see. or the Solow residual. We will examine a few of these directions in later chapters. Productivity. It also failed to account for the correlation between hours and productivity. A deeper problem lay in modeling the movement of economy-wide averages (possibly fallaciously) in terms of the behavior of a representative or stand-in household. and they have constituted the topic for much further study. The RBC literature has also generated international business cycle models to replicate findings on current account dynamics. the observed procyclical movements in productivity should merely be a response to exogenous technology shocks (see Prescott [173]). Other skeptics contested the implications of the RBC approach for the observed behavior of productivity. The model lacked money. is known to be procyclical. international risk sharing. Heathcote and Perri [117]). and variable factor . markups. 111] argued persuasively that there were most likely endogenous components to the cyclical movement of productivity arising from imperfect competition at the firm level and internal increasing returns to scale in production. Instead. financial diversification. it faced the problem of reconciling observations on money–output correlations within a model where the main driving force was real productivity shocks. On the one hand. In a series of papers.4 Business Cycles: Fact. hence.

Smets and Wouters [194. The controversy over the calibration approach also resulted in work on alternative methods for empirically analyzing business cycle phenomena. vector autoregression (VAR) and structural VAR (SVAR) have also proven to be popular in empirical macroeconomic research. More recently. SVAR also permits an identification of shocks. or monetary versus technology shocks. dynamic stochastic general equilibrium (DSGE) models have been developed to identify shocks and propagation mechanisms of business cycles models (see. This facilitates the analysis of the different effects of government versus technology shocks. for example. We will discuss New Keynesian models in detail in Chapter 5.2 We will discuss the issues involved in matching the model with the data in Chapter 7. 2 Canova [58] is an excellent reference source on the quantitative and empirical analysis of dynamic stochastic general equilibrium models. 195]). . and Kimball [32]). Following Sims [191]. Fernald. The New Keynesian challenge has proceeded along theoretical lines (see Rotemberg and Woodford [182]) and empirical considerations (see Gali [96] or Basu. One approach that is popular in the business cycle literature is the method of unobservable index models or dynamic factor analysis developed by Sargent and Sims [185] and others. While both models allow for rich dynamic interrelationships among a set of endogenous variables and an examination of business cycle dynamics based on impulse response functions.Introduction 5 utilization.

This page intentionally left blank .

or lagging indicators of the business cycle. Stock and Watson [197] present a modern methodology to describe business cycles in terms of the cyclical time series behavior of the main macroeconomic series and their comovement with cyclical output. 2. In this chapter. This framework is based on the principle of identifying turning points in economic activity and determining which series constitute leading. a cycle consists of expansions occurring at about the same time in many 7 .Chapter 2 Facts The vast literature on business cycles has focused on generating the stylized facts regarding cyclical fluctuations. and Stock and Watson [199].1. and Osborn [20]. Basu and Taylor [31] have examined business cycles in an international historical context. we describe the National Bureau of Economic Research (NBER) methodology for dating business cycles and present the basic facts regarding business cycles. and Burns and Mitchell [50] provide a framework to describe the main features of business cycles. Mitchell and Burns [164]. Mitchell [163]. Kontolemis. Artis and Zhang [18]. Much of the early work on business cycles was implemented for the US economy. DEFINING A BUSINESS CYCLE The notion that market economies are subject to recurring fluctuations in a large set of variables was formalized by Burns and Mitchell [50] in their 1946 work entitled Measuring Business Cycles as follows: Business cycles are a type of fluctuation found in the aggregate economic activity of nations that organize their work mainly in business enterprises. the European or euro area business cycle has also been the topic of much recent study (see Artis. among others). However. We will also discuss some of the empirical findings in these regards. coincident.

or secular trends? Finally. In seeking to identify “recurrent changes”. random fluctuations. In their words (Burns and Mitchell [50]. Once the reference dates are found. 14): . The NBER approach to identifying business cycles as outlined by Mitchell [163]. Burns and Mitchell [50] stress that the notion of a reference cycle should not be equated with an observable construct. and revivals which merge into the expansion phase of the next cycle. find the cyclical peaks and troughs in a given set of economic variables. Ch. followed by similarly general recessions. As part of this analysis. Some of these questions are precisely the ones that we seek to answer in this book. then should one worry about differences in business cycle activity across regions? Should business cycles be considered in an international context? How about the historical nature of business cycles? Have business cycles moderated over time? Likewise. and also lay down rules for excluding irregular movements and other similar changes. the duration. If the answer to the latter question is in the affirmative. Burns and Mitchell [50] themselves noted that this definition raised as many questions as it sought to answer. p. Fallacy and Fantasy economic activities. how broadly should the aggregates that are being considered be defined? The notion that changes in economic activity occur “at about the same time” admit the possibility of economic variables that lead or lag the cycle. when one considers the statement regarding expansions occurring in “many economic activities”. and Burns and Mitchell [50] is comprised of two mutually reinforcing acts: first. Mitchell and Burns [164]. timing.8 Business Cycles: Fact. this sequence of changes is recurrent but not periodic. then an aggregate business cycle or a reference cycle is identified. If one talks about “fluctuations in the aggregate economic activity of nations”. contractions. and second. whether it pertains to the measurement of business cycles or the construction of models of cyclical fluctuations. determine whether these turning points are sufficiently common across the series. they are not divisible into shorter cycles of similar cycles with amplitudes approximating their own. and amplitude of each specific cycle are compared with that of the reference cycle. 2. This definition has formed the basis of modern thinking about business cycles. the cyclical behavior of each series is then examined relative to the reference cycle. how should we deal with seasonal changes. the comments regarding the duration and amplitude of business cycles are based on actual observations of cyclical phenomena. in duration business cycles vary from one year to ten or twelve years.

2. The NBER business cycle methodology identifies a business cycle based on the (absolute) downturn of the level of output. Point A defines a trough for a classical cycle while point B defines a peak. By contrast. like other concepts. One advantage of using growth cycles is that they have expansions and contractions that are approximately of the same duration. For. a trough occurs at point C for a growth cycle while point D defines a peak. . There is an alternative approach which considers the decline in the series measured as a deviation from its long-run trend. but changes in readings taken from many recording instruments of varying reliability. we say that it is Fig.Facts 9 When we speak of ‘observing’ business cycles we use figurative language.1 displays the difference between classical and growth cycles.1. This is known as a classical business cycle. classical cycles typically have recessions that are shorter than expansions because of the growth effect. business cycles can be seen only ‘in the mind’s eye’. Figure 2. Classical and Growth Cycles. By contrast. What we literally observe is not a congeries of economic activities rising and falling in unison. such cycles are known as growth cycles. When the economy is moving from a trough to a peak. Following the terminology in Zarnowitz [211].

One approach is to use a linear de-trending procedure which assumes that the underlying series possesses deterministic time trends. As King. national income. The turning points are determined judgmentally. and Watson [133] argue. This period was dubbed as the period of the “Great Moderation”. The amplitude of a business cycle is the deviation from trend. This committee uses data on real output. equivalently. employment. in their survey of empirical business cycles. and the longest cycle is 128 months or more than ten years. However. Similarly. real business cycle (RBC) models which allow for trends in the technology shock imply that growth and business cycles are jointly determined. the practice of separating the trend and cyclical component using linear time series methods is well established. An alternative approach is to assume a stochastic trend modeled as a unit root in the series at hand. Stock. The shortest cycle is 17 months or nearly six quarters. Post WWII. 38 months pre-WWII. The dating of business cycles for the US is done formally by the NBER Business Cycle Dating Committee. the average length of an expansion has increased to 57 months from. We can also observe expansionary and contractionary phases of the business cycle from this table. Plosser. the average length of a contraction has decreased to ten months from 17 months or more in the years preceding 1945. quarters. Nevertheless. Table 2. As an example. Fallacy and Fantasy in an expansion. The duration of the business cycle is the length of time (in months. at most. If one chooses to use growth cycles. One of the longest expansions to have occurred post-WWII is between March 1991 and November 2001 — a period of 120 months. There are several approaches to de-trending economic time series. although a computer algorithm exists that can approximate the results (see Bry and Boschan [49]). or years) that the economy spends between two troughs or. two peaks.10 Business Cycles: Fact. The contribution of Nelson and Plosser [167] was to show that economic time series such as real GDP typically possess unit roots.1 gives the dates of business cycles or the so-called “reference dates” for the US economy since 1857. The average length of a cycle (peak from previous peak or trough from previous trough) in the post-WWII era is 67 months or over five years. this committee recently announced that the US economy had formally been in a recession since December 2007. there is an issue of how to identify the cyclical component of a given series. and trade at the sectoral and aggregate levels to identify and date business cycles. and a recession is said to occur when the economy is moving from a peak to a trough. Stock and . There are 32 complete cycles in the entire sample period.

Duration in Months Contraction Peak to Trough Expansion Previous Trough to This Peak — 30 22 46 18 34 36 22 27 20 18 24 21 33 19 12 44 10 22 27 21 Cycle Trough from Previous Trough — 48 30 78 36 99 74 35 37 37 36 42 44 46 43 5 51 28 36 40 64 Peak from Previous Peak — — 40 54 50 52 101 60 40 30 35 42 39 56 32 36 67 17 40 41 34 (Continued) June 1857(II) October 1860(III) April 1865(I) June 1869(II) October 1873(III) March 1882(I) March 1887(II) July 1890(III) January 1893(I) December 1895(IV) June 1899(III) September 1902(IV) May 1907(II) January 1910(I) January 1913(I) August 1918(III) January 1920(I) May 1923(II) October 1926(III) August 1929(III) December 1854 (IV) December 1858 (IV) June 1861 (III) December 1867 (I) December 1870 (IV) March 1879 (I) May 1885 (II) April 1888 (I) May 1891 (II) June 1894 (II) June 1897 (II) December 1900 (IV) August 1904 (III) June 1908 (II) January 1912 (IV) December 1914 (IV) March 1919 (I) July 1921 (III) July 1924 (III) November 1927 (IV) March 1933 (I) — 18 8 32 18 65 38 13 10 17 18 38 23 13 24 23 7 18 14 13 43 Facts 11 .Table 2. Business Cycle Reference Dates Peak Trough US Business Cycle Expansions and Contractions.1.

1. ∗∗ 15 cycles. (Continued) Business Cycle Reference Dates Peak Trough Contraction Peak to Trough May 1937(II) February 1945(I) November 1948(IV) July 1953(II) August 1957(III) April 1960(II) December 1969(IV) November 1973(IV) January 1980(I) July 1981(III) July 1990(III) March 2001(I) December 2007 (IV) Average. all cycles: 1854–2001 (32 cycles) 1854–1919 (16 cycles) 1919–1945 (6 cycles) 1945–2001 (10 cycles) ∗ 13 cycles. Source: NBER. Fallacy and Fantasy 17 22 18 10 55 48 53 67 .12 Table 2. Duration in Months Expansion Previous Trough to This Peak 50 80 37 45 39 24 106 36 58 12 92 120 73 38 27 35 57 Cycle Trough from Previous Trough 63 88 48 55 47 34 117 52 64 20 100 128 Peak from Previous Peak 93 93 45 56 49 32 116 47 74 18 108 128 81 56∗ 49∗∗ 53 67 June 1938 (II) October 1945 (IV) October 1949 (IV) May 1954 (II) April 1958 (II) February 1961 (I) November 1970 (IV) March 1975 (I) July 1980 (III) November 1982 (IV) March 1991(I) November 2001 (IV) 13 8 11 10 8 10 11 16 6 16 8 8 Business Cycles: Fact.

if the underlying data are . The first approach tends to generate spurious business cycle effects in the de-trended series.1) Using the lag operator notation and simplifying. The parameter µ controls the smoothness of the trend component. If these are trend-stationary. µ is typically recorded as 1600.1). The Hodrick–Prescott filter defines gt to solve ∞ ∞ min gt t=−∞ (yt − gt ) + µ t=−∞ 2 (gt+1 − gt ) − (gt − gt−1 ) . King and Rebelo [130]. in particular. 1 + µ(1 − L−1 )2 (1 − L)2 (1. Several alternative approaches have been proposed in the recent business cycle literature to separate the trend versus cyclical component of a time series. The general equation characterizing the filter. For quarterly data. 2 The cyclical component of the series is defined as ytc = yt − gt . then the de-trending procedure has favorable properties. Cogley and Nason have argued. that business cycle dynamics obtained by using a HP de-trending procedure depend on the properties of the underlying data. which involves minimizing a quadratic form to determine the trend component in a given series. One of these approaches is the so-called Hodrick–Prescott [120] (HP) filter. (1. the cyclical component can be represented as ytc = yt − gt = µ(1 − L−1 )2 (1 − L)2 yt . let yt denote the series in question and gt denote the unknown trend component. Specifically. The properties of the HP filter have been studied by many authors. including Singleton [192]. Eq.Facts 13 Watson [197] argue that neither linear time trends nor first-differencing to eliminate unit roots provides a satisfactory approach to identifying the cyclical component of a series. is modified at the beginning and end of the sample. whereas the second exacerbates the role of short-term noise. and Cogley and Nason [69]. (1.2) In practice. Taking the derivative with respect to gt yields ∞ ∞ µgt+2 − 4µgt+1 + (1 + 6µ)gt − 4µgt−1 + µgt−2 = t=−∞ t=−∞ yt . the filter is applied using a finite sample.

once the ideal filter’s weights are found in the time domain.4) where ψ(L) is a (K − 1)-order symmetric moving average polynomial. Baxter and King derive the band-pass filter by considering low-pass and high-pass filters with the required properties. the Baxter–King filter will eliminate deterministic quadratic trends or render stationary series that are integrated up to order two. The band-pass filter retains the movements in a series for the frequencies associated with the periodicities of 0. The band-pass filter of Baxter and King [37] is obtained by applying a K th-order moving average to a given time series: yt∗ K = k=K ak yt−k . however. The filter is designed to have a number of other properties. Fallacy and Fantasy difference-stationary.3) where the moving average coefficients are chosen to be symmetric. including the property that the results should not depend on the sample size and that it does not alter the timing relations between series at any frequency. the optimal . periodicities between six quarters and eight years. It turns out that the time-domain representation of the band-pass filter is an infinite moving average. In particular. where I (·) is the indicator function. . (1. . ak = a−k for k = 1. (1. Let s denote the number of observations in a year. . Hence. I (2) or less. The band-pass filter developed by Baxter and King [37] “filters” out both the long-run trend and the high-frequency movements in a given time series while retaining those components associated with periodicities of typical business cycle durations. K .5s and 8s. A second approach is based on the spectral analysis of economic time series. namely. i. However.14 Business Cycles: Fact. application of the HP filter induces spurious business cycle fluctuations. then it has trend elimination properties. . k=K they show that the lag polynomial describing the K th-order moving average can be written as a(L) = (1 − L)(1 − L−1 )ψ(L).5s)). The frequency response function of the ideal band-pass filter is defined as βbp (ω) = I (2π/(8s) ≤ ω ≤ 2π/(0. K ak = 0. They show that if the sum of the moving average coefficients is zero.e.

Nevertheless. We also observe the cyclical troughs and peaks of economic activity associated with the business cycle.01 8. We can observe the cyclical downturns and upturns corresponding to the NBER business cycle reference dates from this graph.03 15 50 55 60 65 70 75 80 85 90 95 00 05 BPCYCLEUSA Fig. approximating filter with maximum lag K is obtained by truncating the ideal filter’s weights at lag K . We consider a measure of the business cycle as the co-movement of the cyclical behavior of individual series with . These facts constitute important benchmarks by which to judge the performance of alternative models of business cycles. 2.2 discusses these properties of business cycles.6 -0.2 50 55 60 65 70 75 80 85 90 95 00 05 BPTEST USA -0.00 0.0 7. STYLIZED FACTS In this section.8 8.02 -0.2 illustrates the logarithm of real GDP for the US and its trend and components estimated according to the Baxter–King filter for the period 1947(I)–2005(III). Figure 2.04 7. In what follows. we use this approach to generate the cyclical components of the different series.2 0.01 0.6 9.2.04 0. Section 2.4 -0.05 -0. 2.03 0.02 8. the notion of a “business cycle” is also concerned with the co-movement of a large number of economic variables. showing the large gains in productivity and growth attained over this period.2.Facts 9. We observe that real GDP displays a marked positive trend over the sample period. The cyclical component tracks very well the US business cycle as identified by the NBER. we provide some stylized facts of business cycles. Trend and Cyclical Components in USA GDP. Baxter and King [37] employ a finite-order approximation to obtain the coefficients of this filter with a truncation of K = 3 years or K = 12 quarters.

and the cross-correlations of the cyclical component of GDP with the cyclical component of each series for lags and leads up to six periods.1 Information on leading indicators is also collected by the Conference Board. Exceptions are production of agricultural goods and natural resources. Among other measures. hours. investment. a leading indicator reaches a peak before real GDP reaches its peak and bottoms out (reaches a trough) before real GDP. . and separate their sample period into the pre-World War I era. These have been described in a number of economic studies. and passed over to the Conference Board in 1995. 2 Early work on developing a composite index of leading indicators is due to Mitchell and Burns [164]. asset returns and prices. We can also examine whether different time series are out of phase with real GDP. The stylized facts of business cycles are typically defined in terms of the behavior of the main components of GDP. lagging indicators reach a peak or trough after real GDP. Leading indicators are useful for predicting subsequent changes in real GDP. They use the HP filter to derive the cyclical components of the different series. real wages. and monetary aggregates. They make use of the Baxter–King band-pass filter to identify the trend versus cyclical components of each series. variables that move in the same direction over the cycle as real output are procyclical. 2.16 Business Cycles: Fact. they consider the co-movement of the cyclical component of GDP with the cyclical component of each series. Backus and Kehoe [23] analyze the properties of historical business cycles for ten developed countries using a century-long dataset up to the 1980s. and the post-World War II era. Consumption of 1 Stock and Watson [197] use data on 71 variables to characterize US business cycle phenomena over the period 1953–1996. the interwar era between World War I and II.2 The salient facts of a business cycle can be described as follows: 1. Such a composite index was made official by the US Department of Commerce in 1968. Consumption. In other words. Fallacy and Fantasy the cyclical component of real output. Variables that display little correlation with output over the cycle are called acyclical. and imports are all strongly procyclical. Consumption of durables is much more volatile than consumption of nondurable goods and services. which are not especially procyclical. inventories. Coincident indicators reach a peak or a trough at roughly the same time as real GDP. According to this approach. Real output across virtually all sectors of the economy moves together. productivity. For example. the contemporaneous correlation of output in different sectors of the economy is large and positive. Variables that move in the opposite direction (rise during recessions and fall in expansions) are countercyclical. Finally.

an inverted yield curve. an expansion is characterized by expectations of higher interest rates at longer horizons whereas a recession typically signals a decline in long-term interest rates relative to short-term rates. 6. Productivity is slightly procyclical. Investment in equipment and nonresidential structures is procyclical with a lag. Employment fluctuates almost as much as output and total hours of work. 8. while average weekly hours fluctuate much less. and capacity utilization are all strongly procyclical. 13. while capacity utilization tends to be coincident. 4. or lagging indicators. Investment in residential structures is procyclical and highly volatile. Government spending tends to be acyclical. They have not displayed a steady pattern in terms of variability to GDP or in terms of leading. Profits are highly volatile. or the yield spread between corporate paper and Treasury bills with six months’ maturity. 7. employee hours. 12. Velocity and the money supply are procyclical.Facts 17 3. Nominal interest rates tend to be procyclical. durable goods fluctuates more than GDP. but both real wages and productivity vary considerably less than output. Net exports are countercyclical. 5. namely. the trade balance tends to be countercyclical. The implication is that most fluctuations in total hours result from movements in and out of the work force rather than adjustments in average hours of work. The employment series lags the business cycle by a quarter. Since imports are more strongly procyclical than exports. The correlation with output is generally negative. coincident. The yield curve which shows the rates of return on bonds of different maturities tends to be upwardsloping during an expansion and downward-sloping at the onset of a recession. 11. 9. . The risk premium for holding private debt. That is. but weakly so. The reason for this countercyclical behavior is likely to be changes in default risk. tends to shrink during expansions and increase during recessions. Real wages are procyclical or acyclical. whereas nondurables fluctuate considerably less. Total employment. The correlation between government expenditures and output is nearly zero. 10.

contemporaneous correlations between output fluctuations in different countries were highest in the interwar period. The behavior of prices and inflation appears to have changed over time. inflation was procyclical with a very low mean. The stock market is positively related to the subsequent growth rate of real GDP. a topic to which we will return. In the postWWII era. reflecting the common experience of the Great Depression. hence. The correlation is typically larger in the post-war period than in the pre-war period.18 Business Cycles: Fact. inflation appears to be countercyclical. In this sense. and productivity have proved among the most difficult to reconcile by current business cycle theories. In the pre-WWI period and interwar period. Chadha and Nolan [62] identify the stylized facts . Other papers have generated business cycle facts using data over long samples. Between 1945 and 1980. 15. with the exception of Germany and Japan. although it is important to emphasize that the market has fallen without a subsequent recession. The changes in the cyclical behavior of prices and inflation also have implications for the socalled impulses and propagation mechanisms of business cycles. the stock market fell in the quarter before each of the eight recessions. changes in stock prices have been taken as providing information about the future course of the real economy. Finally. Money (M2) is procyclical and tends to be a leading indicator of output. Fallacy and Fantasy 14. prices and inflation were procyclical in the pre-WWII era. 17. However. real wages. 20. One way to rationalize this phenomenon may be in terms of monetary policy that is more accommodative — under a gold standard. there is more evidence of supply-side-driven fluctuations in output. The findings concerning the behavior of hours. the procyclicality of M2 has diminished since the 1980s. for example. Witness the impact of the large oil shocks in the 1970s. 16. employment. the increased persistence of inflation and the countercyclical behavior of prices observed in this era. By contrast. This set of facts constitutes a benchmark which a successful business cycle model should meet. 18. A similar change appears to have characterized the behavior of price-level fluctuations. The standard deviation of inflation is lower than that of real GDP. 19. Since the early 1980s. There is also a marked increase in the persistence of inflation after WWII. Inflation is a coincident indicator.

and Osborn [20] propose business cycle turning points for a number of countries based on industrial production. Backus and Kehoe [23] argue that it is difficult to reach a firm conclusion on this point based on US data alone. Christina Romer [180. and compare the volatility of 21 different variables in the 20 years since the 1980s and the 40 years in the period following WWII up to the 1980s. 15% to smaller shocks to productivity. The above findings can also shed light on whether output fluctuations have moderated in the post-WWII era. While her findings have garnered much interest. 2. then one discerns no change in output fluctuations across the two periods. and another 15% to reduced shocks to food and commodity prices. Artis and Zhang [18. Their findings also provide evidence on the factors behind the “Great Moderation”. Kontolemis.3. They attribute around 20–30% of the reduction in output volatility to better monetary policy. Stock and Watson [198] seek to avoid the problem of poor data quality in the pre-WWII period. On the one hand. They suggest that if one considers additional evidence based on a larger sample of countries over the different periods.and post-WWII data of comparable quality. They find that the standard deviations of a typical variable in their sample declined by about 20–40% since the 1980s. THE EURO AREA BUSINESS CYCLE Much of the early work on the European business cycle was concerned with examining the impact of monetary union arrangements on economic activity. Artis. If the higher-quality data in the post-WWII period are subjected to a transformation that makes the pre. 19] investigate the relationship of the European Exchange Rate Mechanism (ERM) to the international business cycle in terms of the linkage and synchronization of cyclical fluctuations between countries. Their findings suggest the emergence of a group-specific European business cycle since the formation of the ERM. the variability of output appears to have diminished in the post-WWII period. A’Hearna and Woitek [2] establish the facts of business cycles in the late 19th century using spectral techniques. which is independent of the US cycle. The countries selected are . 181] has argued that the finding of lower output variability in the postWWII period is due to changes in the measurement of output in the preand post-WWII eras.Facts 19 of business cycles for the UK economy over a long sample period.

. they distinguish between classical and growth cycles. conditional on being in a contraction. Fallacy and Fantasy the G7 countries along with the prominent European countries. Artis. They use this information to examine the international nature of cyclical movements. The euro area experience makes consideration of both types of business cycles relevant. suppose that the economy can be in one of two mutually exclusive states. and a trough terminates a recession. The Markov process distinguishes between turning points within the two states by assuming that Et = Rt = ECt expansion continuation . making growth cycles the relevant concept. They also consider the lead/lag relationships between countries at peaks and troughs. define pRT = Pr(Tt+1 |RCt ) as the probability of transiting to a trough. Then. Pt peak RCt recession continuation . Tt trough Let pEP = Pr(Pt+1 |ECt ) denote the probability of transiting to a peak. implying that classical cycles should also be considered. and to determine whether cyclical movements are similar across different countries. Whereas in the post-WWII era the euro area countries exhibited high rates of growth. These authors also articulate a formal model of turning points based on restrictions imposed on a Markov process. which implies that the probability of continuing an expansion pEE = Pr(ECt+1 |ECt ) is given by pEE = 1 − pEP . expansion Et or recession Rt . in recent years growth has slowed and even absolute declines in real GDP have been observed. the probability of continuing a contraction pRR = Pr(RCt+1 |RCt ) is pRR = 1 − pRT .3 To describe this algorithm.20 Business Cycles: Fact. Likewise. conditional on being in an expansion. Let St denote a discrete random variable that follows a first-order Markov process 3This algorithm follows Harding and Pagan [115]. Marcellino. As in the Stock and Watson [197] approach. and Proietti [21] discuss alternative approaches to dating euro area business cycles. who extended the Bry and Boschan [49] algorithm to a quarterly setting. Suppose that a peak terminates an expansion.

St−1 . Pt+1 . can only be followed by a peak at t + 1. St ). Hence. expansion or contraction. conditional on having been in an expansionary phase. the expansion continues. Thus. is pRT . . Likewise. (St−4 . conditional on having been in a contractionary phase. St = RCt . as follows: ETSt = {( yt+1 < 0) ∪ ( RTSt = {( yt+1 > 0) ∪ ( 2 2 yt+2 < 0)} yt+2 > 0)}. Let yt denote the underlying series. respectively. A similar condition exists for troughs. (St−4 . St−3 . yt+1 = yt+1 − yt < 0 and 2 yt+2 = yt+2 − yt < 0 represent the candidate sequence for terminating an expansion. ETSt . a peak at date t − 4. Now consider applying this algorithm to dating classical cycles. which defines a peak. Tt ) belong to a contraction. and on complete cycles. The duration of a phase is required to be two quarters. the probability that the economy will transit to a trough. and that ETSt is true. Then. This is the joint event that the history at time t. and a recession termination sequence. we can define an expansion termination sequence. St−1 . which defines a trough. St ). The minimum duration of a complete cycle is given as five quarters. pEP . St−2 . The algorithm is completed by finding the probability that the economy will transit to a peak. St−2 . This is the joint event that the history at time t. RTSt . yt+1 = yt+1 − yt > 0 and 2 yt+2 = yt+2 − yt > 0 represent the candidate sequence for terminating a contraction. features a contractionary state at time t.Facts 21 with four states and transition probability matrix as follows: ECt+1 ECt Pt RCt Tt pEE Pt+1 pEP RCt+1 Tt+1 0 0 1 0 0 0 0 1 pRR 0 0 pRT 0 0 The dating rules impose minimum durations on a phase. Pt ) both belong to an expansion and (RCt . peak-to-peak or trough-to-trough. features an expansionary state at time t. Likewise. Pt−4 . which is automatically satisfied since the states (ECt . if ETSt is false. St−3 . St = ECt . Else.

Fallacy and Fantasy and that RTSt is true.22 Business Cycles: Fact. The Centre for Economic Policy Research (CEPR) has recently formed a committee to set the dates of the euro area business cycle. They implement their procedures using data on euro area GDP and its components for the European Central Bank’s (ECB) Area-Wide Model for the period 1970–2001. the contraction continues. and Proietti [21]. The turning points determined by the CEPR Business Cycle Dating Committee are also indicated in this table.2. Table 2. We note that these are similar to the classical cycles identified by Artis et al. Peak (1) Classical Cycles∗ 1974 (III) 1980 (I) 1982 (II) 1992 (I) (2) Deviation Cycles∗ 1974 (I) 1976 (IV) 1980 (I) 1990 (II) 1995 (I) 1998 (I) (3) CEPR Dating Committee 1974 (III) 1980 (I) 1992 (I) ∗ Artis. and a trough cannot occur if output is above it.2 shows the business cycle turning points for both classical and deviation or growth cycles obtained using this approach. Otherwise. . The dating of deviation cycles is achieved by modifying this algorithm to account for the fact that a peak cannot occur if output is below trend. Marcellino. Euro Area Business Cycle Turning Points. Finally. Trough 1975 (I) 1981 (I) 1982 (IV) 1993 (I) 1975 (III) 1977 (IV) 1982 (IV) 1993 (II) 1997 (I) 1975 (I) 1982 (IIII) 1993 (III) 4 See Fagan. and Mestre [85].4 Table 2. Its stated mission is to establish the chronology of recessions and expansions of the 11 original euro area member countries for 1970–1998 and of the current euro area as a whole since 1999. [21] also examine the properties of alternative filters such as the Hodrick–Prescott and Baxter–King filters for decomposing a time series into trend and cyclical components. [21]. Henry. if RTSt is false. Artis et al.

03 0. they find no evidence for closer international synchronization over their period of study. Fig. they use a socalled factor-structural vector autoregression (FSVAR). they find evidence on the emergence of two cyclically coherent groups. Germany. Stock and Watson [199] provide a comprehensive analysis of the volatility and persistence of business cycles in G7 countries (defined to include the US. This is a standard structural vector autoregression (VAR) with an unobserved factor structure imposed on the VAR innovations.6 -0. and the US).3. which is specified in terms of the growth rates of quarterly GDP for the G7 countries. 2. First. [20].01 13.01 14. France.8 -0. Let Yt denote the n × 1 vector of .0 0. and Canada) over the period 1960– 2002.02 0.4 0. Trend and Cyclical Components in Euro Area GDP. UK. namely.3 illustrates the logarithm of real GDP for the euro area and its trend and cyclical components estimated according to the Baxter–King filter for the period 1970(I)–2005(III). They find that the volatility of business cycles has moderated in most G7 countries over the past 40 years.2 13. They also provide evidence on the synchronization of international business cycles. 2.Facts 14. the eurozone countries and English-speaking countries (including Canada. This is similar to the findings of Köse. However. do they arise from changes in the magnitudes of the shocks or the nature of the propagation mechanism? Are the sources of the changes domestic or international? To answer these questions.00 13. and Terrones [139] and others.02 -0. Japan. Italy. in sync with Artis et al. the UK.03 1970 23 14. They base their results on various measures of correlation of GDP growth across countries. Prasad. Stock and Watson [199] also seek to provide evidence on the sources of the changes.4 1970 1975 1980 1985 1990 1995 2000 2005 1975 1980 1985 1990 1995 2000 2005 BPTRENDEURO EUROAREA BPCYCLEEURO Fig.

First. . Stock and Watson [199] find evidence for two common shocks. These may arise from the role of international trade. and (iii) the spillover effects of the domestic shocks on the h-step-ahead forecast error for each country. they find that most of the variance of GDP growth can be attributed to common and idiosyncratic domestic shocks. Notice that the common shocks affect the output of multiple countries contemporaneously. the impact of international shocks is estimated to be greatest for countries such as Canada. the elements of t denote the country-specific idiosyncratic shocks. . the role of international sources of fluctuations arising from common shocks or from . σνk ).6) (3. However. The standard VAR model is given by Yt = A(L)Yt−1 + νt . Fallacy and Fantasy real GDP growth for the G7 countries considered in this study. where the vector of reduced-form errors vt has the factor structure ν t = ft + t . Second. . In the first period. . They also provide a variance decomposition for the impact of (i) the common international shocks. They consider two sample periods: 1960–1983 and 1984–2002. In this representation. . France. This provides the identification scheme to help identify the common international shocks. and the least for Italy and Japan.24 Business Cycles: Fact. .5) where ft is a k × 1 vector with k ≤ n. σfk ) and E (νt νt ) = diag (σν1 . reflecting the impact of the ten-year-long deflationary episode for this country. . their relative variance varies by country and by time period. where the elements of ft and t are assumed to be mutually independent. for example. whereas the idiosyncratic shocks affect them with a lag. (3. and Germany. The model allows for spillover effects to occur through the lagged effect of an idiosyncratic shock. The covariance structure of the shocks is given by E (ft ft ) = diag (σf 1 . and the elements of ft denote the common international shocks. (ii) the domestic shocks. domestic shocks explain almost all of the variance for Japan. . In the second period.

(4. they also show that the decline in overall volatility of GDP growth for countries such as the US. 2. . Their overall results indicate that the magnitudes of common international shocks appear to have become smaller in the 1980s and 1990s than they were in the 1960s and 1970s. M the number of time series per country. . and the UK is due to the decline in the variance arising from international shocks. and Whiteman [138] consider the issue of a world business cycle.t + i. Canada. region . Latin America.t . and investment) across all countries and regions as well as across countries and aggregates separately. • the single world factor ( f world ).t country + bi country country fn. . Developing Asia. and the UK. However.Facts 25 spillovers appears to have increased for the US. . and also of the failure of business cycles to become more synchronized despite the great increase in trade flows over this period. and use data on 60-odd countries covering seven regions of the world to determine common factors underlying the cyclical fluctuations in the main macroeconomic aggregates (output. IS THERE A WORLD BUSINESS CYCLE? Köse. or Oceania). There are three types of factors: • N country-specific factors ( fn .t denote the observable variables for i = 1. one per each country). N × M . Finally. Let yi. Germany. France. Developed Asia.t = ai + biworld ft world + bi region region fr. t = 1. . for example. . • R regional factors ( fr Africa. The behavior of each observable variable is related to the factors as follows: yi. and Raynauld [109] or Lumsdaine and Prasad [154].5 Their approach assumes that there are K unobservable factors that are hypothesized to characterize the dynamic interrelationship among a cross-country set of economic time series. Europe. where r stands for North America. see. This is the source of the moderation of individual business cycles. Let N denote the number of countries.7) 5 For recent studies. Head. . and Italy. they also find that shocks have become more persistent in countries such as Canada. . consumption. . Gregory. and T the number of time periods. T . Otrok.4. They argue that many recent studies of international business cycles focus on a subset of countries — not the world.

E (ufk . The innovations k ui.t . (4. fk . Thus. scales are identified by assuming that each σf2 is equal to a constant. .2 fk .26 Business Cycles: Fact. This procedure yields a joint posterior distribution for the unobserved factors and the unknown parameters. . are mean zero. There are k several ways to estimate this model. which themselves may be serially correlated. we cannot identify uniquely the sign or scale of the factors or the factor loadings.8) where E (ui.qk fk .t can be explained by each factor. respectively.t j.t ) = 0 for i = j. and defer a discussion of the estimation techniques to Chapter 7. and to follow autoregressions of orders pi and qk .t−s ) = σi2 for i = j and s = 0.t ) = σf2 .t = φi. . Thus. all the covariation among the observable series yi.t−1 + φfk . the sign of the factor loading for US output is considered positive. (4.t−qk + ufk . and ufk . we report results based on the median of the posterior distribution for the factors. . and show how much of the variation in yi. i.2 i. conditional on the data. One identification device is to assume that the factor loading on one of the factors is positive.t−2 + · · · + φi. . the model assumes that the behavior of M × N time series can be explained by N + R + 1 factors. as i.10) where E (ufk . . The coefficients bi denote the factor loadings. Both the idiosyncratic errors and the factors are allowed to be serially correlated.9) = φfk . M × N . for the regional factor. s.t . (4. Since the common factors are unobserved.1 i. For the world factor. . one approach is to use a Kalman filtering algorithm together with classical statistical techniques to estimate the model’s parameters. where N + R + 1 < MN . Let fk.t .t fk . i = 1.t . an alternative is to use Bayesian estimation techniques.t is due to the common factors. contemporaneously uncorrelated random variables. . k = 1.t ui. Since the factors are unobservable. K . Fallacy and Fantasy j where E ( i. and zero otherwise. In what follows.t . and the country factors are identified by positive factor loadings on the output of each country. Likewise.1 fk . Finally.t−2 + · · · + φfk .t ui.t−pi + ui.t−1 + φi. the sign of the factor loading corresponding to North America is considered positive.t = Then.t−s ) = 0 for all k. .pi i.t ufk .

country factors account for a much larger fraction of consumption growth than output growth. regional. the recession of the 1970s associated with the first oil shock. the world factor is more successful in explaining economic activity in developed relative to developing countries. if anything. and the downturn and recession of the early 1990s. Specifically. and country factors are modeled as autoregressive processes which may display substantial persistence depending on the estimated values of the parameters. Second. we have that Var(yi. as severe as the recession of the mid-1970s. Recall that the world. 9% of consumption growth. Var(yi. Using the representation of the model.11) country 2 country ) Var(fn. By contrast. idiosyncratic . The paper also allows for a simple decomposition of variance attributed to the different factors. the world factor based on a large set of developed and developing countries implies that the recession of the 1980s was.t ) They find evidence that.t ) + Var( i.Facts 27 Köse et al. first. in contrast to models estimated using a smaller sample of countries. + (bi Thus. and 7% of investment growth. They interpret this finding as evidence for a world business cycle. This finding has been documented further in the international business cycle literature. which is discussed in detail in Chapter 4. the world factor explains nearly 15% of variation in output growth. Third. the regional and country factors explain covariation or co-movement that is less persistent. Second.’s paper [138] shows that the world factor identifies many of the major cyclical events over a 30-year period: the expansionary periods of the 1960s. the fraction of variance attributed to the world factor can be expressed as (biworld )2 Var(ft world ) . The authors also document some noteworthy findings regarding the sources of volatility of investment growth.t ). the recession of the 1980s associated with the debt crisis in developing countries and the tight monetary policies in the major developed countries. they find that country and idiosyncratic factors account for a much larger share of variability in investment growth than the world and regional factors.t ) = (biworld )2 Var(ft world ) + (bi region 2 ) Var(fr.t region ) (4. However. A second important finding from the paper is that most of the persistent co-movement across countries and aggregates is captured by the world factor.

They consider the time series behavior of output. disruptions to market access. Yet.7 We conjecture that variability in investment due to such factors is a key channel for inducing idiosyncratic volatility in developing economies’ investment behavior. especially in an era of increased trade flows and financial integration. and debt and currency crises. and the current account for 15 countries including the US. there is little evidence that the volatility of European aggregates can be attributed to the European regional factor. total consumption. They divide this period into four periods that also reflect the monetary and capital account regimes prevailing in them. and other European countries plus Argentina for the period since 1870. exchange rates. and Altug [79]. for example. This last result is interpreted as evidence against the existence of a European business cycle. policy reversals. . The authors pose this as a puzzle. Yet this finding could also be due to model misspecification. Serven and Solimano [189]. Rodrik [178] emphasizes the importance of political risk (in his case. Several studies document the importance of political risk in the unsuccessful recovery of private investment following the adoption of IMF stabilization packages in various countries. It featured fixed exchange rates and worldwide capital market integration. 7 In a related literature. Fallacy and Fantasy shocks explain a much larger fraction of the volatility in investment growth in developing countries relative to developed countries. see Caballero [53] or Demers.28 Business Cycles: Fact. 6 For recent reviews. 2. The last two findings of the paper imply that regional factors play a minor role in aggregate output fluctuations. See. there is also the possibility that some observed covariation could result from the spillover effects of idiosyncratic or country shocks. HISTORICAL BUSINESS CYCLES Basu and Taylor [31] examine business cycles within an international historical perspective. and Demers [10] argue that an important source of risk for developing countries is political risk or risk arising from the threat of expropriation. As Stock and Watson [199] note. prices.5. Demers. Paralleling this finding. the dynamic factor model suffers from the shortcoming that all covariation is attributed to the common factors. much work on investment behavior shows that irreversible investment decisions are particularly susceptible to risk and uncertainty. • 1870–1941: This era corresponded to the classic gold standard. the UK. However.6 Altug. real wages. the risk of policy reversal) for irreversible investment decisions. investment. Demers.

which simultaneously occurred in a number of countries. Germany. Spain. Norway. the world economy shifted from a globalized regime to an autarkic regime. Italy. which corresponded to the post-World War II era of reconstruction and a resumption of global trade and capital flows. The countries that they examine are Australia. Japan. This period also corresponded to the Great Depression. which examines whether the turning points in the different series occur at similar dates. Basu and Taylor [31] also argue that their periodization allows for an analysis of international capital flows and capital mobility in affecting cyclical fluctuations. Consideration of alternative historical periods. the so-called concordance correlation. Denmark. Instead of using information on NBER-type reference dates. Bordo and Heibling [44] use real GDP or industrial production series to determine synchronization. Switzerland. It also provides a way to identify the importance of demand. They examine business cycles in 16 countries during the period 1880–2001. France. to monetary phenomena. They also use standard output correlations and factor-based measures. For example. Bordo and Heibling [44] ask whether national business cycles have become more synchronized. 1945–1971: The third regime is the Bretton Woods era. and the US. Basu and Taylor [31] argue that considering such a breakdown allows for the analysis of the impact of different regimes on cyclical phenomena. the UK.” Among other concepts. Early 1970s–present: This period features a floating exchange regime and a period of increasing globalization.Facts 29 • • • 1919–1939: During this period. Synchronization refers to the notion that “the timing and magnitudes of major changes in economic activity appear increasingly similar. Sweden. the Netherlands. most explanations of the Great Depression attribute the source of this massive downturn. . including the era of the Great Depression. Finland. Portugal. they use a statistical measure of synchronization developed by Harding and Pagan [115]. and consider the four exchange rate regimes also examined by Basu and Taylor [31].versus supply-side factors or shocks and the role of alternative propagation mechanisms such as price rigidity. thus allows for an examination of such mechanisms as nominal rigidities in propagating shocks. Canada.

A simple measure of concordance is defined by the variable 1 Iij = T T [Sit Sjt − (1 − Sit )(1 − Sjt )]. (5.30 Business Cycles: Fact. Note ∗ that the variance of Iij is given by ∗ Var(Iij ) 4 = 2E T T t=1 ¯ ¯ (Sit − Si )(Sjt − Sj ) . let Sit and Sjt denote binary-cycle indicator variables which assume a value of 1 if the economy is in an expansion. they find that the average of the correlation coefficients is zero. we divide Iij by a consistent estimate of its standard error under the null hypothesis of independence. as half of all pairs of business cycles are negatively related to each other while the other half are positively related. then under the assumption that Sit and Sjt are independent. 0 otherwise. then the standardized index equals −1. say. They argue that Sit = 1 for most countries during the Bretton Woods era of 1948–1972 and hence leave this period out. if they are in different states in each period (so that Sit = 1 − Sjt ).12) ¯ Let Si = (1/T ) T Sit denote the estimated probability of being in an t=1 expansion. In the interwar and post-Bretton Woods era. Fallacy and Fantasy To briefly describe the concordance correlations. Bordo and Heibling [44] calculate the business cycle indicators Sit for each of the four exchange rate regimes. They define a recession as one or more consecutive years of real GDP growth. cycles were. more than half of all national business cycles become positively related to each other. Subtracting this from Iij yields the mean-corrected concordance index: ∗ Iij 1 =2 T T t=1 ¯ ¯ (Sit − Si )(Sjt − Sj ). The key finding is that during the classical gold standard. uncorrelated . For the gold standard era. while an expansion is defined as one or more years of positive GDP growth. ¯ ¯ ¯ ¯ the expected value of the concordance index is 2Si Sj + 1 − Si − Sj . If the two cycles are perfectly synchronized (so that Sit = Sjt for all t). then the standardized index equals 1. for countries i and j.13) ∗ To derive the concordance correlation coefficient. on average. t=1 (5. and if the two cycles are unrelated. the standardized index equals 0.

they started becoming synchronized with each other. However. They find that there has been a tendency for higher. They also find higher output correlations for the core European countries (the EEC) and the Continental European countries. They find that both global (common) shocks and transmission have become more important. In the trade linkages version. They consider a so-called center country version of this model. and that the increasing importance of global shocks reflects the forces of globalization. but this was not so for financial or twin crises. whereas beginning with the interwar period. the importance of transmission for peripheral countries arises only in the trade model. and a trade linkages version. especially the integration of goods and services through international trade and the integration of financial markets. We discuss emerging market business cycles in detail in Chapter 6. . The authors also examine standard output correlations. They find that the variability of output due to variability in the common factors has “doubled from about 20 percent during the Gold Standard era to about 40 percent during the modern era of flexible exchange rates. and the idiosyncratic shocks are allowed to be serially correlated and heteroscedastic. more positive bilateral output correlations by era. They argue that banking crises were less severe in the period before 1915. lagged GDP growth of the center country is included alongside the lagged value of the own country’s GDP growth in the VAR representation. suggesting that it is not transmission from the center country that accounts for the increased role of transmission. In this model. before 1914 and after 1971. as in Stock and Watson [199]. idiosyncratic shocks. which measure the magnitude as well as the direction of output changes. Finally. such crises have figured importantly in emerging market output fluctuations.Facts 31 with each other. and spillover effects in generating this result. Eichengreen and Bordo [84] examine the nature of crises in two periods of globalization. Typically.” They also estimate restricted versions of a FSVAR model along the lines of Stock and Watson [199] to determine the role of common shocks. lagged GDP growth of the major trading partner is included in place of the center country’s. Bordo and Heibling [44] conclude by noting that global shocks are the dominant influence across all regimes. they find an increase in correlations for the Anglo-Saxon countries. The authors also estimate a so-called static approximate factor model for the growth rates of GDP. In the former. there are no dynamics in the relation between output growth rates and the factors.

This page intentionally left blank .

This approach emphasizes the role of intertemporal substitution motives in propagating shocks. Current real business cycle (RBC) theory argues that business cycles can arise in frictionless. More controversially. complete markets in which there are real or technology shocks. Keynesian and New Keynesian models stress the role of frictions such as price stickiness. By contrast. have figured prominently in the recent literature. whether they are permanent or transitory such as oil shocks. Political shocks. perfectly competitive. as Keynes [127] did. The Great 33 . that investors’ “animal spirits” or. wars. and other disruptions in market activity have also been acknowledged to play a role. one could also assign a role to taste shocks or changes in preferences of consumers. more precisely.Chapter 3 Models of Business Cycles The standard approach to classifying business cycle models follows Ragnar Frisch’s [95] terminology of impulses and propagation mechanisms. changes in their subjective beliefs could help to trigger business cycles. Technology shocks which alter a society’s production possibilities frontier. Weather shocks have always had a place among the impulses thought to trigger fluctuations in economic activity. the result is unemployment and greater declines in output relative to a situation with flexible prices. Much of the controversy in the business cycle literature has stemmed from differences attached to the importance of alternative propagation mechanisms and the associated shocks. if real wages do not adjust downward when there is a negative shock to demand. one could argue. Equivalently. The shocks or impulses thought to instigate business cycles have typically been varied and diverse. In a simple labor market model.

34 Business Cycles: Fact. As Kydland and Prescott argue. Since the debate has revolved around the efficacy of technology shocks and intertemporal substitution effects in replicating business cycles. The standard RBC model also assumes that there is no government. Some of the extensions of the RBC approach have modified the basic framework to incorporate the role of government. and that there is no trade with the rest of the world. for example. In this case. 2 Rebelo [177] provides a recent review of alternative models and directions associated with the RBC agenda. AN RBC MODEL RBC theory has become popular in recent years because its micro foundations are fully specified and it links the short run with the neoclassical growth model. that households consume only goods produced in the market. this is a crucial modification as “more than half of business cycle fluctuations are accounted for by variations in the labor input” ([143]. A number of these assumptions have come under criticism in the recent literature. Bernanke and Gertler [40] or Kiyotaki and Moore [134].1 In this chapter. all factors are fully employed. 173). and international trade. . There is a representative consumer who derives utility from consumption and leisure. We will discuss these extensions in the following sections. but his discussion is abbreviated to satisfy the constraints imposed by a journal article. shocks originating in various financial markets lead to a widening circle of bankruptcies and bank failures as well as large and negative effects on real output. we discuss variants of the RBC model. and all prices adjust instantaneously to clear markets. 1 See.2 3. it seems imperative to discuss the main propagation mechanisms of the RBC model. and a constant-returns-to-scale (CRTS) production technology for producing the single good using labor and capital. p. All markets are perfectly competitive.1. In the following chapter. we discuss an international RBC model to describe how the transmission mechanisms proposed by this approach translate into an open-economy setting. home production. Fallacy and Fantasy Depression and the recent financial crises also indicate the importance of credit market frictions and financial accelerator-type effects as a potential propagation mechanism for business cycles. The prototypical RBC model has the structure of a standard neoclassical growth model with a labor/leisure choice incorporated.

3 See. which is assumed to be stationary but persistent as long as ρ = 1. 0 < θ < 1. There is a representative firm with CRTS production that is affected by a stochastic technology shock each period: yt = exp (zt )ktα ht1−α . 0 < ρ < 1.1) where 0 < β < 1 is the discount factor. (1.4) Capital evolves according to kt+1 = (1 − δ)kt + it . (1. time spent not working is taken as leisure. The aggregate feasibility constraint is defined as ct + it = yt .3) Assume that the technology shock follows an AR(1) process: t+1 .d. The time constraint requires that the sum of leisure and labor hours equals 1: lt + ht = 1. zt+1 = µ+ρzt + 0 < α < 1. and N (0. The problem described above is an infinite-horizon dynamic stochastic optimization problem. For ease of exposition. lt ) = ctθ lt1−θ . We will discuss the propagation mechanism when describing the solution for the model.. (1. t ∼ i. (1. The existence of a solution can be shown using standard recursive methods for analyzing such problems. Altug and Labadie [6] or Stokey and Lucas with Prescott [201]. for example. (1. σ 2 ).6) Notice that the only shock in this model is the technology shock. lt ) . .3 Assuming a solution exists.Models of Business Cycles 35 Turning to the basic model. (1.2) Hence.i. the representative agent has time-separable preferences over consumption and leisure choices given by ∞ U = E0 t=0 βt u(ct . we will suppose that the utility function has the form U (ct .5) where it denotes economy-wide investment and 0 < δ < 1 is the depreciation rate.

zt+1 ) = λt . U1t (1. (1.t+1 [exp (zt+1 )α(kt+1 /ht+1 )α−1 + (1 − δ)] = 1. which he then consumes.8) (1.7) subject to ct + kt+1 = exp (zt )ktα ht1−α + (1 − δ)kt . Notice that this is just a simple Robinson Crusoe economy in which the Crusoe produces output using capital and labor. There is no unemployment in the model as time not spent working is (optimally) taken as leisure. zt ) = max {U (ct . βEt Vk (kt+1 . and ct = yt . Thus. the first-order conditions with respect to ct .10) (1. lt ) = λt exp (zt )(1 − α)ktα ht−α . lt + ht = 1. kt+1 and lt and the envelope condition are U1 (ct . Vk (kt . his choice is between how much to work and how much time to spend in leisure.lt . that is. Let λt denote the Lagrange multiplier on the resource constraint. zt+1 )} ct . lt ) = λt . U2 (ct . Observe that the ratio U2 /U1 is a function of the capital-labor ratio.36 Business Cycles: Fact. ¯ it = 0. respectively.11) where U1t and U2t denote the marginal utility of consumption and leisure. Suppose for the moment that there is no investment in the model. zt ) = λt exp (zt )αktα−1 ht1−α + (1 − δ) .13) The first equation shows that the marginal rate of substitution between consumption and leisure is equal to the marginal product of labor. This means that the capital accumulation process will also be a function of the capital-labor ratio.kt+1 (1. Substituting for lt = 1−ht using the time constraint. lt ) + βEt V (kt+1 . kt /ht . Fallacy and Fantasy let the value function associated with the problem be expressed as V (kt . To analyze the optimal consumption-leisure .12) (1.9) (1. The second equation is the intertemporal Euler equation. These conditions can be rewritten as U2t = exp (zt )(1 − α)(kt /ht )α U1t βEt U1. kt = k.

1 − θ + θα (1. . we can rewrite the intertemporal Euler equation by noting that U1 (ct+i . However. (1. Notice that the technology shock does not affect optimal hours worked because. (1.12). for this preference specification. then Crusoe spends 50% of his time working and 50% taking leisure.16) Now a temporary negative technology shock also affects investment or saving. we have the intertemporal Euler equation characterizing the behavior of the optimal capital stock over time. which under our preference specification becomes (1 − θ)ct = exp (zt )(1 − α)(kt /ht )α . the income and substitution effects of an increase in productivity on hours worked cancel each other out. Figure 3.14). for example. in addition to equation (1. θ(1 − ht ) We can solve this equation for the optimal labor hours ht∗ as ht∗ = θ(1 − α) . This yields βEt exp (zt+1 ) kt+1 ht+1 α θ−1 α exp (zt+1 ) kt+1 ht+1 α−1 +1−δ kt = exp (zt ) ht α θ−1 . Then. Now suppose output can be consumed or invested so that it = 0. Using our preference specification. the impact of these changes is to induce fluctuations in labor supply and employment as well as investment. lt+i ) = θ(ct+i /lt+i )θ−1 for i ≥ 0 and using the result in equation (1.1 describes the representative agent’s optimum.Models of Business Cycles 37 choice. Crusoe will typically respond to a temporary negative shock by lowering savings and also by working less.15) Thus.14) θlt Substituting for ct = yt = exp (zt )ktα ht1−α yields (1 − θ) exp (zt )ktα ht1−α = exp (zt )(1 − α)(kt /ht )α . Thus.14). we note that if the substitution effect of an increase in productivity dominates the income effect. then Crusoe will work less in response to a temporary negative technology shock and will also consume and produce less. we consider equation (1. if θ = 2/3 and α = 1/2.

Furthermore. 3. it will have a lower capital stock in the future. if the economy experiences a period with lower investment. Fallacy and Fantasy Fig. where sht = wt ht /pt yt denotes the share of labor in national income.1. the parameter α is typically measured as the share of capital in real output defined as skt = rt kt /pt yt . We can also derive a measure of the Solow residual for this model by using the form of the production function. Under constant returns to scale. Under these assumptions.3) and then take the first differences: ln (yt+1 ) = zt+1 + α ln (kt+1 ) + (1 − α) ln (ht+1 ). where rt denotes the competitively determined rental rate on capital and pt denotes the product price normalized here as unity. To derive an observable measure of the Solow residual. the growth in the total factor productivity (TFP) is measured as a residual: zt+1 = ln (yt+1 ) − skt ln (kt+1 ) − (1 − skt ) ln (ht+1 ). Robinson Crusoe’s Optimum. skt + sht = 1. take the logarithm of both sides of equation (1.38 Business Cycles: Fact. implying that the effect of the shock persists. . To derive the Solow residual.

Notice that in the RBC framework the Solow residual is typically treated as exogenous. The quadratic approximation procedure is implemented as follows. 3. we can substitute for consumption in the utility function to obtain u(ct ) = u( exp (zt )ktα ht1−α − it .16) together with the feasibility . We implement a quadratic approximation procedure that was initially suggested by Kydland and Prescott [141]. Letting u(ct .Models of Business Cycles 39 Solow [196] showed that the residual accounted for about one half of the US GDP growth between 1909 and 1949.14) and (1. Step 1: Compute the deterministic steady state for the model. This is an issue that has been contested in the business cycle literature.17) (2. rendering the RBC interpretation invalid. 40% of GNP growth in the USA between 1929 and 1957 resulted from technical development. Similarly. according to Denison [80]. and convert the original nonlinear dynamic optimization procedure to one which has a quadratic objective and linear constraints (see also Christiano [64]). 0 < θ < 1. We assume that capital depreciates at the rate 0 < δ < 1 and the technology shock follows the process in (1. zt+1 = µ + ρzt + ct ≥ 0. This is obtained by setting the exogenous technology shock equal to its mean value and evaluating the conditions (1. the problem now becomes ∞ {it . (2. t+1 . A NUMERICAL SOLUTION We now provide a numerical solution to further illustrate the properties of the standard RBC model.18) (2. consumption plus investment equals output at the optimum. Critics of the RBC approach argue that there are endogenous components to the fluctuations in the Solow residual. Using this fact. 1 − ht ) = θ ln (ct ) + (1 − θ) ln (1 − ht ).2. Since the utility function is strictly increasing.ht }t=0 max E0 ∞ t=0 θ ln ( exp (zt )ktα ht1−α − it ) + (1 − θ) ln (1 − ht ) subject to kt+1 = (1 − δ)kt + it .4).19) 0 ≤ ht ≤ 1. 1 − h).

kt . substituting for c first and then dividing through by y . To find an explicit expression for h parameters. Let z = µ/(1 − ρ) denote the ¯ unconditional mean for the (log of ) technology process. ¯ ¯ α h 1−α y ¯ exp (¯ ) z r +δ exp (¯ )k z ¯ Substituting back into the equation defining h and simplifying yields α ¯ = θ(1 − α). Using (1. we obtain ¯ ¯ ¯ k ¯ (1 − θ)h 1 − δ y ¯ Now note that ¯ ¯ ¯ ¯ k (k/h)1−α k α = = = . investment is equal to depreciation: ¯ ¯ i = δk. Fallacy and Fantasy conditions at constant values for ct .22) as ¯c ¯ y (1 − θ)h¯ = (1 − α)θ(1 − h)¯ . ¯ θ 1−h c ¯ (2.22) (2.16). . use the relation in (2. note that in the deterministic steady state. (2. we obtain ¯ y ¯ 1−θ h = (1 − α) .40 Business Cycles: Fact. (r + δ)(1 − θα) − (1 − θ)δα (2. Rearranging.21) ¯ where the aggregate feasibility constraint c + δk = y ≡ exp z k α h 1−α is also ¯ ¯ ¯¯ ¯ ¯ in terms of the underlying assumed to hold. Simplifying the result in (1. we can solve for the steady-state capital-labor ratio as ¯ exp (¯ ) z k = α ¯ r +δ h 1/(1−α) . and ht .14). h (1 − θ) + θ(1 − α) − (1 − θ)δ r +δ or ¯ h= θ(1 − α)(r + δ) .20) Next.23) ¯ + θ(1 − α)h = θ(1 − α).

k. i. See Campbell [55]. u) . k.4 For this purpose. ht − h) . . Thus. ∂s∂s where all partial derivatives are evaluated at ¯. kt − k. z . the quadratic form (s − ¯) T (s − ¯) can be s written as s (s − ¯) T (s − ¯) = s = x u x u T11 T12 T21 T22 R W W Q x u x . h). zt+1 zt 0 µ 0 ρ 000 t+1 s s Notice that s − ¯ = (x. h) and ¯ ≡ (0. u 4 An alternative approach is to log-linearize the model. Approximate u(s) near the deterministic steady state ¯ s ¯ ¯ ¯ ¯ s using a second-order Taylor series approximation as u(s) = u(¯) + (s − ¯) s s ∂u(s) 1 ∂2 u(s) (s − ¯). s + (s − ¯) s ∂s 2 ∂s∂s Define e as the 5 × 1 vector with a 1 in the first row and zeros elsewhere.Models of Business Cycles 41 Step 2: Approximate the original utility function by a quadratic function around the deterministic steady state. zt − z ) and the ¯ ¯ vector of control variables as ut = (it − i. i. z. let s ≡ (1. s ¯ ¯ Define the vector of state variables as xt = (1. We can write the law of motion for the state variables as           0 1 0 0 000 1 1 0  kt+1  =  0 (1 − δ) 0   kt  +  0 1 0   it  +  0  . s where 1 ∂2 u(s) T = e u(¯) + ¯ s ¯ e s s 2 ∂s∂s + 1 2 ∂u(s) ∂u(s) e +e ∂s ∂s + 1 2 ∂2 u(s) . Notice that the utility function can be written as s u(s) = (s − ¯) T (s − ¯).

xt . (2. Given the structure of the problem. McGrattan. the solution for the stochastic optimal control problem is identical to the solution for the deterministic version of the problem with the shocks εt+1 replaced by their expectation E (εt+1 ). This class of problems is known as optimal linear regulator problems. Such problems can be solved using the methods for solving dynamic optimization problems that satisfy a certainty equivalence property. Hansen. Using the definition of T . Fallacy and Fantasy Step 3: Convert the original dynamic optimization problem into a problem with linear constraints and a quadratic objective function. This is now an optimal control problem with a quadratic objective and linear constraints. t ≥ 0.42 Business Cycles: Fact. (2. where d and P are quantities to be determined. Substituting for next period’s state variables and using this expression for the value function. Bellman’s equation becomes x Px + d = max x Rx + u Qu + 2x Wu u + βE (Ax + Bu + ε) P(Ax + Bu + ε) + d . and ut .24) subject to the linear law of motion xt+1 = Axt + But + εt+1 . and Sargent [16] provide further discussion of the formulation and estimation of linear dynamic economic models.26) . Ljungqvist and Sargent [146] and Anderson. in other words. notice that the value function will be a quadratic function in the state variables: V (x) = x Px + d . Bellman’s equation for this problem is given by V (xt ) = max xt Rxt + ut Qut + 2ut Wxt + βE [V (xt+1 )] ut subject to xt+1 = Axt +But +εt+1 . the dynamic optimization problem can now be written as ∞ {ut }t=0 max E0 ∞ t=0 βt [xt Rxt + ut Qut + 2xt Wut ] (2.25) where E (εt+1 ) = 0 and E (εt εt ) = .

5To derive this result. we have used the rules for differentiating quadratic forms as ∂u Qu ∂x Tu ∂u Tx = [Q + Q ]u = 2Qu.27) Notice that the shocks εt+1 do not affect the optimal choice of ut . output. Q . the model is then calibrated with the data. A. ∂u ∂u ∂u 6 Notice that the matrices R. The equation for P is known as the algebraic matrix Riccati equation for so-called optimal linear regulator problems. and B must be further restricted to ensure the existence of a solution to the matrix difference equation defining Pn . The simulated series are used to generate time series for consumption. The policy function associated with Pn is Fn+1 = (Q + βB Pn B)−1 (βB Pn A + W ). Substituting for u back into the definition of V (x) yields P = R + βA PA − (βA PB + W )(Q + βB PB)−1 (βB PA + W ).6 The solution for P can be obtained by iterating on the matrix Riccati difference equation: Pn+1 = R + βA Pn A − (βA Pn B + W )(Q + βB Pn B)−1 (βB Pn A + W ). A stochastic process for zt is also specified and a random number generator is used to simulate the TFP time series. given particular functional forms for the production function and the utility function (see Cooley and Prescott [75]).28) (2. Calibration refers to the practice of determining the parameters of the model based on its steady state properties and the results of other studies. In the RBC literature. (2. starting from P0 = 0. One condition that suffices is that the eigenvalues of A are bounded in modulus below unity. where F = (Q + βB PB)−1 (βB PA + W ). = T x and = Tx. . d = (1 − β)−1 [βE (ε Pε)].Models of Business Cycles 43 The first-order conditions with respect to u are Qu + W x + β[B PAx + B PBu] = 0.5 Solving for u yields u = −(Q + βB PB)−1 (W x + βB PAx) = −Fx.29) (2.

The model is then judged to be close or not close based on this comparison. The steady values for the variables are given by k = 1.9163 0. Standard Deviation (%) Output Consumption Investment Hours Productivity 8.1128. In terms of the correlation of each series with output. Trends in the data are removed using a given filter.9238 1. with hours showing the least procyclicality.36.1.4934 0. we calculate unconditional moments for the different series after dropping the first 1000 observations. θ = 0.10. we find that all series are procyclical. and generate a sequence of ˆ technology shocks beginning from some initial value z0 as zt+1 = ρzt + ˆ t+1 .7496 Correlation with Output 1 0. δ = 0.2952.1.028. α = 1/3. We first calculate a i ¯ ¯ set of unconditional moments that have been used in the RBC literature to match the model with the data. we draw a sequence of shocks {ˆ t+1 }2999 that are normally distributed t=0 with mean zero and standard deviation 0.1134 12.6684 7. so the approach differs significantly from standard econometrics. Specifically. productivity is almost as variable as output.1281. ¯ ¯ = 0. The typical set of unconditional moments used in the RBC literature is displayed in Table 3.95. By contrast. Likewise. These are obtained by simulating the behavior of the different series across a given history of the shock sequence. variances. covariances.3655. and labor. ρ = 0. We then use the linear decision rules for all variables to simulate for the endogenous variables based on the same history of shocks.8616 8. and cross-correlations for the simulated time series are compared to the comparable statistics in the data. Cyclical Properties of Key Variables.95.028.9840 . and h = 0. The means.44 Business Cycles: Fact. Consider the parameter values β = 0. y = 0. The RBC theorists have interpreted these results as implying that.9807 0. Fallacy and Fantasy investment. c = 0. and ¯ σ = 0. We now illustrate this solution method for the standard RBC model with a labor-leisure choice presented at the beginning of this section. We find that consumption fluctuates slightly less than output and investment fluctuates significantly more. in the first instance. we find that the variation in hours is typically quite low.4783. Finally. the model is capable of delivering some of the salient features Table 3.

Consumption.Models of Business Cycles 45 of the data. the RBC model has been viewed as being able to generate cycles endogenously and having economic significance even if the calibration method is controversial. Consumption shows a gradual positive response because investment is initially high. rising first and then declining back to a lower level. The percentage change in output exceeds the percentage change in the technology shock because optimal hours also show a positive response to the technology shock.5 0. and productivity are all strongly procyclical. investment. This is discussed further in the following sections.2. the model predicts that consumption fluctuates less than output whereas investment fluctuates more. capital and consumption show a humped-shaped response. Specifically. By contrast. Figure 3. Thus.1 1 % deviations form steady state 0. These responses have been widely Hours Output Consumption Capital Technology 1. Impulse Responses to a Shock in Technology. .8 0.6 0.7 0. the technology shock explains much but not all of the variation in output. 3. In terms of the variation accounted by the technology shocks.2 illustrates the response of all the variables to a 1% shock in technology starting from the steady-state capital stock.4 0 5 10 Years after shock 15 20 Fig. We note that hours and output both increase and then fall back to a lower level. as we observe in the data.9 0.

46

Business Cycles: Fact, Fallacy and Fantasy

documented in the RBC literature (see, for example, Uhlig [206]). The positive response of hours to a positive productivity shock has also become an important point of difference between models that assume perfectly flexible prices versus those that assume nominal price rigidity. We discuss New Keynesian models with monopolistic competition and prices that are fixed in the short run in Chapter 5.

3.3. INITIAL CRITICISMS
In perhaps what is the most famous example of an RBC model, Kydland and Prescott [141] formulated a real business cycle model in which preferences are not separable over time with respect to leisure and there exists a time-tobuild feature in investment for new capital goods. The production function displays constant returns to scale with respect to hours worked and a composite capital good. The only exogenous shock to their model is a random technology shock which follows a stationary first-order autoregressive process. They use the quadratic approximation procedure to obtain linear decision rules for a set of aggregate variables, and generate time series for the remaining series by drawing realizations of the innovation to the technology shock. They calculate a small set of moments associated with each series to match the model with the data. When calibrating their model, Kydland and Prescott choose the variance of the innovation to the technology shock to make the variability of the output series generated by their model equal to the variability of observed GNP. As McCallum [156] notes, this feature of their analysis makes it difficult to judge whether “technology shocks are adequate to generate output, employment, etc. fluctuations of the magnitude actually observed.” One of the most pervasive criticisms of the model is that there is no evidence of large, economy-wide disturbances that can play the role of the technology shocks posited by Kydland and Prescott. The only exception is oil price shocks, leading one to question whether the model was developed in reaction to the supply-side disturbances that characterized the 1970s. The RBC approach in general has difficulty in answering what some other examples of big shocks are. Summers [202] argues that “the vast majority of what Prescott labels technology shocks are in fact observable concomitants of labor hoarding and other behavior which Prescott does not allow for in his model.” This point was subsequently

Models of Business Cycles

47

elaborated on by Hall [110, 111], who argued that the procyclical movements in the Solow residual were, in fact, endogenous changes in efficiency arising from labor hoarding, increasing returns to scale, or price markups. Summers [202] also took issue with the parameters of the model that Kydland and Prescott [141] or Prescott [173] advocated in their calibration exercise. He argued that the econometric evidence presented by Eichenbaum, Hansen, and Singleton [83] on the intertemporal substitution of leisure hypothesis pointed to a share of time allocated to market activities that was more in the range of one-sixth, not one-third. He also noted that a real interest rate of 4% was inconsistent with the historical evidence of very low yields averaging around 1%. Singleton [192] raised the problem of inference and sampling uncertainty surrounding the various measures of fit proposed in the RBC literature. He noted that the calibration approach did not provide a straightforward and transparent way of judging whether a given model constituted an improvement over another model. Eichenbaum [82] raised the issue of the model’s fit based on the ratio of variance for GDP implied by the model and the data. Eichenbaum showed that the standard error of this ratio was heavily influenced by the parameters of the assumed stochastic process for the technology shock. The policy implications of the RBC model have also come under attack. According to the RBC model, business cycle fluctuations are Pareto optimal and there is no role for the government to try to smooth or mitigate the fluctuations. In fact, such policy efforts are inefficient. Prescott [173] commented as follows:
Economic theory predicts that, given the nature of the shocks to technology and people’s willingness and ability to intertemporally and intratemporally substitute, the economy will display fluctuations like those the U.S. economy displays. . . . Indeed, if the economy did not display the business cycle phenomena, there would be a puzzle.

Thus, according to this approach, cyclical fluctuations are viewed as the natural response of the economy to changes that affect individuals’ consumption or production decisions. However, if prices or wages are fixed or rigid or if there are other types of frictions arising from credit markets, then the policy implications of the RBC approach may not be valid and the economy may operate with high levels of unemployment and excess capacity over long periods. In this

48

Business Cycles: Fact, Fallacy and Fantasy

case, government policies may be required to move the economy towards a full employment level. In the chapters that follow, we will discuss the efficacy of the RBC model in serving as a model of aggregate fluctuations as well as questions that have lingered to this day about the adequacy of the calibration approach in matching the model to the data.

3.4. “PUZZLES”
Many of the initial criticisms which had been voiced by opponents of the RBC approach took on the character of “puzzles”. These puzzles have constituted the basis of much further research in the area. We can list some of these puzzles as follows: • The variability of hours and productivity: As seen in Table 3.1, one of the main problems with the standard RBC model is that it cannot capture the variation in the aggregate labor input (see also Kydland [140]). In the data, employment is strongly procyclical and almost as variable as output while real wages are weakly procyclical. In the standard model, a productivity shock shifts the marginal product of labor so that the observed variations in employment can only occur if the labor supply curve is relatively elastic. Yet, micro studies find that wage elasticity of labor supply is quite low. In this case, the marginal productivity shock should lead to most of the adjustment in real wages and less in the quantity of labor. Furthermore, in the data, hours of work per worker adjusts very little over the cycle. About two-thirds of the variability in total hours worked comes from movements into and out of the labor force, and the rest is due to adjustment in the number of hours worked per employee. These findings create a puzzle for the standard RBC model. • The productivity puzzle: In the data, the correlation between productivity and hours worked is near zero or negative, while the correlation between productivity and output is positive and around 0.5.7 By contrast, the RBC model, which is driven entirely by productivity shocks, generates correlations that are large and positive in both cases. Another problem that
7 See, for example, Christiano and Eichenbaum [65], who measure hours worked and productivity based on both household and establishment-level surveys conducted by the US Department of Labor.

31) . the individual gets paid regardless of whether he works or not. whereas in the RBC model labor’s share is fixed. (4. labor’s share of income moves countercyclically. • Reverse causality: The stylized facts of business cycles state that money. The conventional wisdom regarding the money-output correlation. This framework also provides a way for reconciling large labor supply elasticities at the aggregate level with low labor supply elasticities at the individual level. ˆ Suppose that workers are constrained to work either zero or h hours. A lottery then determines whether an individual actually works.4. This is accomplished by assuming that individuals can work all the time or not at all. Let πt denote the probability that a given agent is employed in period t so that the number of per capita hours worked is given by ˆ Ht = πt h.30) The main idea is that there are nonconvexities or fixed costs that make varying the number of employed workers more efficient than varying hours per worker. To account for the nonconvexities introduced by the work/nonwork decision. lottery models studied by Gary Hansen [112] and Richard Rogerson [179] have been used to explain the stylized fact that aggregate hours vary more than productivity does. (4. especially M2. is that the causality goes from money to output but with “long and variable lags”. appears to be a leading indicator of aggregate output. 3. it is assumed that individuals choose the probability of working. We now describe a version of the RBC model due to Rogerson [179] and Hansen [112] that allows for fixed costs and nonconvexities in labor supply.Models of Business Cycles 49 arises in matching the model and the data is that in the data. However. A Model with Indivisible Labor Supply The indivisible labor. where ˆ 0 < h < 1. Since what is being traded is the contract. this is inconsistent with the notion that TFP shocks are the driving force behind business cycle activity. This economy is one in which individuals and a firm trade a contract that commits the household to work h0 hours with probability πt . summarized by Friedman and Schwartz [94] in their study of monetary history.1. πt .

ex ante all individuals are alike. Notice that the social planner chooses the consumption allocations of each agent plus the probability of their working.t = ct so that the agent consumes the same amount whether or not he is working.34) It follows that c0. Hence. but ex post they differ because some work while others enjoy leisure. πt c1.t .t = c1. l ) = ln (c) + A ln (l ).t denote the consumption of an unemployed worker and c1. the unemployed worker enjoys higher utility since working causes disutility.c0. the first-order conditions with respect to (c0.32) max E [u(ct .t denote the consumption of an employed agent. lt )] = πt u(c1.t .t ) are 1 − πt = (1 − πt )λt . Omitting the work/leisure decision for the moment. This is a feature that is counterfactual to the working of actual labor markets. Then the expected utility of the representative consumer.50 Business Cycles: Fact. they ˆ must supply h hours of work so that there is no choice over hours of work directly. 1 − h) + (1 − πt )u(c0. is given by ˆ E [u(ct . Fallacy and Fantasy Let c0. (4. c1. Assume that the individual utility function has the form u(c. When agents do work.t πt = πt λt . c1. 1). With complete insurance and identical preferences that are separable with respect to consumption and leisure. all individuals have the same consumption but the unemployed are better off.t = ct .c1.t + (1 − πt )c0.t . Notice that the agent will consume ct whether or not he is working.33) (4.t (4.t . Hence. lt )] s.t. The social planner solves the problem πt . In this model. c0. expected utility (where the expectation is over whether or not you work) is ˆ ln (ct ) + πt A ln (1 − h) + (1 − πt )A ln (1).t (4. taking into account the work versus nonwork decision. Let λt denote the Lagrange multiplier on the feasibility constraint for consumption.35) .

Ht . The preferences can now be written as ∞ U = E0 t=0 βt u(ct . (4. By contrast. Ht ) . lt )] = ln (ct ) + πt A ln (1 − h) = ln (ct ) − BHt . Ht ) = ln (ct ) − BHt .32) with equation (4. Using the definition of Ht . We now solve the model subject to the time constraint. ˆ h (4. which is consistent with assumptions about individual behavior.kt+1 } max E0 t=0 βt ln (ct ) − BHt + λt [exp (zt )ktθ Ht1−θ +(1 − δ)kt − ct − kt+1 ] . where u(ct . resource constraint.36) Comparing equation (4. and law of motion for the capital stock described earlier. ct B = λt exp (zt )(1 − θ)ktθ Ht−θ . πt = Ht /h.37) (4. The first-order conditions are 1 = λt .39) .38) (4. the latter specification — which is linear in total hours Ht — implies a high intertemporal elasticity at the aggregate level. where B= ˆ −A ln (1 − h) . θ−1 1−θ λt = βEt λt+1 exp (zt+1 )θkt+1 Ht+1 + (1 − δ) .36) shows the effect of the lottery assumption.Models of Business Cycles 51 ˆ where A is a positive constant.35) and use ln (1) = 0 to obtain ˆ E [u(ct . production function. The former specification for preferences implies a low intertemporal elasticity of substitution in labor supply. Now substitute for πt in (4. The problem is ∞ {ct .

(4. With Ht determined in (4. These typically specify an extra margin regarding individuals’ choices that helps break the close link between hours and productivity implied by the standard model. and the stochastic process for the technology shock using the approach in Kydland and Prescott [141]. δ.7 compared to the model without indivisibilities which implies a value near unity.4.37) and (4.41) where the term in square brackets shows the rate of return to investing in the aggregate production technology.39) yield the intertemporal Euler equation as 1=β ct ct+1 θ−1 1−θ [exp (zt+1 )θkt+1 Ht+1 + (1 − δ)] . (4.40) Equations (4. we can use the resource constraint to solve for ct and then substitute for ct into the intertemporal Euler equation to obtain a nonlinear stochastic in kt+1 with forcing process {zt }. Hansen.8 3. β. The purpose of the framework that Hansen [112] and Rogerson [179] consider is to generate the stylized fact with respect to the relative variability of hours versus productivity. Hansen argues that the model with indivisibilities can generate a variability of hours relative to productivity around 2. .40). A. The Productivity Puzzle Several resolutions of the productivity puzzle have been suggested. and it is not intended to incorporate the microeconomic foundations of the labor market. Hansen [112] calibrates this model by specifying values for the unknown parameters θ.2.38) can be used to solve for Ht as Ht = Bct exp (zt )(1 − θ) ˆ A ln (1 − h)ct ˆ exp (zt )(1 − θ)h −1/θ kt −1/θ = − kt . Fallacy and Fantasy Notice that (4. 8 See Browning. and Heckman [48] for further discussion regarding the reconciliation of micro evidence with dynamic general equilibrium models.52 Business Cycles: Fact.

Let N denote the time endowment of the representative household per period. ¯ ln (N − nt ) ¯ (4. the standard model cannot deliver the strong procyclical response of hours without procyclical behavior in productivity. and α is a parameter that governs the impact of government consumption on the marginal utility of private consumption.45) where zt is a technology shock which evolves according to the process zt = zt−1 exp (λt ). They consider a prototypical RBC model with a labor/leisure choice. (4. p where ct is private consumption.Models of Business Cycles 53 Government Consumption Shocks Christiano and Eichenbaum [65] note that as long as there is a single shock that drives the behavior of both hours and productivity.42) ¯ where ct denotes consumption and N −nt denotes leisure of the representative household. namely. They generate the negative correlation between hours worked and real wages or productivity by introducing government consumption shocks. Consumption services ct are related to private and public consumption as follows: ct = ct + αgt . and utilize the solution of the social planner’s problem to derive the ¯ competitive equilibrium allocations.46) .43) is public consumption. (4. They consider two different specifications for the labor/leisure choice. one which is based on a time-separable logarithmic specification and a second which incorporates the Hansen indivisible labor assumption. gt p (4. 0 < θ < 1.44) V ( N − nt ) = ¯ N − nt for all t. Per capita output is produced according to the Cobb–Douglas production function yt = (zt nt )1−θ ktθ . The social planner ranks alternative consumption/leisure streams according to ∞ E0 t=0 ¯ βt [ln (ct ) + γV (N − nt )] . (4.

and gt = gt /zt . and u4 < 0. u2 > 0. cht . hours of work can increase along a downward-sloping labor demand curve. p (4.54 Business Cycles: Fact. yt = yt /zt . hours increase and average productivity declines in response to a positive government consumption shock. Home Production Another way of improving the model’s ability to match the data is to introduce home production. and hht is labor time spent in home work. t=0 where 0 < β < 1. the solution for the model can be more fruitfully expressed in terms of the transformed variables ¯ ¯ ¯ ¯ kt+1 = kt+1 /zt . Assume that u1 > 0. . cmt is the consumption of a market good. hmt is labor time spent in market work. In this model. λt is an independent and identically distributed process with mean λ and standard deviation σλ . The total amount of time available to the household is normalized as unity. The aggregate resource constraint stipulates that consumption plus investment cannot exceed output in each period: ct + gt + kt+1 − (1 − δ)kt ≤ yt . cht is consumption of the home-produced good. consider a decision-maker who has preferences ∞ βt u(cmt . and µt is the innovation to ln (¯t ) g g g with standard deviation σµ . u3 < 0. and leisure is defined as time not spent working in the market or at home: lt = 1 − hmt − hht . government consumption shocks lead to shifts in the labor supply curve so that. Following Benhabib. Hence. Fallacy and Fantasy In this equation. and Wright [39]. The specification of the model is completed by assuming a stochastic law of motion for gt as ¯ g g ln (¯t ) = (1 − ρ) ln (¯ ) + ρ ln (¯t−1 ) + µt . The key assumption is that private and government consumption are not perfect substitutes. ct = ct /zt . in the absence of technology shocks. Rogerson. If leisure is a normal good. hmt . |ρ| < 1. an increase in government consumption leads to a negative wealth effect for consumers through the economy-wide resource constraint.47) Since the technology shock follows a logarithmic random walk.48) where ln (¯ ) is the mean of ln (¯t ). hht ). g (4. In this expression.

kht . However. then firms may retain workers even though they are not exerting much effort.Models of Business Cycles 55 At each date. A rise in market productivity may induce households to substitute away from home production towards market production. Alternative measures put home production to be in the range of 20–50% of GDP. home production is used to produce a nontradeable consumption good. but market capital goods are rented to firms at the competitive rental rate rt . one criticism of the home production theory is that it suggests that all movements out of the labor force (toward home production) are voluntary. and household capital goods kht . if there are costs to hiring or laying workers off. the labor supply curve also shifts in response to a good productivity shock. Household capital goods are used in home production. labor effort is likely to be adjusted first in response to a productivity shock. . This gives us another margin to substitute market labor and improves the model’s predictions. the household’s budget constraint is cmt + km.t+1 + kh. Letting wt denote the wage rate and δm and δh denote the depreciation rates on market and home capital. However. Labor Hoarding A third way to resolve the productivity puzzle is through a labor hoarding argument. respectively.t+1 ≤ wt hmt + rt kmt + (1 − δm )kmt + (1 − δh )kht . market capital goods kmt . thereby leading to greater variability in labor. This says that the effective labor input can be altered even though the total number of workers is fixed. zht ). where zht is a shock to home production and g is increasing and concave in labor and capital. The firm may not alter its work force every time there is a productivity shock (which would occur through a shift in the marginal product of labor curve). Eventually more workers may be hired or fired. the household can purchase market goods cmt . Hence. Unlike the standard model. In the model. Home goods are produced according to the home production function: cht = g (hht . but only after longer periods of time.

the total number of workers. then the productivity/hours correlation is reduced and more closely matches the data. and Rebelo [52]. hence.3. let f denote a fixed shift length. the production function can be written as Yt = St Ktα [Ht ]1−α .4. Fallacy and Fantasy To illustrate the role of labor hoarding. The comments made regarding fluctuations in the effort level of labor also apply to capital. If labor hoarding is included in the model. Specifically. Reverse Causality One approach to dealing with this criticism is to introduce money and banking into the standard RBC model following King and Plosser [129]. Notice that the conventionally measured Solow residual St is related to true technology shock At as ln (St ) = ln (At ) + α ln (Kt ) + (1 − α)[ln (f ) + ln (Nt ) + ln (Wt )] − α ln (Kt ) − (1 − α)[ln (f ) + ln (Nt )] = ln (At ) + (1 − α) ln (Wt ). This can lead to variation in the Solow residual that is not related to changes in productivity or technology. the work effort of each individual. capital Kt . Ht = fNt . Decisions to alter effort levels will be the outcome of a maximizing decision. Yt = At Ktα [fNt Wt ]1−α . where Ht denotes the total hours worked. The measured capital in the Solow residual does not take into account optimal fluctuations in capital utilization rates. and a labor input that reflects variations in work effort following Burnside. In the standard model. Eichenbaum. suppose . These authors observe that transactions services (as provided by money and the banking sector) can be viewed as an intermediate input that reduces the cost of producing output and. can be treated as a direct input into the aggregate production function. Nt . In the absence of variations in work effort. 3. Thus. consider a production function for aggregate output Yt which depends on an exogenous technology shock At . To briefly describe their framework. 0 < α < 1. so that the movements in the Solow residual are not entirely exogenous.56 Business Cycles: Fact. and Wt .

The household chooses an amount of transactions services dht so as to minimize the total transactions costs. By contrast.51) Households are assumed to own the capital stock and to make investment decisions it subject to the resource constraint ct + it ≤ yt + (1 − δ)kt . lt ) .50) where ndt and kdt denote the amounts of labor and capital allocated to the financial sector. In this expression. The production of the intermediate good is given by dt = g (ndt . Households maximize the expected discounted value of utility from consumption ct and leisure lt as ∞ E0 t=0 βt U (ct . where 0 < δ < 1 is the depreciation rate on capital.52) where τ < 0. Likewise. and transactions services to maximize profits on a period-by-period basis. where wt is the real wage and ρt is the rental price of transactions services. This implies a demand for transactions services that can be obtained from the first-order condition ρt = wt τ (dht /(ct + it )) (4. respectively. and dft is the amount of transactions services used in the final goods industry. capital. we .Models of Business Cycles 57 the final good is produced according to the production technology yt = f (kft . nft . 0 < β < 1. dft )φt . τ < 0.53) ∗ as dht = h(ρt /wt )(ct + it ). The time required for this activity is nτt = τ(dht /(ct + it ))(ct + it ). and λt captures technological innovations to the financial services industry. (4. The financial industry is assumed to provide accounting services that facilitate the exchange of goods by reducing the amount of time that would be devoted to market transactions. φt is a shock to production of the final good at time t. (4. (4. firms rent labor. hours allocated to ∗ producing transactions services is nτt = τ ∗ (h(ρt /wt ))(ct + it ). kdt )λt . (4. Finally. wt nτt + ρt dht . Households are also assumed to combine time and transactions services to accomplish consumption and investment purchases.49) where kft is the amount of capital. where h = (τ )−1 . nft is the amount of labor services.

. The increase in output will also stimulate the demand for transactions services by households and firms. is more closely related to output than outside money. commercial credit. equivalently. One 9 See Sims [190]. As a result. or inside money. if interest rates are rising and firms wish to invest more in anticipation of higher expected profits. so firms will wish to finance a greater volume of goods in process. Fallacy and Fantasy require that the household’s time allocated to the different activities sums to 1. Their results are derived from a structural vector autoregression (VAR) with long-run restrictions to identify the alternative structural shocks. the central bank may expand the money supply to keep interest rates from rising too rapidly. investment demand rises and output will also increase in response to the additional hours worked. nt = nft + ndt + nτt . Consistent with the RBC view. Another possibility is that the central bank uses the accommodative monetary policy. responds to the positive productivity shock. the causality runs from the productivity shock to money even though the increase in money occurs before the higher productivity shock. This framework can be used to rationalize the observations regarding money and output. Then. As a consequence. Suppose that a shock to the production of final goods or. or a broad measure of money that includes commercial credit. If the substitution effect of an increase in the marginal product of labor outweighs the wealth effect of the productivity shock. there will be an increase in hours of work. they find that an important source of the money-output correlation is output shocks affecting inside money in the short run.9 During an expansion.5. a positive shock to productivity occurs. Specifically. THE SOURCE OF THE SHOCKS The RBC model has come under much criticism for assuming that technology or productivity shocks are the sole driving force of cyclical fluctuations. inside money. Ahmed and Murthy [3] provide a test of this hypothesis using a small open economy such as Canada to evaluate the impact of exogenously given terms of trade and real interest rates versus domestic aggregate demand and supply disturbances. 3. consumption and leisure of the representative consumer will rise but so will investment demand as consumers seek to spread the extra wealth over time.58 Business Cycles: Fact. Hence.

the relative price of new equipment declined by 3% at an average annual rate while the equipment-GNP ratio increased substantially. In this vein.46) between a de-trended measure of the relative price of capital and investment expenditures. during the period 1950–1990. these authors motivate their analysis by noting that. 10 Altug. Following Gordon [104]. and Demers [10] examine the impact of political risk and regime changes due to changes in the party in power on real investment decisions under irreversibility.5. we discussed the role of government or fiscal shocks in generating observed co-movements of the data. Furthermore. War shocks can also be modeled as a source of cyclical fluctuations (see Barro [27] and Ohanian [169]). 107]. and Krusell [106. Hercowitz. 11 In Greenwood. thereby stimulating output growth. the authors use a growth accounting approach to show that investment-specific shocks can account for 60% of postwar US growth of output per man-hour. there was a negative correlation (−0.10 3. 0 < θ < 1. Although their analysis is not general equilibrium. These authors consider the impact of distortionary taxes on real outcomes. Hercowitz. Demers. Consider a standard RBC model with a representative consumer who derives utility from consumption and leisure as ∞ E0 t=0 βt [θ ln (ct ) + (1 − θ) ln (1 − lt )] . Other papers that have employed the RBC framework with fiscal shocks include Braun [46] and McGrattan [158]. Christiano and Eichenbaum [65] introduce government consumption shocks as a way of breaking the strong and positive relationship between hours worked and productivity implied by the basic RBC model.Models of Business Cycles 59 way of dealing with this criticism is to introduce alternative sources of shocks into RBC models. . and Krusell [106]. Investment-Specific Technological Shocks An important source of shocks is the investment-specific technological change. [107]. This has been examined by Greenwood. the role of investment-specific technological shocks is used to account for cyclical fluctuations. The notion behind this type of change is that production of capital goods becomes increasingly efficient over time. In the previous section.11 In Greenwood et al. they nevertheless show that political events can have non-negligible effects on real economic variables.1.

57) (5. where ae = exp (η)φe (ke /q − κe ke /q)2 /(ke q). and γz and γq denote the average gross growth rates of the two processes.55) where the variable q shows the investment-specific technological change in equipment and δe (h) = b h. Investment in both structures and equipment is also subject to convex costs of adjustment denoted as a = ae + as . αe + αs < 1. φs . the service flow is denoted hke .56) Thus. (5. κe > 0.54) The law of motion for equipment is given by ke = (1 − δe (h))ke + ie q. The production function is given by y = F (hke . ω ω > 1.60 Business Cycles: Fact. Structures evolve according to the standard law of motion: ks = (1 − δs )ks + ii . there is a government which levies taxes on labor and capital income at the rates τl and τk . Equipment is utilized at the time-varying rate h. where z is a measure of total factor productivity. αs . respectively. The shocks to technology and to the efficiency of equipment capital constitute the drivers of the t+1 t+1 model. z) = z(hke )αe ksαs l 1−αe −αs . l . equipment denoted ke and structures denoted ks . They evolve as zt+1 = γz exp (ζt+1 ) and qt+1 = γq exp (ηt+1 ). 0 < αe . ks . (5. Fallacy and Fantasy where ct denotes consumption and lt denotes labor supply. respectively. κs > 0. as = φs (ks − κs ks )2 . investment in equipment is subject to variable rates of utilization and depreciation which are convex in the utilization rate. that follow first-order Markov processes. φe . where ζt+1 and ηt+1 are innovations to productivity and investmentspecific technological change.58) Finally. 0 < δs < 1. hence. (5. (5. The revenue raised by the government is . There is a production technology that depends on labor and the services from two types of capital goods.

respectively. The first parameter is determined using data on the inverse of the relative price of investment goods following Gordon [104]. respectively. ie . the quantity of labor l and the utilization rate h are constant. Equipment. c. this quantity is determined as the ratio of Gordon’s [104] price index of quality-adjusted equipment and the price index for consumption. as . at the rate ge = γq g . Since the inverse of q denotes the relative price of equipment in terms of consumption goods.60) In this analysis. the depreciation rate on structures δs . it also affects the utilization cost of existing equipment this period by lowering the replacement cost of old capital goods. grows faster. and it is set at 12 From the production function. or z q g = γz ge = γz 1/(1−αe −αs ) αe /(1−αe −αs ) γq . In the balanced growth path equilibrium. . (5. ae . denote this rate by g .59) where re and rs denote the rental rate on equipment and structures. (5.Models of Business Cycles 61 returned to the private sector through lump-sum transfers τ.12 A key parameter to be estimated from the data is γq . 1/(1−αe −αs ) (1−αs )/(1−αe −αs ) γq . A higher value of q directly affects the quantity of equipment that will be available to produce output next period. and hence increases its utilization. and the tax rate on wage income τl . the investment-specific technological shock plays a key role. However. The resource constraint and the accumulation equation for structures imply that y. however. we have that g = γ (γ g )αe g αs . and ks all have to grow at the same rate. investment-specific technological change increases the services from old equipment at the same time as it increases the quantity of equipment available for next period. The resource constraint requires that consumption plus investment in the two types of capital equal output net of adjustment costs: c + ie + is = y − ae − as . and w denotes the real wage rate. is . Hence. We note that the economy displays balanced growth as zt and qt grow at the rates γz and γq . The parameters that are determined using a priori information include the average growth rate of the investment-specific shock γq . This yields the government budget constraint: τ = τk (re hke + rs ks ) + τl wl .

g = 1. They conclude that though investment-specific shocks contribute to business cycle fluctuations. the symmetry condition φe = (g /ge )2 φs ≡ φ leaves only the parameter φ to be determined. Fallacy and Fantasy 1. this figure is 32%. Hence. Simultaneously. the parameter φ is set to equalize the consumption/output correlation implied by the model to that in the data.18. αs = 0. they are not the main factor. which tends to raise the stock of equipment next period. δs = 0.40. This leads to increased employment and output. the investment-specific shock affects uses of factors such as labor and capital through changed investment opportunities. This constitutes an upper value for the contribution of the shocks q. This standard calibration exercise yields θ = 0. a low value of the adjustment cost parameter φ is chosen to make the volatility of investment in the model equal to that in the data. Unlike the standard technology shock in RBC models.20.032. . and bh ω = 0. there is a decline in the equipment’s replacement value. which raises utilization of equipment today.97. Given that equipment investment is only 7% of GNP and 18% of the value of output being derived from the use of equipment.53. β = 0. A high value of φ works to reduce investment and to increase the procyclicality of consumption. Furthermore. the investment-specific technological shocks explain around 28% of business cycle fluctuations as captured by the standard deviation of output. αe = 0. the fraction of output directly affected by the shock will typically be small. This provides a lower bound on the contribution of the q shocks because a positive shock to q reduces consumption at the same time as it increases investment. The model is simulated to understand the contribution of the investmentspecific technological shocks in generating cyclical fluctuations.0124. Greenwood et al. By contrast. only the q shock is assumed to be operative.62 Business Cycles: Fact. [107] calibrate the adjustment cost parameter under two alternative assumptions. Hence. The second set of parameters is calculated using information on the long-run behavior of a set of the model’s variables. Likewise. According to the first. They find that under a high value of φ.12. where the parameters b and h cannot be identified separately. For this purpose. the impact of the investment-specific technological shock depends on the quantitative importance of the transmission mechanism. ω = 1. The restriction that adjustment costs are zero along the balanced growth path yields the values κe = κs = g . τk = 0. A positive shock to this variable works by increasing the rate of return to equipment investment.056 and τl = 0.40. For a low value of φ.59.

kt ν1 ν0 > 0. . there is a representative consumer with preferences: ∞ E0 t=0 βt u(ct . Barsky and Kilian [28] analyze the relationship between oil shocks and changes in economic activity. 0 < θ < 1. kt is the stock of capital. δ(ut ) = ω0 utω1 . capital utilization is the channel through which energy shocks enter the model. 0 < β < 1. lt ) . As in the standard RBC framework. As in the previous model. the depreciation of physical capital depends on its utilization rate so that kt+1 = (1 − δ(ut ))kt + it . Rotemberg and Woodford [183] provide a rationale for the role of energy price shocks in a model with imperfect competition. ω0 > 0. the source of large supply-side shocks for the RBC agenda has been sought in oil shocks. The production function is given by yt = (zt lt )θ (kt ut )1−θ . and ut is the rate of utilization of capital. 0 < α < 1. lt ) = t 1−σ 1−σ −1 . We briefly describe her model before concluding this chapter.2. with c α (1 − lt )1−α u(ct .5. ν1 > 1. Finn [92] argues that it is possible to rationalize the role of energy shocks in models with perfect competition.Models of Business Cycles 63 3. σ > 0. where zt is an exogenous technology shock. and ω1 > 1. In her model. ω1 where 0 < δ(ut ) < 1. Energy Shocks From the beginning. By contrast. The new feature in Finn’s analysis is that capital utilization requires energy: et ν0 utν1 = .

then the decrease in the values of et . Fallacy and Fantasy We can solve this equation for the utilization rate ut and substitute it into the production function to obtain yt = (zt lt ) θ 1− ν1 ν1 kt 1 et 1 ν1 ν0 1 ν1 1−θ . This tends to reduce the investment it and the future capital stock kt . Thus. . and lt continue into the future and reduce the future marginal productivity of capital. Finn [92] shows that the quantitative response of value-added and real wages to a shock in energy prices implied by the model is in line with that in the data. Hence. the economy-wide resource constraint is given by ct + it + pt et ≤ yt . and energy.64 Business Cycles: Fact. First. and hence the real wage wt and work effort lt . Thus. The decrease in et further reduces the marginal product of labor. Finn [92] describes qualitatively and quantitatively the impact of increases in the price of energy. If the price shock is persistent. ut . This shows the direct effect of energy. a positive energy price shock operates in a similar way as a negative technology shock. capital. there are two channels for the transmission of energy shocks in the model. but there is also an indirect effect which works through the effect of the utilization rate on the depreciation of capital. Finally. an increase in pt reduces et and ut through the production function. the production of output depends on labor. The indirect effect of the increase in the price of energy on capital’s future marginal energy cost also tends to reduce investment and the future capital stock. where pt is the price of energy.

open-economy business cycle models allow for the role of technology processes in different countries in generating cyclical fluctuations. Kehoe. Backus. One of the aims of the RBC approach has been to determine whether two-country versions of the basic model can account for both the co-movements studied in closed-economy models as well as international co-movements.1. The closed-economy model cannot be used to understand the propagation of shocks across countries or regions. FACTS One of the important areas of research has been to derive the so-called stylized facts of international business cycles. They study the properties of a baseline model against the actual features in the data. nor can it be used to discuss notions of international risk sharing. They consider a two-country RBC model that has the features of the original Kydland–Prescott model and that assumes a single homogeneous good and complete contingent claims for allocating risk. and Kydland [25] consider data on 12 developed countries over a sample period dating from 1960 to 1990. we review some of the facts of international business cycles and discuss models that have been used to rationalize the findings. Backus et al.Chapter 4 International Business Cycles The international business cycle literature provides an extension to the original real business cycle (RBC) approach by allowing for open-economy considerations. In this chapter. Unlike the closed-economy model. including correlations of macroeconomic aggregates across countries and movements in the balance of trade. [26] study 65 . They can be used to model the impact of the ability to trade in goods and to borrow and lend internationally. 4.

respectively. Fallacy and Fantasy the properties of the same model for ten developed countries plus a European aggregate developed by the OECD between 1970 and 1990. Japan. zt∗ denote the technology shocks to the domestic and foreign countries. Thus. The large autoregressive coefficients displayed on the diagonal capture the persistence in observed productivity. 0. where zt . defined as the relative price of imports to exports. ∗ denote the respective innovations to these t shocks. • . the United Kingdom.088 0. The findings can be summarized under two headings: • The quantity anomaly: In the data. and the cross-correlations of productivity are greater than those of consumption. France. The price anomaly: The volatility of the terms of trade. Austria. Backus et al. measured as the Solow residual. the cross-country correlations of output are greater than those of productivity. the theoretical model implies the reverse ranking. By contrast. The individual countries they consider are Australia. 26] estimate a bivariate AR(1) process for the domestic and foreign shocks as zt zt∗ =A zt−1 + ∗ zt−1 t ∗ t . Switzerland. In this regard. ∗ ) = 0. Italy.906 0.00852 and Corr( t . Based on several estimated specifications for the US and the European aggregate. and the United States. Among the important features of their analysis is the estimation of the correlation properties of international productivity shocks using data on estimated Solow residuals. The data are filtered using the Hodrick–Prescott filter.906 with Std ( t ) = Std ( ∗ ) = 0. Germany. is much higher in the data than it is in the theoretical model.088 .66 Business Cycles: Fact. these authors use a symmetric specification with A= 0. the exercise in this literature is to replicate the correlation properties among the different series given the variability and correlation properties of the shocks. [25. The positive t t off-diagonal entries allow for spillover effects of the shock to productivity in one country to have a positive effect on the productivity of the other country.258. Canada. and t .

First. This feature of the model leads to the cross-country correlations of consumption being much higher in the model than in the data. By contrast. investment. but the variability of investment and the trade balance continue to remain . The price anomaly arises because the real exchange rate in the baseline international RBC model is closely related to the ratio of consumption across the two countries. and employment in the theoretical model. what is typically observed in the data are positive comovements in these series across countries. suppose a positive technology shock occurs in one country. [14] confirm the findings of Backus et al. Since investment responds strongly to a positive productivity shock. [25. which leads to a trade deficit in the domestic country. they consider asymmetric spillovers between the US and the European aggregate in which the response of US productivity to shocks to European productivity is smaller than the response of European productivity to shocks to US productivity. This changes the outputinvestment correlation from positive to negative in the domestic country. investment and the trade balance are more volatile in the model than they are in the data. say. Cardia. this ratio displays little volatility. which allows them to obtain standard errors. [26] sample to 20 industrialized countries over the period 1960–2004. This fact leads to negative cross-correlations between output. 26]. and Zimmermann [14] extend the Backus et al. they are negative. They estimate the moments of interest using the generalized method of moments (GMM) estimation. whereas in the model. the increase in investment plus consumption is greater than the increase in output. Backus et al. The international RBC model produces these results because it implies that the incentive is to use inputs where they are most productive. which is accompanied by output and employment increases. Furthermore. The findings of Ambler et al.International Business Cycles 67 Ambler. The two-country frictionless international business cycle model also implies that consumers in different countries can share risk perfectly. [25] also examine a variety of perturbations of their original setup. In other words. though with lower magnitudes. implying that the real exchange rate is also less volatile in the model than it is in the data. and employment are positive and generally high in the data. They also find that the cross-country correlations of output. investment. the US. In a model with perfect capital mobility. In their original analysis. With perfect risk sharing. This leads to a strong investment response in the US. we thus observe a flow of factors from other countries or regions to the US.

Cole [71] and Cole and Obstfeld [72] examine a two-country real version of the Lucas asset pricing model for this purpose (see Lucas [152]). The first of these involves introducing a small trading friction in the form of a transport cost. the cross-country correlation of consumption increases further. as does the positive correlation of consumption across countries. Surprisingly. 26] considers the case with complete contingent claims and perfect risk sharing across different countries. and the crosscountry correlation of output goes from being negative to positive. Agents from both . However. 4.2 and the correlation of the innovations from 0. it is the correlation between technology shocks that leads to any connection between countries. Second.258 to 0. The standard international RBC model studied by Backus et al. Fallacy and Fantasy counterfactually high. Next. In this case. Backus et al. it is important to understand the role of alternative financial arrangements in generating the cross-country correlations. they find that investment and the trade balance become much less volatile. a rise in productivity in the domestic country signals a rise in productivity in the foreign country through the spillover effects. they eliminate all trade in goods and assets by considering an autarky situation. They find that this friction has the effect of reducing the volatility of investment and the trade balance.68 Business Cycles: Fact. THE ROLE OF INTERNATIONAL RISK SHARING Before we continue with other extensions that have been proposed in the international RBC literature. but has little effect on the cross-country correlations of consumption and output. [25.5. They also examine various sorts of trading frictions. they consider large spillovers across countries by changing the off-diagonal elements of the A matrix from 0. the foreign agent increases consumption and reduces investment in anticipation of future income increases. [25] argue that this result is not due to international risk sharing considerations. elimination of trade in state-contingent claims does not help to lower the cross-country correlations of consumption. Cole [71] argued that it is important to examine how alternative financial arrangements interact with the preferences and the production possibilities set to determine the behavior of the real variables. With this change.088 to 0. Hence. but stems from the permanent income hypothesis: when domestic and foreign productivity shocks are correlated.2. They find that the results are very similar to the case with a small transport cost.

agents observe the current realization. Let st ∈ S ⊆ R m denote a vector of exogenous shocks that follows a + discrete first-order Markov process with a stationary probability transition ˆ matrix P with (i. c2 > 0. ¯ ¯ Assumption 2. The representative consumer in country j has preferences over random j j sequences {c1. and strictly concave. To do this. We assume that endowments are stationary in levels. Y1 and Y2 .International Business Cycles 69 countries are identical in terms of preferences and differ only in terms of endowments.1) We have the following assumption. U1 (c1 . 4. c2 ) lim U1 (c1 . y ]. c2 ) = 0.2. Neither good is storable.j = p(st = sj |st−1 = si ). For all c1 . j) element pi. where y > 0 and y < ∞. c2 →0 U2 (c1 . (2.t . where pi.1. strictly increasing. 2 Assumption 2. The utility function U : R+ → R is continuously differentiable. We assume that endowment is a time-invariant function of the exogenous shock. c2 ) = ∞.j . c1 →0 U2 (c1 . Pareto Optimal Allocations We begin by characterizing the Pareto optimal allocations. Country 1 has a random endowment of good Y1 and country 2 has a random endowment of good Y2 . In the ˆ ˆ beginning.t }∞ defined by t=0 ∞ E0 t=0 βt U (c1. j j 0 < β < 1. Define Y ≡ [y. The functions y1 : S → Y and y2 : S → Y are continuous functions that are bounded away from zero. we can examine the solution to a central planning problem which maximizes the weighted utility of agents in the domestic and foreign countries subject to the .t ) . Let π denote the vector of stationary probabilities which satisfy π = P π. There are two goods.1.t . c2.2. c2 ) lim This requirement on the utility function ensures that both goods are consumed in equilibrium. c2.

c2.3) subject to 1 2 ci. φ1 φ2 . c2. c2. c2.t . Let µi (st ) denote the Lagrange multiplier for the resource constraint for good i.t .t ) = µ2 (st ) for j = 1.5) . 2. φj U2 (c1. ci2 (s t )}∞ to maximize the weighted t=0 sum of utilities subject to the sequence of resource constraints ∞ 1 1 2 2 π(s t )βt φ1 U (c1 (s t ). where s t = (st . st−1 .t + ci.t . the social planner’s problem at time t involves solving the static problem 1 1 2 2 max[φ1 U (c1. .t . or 1 1 U1 (c1.t )] (2.t ) 2 2 U2 (c1. the endowment is nonstorable.t ) 1 1 U2 (c1. and there is no investment process. Define ˆ π(s t ) = π(st |st−1 )π(st−1 |st−2 ) · · · π(s1 |s0 )π(s0 ).t .70 Business Cycles: Fact. . c2 (s t )) t=0 st (2. The social planner chooses sequences {ci1 (s t ).2) subject to ci1 (s t ) + ci2 (s t ) = yi (st ).t = yi. The first-order conditions imply that φj U1 (c1. j i = 1.t ) 2 2 U1 (c1. c2. Fallacy and Fantasy resource constraints. 2. c2 (s t )) + φ2 U (c1 (s t ). i = 1.t . Since preferences are time-separable. c2. c2. where ci.t denotes the consumption of good i by residents of country j.t . Let φj denote the Pareto weight for country j. s0 ) denotes the history of the shocks at time t.t ) = µ1 (st ). .t . c2.t . 2. .4) (2.t ) j j j j = = φ2 . φ1 (2.t ) + φ2 U (c1.

t = (1 − δ)y1.t = δy1.t ) = α 1−α (c1. (2.t .9) (2. Also note that a shock to the output of country 1. we will examine the role of alternative financial arrangements that can be used to rationalize the observations.t = (1 − δ)y2.2. This is an implication of perfect risk sharing that arises in a complete contingent claims equilibrium. It is straightforward to demonstrate . It is convenient to assume that preferences are of the Cobb–Douglas variety: U (c1. (2.t = δy2. In what follows.7) (2. c2. the Pareto optimal consumption allocations satisfy 1 c1. Complete Contingent Claims Consider first the behavior of allocations in a competitive equilibrium in which agents can trade claims that pay off contingent on all possible realizations of the income processes in each country. will cause consumption in both countries to increase. we find that the ratio of marginal utilities across agents in different countries is equalized for all states and goods.International Business Cycles 71 Hence.6) Assuming that preferences are of Cobb–Douglas variety. 4.8) (2. These expressions show that consumers in each country consume a constant fraction of output in each period.t )1−ρ 1−ρ .t 1 c2.t .10) where δ= 1+ φ2 φ1 σ 1 and σ = ρ .2.t 2 c2. which also causes the output of country 2 to increase through positive spillover effects. This implication of the model has been shown to lead to the counterfactually high cross-country correlations of consumption implied by the standard international RBC model.t 2 c1.t c2.

denoted as CAj . Consider first the case when agents can trade one-period contingent claims that pay off in each state next period on the consumption good of each country. The current account for each country. and that the competitive equilibrium allocations are Pareto optimal. .t − 2 2 c1. 2.t q(st+1 .t (2.t + st+1 ∈S 2 z1.t − c1.t = p(st ) 2 + c2.t . The sequential budget constraints for residents of country j for j = 1. Ch. it is without loss of generality to consider only one-period contingent claims. these equations indicate that the output produced in country 1 plus the receipts (or payments) on assets denominated in the goods of countries 1 and 2 must be equal to consumption of domestic goods plus imports of foreign goods plus purchases (sales) of contingent assets denominated in the goods of countries 1 and 2. Thus.t 1 1 q(st+1 . 2 are given by 1 1 1 1 y1.t + z1.t + p(st )c2. st ) z1 (st+1 ) + p(st )z2 (st+1 ) 2 c1.72 Business Cycles: Fact.t + p(st )c2. st ) st+1 ∈S 2 z1 (st+1 ) 2 + z2 (st+1 ) . Fallacy and Fantasy that there exists a contingent claims equilibrium for this economy. CA = y2.12) where q(st+1 . j = 1. p(st ) (2.t . 7. Let p(st ) denote the relative price of good y2 in terms of the numéraire good y1 .1 Agents can also trade in the goods of the other country.t p(st ) 2 + c2. is given by 1 1 CA1 = y1.11) y2.t + + p(st ) 2 + z2. 1 For further discussion.t + p(st )z2. st ) is the price of a contingent claim that pays off in each possible j state next period conditional on the state today and zi (st+1 ) are the holdings of country i’s assets by consumers from country j for i. see Altug and Labadie [6]. Given the Markov nature of uncertainty.t = c1.

International Business Cycles

73

and the capital account denoted KAj is given by
1 1 KA1 = z1,t + p(st )z2,t − st+1 ∈S 2 z1,t 1 1 q(st+1 , st ) z1 (st+1 ) + p(st )z2 (st+1 ) , 2 z1 (st+1 ) 2 + z2 (st+1 ) . p(st )

KA2 =

p(st )

2 + z2,t − st+1 ∈S

q(st+1 , st )

We can interpret the budget constraints for each country j by stating that the sum of the current account plus the capital account must be zero: CAj + KAj = 0, j = 1, 2.

In an economy with trade in international assets, a current account deficit (CA < 0) must be balanced with a capital account surplus (KA > 0). This implies that a country which consumes more than it produces must be a net borrower or, equivalently, it must sell more assets to the rest of the world than it purchases from the rest of the world. The market-clearing conditions are given by
1 2 ci,t + ci,t = yi,t , 1 2 zi,t + zi,t = 0,

zi1 (st+1 ) + zi2 (st+1 ) = 0,

st+1 ∈ S

for i = 1, 2. Given the Markov nature of uncertainty, we can define the value function for the problem of each consumer in country j as   j j j j max U c1,t , c2,t V st , z1,t , z2,t = j j  {c ,z }i=1,2
i,t i,t+1


st+1 ∈S

π(st+1 |st )V st+1 , z1,t+1 , z2,t+1

j

j

  

subject to the budget constraint (2.11) (or (2.12)). Let λi (st ) denote the Lagrange multiplier for country i in state st . The first-order conditions with

74

Business Cycles: Fact, Fallacy and Fantasy

envelope conditions substituted in are given by
1 1 U1 (c1,t , c2,t ) = λ1 (st ), 1 1 U2 (c1,t , c2,t ) = p(st )λ1 (st ), 2 2 U1 (c1,t , c2,t ) =

λ2 (st ) , p(st ) i = 1, 2.

2 2 U2 (c1,t , c2,t ) = λ2 (st ),

λi (st )q(st+1 , st ) = βπ(st+1 |st )λi (st+1 ), Simplifying these conditions yields
i i U2 (c1,t , c2,t ) i i U1 (c1,t , c2,t )

= p(st ),

i = 1, 2

(2.13)

and
i i βπ(st+1 |st )U1 (c1,t+1 , c2,t+1 ) i i U1 (c1,t , c2,t )

= q(st+1 , st ),

i = 1, 2.

(2.14)

Thus, we see that the Pareto optimal allocations can be supported in a complete contingent claims equilibrium if λi (st ) = µi (st ) , φi i = 1, 2, where p(st ) = µ2 (st ) . µ1 (st )

In this equilibrium, we find that consumers in each country equate their marginal rates of substitution for each good across all possible current states. If preferences satisfy the Cobb–Douglas assumption, then consumers in each country consume a constant fraction of current output as specified in equations (2.7)–(2.10). This follows from the fact that the competitive equilibrium is Pareto optimal. Hence, p(st ) =
i (1 − α)c1,t i αc2,t

=

(1 − α)y1,t . αy2,t

In this case, the contingent claims contracts are considered to be consistent with the results on consumption for each country. Given the solution for

International Business Cycles

75

the relative price p(st ) and the consumption allocations given in equations (2.7)–(2.10), notice that the solution
i i z1 (st+1 ) = z2 (st+1 ) = 0, i i z1,t = z2,t = 0

with δ = α satisfies the consumers’ budget constraints and the market-clearing conditions. But this is a contingent claims equilibrium in which no assets are traded! In such an equilibrium, it is still possible to price the contingent claims using the expression in (2.14). In the next section, we will demonstrate explicitly that the Pareto optimal allocations can indeed be attained in a noasset-trading equilibrium. Since the price of any financial asset can be obtained as a function of the contingent claims prices, it follows that any financial asset can also be priced in such an equilibrium.

4.2.3. No Asset Trading
Now suppose that there is trade in goods but no trade in international assets. In this case, we are forcing the current account to equal zero in each period. Country 1 and 2’s budget constraints can be expressed, respectively, as
1 1 c1,t + pt c2,t = y1,t , 2 c1,t 2 + c2,t = y2,t .

(2.15) (2.16)

pt

Once again, this is a static problem because there are no assets for intertemporal j consumption smoothing and the endowment is nonstorable. Let µt denote the Lagrange multiplier on the budget constraint of country j. If we assume the Cobb–Douglas functional form for preferences, then the consumption allocations are
1 c1,t = αy1,t , 1 c2,t =

(1 − α)y1,t , pt

2 c1,t = αpt y2,t , 2 c2,t = (1 − α)y2,t .

Fallacy and Fantasy Equilibrium in the two goods markets requires that 1 2 c1.t = y2. ρ −1 1−α α and φ2 = 1 − φ1 .t .t This price can be substituted into the consumption functions above to show that 1 c2. The international ratio of marginal utilities across countries is identical across goods and states.t is positively transmitted to the residents of country 2 by the increase in demand (and hence the relative price) for good y2.t . • . (2.t + c1.t . 1 2 c2.17) (2.t . αy2.t = (1 − α)y1.18) Substitute the consumption functions into the market-clearing conditions and solve for the relative price to show that pt = (1 − α)y1. Hence. Efficient risk sharing can occur despite the lack of financial assets and insurance.76 Business Cycles: Fact.t = y1. Notice that the no-asset-trading allocation is identical to the central planning allocation if α=δ= 1+ or φ = 1+ 1 φ2 φ1 σ .t = αy2. A large negative shock in the amount of a good is similarly transmitted across borders. This exercise illustrates several points: • The absence of international capital mobility does not necessarily imply that the allocation is not Pareto optimal.t . 2 c1. despite the absence of trade in financial assets.t . efficient risk sharing occurs through changes in the relative price of goods. A large and positive shock in the amount of good y1.t + c2.

t = (1 − α)y1. αw denote the expenditure shares under the assumption of Cobb–Douglas preferences and let w be the numéraire good.International Business Cycles 77 • • Notice that we can price financial assets in the model. 2 2 c1. Call this third good w.t in terms of wt and p2.t + wtb .t = p2. αw y2.t in units of wt .t + cw.4. The total consumption of countries 1 and 2 is 1 1 c1. 4. Assume that both countries receive an exogenous and stochastic endowment of good w and let w j : S → W = ¯ [w.t y2. Nonspecialization in Endowments To illustrate the impact of nonspecialization. As Cole and Obstfeld [72] point out.t + pt c2. by 1 1 1 p1.t + p2. 2 2 2 p1.t y1.t denotes the relative price of good y2.t ≤ p1.t c2. α2 .t + (1 − α)y2. Notice that correlation of the total value of consumption of country 1 and country 2 is positive and equal to one. αw y1.t + wta . The equilibrium relative prices satisfy p1. the positive transmission of shocks occurs because countries specialize in the production of goods. Let α1 .t + p2.t ≤ p2.t α2 [wt1 + wt2 ] .t c1.t + pt c2. despite the absence of financial capital mobility.t = αy1. under the assumption that the current account is zero. Agents in countries 1 and 2 have budget constraints given.t denotes the relative price of good y1.t + cw.t + αy2. w] denote the realization of good w in country j.t .t . and can show that real interest rates and real asset returns will be equal for the two countries.2.t c2.t c1. respectively. so that p1.t = α1 [wt1 + wt2 ] .t . Cole and Obstfeld [72] introduce a third good.

are sector-specific shocks. when there are sector-specific shocks. Notice that the relative price of wt may not adjust much if wt1 and wt2 are negatively correlated but the sum wt1 + wt2 fluctuates very little. Shocks to y1. can be shown to satisfy 1 c1. there are gains to asset trading that improve risk sharing and allow diversification. country-specific shocks.t or y2. then it is beneficial to trade equity shares or other forms of financial assets. Call this good n and assume that nj : S → N = [n. by definition.2. This helps to clearly distinguish between country-specific shocks.5. in the absence of trade in financial assets. t t share of the endowment of wt . the country must export more of the good in which it specializes in production to finance the import of good wt . whereas shocks to wt1 .t are. only if the wt varies with st . The intuition is that. Let α1 .t = αw wt1 wt1 + wt2 wt2 wt1 + wt2 + α1 y1. The ratio of marginal utilities of both countries for each good will be equal to a constant across all states.78 Business Cycles: Fact.t . wt2 . n] denotes the realization of good n in ¯ country j. Hence.t = αw 2 c1. is constant as the total t t t t w1 w2 4. If these shocks are not perfectly correlated. Similar expressions can be derived for the consumption of goods y2 and w. which affect all sectors within a country. Under the . a condition for Pareto optimality. ( w 1 +w 2 ) and ( w 1 +w 2 ). Since the current account must always be balanced. under the assumption of Cobb–Douglas preferences. the country with a negative shock to the endowment of wt would like to run a current account deficit by importing wt and borrowing against future endowment. Nontraded Goods Suppose now that the third good is a nontraded good. The shocks to wt1 and wt2 must be perfectly correlated for the allocation with no trade in financial assets to be Pareto optimal. α2 .t . wt2 are not perfectly correlated. where wt1 . Fallacy and Fantasy The consumption of good 1 by agents in countries 1 and 2. αn denote the expenditure shares under the assumption of Cobb–Douglas preferences and let y1 be the numéraire good. + α2 y1. and industry-specific shocks.

j Let zi. 1 − αn α1 = y2. the correlation of national consumption levels may be close to zero. n1 4.t + q2.t z1. 2 denote the shares of good i held by an agent in country j at the beginning of period t.t+1 + q2.t for j = 1. If this sector constitutes a large fraction of consumption. 1 − αn α2 = y1.t . The ratio of marginal utility across countries for a traded good will now depend 1 2 on the ratio nt2 . Unless nt . 1 − αn α2 = y2.t c2.t .t y2. An agent’s budget constraint is z1.t [y1.t [p2.t 2 c2. then the resulting t allocations will not be Pareto optimal. then even if consumption of traded goods is perfectly correlated.t ] ≥ c1. nt are perfectly correlated.t+1 .6. the demands for the goods satisfy 1 c1.t + q1.t + p2. There are gains from international risk sharing through asset trade.t α1 y1.2.t .t + q1.t = 1 c2. 2 and i = 1.t 2 c1. nt .t z2.t ] + z2. Thus.t . nontraded goods provide one vehicle for reducing the large cross-country correlations implied by the baseline model. nt . 1 − αn 1 2 Each country consumes its endowment of the nontraded good. An agent in country 1 holds equity shares that are claims to the endowment stream for good y1 (the domestic good) and claims to y2 (the foreign good). j j j j j j j (2. Trade in Equity Shares We now introduce trade in financial assets. Notice that the correlation of consumption across countries will now depend on the proportion of a country’s consumption that is nontradeable.19) . We now discuss the impact of asset trading and relate it to the model without asset trade discussed earlier.International Business Cycles 79 assumption of balanced trade and Cobb–Douglas preferences.

We commented that we could price financial assets even 1 1 if these assets were not traded. c2. all equity shares are held and the endowment of each good is completely consumed.t .t )q2. c2. If we assume that z1.t = α and 2 2 z1. we can achieve at least two stationary allocations. Agents across countries have identical consumption in this case.t )p2. then portfolio diversification may be achieved by specializing in the holding of equity shares of one country only. 2 so that φj = 1/2 and δ = 1/2. national wealth is equal across countries and agents have perfectly diversified portfolios.t . it does not require the existence of a full set of contingent claims.t+1 + p2.t = 1/2 for j = 1. then portfolio diversification may require that an agent hold equity shares for w 1 and w 2 if the endowment shocks for w are not perfectly correlated. In particular. Hence. depending on the initial distribution of the claims. In the initial model described earlier.t+1 .80 j Business Cycles: Fact. If we introduced trade in equity shares when there is a third good w that is produced by both countries.t+1 ]. In such a world. j j j j j j j j j j j j (2. This simple example. c2.t .t ) = U2 (c1. c2.t+1 . The presence of nontraded goods or lack of specialization in production of a good can affect how much consumption insurance can be achieved through relative price fluctuations. Fallacy and Fantasy Let µt denote the Lagrange multiplier.t = z2.t+1 )[q1. c2.t+1 y2.21) (2.20) (2. U1 (c1. so that zi.t . illustrates an important point. when combined with our discussion of the equilibrium with no trade in financial assets.t . there was no trade in financial assets and specialization in endowments and the allocation was Pareto optimal. 2 and i = 1. Lucas [152] assumes that agents hold identical j portfolios. The first-order conditions are U1 (c1. If these shocks are perfectly correlated.t = 1 − α.t+1 ].22) In equilibrium.t+1 + y1.t = βEt U1 (c1. U1 (c1. unlike the economy in which there is no trade in financial assets. The . then the equilibrium allocation with no trade in financial assets can be achieved. c2.t+1 )[q2. There has been substantial literature on the lack of international portfolio diversification and the degree of international consumption risk sharing.t = z2.t = βEt U1 (c1. The agent maximizes his objective function subject to the constraint. International risk sharing can be achieved through fluctuations in relative prices in the current account and by trade in financial assets.t )q1.

the conclusion on whether there is sufficient or insufficient risk sharing is very sensitive to model specifications. who show that the data are inconsistent with the implications for consumption for the baseline international RBC model. she cannot reject the hypothesis that there is risk sharing. Clearly. By contrast. combined with the assumption that utility is nonseparable in traded and nontraded goods. Baxter and Jermann [36] argue that the failure of international diversification is substantial. as we have noted above. As we have noted above. within the context of their model. Gregory. but that when nonseparability and asset trade restrictions are combined. Devereux.2. 4. Lewis [145] documents that the nonseparability of utility or the restriction of asset trade alone is not enough to explain the risk sharing that we observe. the existence of nontraded goods. Lewis [145] also documents that there is insufficient intertemporal risk sharing in consumption. A more recent paper by Heathcote and Perri [117] extends the Baxter and Jermann model to include more than one traded good. and Smith [81] show that preferences that display a nonseparability between consumption and leisure can help to reduce the consumption correlations implied by the model. whether there exists perfect risk sharing in the data appears to depend on the model specification adopted by the researcher. [25]. Limited Risk Sharing As described earlier.7.International Business Cycles 81 international portfolio diversification puzzle is the notion that investors hold too little of their wealth in foreign securities to be consistent with the standard theory of portfolio choice. optimal behavior would lead to a short position in domestic assets because of a strong positive correlation between the returns to human and physical capital. Empirical evidence on international consumption risk sharing is provided in Backus et al. They find. . Nevertheless. one could ask whether limited risk sharing opportunities among consumers in different countries could help to better reconcile the data with the model. We have also shown above that relative price fluctuations can be a substitute for trade in financial assets in achieving consumption insurance. Their model incorporates human and physical capital and. makes it more difficult to determine the optimal degree of consumption risk sharing. that consumption insurance is available through relative price fluctuations and that these price fluctuations are capable of achieving efficient risk sharing.

82 Business Cycles: Fact. Kollman [136] and Baxter and Crucini [34] examine models where only non-contingent bonds can be traded internationally. In their framework. Output in each country is produced using capital and labor according to the constant-returns-to-scale production function: F (ki (s t−1 ). 125]. and others on debt-constrained asset markets. given a trade structure. Heathcote and Perri [117] examine the implications of a two-country model under alternative assumptions about the financial structure. investment. These authors find that incomplete asset markets help to reduce the cross-country correlation of consumption. where Ai (s t ) is a random shock. Kehoe and Perri [126] consider a model where market incompleteness is obtained endogenously through the introduction of imperfectly enforceable international loans. Alvarez and Jermann [12]. the inability of agents to share risk perfectly and to fully offset the effects of idiosyncratic shocks leads to lower cross-country correlations of consumption and higher cross-country correlations of output. Their analysis involves extending the analysis in Cole and Obstfeld [72] to incorporate an explicit production side. a country can incur international indebtedness only to the extent that such indebtedness can be enforced through the threat of exclusion from future intertemporal and interstate trade. t=0 s t . but the cross-country correlations of output. They show that the model matches the correlations in the data under a financial autarky assumption. Kehoe and Perri [126] consider a standard international RBC model with production and capital accumulation. This analysis builds on the earlier works of Kehoe and Levine [124. Moreover. The preferences of the representative consumer in each country are given by ∞ π(s t )U (ci (s t ). Ai (s t )li (s t )). credit market frictions and restrictions on international capital flows constitute a proposed channel for the propagation of international shocks. the results for the cross-country correlations are obtained only if the productivity shocks are highly persistent and there are very little spillover effects across countries. and hours worked remain counterfactually negative. Kocherlakota [135]. Fallacy and Fantasy In this vein. In these models. li (s t )).

li (s r )) ≥ Vi (ki (s t−1 ). the competitive equilibrium allocations solve the social planner’s problem defined as ∞ π(s t )βt φ1 U (c1 (s t ). li (s r )) ci (s r ) + ki (s r ) ≤ F (ki (s r−1 ). The enforcement constraints are expressed as ∞ βr−t π(s r |s t )U (ci (s r ). s t ). ci (s r ) for r ≥ t to solve ∞ Vi (ki (s subject to t−1 ). the social planner’s problem with enforcement constraints cannot be formulated recursively as a dynamic programming problem. s t ) involves choosing ki (s r ).2 2 F (ki (s t−1 ). An . s t ) denotes the value of autarky from s t onwards. Under complete markets. s ) = max r=t sr t βr π(s r |s t )U (ci (s r ). li (s r ). where Vi (ki (s t−1 ). i=1 (2. The reason is that future decision variables such as ci (s r ) and li (s r ) for r > t enter the current enforcement constraints. l1 (s t )) + φ2 U (c2 (s t ). As Kehoe and Perri [126] explain. t=0 s t (2.23) subject to the resource constraints ci (s t ) + ki (s t ) = i=1.25) where π(s r |s t ) denotes the conditional probability of the history s r given s t and Vi (ki (s t−1 ).24) The innovation in Kehoe and Perri [126] is to formulate a version of this problem that allows for enforcement constraints. l2 (s t )) . r=t sr (2. Ai (s r )li (s r )) + (1 − δ)ki (s r−1 ). Ai (s t )li (s t )) + (1 − δ)ki (s t−1 ) . These require that each country prefers the allocation it receives to the one that it could attain if it were in (financial) autarky from then onwards. r≥t given ki (s t−1 ).International Business Cycles 83 where ci (s t ) denotes consumption of residents of country i.

li (s t )) − Vi (ki (s t−1 ). Mi (s t−1 ) satisfies Mi (s t ) = Mi (s t−1 ) + µi (s t ). Therefore. Since π(s r ) = π(s r |s t )π(s t ). li (s t )) t=0 s t i + µi (s t )[U (ci (s t ). In this expression. plus the term ∞ βt π(s t )µi (s t ) r=t sr βr−t π(s r |s t )U (ci (s r ). the budget constraints for households in each country are expressed as ci (s t ) + ki (s t ) + q(s t )bi (s t ) ≤ wi (s t )li (s t ) + [ri (s t ) + (1 − δ)]ki (s t−1 ) + bi (s t−1 ).24).23). It is beyond the scope of this book to derive a solution for the model with enforcement constraints. Thus. Kehoe and Perri [126] generate solutions for three different economies: complete markets. t ≥0 with Mi (s t−1 ) = φi . li (s r )) − Vi (ki (s t−1 ). see Marcet and Marimon [155]. 2 For other applications. The bond economy stipulates that all intertemporal trades must be implemented with an uncontingent bond.84 Business Cycles: Fact. The Lagrangian function becomes comprised of three terms: the weighted sum of utilities defined in (2. s t )] plus the terms relating to the resource constraints. we will focus on the results of simulating this model and compare it with alternatives that have been obtained in the literature. Mi (s t ) are just the original planning weights plus the sum of the multipliers µi (s t ) along the path s t . Thus. and an economy with enforcement constraints. Fallacy and Fantasy alternative approach involves adding a new pseudo state variable to achieve a recursive formulation of the original problem. the Lagrangian function is written as ∞ βt π(s t ) Mi (s t−1 )U (ci (s t ). a bond economy. s t ) . . the standard resource constraints defined by (2.2 This approach is implemented by defining βt π(s t )µi (s t ) as the Lagrange multiplier on the enforcement constraints.

the cross-country correlations of employment and investment are negative in the model.85 and the off-diagonal elements equal to 0. σ ≥ 0.36. whereas these correlations are the opposite in the data. and a third one allows for high spillover with the diagonal elements equal to 0. the model displays many of the anomalies reported in the literature. where 0 < b < ∞.International Business Cycles 85 where wi (s t ) and ri (s t ) denote the wage rate and the rental rate on capital in country i. AL) = k α (Al )1−α . [25]. There is also a borrowing constraint that requires that ¯ ¯ borrowing cannot exceed some finite bound.36. α = 0.25. net exports and investment are much more volatile in the model than they are in the data. and bi (s t ) is the quantity of the bond by consumers in country i. (2. and δ = 0. b(s t ) ≥ −b. l ) = 1−σ and the production function as F (k. In the economy with enforcement constraints. 0 < γ < 1. q(s t ) is the period t price of an uncontingent that pays off one unit in period t + 1 regardless of the state. the variance of the innovations to the productivity shocks in each country is set at Var( i. the three discrepancies between the model and the data remain. A second parameterization allows for high persistence with the diagonal elements equal to 0. (2. the output and .99 and the off-diagonal elements equal to zero.t ) = 0. The findings from the model echo many of the earlier findings.26) The parameter values that are considered are standard. First. Third. Preferences in each country are taken to be of the form c γ (1 − l )1−γ U (c.15. Similar to Backus et al. According to one parameterization intended to capture substantial persistence. The coefficients of the A matrix. which governs the persistence properties of the shocks and the spillovers between them. In the bond economy. These values are consistent with those assumed by Kollman [136] and Baxter and Crucini [34]. 0 < α < 1.27) 1−σ . and given by β = 0.9 and the off-diagonal elements are set at zero.99. the autoregressive coefficients of the shocks for each country captured by the diagonal elements of the A matrix are set at 0. Second. whereas they are positive in the data.007 and the correlation of the shocks is set at 0. albeit with some minor improvements in the various statistics. Under the complete markets specification. are parameterized in different ways. γ = 0.025. the consumption correlations are significantly higher than the output correlations in the model. σ = 2.

but that the anomalies regarding the cross-correlations of output. Finally. This increases the productivity of both capital and labor in the domestic country. they find that the volatilities of investment and net exports are diminished. as domestic residents save more and also as investment from abroad flows there. we see that the differing responses of the domestic versus the foreign country lead to the negative cross-country correlations of employment and investment. Fallacy and Fantasy consumption correlations are both positive and closer to each other. and employment remain. The capital stock in the domestic country increases. Considering economies with exogenous adjustment costs to investment. Since residents of each country can acquire contingent claims that allow them to smooth consumption across all possible history of shocks. the volatility of net exports and investment has declined dramatically. risk sharing across countries implies that consumption increases in the foreign country. investment. so resources are optimally shifted there. although the cross-country correlation of consumption is greater than that of output. let us consider the impact of a positive productivity shock in the enforcement economy. Consumption increases substantially more in the domestic country because consumption and labor complement each other. Thus. Kehoe and Perri [126] note that the enforcement constraint acts as an endogenous inhibiting factor. Finally. Suppose that the social planner tries to implement . consumption. In the literature. and third. the responses are similar to those for the complete markets economy. exogenous adjustment costs are typically added to inhibit the flow of investment. leading to very volatile investment and net exports. Hence.86 Business Cycles: Fact. they are somewhat dampened. In the foreign country. we observe the high cross-country correlations between domestic and foreign consumption. As discussed earlier. The net flow of investment increases the trade deficit in the domestic country. the cross-country correlations of employment and investment have now switched from being negative to positive. Second. the frictionless international RBC model implies that investment flows to the country with the higher productivity shock. we notice declines in employment and investment. but because of restrictions on asset trading. In the bond economy. The only remaining discrepancy is that net exports are procyclical in the model whereas they are countercyclical in the data. the authors examine the response of the variables in domestic and foreign countries to a positive shock to productivity in the domestic country.

Over time. 4. Hence. and the real exchange rate. the social planner increases the relative weight on the utility of domestic country residents (recall that the social planner’s weight is the inverse of the marginal utility of consumption). to ensure that domestic country residents consume more than foreign country residents. They then compare these findings for the international business cycle to those obtained for data between regions within a country — the so-called “intranational business cycle”. the value of autarky diminishes and the relative weight on the home country also falls. increasing the incentives for default. We describe a few of these extensions in what follows. this is not possible. As a result. the complete markets allocation implies that an increase in output in the domestic country will lead to increases in consumption in both the domestic and foreign countries. the production side. and the presence of additional shocks have been developed. however.3.They argue that the intranational business cycle is a natural environment for thinking about the interactions between economies when there are no trade frictions and when there are not multiple currencies. OTHER EXTENSIONS The international RBC literature has sought to reconcile the findings in the data not only in terms of the cross-country consumption correlations. Hence. but also for the behavior of output. The increased investment flows to the domestic country further increase the value of autarky. The high and persistent productivity shock in the domestic country increases the value of autarky. investment. In the presence of an enforcement constraint. [13] modify the supply side of a two-country model by adding multiple . as the productivity shock dies out. the planner must restrict investment to the domestic country to prevent the default option from being exercised. the planner also builds up the capital stock in the foreign country to ensure that the risk-sharing arrangement between the domestic and foreign countries is not reneged upon. In terms of the behavior of consumption. Hess and Shin [119] re-explore the two international business cycle anomalies emphasized by Backus et al. Furthermore.International Business Cycles 87 the complete markets allocation for this economy. Ambler et al. models that relax assumptions regarding preferences. [25] as well as establish the pattern of productivity growth between industries and countries.

they argue that introducing nontraded goods may be a way of re-establishing the link between a nation’s output and its spending. However. they argue that what is needed is a source of nation-specific shocks that shift demand. overstates the cross-country correlation of aggregate consumption. The notion is that a positive shock to productivity in the domestic country will tend to reduce the investment . they find that the model overstates the negative correlation between the relative price of nontradable to tradable goods and relative consumption of nontraded to traded goods. Stockman and Tesar [200] introduce a nontraded goods sector in each country. a nontradable good. Likewise. The model generates a higher crosscountry correlation of output than standard one-sector models. consisting of the tradable goods of countries 1 and 2. Hence. and greatly overstates the cross-country correlation of tradable consumption. Basu [29]. and leisure. there is no international capital or labor mobility. Instead. They argue that variable capacity utilization has the potential to account for the co-movement of factor inputs across countries. Fallacy and Fantasy sectors and trade in intermediate goods. Preferences are assumed to be nonseparable with respect to a composite good. output in the traded and nontraded goods sector of each country is produced using sector-specific capital and labor which is mobile between sectors. It also predicts cross-country correlations of employment and investment that are closer to the data. or Basu and Kimball [30]). They conclude that an international business cycle model driven solely by productivity shocks cannot account for the findings. We already considered the role of variable factor utilization in reconciling the behavior of procyclical productivity in closed-economy business cycle models (see also Burnside and Eichenbaum [51]. The authors find that the model predicts the correlation between home and foreign output. Baxter and Farr [35] develop a model with variable capital utilization as a way of accounting for the international correlations. They employ preferences that depend on consumption of the tradable goods of countries 1 and 2. Similar to our discussion above. They introduce taste shocks that affect the utility of traded versus nontraded goods and find that this feature brings the data more in line with the implications of the model. In their model.88 Business Cycles: Fact. they find that introducing nontradable goods does not suffice to reduce the cross-country correlation of consumption. and understates the variability of the trade balance. and a nontradable good.

Kt . As we described above. φ > 0.26). Assuming that the household in each country owns the capital stock and makes real investment decisions. Finally. the variance of the innovations is set so that the variance of output in the model exactly matches the volatility . where δ > 0.International Business Cycles 89 flow across countries by leading to an increase in capital utilization rather than investment. where capital services are the product of the capital stock. Baxter and Farr [35] assume that preferences are given by the specification in equation (2. Output in the domestic country is produced using capital services St and labor Lt as Yt = At St1−α (Xt Lt )α . Kehoe and Perri [126] achieve this through an enforcement constraint which requires that the utility under the constrained allocation exceed the utility that country could obtain under autarky after defaulting on its international debt obligations. the budget constraint for the representative household is given by Ct + It + qt Bt+1 ≤ Yt + Bt . which is consistent with earlier studies. real pure discount bonds which have the price qt = [1 + rt ]−1 . The parameterization of the capital utilization function and the adjustment cost function are key aspects of the quantitative analysis. The capital accumulation process displays both costly utilization of capital and adjustment costs: Kt+1 = [1 − δ(Zt )] + φ(It /Kt )Kt . Zt : St = Zt Kt . and φ < 0.258. Similar functions characterize the foreign country. where Bt+1 is the quantity of bonds carried over into the next period. International trade in assets is made solely with one-period. The correlation of the innovations to the technology shocks is set at 0. and the utilization rate. δ > 0. The covariance properties of the technology processes for the domestic and foreign technology shocks are parameterized to be near unit roots with zero spillover effects. 0 < α < 1. many of the puzzles of the international business cycle literature can be addressed successfully by devising a mechanism to limit investment flows across borders.

though they also relate their discussion to the international RBC literature. One of the key findings regarding the role of variable capital utilization is that. McCallum [157] or Helliwell [118].90 Business Cycles: Fact. the volatility of the shocks necessary to match output volatility is substantially reduced. introducing variable capital utilization reduces the volatilities of consumption. Obstfeld and Rogoff [168] provide a broader view of the empirical puzzles in the international macroeconomics literature. Fallacy and Fantasy of the US economy. we would expect capital to flow to regions where the rates of return are highest. in a world of integrated capital markets. Intuitively. 4. In our earlier discussion. Obstfeld and Rogoff [168] argue that many of the puzzles in international macroeconomics may just be specific to the models researchers are using. capital. for example. Specifically. we described the puzzles that arose when trying to match the international RBC model to the data. the cross-correlation of hours and investment become positive. Yet. The Feldstein–Horioka puzzle: According to this puzzle. They enumerate the following discrepancies between data and existing theory as “puzzles”: • Home bias in trade: A number of authors have documented that intranational trade is typically much greater than international trade. When a positive productivity shock occurs in one country. PUZZLES REVISITED The puzzles in the international business cycle literature have been the topic of much research (see. which is not. and exports. long averages of saving rates and investment rates tend to be correlated for the OECD countries (see Feldstein and Horioka [89]). capital utilization rates increase. and also those of hours and employment. Even with a modest cross-country correlation of the innovations. the survey by Baxter [33]). which in turn increase employment. which is desirable. Second. variable capital utilization takes the place of investment flows across countries. • . See. for example. first. The most important impact of variable factor utilization is on the cross-correlations of the factor inputs. In their review. this is accompanied by increases in investment and employment in the other country.4.

whereas the opposite is true in the data. They also note that the pricing puzzles require such features as nominal rigidities of the type that we will consider in the next chapter. first. . and does not have the same weight as. They argue that. In Section 4. perfect risk sharing across countries implies that countries which experience declines in the relative price of consumption should receive large transfers of goods. see Tesar and Werner [203]). the analysis of Obstfeld and Rogoff [168] is an attempt to unify the explanations of a related but disparate set of findings. Obstfeld and Rogoff [168] find that adding transport costs goes a long way towards accounting for the first five puzzles. as we described above. the literature that we have described has revealed a variety of promising directions for future research. The exchange rate disconnect puzzle: This puzzle captures the notion that there exists a relationship between exchange rates and any macroeconomic variable. Clearly. Second.2. say. whereas the other four refer to the behavior of quantities. the international consumption correlation puzzle tends to be specific to a model’s assumptions. The purchasing power parity (PPP) puzzle: The PPP puzzle arises from the fact that shocks to the real exchange rate are very persistent. Backus and Smith [24] show that in an economy with traded and nontraded goods. Meese and Rogoff [161] further show that most exchange rate models forecast exchange rates no better than a naive random walk. The international consumption correlation puzzle: We already discussed the finding that cross-country consumption correlations exceed the crosscountry output correlations in a typical international RBC model. the international portfolio diversification puzzle.International Business Cycles 91 • • • • Home bias in equity portfolios: This reflects the puzzling preference of stock market investors for home assets (for a recent elaboration. In contrast to some of the other contributions that we have discussed. Surprisingly. we discussed various explanations that account for this puzzle — the presence of human capital as discussed by Baxter and Jermann [36] or nontraded goods. the last two puzzles are pricing puzzles.

This page intentionally left blank .

Prototypical New Keynesian models such as that by Rotemberg and Woodford [182] allow for imperfect competition and markups to capture alternative propagation mechanisms in response to technology shocks or shocks to government expenditures. The RBC conclusions regarding the response of hours. see Christiano.Chapter 5 New Keynesian Models The New Keynesian approach has gained significance in the modern macroeconomics literature. output. More recently. Fernald. and other variables to a technology shock have been questioned by empirical results obtained along several different lines. Their approach involves purging the standard Solow residual of factors that might lead to procyclicality. such as variable factor utilization. the response of hours to productivity shocks is negative. taken issue with the notion that prices can adjust costlessly to clear markets. which predicts that hours rise on impact to a positive technology shock. Eichenbaum. and other variables. Critics of the original real business cycle (RBC) approach had.There also exist open-economy versions of the New Keynesian model that have been used for policy analysis 1 For a review and discussion of these models. and Evans [66].1 The behavior of hours and productivity has been a topic of debate. output. On the one hand. New Keynesian theories have revived interest in business cycle models that are capable of producing shortrun economic fluctuations based on the types of forces that Keynes had initially postulated. 93 . Limited participation models also allow alternative mechanisms for the propagation of real and monetary shocks. Gali [96] has argued that in a suitably restricted vector autoregression (VAR) including measures of hours. Basu. and Kimball [32] also present evidence that hours worked and other variables fall in response to technology improvements in the short run. from the onset. This is in contrast to the RBC model. productivity.

and > 1 is the elasticity of substitution among consumption goods. 1 + σn 1 + σu t (1. Ct is a composite consumption good defined as Ct = 1 0 Cit ( −1)/ /( −1) di . and effort levels Ut to maximize ∞ E0 t=0 βt ln (Ct ) + λm ln Mt Pt − H (Nt . THE BASIC MODEL Consider a simple New Keynesian model with monopolistic competition.. 1] consumed in period t. and then discuss the empirical findings in more detail. Fallacy and Fantasy and for understanding the role of monetary shocks. The functional form for H (Nt . see Bowman and Doyle [45]. 2 For a further discussion. Ut ) (1.3) where Cit is the quantity of good i ∈ [0. and variable labor effort due to Gali [96]. respectively. Suppose that a representative household chooses consumption Ct .2) for t = 0. (1. 1. 2. respectively.1.5) ϒt and t denote monetary transfers and profits. . hours worked Nt . Wt and Vt denote the nominal prices of an hour of work and effort. Ut ) is given by H (Nt .1) subject to 1 0 Pit Cit di + Mt = Wt Nt + Vt Ut + Mt−1 + ϒt + t (1. 5. price rigidities. Ut ) = λn λu Nt1+σn + U 1+σu . and Pt = 1 0 1/(1− ) 1− Pit di (1.2 We first describe a simple New Keynesian framework that can be used to rationalize the observations.4) is the aggregate price index. The price of good i is given by Pit . money holdings Mt . . . In this expression. .94 Business Cycles: Fact.

New Keynesian Models

95

The first-order conditions with respect to the household’s problem are given by 1 µt Pit = Ct λm
1 0 ( −1)/ Cit 1/( −1)

di

Cit

−1/

,

i ∈ [0, 1],

(1.6) (1.7) (1.8) (1.9)

1 = µt − βEt (µt+1 ), Mt

λn Ntσn = µt Wt , λu Utσu = µt Vt ,

where µt denotes the Lagrange multiplier on the period-by-period budget constraint. We can solve for µt from the first-order condition corresponding to the consumption choice as µt = 1/(Pt Ct ). Substituting for this variable and simplifying yields Cit = Pit Pt

Ct ,

i ∈ [0, 1], Pt , Ct+1 Pt+1 1

(1.10) (1.11) (1.12) (1.13)

1 Pt = λm + βEt Ct Mt λn Ntσn Ct = Wt , Pt Vt λu Utσu Ct = . Pt
α Yit = Zt Lit ,

In this economy, good i is produced by firm i using the production function (1.14)

where Lit is the quantity of effective labor used by firm i:
θ 1−θ Lit = Nit Uit ,

0 < θ < 1.

(1.15)

Zt is an aggregate technology shock whose growth rate follows an i.i.d. process 2 {ηt } with ηt ∼ N (0, ση ): Zt = Zt−1 exp (ηt ). (1.16)

Consider first the choice of optimal inputs of hours and effort chosen by the firm to minimize its costs subject to the production technology. Let λ be the

96

Business Cycles: Fact, Fallacy and Fantasy

Lagrange multiplier on the technology constraint. The first-order conditions are given by
θα−1 (1−θ)α Uit , Wt = λZt θαNit θα (1−θ)α−1 . Vt = λZt (1 − θ)αNit Uit

(1.17) (1.18)

The ratio of these conditions yields θ Uit Wt = . 1 − θ Nit Vt (1.19)

Notice that the firm will be willing to accommodate any changes in demand at the given price Pit as long as this price is above marginal cost. Hence, the firm chooses the output level Yit = Pit Pt

Ct .

(1.20)

Thus, when choosing price, the firm will solve the problem max Et−1 {(1/Ct )(Pit Yit − Wt Nit − Vt Uit )}
Pit

(1.21)

subject to (1.19) and (1.20). To find the first-order condition for this problem, we will use the last two constraints to solve for the firm’s cost function as 1 − θ Wt Vt Nit Wt Nit + Vt Uit = Wt Nit + θ Vt = = Wt Yit Wt Nit = 1/α θ θZt ((1 − θ)Wt /θVt )1−θ Wt (Pit /Pt )− θZt
1/α /α C 1/α t 1/α

((1 − θ)Wt /θVt )1−θ

.

Using this result, the first-order condition is Et−1 (1/Ct ) αθPit Yit − −1 Wt Nit = 0. (1.22)

Finally, the quantity of money is determined as
s Mts = Mt−1 exp (ξt + γηt ),

(1.23)

2 where {ξt } is a white noise that is orthogonal to {ηt }, with ξt ∼ N (0, σm ).

New Keynesian Models

97

In a symmetric equilibrium, all firms charge the same price Pt and choose the same levels of the inputs and output Nt , Ut , and Yt . Market clearing in the goods market requires that Ct = Cit = Yit = Yt . Finally, equilibrium in the money market requires that the growth rate of the money stock evolve exogenously as Mt /Mt−1 = exp (ξt + γηt ). Next, assume that consumption is proportional to real balances in equilibrium, Ct = Mt /Pt . This, together with market-clearing conditions, implies that Pt Yt = Mt . Using the first-order condition for consumption in (1.11) and the money growth rule in (1.23) yields Ct = λ−1 m = λ−1 m = Mt 1 − βEt Pt Pt Yt Pt+1 Yt+1

Mt 2 2 1 − β exp (σm + γ 2 ση )/2 Pt Mt , Pt

(1.24)

2 2 where = λ−1 [1 − β exp (σm + γ 2 ση )/2]. Next, use (1.12), (1.13), and m (1.19). Taking the ratio of the first two conditions yields

λn Ntσn Wt = . Vt λu Utσu Equating this with (1.19) yields Ut 1 − θ λn Ntσn = . Nt θ λu Utσu Solving for Ut yields Ut = A1/α(1−θ) Nt
(1+σn )/(1+σu )

,

(1.25)

where A = ((1 − θ)/θ(λn /λu ))α(1−θ)/(1+σu ) . Substituting this result into the equation for the production function yields Yt = AZt Nt ,
ϕ

(1.26)

where ϕ = αθ + (1 + σn )α(1 − θ)/(1 + σu ). Using the price-setting rule in (1.22) together with (1.12) and the expression for equilibrium consumption

27)–(1. A positive technology shock defined by ηt > 0 has a permanent positive one-for-one impact on output and productivity and a permanent negative impact on the price level if γ < 1. an increase in ξt causes output and employment to go up for one period and then to revert back to their initial values. This result can be best understood by considering the case of γ = 0. that is. and xt depend only on the current ξt . Hence. though with a one-period lag. and hence a decline in employment will occur.27) (1. The conditions in (1. but the sign depends on whether ϕ < (>)1. a positive technology shock has a negative short-run impact on employment.30) (1.98 Business Cycles: Fact. However. with a positive shock to technology. We note that measured labor productivity responds positively whenever ϕ > 1. real balances remain unchanged in the face of a positive technology shock. which corresponds to the situation of short-run increasing returns to labor. we can show that pt = ξt−1 − (1 − γ)ηt−1 . yt = nt = ξt + γηt + (1 − γ)ηt−1 . Hence. A positive monetary shock defined by ξt > 0 has a temporary impact on output. ϕ ϕ 1 ϕ ξt + 1−γ + γ ηt ϕ (1. producing the same output will require less labor input. given a constant money supply and predetermined prices.28) (1. 1 1−γ ξt − ηt . when there is no accommodating response in the money supply to real shocks. the price level responds one-for-one to an increase in ξt . nt .29) xt = 1 − + (1 − γ) 1 − 1 ηt−1 . ϕ where x = y − n is the log of labor productivity. Finally. demand remains unchanged so that firms will be able to meet demand by producing an unchanged level of output. In such a case. Fallacy and Fantasy and output derived above. This can be observed by noting that the levels of yt . and productivity. . The impact of ξt on labor productivity is also transitory. employment. More interestingly.30) can be used to describe the impact of monetary versus technology shocks.

e. where z and m denote the sequences of technology and non-technology t t shocks. First. the difference-stationary version of the model implies that the response 3The value of the matrix C (L) evaluated at L = 1 gives the long-run multipliers for the model. Christiano. he finds that alternative measures of labor input decline in response to a positive technology shock while GDP adjusts only gradually to its long-run level. This is similar to the approach in Blanchard and Quah [43] for identifying demand versus supply shocks using long-run restrictions on estimated VARs. EMPIRICAL EVIDENCE Gali [96] estimates a structural VAR (SVAR) and identifies technology shocks as the only shocks that are allowed to have a permanent effect on average labor productivity. and McGrattan [63] have challenged the findings derived from so-called SVARs. The identifying assumption is that only technology shocks have a permanent effect on productivity. The orthogonality assumption..New Keynesian Models 99 5.3 Gali [96] estimates this model using postwar US data. and Vigfusson [68] suggest that the standard RBC results hold if per capita hours are measured in log-levels as opposed to differences. Second. respectively. By contrast. Eichenbaum. the long-run impact of a given shock. which can be expressed as the restriction C 12 (1) = 0. Furthermore. together with a normalization.2. Gali [96] finds that variables that have no permanent effects on employment (and which are referred to as demand shocks) explain a substantial fraction of the variation in both employment and output. implies that E ( t t ) = I . technology shocks explain only a small fraction of employment and output fluctuations. Kehoe. i. . they argue that the difference specification for aggregate hours is a priori misspecified because all RBC models imply that per capita hours is a stationary variable. they argue that if the simulated data from a simple RBC model are subjected to a SVAR-based test. Chari. Surprisingly. The SVAR model interprets the behavior of (log) hours nt and (log) productivity xt in terms of two types of exogenous disturbances — technology and non-technology shocks — which are orthogonal to each other and whose impact is propagated through time based on some unspecified mechanisms as xt nt = C 11 (L) C 12 (L) C 21 (L) C 22 (L) z t m t = C (L) t .

durables and residential investment. These findings have. variable factor utilization. They also examine the response of a key set of variables such as output. utilization. They trace the source of the difference to the failure to include a sufficient number of lags in the estimated VAR specifications so as to adequately capture the behavior of hours implied by the underlying theoretical model. and sectoral re-allocation and aggregation effects. the evidence obtained from this approach is robust to long-run identifying assumptions or to the inclusion of new variables in the estimated dynamic system. they advance price rigidity as the major reason for these deviations from the RBC predictions in the short run. They use both standard regression analysis and simple bivariate VARs for this purpose. They also uncover further evidence for the negative response of non-residential investment to such shocks. on the other. Basu et al. and various prices and interest rates to changes in the purified Solow residual. hours worked. non-constant returns to scale. on the one hand.100 Business Cycles: Fact. employment. [32] present evidence that support the SVAR findings by generating a modified Solow residual that accounts for imperfect competition. in their view. They find that purging the standard Solow residual of these effects eliminates the phenomenon of “procyclical productivity”. leads to an erroneous conclusion of the negative response of hours worked to a productivity shock. even though the data underlying this test are drawn from a standard RBC model which implies the opposite response! Despite the controversies surrounding the SVAR approach. if hours worked is considered to be stationary in levels. non-residential investment. then they argue that the confidence bands for the impulse response functions from the SVAR are so wide that the procedure is uninformative about the question at hand. generated substantial controversy and. Their findings corroborate the findings from the SVAR approach regarding the negative response of hours to technology improvements in the short run. However. Unlike the SVAR approach. non-durables and services. Following Gali [96] and Gali and Rabanal [98]. cast further doubt on the ability of the RBC model driven by technology shocks to provide a convincing explanation of economic fluctuations for the major developed countries. This misspecification. Fallacy and Fantasy of hours to a shock to productivity is negative as in Gali and Rabanal [98]. .

More recently. They argue that the large shortfalls in GDP suffered by Argentina in the 1980s can be explained in a simple growth model framework using the observed measure of productivity or the Solow residual. The question then arises whether a small open economy-type RBC model can account for both developed and emerging market business cycles. 176]). On the other hand. Ratfai and Benczur [175. the real business cycle (RBC) agenda has been increasingly applied in emerging market contexts. for example. and income and exports are typically highly volatile. Earlier RBC models for small open economies were developed by Mendoza [162] and Correia.Chapter 6 Business Cycles in Emerging Market Economies In recent years. they have been examining the efficacy of RBC-type models in accounting for these facts. for example. Kydland and Zaragaza [144] also study the behavior of an emerging market economy. Business cycles in emerging markets exhibit different characteristics compared to those in developed economies. consumption varies more than output. Argentina. Arellano and Mendoza [17]). researchers have begun generating business cycle facts for such economies (see. Emerging market business cycles also display a different set of stylized facts relative to developed economies. For instance. the trade balance is strongly countercyclical. On the one hand. The recent literature on the “Sudden Stop” phenomenon has emphasized the large reversals in current accounts and the incidence of capital outflows that have become identified with many recent emerging market economies’ experience (see. namely. Aguiar and Gopinath 101 . and Rebelo [77]. Neves. but their analysis is based on the one-sector optimal growth model.

a stationarity-inducing transformation is used to remove the stochastic trend from all variables as xt xt = ˆ . t t=0 t t The permanent shock to productivity evolves as t t = gt t−1 = s=0 gs . let output be produced according to the Cobb– Douglas production technology as Yt = exp (zt )Kt1−α ( t Lt )α . gt denotes the shocks to the growth rate of productivity. Pancrazi. and µg denotes average long-run productivity growth.4) 2 where { is distributed as i. and Uribe [101] have examined versions of a small open-economy business cycle model with permanent and transitory shocks to account for emerging versus developed economy experiences.d.2) 2 where { z }∞ is distributed as i. with E ( z ) = 0 and Var( z ) = σz . (1. Since productivity and output depend on the cumulation of the shocks gt . g t) ln (gt ) = (1 − ρg ) ln (µg ) + ρg ln (gt−1 ) + g ∞ t }t=0 g t) (1.102 Business Cycles: Fact. here modeled as changes in trend productivity growth.1) where {zt } and { t } represent two alternative productivity processes.3) |ρg | < 1. A SMALL OPEN-ECONOMY MODEL OF EMERGING MARKET BUSINESS CYCLES Aguiar and Gopinath [1] present a small open-economy model of business cycles that allows for permanent and transitory changes to productivity. Fallacy and Fantasy [1] and Garcia-Cicco. Thus. (1. By contrast. |ρz | < 1. g t.1. emerging market economies’ experience can be distinguished by the large number of regime shifts. developed economies typically face stable political and economic policy regimes so that changes to productivity are transitory. In their view.i. t−1 .i. with E ( = 0 and Var( = σg . The shock zt represents the transitory component of productivity. 0 < α < 1. To describe the model. (1. 6. and evolves as a stationary AR(1) process: zt = ρz zt−1 + z t.d.

expected discounted utility remains bounded provided βE (gt ) = βµg < 1. The GHH preferences are defined as ut = (Ct − τ t−1 Ltν )1−σ .5) Here. and that the country can borrow using one-period debt that sells for the price .}}. preferences are given by ut = (Ct (1 − Lt )1−γ )1−σ . .6) The expected discounted utility of the representative consumer can be written as 1 ν ν ˆ E0 {(C0 − τL0 )1−σ + β 1−σ (C1 − τL1 )1−σ 0 1−σ ν ˆ + β2 1−σ (C2 − τL2 )1−σ . Thus. . (1. the elasticity of labor supply is given by 1/(ν − 1). .}} 1 1 1−σ ˆ ν ν = E0 {(C0 − τL0 )1−σ + βg0 (C1 − τL1 )1−σ 1−σ ν ˆ + βg0 E1 {βg 1−σ (C2 − τL2 )1−σ . 1−σ ν > 1. The economy-wide resource constraint is given by Ct + Kt+1 = Yt + (1 − δ)Kt + φ 2 Kt+1 − µg Kt 2 Kt − Bt + qt Bt+1 . 1 where the last line is obtained by substituting recursively and making use of an iterated expectations argument. (1.7) This shows that investment is subject to quadratic costs of adjustment. The first is from the class of so-called GHH (after Greenwood. In the Cobb–Douglas case. (1.Business Cycles in Emerging Market Economies 103 Aguiar and Gopinath [1] consider two types of preferences over consumption and leisure. σ ≥ 0. and Huffman [105]) and the second are standard Cobb–Douglas preferences. 1−σ γ 0 < γ < 1. and the intertemporal elasticity of substitution is given by 1/σ. . τ > 0. . . Hercowitz.} 1 = 1 1−σ ˆ ν ν E0 {(C0 − τL0 )1−σ + βg0 (C1 − τL1 )1−σ 1−σ ν ˆ + E1 {β2 g0 g 1−σ (C2 − τL2 )1−σ .

Fallacy and Fantasy qt . Bt .6) in the case of Cobb–Douglas preferences.5) in the case of GHH preferences and by (1. zt . ˆ Substituting for Ct using the resource constraint. ∂V . Bt+1 . Lt ) gt + φ gt ˆ Kt+1 − µg gt ˆ Kt ∂V ˆ ∂Bt+1 = f (β. gt ) is given by γ(1−σ) βgt1−σ in the former case and by βgt in the latter. gt )Et = 0.Bt+1 max ˆ ˆ ˆ {u(Ct . Lt ) + uc (Ct . gt ) = ˆ ˆ ˆ Ct . Assuming it exists. Lt ) + f (β. The price of debt for the country depends on the quantity of debt outstanding as Bt+1 1 = 1 + rt = 1 + r ∗ + ψ exp −b −1 .104 Business Cycles: Fact. gt+1 )}. and ψ > 0 governs the elasticity of the interest rate to changes in indebtedness. Likewise. b represents the steady-state level of debt. Lt ) is defined by (1. and Lt are ˆ uc (Ct . The optimization is subject to the transformed budget constraint ˆ ˆ ˆ ˆ Ct + gt Kt+1 = Yt + (1 − δ)Kt ˆ φ Kt+1 − − µg gt ˆ 2 Kt 2 ˆ ˆ ˆ Kt − Bt + qt gt Bt+1 .9) (1.Kt+1 . the value function defined in terms of the transformed variables is given by ˆ ˆ V (Kt . ∂Lt .Lt . Bt+1 . ˆ where u(Ct . The model expressed in terms of the transformed or “hatted” variables is solved by using standard recursive methods. f (β. Lt ) ˆ ∂Yt = 0. gt )Et ˆ ˆ uL (Ct . the individual country does not internalize the fact that an increase in borrowing will lead to an increase in the interest rate on those loans.10) (1. the first-order conditions ˆ ˆ with respect to Kt+1 . qt t (1. when choosing how much debt to take on. zt+1 . gt )Et V (Kt+1 . Furthermore. Lt )gt qt + f (β. ˆ ∂Kt+1 (1.8) where r ∗ is the world interest rate.11) ˆ uc (Ct .

The remainder of the parameters. ˆ ˆ ∂V (Kt .Business Cycles in Emerging Market Economies 105 The envelope conditions are given by ˆ ˆ ˆ ∂Yt ∂V (Kt . ˆ ∂ Bt For simplicity. Bt . zt . the resource constraint (1. a subset of the parameters is set at values determined a priori. The solution for the model is obtained by implementing a log-linear approximation to the equilibrium conditions described above.1). including the parameters of the shock processes. gt ) ˆ = −uc (Ct .8).4). In the recent applications. and the equation describing the real interest rate (1. gt ) ˆ = uc (Ct .13) . gt+1  ˆ ˆ ˆ 2 Kt+1 Kt+1 Kt+1 ˆ (Ct − τLtν )−σ = τνLtν−1 = β (1 + r) ν ˆ Et (Ct+1 − τLt+1 )σ . . zt . gtσ exp (zt )αgtα ˆ Kt Lt 1−α (1. The stationary competitive equilibrium satisfies the following conditions: ˆ (Ct − τLt )−σ 1 + φ gt = ˆ Kt+1 − µg ˆ Kt α β Lt+1 α ˆ E (Ct+1 − τLt+1 )−σ 1 − δ + exp (zt+1 )(1−α)gt+1 σ t ˆ gt Kt+1  2  ˆ ˆ ˆ Kt+2 Kt+2 φ Kt+2 + φ gt+1 − µg gt+1 − − µg  . Lt ) + (1 − δ) ˆ ˆ ∂ Kt ∂ Kt + φ gt ˆ ˆ ˆ Kt+1 Kt+1 φ Kt+1 − µg gt − − µg gt ˆ ˆ ˆ 2 Kt Kt Kt 2   . consider only the case with GHH preferences. Lt ). the laws of motion for the shocks (1. (1.7). are estimated using a generalized method of moments (GMM) approach that we describe in detail in the following chapter.2)–(1.12) (1.14) subject to the production function (1. Bt .

Since the number of parameters is less than the number of moment conditions. as is the capital adjustment parameter φ. By contrast. ρg . Furthermore. By contrast. DO SHOCKS TO TREND PRODUCTIVITY EXPLAIN BUSINESS CYCLES IN EMERGING MARKET ECONOMIES? Aguiar and Gopinath [1] consider two versions of their model. ρz . φ). and 2.5 or 5. consumption. the shock to trend productivity can generate consumption booms that tend to appear side-byside with current account deficits in emerging market economies. net exports-to-GDP. The moments that they use for estimation include the standard deviations of log (filtered) income. σg /σz . there are also overidentifying conditions which can be used to test the model’s implications. These findings have been challenged by Garcia-Cicco et al. The most noteworthy finding that emerges from the estimation is that the ratio of shock standard deviations. who argue that the results of Aguiar and Gopinath [1] are due to their use of . The parameters to be estimated are given by (µg . an emerging market economy version estimated using quarterly data for Mexico and a developed country version estimated for Canada between 1980 and 2003. σg .41 for Canada depending on the specification for preferences. They also consider the mean and standard deviation of (unfiltered) income growth as well as the autocorrelations of (filtered) income and (unfiltered) income growth. a positive shock to trend productivity causes consumption to respond more than output in the expectation that income will be higher in the future. In particular. is 0. the autocorrelations of transitory shocks are roughly similar. as well as the correlations of the latter three with output. The authors argue that differences in the magnitude of shocks to trend productivity can account for some of the salient features of business cycles in emerging market economies. investment.25 or 0. these results can also explain why consumption is more volatile than income in economies where shocks to productivity growth are more important than transitory shocks. they show that a positive transitory shock to productivity reduces consumption in anticipation of lower income in the future. Thus. This leads to a large trade deficit which tends to persist for a considerable period of time (16 quarters).4 for Mexico. σz . This yields 11 moment conditions.2. This ratio captures the importance of shocks to trend productivity.106 Business Cycles: Fact. Fallacy and Fantasy 6. [101].

whereas in the data these quantities are roughly equal. the second result suggests that it overemphasizes them. First. and find that the consumption-smoothing motive in response to a transitory shock dominates the anticipatory effect of consumption to a permanent productivity shock. Whereas the first result suggests that the estimated model does not emphasize the role of permanent shocks sufficiently. They consider a model that is very similar to the one in Aguiar and Gopinath [1]. They determine this empirically. the permanent shock is estimated to be more volatile and persistent than the transitory shock. However. [101] find that the overidentifying restrictions of the model are rejected. investment growth. much of the business cycle literature for developed countries has concentrated on the post-World War II period. Second. in contrast to what is observed in the data. and. the correlations of output growth with consumption growth. and the unconditional mean of output growth.and second-order autocorrelations of output growth. [101] use annual data for Argentina over the period 1913–2005. Garcia-Cicco et al. consumption growth in the estimated model is calculated to be less volatile than output growth. By contrast. As we described earlier. The authors also find that the trade balance-to-output ratio is estimated to be around four times as volatile as output growth. and estimate the parameters of the stochastic processes for the permanent and transitory shocks using 16 moment conditions. in terms of the parameter estimates. this choice of sample period does not seem out of line. In particular. investment growth is found . [101] argue that a similar finding is not true for an emerging market economy such as Argentina. and the trade balance-to-output ratio. which is not significantly different from zero. consumption growth. In particular. The standard deviations of the shocks and the autoregressive parameter for the permanent shock are estimated precisely. unlike Aguiar and Gopinath [1]. and less volatile than output growth if the opposite is true. Given the evidence that business cycles have moderated during this period.Business Cycles in Emerging Market Economies 107 a short sample to estimate low-frequency movements in productivity. Garcia-Cicco et al. and the trade balance-to-output ratio. they consider the variances and first. investment growth. As we described above. this is not the case for the autoregressive coefficient for the transitory shock. Garcia-Cicco et al. Third. they show that output fluctuations in the post-World War II period are as large as those in the pre-World War II period. consumption growth will be more volatile than output growth if permanent shocks are more important than transitory shocks.

as households perceive the income increase to be permanent. if the shocks are transitory. the trade balance-to-output ratio displays a near random walk behavior even though the estimated autocorrelation function in the data is downward-sloping. consumption in the model follows a near random walk even though output and investment become stationary variables. suggesting in this case that neither source of shocks is sufficiently volatile. with a constant world interest rate. In particular.108 Business Cycles: Fact. they argue that some of their findings that we described above point to the importance of shocks that are different from productivity shocks. their findings cannot be used to disentangle which of these factors is responsible for the failure of the model to replicate the observations. If the shocks are permanent. Hence. they also argue that their results are derived under the joint hypothesis of business cycles driven by productivity shocks and the propagation mechanisms of the standard RBC approach. Garcia-Cicco et al. The authors also show that the random walk behavior of the trade balance-to-output ratio remains. Fallacy and Fantasy to be insufficiently volatile. By contrast. regardless of whether shocks to productivity are permanent or transitory. The authors show that the estimated model also fails to replicate the autocorrelations of output growth and the trade balance-tooutput ratio. However. consumption increases in response to an innovation to output in roughly the same magnitude as output. Hence. [101] conclude that a pure RBC model driven by permanent and/or transitory exogenous shifts in productivity does not provide an adequate explanation of business cycles in emerging markets. In particular. the trade balance is unaffected and the trade balance-tooutput ratio inherits the behavior of output. . the behavior of the trade balance-to-output ratio again follows a near random walk because in this case.

Yet one of the most important aspects of the debate regarding the RBC approach lies in the use of calibration as a way of matching the model to the data. we described some criticisms leveled against the assumptions of the real business cycle (RBC) framework. One of the most popular approaches has been based on the dynamic factor model. Their approach may be viewed as an extension of the standard RBC approach. which seeks to describe the joint cyclical behavior of a key set of time series in terms of a lowdimensional vector of unobservable factors and a set of idiosyncratic shocks. Watson [209] extended this approach to derive measures of fit for an underlying economic model. The recent business cycle literature has witnessed the use of a variety of techniques regarding the empirics of business cycles. Altug [5] estimated an unobservable index model for a key set of aggregate series based on a modified version of the Kydland and Prescott model [141] using maximum likelihood. This model was initially proposed by Sargent and Sims [185] for describing cyclical phenomena. An alternative approach was proposed by Christiano and Eichenbaum [65]. Amongst the most prominent of these assumptions is the assumption of perfect price flexibility. In her original contribution. and also provide an overview of the estimation versus calibration debate. In this chapter.Chapter 7 Matching the Model to the Data In the previous chapters. who used the generalized method of moments (GMM) approach (see Hansen [113]) to match a selected set of unconditional first and second moments implied by their model. We also discussed variations of the model that could be used to account for specific correlations in the data. we will study alternative approaches for matching the implications of a theoretical model with the data. which assesses the adequacy of the model based on the behavior of the relative variability and co-movement of 109 .

110 Business Cycles: Fact. . the variances and autocovariances of the observed series {wt } can be decomposed in terms of the variances and ˜ autocovariances of a low-dimensional set of unobserved common factors and . and νt is an n × 1 vector of idiosyncratic shocks ˜ that are mutually uncorrelated and uncorrelated with common factors. Under these assumptions. To describe how to formulate unobservable index models.3) Both the common factors and the idiosyncratic factors may be serially correlated.t νi. . that is. covariation among the elements of wt can ˜ arise because they are functions of the same common factor or because they are functions of different factors which are themselves correlated at different leads and lags. According to this model. E (f˜t f˜t−s ) = 0 for t = s and E (˜ i.1. j. Finally. . More precisely. 7.t νj.t−s ) = 0 for all ν ˜ i. Fallacy and Fantasy a small set of moments. DYNAMIC FACTOR ANALYSIS Dynamic factor analysis seeks to describe the joint cyclical behavior of a key set of time series in terms of a low-dimensional vector of unobservable factors and a set of idiosyncratic shocks that are mutually uncorrelated and uncorrelated with the factors. A k-factor unobservable index model for wt is given by ˜ ∞ wt = ˜ s=−∞ ˜ ˜ H (s)f˜t−s + νt . we require that ˜ E (f˜t νi. let {wt }∞ denote an n-dimensional mean zero. . Canova [56.2) (1.t ) = 0 ν ˜ for i = 1. (1. covariance stationary stochastic ˜ t=0 process used to describe observations on the (possibly de-trended) values of a set of variables.1) ˜ ˜ where {H (s)}∞ s=−∞ is a sequence of (n × k)-dimensional matrices. ft is a k × 1 vector of common factors. 57] discusses alternative approaches for conducting statistical inference in calibrated models. (1. n for i = j. t = s. Bayesian estimation of the so-called dynamic stochastic general equilibrium (DSGE) models provides an alternative to incorporating prior beliefs about various parameters that formalize some of the practices in the calibration approach.t ) = 0 E (˜ i.

Substituting for Rw (r) into (1. n.Matching the Model to the Data 111 the idiosyncratic shocks. 0. be the autocovariance function of {wt }. 1.1). . .4) yields ∞ ∞ Sw (ω) = r=∞ s=−∞ ∞ ˜ H (s) ∞ v=−∞ ˜ Rf (r + s − v)H (v) exp (−iωr) + r=∞ Sν (r) exp (−iωr) ∞ 2 r=−∞ Rxl xh (r) < ∞ for each l . Under the assumptions ˜ underlying (1. Let Rw (r) = E (wt (wt+r ) ). in general. 1This function is well-defined as long as . However. . h = 1. . since Rxl xh (r) = Rxl xh (−r). . . notice that offdiagonal elements of Sw (ω) are. . ˜ ˜ r = .4) Assuming that Sw (ω) is non-singular at each frequency ω. .1 (1. = s=−∞ An alternative representation of the autocovariance function is provided by the spectral density function ∞ Sw (ω) = r=∞ Rw (r)e −irω . . complex numbers. the diagonal elements of Sw (ω) are real. . −1. |ω| ≤ π. ∞ Rw (r) = E s=−∞ ˜ ˜ H (s)f˜t−s + νt ∞ v=−∞ ˜ ˜ H (v)f˜t+r−v + νt+r =E ˜ ˜ ˜ · · · + H ( − 1)f˜t+1 + H (0)f˜t + H (1)f˜t−1 + · · · + vt ˜ ˜ ˜ ˜ v × · · · + H ( − 1)f˜t+r+1 + H (0)f˜t+r + H (1)f˜t+r−1 + · · · +˜t+r ∞ = s=−∞ ∞ ˜ H (s) ˜ H (s) ∞ v=−∞ ∞ v=−∞ ˜ E (f˜t−s f˜t+r−v )H (v) + E (˜ t νt+r ) ν ˜ ˜ Rf (r + s − v)H (v) + Rν (r).

He discusses a classical measurement error case. The use of the dynamic factor model in business cycle analysis dates back to the work of Sargent and Sims [185]. who derives an unobservable index model for a key set of aggregate series by augmenting the approximate linear decision rules for a modified version of the Kydland and Prescott [141] model with i.d.d. Altug also uses this representation to estimate the model using maximum likelihood (ML) estimation in the frequency domain. The common factor is identified as the innovation to the technology shock. which describe how the common factors affect the behavior of the elements of wt at all leads and lags.112 ∞ Business Cycles: Fact. it is not possible to identify ˜ the common factors with different types of shocks to the economy. As these authors observe. ˜ ˜ where H (ω) denotes the Fourier transform of H (s). Another well-known application of this approach is due to Altug [5].i. Hence. Fallacy and Fantasy ∞ = s=−∞ ˜ H (s) exp (−iωs) ∞ Rw (u) exp (−iωu) u=−∞ × z=−∞ ˜ H (z) exp (−iωz) + Sν (ω) ˜ ˜ = H (ω)Sw (ω)H (ω) + Sν (ω). as well as a case with orthogonal prediction errors. Sargent [184] also employs the device of augmenting a singular model with additional idiosyncratic shocks. the dynamic factor model provides decomposition at each frequency that is analogous to the decomposition of variance in the conventional factor model. as in Altug [5]. Also. measurement errors or idiosyncratic components not captured by the underlying RBC model. dynamic factor analysis may be linked to the notion of a “reference cycle” underlying the methodology of Burns and Mitchell [50] and the empirical business cycle literature they conducted at the National Bureau of Economic Research. The dynamic factor model can be estimated and its restrictions tested across alternative frequencies using a frequency domain approach to time series analysis. The unrestricted version of the dynamic factor model does not place restrictions on ˜ the matrices H (s). and the idiosyncratic shocks are interpreted as i.i. Unlike the unrestricted factor model which can be estimated frequency by frequency. The restricted factor model makes use of the cross-equation restrictions across the linear decision rules implied by the original model. this model must be estimated jointly across . error terms.

The question at hand is whether the data generated by the model are able to reproduce the behavior of the observed series. Also. denotes the stochastic process generated by some underlying Here. In the data. when the restrictions of the underlying model are imposed. the vector of variables yt corresponds to the empirical counterpart of xt . define the n × 1 vector of 2 For further discussion of maximum likelihood estimation in the frequency domain. Unlike the approach adopted by Altug and Sargent. The ACGF for yt is similarly denoted Ay (z). . investment in equipment.Matching the Model to the Data 113 all frequencies because the underlying economic model constrains the dynamic behavior of the different series as well as specifies the nature of the unobserved factor. Measures of Fit for Calibrated Models Watson [209] extended the approach in Altug [5] and Sargent [184] to derive measures of fit for an underlying economic model. his approach involves choosing the correlation properties of the error process between the actual data and the underlying model such that its variance is as small as possible. let xt denote an n × 1 vector of covariance stationary random variables. not to describe a statistical model that can be used to conduct statistical tests. model. Watson’s analysis does not depend on assuming that the unobserved factors and the idiosyncratic shocks are uncorrelated.1. and Ax summarizes the unconditional second-moment properties of this model. investment in structures. the model cannot explain the cyclical variation of the observed variables.2 Altug [5] initially estimates an unrestricted dynamic factor model for the level of per capita hours and the differences in per capita values of durable goods consumption. the joint process for the data and the error is introduced to motivate goodness-of-fit measures. and aggregate output. Define the autocovariance generating function (ACGF) for xt by ∞ Ax (z) = r=−∞ E (xt xt−r )z r . For this purpose. However. She finds that the hypothesis of a single unobservable factor cannot be rejected at conventional significance levels for describing the joint time series behavior of the variables. see Hansen and {xt }∞ t=0 Sargent [114]. 7. To describe Watson’s approach. Instead.1.

In the standard dynamic factor model. Hence. and ut are assumed to be serially uncorrelated scalar random variables.5) where Axy (z) is the joint ACGF for xt and yt . If.7) ˆ a given model fits that data. a lower bound may be deduced for the variance of ut without imposing any restrictions on Axy (z). in addition. let Ay (z) denote the population autocovariance function and Ay (z) ˆy (z) is ascribed to sampling denote its estimated counterpart. xt . Ay (z) = Ax (z). This implies a version of a classical errors-in-variables approach. 4This is similar to the interpretation provided by Altug [5]. Nevertheless. the assumption regarding Axy (z) is that it is zero.114 Business Cycles: Fact.4 However. the error term ut is the approximation error in describing the observed data with the underlying economic model. or |σxy | ≤ σx σy : 2 2 2 min σu = σy + σx − 2σxy s. notice that Ay (z) can be determined from the data. since Axy (z) is unknown. This bound is calculated by choosing Axy (z) to minimize the variance of ut subject to the constraint that the implied joint ACGF for xt and yt is positive definite. To illustrate this approach. define ut by the relation ut = yt − xt . (1. yt . Axy (z) = 0.3 To continue. some additional assumptions are needed to operationalize this approach.t. let us consider two cases.6) (1. and Ax (z) from the model. In this case. it is the size of the sampling error that is used to judge whether . σxy (1. 3 In a standard goodness-of-fit approach. However. the difference between Ay (z) and A error.5 In the first case. Then. which implies that the ACGF for ut can be expressed as Au (z) = Ay (z) + Ax (z) − Axy (z) + Ayx (z). To calculate Au (z). |σxy | ≤ σx σy . The problem is to choose σxy to minimize the variance of ut subject to the constraint that the covariance matrix of xt and yt remains positive definite. and the size of the sampling error can be determined from the data-generating process for yt . then the sampling error in estimating Ay (z) from actual data can also be used to determine how different Ay (z) is from Ax (z). Fallacy and Fantasy errors ut that are required to reconcile the ACGF implied by the model with that of the data. 5 We omit the third case that Watson considers because it requires results for the Cramer representation of stationary time series. the assumption of pure measurement error which is uncorrelated with the true variable is not appropriate. in Watson’s framework.

The matrices U and V are defined from the singular value decomposition USV of Cy WCx such that U is a n × k orthogonal matrix (with U U = Ik ).u . As a consequence. Watson [209] uses this methodology to generate goodness-of-fit measures for a standard RBC model with a stochastic trend in technology. is given by xy = Cx VUCy .10) where = Cx UVCy−1 . He log-linearizes . The covariance matrix of ut is given by u = y + x − xy + yx . we consider a transformation that allows us to determine the size of ut . this result corresponds to the one in the scalar case. the joint covariance matrix of (xt . yt ) is singular. j element of . tr( u ) = n i=1 ij. An alternative approach is to minimize the weighted sum of the variances as tr(W u ).Matching the Model to the Data 115 It is easy to see that the solution for this problem is to set σxy = σx σy . (1. given the properties of the U and V matrices. As a 2 consequence. (1. we cannot minimize the variance of ut directly. Plosser. Furthermore. Watson [209] provides a solution for the case in which x has rank k ≤ n so that the number of variables is typically less than the number of shocks. In this case. and S is a k × k diagonal matrix. he considers the model in King. The problem in this case is to choose xy to minimize tr(W u ) subject to the constraint that the covariance matrix of (xt . where ij.8) where γ = σx /σy . yt ) is positive semidefinite. where W is an n × n matrix. One convenient transformation is the trace of u . Specifically. Instead. For k = 1. Now suppose that xt and yt are serially uncorrelated random vectors with covariance matrices x and y . An implication of this result is that. 132]. and Rebelo [131.9) In this expression. Cx is the n × k matrix square root of x as x = Cx Cx and Cy is the n × n matrix square root of y . V is a k × k orthonormal matrix (with VV = Ik ). which minimizes the weighted sum of the variances of the elements of ut subject to the constraint that this matrix is positive semidefinite. xt and yt are perfectly correlated so that xt = γyt . σu = (σy − σx )2 at the minimum. xt can be represented as xt = yt . as in the scalar case. (1.u denotes the i. Proposition 1 in Watson [209] demonstrates that the unique matrix xy .

However. see also Altug [4]. Reichlin. His approach allows for a decomposition by frequency of the variance in each observed series. 7. and Sala [103] show how more general classes of equilibrium business cycle models can be cast in terms of the dynamic factor representation. suggesting that the stochastic trend properties of the model do not hold in the data. their results also indicate that sector-specific shocks are needed to explain the variance of the series. . Fallacy and Fantasy the first-order conditions to obtain the solution for log-differences of output.6 He finds that the biggest differences between the spectra for output. They examine the behavior of four-digit industrial output and productivity for the US economy for the period 1958–1986 and find evidence in favor of at least two economy-wide shocks. that is.2. the variance explained by the model. both having a long-run effect on sectoral output.116 Business Cycles: Fact. and the unobserved factor model can be estimated by using equation-by-equation ordinary least squares (OLS). Watson’s analysis shows that focusing only on a small subset of moments implied by the model can be misleading because the RBC model is unable to reproduce the typical spectral shape of economic time series. Based on a law of large numbers argument. models for which the number of exogenous shocks is less than the number of series. The model also implies that the number of hours worked is stationary whereas there is considerable power at the low frequencies for this series in the data. Other Applications Forni and Reichlin [93] use the dynamic factor model to describe business cycle dynamics for large cross-sections. They also describe how to derive impulse response functions for time series models which have reduced rank. and the number of hours worked.1. Giannone. and the error in reconciling the model with the data. they show that the number of common factors can be determined using the method of principal components. the economy-wide shocks can be identified using structural vector autoregression (SVAR) techniques. investment. 6 A comparison of the spectra implied by the model and those generated by actual data also figured in early versions of Altug [5]. consumption. consumption. and investment occur at frequencies corresponding to the business cycle periodicities of 6–32 quarters.

Some of the parameters included in 1 may be the depreciation rate of capital. the depreciation rate δ is set to reproduce the average depreciation on capital as E δ− 1− kt+1 it − kt kt = 0. g . ρ. δ. More generally. θ. Since there is a stochastic trend in this economy arising from the nature of the technology shock process. λ. the elements of restrictions E H1t ( 1) satisfy the unconditional moment = 0. the share of capital satisfies the Euler equation E β−1 θ yt+1 kt+1 +1−δ 1 ct ct+1 = 0.11) . Proceeding in this way. (2. GMM ESTIMATION APPROACHES Other papers have employed nonlinear estimation and inference techniques to match equilibrium business models with the data. the deterministic steady states are derived for the transformed variables. technology. Christiano and Eichenbaum [65] derive a solution for the social planner’s problem by implementing a quadratic approximation to the original nonlinear problem around the deterministic steady states. Likewise. σλ ) . σµ . and the share of capital in the neoclassical ¯ production function. given data on gross investment it and the capital stock kt+1 .2. and the exogenous stochastic processes. θ. let 1 denote a vector of parameters determining preferences. For example. As in the standard RBC approach.Matching the Model to the Data 117 7. 1 = (δ. To describe how their approach is implemented. Their estimation strategy is based on a subset of the first and second moments implied by their model. which assesses the adequacy of the model based on the behavior of the relative variability and co-movement of a small set of time series. Christiano and Eichenbaum [65] consider the hours-productivity puzzle and use the generalized method of moments (GMM) approach (see Hansen [113]) to match a selected set of unconditional first and second moments implied by their model. Their approach may be viewed as an extension of the standard RBC approach. the parameters in 1 are estimated using simple first-moment restrictions implied by the model. γ.

σn /σy/n ] (2. the average productivity of labor. F ( ) = 0.12) (2. nt . n) and σn /σy/n .16) Christiano and Eichenbaum [65] are interested in testing restrictions for the correlation between hours and productivity. gt . 2 2 E nt − σn = 0. f2 ( 1 )] (2. (2. labor hours. σn /σy/n . gt . it .118 Business Cycles: Fact. The unconditional second moments are obtained through simulating the model’s solution based on the linear decision rules obtained from the approximate social planner’s problem. The restrictions of the model can be summarized as E H2t ( 2) = 0. (2. public consumption. (2.13) (2. let f( 1) = [f1 ( 1 ).20) . n). n) − (y/n)t nt = 0.19) Under the null hypothesis that the model is correctly specified. Fallacy and Fantasy 2 consist of the standard RBC second-moment restrictions as The elements of E yt2 (σx /σy )2 − xt2 = 0.18) . Also. private investment. 2 ] as the k × 1 vector containing the true values of the parameters and second moments for the model. and the relative variability of hours versus average productivity. 2 σn /(σn /σy/n ) corr(y/n. it . corr(y/n. (2. For any parameter vector 1 . let A be a 2 × k matrix of zeros and ones such that A and F( ) = f ( 1) − A = [corr(y/n.17) represent the model’s restrictions for corr(y/n. where ct denotes private consumption. and (y/n)t . n). they use a Wald-type test based on the orthogonality conditions implied by the relevant unconditional moments. Define = [ 1 .15) E (y/n)2 σn /σy/n t E 2 − nt = 0. 2 x = ct . To do this.14) (2.

we will deal with a variety of criticisms aimed directly at the calibration approach and some suggested alternatives to it. and average productivity. He took issue with the calibration approach by showing that the model’s implications for the variance of output relative to its value in the data are heavily dependent on assumed values for the underlying parameters for the technology process. calibrated or otherwise. Aguiar and Gopinath [1] and Garcia-Cicco. Christiano and Eichenbaum [65] estimate the parameters of the model using GMM and examine the various unconditional second moments implied by the model. once we quantify the uncertainty in model predictions arising from uncertainty about model parameter values. Even if 7 See Kydland and Prescott [142. In a highly suggestive analysis. Using data on private consumption. Kydland and Prescott [141] initiated this debate in the modern business cycle literature. Since then. there have been ongoing discussions on both sides of the issue. 7.21) is asymptotically distributed as a χ2 random variable with two degrees of freedom.7 In this section. the authors show that the statistic J = F ( ˆ T ) Var[F ( ˆ T )]−1 F ( ˆ T ) (2. Letting test statistic for the second-moment restrictions is based on the distribution of F ( ˆ T ) under the null hypothesis. As we described earlier. investment. in fact. the GMM approach has been used in other recent applications. and Uribe [101] employ this approach in their analysis of business cycles in emerging markets. THE CALIBRATION VERSUS ESTIMATION DEBATE The crux of the recent business cycle debate centers on how a given theoretical model should be matched with the data. Eichenbaum [82] asked whether RBC analysis.Matching the Model to the Data 119 In practice. there is sampling error in estimating from a finite data set ˆ T denote the estimated value of . our view of what the data is [sic] telling us is affected in a first-order way. the containing T observations. . aggregate hours. 143] for a further discussion and defense of their approach. For example. Pancrazi.3. In his words: Indeed. government expenditures. constituted “wisdom or whimsy”. Using this distribution.

The autocorrelation function from the model is generated by simulating the model 1000 times. the authors use the SVAR technique developed by Blanchard and Quah [43]. 7. Since the single technology shock model is singular. they use the two-shock version of the RBC model proposed by Christiano and Eichenbaum [65]. To estimate impulse response functions from the data.1.120 Business Cycles: Fact. The answer could be 70% as Kydland and Prescott [142] claim. As part of their diagnostics. but the data contain almost no evidence against either the view that the answer is really 5% or that the answer is really 200%. while technology shocks almost certainly play some role in generating the business cycle. its power spectrum. for example. They consider a two-variable VAR with output and hours worked (or consumption). The striking results in Cogley and Nason [70] show that the autocorrelation function . Fallacy and Fantasy we do not perturb the standard theory and even if we implement existing formulations of that theory on the standard postwar sample period and even if we use the stationary inducing transformation of the data that has become standard in RBC studies — even then the strong conclusions which mark this literature are unwarranted. They assume that the technology shocks are difference-stationary. whereas government shocks evolve as a persistent AR(1) process. they consider the autocorrelation function for output. They compare the autocorrelation function and the impulse response function in the data with those generated by the model using generalized Q statistics. Eichenbaum [82] also presented evidence to show that the performance of the standard model as well as modifications that allow for labor hoarding. deteriorate considerably when a break in the sample is allowed for. The Dynamics of Output Cogley and Nason [70] examine the dynamics of output as a way of determining the efficacy of the RBC model in describing the data. and identify the technology shock under the assumption that it has a permanent effect on output. Such a break accounts for the slowdown in productivity growth in the US that occurred in the late 1960s. and impulse response functions in response to shocks. there is simply an enormous amount of uncertainty of just what percent of aggregate fluctuations they actually do account for. which includes shocks to productivity and government consumption as well as the indivisible labor feature of Hansen [112] and Rogerson [179]. What the data are actually telling us is that.3.

let Xt = f (Zt . Let G(Xt |f . β|f . f )d β. conditional on the function f and the parameters β. the data display a hump-shaped response to transitory shocks whereas the model implies a much smaller monotonic decay. . Cogley and Nason [70] also argue that the model displays weak propagation mechanisms which arise from intertemporal substitution. let π(β|I . GDP. Also.33–7 years per cycle. Denote the predictive density p(Xt |I . f ) denote the density for the parameters β.2. f ) = H (Xt . The test statistic also shows that the implications of the model are rejected at significance levels of 1% or less. and adjustment costs. let Xt denote a vector stochastic process with a known distribution that describes the evolution of a set of observed variables.8 By contrast. the autocorrelations of output at lags 1 and 2 are significant and positive. and let H (Xt . conditional on the information set available to the researcher I and the function f .3. Define expectations of the functions of the simulated data 8These results are. capital accumulation. investment. β) denote the density of the vector Xt .Matching the Model to the Data 121 for output implied by the model is nearly flat. in some sense. 7. Calibration as Estimation Canova [56. β) denote the process for these variables generated by a specific economic theory or model as a function of exogenous and predetermined variables Zt and the parameters β. but it is unable to match the impulse response function for the transitory component. as is the power spectrum. In particular. complementary to those regarding the role of filtering on the cyclical properties generated from a standard RBC model. 57] proposes an alternative approach for providing statistical inference in calibrated models. say. See Cogley and Nason [69]. and interest rates. and the spectrum for output displays significant power at the business cycle frequencies of 2. I ) denote the joint density for the data and the parameters. β|I . His approach involves constructing prior distributions for the parameters used in calibrated models based on existing estimates in the literature or other a priori information available to the ¯ researcher. To describe his approach. In terms of the impulse response functions. the model has some success in matching the estimated response functions in the data for the permanent component of GDP.

or estimation techniques. Simulation exercises conducted after a subset of the parameters has been estimated (using GMM. In general. Notice also that the function f is typically unknown and must be obtained using some approximation method. the model can be simulated repeatedly for different values of β and Zt to compute sample paths for Xt . f )dXt H (Xt . If such information is hard to obtain. where ν(Xt ) is some moment of interest and D is a bounded set. f ) by E (µ(Xt )|f . f ). suppose that we are interested in evaluating Pr(ν(Xt ) ∈ D). for . While the densities H and p are unknown. • For each drawing of β and zt . • Repeat the above two steps N times. one could count estimates of elements of β obtained from alternative studies and smooth the resulting histogram to obtain the density π(β|I . define µ(Xt ) = 1 if ν(Xt ) ∈ D and zero otherwise. Then. f )d βdXt . β|I . the approach that Canova advocates for conducting statistical inference in calibrated models is as follows: • Select a density π(β|I . Canova [56] describes how to account for such an approximation error when computing the moments of the simulated data.122 Business Cycles: Fact. f ) and zt from k(Zt ). • Draw vectors β from π(β|I . One of the key aspects of Canova’s approach is the selection of a density π that summarizes information about existing parameter estimates in an efficient manner. The calibration approach involves putting a point mass on a particular value of β. (3. This information may be derived from alternative data sets. given the information I available to the researcher and a density k(Zt ) for the exogenous processes. then one can use a uniform distribution. f ) for the unknown parameters. model specifications.22) This approach allows for a probability statement regarding the behavior of various moments implied by the model. β) or its approximation. • Construct the frequency distribution of µ(xt ) and other statistics of interest that can be used to evaluate the performance of the model. More precisely. Fallacy and Fantasy (denoted by µ(Xt )) under the predictive distribution p(Xt |I . I ) = = µ(Xt )p(Xt |I . generate {xt }T and compute µ(xt ) using t=1 the model xt = f (zt . For example.

and argued that any reasonable macroeconomic model should display some form of nonlinear dynamics (see Potter [172] for a review). They found strong evidence of nonlinearity in industrial production. and time-varying markups. unemployment. Yet there is a new literature that shows that macroeconomic time series may exhibit marked nonlinear behavior. and the underlying productivity shocks using a simple production function framework. Such nonlinearities may take the form of conditional heteroscedasticity such as ARCH or GARCH effects. . we discussed the role of alternative factors that could give rise to endogenous changes in productivity such as labor hoarding. markups.Matching the Model to the Data 123 example) involve putting a point mass on β∗ estimated according to a particular method. they note that observed measures of output and factor inputs such as the capital stock and hours worked display marked nonlinear behavior. they do not find evidence of nonlinearity in the Solow residuals measured after allowing for increasing returns to scale. they nevertheless showed that postwar employment. In our discussion in Chapter 3.3.and second-moment properties of aggregate economic variables for the purpose of matching a model to the data. either in the form of nonlinear stochastic dynamics or deterministic chaos. and Patterson [9] examine the implied behavior of output. Alternatively. GMM. Specifically. One can argue that the approach described above is a global sensitivity exercise that also allows for an evaluation of the model based on the probabilities attached to events that the researcher is interested in. increasing returns.3. Neftci [166] was among the first to demonstrate that unemployment fluctuations were asymmetrical along the business cycle. and while they could not find evidence of chaotic behavior. Using an array of diagnostic tests. They conclude that the nonlinearity must lie in the transmission mechanism. Nonlinearity in Macroeconomic Time Series Another criticism of the RBC model literature is that it has typically been concerned with examining the first. or variable capacity utilization. and industrial production could be described as nonlinear stochastic processes. In a novel analysis. 7. Ashley. variable capacity utilization. the factor inputs. Brock and Sayers [47] tested real macroeconomic variables displayed by deterministic chaotic dynamics. Altug. there may exist asymmetries in various economic variables. Ashley and Patterson [22] developed a test to test for deviation from linear stochastic processes. say.

. Mexico. and excess volatility (i. and conditional heteroscedasticity are common for many of these aggregates. The last feature is achieved by taking a transformation of the normal density using Hermite polynomials (see Gallant and Tauchen [100]). the candidate parameters of the simulated model are adjusted until the simulations of the economic model have statistical properties similar to those of the data. The statistical models are flexible enough to allow for increasing time dependence in the mean of the process (captured through a VAR).. and for nonnormal disturbances. The approach to matching the model to the data is through the efficient method of moments (EMM) developed by Gallant and Tauchen [99]. skewness).and second-moment properties of such variables. Valderrama [207] considers a simple Brock–Mirman-type growth model with GHH preferences (see Greenwood. This ensures that the nonlinearities in the underlying model are not eliminated through a linearization procedure. it also allows for periods of high volatility followed by low volatility (conditional heteroscedasticity).e. excess kurtosis). asymmetric business cycles (i. A comparison is made between the statistical parameter estimates in the SNP step and the statistical parameters obtained using the simulated data and the same statistical model. Fallacy and Fantasy Valderrama [207] examines the statistical behavior of national income and product account aggregates for the US and a set of OECD countries including France. Italy. the economic model is simulated for a given set of parameters. but they are unable to reproduce nonlinearities in these time series. which is a two-step procedure. In the second step. He argues that standard general equilibrium models are able to replicate the first.124 Business Cycles: Fact. Then. Valderrama [207] selects three statistical models to describe the properties of the data. GARCH). He poses this as a “canonical” challenge to the RBC approach. Hercowitz. a seminonparametric (SNP) model is estimated to characterize the statistical properties of the data. and shows that nonlinearities such as skewness. However. and it allows for a non-Gaussian error term. and Huffman [105]) and adjustment costs in investment. for conditional heteroscedasticity (ARCH.e. In the first step. Japan. kurtosis. The objective function is distributed as a X 2 statistic. as in the GMM approach. The SNP approach nests standard VARs. The first model selected by the SNP is a VAR(3) with conditional heteroscedasticity and a Hermite . A value iteration approach is used as the solution procedure. and the UK. The SNP approach uses a standard VAR to model the conditional mean and an ARCH-GARCH structure to model the conditional variance.

whereas in the other two models. but not the nonlinearity in consumption. Valderrama [207] finds that the irreversibility constraint does not bind during the solution of the model. and arrive at results that indicate little or no significant impact of irreversibility or lumpiness on aggregate investment dynamics. Surprisingly. The linear VAR(3) model also implies a much lower volatility for the consumption and investment series than the other two models. which allows for nonlinearity in terms of the ARCH effects but continues to assume that the errors are Gaussian. its estimates are around 3. Valderrama finds that the RBC model is not successful at generating the conditional variance of investment. it can capture the nonlinearity in investment. this parameter is estimated to be around 10. [9] who show that the nonlinearities in the observed series must lie in the propagation mechanism. these results suggest that such features as financial frictions or irreversibility in investment are required to match the nonlinearities of consumption and investment. φ. The parameters of the underlying economic model are chosen based on standard calibration exercises and also to match the three statistical models using a simulated method of moments approach.Matching the Model to the Data 125 polynomial of degree 4. Veracierto [208] argues that the reason why irreversibility has an impact on aggregate investment in various . Valderrama [207] finds that the biggest difference among the three statistical models is in terms of the adjustment cost parameter. Two other models are selected for the purpose of describing the data. The second model is a VAR(3) with ARCH(1) errors. By contrast. This is similar to some earlier results in the literature. When the statistical model is forced to be a linear VAR(3). Veracierto [208] and Thomas [205] compare model economies with irreversible investment (in the case of Veracierto) or lumpy investment (in the case of Thomas) with the actual economy and with a baseline model of flexible investment. The first of these is a linear VAR(3) so that the statistical procedure is similar to the RBC approach of matching the impulse responses from a standard VAR with those generated from the underlying economic model. the other two statistical models allow for these features and hence do not require such a high adjustment cost parameter. The intuition for this result stems from the fact that a high adjustment cost parameter helps to smooth the implied behavior of investment and to reduce conditional volatility or kurtosis in the investment series. Combined with the findings of Altug et al. implying that the irreversibility feature is not important for matching the model to the data.

Veracierto [208] argues that when the size of sectoral shocks is consistent with that observed in the data. Fallacy and Fantasy multi-sector growth models considered in the literature (see.126 Business Cycles: Fact. and Gilchrist [41]). Gertler. monetary shocks. for otherwise they would not be identified. Thus. financial accelerator models of investment find that financially constrained firms’ investment responds three times as strongly to a monetary expansion than that of firms which are not constrained (see Bernanke. and may be responsible for softening the impact of the inflexibility faced by plants in terms of their capital adjustments. these microeconomic nonlinearities must matter at the aggregate level. our analysis of the simple RBC model with flexible prices and wages suggests that it cannot successfully match many features of the data. for example. we suggest that the reason why irreversibility may not matter in a model with flexible prices is that the introduction of imperfect competition with optimal price-setting behavior on the part of firms or. this is a puzzling finding. the aggregate impact of irreversible investment disappears in a general equilibrium framework. However. Moreover. alternatively. irreversibility may play a more important role when financial constraints at the plant level and the household level are taken into account. This latter feature may compensate for the rigidity faced by plants given that capital and labor are substitutable according to the Cobb–Douglas specification of technology. These arguments suggest that nonlinearities are another . changing the way in which the labor market is modeled and introducing wage contracts in the RBC framework may lead to more pronounced quantity adjustments. as Valderrama [207] states. At the same time. as in Bertola and Caballero [42]. As Caballero [53] emphasizes: “An important point to note is that since only aggregate data were used. Coleman [73] or Ramey and Shapiro [174]) is due to their assumption of unrealistically large sectoral shocks. More generally. is due to their assumption of a very large variance for these idiosyncratic shocks. both Veracierto and Thomas find that irreversibility or lumpiness at the plant level does not affect aggregate investment significantly once general equilibrium effects such as endogenous price adjustments are taken into account. Given the importance of nonlinearity in macroeconomic time series documented in a number of studies. Similarly. namely. one may consider the impact of another type of aggregate shock. for example. the reason why aggregate investment displays lower variability in the presence of irreversibility when firms are subject to idiosyncratic shocks.” Yet. Both Veracierto [208] and Thomas [205] assume that labor is perfectly mobile across plants.

Hoover. Many have argued that this was to be expected (see Hartley. highlighting problems in finding appropriate 9 See Altug [4]. Yet. we note that Altug’s [4] results did not lead to support for the model.4. The Debate Reconsidered King [128] discusses the arguments on different sides of the debate under the heading “Quantitative Theory and Econometrics”. for example. The factor representation underlying Altug’s analysis has also been resuscitated by Giannone et al. may be preferable.3. and Salyer [116].Matching the Model to the Data 127 area where the assumptions of the simple RBC model fail to hold. Yet.9 In the absence of the initial frequency-domain estimation based on the factor representation. 7. in what sense does adding another simple feature to such a contrived model “resolve” a very specific empirical puzzle? Adding another shock may “loosen” the behavior of the model. but in what sense does this make the model a better interpretation of reality? King also discusses the shortcomings of this approach. it is unlikely that RBC models would be subjected to the types of diagnostic tests proposed by Watson [209]. For some time. p. King [128] advocates such an approach as an alternative to calibration that does not face the problems of “testing highly restrictive linear models”. one could argue that the GMM approach in Christiano and Eichenbaum [65] involves choosing a subset of moments of an equally restrictive nonlinear model. the outcome of this procedure will be known in advance of the test: the probability that the model is true is zero for stochastically singular models and nearly zero for all other models of interest. an RBC model with some or all of its parameters estimated) to an unrestricted time series model. and a potential direction for improving the model’s fit. . One can also examine the claim that a less restrictive approach to estimation and model evaluation following the approach in Christiano and Eichenbaum [65]. King claims to be on the quantitative theory side of the debate. Watson’s [209] insights were partly due to Altug’s initial analysis based on the spectra implied by the model and those in the data.” Returning to the inception of the estimation versus calibration debate. and argues against the estimation and comparison of a “heavily restricted linear time series model (for example. [103] in a VAR framework. If the simple model is counterfactual. 17).

The initial criticisms of .128 Business Cycles: Fact. The simple propagation mechanisms inherent in the model have been deemed incapable of reproducing business cycle dynamics of key variables such as output. These developments show that the contribution of the more formal econometric techniques need not be dismissed as cursorily as is sometimes the case. the estimation versus calibration debate has led to a panoply of research that has extended the use of these models in a variety of directions. It is also worth noting that many of the implications of the standard RBC model have not withstood the scrutiny conducted under a variety of approaches. examining a broader set of moments may make more sense because this yields more information about the underlying behavior of the model. As Geweke [102] notes. Another criticism of the RBC approach is that it fails to capture nonlinearities in key macroeconomic time series. far from this occurring. The estimation and testing of a highly restrictive linear or linearized model may yield many insights regarding the failure of a model as much as examining the performance of a model based on a small subset of moments. Gregory and Smith [108] have also argued that the smallsample properties of the estimators obtained with the approach advocated by Christiano and Eichenbaum [65] may be far from reasonable if the calibrated parameters do not consistently estimate the true values. Fallacy and Fantasy instruments and in the appropriate selection of moment conditions for model evaluation. one could argue that the approach in Christiano and Eichenbaum [65] is too restrictive. In this sense. and the New Keynesian challenge has shown that the model fails in generating the observed negative response of hours to productivity improvements. This interpretation precludes the notion that quantitative theorizing can take the place of formal econometric methods. One interpretation of quantitative theorizing is that it helps researchers understand the workings of highly nonlinear dynamic stochastic models and gain some intuition about different model features. In fact. Perhaps initially the RBC theorists were concerned that the estimation versus calibration debate would discredit the use of well-specified dynamic general equilibrium models for the purpose of capturing the quantitative behavior of economic variables. These arguments show that there is no clear-cut dichotomy between these approaches. Conversely. examining a small set of moments does not necessarily lead to a less restrictive approach because the moments under consideration typically incorporate all the implications of the underlying theoretical model.

Pagan. adjustment costs in investment. Above all. the model incorporates wage and price contracts following the approach in Calvo [54]. Finally. There is also an interest-rate-setting rule that defines . For example. sunspot equilibria. Yet this model has a core set of 26 equations! As another example. 7. The recent class of models developed for this purpose has vastly more features than any simple RBC model. as all agents have complete knowledge of the economy and complete understanding of shocks when they hit. There are no market frictions and no locational specificity. In sum. The aim of this model is to reproduce the inertial behavior of inflation and persistence in real quantities. Eichenbaum. One can ask whether this class of models overcomes some of the criticisms leveled against the original RBC approach. DSGE MODELING Another offshoot of the original RBC debate is the development of dynamic stochastic general equilibrium (DSGE) models that can be used for policy analysis. there is no banking sector and no specific role for money and credit in the monetary transmission mechanism. no fixed costs or discontinuities. the model recently proposed by Kapetanios. especially in the short run. To capture the first feature. consider the model proposed by Christiano.4. the original RBC approach has even instigated a new generation of Keynesian models that offer features such as credit accelerators. and a variety of techniques have been developed for the purpose of matching the model to the data. a topic to which we turn next. the model contains assumptions that are almost guaranteed to be violated by the data. and animal spirits. and Scott [123] for policy analysis of a small open economy.Matching the Model to the Data 129 the RBC approach have led to a wide set of extensions of the original RBC approach. and variable capital utilization. According to them: [This] model is stark in its assumptions. Consider. As we discuss in the next chapter. The model also incorporates a variety of “real” frictions such as habit persistence in consumption. The model assumes a representative household and a symmetric equilibrium for firms. markets are assumed to clear at all times. and Evans [67] for the analysis of the effects of monetary policy shocks on real and nominal variables. for example. Another extension of the original approach has been the development of applications that are useful for policy analysis. There are no market frictions and distortions. firms are assumed to borrow working capital to finance their wage bill.

an exploratory analysis of the role of nominal and real rigidities in propagating shocks. one could also question the implications of the model should there be changes in the distribution of the exogenous processes. Furthermore. This raises 10 For a recent example of the method of indirect inference applied to a model of the EU economy. in many ways. see also Demers. Christiano et al. Christiano et al. the impulse responses to a given monetary shock. Demers. 11 For a review and discussion.130 Business Cycles: Fact. provide some evidence on the role of price and wage rigidities. By contrast. with the latter result derived from the existence of state-contingent securities for consumption. and Altug [79]. but in a highly restrictive environment. For applications.10 The authors’ analysis is. Christiano et al. For example.11 Christiano et al. Minford. the assumption of a constant adjustment cost parameter would fail to hold. [67] pursue a limited information estimation strategy by estimating a subset of the parameters of the model to match the impulse responses of eight key macroeconomic variables to a monetary policy shock with those implied from an identified VAR using the so-called method of indirect inference. . estimated models of adjustment costs have been shown to imply implausible degrees of costs of adjustment. but homogeneous in their consumption and asset holdings. In such cases. Yet. It is also very much in the spirit of the original Kydland–Prescott approach [141] in that the goal is to match a set of observed characteristics — in this case. see Altug. there are shortcomings in their approach. and Demers [10. Fallacy and Fantasy monetary policy. implying that the propagation mechanisms would vary with changes in the distribution of exogenous variables facing agents. the model assumes a continuum of households which are heterogeneous in their wage and the hours that they work. 11]. For example. and Wickens [160]. the adjustment cost model has been criticized on the grounds that it implies a constant cost of adjustment which does not vary with economic conditions. Demers. see Meenagh. [67] assume that monetary policy shocks are drawn from a given and known distribution. [67] derive much of their evidence regarding the types of real rigidities on the results of calibration-type exercises of the current generation of macroeconomic models such as habit persistence or adjustment costs. However. the irreversible investment model studied by Demers [78] and others has been shown to generate a time-varying adjustment cost that varies in response to changes in objective and subjective forms of risk and uncertainty.

and . The recent research on DSGE models has also employed Bayesian analysis as a way of incorporating prior uncertainty about the parameters of interest (see. Second. First. the posterior distribution for the parameters. the production technology. Judd [122] provides a textbook treatment of many currently used solution methods. The Bayesian estimation of DSGE models involves several steps. which is typically a high-dimensional object. and θ is a vector of parameters characterizing preferences. they employ time domain methods by making use of the Kalman filter with a linear law of motion for the state variables and a linear measurement equation for a key set of observables. As in the analysis of Altug [5]. the method presupposes that a solution can be found for the underlying economic model of interest. suppose that the state space representation for the model’s solution consists of the following: • a transition equation St = g (St−1 . vt . structural or invariant to interventions.Matching the Model to the Data 131 the issue of whether DSGE models are. they augment the approximate linear laws of motion with additional shocks. for example. the likelihood function for the observations must be formed. amongst others. Second. and Wright [159] consider an equilibrium business cycle model with household production and distortionary taxation. vt is a vector of innovations. Schorfhiede [186] and An and Schorfhiede [15]). θ). However. There are a variety of approaches for solving nonlinear dynamic stochastic models. examining the posterior distribution calculated as the prior distribution times the likelihood function is often more straightforward computationally than maximizing the likelihood function. this approach first provides a link with the calibration approach by allowing the researcher to use information from microeconomic studies or macroeconometric estimates. To illustrate this approach. the use of Bayesian methods and. in particular. where St is a vector of state variables that describe the evolution of the model over time. Rogerson. Following the initial contribution of Altug [5]. a variety of other papers have implemented maximum likelihood estimation of dynamic equilibrium models. must be examined. Third. which is obtained from the prior distribution and the likelihood function. More recent applications include Canova [60] and Ireland [121]. in fact. McGrattan. and information. Similar to Canova [56].

Given the probability density functions for the shocks fv ( · ) and fw ( · ). θ)p(St |y t . The excellent survey by Fernandez-Villaverde [90] describes in detail how to implement Bayesian estimation of DSGE models. θ). θ) = p(y1 |θ) t=2 T p(yt |y t−1 .132 Business Cycles: Fact. θ). θ). θ)dSt . θ) T = p(S1 |S0 . the state space representation allows us to compute the likelihood of the observations y T = (y1 . yT ) at the parameter values θ as T p(y . wt . θ) and p(Yt |St . θ)p(St |y t−1 . Hence. the posterior distribution can be expressed as π(θ|y T ) = p(y T |θ)π(θ) . wt . This state space representation can be used to derive the likelihood function of the observables and to obtain the posterior distribution for the parameters conditional on the observations. Given a prior distribution π(θ) for the parameters θ. . θ) = p(St+1 |St .23) The remaining task is to compute the sequence {p(St |y t−1 . Likewise. θ)}T . θ). where Yt are the observables and wt are the shocks to the observables (which may take the form of measurement errors. . Following his approach. we can compute p(Yt |St−1 . . this sequence is computed by making use of the relation12 p(St+1 |y t . vt . θ). . θ)dS0 t=2 p(Yt |St . Fallacy and Fantasy • a measurement equation Yt = h(St . Notice that Yt = h(g (St−1 . p(y T |θ)π(θ)d θ (4. which shows t=1 the conditional distribution of the states given the observations. we can also compute the probabilities p(St |St−1 . among others). . θ)dSt 12This is known as the Chapman–Kolmogorov equation.

In contrast to the approach in Christiano et al. an investment-specific shock. θ)p(St |y t−1 . and the nominal short-term interest rate — by minimizing the posterior distribution model’s parameters based on a linearized state space representation for the model. Smets and Wouters [194] argue that their approach offers several important advances. the estimated DSGE model performs as well as standard and Bayesian VARs. a temporary contractionary monetary shock has a temporary negative effect on output and inflation. [67].Matching the Model to the Data 133 and an application of Bayes’ theorem as p(St |y t . they consider a full set of structural shocks. investment. They consider the behavior of seven key macroeconomic time series in the euro area — real GDP. Fourth. They extend the approach in Christiano et al. θ) = p(yt |St . based on Bayesian model evaluation criteria. they find that the estimates of the structural parameters of the model are plausible and. Third. These include a productivity shock. exploring it for different values of θ typically involves the use of sampling through Markov chain Monte Carlo (MCMC) methods. their findings attribute the largest role 13 See Schorfhiede [186] or Fernandez-Villaverde [90]. a labor supply shock. on the whole. . the GDP deflator. Their estimates imply more price stickiness than the estimates of Christiano et al. 195] provide two important examples of this approach. then the Kalman filter is used to obtain the likelihood function. Second. employment. θ) . they argue that the effects of the different shocks on the variables of interest are consistent with existing evidence that does not rely on a DSGE framework. [67] to estimate the parameters of a DSGE model with real and nominal frictions for the euro area using Bayesian methods. similar to those for the US economy. However. they find that. if the transition and measurement equations are linear and if the shocks are normally distributed. θ)dSt Finding the posterior distribution of the parameters displayed in (4. and a government consumption shock plus three “cost-push” shocks. a shock to the household’s discount factor. even if we can evaluate the posterior distribution. Smets and Wouters [194. p(yt |St . [67]. θ)p(St |y t−1 . Otherwise. however. consumption. a method known as the particle filter13 is used to evaluate the likelihood function.23) involves a few remaining steps. For example. First. First. the real wage.

134

Business Cycles: Fact, Fallacy and Fantasy

to labor supply and monetary shocks in accounting for the variation in output in the euro area. One problem with this approach, as we discussed above, is that there tends to be a circularity in building a model to match findings, say, from a standard VAR, which may not be robust themselves. In his survey written in 1992, Fair [86] expressed his disappointment with both the RBC and New Keynesian research agendas in the following way: “The RBC literature is interested in testing in only a very limited way, and the New Keynesian literature is not econometric enough to even talk about serious testing.” In the last 20 years or so, the New Keynesian agenda has progressed much closer towards gaining an explicitly econometric focus. Yet, it is difficult to say how much headway has been made in understanding the propagation mechanisms underlying business cycle fluctuations. Simple models with features ranging from home production to government consumption shocks to nonseparable preferences have been examined to understand their inner workings, as have the roles of nominal rigidities and price-setting behavior. While these model features have proved useful for accounting for some stylized facts, they have often continued to retain counterfactual implications. For example, Cogley and Nason [70] show that the home production framework produces a negative autocorrelation for output. As discussed earlier, many of the model features, such as symmetric and constant adjustment costs in investment, are essentially ad hoc and are at variance with empirical evidence at the micro level. The recent DSGE models include a large set of shocks that have little economic meaning. By its very construction, the DSGE framework cannot easily move into a setup that relaxes the representative consumer assumption, be this through observed and unobserved forms of heterogeneity or the presence of uninsurable risk. Yet, even under the complete markets assumption, Altug and Miller [7, 8] show that exogenous and endogenous forms of heterogeneity matter for the behavior of individual allocations. Browning, Hansen, and Heckman [48] provide a comprehensive discussion regarding the role of heterogeneity in general equilibrium modeling, and raise cautionary points about linking the current class of dynamic equilibrium models that dominate macroeconomics with microeconomic models and evidence. Could the entire general equilibrium macroeconomic quantitative theorizing approach — despite the use of much bigger models and more powerful computational tools — be a detour, as Fair [86] seems to indicate?

Matching the Model to the Data

135

Should economists build intuition from the workings of well-articulated economic models at the same time as they use more purely econometric approaches for quantitative policy analysis? Kapetanios et al. [123] note that the current class of fully articulated dynamic equilibrium models that are increasingly being used by policy-makers such as central banks help to provide restrictions on a reduced form, a practice very much in the spirit of the Cowles Commission approach. Thus, the exercise is not necessarily to recover structure in the form of the parameters of preferences and technology, as argued by Lucas [150] in his famous critique of econometric policy evaluation. More tellingly, can economists hope to recover “true structure” using macroeconomic data in the presence of observed and unobserved forms of heterogeneity? Or, alternatively, when subjective beliefs by economic agents in a changing stochastic environment constitute “hidden structure”, as Martin Weitzman [210] claims? The notion that structural models may not indeed be structural has begun to be discussed in the literature. Fernandez-Villaverde and Rubio-Ramirez [91], for example, discuss the stability over time of estimated parameters in DSGE models, and show that the parameters characterizing the pricing behavior of households and firms in Calvo-type pricing functions are correlated with inflation. Canova and Sala [61] investigate identifiability issues in DSGE models and their consequences for parameter estimation and model evaluation when the objective function measures the distance between estimated and model impulse responses. They show that observational equivalence as well as partial and weak identification problems are widespread, lead to biased estimates and unreliable t-statistics, and may induce investigators to select false models. Canova [59] also examines the issue of identifiability of DSGE models, and argues that researchers should be careful in interpreting the diagnostics obtained from the estimation of structural models and make use of additional data in the form of micro data or data from other countries to augment the information obtained by formal estimation.

This page intentionally left blank

In a recent analysis. I have omitted a discussion of multi-sector models in generating business cycle behavior. for example. he considers the role of a search externality and a lemons problem in the market for search inputs. Many have argued that such models are more in the spirit of true Keynesianism because they stress the role of “animal spirits”. Farmer [88] generates a model of self-fulfilling equilibria which arises from market failure associated with the labor market. Nevertheless. multi-sector models may also have shortcomings because they typically imply that overall expansions in the economy are associated with contractions in one sector. even the experience of reviewing some of this literature shows that it is lively. More generally.Chapter 8 Future Areas for Research The literature on business cycles is vast. the wage is then set by a 1 See. For example.1 In such cases. but always thought-provoking. it may be that shocks are concentrated in a particular sector and are propagated to the rest of the economy from that sector. However. I have offered a perspective based on my own perceptions and mindsets. one needs to consider multi-sector models since the one-sector growth model does not allow for a consideration of these issues. at times contentious. In this book. Specifically. 137 . There are many other topics of interest that we could have discussed regarding business cycles. and contrasts it with the standard real business cycle (RBC) approach. Benhabib and Farmer [38]. For one. Farmer [87] provides an eloquent enunciation of this approach. Many recent works assume that there is a search technology such that firms and workers are randomly matched. I have not provided a discussion of models with selffulfilling expectations and multiple equilibria.

the model with a search externality emphasizes the role of confidence in restoring full-employment output. and Rouabah [171]. This leads to a model with a continuum of steady-state equilibria. The policy implications of such a model differ from those of the standard textbook Keynesian model. he assumes that firms offer the same wage in advance. Undoubtedly. new techniques. For a recent application in the context of a standard RBC model. and new controversies regarding both. By contrast. see Cooley and Quadrini [76]. irrespective of what the new models or techniques will look like. we may conclude that the debate surrounding their development will not cease to arouse interest or to generate future avenues for research. This book has also omitted a discussion of models with credit and collateral constraints.138 Business Cycles: Fact. which is transmitted to real variables through investors’ beliefs about the stock market. Farmer [88] concludes his analysis by commenting on the potential efficacy of the Obama fiscal stimulus package(s) in light of the findings from his model. all firms earn zero profits but not all equilibria have the same welfare properties. this discussion will be accompanied by the development of new models.3 The above considerations show that business cycles will continue to be a topic of discussion among economists for many years to come. which seeks to alleviate demand-induced shortfalls in output through large fiscal stimuli. see Pierrard. However. One could argue that this type of model deserves much more attention given the global financial crisis that originated from the subprime housing market in the US in 2007. . This leads to a demand-constrained equilibrium. Farmer [88] drops the assumption that wages are determined as a result of a Nash bargaining process. This feature of the model allows for the role of “confidence”. 2This approach owes its origins to the analysis in Mortensen and Pissarides [165].2 In contrast to this literature. In these equilibria. de Walque. Fallacy and Fantasy Nash bargaining process. 3 For a recent analysis of models with financial frictions in a dynamic general equilibrium setting. Instead.

S. 115. S. 30. and R.. Ahmed. 506–533. Ambler. S.” CESifo Economic Studies. Altug. R. “The Investment Tax Credit and Irreversible Investment.” Journal of Macroeconomics. “Gestation Lags and the Business Cycle. 11. 13. 321–346. Demers (2007). 543–570. and U.” Journal of Monetary Economics. Altug. S. Alvarez. “Emerging Market Business Cycles: The Trend Is the Cycle. M.” Canadian Journal of Economics. A’Hearna. Asset Pricing for Dynamic Economies. pp. 46. pp. 10. “Efficiency. “International Business Cycles: What Are the Facts?” Journal of Monetary Economics. 8. S. Gopinath (2004). S.” European Economic Review. pp.. “International Transmission of the Business Cycle in a Multi-sector Model. and M. Woitek (2001). Demers (2009). 430–465. 53. and D. pp. and R. Murthy (1994). 9. Ashley. 3. Output. 4. Equilibrium. F. “Political Risk and Irreversible Investment. 68. “The Effect of Past Experience on Female Wages and Labor Supply. S. pp. and Asset Pricing with Risk of Default.” Journal of Political Economy. 45–85. F. 6. 775–797. 65. 12. (1989). “Are Technology Shocks Nonlinear?” Macroeconomic Dynamics.” Econometrica. 273–300. and P. 3. forthcoming. Altug.” Review of Economic Studies. and U.Bibliography 1. Patterson (1999). E. Labadie (2008).. Altug. 14. “More International Evidence on the Historical Properties of Business Cycles. “Household Choices in Equilibrium. E. “Time-to-Build and Aggregate Fluctuations: Some New Evidence. Miller (1990). Demers. and C. and G. pp. 5. pp. Zimmermann (2002). 69–102. 51. 27. Demers. S. Altug. Altug. S. S. pp. S. 889–920. 982–993. “Money. 7. and Real Business Cycles in a Small Open Economy.” International Economic Review. (1985). Miller (1998). Aguiar. 2. Cardia.. F. pp. and M.” Unpublished manuscript. Cambridge: Cambridge University Press. Altug. 58. Altug. and R. Jermann (2000). pp. 47.” Econometrica. Cardia. 139 . Ambler. pp.. and C. B. 257–276. Zimmermann (2004).

“International Trade and Business Cycles. “Credit Frictions and ‘Sudden Stops’ in Small Open Economies: An Equilibrium Business Cycle Framework for Emerging Markets Crises. Barsky. Vol. 21. and L.” Econometric Reviews. and J. and F. (1987). Kydland (1992). 13. Smith (1993). Prices and Budget Deficits in the United Kingdom. and E. pp. E. C. 335–405. 249–279. “Bayesian Analysis of DSGE Models. 1701–1918.” Journal of Business. Artis. Chadha. 100. pp.” Journal of Economic Perspectives. pp. McGrattan.). 745–775.” Innocenzo Gasparini Institute for Economic Research (IGIER) Working Paper No. 297–316..” In S. “Are Technology Improvements Contractionary?” American Economic Review. E. Basu. Osborn (1997).” NBER Working Paper No. pp. . and F. pp. pp. 1. Schorfhiede (2007). S. 864–888. 26. Amman.. Cambridge: Cambridge University Press. Backus. 2. S. 120–132. Sargent (1996). and P.” International Economic Review. Anderson. and A. 29. Handbook of Computational Economics Vol. Artis. D. Princeton: Princeton University Press. D. 113–172.. Interest Rates. pp. 1418–1448. “Procyclical Productivity: Increasing Returns or Cyclical Utilization?” Quarterly Journal of Economics. 20. J. Artis. M. 35. Fernald. pp.” Journal of Economic Perspectives. P. An. “Dating the Euro Area Business Cycle. Zhang (1999). 18.” In H. Arellano. Patterson (1989). Z. and W. 23. and W. Handbooks in Economics. 111. and T. Handbook of International Economics. 24. “Cyclical Productivity with Unobserved Input Variation. Basu. J. “International Real Business Cycles. Altug. pp. R.). 115–134. (1996). M. Barro. Amsterdam: North-Holland. “Consumption and Real Exchange Rates in Dynamic Economies with Non-traded Goods. Cooley (ed.” Journal of Political Economy. “Business Cycles for G7 and European Countries. Taylor (1999). 22. S.” Journal of International Economics. Basu. Grossman and K. Rust (eds.. 51. 19. 82. Kimball (1997). 5915. S. Proietti (2003). Rogoff (eds. 18. 171–252. Kehoe (1992). Ashley. 17. 1801–1864. (1995). Kydland (1995). and M. and C. “Government Spending. Kehoe. “International Evidence on the Historical Perspective of Business Cycles. pp. pp. Amsterdam: Elsevier Science/North-Holland. Kontolemis. Fallacy and Fantasy 15. D. Backus. Dynamic Macroeconomic Analysis: Theory and Policy in General Equilibrium. and M. 30. R. 26. 331–356. 31. Mendoza (2003). Kehoe. Baxter. D. Marcellino. Hansen. 45–68. and F. 13. Artis. 32. pp. and G. Frontiers of Business Cycle Research. M. “International Business Cycles: Theory and Evidence. 33. 221–247. 1–16. “Business Cycles in International Historical Perspective.). Zhang (1997). pp. M. “Mechanics of Forming and Estimating Dynamic Linear Economies. Nolan (eds. and D. and T. pp. 685–704. Kimball (2006). R. M. Basu. and D. P. M. 237. “Oil and the Macroeconomy Since the 1970s. Backus. Kendrick.).” Journal of Monetary Economics. 25. Kilian (2004). 27.. L. 96. 20. 719–751. “International Business Cycles and the ERM: Is There a European Business Cycle?” International Journal of Finance and Economics. 16. “Linear versus Nonlinear Macroeconomics: A Statistical Test. 70. D.” In G. 28.140 Business Cycles: Fact. Backus. pp.” In T. S. pp. “Further Evidence on the International Business Cycle and the ERM: Is There a European Business Cycle?” Oxford Economic Papers. pp.. 30.” American Economic Review.

pp. Gertler (1989). King (1999).” American Economic Review. pp. R. 36. 335–347. M.” Federal Reserve Board (FRB) International Finance Discussion Paper No. Bertola. 79. . D. 37. L. M.Bibliography 141 34. “The Dynamic Effects of Demand versus Supply Disturbances.” Review of Economics and Statistics.” Journal of Political Economy.. (1999).” Journal of Monetary Economics. and S. “Irreversibility and Aggregate Investment. and U. G. Net Worth and Business Fluctuations. and M. “Labor Hoarding and the Business Cycle. 42. 45. 1A. and M. 65. Bordo. Jermann (1997). 52. Bernanke. 37. and R. 40. 441–462. 36. 2003-762. “Homework in Macroeconomics: Household Production and Aggregate Fluctuations. (1983). Doyle (2003). 654–673. and S. Woodford (eds. 12. 46. Mitchell (1946). 575–593. R. A. Quah (1989). “Variable Factor Utilization and International Business Cycles. Gertler. and D. 43. “The Financial Accelerator and the Flight to Quality.” Journal of International Economics. 245–274. Boschan (1971). 1–15. Baxter. 48. New York: Columbia University Press for the NBER. pp. Brock. 35. (1994). 50. “Factor Hoarding and the Propagation of Business Cycle Shocks.. J. North Holland: Elsevier Science. 87(1).” Review of Economic Studies. and J.. Measuring Business Cycles. J.” Journal of Monetary Economics. “The International Diversification Puzzle Is Worse Than You Think. 86. Browning. “Is the Business Cycle Characterized by Deterministic Chaos?” Journal of Monetary Economics. 22. M. W10130. 14–31. C. Sayers (1988). Eichenbaum. 47. Hansen. 101. and T. Farmer (1996). pp. 383–398. pp. Bernanke. 41. Burns. 1B. Burnside. Benhabib. Calvo.” Journal of Monetary Economics. A. O. M. and B. 1154–1174. 33. 223–246.” In J. Handbook of Macroeconomics. 421–443. W. 8. G. 39. “Staggered Prices in a Utility-Maximizing Framework. 61. “Tax Disturbances and Real Activity in the Postwar United States. “Have National Business Cycles Become More Synchronized?” NBER Working Paper No. pp. pp. 51.” American Economic Review.). Handbook of Macroeconomics. Vol. Wright (1991). 79. 1166–1187. 78. Baxter. Gilchrist (1996). 49. Taylor and M. and R. Heckman (1999). pp. Ch. Taylor and M. pp. “Micro Data and General Equilibrium Models. Amsterdam: Elsevier Science. B. 44. Baxter. and D. “Measuring Business Cycles: Approximate Bandpass Filters for Economic Time Series. Vol. 99. “New Keynesian. Open-Economy Models and Their Implications for Monetary Policy. M. Benhabib. Farr (2005). and R. M. 170–180. Cyclical Analysis of Time Series: Selected Procedures and Computer Programs. M. “Aggregate Investment. 53.” The Review of Economics and Statistics. “Indeterminacy and Sector-Specific Externalities. Heibling (2003).” American Economic Review. C. Rogerson. Baxter. pp. 71–90. pp. Bry. 12. Caballero (1994). 54. Eichenbaum (1996). “Agency Costs.). 821–854. Rebelo (1993). and C. “Business Cycles and the Asset Structure of Foreign Trade. Caballero. B. and C. G.” In J. and M.. 81. Blanchard. pp. Ch. pp.” Journal of Political Economy. New York: NBER. M.” International Economic Review. Woodford (eds. Braun. Crucini (1995). Bowman. and W. pp. 38. and R.” American Economic Review. pp. Burnside.

“Back to Square One: Identification Issues in DSGE Models. “What Explains the Great Moderation in the US? A Structural Analysis. 58.” Journal of Political Economy.” Journal of Economic Dynamics and Control. Princeton: Princeton University Press. 28. 1–38. and C. 29. “Are Structural VARs Useful Guides for Developing Business Cycle Theories?” Federal Reserve Bank of Minneapolis Research Department Working Paper No. Canova. Cooley. “How Much Structure in Empirical Models?” CEPR Discussion Paper No. M. and E. F. 477–501. 70. 65.” In T.” National Institute Economic Review. 697–721. 73. “Inspecting the Mechanism: An Analytical Approach to the Stochastic Growth Model. 492–511. Eichenbaum (1992). Eichenbaum. “Effects of the Hodrick–Prescott Filter on Trend and Difference Stationary Time Series: Implications for Business Cycle Research. and M. 237–259. pp. McGrattan (2004). 66. 59. Canova. pp. 72–89. Christiano. 41. Fallacy and Fantasy 55. and J. (2006).” American Economic Review. J. F. . 85. 113. Canova. 1. 57.” American Economic Review. Princeton: Princeton University Press.. Chari. Methods for Applied Macroeconomic Analysis. Cole. Coleman. T. 62. Christiano. 67. 3–24. 56. and C. pp.) (1995). 33. 72. “Statistical Inference in Calibrated Models. 253–278. pp. 1–45. 68. 463–506. 82. (1988). (1995). Christiano. Canova. and C. 71. 61.. pp. L. L. Canova. Evans (1997). 75. Evans (2005). 7. M. L. T.142 Business Cycles: Fact. Cogley. L. 1201–1249. Cogley. and J.” Journal of the European Economic Association. 60. “Sensitivity Analysis and Model Evaluation in Simulated Dynamic General Equilibrium Economies. 74. (2009). Frontiers of Business Cycle Research. “Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy. (1994). “Output Dynamics in Real Business Cycle Models.” Journal of Monetary Economics.” Journal of Applied Econometrics. Nolan (2002). pp. pp. Obstfeld (1991). pp. pp. 63. Chadha. Eichenbaum. T. Frontiers of Business Cycle Research. H.” European Economic Review. “Why Does Inventory Investment Fluctuate So Much?” Journal of Monetary Economics. 247–280. F. P. M. 9819. pp. (1997). Vigfusson (2003). pp. pp.” International Economic Review.. pp. Cooley (ed. Cole. 6791. 583. “Financial Structure and International Trade. Nason (1995). F. pp. 430–450. and R. J. Sala (2006). Christiano. 206–227. W. 9. 631. (2008). H. 64.” Journal of Monetary Economics. Campbell. “Sticky Price and Limited Participation Models of Money: A Comparison. Kehoe. and E. L. 182.” ECB Working Paper No. “A Long View of the UK Business Cycle. T. 21. “Current Real Business Cycle Theories and Aggregate Labor Market Fluctuations. “Economic Growth and Business Cycles. “What Happens After a Technology Shock?” NBER Working Paper No. and M. F. and L. 19. (ed. “The Behavior of Interest Rates in a General Equilibrium Multisector Model with Irreversible Investment. Nason (1995). V. 36. Christiano.).” International Economic Review. Eichenbaum. Prescott (1995). “Commodity Trade and International Risk Sharing. Canova. 123–144. (1988). 69. pp.” Macroeconomic Dynamics.. F. Princeton: Princeton University Press. Cooley. (1994).

” NBER Working Paper No. Farmer. Schwartz (1963).. “Realistic Cross-Country Consumption Correlations in a Two-Country. “An Area-Wide Model (AWM) for the Euro Area. Quadrini (1999). 3–16. Nolan (eds. A Monetary History of the United States. and the Business Cycle: Do Technology Shocks Explain Aggregate Fluctuations?” American Economic Review. 91.. Demers. and A. Cambridge: Cambridge University Press. 58. Gregory. 103. M. “Irreversibility and the Arrival of Information. 87. 400–416. 89. J. pp. “Perfect Competition and the Effects of Energy Price Increases on Economic Activity. pp. and C. “How Structural Are Structural Parameters?” NBER Working Paper No. 607–626. Washington. “Domestic Saving and International Capital Flows. Demers. 77. E. Forni. 85.” Economic Journal. Finn. 1867– 1960. Altug. 249–271. 82. Cooley. Feldstein. 14846.” Review of Economic Studies. Altug (2003). “Confidence. 81. G. (2009). Bordo (2002). Devereux. Dynamic Macroeconomic Analysis: Theory and Policy in General Equilibrium. 39. 1929–1982. 89. 453–473. Eichenbaum. 333–350. 34–154. 1004. 92. Eichengreen. M. M..” NBER Working Paper No. Mestre (2001). 90. M. and R. Friedman. M.. J. and C. Demers. DC: Brookings. Correia. 84. Rebelo (1995). 32. J. M.” Review of Economic Studies. London: George Allen & Unwin. Cambridge. 86. pp. and R. (1991).Bibliography 143 76. 11. The Macroeconomics of Self-Fulfilling Prophecies.). J. Neves. and K. 1089–1113. “Crises Now and Then: What Lessons from the Last Era of Financial Globalization?” NBER Working Paper No. Horioka (1980). (1999). and Animal Spirits. (1991). pp. pp. Fernandez-Villaverde. Trends in American Economic Growth. Denison. J. “Let’s Get Real: A Factor Analytical Approach to Disaggregated Business Cycle Dynamics. R. Fagan.” In S. “The Cowles Commission Approach. Credit. Farmer. 79. . Frisch. (1999). MA: MIT Press. and S. T. 314–329. 2nd ed. B. “Technology.” Quarterly Journal of Economics.” Journal of Money. Eichenbaum. 13166. “Propagation Problems and Impulse Problems in Dynamic Economies. Reichlin (1998).” European Economic Review.” Cowles Foundation Discussion Paper No. “Investment Dynamics. (1985). and V. M. 80. 78. 14677. and M. “Real Business Cycles: Wisdom or Whimsy?” Journal of Economic Dynamics and Control. pp. M. L.” ECB Working Paper No. Crashes. pp. A. R. pp. 94. and S. 42. Singleton (1988). “Business Cycles in Small Open Economies. pp. 83. (1933). and L. Real Business Cycle Theories. Hansen. Chadha. Employment. Princeton: Princeton University Press. Smith (1992). 96. and G. Fernandez-Villaverde.” Journal of International Money and Finance. 90.” In Economic Essays in Honor of Gustav Cassel. (2009). 51–78. 95. 15. Gali. W8716. Fair (1992). F. J. and New Keynesian Economics. (2000). 88. and Banking.” Journal of Monetary Economics. Rubio-Ramirez (2007). Henry. 44. R. M. “The Econometrics of DSGE Models. pp.. “A Time Series Analysis of Representative Agent Models of Consumption and Leisure Choice Under Uncertainty. pp. 93. 65. “A Neoclassical Model of the Phillips Curve Relation. 165–193. Equilibrium Business Cycle Model. I.

(1982). Smith (1989).” European Economic Review. Hercowitz. R. “Invariance Properties of Solow’s Productivity Residual. “The Role of Investment-Specific Technological Change in the Business Cycle. 49. 57–89. “Computational Experiments and Reality. 38. User’s Guide. and J. 107. Monetary Policy. pp. NBER Macroeconomics Annual 2004. Version 8. Common Factors and the Empirical Validation of Equilibrium Business Cycle Models. 12. Hall. Tauchen (1996). Head. 2. Z. A. 99.. 117. Heathcote. Hoover. Krusell (2000).” In P. J. “Real Business Cycles in Emerging Countries?” NBER Working Paper No. . 98. 106. Rogoff (eds. Hansen. “Indivisible Labor and the Business Cycle. 71–112. 44.. Gallant. R. 401..S. “Long-Run Implications of Investment-Specific Technological Change.” American Economic Review. R. and A. L. The Measurement of Durable Goods Prices.” Journal of Economic Dynamics and Control. pp. pp.” Econometrica. D.). Geweke. Gallant. 657–681. and P. “Large Sample Properties of Generalized Method of Moments Estimators.. Giannone. “Formulating and Estimating Dynamic Linear Rational Expectations Models.7. Hall. Uribe (2006). Fallacy and Fantasy 97. 105. Harding. and M. Capacity Utilization. pp. 110. 1029–1054. pp.” International Economic Review.” University of North Carolina Working Paper. and G. Real Business Cycles: A Reader. Salyer (eds. 101. 116. A. Huffman (1988). 402–417. and K. Diamond (ed. Cambridge. 91–115. Z. pp. J.. 225–318. Greenwood. 342–362. 114. and L. “VARs. MA: MIT Press. 601–627. Hansen. L. 109. Sala (2006). 1681–1690.). pp.” Econometric Reviews. pp. Pagan (2003). Sargent (1980). “Which Moments to Match?” Econometric Theory. “Financial Autarky and International Business Cycles. pp. 50. Gali. Greenwood. J. Industry.” Journal of Econometrics. Cambridge.” Journal of Political Economy. and P. D. 921–947. “Technology Shocks and Aggregate Fluctuations: How Well Does the RBC Model Fit Postwar US Data?” In M. 103. and G. (1999). “SNP: A Program for Nonparametric Time Series Analysis. and the Real Business Cycle. “Calibration as Estimation. London: Routledge. and T..) (1998).” Journal of Monetary Economics. Raynauld (1997). “Investment. L. Pancrazi. J. 16. (1988).” Journal of Economic Dynamics and Control. pp. K. (2008). pp. Greenwood. Chicago: University of Chicago Press. 96.” Journal of Monetary Economics. 115. and P. G. 78. Growth/Productivity/Unemployment. Rabanal (2005). Princeton: Princeton University Press. 27. “A Comparison of Business Cycle Dating Methods. Krusell (1997). Gregory. 112. pp. J. “Measuring World Business Cycles. 87. 113. 7–46. J. and F. Hartley. Reichlin. J. 12629. (1990). pp. 132. 111.” American Economic Review. Z. J. Hansen. (1990). Perri (2002). Hercowitz. pp. Gali. 102. and the Business Cycle. “The Relation Between Price and Marginal Cost in U. Gertler and K. A. 104. 108. and G. 257–279. J. Inflation. Tauchen (1998).. R. 677–702.” SCE Computing in Economics and Finance Working Paper No. Gordon. pp. 100.144 Business Cycles: Fact. Gregory. R. (1985). 309–327. MA: MIT Press. 9. Hercowitz. Garcia-Cicco. R. and A.

309–341. Kehoe. and S. 207–231. and A. and Money.” Economic Quarterly. (2004). (1995). Whiteman (2003). “Making a Match: Combining Theory and Evidence in Policy-Oriented Macroeconomic Modeling. R. Terrones (2003). and C. pp. 211–248. 138. J. 139. (1995). R. 120. G. 819–840. Interest. pp. Rebelo (1988).” Journal of Political Economy. and Business Cycles: I. 56. (1926). and Business Cycles: II. Pagan.” Review of Economic Studies. G. and D. 21. Kiyotaki. 81. Prescott (1997). pp.. T. 127. and E. 21. “How Does Globalization Affect the Synchronization of Business Cycles?” IZA Discussion Paper No. Rebelo (1988). 29. 53–105. N. 60.Bibliography 145 118.” Journal of Monetary Economics. “Stochastic Trends and Economic Fluctuations. 70. 86. Helliwell. Real Exchange Rates. “Implications of Efficient Risk Sharing without Commitment. Levine (1993). “Technology Shocks in the New Keynesian Model. and J. “International Business Cycles: World. 595–609. King.. 907–928. 69(3). King. pp. King. Kapetanios. C. 129. and K. “Postwar U. pp.” Econometrica. P. R. J. Scott (2007). King.T. 130. 191–211. London: Macmillan.” Archiv für Sozialwissenschaft und Sozialpolitik. pp. Shin (1997).” Econometrica. How Much Do National Borders Matter? Washington. 81. 63. J. pp. 131. and Country-Specific Factors. Kehoe. Kondratiev. “Liquidity Constrained Markets versus Debt Constrained Markets. “Money. pp. N. Growth. Growth. 122. 363–380. 13. Business Cycles. and the Structure of International Asset Markets.” Review of Economic Studies.. 123. R. 125. and M. 565–594. MA: MIT Press. “International Business Cycles with Endogenous Incomplete Markets. A. E. pp. C. 136.. A. Judd. and C. The Basic Neoclassical Model. 119. 702. Keynes. “Production. 923–936.S. 1–16. pp. Kehoe. 17. (1936). . Köse. P. and Banking. R. Hess. Prasad. 93–109. “International and Intranational Business Cycles. (1998). and S. 93. C. 134. 105. C. pp. 195–232.” Journal of Econometrics. 573–606. Moore (1997). 128. “Production. Credit. A. 124. (1996). Numerical Methods in Economics.” American Economic Review. and M. Cambridge. Region. Plosser. pp.” Journal of Money. pp. “Die Langen Wellen der Konjunktur [The Long Waves of the Business Cycle]. “Quantitative Theory and Econometrics. 121. R. Stock. K. Watson (1991). Rebelo (1993). Köse. Ireland. Perri (2002).” Review of Economics and Statistics. 575–598.” American Economic Review. “Credit Cycles. and S. 132. 137. Plosser.” Journal of Monetary Economics.” Oxford Review of Economic Policy. Kocherlakota. pp. Hodrick. 135. (1998). Otrok. and D... N. R. Plosser. Levine (2001). pp. pp. Plosser (1984). “Consumption. 74. and F. Kollman. King.” Journal of Economic Dynamics and Control. 865–888. pp. New Directions. M. 1216–1239. DC: Brookings Institution. “Low Frequency Filtering and Real Business Cycles. R. Credit. 136.” American Economic Review. 14.” Journal of International Money and Finance. pp. 126. The General Theory of Employment. and Prices in a Real Business Cycle Model. King. 133. “Debt Constrained Asset Markets.

R. Lucas. 101–127. Cambridge. “On ‘Real’ and ‘Sticky-Price’ Theories of the Business Cycle.” Scandinavian Journal of Econometrics. (1975). R. 153. Lewis. F. McGrattan.146 Business Cycles: Fact. Fallacy and Fantasy 140. Kydland. 3–18. R. E2008/11. and L. 154. “Testing a DSGE Model of the EU Using Indirect Inference.” Review of Economic Dynamics.” Journal of Political Economy. (1994). pp. 14. Credit. “Expectations and the Neutrality of Money. pp. pp. 146. pp. 267–297. R. “What Can Explain the Apparent Lack of International Consumption Risk Sharing?” Journal of Political Economy. and C. 335–359. Prasad (2003). Prescott (1990). 1345–1370. R. 50. and C. Wright (1997). 267–290. 161–178.” Journal of Monetary Economics. Kydland. and T. M. 157. F. “An Equilibrium Model of the Business Cycle.” Journal of Monetary Economics. F. E. E. 1113–1144. pp. Lucas. Rapping (1969). Princeton: Princeton University Press. “Business Cycles: Real Facts and a Monetary Myth. pp. (1996). Frontiers of Business Cycle Research. Brunner and A. B.). “The Macroeconomic Effects of Distortionary Taxation. McCallum. pp. J. 33. MA: MIT Press.” In K. (1986). 77. pp. pp. 149. 152. 156. pp. F. pp. 104(2). 38. 126–156. (1977). “Econometric Policy Evaluation: A Critique. 7–29. and M. Wickens (2008). Ljungqvist. Meenagh.” Journal of Economic Theory.” Federal Reserve Bank of Minneapolis Quarterly Review. 151. 83. (1996).” Cardiff Business School Economics Working Paper No. “Identifying the Common Component in International Economic Fluctuations. pp. 144. pp. Long.). pp. (1995). 147. (1972). 141. Marimon (1999). Kydland. R.” International Economic Review. 85. 19–46. J.” American Economic Review.” Journal of Money. Zaragaza (2002). and R. 1(2) (Supplementary Series). McCallum. 615–623. 143. Amsterdam: North-Holland. Prescott (1982). Cooley (ed. 152–165. E. D. McGrattan. Prescott (1991). Meltzer (eds.. 18. “An Equilibrium Model of the Business Cycle with Household Production and Fiscal Policy. J. 148.” Universitat Pompeu Fabra (UPF) Economics Working Paper No. Lucas.” Journal of Monetary Economics. A. Lumsdaine. “Understanding Business Cycles. F. “Recursive Contracts. Employment and Inflation. pp. L. R. Stabilization of the Domestic and International Economy. “Time-to-Build and Aggregate Fluctuations. 573–601. Sargent (2000). pp. Rogerson. 10. 113. 91. 39–69. “Real Wages. 5. Lucas. P. “National Borders Matter: Canada-US Regional Trade Patterns. and Banking. and E. and E. 93. Kydland. and R. Minford.” Economic Journal. and E. Marcet. “Interest Rates and Currency Prices in a Two-Country World. Lucas. 4. pp. K. 160.” Journal of Political Economy. Lucas.. 103–123. (1976). pp. Kydland. 150. R. 158. (1982). 155.” In T. . Plosser (1983). 337.” Journal of Political Economy. 145. “Business Cycles and Aggregate Labor Market Fluctuations. “Argentina’s Lost Decade. and E.” Econometrica. “The Econometrics of the General Equilibrium Approach to Business Cycles. 142. 159. “Real Business Cycles. 721–754. 397–414. Recursive Macroeconomic Theory.

Neftci.” Unpublished manuscript. Plosser (1982). 3–16. “Policy Uncertainty and Private Investment in Developing Countries. pp. pp. 175. pp. 797–818. S. Rogoff (1983). “Facts of Business Cycles Around the Globe. (1999). V. and Future. pp. and K. Rogoff (2001). 13. E. (1986). G.” Review of Economic Studies. “Is the Stabilization of the Postwar Economy a Figment of the Data?” American Economic Review. 172. S. L.” Journal of Political Economy. 3–24. 139–162.” Paper presented at the IEA 15th World Congress. Ratfai. “Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample?” Journal of International Economics. 10. Romer. Supervision and Liquidity Injections: A Dynamic General Equilibrium Approach.” American Economic Review. Bernanke and K. 179. 180. (1989). R. Shapiro (2001). Potter.W. New York: W. C. and P. “Are Economic Time Series Symmetric over the Business Cycle?” Journal of Political Economy. “Trends and Random Walks in Macroeconomic Time Series. 958–992. pp. pp. “Nonlinear Time Series Modelling: An Introduction. “The Macroeconomic Effects of War Finance in the United States: World War II and the Korean War. 181. 316–334. and K. Mendoza. 169.” CEPR Discussion Paper No. “Indivisible Labor. Turkey. W. 163. 21. New York: National Bureau of Economic Research. Pierrard. 76. Prescott. (1991). Meese. 164. pp. “Economic Fluctuations in Central and Eastern Europe: The Facts. pp. 92. 1869–1908. Statistical Indicators of Cyclical Revivals (NBER Bulletin No. “Real Business Models: Past. and A. Rogoff (eds. and C. Nelson. 14. pp. 69). Ohanian. Lotteries and Equilibrium.). 178.” Journal of Monetary Economics. 81. pp. 173. and P. 171. (1988). New York: National Bureau of Economic Research. S. C. .” Journal of Monetary Economics. 217–238. “Financial (In)stability. 4846. Burns (1938). Ratfai. 391–415. 177.” Scandinavian Journal of Economics. 109. E. 36. (2005). 107. “The Prewar Business Cycle Reconsidered: New Estimates of GNP. Romer. 23–40. pp. pp. (1984). M.” Carnegie-Rochester Conference Series on Public Policy. Benczur (2005). Cambridge. Obstfeld. Pissarides (1994). Rouabah (2008). 505–528. O. 176. 61. C. pp. Phelps. 25.” Journal of Political Economy. Mortensen. 11–44. Ramey. 97. Benczur (2008). Rogerson. (1991). and M.Bibliography 147 161. 170. (1997).” American Economic Review. 166. Microeconomic Foundations of Employment and Inflation Theory. 168. “Real Business Cycles in a Small Open Economy. 174. 307–328. and A. 87. E. D.” Journal of Economic Surveys. 1–37. 339–390. Istanbul. (1970). Rodrik. 162. R. 167. and C. Business Cycles: The Problem and Its Setting. Rebelo. 229–242. “Displaced Capital: A Study of Aerospace Plant Closings. “Theory Ahead of Business Cycle Measurement. Mitchell. 165. A. Norton & Co.. (1927).” Journal of Development Economics. pp. Mitchell. Present. MA: MIT Press. NBER Macroeconomics Annual 2000. pp. “Job Creation and Job Destruction in the Theory of Unemployment. “The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?” In B. D. W. (1986). A. de Walque.

62. Watson (1999).” In T. Striving for Growth After Adjustment: The Role of Capital Formation. 200. pp. Handbook of Macroeconomics. 189. 1A. Smets.” Review of Economics and Statistics. 243–293. Smets. Stock. Rotemberg. pp.” Journal of Money. North Holland: Elsevier Science. Credit. Frontiers of Business Cycle Research. pp. Schumpeter. (1957). “An Estimated Dynamic Stochastic General Equilibrium Model of the Euro Area. 45–109. 39. “Private Investment and Macroeconomic Adjustment: A Survey. 15. NBER Macroeconomics Annual 2002. “Dynamic General Equilibrium Models of Imperfectly Competitive Product Markets. Solimano (1993). F. 199. (1939). E. MA: MIT Press. 312–320. and M. J. (1937). J. 5. pp. 188. K. 968–1006. Serven. A. 194. 185. Schorfhiede. 1123–1175. (1972). Slutsky. pp.” Journal of Political Economy. “Money. Stockman. “Econometric Issues in the Analysis of Equilibrium Business Cycle Models. DC: The World Bank. Cambridge. Tesar (1995). Rotemberg. and Banking. 85. F. “Has the Business Cycle Changed and Why?” In M. 1–48.” In J. “The Summation of Random Causes as the Source of Cyclic Processes. pp. Fallacy and Fantasy 182. 540–553.” Journal of the European Economic Association. Wouters (2003). Washington. 28. The Theory of Economic Growth. pp. pp. MN: Federal Reserve Bank of Minneapolis. Stock.” Journal of Monetary Economics. pp. 196. and A.” American Economic Review. Cambridge. and C. (1980). (1934). 195. Business Cycles. 361–387. 193. pp. Solow. Vol. Sargent. J. “Two Models of Measurement Error and the Investment Accelerator. 191. 645–670. New York: McGraw-Hill. 198. Income and Causality. 1. pp. Woodford (1995). Cooley (ed. “Business Cycle Modeling Without Pretending to Have Too Much A Priori Economic Theory. “Loss Function-based Evaluation of DSGE Models. Princeton: Princeton University Press. Taylor and M. pp. and R. 186. Wouters (2005). Woodford (1996).148 Business Cycles: Fact. 5. Sims (1977). 183. F. Watson (2005). Watson (2003). Macroeconomic Time Series. “Comparing Shocks and Frictions in US and Euro Area Business Cycles: A Bayesian DSGE Approach. L. Sims. Gertler and K. (1988). “Technical Change and the Aggregate Production Function. Rogoff (eds. 168–185. and R. J. 97. 48. pp. Sargent. 105–146. (1989). . pp. pp.” In New Methods in Business Cycle Research: Proceedings from a Conference.). 197. “Understanding Changes in International Business Cycle Dynamics. T. pp. Singleton. J. Minneapolis. 11–30. “Business Cycle Fluctuations in U. 187.” Econometrica.” Journal of Applied Econometrics. J. “Tastes and Technology in a Two-Country Model of the Business Cycle: Explaining International Comovements. and M.” In L. Sims. “Macroeconomics and Reality.). 161–183.).” Journal of Applied Econometrics. and M. Stock. Woodford (eds. MA: Harvard University Press. C. R. T. 251–287. 192. 190.).” Econometrica. 21.” Journal of the European Economic Association.” American Economic Review. 159–230. Schumpeter. (2000). 549–577. 20. Solimano (eds. C. 184.S. 3–64. Serven and A. and M. and M. and L. “Imperfect Competition and the Effects of Energy Price Increases on Economic Activity. pp. J.

Bibliography 149 201. Zarnowitz. Summers. 1102–1130. . pp. “Subjective Expectations and Asset-Return Puzzles. 204.” Journal of Political Economy. Santomero (eds. “Measures of Fit for Calibrated Models. Recursive Methods in Economic Dynamics. Litan and A. and Forecasting.” Available at http://www2. (1986). 206. D. Watson. Lucas.de/institute/wpol/html/toolkit. 281–372. 205. 203. 181–197.” Federal Reserve Bank of Minneapolis Quarterly Review. Finn Kydland and Edward Prescott’s Contribution to Dynamic Macroeconomics: The Time Consistency of Economic Policy and the Driving Forces Behind Business Cycles. Uhlig. Business Cycles: Theory. 1011–1041. and R. (2007). Tesar. “Statistical Nonlinearities in the Business Cycle: A Challenge for the Canonical RBC Model. History. Advance Report on the Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel. V.). 202. Chicago: The University of Chicago Press. M. with E. M. pp. Washington. M. Stokey.wiwi. 101. Vol. (1993). 92. pp. 207. Veracierto. “Some Skeptical Observations on Real Business Cycle Theory. (1992). and I. 208.htm/. Valderrama. 2957–2983. 211. 97. “Plant-Level Irreversible Investment and Equilibrium Business Cycles. 23–27. 210. Werner (1998). (2002). pp. 1.hu-berlin. L. pp. DC: The Brookings Institution.” Journal of Economic Dynamics and Control.” In R. The Royal Swedish Academy of Sciences (2004). pp. 110. 508–534. (2007). Thomas. L. (1997). Brookings-Wharton Papers on Financial Services. Weitzman. MA: Harvard University Press. 31(9). 209. (2002). 10. pp.” American Economic Review. N. J. “Is Lumpy Investment Relevant for the Business Cycle?” Journal of Political Economy. “The Internationalization of Securities Markets Since the 1987 Crash. Prescott (1989). “A Toolkit for Analyzing Nonlinear Dynamic Stochastic Models Easily. Cambridge. H. Indicators.” American Economic Review.

This page intentionally left blank .

94 animal spirits. 80 international risk sharing. 110 for large cross-sections. 124 credit market frictions. 138 deterministic steady state. 73 demand-constrained equilibrium. 39 dynamic factor analysis. 58 international portfolio diversification. 71 conditional heteroscedasticity. 129 identifiability of. 130 indivisible labor. 45 indirect inference. 124 . 54 home-produced good. 131 kurtosis. 16 leading indicator. 16 in emerging markets. 66 current account. 117 goodness-of-fit measures. 113 home production. 34 cross-country correlations. 59 irreversible investment. 49 inside money. 125 Kalman filter. 42 business cycles coincident indicator. 109. 73 common factors. 54 impulse response functions.Index accommodative monetary policy. 101 international. 110 complete contingent claims. 16 calibration. 135 dynamics of output. 43 capital account. 19. 116 151 dynamic stochastic general equilibrium (DSGE) models. 33 Bellman’s equation. 68 investment-specific technological change. 83 generalized method of moments (GMM). 58 aggregate price index. 120 enforcement constraints. 65 lagging indicator.

137 New Keynesian model. 78 optimal linear regulator problem. Fallacy and Fantasy Riccati equation. 56 . 66 price rigidities. 121 price anomaly. 88 vector autoregression (VAR). 38 trade balance. 33 reverse causality. 81 lottery models. 132 predictive density. 101 symmetric equilibrium. 53 industry-specific. 82 maximum likelihood estimation frequency domain. 38 spectral density function. 49 nonlinear time series models. 124 shocks. 50 Solow residual. 123 nonnormal disturbances. 131 unconditional moments. 77 nontraded good. 98 technology. 2 state space representation. 111 stabilization policy. 94 multi-sector models. 58 structural. 97 real business cycle (RBC) theory numerical solution. 101 transactions services from money and banking. 137 seminonparametric (SNP) model. 43 search models. 135 Sudden Stops. 44 unobservable index models. 78 energy. 132 structural models. 42 posterior distribution. 63 government consumption. 67 puzzles. 137 self-fulfilling expectations. 124 nonspecialization in endowments. 33 skewness. 49 Business Cycles: Fact. 33 nonconvexities in labor supply. 56 transition equation. 54 market incompleteness. 112 measurement equation. 132 monopolistic competition. 94 productivity spillovers. 66 real balances. 110 variable capacity utilization. 97 total factor productivity (TFP).152 labor hoarding. 48 quantity anomaly. 33 country-specific. 78 monetary. 124 social planner’s problem. 55 limited risk sharing. 137 multiple equilibria. 99 market good.

Sign up to vote on this title
UsefulNot useful