This action might not be possible to undo. Are you sure you want to continue?
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 CataloginginPublication Data
Altug, Sumru.
Business cycles : fact, fallacy, and fantasy / by Sumru G. Altug.
p. cm.
Includes bibliographical references and index.
ISBN13: 9789812832764 (hardcover)
ISBN10: 9812832769 (hardcover)
1. Business cycles. I. Title.
HB3711.A424 2009
338.5'42dc22
2009034620
British Library CataloguinginPublication 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 401402, 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 b808fm
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. Largescale computers in
the postWorld 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 “TimetoBuildandAggregate
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 b808fm
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 socalled 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 b808fm
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. InvestmentSpeciﬁc 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 b808fm
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 OpenEconomy 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 b808ch01
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 b808ch01
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 secondorder
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 policymakers.
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 shortrun 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 b808ch01
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 Phillipscurve 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, timeto
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 b808ch01
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
economywide averages (possibly fallaciously) in terms of the behavior of a
representative or standin 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 b808ch01
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 b808ch01
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808ch02
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 b808ch02
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 b808ch02
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 longrun 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 b808ch02
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 socalled
“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 postWWII 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
preWWII. One of the longest expansions to have occurred postWWII 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
detrending economic time series. One approach is to use a linear detrending
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 b808ch02
Facts 13
Watson [197] argue that neither linear time trends nor ﬁrstdifferencing 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 detrended series, whereas the second exacerbates the role
of shortterm 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 socalled 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
+µ
∞
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 detrending procedure depend on
the properties of the underlying data. If these are trendstationary, then the
detrending procedure has favorable properties; if the underlying data are
October 9, 2009 15:12 9in x 6in b808ch02
14 Business Cycles: Fact, Fallacy and Fantasy
differencestationary, however, application of the HP ﬁlter induces spurious
business cycle ﬂuctuations.
Asecond approach is based on the spectral analysis of economic time series.
The bandpass ﬁlter developed by Baxter and King [37] “ﬁlters” out both the
longrun trend and the highfrequency 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
bandpass ﬁlter of Baxter and King [37] is obtained by applying a K thorder
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 thorder 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
bandpass ﬁlter by considering lowpass and highpass ﬁlters with the required
properties. Let s denote the number of observations in a year. The bandpass
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
bandpass 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 timedomain
representation of the bandpass 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 b808ch02
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 ﬁniteorder
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 comovement 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 comovement of the cyclical behavior of individual series with
October 9, 2009 15:12 9in x 6in b808ch02
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 bandpass ﬁlter to identify the trend versus
cyclical components of each series. Among other measures, they consider the comovement of the cyclical
component of GDP with the cyclical component of each series, and the crosscorrelations 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
centurylong 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 preWorld War I era, the interwar era between
World War I and II, and the postWorld 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 b808ch02
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 downwardsloping 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 longterm interest rates relative to shortterm 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 b808ch02
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 preWWI 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 pricelevel 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 postwar
period than in the prewar 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 supplysidedriven 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 preWWII 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 b808ch02
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 postWWII 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 postWWII eras. If the higherquality data in the postWWII period are
subjected to a transformation that makes the pre and postWWII 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 postWWII period. Stock and
Watson [198] seek to avoid the problemof poor data quality in the preWWII
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 groupspeciﬁc 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 b808ch02
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 postWWII 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 ﬁrstorder 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 b808ch02
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, peaktopeak or troughtotrough. 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 b808ch02
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)
AreaWide 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 b808ch02
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
Englishspeaking 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 factorstructural 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 b808ch02
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
+ν
t
, (3.5)
where the vector of reducedform 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 countryspeciﬁc 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 hstepahead 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 tenyearlong 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 b808ch02
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 60odd 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
crosscountry 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 countryspeciﬁc 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 b808ch02
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
+φ
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
+φ
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 b808ch02
Facts 27
Köse et al.’s paper [138] shows that the world factor identiﬁes many of
the major cyclical events over a 30year 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 mid1970s. 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 comovement across countries and aggregates is captured by the
world factor. By contrast, the regional and country factors explain covariation
or comovement 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 b808ch02
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 b808ch02
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 postWorld 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 supplyside 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 socalled concordance correlation, which
examines whether the turning points in the different series occur at similar
dates. Instead of using information on NBERtype reference dates, Bordo
and Heibling [44] use real GDP or industrial production series to determine
synchronization. They also use standard output correlations and factorbased
measures.
October 9, 2009 15:12 9in x 6in b808ch02
30 Business Cycles: Fact, Fallacy and Fantasy
To brieﬂy describe the concordance correlations, let S
it
and S
jt
denote
binarycycle 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 meancorrected 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 postBretton 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 b808ch02
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 AngloSaxon countries.
The authors also estimate a socalled 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 socalled 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 b808ch02
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808ch03
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 b808ch03
34 Business Cycles: Fact, Fallacy and Fantasy
Depression and the recent ﬁnancial crises also indicate the importance of credit
market frictions and ﬁnancial acceleratortype 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 openeconomy
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 constantreturnstoscale (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 b808ch03
Models of Business Cycles 35
Turning to the basic model, the representative agent has timeseparable
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 economywide 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 inﬁnitehorizon 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 b808ch03
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 ﬁrstorder 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 capitallabor 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
capitallabor 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 consumptionleisure
October 9, 2009 15:12 9in x 6in b808ch03
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 b808ch03
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 b808ch03
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 b808ch03
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 steadystate capitallabor 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 b808ch03
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 secondorder 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 loglinearize the model. See Campbell [55].
October 9, 2009 15:12 9in x 6in b808ch03
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
+ε
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 b808ch03
Models of Business Cycles 43
The ﬁrstorder 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
socalled 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 b808ch03
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 crosscorrelations 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 laborleisure 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
+ˆ
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 b808ch03
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 steadystate 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 humpedshaped 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 b808ch03
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 timeto
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 ﬁrstorder 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, economywide
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 supplyside
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 b808ch03
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 onesixth, not onethird. 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 b808ch03
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 twothirds
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 establishmentlevel surveys conducted by the US Department of Labor.
October 9, 2009 15:12 9in x 6in b808ch03
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
moneyoutput 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 b808ch03
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
ﬁrstorder conditions with respect to (c
0,t
, c
1,t
) are
1 − π
t
c
0,t
= (1 − π
t
)λ
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 b808ch03
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 ﬁrstorder 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 b808ch03
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 b808ch03
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 timeseparable 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 b808ch03
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 downwardsloping 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 economywide 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 decisionmaker 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 homeproduced 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 b808ch03
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 b808ch03
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 b808ch03
Models of Business Cycles 57
the ﬁnal good is produced according to the production technology
y
t
= f (k
ft
, n
ft
, d
ft
)φ
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
)λ
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 periodbyperiod
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
ﬁrstorder 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 b808ch03
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 longrun restrictions to identify the alternative structural shocks.
Consistent with the RBC view, they ﬁnd that an important source of the
moneyoutput 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 b808ch03
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 comovements 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. InvestmentSpeciﬁc Technological Shocks
An important source of shocks is the investmentspeciﬁc 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 equipmentGNP ratio increased
substantially. Furthermore, there was a negative correlation (−0.46) between a
detrendedmeasure 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 investmentspeciﬁc 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 nonnegligible effects on real
economic variables.
11
In Greenwood, Hercowitz, and Krusell [106], the authors use a growth accounting approach to
show that investmentspeciﬁc shocks can account for 60% of postwar US growth of output per manhour.
October 9, 2009 15:12 9in x 6in b808ch03
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 timevarying 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 investmentspeciﬁc 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 ﬁrstorder 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 b808ch03
Models of Business Cycles 61
returned to the private sector through lumpsum 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 investmentspeciﬁc 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, investmentspeciﬁc
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 qualityadjusted equipment and
the price index for consumption. The parameters that are determined using
a priori information include the average growth rate of the investmentspeciﬁc
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
(γ
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 b808ch03
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 longrun 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 investmentspeciﬁc 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 investmentspeciﬁc 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 investmentspeciﬁc
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 investmentspeciﬁc shocks contribute to
business cycle ﬂuctuations, they are not the main factor.
October 9, 2009 15:12 9in x 6in b808ch03
Models of Business Cycles 63
3.5.2. Energy Shocks
From the beginning, the source of large supplyside 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 b808ch03
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 economywide 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 valueadded 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 b808ch04
Chapter 4
International Business Cycles
The international business cycle literature provides an extension to the
original real business cycle (RBC) approach by allowing for openeconomy
considerations. One of the aims of the RBC approach has been to determine
whether twocountry versions of the basic model can account for both the
comovements studied in closedeconomy models as well as international
comovements, including correlations of macroeconomic aggregates across
countries and movements in the balance of trade. The closedeconomy 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 closedeconomy model, openeconomy 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 socalled
stylized facts of international business cycles. Backus, Kehoe, and Kydland [25]
consider data on12developedcountries over a sample perioddating from1960
to 1990. They consider a twocountry 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 b808ch04
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
offdiagonal 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 crosscountry correlations of
output are greater than those of productivity, measured as the Solow
residual, and the crosscorrelations 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 b808ch04
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 crosscountry 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 crosscorrelations 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
twocountry frictionless international business cycle model also implies that
consumers in different countries can share risk perfectly. This feature of the
model leads to the crosscountry 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 b808ch04
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 offdiagonal 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 crosscountry 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 crosscountry 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 statecontingent claims does not help to lower the
crosscountry 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 crosscountry 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 twocountry 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 b808ch04
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 ﬁrstorder 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 timeinvariant 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 b808ch04
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
)β
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 timeseparable, 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 ﬁrstorder 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 b808ch04
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 crosscountry 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 b808ch04
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 oneperiod 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 oneperiod 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 b808ch04
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 marketclearing 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
_
+β
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 ﬁrstorder conditions with
October 9, 2009 15:12 9in x 6in b808ch04
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
)λ
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
)λ
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 b808ch04
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 marketclearing
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
assettrading 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 b808ch04
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 marketclearing 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 noassettrading 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 b808ch04
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 b808ch04
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 countryspeciﬁc shocks, which
affect all sectors within a country, and industryspeciﬁc shocks. Shocks to y
1,t
or y
2,t
are, by deﬁnition, countryspeciﬁc shocks; whereas shocks to w
1
t
, w
2
t
,
where w
1
t
, w
2
t
are not perfectly correlated, are sectorspeciﬁc shocks. Hence,
when there are sectorspeciﬁc 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 b808ch04
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 crosscountry 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 b808ch04
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 ﬁrstorder 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 b808ch04
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 b808ch04
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 noncontingent bonds can be traded internationally. These authors ﬁnd
that incomplete asset markets help to reduce the crosscountry correlation
of consumption, but the crosscountry correlations of output, investment,
and hours worked remain counterfactually negative. Moreover, the results for
the crosscountry 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 twocountry 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 ondebtconstrained
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 crosscountry
correlations of consumption and higher crosscountry 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 constantreturnstoscale 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 b808ch04
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
)β
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 b808ch04
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
)µ
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
)µ
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
))
+µ
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 b808ch04
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 offdiagonal 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 offdiagonal elements equal to zero,
and a third one allows for high spillover with the diagonal elements equal to
0.85 and the offdiagonal 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 crosscountry 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 b808ch04
86 Business Cycles: Fact, Fallacy and Fantasy
consumption correlations are both positive and closer to each other, although
the crosscountry correlation of consumption is greater than that of output.
Second, the crosscountry 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 crosscorrelations 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 crosscountry 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
highcrosscountry 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 b808ch04
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 risksharing 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 crosscountry 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] reexplore 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 socalled “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 twocountry model by adding multiple
October 9, 2009 15:12 9in x 6in b808ch04
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 onesector models. It also predicts
crosscountry 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 reestablishing 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 sectorspeciﬁc 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 crosscountry correlation of
aggregate consumption, and greatly overstates the crosscountry correlation
of tradable consumption. Hence, they ﬁnd that introducing nontradable
goods does not sufﬁce to reduce the crosscountry 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 nationspeciﬁc 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 comovement
of factor inputs across countries. We already considered the role of variable
factor utilization in reconciling the behavior of procyclical productivity in
closedeconomy 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 b808ch04
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 oneperiod,
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 b808ch04
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 crosscorrelations of
the factor inputs. Speciﬁcally, the crosscorrelation 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 crosscountry 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 b808ch04
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 crosscountry 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 b808ch04
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808ch05
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 openeconomy
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 b808ch05
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
+ϒ
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 b808ch05
New Keynesian Models 95
The ﬁrstorder 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
(µ
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 periodbyperiod budget
constraint. We can solve for µ
t
from the ﬁrstorder 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
{η
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 b808ch05
96 Business Cycles: Fact, Fallacy and Fantasy
Lagrange multiplier on the technology constraint. The ﬁrstorder 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 ﬁrstorder 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 ﬁrstorder 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 b808ch05
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 marketclearing conditions,
implies that P
t
Y
t
= M
t
. Using the ﬁrstorder 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
+γ
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
/λ
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 pricesetting rule in
(1.22) together with (1.12) and the expression for equilibrium consumption
October 9, 2009 15:12 9in x 6in b808ch05
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
+γ
_
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 shortrun increasing returns to labor.
Finally, the price level responds oneforone to an increase in ξ
t
, though with
a oneperiod lag.
A positive technology shock deﬁned by η
t
> 0 has a permanent positive
oneforone 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 shortrun 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 b808ch05
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 longrun 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 nontechnology 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 nontechnology
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 longrun 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 loglevels as opposed to differences.
Chari, Kehoe, and McGrattan [63] have challenged the ﬁndings derived
from socalled 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 SVARbased
test, the differencestationary version of the model implies that the response
3
The value of the matrix C(L) evaluated at L = 1 gives the longrun multipliers for the model, i.e.,
the longrun impact of a given shock.
October 9, 2009 15:12 9in x 6in b808ch05
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, nonconstant returns
to scale, variable factor utilization, and sectoral reallocation and aggregation
effects. Unlike the SVAR approach, the evidence obtained from this approach
is robust to longrun 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, nonresidential investment,
durables and residential investment, nondurables 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 nonresidential
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 b808ch06
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 RBCtype 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
economytype 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 onesector 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 b808ch06
102 Business Cycles: Fact, Fallacy and Fantasy
[1] and GarciaCicco, Pancrazi, and Uribe [101] have examined versions of
a small openeconomy business cycle model with permanent and transitory
shocks to account for emerging versus developed economy experiences.
6.1. A SMALL OPENECONOMY MODEL OF EMERGING
MARKET BUSINESS CYCLES
Aguiar and Gopinath [1] present a small openeconomy 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 longrun productivity growth. Since productivity and output depend
on the cumulation of the shocks g
t
, a stationarityinducing 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 b808ch06
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 socalled 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
+β
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
{β
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 economywide 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 oneperiod debt that sells for the price
October 9, 2009 15:12 9in x 6in b808ch06
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 steadystate 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 ﬁrstorder 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 b808ch06
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)
+φ
_
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
+φ
_
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 loglinear
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 b808ch06
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 exportstoGDP, 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
/σ
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 sideby
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 GarciaCicco 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 b808ch06
Business Cycles in Emerging Market Economies 107
a short sample to estimate lowfrequency movements in productivity. By
contrast, GarciaCiccoet 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 postWorld 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. GarciaCicco 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 postWorld
War II period are as large as those in the preWorld 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 secondorder autocorrelations of output growth,
consumption growth, investment growth, and the trade balancetooutput
ratio; the correlations of output growth with consumption growth, investment
growth, and the trade balancetooutput ratio; and the unconditional mean
of output growth.
First, unlike Aguiar and Gopinath [1], GarciaCicco 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 consumptionsmoothing 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 balancetooutput 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 b808ch06
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 balanceto
output ratio. In particular, the trade balancetooutput ratio displays a near
random walk behavior even though the estimated autocorrelation function in
the data is downwardsloping. The authors also show that the random walk
behavior of the trade balancetooutput 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 balanceto
output ratio inherits the behavior of output. By contrast, if the shocks are
transitory, the behavior of the trade balancetooutput 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.
GarciaCicco 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 b808ch07
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 comovement of
109
October 9, 2009 15:12 9in x 6in b808ch07
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 socalled 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 lowdimensional 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 ndimensional mean zero, covariance stationary stochastic
process used to describe observations on the (possibly detrended) values of a
set of variables. A kfactor 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 lowdimensional set of unobserved common factors and
October 9, 2009 15:12 9in x 6in b808ch07
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 nonsingular 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 welldeﬁned 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 b808ch07
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 wellknown 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 crossequation 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 b808ch07
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 goodnessofﬁt 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 secondmoment 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 b808ch07
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 errorsinvariables 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
σ
xy
 ≤ σ
x
σ
y
:
min
σ
xy
σ
2
u
= σ
2
y
+σ
2
x
− 2σ
xy
s.t. σ
xy
 ≤ σ
x
σ
y
. (1.7)
3
In a standard goodnessofﬁt 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 datagenerating 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 b808ch07
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
/σ
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 goodnessofﬁt measures
for a standard RBCmodel with a stochastic trend in technology. Speciﬁcally, he
considers the model in King, Plosser, and Rebelo [131, 132]. He loglinearizes
October 9, 2009 15:12 9in x 6in b808ch07
116 Business Cycles: Fact, Fallacy and Fantasy
the ﬁrstorder conditions to obtain the solution for logdifferences 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 crosssections. 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 economywide shocks can be identiﬁed
using structural vector autoregression (SVAR) techniques, and the unobserved
factor model can be estimated by using equationbyequation ordinary least
squares (OLS). They examine the behavior of fourdigit industrial output and
productivity for the US economy for the period 1958–1986 and ﬁnd evidence
in favor of at least two economywide shocks, both having a longrun effect on
sectoral output. However, their results also indicate that sectorspeciﬁc 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 b808ch07
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 hoursproductivity 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 comovement 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
ﬁrstmoment 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 b808ch07
118 Business Cycles: Fact, Fallacy and Fantasy
The elements of
2
consist of the standardRBCsecondmoment restrictions as
E
_
y
2
t
(σ
x
/σ
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
/σ
y/n
_
2
− n
2
t
_
= 0, (2.14)
E
__
σ
2
n
/(σ
n
/σ
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
/σ
y/n
. To do this, they use a
Waldtype 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
/σ
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
/σ
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 b808ch07
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 secondmoment 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 GarciaCicco, 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 ﬁrstorder 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 b808ch07
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 twoshock 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 differencestationary, 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 twovariable 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 b808ch07
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 humpshaped 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(βI , 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
, βf , I ) denote the joint density
for the data and the parameters. Denote the predictive density p(X
t
I , f ) =
_
H(X
t
, βI , 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 b808ch07
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
, βI , 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(βI , 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(βI , 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(βI , 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 b808ch07
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 secondmoment 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 timevarying 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 b808ch07
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 secondmoment 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–Mirmantype 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 twostep 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 ARCHGARCH structure to model the
conditional variance, and it allows for a nonGaussian 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 b808ch07
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 b808ch07
126 Business Cycles: Fact, Fallacy and Fantasy
multisector 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 pricesetting 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 b808ch07
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
frequencydomain 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 b808ch07
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 clearcut 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 wellspeciﬁed 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 b808ch07
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 interestratesetting rule that deﬁnes
October 9, 2009 15:12 9in x 6in b808ch07
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 socalled 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 statecontingent securities for consumption.
Christiano et al. [67] derive much of their evidence regarding the types
of real rigidities on the results of calibrationtype 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 timevarying 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 b808ch07
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 highdimensional 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 b808ch07
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(θy
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 FernandezVillaverde [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 b808ch07
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 investmentspeciﬁc shock, and a government consumption
shock plus three “costpush” 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
shortterm 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 FernandezVillaverde [90].
October 9, 2009 15:12 9in x 6in b808ch07
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 pricesetting 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 b808ch07
Matching the Model to the Data 135
Should economists build intuition from the workings of wellarticulated
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 policymakers 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. FernandezVillaverde and RubioRamirez
[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 inCalvotype 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 b808ch07
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:12 9in x 6in b808ch08
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 thoughtprovoking. There are many other topics of interest that
we could have discussed regarding business cycles. For one, I have omitted a
discussion of multisector 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 multisector models since the onesector growth model does
not allowfor a considerationof these issues. However, multisector 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 selffulﬁlling 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 b808ch08
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 steadystate 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 demandconstrained equilibrium. The policy implications of such a model
differ from those of the standard textbook Keynesian model, which seeks to
alleviate demandinduced shortfalls in output through large ﬁscal stimuli. By
contrast, the model with a search externality emphasizes the role of conﬁdence
in restoring fullemployment 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 b808bib
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). “TimetoBuild 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 Multisector 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 b808bib
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/NorthHolland, 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 withNontraded 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: NorthHolland,
pp. 1801–1864.
October 9, 2009 15:13 9in x 6in b808bib
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 SectorSpeciﬁc 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, OpenEconomy Models and Their
Implications for Monetary Policy,” Federal Reserve Board (FRB) International Finance
Discussion Paper No. 2003762.
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 UtilityMaximizing Framework,” Journal of
Monetary Economics, 12, pp. 383–398.
October 9, 2009 15:13 9in x 6in b808bib
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 b808bib
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 CrossCountry Consumption
Correlations in a TwoCountry, 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 AreaWide 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 SelfFulﬁlling 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. FernandezVillaverde, J. (2009). “The Econometrics of DSGE Models,” NBERWorking
Paper No. 14677.
91. FernandezVillaverde, J. and R. RubioRamirez (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 b808bib
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. GarciaCicco, 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). “LongRun Implications
of InvestmentSpeciﬁc Technological Change,” American Economic Review, 87,
pp. 342–362.
107. Greenwood, J., Z. Hercowitz, and P. Krusell (2000). “The Role of InvestmentSpeciﬁc
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 b808bib
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 PolicyOriented 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 CountrySpeciﬁc 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 b808bib
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). “TimetoBuild 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: NorthHolland,
pp. 7–29.
152. Lucas, R. (1982). “Interest Rates and Currency Prices in a TwoCountry 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 ‘StickyPrice’ Theories of the Business Cycle,”
Journal of Money, Credit, and Banking, 18, pp. 397–414.
157. McCallum, J. (1996). “National Borders Matter: CanadaUS 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 b808bib
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,” CarnegieRochester
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 b808bib
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 Functionbased 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: McGrawHill.
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 TwoCountry Model of
the Business Cycle: Explaining International Comovements,” American Economic Review,
85, pp. 168–185.
October 9, 2009 15:13 9in x 6in b808bib
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.), BrookingsWharton 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.huberlin.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). “PlantLevel 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 AssetReturn 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 b808bib
This page intentionally left blank This page intentionally left blank
October 9, 2009 15:13 9in x 6in b808Index
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
crosscountry correlations, 66
current account, 73
demandconstrained equilibrium, 138
deterministic steady state, 39
dynamic factor analysis, 110
for large crosssections, 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
goodnessofﬁt measures, 113
home production, 54
homeproduced good, 54
impulse response functions, 45
indirect inference, 130
indivisible labor, 49
inside money, 58
international portfolio diversiﬁcation, 80
international risk sharing, 68
investmentspeciﬁc technological change, 59
irreversible investment, 125
Kalman ﬁlter, 131
kurtosis, 124
151
October 9, 2009 15:13 9in x 6in b808Index
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
multisector 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
selffulﬁlling expectations, 137
seminonparametric (SNP) model, 124
shocks, 33
countryspeciﬁc, 78
energy, 63
government consumption, 53
industryspeciﬁc, 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
Turkey & Centre for Economic Policy Research.Sumru G Altug Koç University. F World Scientific NEW JERSEY • an F LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TA I P E I ta • sy CHENNAI . UK ess sin Bu les Cyc and y allac Fact.
5'42dc22 2009034620 British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library. USA. Singapore 596224 USA office: 27 Warren Street. 222 Rosewood Drive. London WC2H 9HE Library of Congress CataloginginPublication Data Altug. or parts thereof. I. Danvers. Business cycles : fact. Copyright © 2010 by World Scientific Publishing Co. without written permission from the Publisher..com Printed in Singapore. may not be reproduced in any form or by any means. including photocopying. please pay a copying fee through the Copyright Clearance Center. Hackensack. Includes bibliographical references and index. and fantasy / by Sumru G. Inc. Covent Garden. Pte. For photocopying of material in this volume. Business cycles. HB3711. In this case permission to photocopy is not required from the publisher. 5 Toh Tuck Link. Ltd. p. Title. fallacy. Typeset by Stallion Press Email: enquiries@stallionpress. electronic or mechanical. MA 01923. This book. cm. Pte. NJ 07601 UK office: 57 Shelton Street. ISBN13: 9789812832764 (hardcover) ISBN10: 9812832769 (hardcover) 1. recording or any information storage and retrieval system now known or to be invented. All rights reserved.A424 2009 338. Suite 401402. Sumru. . Altug.Published by World Scientific Publishing Co. Ltd.
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 2004. The oil shocks of the 1970s and the 1980s affected economists’ views regarding the sources of macroeconomic ﬂuctuations. 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 recent years. The model was rejected. in particular — is affected by major changes in economic conditions. The Great Depression greatly inﬂuenced the perceptions of a generation of economists. This book draws upon Kydland and Prescott’s original contribution. Sometimes. This debate continues to this day.Preface How do intellectual disciplines progress? Undoubtedly. v . The ﬁeld of macroeconomics has changed signiﬁcantly due to Kydland and Prescott’s contributions.D. Largescale computers in the postWorld War II era played an important role in the development of simultaneous equation models. I was a Ph. student at the Graduate School of Industrial Administration at Carnegie Mellon University when Kydland and Prescott’s “TimetoBuild and Aggregate Fluctuations” article was published in the early 1980s [141]. the discipline of economics — and macroeconomics. My thesis was on estimating the model in the same article. real business cycle (RBC) analysis has come to provide a ﬂexible and popular approach for examining macroeconomic phenomena. the development of new techniques or new ways of modeling can also affect the course that a discipline takes. beginning with Keynes. Finn Kydland and Edward Prescott received the Nobel Prize in Economics.
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. . Jordi Gali’s [97] recent text articulates an alternative New Keynesian framework for describing aggregate ﬂuctuations. James Hartley. This book takes a more eclectic approach. asking some basic questions about RBC analysis and summarizing the ongoing controversies surrounding it. researchers at central banks have begun using socalled dynamic stochastic general equilibrium (DSGE) models for policy analysis. many researchers also pursued the econometric analysis of RBC models. Not content with the initial rejection of the model. In recent years.vi Business Cycles: Fact. and Kevin Salyer’s [116] collection of articles provides a critique of the calibration approach. Fallacy and Fantasy Kydland and Prescott’s seminal article initiated the school of RBC analysis. This literature evolved in different ways. in many ways. reﬂects my own interests. Talented and creative individuals extended the initial Kydland–Prescott research in different ways. This book attempts to provide an overview of the burgeoning business cycle literature that. Kevin Hoover.
.2. . . . . . . . . . . . . . . . . . . .2. .5. .4.2. . . . . . . . . The Euro Area Business Cycle . . . . . . . . . . . . . .1. . . . . . Is There a World Business Cycle? 2. . . . . 4. . . . . . 2. . . . Reverse Causality . . . . . . . . . 4. . . . . . . . . . . . . 2. . .5. . The Role of International Risk Sharing 4. 3. . . . . . . Deﬁning a Business Cycle . . International Business Cycles 4. 3.3. . . . 3. . . . . . . . . . Models of Business Cycles 3.1. . . . . . . . . . .5. . . Facts . .2.3. . . . . . . . . 3. .4. . . . . . . . . . . . . .2. . . . Initial Criticisms . . . . . . . . . . . Stylized Facts . . . . . . . A Numerical Solution . . . . . . . . . . A Model with Indivisible Labor Supply . .4. . . . . Introduction Facts 2. . . . . . . . .1. . . . . . . . 2. Energy Shocks . . .2. . . .3. . . . . . . .4. . . . . . . . . 3. 3. . . . . . . . . 65 68 69 71 . Pareto Optimal Allocations . . . .Contents Preface 1. . 3. . . . . . . . . 3. . . . . . .4. InvestmentSpeciﬁc Technological Shocks 3. . . Historical Business Cycles .2. . Complete Contingent Claims vii v 1 7 .1. . The Source of the Shocks . . . . . . . . . . . . . . 2. . . . . . . “Puzzles” . .2. . . .1. . . . . . . . 4. . . An RBC Model . .5. . . . . . . .1. . . . . . . . . The Productivity Puzzle . 3. 7 15 19 25 28 33 34 39 46 48 49 52 56 58 59 63 65 . . .
. Puzzles Revisited . . . . Do Shocks to Trend Productivity Explain Business Cycles in Emerging Market Economies? . 7. . .3. . . . .2. . Fallacy and Fantasy 4. . . . . . .1. . .1. 5. . Trade in Equity Shares . . . . . . .3.1. . 7. . . . . . . . . . . . . . . . . . . . . . . . . . Nontraded Goods . . . . . .2. . . 4. . . . . GMM Estimation Approaches . . . .4. . 4. . 7. Empirical Evidence . . .viii Business Cycles: Fact. . A Small OpenEconomy Model of Emerging Market Business Cycles . . . . . . .3. . . . . . . . . . 8. . . . 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. . . . . . . . Matching the Model to the Data 7. . Nonspecialization in Endowments 4. . . . Business Cycles in Emerging Market Economies 6. . Other Applications . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . 6.3. . .2. . .2. . . . . . 7. . . . . . . . . .3. Future Areas for Research Bibliography Index .3. . . . . . Other Extensions .2.4. . .4. The Calibration versus Estimation Debate . . . .3. . . . . . . . 4. . . . . . . . . .5. . . . . .1. . . . The Basic Model . . . Measures of Fit for Calibrated Models .3. . . . . . Nonlinearity in Macroeconomic Time Series 7. . 7. . . The Dynamics of Output . Limited Risk Sharing . No Asset Trading . . . 6. . .2. . . . . . .1. . . . . . . . . . . . .6. . . . . . . . . 4. .4. . . . .2.1. . . . . . . . . . . .7. .1. . . . . . . Calibration as Estimation . . . . DSGE Modeling . 7. . . 5. .2. . . . 7. . . . . . . . . . . . . . . . . . . Dynamic Factor Analysis . . .2. . . . . . 7. . . . . . The Debate Reconsidered . . 7. . .
the notion of business cycles originated from various types of panics. Kondratiev [137] argued that in addition to shorter economic cycles. While some scholars proposed different theories to explain ﬂuctuations in economic activity. and motor engines. corresponding to the introduction of steel and steam engines. “crises” are an endemic feature of capitalist economies. corresponding to the Industrial Revolution. 1840–1890. This book describes these new developments. there have been many developments in the ﬁeld of business cycles. Schumpeter [187. there were periodic movements or “long waves” in economic variables. According to Karl Marx. when one examined the history of commercial cycles for the capitalist economies of the time. other scholars investigated methods for the systematic 1 . As Mitchell [163] recounts. much effort was devoted to understanding the causes of what many viewed as “abnormal” phenomena. and 1890–1950. chemical processing. both growth and business cycles could be ascribed to the process of innovation. In the 1920s. Wesley Mitchell [163] comments as thus: “As knowledge of business cycles grows. He identiﬁed three long waves: 1780–1840. corresponding to the invention of electricity.Chapter 1 Introduction In the opening page of the book Business Cycles published in 1927. other economists observed that the alternating phases of prosperity and depression seemed to follow each other on a regular basis. more effort is required to master it. Historically.” Ever since. In his framework. and crises experienced by market economies in the 19th and early 20th centuries. one of the most prominent thinkers of the time. depressions. However. 188] sought to explain the existence of such long waves as an outcome of technological innovations.
Post World War II. The Norwegian economist Ragnar Frisch [95] developed the notions of impulse and propagation mechanisms for describing business cycles. 149] developed monetary models of the business cycle as a way 1 See also Zarnowitz [211]. Fallacy and Fantasy measurement and identiﬁcation of business cycles. In their seminal contributions. the Keynesian framework was interpreted as a model of output determination at a point in time. Slutsky [193] argued that the sum of a number of uncorrelated shocks could produce serially correlated or smooth movements in the generated series. which continue to form the main vehicle for empirically studying business cycles. post World War II. The Great Depression and World War II were two major events in the development of business cycle analysis. Their inﬂuence on thinking about business cycles has persisted to this day. 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 shortrun economic ﬂuctuations. a business cycle is the simultaneous downturns and upturns of a large number of economic series. Their ideas were formulated in terms of linear time series models. According to them. and modeled business cycles as the response of a secondorder dynamic system to random shocks.1 We discuss the NBER methodology and the stylized facts of business cycles in Chapter 2. the lessons of the Great Depression were vivid in the minds of many policymakers. Phelps [170] and Lucas [148. During this time. the focus shifted to stabilization policy. and it continues to this day in the business cycle dating methodology employed by the NBER.2 Business Cycles: Fact. or stagﬂation as it is popularly known. After the early work of Burns and Mitchell. Writers such as Frisch and Slutsky embedded the notion of business cycles into simple dynamic systems driven by stochastic shocks. Nevertheless. Burns and Mitchell [50] and researchers at the National Bureau of Economic Research (NBER) began to identify the phenomena of business cycles. Their work involved the dating of business cycles and the development of leading indicators for the US economy. The period following World War II was an era of high and sustained growth in many countries. led researchers to account for the observations using new mechanisms for the effects of money on output. The oil shocks of the 1970s and the experience of high inﬂation and high unemployment.
During this period. In Phelps’ and Lucas’ framework. namely. the durability of leisure. Widely known as the calibration approach. we cannot ignore the calibration approach or. The literature on real business cycle (RBC) theory owes its existence to Kydland and Prescott [141]. and inventories.Introduction 3 of providing a consistent theoretical foundation for describing the impact of changes in money on output. One of the major issues with Kydland and Prescott’s contribution. failed to garner sufﬁcient empirical support. They also proposed a methodology for confronting their theory with the data. this involves matching a small set of moments implied by the model with those in the data. the debate on the empirical validation of business cycle models. Though this approach is cited extensively. 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. was identifying technology shocks that could generate cyclical ﬂuctuations of magnitudes observed in the data (see Summers [202]). In his article “Understanding Business Cycles”. 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. timetobuild in investment. more generally. Since this book purports to discuss 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. We will discuss RBC models further in Chapter 3. the speciﬁc mechanisms postulated in this literature as leading to business cycles. unanticipated shocks to money. there was a revival of interest more generally in examining aggregate economic activity as recurrent phenomena characterizing the functioning of economies with optimizing agents. the merits of this approach have been a topic of debate. as identiﬁed by skeptics. In some sense. this feature of Kydland and Prescott’s analysis . 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. Despite providing great theoretical advances in the analysis of aggregative phenomena. the emphasis was on generating the Phillipscurve type of phenomena between inﬂation and unemployment based on informational frictions.
Fallacy and Fantasy was viewed as “fantastical” by many. More recently. it introduces alternative mechanisms for generating price stickiness such as imperfect competition among ﬁrms. On the one hand. is known to be procyclical. The New Keynesian viewpoint breaks with the RBC approach by contesting the view that prices adjust frictionlessly to clear markets. A more general critique to RBC analysis was mounted by the New Keynesian challenge. the original Kydland–Prescott model could not explain the relative variability of hours and productivity or real wages. The model lacked money. 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.4 Business Cycles: Fact. or the Solow residual. Heathcote and Perri [117]). Instead. Other skeptics contested the implications of the RBC approach for the observed behavior of productivity. endogenous changes in efﬁciency due to increasing returns to scale. and variable factor . Subsequent research that evolved from the original Kydland–Prescott exercise has proceeded along several different dimensions. It also failed to account for the correlation between hours and productivity. In a series of papers. A deeper problem lay in modeling the movement of economywide averages (possibly fallaciously) in terms of the behavior of a representative or standin household. 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. and international capital ﬂows. 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. markups. The RBC literature has also generated international business cycle models to replicate ﬁndings on current account dynamics. According to the RBC approach. the observed procyclical movements in productivity should merely be a response to exogenous technology shocks (see Prescott [173]). ﬁnancial diversiﬁcation. models have been developed that study the role of market completeness/incompleteness on cyclical ﬂuctuations (see. for example. it faced the problem of reconciling observations on money–output correlations within a model where the main driving force was real productivity shocks. Productivity. hence. 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. international risk sharing. For example. Hall [110.
Following Sims [191]. More recently. vector autoregression (VAR) and structural VAR (SVAR) have also proven to be popular in empirical macroeconomic research. Smets and Wouters [194. . 2 Canova [58] is an excellent reference source on the quantitative and empirical analysis of dynamic stochastic general equilibrium models. The controversy over the calibration approach also resulted in work on alternative methods for empirically analyzing business cycle phenomena. dynamic stochastic general equilibrium (DSGE) models have been developed to identify shocks and propagation mechanisms of business cycles models (see. or monetary versus technology shocks. Fernald. SVAR also permits an identiﬁcation of shocks.2 We will discuss the issues involved in matching the model with the data in Chapter 7. This facilitates the analysis of the different effects of government versus technology shocks. for example.Introduction 5 utilization. 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. 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. 195]). The New Keynesian challenge has proceeded along theoretical lines (see Rotemberg and Woodford [182]) and empirical considerations (see Gali [96] or Basu. We will discuss New Keynesian models in detail in Chapter 5. and Kimball [32]).
This page intentionally left blank .
We will also discuss some of the empirical ﬁndings in these regards. 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. This framework is based on the principle of identifying turning points in economic activity and determining which series constitute leading. or lagging indicators of the business cycle.1. coincident. a cycle consists of expansions occurring at about the same time in many 7 . However.Chapter 2 Facts The vast literature on business cycles has focused on generating the stylized facts regarding cyclical ﬂuctuations. Kontolemis. 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. and Stock and Watson [199]. Mitchell [163]. among others). and Burns and Mitchell [50] provide a framework to describe the main features of business cycles. In this chapter. Much of the early work on business cycles was implemented for the US economy. the European or euro area business cycle has also been the topic of much recent study (see Artis. Mitchell and Burns [164]. 2. 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. Artis and Zhang [18]. and Osborn [20].
the cyclical behavior of each series is then examined relative to the reference cycle. Some of these questions are precisely the ones that we seek to answer in this book. 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. Mitchell and Burns [164]. and amplitude of each speciﬁc cycle are compared with that of the reference cycle. Ch. In their words (Burns and Mitchell [50]. Burns and Mitchell [50] themselves noted that this deﬁnition raised as many questions as it sought to answer. how should we deal with seasonal changes. then an aggregate business cycle or a reference cycle is identiﬁed. whether it pertains to the measurement of business cycles or the construction of models of cyclical ﬂuctuations. If one talks about “ﬂuctuations in the aggregate economic activity of nations”. when one considers the statement regarding expansions occurring in “many economic activities”. The NBER approach to identifying business cycles as outlined by Mitchell [163]. the duration. or secular trends? Finally. If the answer to the latter question is in the afﬁrmative. this sequence of changes is recurrent but not periodic. This deﬁnition has formed the basis of modern thinking about business cycles. As part of this analysis.8 Business Cycles: Fact. and Burns and Mitchell [50] is comprised of two mutually reinforcing acts: ﬁrst. 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. Once the reference dates are found. they are not divisible into shorter cycles of similar cycles with amplitudes approximating their own. contractions. 2. Fallacy and Fantasy economic activities. and second. in duration business cycles vary from one year to ten or twelve years. and also lay down rules for excluding irregular movements and other similar changes. determine whether these turning points are sufﬁciently common across the series. 14): . timing. Burns and Mitchell [50] stress that the notion of a reference cycle should not be equated with an observable construct. followed by similarly general recessions. p. random ﬂuctuations. 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. the comments regarding the duration and amplitude of business cycles are based on actual observations of cyclical phenomena. In seeking to identify “recurrent changes”.
we say that it is Fig. . business cycles can be seen only ‘in the mind’s eye’. Classical and Growth Cycles. By contrast. classical cycles typically have recessions that are shorter than expansions because of the growth effect. One advantage of using growth cycles is that they have expansions and contractions that are approximately of the same duration. For.Facts 9 When we speak of ‘observing’ business cycles we use ﬁgurative language. Figure 2.1 displays the difference between classical and growth cycles. The NBER business cycle methodology identiﬁes a business cycle based on the (absolute) downturn of the level of output. such cycles are known as growth cycles. There is an alternative approach which considers the decline in the series measured as a deviation from its longrun trend. This is known as a classical business cycle. Following the terminology in Zarnowitz [211]. What we literally observe is not a congeries of economic activities rising and falling in unison. By contrast. When the economy is moving from a trough to a peak. like other concepts. but changes in readings taken from many recording instruments of varying reliability.1. a trough occurs at point C for a growth cycle while point D deﬁnes a peak. 2. Point A deﬁnes a trough for a classical cycle while point B deﬁnes a peak.
The average length of a cycle (peak from previous peak or trough from previous trough) in the postWWII era is 67 months or over ﬁve years. The shortest cycle is 17 months or nearly six quarters. Post WWII. the practice of separating the trend and cyclical component using linear time series methods is well established. If one chooses to use growth cycles. the average length of a contraction has decreased to ten months from 17 months or more in the years preceding 1945. and a recession is said to occur when the economy is moving from a peak to a trough. Table 2. employment. There are several approaches to detrending economic time series. The dating of business cycles for the US is done formally by the NBER Business Cycle Dating Committee. However. We can also observe expansionary and contractionary phases of the business cycle from this table. As King. There are 32 complete cycles in the entire sample period. there is an issue of how to identify the cyclical component of a given series. The amplitude of a business cycle is the deviation from trend.1 gives the dates of business cycles or the socalled “reference dates” for the US economy since 1857. One of the longest expansions to have occurred postWWII is between March 1991 and November 2001 — a period of 120 months. this committee recently announced that the US economy had formally been in a recession since December 2007. and the longest cycle is 128 months or more than ten years. real business cycle (RBC) models which allow for trends in the technology shock imply that growth and business cycles are jointly determined. and Watson [133] argue. and trade at the sectoral and aggregate levels to identify and date business cycles. two peaks. Plosser. the average length of an expansion has increased to 57 months from. This period was dubbed as the period of the “Great Moderation”. in their survey of empirical business cycles. The duration of the business cycle is the length of time (in months. quarters.10 Business Cycles: Fact. at most. equivalently. An alternative approach is to assume a stochastic trend modeled as a unit root in the series at hand. This committee uses data on real output. although a computer algorithm exists that can approximate the results (see Bry and Boschan [49]). Stock. Similarly. Fallacy and Fantasy in an expansion. One approach is to use a linear detrending procedure which assumes that the underlying series possesses deterministic time trends. Stock and . The contribution of Nelson and Plosser [167] was to show that economic time series such as real GDP typically possess unit roots. Nevertheless. 38 months preWWII. or years) that the economy spends between two troughs or. The turning points are determined judgmentally. As an example. national income.
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 . Business Cycle Reference Dates Peak Trough US Business Cycle Expansions and Contractions.1.Table 2.
Source: NBER.12 Table 2. all cycles: 1854–2001 (32 cycles) 1854–1919 (16 cycles) 1919–1945 (6 cycles) 1945–2001 (10 cycles) ∗ 13 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. Fallacy and Fantasy 17 22 18 10 55 48 53 67 . ∗∗ 15 cycles.1. 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.
and Cogley and Nason [69]. whereas the second exacerbates the role of shortterm noise. Cogley and Nason have argued. The parameter µ controls the smoothness of the trend component. If these are trendstationary. Eq. Speciﬁcally. The properties of the HP ﬁlter have been studied by many authors. Several alternative approaches have been proposed in the recent business cycle literature to separate the trend versus cyclical component of a time series.1). The Hodrick–Prescott ﬁlter deﬁnes gt to solve ∞ ∞ min gt t=−∞ (yt − gt ) + µ t=−∞ 2 (gt+1 − gt ) − (gt − gt−1 ) . 1 + µ(1 − L−1 )2 (1 − L)2 (1.2) In practice. then the detrending procedure has favorable properties. µ is typically recorded as 1600. Taking the derivative with respect to gt yields ∞ ∞ µgt+2 − 4µgt+1 + (1 + 6µ)gt − 4µgt−1 + µgt−2 = t=−∞ t=−∞ yt . which involves minimizing a quadratic form to determine the trend component in a given series. King and Rebelo [130]. let yt denote the series in question and gt denote the unknown trend component. The ﬁrst approach tends to generate spurious business cycle effects in the detrended series. is modiﬁed at the beginning and end of the sample. One of these approaches is the socalled Hodrick–Prescott [120] (HP) ﬁlter. For quarterly data. in particular. the ﬁlter is applied using a ﬁnite sample. that business cycle dynamics obtained by using a HP detrending procedure depend on the properties of the underlying data. (1. if the underlying data are .1) Using the lag operator notation and simplifying. 2 The cyclical component of the series is deﬁned as ytc = yt − gt . The general equation characterizing the ﬁlter. the cyclical component can be represented as ytc = yt − gt = µ(1 − L−1 )2 (1 − L)2 yt .Facts 13 Watson [197] argue that neither linear time trends nor ﬁrstdifferencing to eliminate unit roots provides a satisfactory approach to identifying the cyclical component of a series. (1. including Singleton [192].
K . The bandpass ﬁlter developed by Baxter and King [37] “ﬁlters” out both the longrun trend and the highfrequency movements in a given time series while retaining those components associated with periodicities of typical business cycle durations.3) where the moving average coefﬁcients are chosen to be symmetric.4) where ψ(L) is a (K − 1)order symmetric moving average polynomial. It turns out that the timedomain representation of the bandpass ﬁlter is an inﬁnite moving average. In particular. periodicities between six quarters and eight years.5s and 8s. . Baxter and King derive the bandpass ﬁlter by considering lowpass and highpass ﬁlters with the required properties. They show that if the sum of the moving average coefﬁcients is zero. The bandpass ﬁlter of Baxter and King [37] is obtained by applying a K thorder moving average to a given time series: yt∗ K = k=K ak yt−k . k=K they show that the lag polynomial describing the K thorder moving average can be written as a(L) = (1 − L)(1 − L−1 )ψ(L).5s)). the optimal . application of the HP ﬁlter induces spurious business cycle ﬂuctuations. The ﬁlter is designed to have a number of other properties. i. The frequency response function of the ideal bandpass ﬁlter is deﬁned as βbp (ω) = I (2π/(8s) ≤ ω ≤ 2π/(0. . (1. The bandpass ﬁlter retains the movements in a series for the frequencies associated with the periodicities of 0. . Fallacy and Fantasy differencestationary. namely. A second approach is based on the spectral analysis of economic time series.e. the Baxter–King ﬁlter will eliminate deterministic quadratic trends or render stationary series that are integrated up to order two. K ak = 0. Let s denote the number of observations in a year. . then it has trend elimination properties. Hence. (1. once the ideal ﬁlter’s weights are found in the time domain. I (2) or less.14 Business Cycles: Fact. however. ak = a−k for k = 1. where I (·) is the indicator function. 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. However.
Section 2. showing the large gains in productivity and growth attained over this period.01 8. 2.6 0.2 50 55 60 65 70 75 80 85 90 95 00 05 BPTEST USA 0. STYLIZED FACTS In this section.02 0. approximating ﬁlter with maximum lag K is obtained by truncating the ideal ﬁlter’s weights at lag K . 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.0 7.04 0.2 0. We consider a measure of the business cycle as the comovement of the cyclical behavior of individual series with .01 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. Trend and Cyclical Components in USA GDP. we use this approach to generate the cyclical components of the different series.03 15 50 55 60 65 70 75 80 85 90 95 00 05 BPCYCLEUSA Fig.8 8. We can observe the cyclical downturns and upturns corresponding to the NBER business cycle reference dates from this graph. the notion of a “business cycle” is also concerned with the comovement of a large number of economic variables.2.Facts 9. The cyclical component tracks very well the US business cycle as identiﬁed by the NBER. Nevertheless.4 0. We observe that real GDP displays a marked positive trend over the sample period. We also observe the cyclical troughs and peaks of economic activity associated with the business cycle.00 0.04 7.02 8. In what follows.2. 2. Baxter and King [37] employ a ﬁniteorder approximation to obtain the coefﬁcients of this ﬁlter with a truncation of K = 3 years or K = 12 quarters.05 0.03 0.2 discusses these properties of business cycles.6 9.
Leading indicators are useful for predicting subsequent changes in real GDP. They make use of the Baxter–King bandpass ﬁlter to identify the trend versus cyclical components of each series. they consider the comovement of the cyclical component of GDP with the cyclical component of each series. 2. Finally. and imports are all strongly procyclical. Among other measures. . Consumption. Backus and Kehoe [23] analyze the properties of historical business cycles for ten developed countries using a centurylong dataset up to the 1980s. 2 Early work on developing a composite index of leading indicators is due to Mitchell and Burns [164]. Coincident indicators reach a peak or a trough at roughly the same time as real GDP. Exceptions are production of agricultural goods and natural resources. a leading indicator reaches a peak before real GDP reaches its peak and bottoms out (reaches a trough) before real GDP. Real output across virtually all sectors of the economy moves together. Such a composite index was made ofﬁcial by the US Department of Commerce in 1968. In other words.2 The salient facts of a business cycle can be described as follows: 1. and separate their sample period into the preWorld War I era. These have been described in a number of economic studies. hours. Variables that move in the opposite direction (rise during recessions and fall in expansions) are countercyclical. the contemporaneous correlation of output in different sectors of the economy is large and positive. Consumption of durables is much more volatile than consumption of nondurable goods and services. lagging indicators reach a peak or trough after real GDP. inventories. variables that move in the same direction over the cycle as real output are procyclical. and passed over to the Conference Board in 1995. We can also examine whether different time series are out of phase with real GDP. According to this approach. investment. and the postWorld War II era.16 Business Cycles: Fact. the interwar era between World War I and II. which are not especially procyclical. Fallacy and Fantasy the cyclical component of real output. Consumption of 1 Stock and Watson [197] use data on 71 variables to characterize US business cycle phenomena over the period 1953–1996. and monetary aggregates. real wages. The stylized facts of business cycles are typically deﬁned in terms of the behavior of the main components of GDP. and the crosscorrelations of the cyclical component of GDP with the cyclical component of each series for lags and leads up to six periods. productivity. asset returns and prices. Variables that display little correlation with output over the cycle are called acyclical. They use the HP ﬁlter to derive the cyclical components of the different series.1 Information on leading indicators is also collected by the Conference Board. For example.
Proﬁts are highly volatile. 5. 13. or lagging indicators. an expansion is characterized by expectations of higher interest rates at longer horizons whereas a recession typically signals a decline in longterm interest rates relative to shortterm rates. 8. 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. 7. That is. Productivity is slightly procyclical. 12. Government spending tends to be acyclical. Velocity and the money supply are procyclical. an inverted yield curve. Total employment. The yield curve which shows the rates of return on bonds of different maturities tends to be upwardsloping during an expansion and downwardsloping at the onset of a recession. The employment series lags the business cycle by a quarter. Investment in equipment and nonresidential structures is procyclical with a lag. durable goods ﬂuctuates more than GDP. namely. Since imports are more strongly procyclical than exports. 6.Facts 17 3. tends to shrink during expansions and increase during recessions. employee hours. The risk premium for holding private debt. The correlation with output is generally negative. They have not displayed a steady pattern in terms of variability to GDP or in terms of leading. and capacity utilization are all strongly procyclical. coincident. while average weekly hours ﬂuctuate much less. or the yield spread between corporate paper and Treasury bills with six months’ maturity. 11. Net exports are countercyclical. The reason for this countercyclical behavior is likely to be changes in default risk. 4. Investment in residential structures is procyclical and highly volatile. but weakly so. whereas nondurables ﬂuctuate considerably less. Employment ﬂuctuates almost as much as output and total hours of work. . 10. Nominal interest rates tend to be procyclical. but both real wages and productivity vary considerably less than output. while capacity utilization tends to be coincident. 9. Real wages are procyclical or acyclical. The correlation between government expenditures and output is nearly zero.
hence. prices and inﬂation were procyclical in the preWWII era. reﬂecting the common experience of the Great Depression. 19. One way to rationalize this phenomenon may be in terms of monetary policy that is more accommodative — under a gold standard. a topic to which we will return. 15. Witness the impact of the large oil shocks in the 1970s. The correlation is typically larger in the postwar period than in the prewar period. Since the early 1980s. changes in stock prices have been taken as providing information about the future course of the real economy. contemporaneous correlations between output ﬂuctuations in different countries were highest in the interwar period. Chadha and Nolan [62] identify the stylized facts . employment. The standard deviation of inﬂation is lower than that of real GDP. the procyclicality of M2 has diminished since the 1980s. there is more evidence of supplysidedriven ﬂuctuations in output. The changes in the cyclical behavior of prices and inﬂation also have implications for the socalled impulses and propagation mechanisms of business cycles.18 Business Cycles: Fact. inﬂation was procyclical with a very low mean. real wages. 17. In the postWWII era. for example. the stock market fell in the quarter before each of the eight recessions. 16. Fallacy and Fantasy 14. inﬂation appears to be countercyclical. 20. Finally. A similar change appears to have characterized the behavior of pricelevel ﬂuctuations. Money (M2) is procyclical and tends to be a leading indicator of output. Between 1945 and 1980. However. Inﬂation is a coincident indicator. By contrast. Other papers have generated business cycle facts using data over long samples. In this sense. This set of facts constitutes a benchmark which a successful business cycle model should meet. In the preWWI period and interwar period. with the exception of Germany and Japan. The stock market is positively related to the subsequent growth rate of real GDP. The ﬁndings concerning the behavior of hours. There is also a marked increase in the persistence of inﬂation after WWII. The behavior of prices and inﬂation appears to have changed over time. although it is important to emphasize that the market has fallen without a subsequent recession. 18. and productivity have proved among the most difﬁcult to reconcile by current business cycle theories. the increased persistence of inﬂation and the countercyclical behavior of prices observed in this era.
Their ﬁndings also provide evidence on the factors behind the “Great Moderation”. Backus and Kehoe [23] argue that it is difﬁcult to reach a ﬁrm conclusion on this point based on US data alone. On the one hand. and another 15% to reduced shocks to food and commodity prices. 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. Artis and Zhang [18. Stock and Watson [198] seek to avoid the problem of poor data quality in the preWWII period. Their ﬁndings suggest the emergence of a groupspeciﬁc European business cycle since the formation of the ERM. Artis. Christina Romer [180. 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 postWWII eras.and postWWII data of comparable quality. 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. 2. They attribute around 20–30% of the reduction in output volatility to better monetary policy. While her ﬁndings have garnered much interest. Kontolemis. They ﬁnd that the standard deviations of a typical variable in their sample declined by about 20–40% since the 1980s. the variability of output appears to have diminished in the postWWII period. The countries selected are . and Osborn [20] propose business cycle turning points for a number of countries based on industrial production. They suggest that if one considers additional evidence based on a larger sample of countries over the different periods.Facts 19 of business cycles for the UK economy over a long sample period. If the higherquality data in the postWWII period are subjected to a transformation that makes the pre.3. A’Hearna and Woitek [2] establish the facts of business cycles in the late 19th century using spectral techniques. 15% to smaller shocks to productivity. then one discerns no change in output ﬂuctuations across the two periods. which is independent of the US cycle. The above ﬁndings can also shed light on whether output ﬂuctuations have moderated in the postWWII era. 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.
conditional on being in a contraction. suppose that the economy can be in one of two mutually exclusive states. who extended the Bry and Boschan [49] algorithm to a quarterly setting. and a trough terminates a recession. the probability of continuing a contraction pRR = Pr(RCt+1 RCt ) is pRR = 1 − pRT . 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. and Proietti [21] discuss alternative approaches to dating euro area business cycles. The Markov process distinguishes between turning points within the two states by assuming that Et = Rt = ECt expansion continuation . expansion Et or recession Rt . Suppose that a peak terminates an expansion. Likewise. Tt trough Let pEP = Pr(Pt+1 ECt ) denote the probability of transiting to a peak. Whereas in the postWWII era the euro area countries exhibited high rates of growth. Pt peak RCt recession continuation . As in the Stock and Watson [197] approach.3 To describe this algorithm. These authors also articulate a formal model of turning points based on restrictions imposed on a Markov process.20 Business Cycles: Fact. Let St denote a discrete random variable that follows a ﬁrstorder Markov process 3This algorithm follows Harding and Pagan [115]. making growth cycles the relevant concept. conditional on being in an expansion. and to determine whether cyclical movements are similar across different countries. They use this information to examine the international nature of cyclical movements. implying that classical cycles should also be considered. . Fallacy and Fantasy the G7 countries along with the prominent European countries. they distinguish between classical and growth cycles. Then. in recent years growth has slowed and even absolute declines in real GDP have been observed. Marcellino. which implies that the probability of continuing an expansion pEE = Pr(ECt+1 ECt ) is given by pEE = 1 − pEP . Artis. The euro area experience makes consideration of both types of business cycles relevant.
and a recession termination sequence. St−3 . St−3 . St−2 . yt+1 = yt+1 − yt < 0 and 2 yt+2 = yt+2 − yt < 0 represent the candidate sequence for terminating an expansion. (St−4 . St−1 . yt+1 = yt+1 − yt > 0 and 2 yt+2 = yt+2 − yt > 0 represent the candidate sequence for terminating a contraction. which deﬁnes a trough. pEP . . Hence. St ). Pt+1 . peaktopeak or troughtotrough. can only be followed by a peak at t + 1. (St−4 . respectively. Pt ) both belong to an expansion and (RCt . St−2 . expansion or contraction. Then. Else. Let yt denote the underlying series. St = RCt . which deﬁnes a peak. is pRT . the probability that the economy will transit to a trough. RTSt . the expansion continues.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. St = ECt . which is automatically satisﬁed since the states (ECt . Pt−4 . The minimum duration of a complete cycle is given as ﬁve quarters. as follows: ETSt = {( yt+1 < 0) ∪ ( RTSt = {( yt+1 > 0) ∪ ( 2 2 yt+2 < 0)} yt+2 > 0)}. a peak at date t − 4. ETSt . conditional on having been in an expansionary phase. This is the joint event that the history at time t. Thus. Tt ) belong to a contraction. A similar condition exists for troughs. The algorithm is completed by ﬁnding the probability that the economy will transit to a peak. Likewise. Likewise. if ETSt is false. we can deﬁne an expansion termination sequence. The duration of a phase is required to be two quarters. Now consider applying this algorithm to dating classical cycles. conditional on having been in a contractionary phase. St ). features an expansionary state at time t. St−1 . and on complete cycles. This is the joint event that the history at time t. features a contractionary state at time t. and that ETSt is true.
the contraction continues. 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. [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. Marcellino. 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 1999.4 Table 2. 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. Finally. Fallacy and Fantasy and that RTSt is true. and a trough cannot occur if output is above it. Henry. Euro Area Business Cycle Turning Points. if RTSt is false. They implement their procedures using data on euro area GDP and its components for the European Central Bank’s (ECB) AreaWide Model for the period 1970–2001.22 Business Cycles: Fact. We note that these are similar to the classical cycles identiﬁed by Artis et al. Otherwise. Artis et al.2 shows the business cycle turning points for both classical and deviation or growth cycles obtained using this approach.2. The Centre for Economic Policy Research (CEPR) has recently formed a committee to set the dates of the euro area business cycle. and Proietti [21]. . and Mestre [85]. The turning points determined by the CEPR Business Cycle Dating Committee are also indicated in this table. 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. Table 2. [21].
which is speciﬁed in terms of the growth rates of quarterly GDP for the G7 countries.03 1970 23 14.Facts 14. Fig. namely.4 1970 1975 1980 1985 1990 1995 2000 2005 1975 1980 1985 1990 1995 2000 2005 BPTRENDEURO EUROAREA BPCYCLEEURO Fig. Germany. Prasad.00 13. They also provide evidence on the synchronization of international business cycles. 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. and Canada) over the period 1960– 2002. Let Yt denote the n × 1 vector of . in sync with Artis et al. and Terrones [139] and others.02 0. they ﬁnd no evidence for closer international synchronization over their period of study. and the US). the eurozone countries and Englishspeaking countries (including Canada. Italy. the UK.01 13. First.0 0. [20].8 0. they ﬁnd evidence on the emergence of two cyclically coherent groups.2 13.01 14. This is a standard structural vector autoregression (VAR) with an unobserved factor structure imposed on the VAR innovations. France.03 0. This is similar to the ﬁndings of Köse.3.6 0. 2. 2. they use a socalled factorstructural vector autoregression (FSVAR). They ﬁnd that the volatility of business cycles has moderated in most G7 countries over the past 40 years.4 0. Stock and Watson [199] also seek to provide evidence on the sources of the changes. UK. Trend and Cyclical Components in Euro Area GDP. Stock and Watson [199] provide a comprehensive analysis of the volatility and persistence of business cycles in G7 countries (deﬁned to include the US. However. They base their results on various measures of correlation of GDP growth across countries.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). Japan.02 0.
They consider two sample periods: 1960–1983 and 1984–2002. and the least for Italy and Japan. In the second period. the role of international sources of ﬂuctuations arising from common shocks or from . and (iii) the spillover effects of the domestic shocks on the hstepahead forecast error for each country. Notice that the common shocks affect the output of multiple countries contemporaneously. and the elements of ft denote the common international shocks. σfk ) and E (νt νt ) = diag (σν1 . the impact of international shocks is estimated to be greatest for countries such as Canada. (ii) the domestic shocks. However. They also provide a variance decomposition for the impact of (i) the common international shocks. The covariance structure of the shocks is given by E (ft ft ) = diag (σf 1 . . Second. The model allows for spillover effects to occur through the lagged effect of an idiosyncratic shock. . First.5) where ft is a k × 1 vector with k ≤ n. In the ﬁrst period. where the vector of reducedform errors vt has the factor structure ν t = ft + t . reﬂecting the impact of the tenyearlong deﬂationary episode for this country. the elements of t denote the countryspeciﬁc idiosyncratic shocks. and Germany. domestic shocks explain almost all of the variance for Japan. These may arise from the role of international trade. . . (3. . .24 Business Cycles: Fact. This provides the identiﬁcation scheme to help identify the common international shocks. their relative variance varies by country and by time period. Stock and Watson [199] ﬁnd evidence for two common shocks. In this representation. whereas the idiosyncratic shocks affect them with a lag. σνk ). The standard VAR model is given by Yt = A(L)Yt−1 + νt . France. . Fallacy and Fantasy real GDP growth for the G7 countries considered in this study. where the elements of ft and t are assumed to be mutually independent. . for example.6) (3. they ﬁnd that most of the variance of GDP growth can be attributed to common and idiosyncratic domestic shocks.
and the UK is due to the decline in the variance arising from international shocks. Germany. Latin America.7) 5 For recent studies. Finally. Developing Asia. • the single world factor ( f world ). (4. they also ﬁnd that shocks have become more persistent in countries such as Canada. 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. and T the number of time periods. region . . . . Canada. This is the source of the moderation of individual business cycles. and use data on 60odd countries covering seven regions of the world to determine common factors underlying the cyclical ﬂuctuations in the main macroeconomic aggregates (output. • R regional factors ( fr Africa.t . and Raynauld [109] or Lumsdaine and Prasad [154]. The behavior of each observable variable is related to the factors as follows: yi. Gregory. Let N denote the number of countries. France. Let yi. 2. Otrok. . t = 1. and Italy. they also show that the decline in overall volatility of GDP growth for countries such as the US.4. or Oceania). .Facts 25 spillovers appears to have increased for the US. . one per each country). T . IS THERE A WORLD BUSINESS CYCLE? Köse. . Europe. . for example. where r stands for North America. They argue that many recent studies of international business cycles focus on a subset of countries — not the world. N × M . M the number of time series per country.t country + bi country country fn. and Whiteman [138] consider the issue of a world business cycle.5 Their approach assumes that there are K unobservable factors that are hypothesized to characterize the dynamic interrelationship among a crosscountry set of economic time series. and also of the failure of business cycles to become more synchronized despite the great increase in trade ﬂows over this period. and the UK.t = ai + biworld ft world + bi region region fr. However. see. There are three types of factors: • N countryspeciﬁc factors ( fn . Developed Asia. and investment) across all countries and regions as well as across countries and aggregates separately. consumption. Head.t denote the observable variables for i = 1. .t + i.
. conditional on the data. One identiﬁcation device is to assume that the factor loading on one of the factors is positive.t = Then. and to follow autoregressions of orders pi and qk . There are k several ways to estimate this model. and zero otherwise. which themselves may be serially correlated. (4. fk .t−pi + ui. Likewise. i = 1.t .9) = φfk .t−2 + · · · + φfk . an alternative is to use Bayesian estimation techniques. the sign of the factor loading for US output is considered positive.t−1 + φfk .t can be explained by each factor. Thus. Let fk.10) where E (ufk . Thus. Since the factors are unobservable.2 fk . and defer a discussion of the estimation techniques to Chapter 7. .t fk . K . and the country factors are identiﬁed by positive factor loadings on the output of each country. s.t ) = 0 for i = j. Fallacy and Fantasy j where E ( i.qk fk .t ui. (4.t = φi.t . and ufk . are mean zero.t−2 + · · · + φi. (4. For the world factor. and show how much of the variation in yi. .8) where E (ui. the model assumes that the behavior of M × N time series can be explained by N + R + 1 factors. the sign of the factor loading corresponding to North America is considered positive. for the regional factor.t .t .t−s ) = 0 for all k.2 i. contemporaneously uncorrelated random variables. .t−qk + ufk .1 i.t−1 + φi. where N + R + 1 < MN . The coefﬁcients bi denote the factor loadings.pi i. all the covariation among the observable series yi. Both the idiosyncratic errors and the factors are allowed to be serially correlated. we report results based on the median of the posterior distribution for the factors.t ) = σf2 . .26 Business Cycles: Fact.t . . scales are identiﬁed by assuming that each σf2 is equal to a constant. M × N . In what follows.t−s ) = σi2 for i = j and s = 0. . . as i.t is due to the common factors. Finally.t ui. The innovations k ui. . This procedure yields a joint posterior distribution for the unobserved factors and the unknown parameters.t ufk . one approach is to use a Kalman ﬁltering algorithm together with classical statistical techniques to estimate the model’s parameters.1 fk . respectively. k = 1. we cannot identify uniquely the sign or scale of the factors or the factor loadings. E (ufk . i. Since the common factors are unobserved.t j.
which is discussed in detail in Chapter 4. the recession of the 1970s associated with the ﬁrst oil shock. They interpret this ﬁnding as evidence for a world business cycle. Recall that the world.t ) = (biworld )2 Var(ft world ) + (bi region 2 ) Var(fr.Facts 27 Köse et al. they ﬁnd that country and idiosyncratic factors account for a much larger share of variability in investment growth than the world and regional factors. The authors also document some noteworthy ﬁndings regarding the sources of volatility of investment growth. in contrast to models estimated using a smaller sample of countries. idiosyncratic .t ) They ﬁnd evidence that. the world factor explains nearly 15% of variation in output growth. 9% of consumption growth.t ). Speciﬁcally. and country factors are modeled as autoregressive processes which may display substantial persistence depending on the estimated values of the parameters. country factors account for a much larger fraction of consumption growth than output growth. regional. as severe as the recession of the mid1970s. Using the representation of the model. the fraction of variance attributed to the world factor can be expressed as (biworld )2 Var(ft world ) . The paper also allows for a simple decomposition of variance attributed to the different factors. Var(yi.t region ) (4. the regional and country factors explain covariation or comovement that is less persistent. This ﬁnding has been documented further in the international business cycle literature. A second important ﬁnding from the paper is that most of the persistent comovement across countries and aggregates is captured by the world factor. ﬁrst. the world factor is more successful in explaining economic activity in developed relative to developing countries. the world factor based on a large set of developed and developing countries implies that the recession of the 1980s was. and the downturn and recession of the early 1990s. Second. if anything. By contrast. and 7% of investment growth.’s paper [138] shows that the world factor identiﬁes many of the major cyclical events over a 30year period: the expansionary periods of the 1960s. Second. the recession of the 1980s associated with the debt crisis in developing countries and the tight monetary policies in the major developed countries. Third. we have that Var(yi.11) country 2 country ) Var(fn.t ) + Var( i. However. + (bi Thus.
5. 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.28 Business Cycles: Fact. 2. the risk of policy reversal) for irreversible investment decisions. The authors pose this as a puzzle. It featured ﬁxed exchange rates and worldwide capital market integration. As Stock and Watson [199] note. disruptions to market access. The last two ﬁndings of the paper imply that regional factors play a minor role in aggregate output ﬂuctuations. Paralleling this ﬁnding. . Demers. They divide this period into four periods that also reﬂect the monetary and capital account regimes prevailing in them. See. Yet this ﬁnding could also be due to model misspeciﬁcation. for example. Serven and Solimano [189]. there is little evidence that the volatility of European aggregates can be attributed to the European regional factor. much work on investment behavior shows that irreversible investment decisions are particularly susceptible to risk and uncertainty. However. 6 For recent reviews. policy reversals. and Demers [10] argue that an important source of risk for developing countries is political risk or risk arising from the threat of expropriation. This last result is interpreted as evidence against the existence of a European business cycle. Demers. 7 In a related literature. exchange rates. prices. and debt and currency crises. • 1870–1941: This era corresponded to the classic gold standard. investment. and other European countries plus Argentina for the period since 1870. total consumption. and the current account for 15 countries including the US. HISTORICAL BUSINESS CYCLES Basu and Taylor [31] examine business cycles within an international historical perspective. They consider the time series behavior of output. real wages. there is also the possibility that some observed covariation could result from the spillover effects of idiosyncratic or country shocks.6 Altug. Rodrik [178] emphasizes the importance of political risk (in his case. the UK. Fallacy and Fantasy shocks explain a much larger fraction of the volatility in investment growth in developing countries relative to developed countries. Yet. the dynamic factor model suffers from the shortcoming that all covariation is attributed to the common factors. especially in an era of increased trade ﬂows and ﬁnancial integration. and Altug [79].7 We conjecture that variability in investment due to such factors is a key channel for inducing idiosyncratic volatility in developing economies’ investment behavior. see Caballero [53] or Demers.
Basu and Taylor [31] argue that considering such a breakdown allows for the analysis of the impact of different regimes on cyclical phenomena. . to monetary phenomena. France. This period also corresponded to the Great Depression. which simultaneously occurred in a number of countries. and consider the four exchange rate regimes also examined by Basu and Taylor [31]. thus allows for an examination of such mechanisms as nominal rigidities in propagating shocks. Norway. For example. Germany. and the US. Consideration of alternative historical periods.Facts 29 • • • 1919–1939: During this period. the world economy shifted from a globalized regime to an autarkic regime. which examines whether the turning points in the different series occur at similar dates. the socalled concordance correlation. Synchronization refers to the notion that “the timing and magnitudes of major changes in economic activity appear increasingly similar. Italy. The countries that they examine are Australia. It also provides a way to identify the importance of demand. most explanations of the Great Depression attribute the source of this massive downturn. the UK. Japan. Finland. Bordo and Heibling [44] use real GDP or industrial production series to determine synchronization. Instead of using information on NBERtype reference dates. Switzerland. Canada. Sweden. 1945–1971: The third regime is the Bretton Woods era. Portugal.versus supplyside factors or shocks and the role of alternative propagation mechanisms such as price rigidity. the Netherlands. which corresponded to the postWorld War II era of reconstruction and a resumption of global trade and capital ﬂows. Denmark. They also use standard output correlations and factorbased measures. Spain. They examine business cycles in 16 countries during the period 1880–2001. Bordo and Heibling [44] ask whether national business cycles have become more synchronized. Basu and Taylor [31] also argue that their periodization allows for an analysis of international capital ﬂows and capital mobility in affecting cyclical ﬂuctuations. Early 1970s–present: This period features a ﬂoating exchange regime and a period of increasing globalization. including the era of the Great Depression. they use a statistical measure of synchronization developed by Harding and Pagan [115].” Among other concepts.
Fallacy and Fantasy To brieﬂy describe the concordance correlations. then under the assumption that Sit and Sjt are independent. for countries i and j. They deﬁne a recession as one or more consecutive years of real GDP growth. Note ∗ that the variance of Iij is given by ∗ Var(Iij ) 4 = 2E T T t=1 ¯ ¯ (Sit − Si )(Sjt − Sj ) . cycles were. and if the two cycles are unrelated. let Sit and Sjt denote binarycycle indicator variables which assume a value of 1 if the economy is in an expansion. In the interwar and postBretton Woods era. Bordo and Heibling [44] calculate the business cycle indicators Sit for each of the four exchange rate regimes. ¯ ¯ ¯ ¯ the expected value of the concordance index is 2Si Sj + 1 − Si − Sj . 0 otherwise. Subtracting this from Iij yields the meancorrected concordance index: ∗ Iij 1 =2 T T t=1 ¯ ¯ (Sit − Si )(Sjt − Sj ). The key ﬁnding is that during the classical gold standard. on average. as half of all pairs of business cycles are negatively related to each other while the other half are positively related. then the standardized index equals −1. They argue that Sit = 1 for most countries during the Bretton Woods era of 1948–1972 and hence leave this period out.12) ¯ Let Si = (1/T ) T Sit denote the estimated probability of being in an t=1 expansion. (5.30 Business Cycles: Fact. while an expansion is deﬁned as one or more years of positive GDP growth. If the two cycles are perfectly synchronized (so that Sit = Sjt for all t). For the gold standard era. they ﬁnd that the average of the correlation coefﬁcients is zero. then the standardized index equals 1. t=1 (5. if they are in different states in each period (so that Sit = 1 − Sjt ). uncorrelated . A simple measure of concordance is deﬁned by the variable 1 Iij = T T [Sit Sjt − (1 − Sit )(1 − Sjt )]. say. we divide Iij by a consistent estimate of its standard error under the null hypothesis of independence. more than half of all national business cycles become positively related to each other.13) ∗ To derive the concordance correlation coefﬁcient. the standardized index equals 0.
whereas beginning with the interwar period. They ﬁnd that there has been a tendency for higher.” They also estimate restricted versions of a FSVAR model along the lines of Stock and Watson [199] to determine the role of common shocks. 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. but this was not so for ﬁnancial or twin crises. They argue that banking crises were less severe in the period before 1915. idiosyncratic shocks. and a trade linkages version. the importance of transmission for peripheral countries arises only in the trade model. We discuss emerging market business cycles in detail in Chapter 6. In the former. they ﬁnd an increase in correlations for the AngloSaxon countries. they started becoming synchronized with each other. In the trade linkages version. which measure the magnitude as well as the direction of output changes. They also ﬁnd higher output correlations for the core European countries (the EEC) and the Continental European countries. . especially the integration of goods and services through international trade and the integration of ﬁnancial markets. and the idiosyncratic shocks are allowed to be serially correlated and heteroscedastic. lagged GDP growth of the center country is included alongside the lagged value of the own country’s GDP growth in the VAR representation. there are no dynamics in the relation between output growth rates and the factors. before 1914 and after 1971. Eichengreen and Bordo [84] examine the nature of crises in two periods of globalization. as in Stock and Watson [199]. Bordo and Heibling [44] conclude by noting that global shocks are the dominant inﬂuence across all regimes. However. more positive bilateral output correlations by era. They ﬁnd that both global (common) shocks and transmission have become more important. Finally. In this model. such crises have ﬁgured importantly in emerging market output ﬂuctuations. The authors also estimate a socalled static approximate factor model for the growth rates of GDP. They consider a socalled center country version of this model. and that the increasing importance of global shocks reﬂects the forces of globalization. and spillover effects in generating this result. The authors also examine standard output correlations.Facts 31 with each other. suggesting that it is not transmission from the center country that accounts for the increased role of transmission. lagged GDP growth of the major trading partner is included in place of the center country’s. Typically.
This page intentionally left blank .
whether they are permanent or transitory such as oil shocks. the result is unemployment and greater declines in output relative to a situation with ﬂexible prices. More controversially. Weather shocks have always had a place among the impulses thought to trigger ﬂuctuations in economic activity. The shocks or impulses thought to instigate business cycles have typically been varied and diverse. perfectly competitive. Keynesian and New Keynesian models stress the role of frictions such as price stickiness. one could also assign a role to taste shocks or changes in preferences of consumers. This approach emphasizes the role of intertemporal substitution motives in propagating shocks. By contrast.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. Equivalently. one could argue. Political shocks. The Great 33 . Technology shocks which alter a society’s production possibilities frontier. changes in their subjective beliefs could help to trigger business cycles. as Keynes [127] did. and other disruptions in market activity have also been acknowledged to play a role. Current real business cycle (RBC) theory argues that business cycles can arise in frictionless. complete markets in which there are real or technology shocks. have ﬁgured prominently in the recent literature. 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. In a simple labor market model. if real wages do not adjust downward when there is a negative shock to demand. that investors’ “animal spirits” or. wars. more precisely.
There is a representative consumer who derives utility from consumption and leisure. We will discuss these extensions in the following sections. A number of these assumptions have come under criticism in the recent literature. 173). we discuss an international RBC model to describe how the transmission mechanisms proposed by this approach translate into an openeconomy setting. In the following chapter. that households consume only goods produced in the market. In this case.1 In this chapter. Fallacy and Fantasy Depression and the recent ﬁnancial crises also indicate the importance of credit market frictions and ﬁnancial acceleratortype effects as a potential propagation mechanism for business cycles. 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. As Kydland and Prescott argue. but his discussion is abbreviated to satisfy the constraints imposed by a journal article. 2 Rebelo [177] provides a recent review of alternative models and directions associated with the RBC agenda.1. Bernanke and Gertler [40] or Kiyotaki and Moore [134]. it seems imperative to discuss the main propagation mechanisms of the RBC model.34 Business Cycles: Fact. home production. The prototypical RBC model has the structure of a standard neoclassical growth model with a labor/leisure choice incorporated. this is a crucial modiﬁcation as “more than half of business cycle ﬂuctuations are accounted for by variations in the labor input” ([143]. p. 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.2 3. we discuss variants of the RBC model. for example. 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. . All markets are perfectly competitive. and all prices adjust instantaneously to clear markets. 1 See. all factors are fully employed. and international trade. and a constantreturnstoscale (CRTS) production technology for producing the single good using labor and capital. Since the debate has revolved around the efﬁcacy of technology shocks and intertemporal substitution effects in replicating business cycles.
i. The problem described above is an inﬁnitehorizon dynamic stochastic optimization problem. lt ) = ctθ lt1−θ . t ∼ i.. There is a representative ﬁrm with CRTS production that is affected by a stochastic technology shock each period: yt = exp (zt )ktα ht1−α .Models of Business Cycles 35 Turning to the basic model. (1.2) Hence. For ease of exposition.6) Notice that the only shock in this model is the technology shock. we will suppose that the utility function has the form U (ct . (1. (1. 0 < θ < 1.d.4) Capital evolves according to kt+1 = (1 − δ)kt + it . (1. 3 See. and N (0. which is assumed to be stationary but persistent as long as ρ = 1. . Altug and Labadie [6] or Stokey and Lucas with Prescott [201]. zt+1 = µ+ρzt + 0 < α < 1.3 Assuming a solution exists. σ 2 ). lt ) .3) Assume that the technology shock follows an AR(1) process: t+1 . We will discuss the propagation mechanism when describing the solution for the model. 0 < ρ < 1. for example. (1. time spent not working is taken as leisure. (1. The time constraint requires that the sum of leisure and labor hours equals 1: lt + ht = 1.1) where 0 < β < 1 is the discount factor. The aggregate feasibility constraint is deﬁned as ct + it = yt .5) where it denotes economywide investment and 0 < δ < 1 is the depreciation rate. the representative agent has timeseparable preferences over consumption and leisure choices given by ∞ U = E0 t=0 βt u(ct . The existence of a solution can be shown using standard recursive methods for analyzing such problems.
kt+1 (1. To analyze the optimal consumptionleisure . This means that the capital accumulation process will also be a function of the capitallabor ratio. ¯ it = 0. zt+1 ) = λt .lt . These conditions can be rewritten as U2t = exp (zt )(1 − α)(kt /ht )α U1t βEt U1. U1t (1. respectively.12) (1. Thus.13) The ﬁrst equation shows that the marginal rate of substitution between consumption and leisure is equal to the marginal product of labor.t+1 [exp (zt+1 )α(kt+1 /ht+1 )α−1 + (1 − δ)] = 1. The second equation is the intertemporal Euler equation. kt+1 and lt and the envelope condition are U1 (ct . βEt Vk (kt+1 .11) where U1t and U2t denote the marginal utility of consumption and leisure. his choice is between how much to work and how much time to spend in leisure. Suppose for the moment that there is no investment in the model. which he then consumes. There is no unemployment in the model as time not spent working is (optimally) taken as leisure. Substituting for lt = 1−ht using the time constraint. (1. Vk (kt .10) (1. kt = k. U2 (ct . and ct = yt . Notice that this is just a simple Robinson Crusoe economy in which the Crusoe produces output using capital and labor.9) (1. that is.7) subject to ct + kt+1 = exp (zt )ktα ht1−α + (1 − δ)kt . zt ) = λt exp (zt )αktα−1 ht1−α + (1 − δ) . Let λt denote the Lagrange multiplier on the resource constraint. lt ) = λt exp (zt )(1 − α)ktα ht−α . lt ) + βEt V (kt+1 . zt ) = max {U (ct . the ﬁrstorder conditions with respect to ct . Observe that the ratio U2 /U1 is a function of the capitallabor ratio. lt ) = λt .36 Business Cycles: Fact. kt /ht .8) (1. lt + ht = 1. Fallacy and Fantasy let the value function associated with the problem be expressed as V (kt . zt+1 )} ct .
However. (1. .16) Now a temporary negative technology shock also affects investment or saving.1 describes the representative agent’s optimum. lt+i ) = θ(ct+i /lt+i )θ−1 for i ≥ 0 and using the result in equation (1. Thus. Now suppose output can be consumed or invested so that it = 0. (1. Figure 3.12). then Crusoe spends 50% of his time working and 50% taking leisure.14).Models of Business Cycles 37 choice. 1 − θ + θα (1. Crusoe will typically respond to a temporary negative shock by lowering savings and also by working less. 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. in addition to equation (1. if θ = 2/3 and α = 1/2. 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 . we can rewrite the intertemporal Euler equation by noting that U1 (ct+i . Then. for this preference speciﬁcation. we note that if the substitution effect of an increase in productivity dominates the income effect. the impact of these changes is to induce ﬂuctuations in labor supply and employment as well as investment. θ(1 − ht ) We can solve this equation for the optimal labor hours ht∗ as ht∗ = θ(1 − α) .15) Thus. we have the intertemporal Euler equation characterizing the behavior of the optimal capital stock over time. for example.14).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 . we consider equation (1. Using our preference speciﬁcation. then Crusoe will work less in response to a temporary negative technology shock and will also consume and produce less. Notice that the technology shock does not affect optimal hours worked because.
To derive the Solow residual.3) and then take the ﬁrst differences: ln (yt+1 ) = zt+1 + α ln (kt+1 ) + (1 − α) ln (ht+1 ).1. 3. take the logarithm of both sides of equation (1. Furthermore. We can also derive a measure of the Solow residual for this model by using the form of the production function. 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 ). it will have a lower capital stock in the future.38 Business Cycles: Fact. where rt denotes the competitively determined rental rate on capital and pt denotes the product price normalized here as unity. the parameter α is typically measured as the share of capital in real output deﬁned as skt = rt kt /pt yt . skt + sht = 1. To derive an observable measure of the Solow residual. Fallacy and Fantasy Fig. . if the economy experiences a period with lower investment. implying that the effect of the shock persists. Under constant returns to scale. Under these assumptions. where sht = wt ht /pt yt denotes the share of labor in national income. Robinson Crusoe’s Optimum.
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. and convert the original nonlinear dynamic optimization procedure to one which has a quadratic objective and linear constraints (see also Christiano [64]). This is obtained by setting the exogenous technology shock equal to its mean value and evaluating the conditions (1. We assume that capital depreciates at the rate 0 < δ < 1 and the technology shock follows the process in (1.4). The quadratic approximation procedure is implemented as follows. consumption plus investment equals output at the optimum.18) (2. 40% of GNP growth in the USA between 1929 and 1957 resulted from technical development.17) (2. Using this fact. 1 − ht ) = θ ln (ct ) + (1 − θ) ln (1 − ht ). Since the utility function is strictly increasing.14) and (1. Critics of the RBC approach argue that there are endogenous components to the ﬂuctuations in the Solow residual. t+1 .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 . according to Denison [80]. we can substitute for consumption in the utility function to obtain u(ct ) = u( exp (zt )ktα ht1−α − it . (2.2. the problem now becomes ∞ {it . Notice that in the RBC framework the Solow residual is typically treated as exogenous. rendering the RBC interpretation invalid. Step 1: Compute the deterministic steady state for the model. zt+1 = µ + ρzt + ct ≥ 0.19) 0 ≤ ht ≤ 1. 0 < θ < 1.16) together with the feasibility . Letting u(ct . Similarly. A NUMERICAL SOLUTION We now provide a numerical solution to further illustrate the properties of the standard RBC model. 1 − h). This is an issue that has been contested in the business cycle literature. We implement a quadratic approximation procedure that was initially suggested by Kydland and Prescott [141]. 3.
investment is equal to depreciation: ¯ ¯ i = δk.40 Business Cycles: Fact. note that in the deterministic steady state. kt . .22) as ¯c ¯ y (1 − θ)h¯ = (1 − α)θ(1 − h)¯ . To ﬁnd an explicit expression for h parameters. Rearranging. use the relation in (2.23) ¯ + θ(1 − α)h = θ(1 − α). Let z = µ/(1 − ρ) denote the ¯ unconditional mean for the (log of ) technology process. we obtain ¯ y ¯ 1−θ h = (1 − α) . Using (1.14). Simplifying the result in (1.22) (2. (2. and ht .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. substituting for c ﬁrst and then dividing through by y . ¯ θ 1−h c ¯ (2. we obtain ¯ ¯ ¯ k ¯ (1 − θ)h 1 − δ y ¯ Now note that ¯ ¯ ¯ ¯ k (k/h)1−α k α = = = . we can solve for the steadystate capitallabor ratio as ¯ exp (¯ ) z k = α ¯ r +δ h 1/(1−α) . ¯ ¯ 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). h (1 − θ) + θ(1 − α) − (1 − θ)δ r +δ or ¯ h= θ(1 − α)(r + δ) . (r + δ)(1 − θα) − (1 − θ)δα (2.16).20) Next. Fallacy and Fantasy conditions at constant values for ct .
ht − h) . let s ≡ (1. See Campbell [55]. . s + (s − ¯) s ∂s 2 ∂s∂s Deﬁne e as the 5 × 1 vector with a 1 in the ﬁrst row and zeros elsewhere. 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 .Models of Business Cycles 41 Step 2: Approximate the original utility function by a quadratic function around the deterministic steady state. 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 . i. kt − k.4 For this purpose. h) and ¯ ≡ (0. s ¯ ¯ Deﬁne the vector of state variables as xt = (1. zt − z ) and the ¯ ¯ vector of control variables as ut = (it − i. u) . 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) . zt+1 zt 0 µ 0 ρ 000 t+1 s s Notice that s − ¯ = (x. Thus. Approximate u(s) near the deterministic steady state ¯ s ¯ ¯ ¯ ¯ s using a secondorder Taylor series approximation as u(s) = u(¯) + (s − ¯) s s ∂u(s) 1 ∂2 u(s) (s − ¯). z. i. z . Notice that the utility function can be written as s u(s) = (s − ¯) T (s − ¯). k. h). u 4 An alternative approach is to loglinearize the model. ∂s∂s where all partial derivatives are evaluated at ¯. k.
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 . 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.25) where E (εt+1 ) = 0 and E (εt εt ) = . (2. t ≥ 0. Ljungqvist and Sargent [146] and Anderson.26) . (2.24) subject to the linear law of motion xt+1 = Axt + But + εt+1 . notice that the value function will be a quadratic function in the state variables: V (x) = x Px + d . in other words. and ut .42 Business Cycles: Fact. Such problems can be solved using the methods for solving dynamic optimization problems that satisfy a certainty equivalence property. Given the structure of the problem. where d and P are quantities to be determined. Hansen. 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 ). xt . Using the deﬁnition of T . McGrattan. and Sargent [16] provide further discussion of the formulation and estimation of linear dynamic economic models. Substituting for next period’s state variables and using this expression for the value function. This class of problems is known as optimal linear regulator problems. the dynamic optimization problem can now be written as ∞ {ut }t=0 max E0 ∞ t=0 βt [xt Rxt + ut Qut + 2xt Wut ] (2. Bellman’s equation becomes x Px + d = max x Rx + u Qu + 2x Wu u + βE (Ax + Bu + ε) P(Ax + Bu + ε) + d .
d = (1 − β)−1 [βE (ε Pε)]. output. The equation for P is known as the algebraic matrix Riccati equation for socalled optimal linear regulator problems. Q . ∂u ∂u ∂u 6 Notice that the matrices R. starting from P0 = 0. and B must be further restricted to ensure the existence of a solution to the matrix difference equation deﬁning Pn . A. Calibration refers to the practice of determining the parameters of the model based on its steady state properties and the results of other studies. 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 ). In the RBC literature.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 ).27) Notice that the shocks εt+1 do not affect the optimal choice of ut . given particular functional forms for the production function and the utility function (see Cooley and Prescott [75]). = T x and = Tx.5 Solving for u yields u = −(Q + βB PB)−1 (W x + βB PAx) = −Fx. One condition that sufﬁces is that the eigenvalues of A are bounded in modulus below unity.28) (2. the model is then calibrated with the data. 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 ). (2. where F = (Q + βB PB)−1 (βB PA + W ). 5To derive this result. we have used the rules for differentiating quadratic forms as ∂u Qu ∂x Tu ∂u Tx = [Q + Q ]u = 2Qu. .Models of Business Cycles 43 The ﬁrstorder conditions with respect to u are Qu + W x + β[B PAx + B PBu] = 0.29) (2. A stochastic process for zt is also speciﬁed and a random number generator is used to simulate the TFP time series.
so the approach differs signiﬁcantly from standard econometrics.7496 Correlation with Output 1 0. The typical set of unconditional moments used in the RBC literature is displayed in Table 3.9840 . we ﬁnd that the variation in hours is typically quite low.028. α = 1/3. Standard Deviation (%) Output Consumption Investment Hours Productivity 8.95.9807 0. with hours showing the least procyclicality.9238 1. productivity is almost as variable as output. and labor. We ﬁrst calculate a i ¯ ¯ set of unconditional moments that have been used in the RBC literature to match the model with the data. ρ = 0.028.1281.95.9163 0. Speciﬁcally. We now illustrate this solution method for the standard RBC model with a laborleisure choice presented at the beginning of this section. in the ﬁrst instance. y = 0. We then use the linear decision rules for all variables to simulate for the endogenous variables based on the same history of shocks. c = 0. Cyclical Properties of Key Variables. We ﬁnd that consumption ﬂuctuates slightly less than output and investment ﬂuctuates signiﬁcantly more. These are obtained by simulating the behavior of the different series across a given history of the shock sequence.8616 8. The steady values for the variables are given by k = 1.2952. and ¯ σ = 0. δ = 0. In terms of the correlation of each series with output. and crosscorrelations for the simulated time series are compared to the comparable statistics in the data. Fallacy and Fantasy investment.1134 12.1128. and generate a sequence of ˆ technology shocks beginning from some initial value z0 as zt+1 = ρzt + ˆ t+1 .1.44 Business Cycles: Fact. covariances.1. Likewise. Trends in the data are removed using a given ﬁlter. Consider the parameter values β = 0. ¯ ¯ = 0. and h = 0.10. The model is then judged to be close or not close based on this comparison. θ = 0. we draw a sequence of shocks {ˆ t+1 }2999 that are normally distributed t=0 with mean zero and standard deviation 0.4783.6684 7. we calculate unconditional moments for the different series after dropping the ﬁrst 1000 observations.3655. The means. Finally. the model is capable of delivering some of the salient features Table 3.4934 0. By contrast. The RBC theorists have interpreted these results as implying that.36. variances. we ﬁnd that all series are procyclical.
2 illustrates the response of all the variables to a 1% shock in technology starting from the steadystate capital stock. Speciﬁcally. This is discussed further in the following sections. investment. the technology shock explains much but not all of the variation in output.6 0. as we observe in 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. Thus. We note that hours and output both increase and then fall back to a lower level.1 1 % deviations form steady state 0. .Models of Business Cycles 45 of the data. Figure 3. 3.9 0. the model predicts that consumption ﬂuctuates less than output whereas investment ﬂuctuates more. and productivity are all strongly procyclical. rising ﬁrst and then declining back to a lower level. These responses have been widely Hours Output Consumption Capital Technology 1. Consumption.4 0 5 10 Years after shock 15 20 Fig. By contrast.2. 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.8 0. Consumption shows a gradual positive response because investment is initially high.5 0. capital and consumption show a humpedshaped response. In terms of the variation accounted by the technology shocks. Impulse Responses to a Shock in Technology.7 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 timetobuild 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 ﬁrstorder 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, economywide 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 supplyside 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 onesixth, not onethird. 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 twothirds 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 establishmentlevel surveys conducted by the US Department of Labor.
(4. this is inconsistent with the notion that TFP shocks are the driving force behind business cycle activity. labor’s share of income moves countercyclically. πt .4. The conventional wisdom regarding the moneyoutput correlation. is that the causality goes from money to output but with “long and variable lags”. This is accomplished by assuming that individuals can work all the time or not at all. This economy is one in which individuals and a ﬁrm trade a contract that commits the household to work h0 hours with probability πt .1. especially M2. 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. Since what is being traded is the contract. summarized by Friedman and Schwartz [94] in their study of monetary history. ˆ Suppose that workers are constrained to work either zero or h hours.Models of Business Cycles 49 arises in matching the model and the data is that in the data.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. it is assumed that individuals choose the probability of working. A lottery then determines whether an individual actually works.31) . (4. 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. appears to be a leading indicator of aggregate output. To account for the nonconvexities introduced by the work/nonwork decision. where ˆ 0 < h < 1. However. 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. whereas in the RBC model labor’s share is ﬁxed. the individual gets paid regardless of whether he works or not. 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. 3. • Reverse causality: The stylized facts of business cycles state that money. A Model with Indivisible Labor Supply The indivisible labor.
c0.t. Let λt denote the Lagrange multiplier on the feasibility constraint for consumption. πt c1.50 Business Cycles: Fact. 1 − h) + (1 − πt )u(c0. Assume that the individual utility function has the form u(c. This is a feature that is counterfactual to the working of actual labor markets. (4.t denote the consumption of an unemployed worker and c1. c1.t . Then the expected utility of the representative consumer.t πt = πt λt .t denote the consumption of an employed agent. In this model.t (4. c0.t . Notice that the agent will consume ct whether or not he is working.t = ct . Omitting the work/leisure decision for the moment.t (4. lt )] = πt u(c1. the ﬁrstorder conditions with respect to (c0.t = c1.32) max E [u(ct . ex ante all individuals are alike. With complete insurance and identical preferences that are separable with respect to consumption and leisure.33) (4.t . they ˆ must supply h hours of work so that there is no choice over hours of work directly. 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). is given by ˆ E [u(ct .t ) are 1 − πt = (1 − πt )λt . lt )] s. When agents do work.c1. Hence. 1). l ) = ln (c) + A ln (l ).35) .t .34) It follows that c0. The social planner solves the problem πt . taking into account the work versus nonwork decision. Notice that the social planner chooses the consumption allocations of each agent plus the probability of their working. but ex post they differ because some work while others enjoy leisure. Fallacy and Fantasy Let c0.t + (1 − πt )c0. c1. the unemployed worker enjoys higher utility since working causes disutility. Hence.t = ct so that the agent consumes the same amount whether or not he is working. all individuals have the same consumption but the unemployed are better off.
By contrast. Ht ) .36) Comparing equation (4.36) shows the effect of the lottery assumption. lt )] = ln (ct ) + πt A ln (1 − h) = ln (ct ) − BHt . which is consistent with assumptions about individual behavior.35) and use ln (1) = 0 to obtain ˆ E [u(ct .Models of Business Cycles 51 ˆ where A is a positive constant. Using the deﬁnition of Ht . (4. where u(ct . The ﬁrstorder conditions are 1 = λt . ˆ h (4.32) with equation (4. πt = Ht /h.37) (4.Ht .kt+1 } max E0 t=0 βt ln (ct ) − BHt + λt [exp (zt )ktθ Ht1−θ +(1 − δ)kt − ct − kt+1 ] .39) .38) (4. The preferences can now be written as ∞ U = E0 t=0 βt u(ct . θ−1 1−θ λt = βEt λt+1 exp (zt+1 )θkt+1 Ht+1 + (1 − δ) . the latter speciﬁcation — which is linear in total hours Ht — implies a high intertemporal elasticity at the aggregate level. We now solve the model subject to the time constraint. The problem is ∞ {ct . and law of motion for the capital stock described earlier. where B= ˆ −A ln (1 − h) . resource constraint. The former speciﬁcation for preferences implies a low intertemporal elasticity of substitution in labor supply. ct B = λt exp (zt )(1 − θ)ktθ Ht−θ . production function. Ht ) = ln (ct ) − BHt . Now substitute for πt in (4.
. and Heckman [48] for further discussion regarding the reconciliation of micro evidence with dynamic general equilibrium models.4. A. 8 See Browning. δ.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)] . and it is not intended to incorporate the microeconomic foundations of the labor market. (4.38) 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 . Hansen [112] calibrates this model by specifying values for the unknown parameters θ.7 compared to the model without indivisibilities which implies a value near unity. and the stochastic process for the technology shock using the approach in Kydland and Prescott [141]. The Productivity Puzzle Several resolutions of the productivity puzzle have been suggested. 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 }.40) Equations (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.2. β. With Ht determined in (4.52 Business Cycles: Fact.41) where the term in square brackets shows the rate of return to investing in the aggregate production technology. (4. Hansen argues that the model with indivisibilities can generate a variability of hours relative to productivity around 2. 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.40).8 3. Hansen. Fallacy and Fantasy Notice that (4.37) and (4.
(4.46) . The social planner ranks alternative consumption/leisure streams according to ∞ E0 t=0 ¯ βt [ln (ct ) + γV (N − nt )] . They consider a prototypical RBC model with a labor/leisure choice. (4. and α is a parameter that governs the impact of government consumption on the marginal utility of private consumption. (4.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. Per capita output is produced according to the Cobb–Douglas production function yt = (zt nt )1−θ ktθ . 0 < θ < 1.43) is public consumption.42) ¯ where ct denotes consumption and N −nt denotes leisure of the representative household. Let N denote the time endowment of the representative household per period. the standard model cannot deliver the strong procyclical response of hours without procyclical behavior in productivity.45) where zt is a technology shock which evolves according to the process zt = zt−1 exp (λt ). one which is based on a timeseparable logarithmic speciﬁcation and a second which incorporates the Hansen indivisible labor assumption. 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. gt p (4. namely. They generate the negative correlation between hours worked and real wages or productivity by introducing government consumption shocks. p where ct is private consumption. ¯ ln (N − nt ) ¯ (4. They consider two different speciﬁcations for the labor/leisure choice.44) V ( N − nt ) = ¯ N − nt for all t.
The total amount of time available to the household is normalized as unity. hours of work can increase along a downwardsloping labor demand curve. cmt is the consumption of a market good. p (4.54 Business Cycles: Fact. . Home Production Another way of improving the model’s ability to match the data is to introduce home production. In this model. consider a decisionmaker who has preferences ∞ βt u(cmt . Hence. hmt . and u4 < 0. yt = yt /zt . Rogerson. 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 . ρ < 1. hours increase and average productivity declines in response to a positive government consumption shock. t=0 where 0 < β < 1. Assume that u1 > 0. u3 < 0. The aggregate resource constraint stipulates that consumption plus investment cannot exceed output in each period: ct + gt + kt+1 − (1 − δ)kt ≤ yt . the solution for the model can be more fruitfully expressed in terms of the transformed variables ¯ ¯ ¯ ¯ kt+1 = kt+1 /zt . The key assumption is that private and government consumption are not perfect substitutes. and leisure is deﬁned as time not spent working in the market or at home: lt = 1 − hmt − hht . an increase in government consumption leads to a negative wealth effect for consumers through the economywide resource constraint. hmt is labor time spent in market work. In this expression. ct = ct /zt . cht is consumption of the homeproduced good.48) where ln (¯ ) is the mean of ln (¯t ). government consumption shocks lead to shifts in the labor supply curve so that. and hht is labor time spent in home work. Following Benhabib. in the absence of technology shocks. λt is an independent and identically distributed process with mean λ and standard deviation σλ . g (4. and µt is the innovation to ln (¯t ) g g g with standard deviation σµ . and gt = gt /zt . u2 > 0. If leisure is a normal good. cht .47) Since the technology shock follows a logarithmic random walk. hht ). Fallacy and Fantasy In this equation. and Wright [39].
zht ). However. the household’s budget constraint is cmt + km. respectively. labor effort is likely to be adjusted ﬁrst in response to a productivity shock. This says that the effective labor input can be altered even though the total number of workers is ﬁxed. the household can purchase market goods cmt . then ﬁrms may retain workers even though they are not exerting much effort. Unlike the standard model. the labor supply curve also shifts in response to a good productivity shock. home production is used to produce a nontradeable consumption good. . and household capital goods kht . Labor Hoarding A third way to resolve the productivity puzzle is through a labor hoarding argument.t+1 ≤ wt hmt + rt kmt + (1 − δm )kmt + (1 − δh )kht . Home goods are produced according to the home production function: cht = g (hht . if there are costs to hiring or laying workers off. but only after longer periods of time. one criticism of the home production theory is that it suggests that all movements out of the labor force (toward home production) are voluntary. but market capital goods are rented to ﬁrms at the competitive rental rate rt . However.t+1 + kh. where zht is a shock to home production and g is increasing and concave in labor and capital. Letting wt denote the wage rate and δm and δh denote the depreciation rates on market and home capital. kht . 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. 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).Models of Business Cycles 55 At each date. Eventually more workers may be hired or ﬁred. Household capital goods are used in home production. Alternative measures put home production to be in the range of 20–50% of GDP. Hence. In the model. thereby leading to greater variability in labor. market capital goods kmt .
The comments made regarding ﬂuctuations in the effort level of labor also apply to capital. 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]. If labor hoarding is included in the model. 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 ). capital Kt . so that the movements in the Solow residual are not entirely exogenous. where Ht denotes the total hours worked. let f denote a ﬁxed shift length. Ht = fNt . 3. the total number of workers. Fallacy and Fantasy To illustrate the role of labor hoarding. 0 < α < 1. hence. In the absence of variations in work effort. and Rebelo [52]. the work effort of each individual. then the productivity/hours correlation is reduced and more closely matches the data. 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. and Wt . Speciﬁcally.56 Business Cycles: Fact. Nt . To brieﬂy describe their framework. Yt = At Ktα [fNt Wt ]1−α . and a labor input that reﬂects variations in work effort following Burnside. The measured capital in the Solow residual does not take into account optimal ﬂuctuations in capital utilization rates. the production function can be written as Yt = St Ktα [Ht ]1−α .4. Decisions to alter effort levels will be the outcome of a maximizing decision.3. This can lead to variation in the Solow residual that is not related to changes in productivity or technology. consider a production function for aggregate output Yt which depends on an exogenous technology shock At . Eichenbaum. Thus. suppose . can be treated as a direct input into the aggregate production function. In the standard model.
The production of the intermediate good is given by dt = g (ndt . 0 < β < 1. Likewise.52) where τ < 0. wt nτt + ρt dht . where 0 < δ < 1 is the depreciation rate on capital.53) ∗ as dht = h(ρt /wt )(ct + it ). (4. respectively. (4.50) where ndt and kdt denote the amounts of labor and capital allocated to the ﬁnancial sector. and dft is the amount of transactions services used in the ﬁnal goods industry. The time required for this activity is nτt = τ(dht /(ct + it ))(ct + it ).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 . Finally. hours allocated to ∗ producing transactions services is nτt = τ ∗ (h(ρt /wt ))(ct + it ). In this expression. where wt is the real wage and ρt is the rental price of transactions services.49) where kft is the amount of capital. Households maximize the expected discounted value of utility from consumption ct and leisure lt as ∞ E0 t=0 βt U (ct . lt ) . capital. By contrast. and λt captures technological innovations to the ﬁnancial services industry. ﬁrms rent labor. nft is the amount of labor services. (4. where h = (τ )−1 . Households are also assumed to combine time and transactions services to accomplish consumption and investment purchases. dft )φt . kdt )λt . (4. τ < 0. we . φt is a shock to production of the ﬁnal good at time t.Models of Business Cycles 57 the ﬁnal good is produced according to the production technology yt = f (kft . 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. and transactions services to maximize proﬁts on a periodbyperiod basis. This implies a demand for transactions services that can be obtained from the ﬁrstorder condition ρt = wt τ (dht /(ct + it )) (4. nft . The household chooses an amount of transactions services dht so as to minimize the total transactions costs.
If the substitution effect of an increase in the marginal product of labor outweighs the wealth effect of the productivity shock. Speciﬁcally. equivalently. a positive shock to productivity occurs.5. 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. the causality runs from the productivity shock to money even though the increase in money occurs before the higher productivity shock. Consistent with the RBC view. consumption and leisure of the representative consumer will rise but so will investment demand as consumers seek to spread the extra wealth over time. This framework can be used to rationalize the observations regarding money and output. One 9 See Sims [190]. Their results are derived from a structural vector autoregression (VAR) with longrun restrictions to identify the alternative structural shocks. there will be an increase in hours of work. the central bank may expand the money supply to keep interest rates from rising too rapidly. they ﬁnd that an important source of the moneyoutput correlation is output shocks affecting inside money in the short run. Another possibility is that the central bank uses the accommodative monetary policy.58 Business Cycles: Fact. Then. The increase in output will also stimulate the demand for transactions services by households and ﬁrms. Hence. 3. As a consequence. 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. nt = nft + ndt + nτt . Fallacy and Fantasy require that the household’s time allocated to the different activities sums to 1. or a broad measure of money that includes commercial credit. if interest rates are rising and ﬁrms wish to invest more in anticipation of higher expected proﬁts. responds to the positive productivity shock. commercial credit. inside money. Suppose that a shock to the production of ﬁnal goods or. investment demand rises and output will also increase in response to the additional hours worked. .9 During an expansion. so ﬁrms will wish to ﬁnance a greater volume of goods in process. or inside money. As a result. is more closely related to output than outside money.
11 In Greenwood et al. 107]. Following Gordon [104]. during the period 1950–1990. 10 Altug.46) between a detrended measure of the relative price of capital and investment expenditures. 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. Demers. The notion behind this type of change is that production of capital goods becomes increasingly efﬁcient over time. Furthermore. . and Krusell [106. War shocks can also be modeled as a source of cyclical ﬂuctuations (see Barro [27] and Ohanian [169]).10 3. they nevertheless show that political events can have nonnegligible effects on real economic variables. 11 In Greenwood. these authors motivate their analysis by noting that. 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 )] . This has been examined by Greenwood. InvestmentSpeciﬁc Technological Shocks An important source of shocks is the investmentspeciﬁc technological change.5.Models of Business Cycles 59 way of dealing with this criticism is to introduce alternative sources of shocks into RBC models. thereby stimulating output growth. the role of investmentspeciﬁc technological shocks is used to account for cyclical ﬂuctuations. These authors consider the impact of distortionary taxes on real outcomes. 0 < θ < 1. Hercowitz. Hercowitz. [107]. In this vein. we discussed the role of government or ﬁscal shocks in generating observed comovements of the data. 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. the relative price of new equipment declined by 3% at an average annual rate while the equipmentGNP ratio increased substantially. there was a negative correlation (−0. Other papers that have employed the RBC framework with ﬁscal shocks include Braun [46] and McGrattan [158]. In the previous section.1. and Krusell [106]. the authors use a growth accounting approach to show that investmentspeciﬁc shocks can account for 60% of postwar US growth of output per manhour. Although their analysis is not general equilibrium.
58) Finally.54) The law of motion for equipment is given by ke = (1 − δe (h))ke + ie q. They evolve as zt+1 = γz exp (ζt+1 ) and qt+1 = γq exp (ηt+1 ). (5. The production function is given by y = F (hke . there is a government which levies taxes on labor and capital income at the rates τl and τk . respectively. that follow ﬁrstorder Markov processes. investment in equipment is subject to variable rates of utilization and depreciation which are convex in the utilization rate. ks . φs .57) (5. z) = z(hke )αe ksαs l 1−αe −αs . 0 < δs < 1. the service ﬂow is denoted hke . and γz and γq denote the average gross growth rates of the two processes. (5.55) where the variable q shows the investmentspeciﬁc technological change in equipment and δe (h) = b h. There is a production technology that depends on labor and the services from two types of capital goods. αe + αs < 1. (5. Structures evolve according to the standard law of motion: ks = (1 − δs )ks + ii . hence.56) Thus. φe . where ζt+1 and ηt+1 are innovations to productivity and investmentspeciﬁc technological change. The revenue raised by the government is . l . The shocks to technology and to the efﬁciency of equipment capital constitute the drivers of the t+1 t+1 model. κs > 0. Fallacy and Fantasy where ct denotes consumption and lt denotes labor supply. respectively. as = φs (ks − κs ks )2 . αs . equipment denoted ke and structures denoted ks . where ae = exp (η)φe (ke /q − κe ke /q)2 /(ke q). (5. 0 < αe .60 Business Cycles: Fact. ω ω > 1. Investment in both structures and equipment is also subject to convex costs of adjustment denoted as a = ae + as . κe > 0. where z is a measure of total factor productivity. Equipment is utilized at the timevarying rate h.
c. Equipment. and ks all have to grow at the same rate. denote this rate by g . However. we have that g = γ (γ g )αe g αs .59) where re and rs denote the rental rate on equipment and structures. ie . . and w denotes the real wage rate. it also affects the utilization cost of existing equipment this period by lowering the replacement cost of old capital goods. and it is set at 12 From the production function. The parameters that are determined using a priori information include the average growth rate of the investmentspeciﬁc shock γq . In the balanced growth path equilibrium. ae . the investmentspeciﬁc technological shock plays a key role. investmentspeciﬁc technological change increases the services from old equipment at the same time as it increases the quantity of equipment available for next period.Models of Business Cycles 61 returned to the private sector through lumpsum transfers τ. Since the inverse of q denotes the relative price of equipment in terms of consumption goods. however. and hence increases its utilization. Hence. grows faster. respectively. respectively.12 A key parameter to be estimated from the data is γq . (5. the depreciation rate on structures δs . (5. this quantity is determined as the ratio of Gordon’s [104] price index of qualityadjusted equipment and the price index for consumption. The ﬁrst parameter is determined using data on the inverse of the relative price of investment goods following Gordon [104]. the quantity of labor l and the utilization rate h are constant. We note that the economy displays balanced growth as zt and qt grow at the rates γz and γq . as .60) In this analysis. This yields the government budget constraint: τ = τk (re hke + rs ks ) + τl wl . A higher value of q directly affects the quantity of equipment that will be available to produce output next period. at the rate ge = γq g . or z q g = γz ge = γz 1/(1−αe −αs ) αe /(1−αe −αs ) γq . 1/(1−αe −αs ) (1−αs )/(1−αe −αs ) γq . 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 . The resource constraint and the accumulation equation for structures imply that y. and the tax rate on wage income τl . is .
the investmentspeciﬁc technological shocks explain around 28% of business cycle ﬂuctuations as captured by the standard deviation of output. [107] calibrate the adjustment cost parameter under two alternative assumptions. αs = 0. Unlike the standard technology shock in RBC models. A positive shock to this variable works by increasing the rate of return to equipment investment. For a low value of φ. αe = 0. which tends to raise the stock of equipment next period. the symmetry condition φe = (g /ge )2 φs ≡ φ leaves only the parameter φ to be determined. which raises utilization of equipment today. only the q shock is assumed to be operative. . This constitutes an upper value for the contribution of the shocks q. β = 0. this ﬁgure is 32%. The model is simulated to understand the contribution of the investmentspeciﬁc technological shocks in generating cyclical ﬂuctuations. 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. and bh ω = 0. Likewise. Simultaneously. This leads to increased employment and output.056 and τl = 0. the fraction of output directly affected by the shock will typically be small.97. A high value of φ works to reduce investment and to increase the procyclicality of consumption.59. the impact of the investmentspeciﬁc technological shock depends on the quantitative importance of the transmission mechanism.20. The second set of parameters is calculated using information on the longrun behavior of a set of the model’s variables. They conclude that though investmentspeciﬁc shocks contribute to business cycle ﬂuctuations. 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. g = 1. The restriction that adjustment costs are zero along the balanced growth path yields the values κe = κs = g . Greenwood et al. δs = 0. the parameter φ is set to equalize the consumption/output correlation implied by the model to that in the data.032. This standard calibration exercise yields θ = 0.62 Business Cycles: Fact.53. For this purpose. where the parameters b and h cannot be identiﬁed separately. there is a decline in the equipment’s replacement value. Hence. Fallacy and Fantasy 1. According to the ﬁrst.40. Hence. Given that equipment investment is only 7% of GNP and 18% of the value of output being derived from the use of equipment. τk = 0.0124. By contrast.40. they are not the main factor. the investmentspeciﬁc shock affects uses of factors such as labor and capital through changed investment opportunities. They ﬁnd that under a high value of φ.18. ω = 1.12. Furthermore.
lt ) . ν1 > 1. The new feature in Finn’s analysis is that capital utilization requires energy: et ν0 utν1 = . 0 < α < 1. the source of large supplyside shocks for the RBC agenda has been sought in oil shocks.5.2. The production function is given by yt = (zt lt )θ (kt ut )1−θ . As in the standard RBC framework. and ω1 > 1. . By contrast. kt ν1 ν0 > 0. Rotemberg and Woodford [183] provide a rationale for the role of energy price shocks in a model with imperfect competition.Models of Business Cycles 63 3. with c α (1 − lt )1−α u(ct . kt is the stock of capital. 0 < β < 1. where zt is an exogenous technology shock. and ut is the rate of utilization of capital. As in the previous model. lt ) = t 1−σ 1−σ −1 . ω1 where 0 < δ(ut ) < 1. there is a representative consumer with preferences: ∞ E0 t=0 βt u(ct . the depreciation of physical capital depends on its utilization rate so that kt+1 = (1 − δ(ut ))kt + it . capital utilization is the channel through which energy shocks enter the model. In her model. σ > 0. 0 < θ < 1. Finn [92] argues that it is possible to rationalize the role of energy shocks in models with perfect competition. δ(ut ) = ω0 utω1 . Barsky and Kilian [28] analyze the relationship between oil shocks and changes in economic activity. ω0 > 0. We brieﬂy describe her model before concluding this chapter. Energy Shocks From the beginning.
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 .64 Business Cycles: Fact. the economywide resource constraint is given by ct + it + pt et ≤ yt . Finn [92] shows that the quantitative response of valueadded and real wages to a shock in energy prices implied by the model is in line with that in the data. If the price shock is persistent. capital. an increase in pt reduces et and ut through the production function. . there are two channels for the transmission of energy shocks in the model. Finally. and lt continue into the future and reduce the future marginal productivity of capital. This shows the direct effect of energy. Thus. where pt is the price of energy. and energy. First. ut . The decrease in et further reduces the marginal product of labor. Thus. This tends to reduce the investment it and the future capital stock kt . but there is also an indirect effect which works through the effect of the utilization rate on the depreciation of capital. the production of output depends on labor. Hence. 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. then the decrease in the values of et . Finn [92] describes qualitatively and quantitatively the impact of increases in the price of energy. a positive energy price shock operates in a similar way as a negative technology shock. and hence the real wage wt and work effort lt .
Chapter 4 International Business Cycles The international business cycle literature provides an extension to the original real business cycle (RBC) approach by allowing for openeconomy considerations. nor can it be used to discuss notions of international risk sharing. The closedeconomy model cannot be used to understand the propagation of shocks across countries or regions. Backus.1. Backus et al. They can be used to model the impact of the ability to trade in goods and to borrow and lend internationally. Kehoe. They consider a twocountry 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. 4. and Kydland [25] consider data on 12 developed countries over a sample period dating from 1960 to 1990. including correlations of macroeconomic aggregates across countries and movements in the balance of trade. They study the properties of a baseline model against the actual features in the data. we review some of the facts of international business cycles and discuss models that have been used to rationalize the ﬁndings. openeconomy business cycle models allow for the role of technology processes in different countries in generating cyclical ﬂuctuations. FACTS One of the important areas of research has been to derive the socalled stylized facts of international business cycles. In this chapter. One of the aims of the RBC approach has been to determine whether twocountry versions of the basic model can account for both the comovements studied in closedeconomy models as well as international comovements. [26] study 65 . Unlike the closedeconomy model.
In this regard. Canada. ∗ ) = 0. 26] estimate a bivariate AR(1) process for the domestic and foreign shocks as zt zt∗ =A zt−1 + ∗ zt−1 t ∗ t . The positive t t offdiagonal 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. Italy. Germany. is much higher in the data than it is in the theoretical model.906 0. Thus. where zt . 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 ﬁndings can be summarized under two headings: • The quantity anomaly: In the data. The individual countries they consider are Australia. Based on several estimated speciﬁcations for the US and the European aggregate. zt∗ denote the technology shocks to the domestic and foreign countries. and the United States.66 Business Cycles: Fact. the crosscountry correlations of output are greater than those of productivity. Austria. ∗ denote the respective innovations to these t shocks.088 .088 0. the theoretical model implies the reverse ranking. Switzerland. The large autoregressive coefﬁcients displayed on the diagonal capture the persistence in observed productivity. the exercise in this literature is to replicate the correlation properties among the different series given the variability and correlation properties of the shocks.00852 and Corr( t . the United Kingdom. and t . Backus et al. By contrast. 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. and the crosscorrelations of productivity are greater than those of consumption. deﬁned as the relative price of imports to exports. 0. The price anomaly: The volatility of the terms of trade. [25. The data are ﬁltered using the Hodrick–Prescott ﬁlter. respectively. measured as the Solow residual. these authors use a symmetric speciﬁcation with A= 0. Japan.906 with Std ( t ) = Std ( ∗ ) = 0. France. • .258.
In other words. 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. which is accompanied by output and employment increases. and employment in the theoretical model. investment and the trade balance are more volatile in the model than they are in the data. this ratio displays little volatility. which allows them to obtain standard errors. [26] sample to 20 industrialized countries over the period 1960–2004. the US. In their original analysis. Furthermore. The ﬁndings of Ambler et al. which leads to a trade deﬁcit in the domestic country. what is typically observed in the data are positive comovements in these series across countries. This changes the outputinvestment correlation from positive to negative in the domestic country. [25] also examine a variety of perturbations of their original setup. whereas in the model. Backus et al. This feature of the model leads to the crosscountry correlations of consumption being much higher in the model than in the data.International Business Cycles 67 Ambler. Cardia. they are negative. [14] conﬁrm the ﬁndings of Backus et al. First. The twocountry frictionless international business cycle model also implies that consumers in different countries can share risk perfectly. investment. and Zimmermann [14] extend the Backus et al. implying that the real exchange rate is also less volatile in the model than it is in the data. They also ﬁnd that the crosscountry correlations of output. investment. They estimate the moments of interest using the generalized method of moments (GMM) estimation. Since investment responds strongly to a positive productivity shock. This fact leads to negative crosscorrelations between output. 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. By contrast. say. With perfect risk sharing. suppose a positive technology shock occurs in one country. [25. 26]. This leads to a strong investment response in the US. In a model with perfect capital mobility. we thus observe a ﬂow of factors from other countries or regions to the US. the increase in investment plus consumption is greater than the increase in output. and employment are positive and generally high in the data. The international RBC model produces these results because it implies that the incentive is to use inputs where they are most productive. though with lower magnitudes.
but stems from the permanent income hypothesis: when domestic and foreign productivity shocks are correlated. They ﬁnd that this friction has the effect of reducing the volatility of investment and the trade balance. it is the correlation between technology shocks that leads to any connection between countries. The standard international RBC model studied by Backus et al.088 to 0. they consider large spillovers across countries by changing the offdiagonal elements of the A matrix from 0.68 Business Cycles: Fact. The ﬁrst of these involves introducing a small trading friction in the form of a transport cost. the foreign agent increases consumption and reduces investment in anticipation of future income increases. and the crosscountry correlation of output goes from being negative to positive. they eliminate all trade in goods and assets by considering an autarky situation. Hence. 26] considers the case with complete contingent claims and perfect risk sharing across different countries. elimination of trade in statecontingent claims does not help to lower the crosscountry correlations of consumption. Fallacy and Fantasy counterfactually high. 4. they ﬁnd that investment and the trade balance become much less volatile. [25. Cole [71] and Cole and Obstfeld [72] examine a twocountry real version of the Lucas asset pricing model for this purpose (see Lucas [152]). THE ROLE OF INTERNATIONAL RISK SHARING Before we continue with other extensions that have been proposed in the international RBC literature. They ﬁnd that the results are very similar to the case with a small transport cost.2 and the correlation of the innovations from 0. Agents from both . but has little effect on the crosscountry correlations of consumption and output. Surprisingly. Next.2. the crosscountry correlation of consumption increases further. as does the positive correlation of consumption across countries. With this change. Backus et al. In this case. 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.258 to 0. [25] argue that this result is not due to international risk sharing considerations. They also examine various sorts of trading frictions. However. a rise in productivity in the domestic country signals a rise in productivity in the foreign country through the spillover effects. Second.5. it is important to understand the role of alternative ﬁnancial arrangements in generating the crosscountry correlations.
j = p(st = sj st−1 = si ). c2 ) lim This requirement on the utility function ensures that both goods are consumed in equilibrium. The functions y1 : S → Y and y2 : S → Y are continuous functions that are bounded away from zero.t ) . The utility function U : R+ → R is continuously differentiable. y ]. We assume that endowments are stationary in levels. To do this.2. strictly increasing. where pi. c2 ) = 0. Let π denote the vector of stationary probabilities which satisfy π = P π. j) element pi.1. Let st ∈ S ⊆ R m denote a vector of exogenous shocks that follows a + discrete ﬁrstorder Markov process with a stationary probability transition ˆ matrix P with (i. c2 > 0. and strictly concave.t }∞ deﬁned by t=0 ∞ E0 t=0 βt U (c1.t . Country 1 has a random endowment of good Y1 and country 2 has a random endowment of good Y2 . c2. 4. For all c1 . j j 0 < β < 1. Y1 and Y2 . c2 ) = ∞. (2. U1 (c1 . The representative consumer in country j has preferences over random j j sequences {c1. We assume that endowment is a timeinvariant function of the exogenous shock.2.International Business Cycles 69 countries are identical in terms of preferences and differ only in terms of endowments. Neither good is storable. 2 Assumption 2. c2. ¯ ¯ Assumption 2. agents observe the current realization. There are two goods. Pareto Optimal Allocations We begin by characterizing the Pareto optimal allocations.j . c2 ) lim U1 (c1 . 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 . c1 →0 U2 (c1 .1) We have the following assumption.t . where y > 0 and y < ∞. Deﬁne Y ≡ [y.1. In the ˆ ˆ beginning. c2 →0 U2 (c1 .
Fallacy and Fantasy resource constraints. c2. i = 1. c2.t )] (2. where ci.t . Since preferences are timeseparable.5) . 2. c2 (s t )) t=0 st (2. . the endowment is nonstorable. Let φj denote the Pareto weight for country j.t denotes the consumption of good i by residents of country j. where s t = (st . .t . and there is no investment process. φ1 φ2 . Deﬁne ˆ π(s t ) = π(st st−1 )π(st−1 st−2 ) · · · π(s1 s0 )π(s0 ).t .t . c2.t ) 2 2 U2 (c1.t + ci.t = yi. φj U2 (c1. c2.t ) 1 1 U2 (c1. Let µi (st ) denote the Lagrange multiplier for the resource constraint for good i.t ) = µ1 (st ). c2 (s t )) + φ2 U (c1 (s t ). c2. c2. the social planner’s problem at time t involves solving the static problem 1 1 2 2 max[φ1 U (c1. st−1 . The ﬁrstorder conditions imply that φj U1 (c1.t ) = µ2 (st ) for j = 1. or 1 1 U1 (c1.t .3) subject to 1 2 ci.t .t .4) (2. s0 ) denotes the history of the shocks at time t. φ1 (2. c2. 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 ). 2. .t ) + φ2 U (c1.70 Business Cycles: Fact.t . .2) subject to ci1 (s t ) + ci2 (s t ) = yi (st ).t . 2. j i = 1. The social planner chooses sequences {ci1 (s t ).t ) j j j j = = φ2 . c2.t ) 2 2 U1 (c1.
which also causes the output of country 2 to increase through positive spillover effects. 4. we will examine the role of alternative ﬁnancial arrangements that can be used to rationalize the observations.2. In what follows.t )1−ρ 1−ρ . Also note that a shock to the output of country 1.t .t c2. the Pareto optimal consumption allocations satisfy 1 c1.t 2 c1. (2.6) Assuming that preferences are of Cobb–Douglas variety.t = δy2. It is straightforward to demonstrate .8) (2. This implication of the model has been shown to lead to the counterfactually high crosscountry correlations of consumption implied by the standard international RBC model. c2.t = (1 − δ)y1.t = δy1.7) (2.International Business Cycles 71 Hence. (2.t = (1 − δ)y2. we ﬁnd that the ratio of marginal utilities across agents in different countries is equalized for all states and goods.2. This is an implication of perfect risk sharing that arises in a complete contingent claims equilibrium.10) where δ= 1+ φ2 φ1 σ 1 and σ = ρ .t 2 c2. These expressions show that consumers in each country consume a constant fraction of output in each period.9) (2.t ) = α 1−α (c1. It is convenient to assume that preferences are of the Cobb–Douglas variety: U (c1.t 1 c2. will cause consumption in both countries to increase. 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.t .
t + p(st )c2. 1 For further discussion. j = 1. Given the Markov nature of uncertainty.t + p(st )z2.t p(st ) 2 + c2.t + st+1 ∈S 2 z1. Consider ﬁrst the case when agents can trade oneperiod contingent claims that pay off in each state next period on the consumption good of each country.t = c1.t (2. 7.12) where q(st+1 .t + p(st )c2.t + z1.1 Agents can also trade in the goods of the other country. 2 are given by 1 1 1 1 y1.t + + p(st ) 2 + z2. Fallacy and Fantasy that there exists a contingent claims equilibrium for this economy.t − 2 2 c1. is given by 1 1 CA1 = y1.t − c1. see Altug and Labadie [6].t . p(st ) (2.t . CA = y2. st ) st+1 ∈S 2 z1 (st+1 ) 2 + z2 (st+1 ) . The current account for each country.72 Business Cycles: Fact. 2. Thus.t 1 1 q(st+1 . Ch. and that the competitive equilibrium allocations are Pareto optimal.t q(st+1 . st ) z1 (st+1 ) + p(st )z2 (st+1 ) 2 c1. The sequential budget constraints for residents of country j for j = 1.11) y2. Let p(st ) denote the relative price of good y2 in terms of the numéraire good y1 . it is without loss of generality to consider only oneperiod contingent claims. denoted as CAj .t = p(st ) 2 + c2. 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. . 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.
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 marketclearing 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
+β
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 ﬁrstorder 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 marketclearing 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 noassettrading 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 .
despite the absence of trade in ﬁnancial assets. A large negative shock in the amount of a good is similarly transmitted across borders.t .t + c1.t .t + c2. (2. Efﬁcient risk sharing can occur despite the lack of ﬁnancial assets and insurance.t = y2.t = αy2.t This price can be substituted into the consumption functions above to show that 1 c2. Notice that the noassettrading allocation is identical to the central planning allocation if α=δ= 1+ or φ = 1+ 1 φ2 φ1 σ .t = (1 − α)y1.t . A large and positive shock in the amount of good y1.t . The international ratio of marginal utilities across countries is identical across goods and states.18) Substitute the consumption functions into the marketclearing conditions and solve for the relative price to show that pt = (1 − α)y1.t . • .t is positively transmitted to the residents of country 2 by the increase in demand (and hence the relative price) for good y2. ρ −1 1−α α and φ2 = 1 − φ1 . Fallacy and Fantasy Equilibrium in the two goods markets requires that 1 2 c1. efﬁcient risk sharing occurs through changes in the relative price of goods. This exercise illustrates several points: • The absence of international capital mobility does not necessarily imply that the allocation is not Pareto optimal.76 Business Cycles: Fact. 2 c1.t . 1 2 c2.t = y1.17) (2. Hence. αy2.
t = α1 [wt1 + wt2 ] .t + pt c2.t . 4. so that p1.t c2. 2 2 c1.t = αy1.t denotes the relative price of good y1.4. the positive transmission of shocks occurs because countries specialize in the production of goods.t α2 [wt1 + wt2 ] .t c1.t .t ≤ p1. α2 . Nonspecialization in Endowments To illustrate the impact of nonspecialization. despite the absence of ﬁnancial capital mobility.t y2.2.International Business Cycles 77 • • Notice that we can price ﬁnancial assets in the model. 2 2 2 p1. αw y2.t = (1 − α)y1.t c2. αw y1.t c1.t ≤ p2. Let α1 . by 1 1 1 p1. The total consumption of countries 1 and 2 is 1 1 c1.t in terms of wt and p2.t + wtb .t + cw.t + p2. under the assumption that the current account is zero. and can show that real interest rates and real asset returns will be equal for the two countries. w] denote the realization of good w in country j. αw denote the expenditure shares under the assumption of Cobb–Douglas preferences and let w be the numéraire good. Assume that both countries receive an exogenous and stochastic endowment of good w and let w j : S → W = ¯ [w. Call this third good w.t + p2.t denotes the relative price of good y2. respectively.t . Cole and Obstfeld [72] introduce a third good.t + wta . Agents in countries 1 and 2 have budget constraints given. As Cole and Obstfeld [72] point out.t + (1 − α)y2.t + αy2.t y1.t + pt c2. The equilibrium relative prices satisfy p1.t in units of wt .t = p2. Notice that correlation of the total value of consumption of country 1 and country 2 is positive and equal to one.t + cw.
2. ( w 1 +w 2 ) and ( w 1 +w 2 ). a condition for Pareto optimality. in the absence of trade in ﬁnancial assets. countryspeciﬁc shocks.t = αw wt1 wt1 + wt2 wt2 wt1 + wt2 + α1 y1. α2 . Hence. If these shocks are not perfectly correlated. wt2 are not perfectly correlated. only if the wt varies with st . Similar expressions can be derived for the consumption of goods y2 and w. Call this good n and assume that nj : S → N = [n. n] denotes the realization of good n in ¯ country j. Nontraded Goods Suppose now that the third good is a nontraded good. the country must export more of the good in which it specializes in production to ﬁnance the import of good wt . by deﬁnition.t = αw 2 c1.t or y2. are sectorspeciﬁc shocks. The intuition is that. there are gains to asset trading that improve risk sharing and allow diversiﬁcation.5. when there are sectorspeciﬁc shocks.t . + α2 y1. αn denote the expenditure shares under the assumption of Cobb–Douglas preferences and let y1 be the numéraire good. Let α1 . Fallacy and Fantasy The consumption of good 1 by agents in countries 1 and 2. can be shown to satisfy 1 c1. then it is beneﬁcial to trade equity shares or other forms of ﬁnancial assets. The ratio of marginal utilities of both countries for each good will be equal to a constant across all states. and industryspeciﬁc shocks. which affect all sectors within a country. Since the current account must always be balanced. 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.78 Business Cycles: Fact. The shocks to wt1 and wt2 must be perfectly correlated for the allocation with no trade in ﬁnancial assets to be Pareto optimal. is constant as the total t t t t w1 w2 4. where wt1 . whereas shocks to wt1 . This helps to clearly distinguish between countryspeciﬁc shocks. wt2 . Under the .t are.t . t t share of the endowment of wt . under the assumption of Cobb–Douglas preferences. 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.
j Let zi.t+1 + q2.t + p2. 1 − αn α2 = y2.t ] + z2. 1 − αn α2 = y1.t for j = 1.t . n1 4.t c2. 1 − αn α1 = y2. nt are perfectly correlated. The ratio of marginal utility across countries for a traded good will now depend 1 2 on the ratio nt2 . the demands for the goods satisfy 1 c1.t = 1 c2.2. If this sector constitutes a large fraction of consumption.t ] ≥ c1. nontraded goods provide one vehicle for reducing the large crosscountry correlations implied by the baseline model. 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).19) . then even if consumption of traded goods is perfectly correlated. We now discuss the impact of asset trading and relate it to the model without asset trade discussed earlier.t y2.t 2 c2.t . j j j j j j j (2. There are gains from international risk sharing through asset trade.t [y1.t + q1. 1 − αn 1 2 Each country consumes its endowment of the nontraded good.t+1 . An agent’s budget constraint is z1. 2 and i = 1.t z2.International Business Cycles 79 assumption of balanced trade and Cobb–Douglas preferences. 2 denote the shares of good i held by an agent in country j at the beginning of period t.t .6. the correlation of national consumption levels may be close to zero. Notice that the correlation of consumption across countries will now depend on the proportion of a country’s consumption that is nontradeable.t 2 c1. Thus.t z1. nt .t + q2.t + q1.t . nt .t α1 y1. Unless nt . then the resulting t allocations will not be Pareto optimal.t [p2. Trade in Equity Shares We now introduce trade in ﬁnancial assets.
c2. The . c2. We commented that we could price ﬁnancial assets even 1 1 if these assets were not traded. depending on the initial distribution of the claims. In such a world.t )p2.20) (2. 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.t+1 + p2.t+1 ].t = βEt U1 (c1. then the equilibrium allocation with no trade in ﬁnancial assets can be achieved.t = 1/2 for j = 1. national wealth is equal across countries and agents have perfectly diversiﬁed portfolios.t+1 ].t .t+1 )[q1. illustrates an important point.t+1 . when combined with our discussion of the equilibrium with no trade in ﬁnancial assets. so that zi.t = α and 2 2 z1. it does not require the existence of a full set of contingent claims. U1 (c1.t = 1 − α. This simple example.t . all equity shares are held and the endowment of each good is completely consumed. U1 (c1. If we assume that z1.80 j Business Cycles: Fact. we can achieve at least two stationary allocations.t .t = z2.21) (2. International risk sharing can be achieved through ﬂuctuations in relative prices in the current account and by trade in ﬁnancial assets.t )q2.t . The ﬁrstorder conditions are U1 (c1. Fallacy and Fantasy Let µt denote the Lagrange multiplier. 2 and i = 1. 2 so that φj = 1/2 and δ = 1/2.t . Agents across countries have identical consumption in this case. The agent maximizes his objective function subject to the constraint. Lucas [152] assumes that agents hold identical j portfolios. then portfolio diversiﬁcation may be achieved by specializing in the holding of equity shares of one country only.t+1 + y1. Hence. there was no trade in ﬁnancial assets and specialization in endowments and the allocation was Pareto optimal.t = z2. c2.t+1 )[q2. In particular. There has been substantial literature on the lack of international portfolio diversiﬁcation and the degree of international consumption risk sharing. In the initial model described earlier. c2. unlike the economy in which there is no trade in ﬁnancial assets. c2.t ) = U2 (c1. 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.22) In equilibrium.t )q1. c2. If we introduced trade in equity shares when there is a third good w that is produced by both countries. j j j j j j j j j j j j (2.t+1 .t+1 y2.t = βEt U1 (c1. If these shocks are perfectly correlated.
. Baxter and Jermann [36] argue that the failure of international diversiﬁcation is substantial. By contrast. 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.2. As we have noted above. Nevertheless. Clearly. the existence of nontraded goods. Lewis [145] also documents that there is insufﬁcient intertemporal risk sharing in consumption. 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. Limited Risk Sharing As described earlier. 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. as we have noted above. A more recent paper by Heathcote and Perri [117] extends the Baxter and Jermann model to include more than one traded good.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. but that when nonseparability and asset trade restrictions are combined. 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. Devereux. [25]. Their model incorporates human and physical capital and. that consumption insurance is available through relative price ﬂuctuations and that these price ﬂuctuations are capable of achieving efﬁcient risk sharing. combined with the assumption that utility is nonseparable in traded and nontraded goods. Gregory.7. 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. she cannot reject the hypothesis that there is risk sharing. who show that the data are inconsistent with the implications for consumption for the baseline international RBC model. makes it more difﬁcult to determine the optimal degree of consumption risk sharing. 4. whether there exists perfect risk sharing in the data appears to depend on the model speciﬁcation adopted by the researcher. They ﬁnd.
Their analysis involves extending the analysis in Cole and Obstfeld [72] to incorporate an explicit production side. the inability of agents to share risk perfectly and to fully offset the effects of idiosyncratic shocks leads to lower crosscountry correlations of consumption and higher crosscountry correlations of output. the results for the crosscountry correlations are obtained only if the productivity shocks are highly persistent and there are very little spillover effects across countries. 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 noncontingent bonds can be traded internationally. 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 )). Heathcote and Perri [117] examine the implications of a twocountry model under alternative assumptions about the ﬁnancial structure. In these models. Kehoe and Perri [126] consider a model where market incompleteness is obtained endogenously through the introduction of imperfectly enforceable international loans. In their framework. li (s t )). investment. t=0 s t . They show that the model matches the correlations in the data under a ﬁnancial autarky assumption. This analysis builds on the earlier works of Kehoe and Levine [124.82 Business Cycles: Fact. given a trade structure. where Ai (s t ) is a random shock. Output in each country is produced using capital and labor according to the constantreturnstoscale production function: F (ki (s t−1 ). 125]. 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. but the crosscountry correlations of output. Kehoe and Perri [126] consider a standard international RBC model with production and capital accumulation. Alvarez and Jermann [12]. Fallacy and Fantasy In this vein. Moreover. Kocherlakota [135]. These authors ﬁnd that incomplete asset markets help to reduce the crosscountry correlation of consumption. and others on debtconstrained asset markets. and hours worked remain counterfactually negative.
Ai (s r )li (s r )) + (1 − δ)ki (s r−1 ). the competitive equilibrium allocations solve the social planner’s problem deﬁned as ∞ π(s t )βt φ1 U (c1 (s t ). r≥t given ki (s t−1 ). t=0 s t (2. s ) = max r=t sr t βr π(s r s t )U (ci (s r ).International Business Cycles 83 where ci (s t ) denotes consumption of residents of country i. li (s r )) ci (s r ) + ki (s r ) ≤ F (ki (s r−1 ). where Vi (ki (s t−1 ). As Kehoe and Perri [126] explain. ci (s r ) for r ≥ t to solve ∞ Vi (ki (s subject to t−1 ). l2 (s t )) . The reason is that future decision variables such as ci (s r ) and li (s r ) for r > t enter the current enforcement constraints. li (s r )) ≥ Vi (ki (s t−1 ). s t ) denotes the value of autarky from s t onwards. li (s r ).2 2 F (ki (s t−1 ). s t ) involves choosing ki (s r ). The enforcement constraints are expressed as ∞ βr−t π(s r s t )U (ci (s r ).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 ). An . Ai (s t )li (s t )) + (1 − δ)ki (s t−1 ) . r=t sr (2. s t ). i=1 (2. 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.23) subject to the resource constraints ci (s t ) + ki (s t ) = i=1. l1 (s t )) + φ2 U (c2 (s t ).24) The innovation in Kehoe and Perri [126] is to formulate a version of this problem that allows for enforcement constraints. the social planner’s problem with enforcement constraints cannot be formulated recursively as a dynamic programming problem. Under complete markets.
84 Business Cycles: Fact. a bond economy. s t ) . li (s r )) − Vi (ki (s t−1 ). The bond economy stipulates that all intertemporal trades must be implemented with an uncontingent bond. 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 ).2 This approach is implemented by deﬁning βt π(s t )µi (s t ) as the Lagrange multiplier on the enforcement constraints. Mi (s t−1 ) satisﬁes Mi (s t ) = Mi (s t−1 ) + µi (s t ). and an economy with enforcement constraints. Kehoe and Perri [126] generate solutions for three different economies: complete markets. t ≥0 with Mi (s t−1 ) = φi . . s t )] plus the terms relating to the resource constraints. Fallacy and Fantasy alternative approach involves adding a new pseudo state variable to achieve a recursive formulation of the original problem. Thus. Mi (s t ) are just the original planning weights plus the sum of the multipliers µi (s t ) along the path s t . see Marcet and Marimon [155].24). li (s t )) − Vi (ki (s t−1 ). we will focus on the results of simulating this model and compare it with alternatives that have been obtained in the literature. 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 standard resource constraints deﬁned by (2. Therefore. The Lagrangian function becomes comprised of three terms: the weighted sum of utilities deﬁned in (2.23). Since π(s r ) = π(s r s t )π(s t ). Thus. li (s t )) t=0 s t i + µi (s t )[U (ci (s t ). In this expression. It is beyond the scope of this book to derive a solution for the model with enforcement constraints. 2 For other applications. the Lagrangian function is written as ∞ βt π(s t ) Mi (s t−1 )U (ci (s t ).
Under the complete markets speciﬁcation. AL) = k α (Al )1−α . and a third one allows for high spillover with the diagonal elements equal to 0. Similar to Backus et al. α = 0. σ = 2. and δ = 0. b(s t ) ≥ −b. Preferences in each country are taken to be of the form c γ (1 − l )1−γ U (c. the variance of the innovations to the productivity shocks in each country is set at Var( i.9 and the offdiagonal elements are set at zero. and bi (s t ) is the quantity of the bond by consumers in country i. In the bond economy. are parameterized in different ways. the autoregressive coefﬁcients of the shocks for each country captured by the diagonal elements of the A matrix are set at 0.36. In the economy with enforcement constraints. γ = 0. net exports and investment are much more volatile in the model than they are in the data. the model displays many of the anomalies reported in the literature.025.36. albeit with some minor improvements in the various statistics.26) The parameter values that are considered are standard. According to one parameterization intended to capture substantial persistence. the consumption correlations are signiﬁcantly higher than the output correlations in the model. (2. which governs the persistence properties of the shocks and the spillovers between them. whereas they are positive in the data. q(s t ) is the period t price of an uncontingent that pays off one unit in period t + 1 regardless of the state. First.85 and the offdiagonal elements equal to 0. The ﬁndings from the model echo many of the earlier ﬁndings.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. 0 < α < 1. Third. where 0 < b < ∞. the three discrepancies between the model and the data remain.25.27) 1−σ .15. and given by β = 0.t ) = 0.99. [25].99 and the offdiagonal elements equal to zero. l ) = 1−σ and the production function as F (k. (2. There is also a borrowing constraint that requires that ¯ ¯ borrowing cannot exceed some ﬁnite bound. σ ≥ 0. These values are consistent with those assumed by Kollman [136] and Baxter and Crucini [34]. the output and . 0 < γ < 1.007 and the correlation of the shocks is set at 0. Second. The coefﬁcients of the A matrix. A second parameterization allows for high persistence with the diagonal elements equal to 0. whereas these correlations are the opposite in the data. the crosscountry correlations of employment and investment are negative in the model.
In the foreign country. Consumption increases substantially more in the domestic country because consumption and labor complement each other. Hence. Suppose that the social planner tries to implement . investment. The net ﬂow of investment increases the trade deﬁcit in the domestic country. The only remaining discrepancy is that net exports are procyclical in the model whereas they are countercyclical in the data. and employment remain. The capital stock in the domestic country increases. they ﬁnd that the volatilities of investment and net exports are diminished. they are somewhat dampened. Kehoe and Perri [126] note that the enforcement constraint acts as an endogenous inhibiting factor. Finally. risk sharing across countries implies that consumption increases in the foreign country. let us consider the impact of a positive productivity shock in the enforcement economy. In the literature. Considering economies with exogenous adjustment costs to investment. the frictionless international RBC model implies that investment ﬂows to the country with the higher productivity shock. the responses are similar to those for the complete markets economy. This increases the productivity of both capital and labor in the domestic country. we see that the differing responses of the domestic versus the foreign country lead to the negative crosscountry correlations of employment and investment. the authors examine the response of the variables in domestic and foreign countries to a positive shock to productivity in the domestic country. exogenous adjustment costs are typically added to inhibit the ﬂow of investment. the crosscountry correlations of employment and investment have now switched from being negative to positive. we observe the high crosscountry correlations between domestic and foreign consumption. although the crosscountry correlation of consumption is greater than that of output. but that the anomalies regarding the crosscorrelations of output. leading to very volatile investment and net exports. we notice declines in employment and investment. consumption.86 Business Cycles: Fact. Second. but because of restrictions on asset trading. so resources are optimally shifted there. Thus. 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. and third. Finally. As discussed earlier. Fallacy and Fantasy consumption correlations are both positive and closer to each other. as domestic residents save more and also as investment from abroad ﬂows there. In the bond economy.
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. investment. but also for the behavior of output. OTHER EXTENSIONS The international RBC literature has sought to reconcile the ﬁndings in the data not only in terms of the crosscountry consumption correlations. Over time. however. the planner also builds up the capital stock in the foreign country to ensure that the risksharing arrangement between the domestic and foreign countries is not reneged upon. In terms of the behavior of consumption. The increased investment ﬂows to the domestic country further increase the value of autarky. as the productivity shock dies out. 4. this is not possible. As a result. Hence. In the presence of an enforcement constraint. They then compare these ﬁndings for the international business cycle to those obtained for data between regions within a country — the socalled “intranational business cycle”. increasing the incentives for default. Hess and Shin [119] reexplore the two international business cycle anomalies emphasized by Backus et al. and the real exchange rate. to ensure that domestic country residents consume more than foreign country residents.International Business Cycles 87 the complete markets allocation for this economy. [25] as well as establish the pattern of productivity growth between industries and countries. Furthermore. models that relax assumptions regarding preferences. We describe a few of these extensions in what follows. 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).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.3. and the presence of additional shocks have been developed. Hence. [13] modify the supply side of a twocountry model by adding multiple . Ambler et al. the value of autarky diminishes and the relative weight on the home country also falls. The high and persistent productivity shock in the domestic country increases the value of autarky. the planner must restrict investment to the domestic country to prevent the default option from being exercised. the production side.
and a nontradable good. It also predicts crosscountry correlations of employment and investment that are closer to the data. 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. Hence. consisting of the tradable goods of countries 1 and 2. Basu [29]. a nontradable good. The authors ﬁnd that the model predicts the correlation between home and foreign output. They employ preferences that depend on consumption of the tradable goods of countries 1 and 2. We already considered the role of variable factor utilization in reconciling the behavior of procyclical productivity in closedeconomy business cycle models (see also Burnside and Eichenbaum [51]. However. output in the traded and nontraded goods sector of each country is produced using sectorspeciﬁc capital and labor which is mobile between sectors. The notion is that a positive shock to productivity in the domestic country will tend to reduce the investment .88 Business Cycles: Fact. Likewise. Stockman and Tesar [200] introduce a nontraded goods sector in each country. and understates the variability of the trade balance. Similar to our discussion above. Baxter and Farr [35] develop a model with variable capital utilization as a way of accounting for the international correlations. or Basu and Kimball [30]). In their model. and leisure. 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. They argue that variable capacity utilization has the potential to account for the comovement of factor inputs across countries. Preferences are assumed to be nonseparable with respect to a composite good. there is no international capital or labor mobility. Instead. overstates the crosscountry correlation of aggregate consumption. The model generates a higher crosscountry correlation of output than standard onesector models. They conclude that an international business cycle model driven solely by productivity shocks cannot account for the ﬁndings. they argue that what is needed is a source of nationspeciﬁc shocks that shift demand. they ﬁnd that introducing nontradable goods does not sufﬁce to reduce the crosscountry correlation of consumption. and greatly overstates the crosscountry correlation of tradable consumption. they argue that introducing nontraded goods may be a way of reestablishing the link between a nation’s output and its spending. Fallacy and Fantasy sectors and trade in intermediate goods.
where Bt+1 is the quantity of bonds carried over into the next period. Kt . The correlation of the innovations to the technology shocks is set at 0. Zt : St = Zt Kt . Similar functions characterize the foreign country. and φ < 0. International trade in assets is made solely with oneperiod. Baxter and Farr [35] assume that preferences are given by the speciﬁcation in equation (2. which is consistent with earlier studies. and the utilization rate. real pure discount bonds which have the price qt = [1 + rt ]−1 . 0 < α < 1.International Business Cycles 89 ﬂow across countries by leading to an increase in capital utilization rather than investment. 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.258. many of the puzzles of the international business cycle literature can be addressed successfully by devising a mechanism to limit investment ﬂows across borders. The parameterization of the capital utilization function and the adjustment cost function are key aspects of the quantitative analysis. As we described above. δ > 0. 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 Ct + It + qt Bt+1 ≤ Yt + Bt . 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. Finally. The capital accumulation process displays both costly utilization of capital and adjustment costs: Kt+1 = [1 − δ(Zt )] + φ(It /Kt )Kt . the variance of the innovations is set so that the variance of output in the model exactly matches the volatility . where capital services are the product of the capital stock.26). φ > 0. Output in the domestic country is produced using capital services St and labor Lt as Yt = At St1−α (Xt Lt )α . where δ > 0.
long averages of saving rates and investment rates tend to be correlated for the OECD countries (see Feldstein and Horioka [89]). When a positive productivity shock occurs in one country. for example. Even with a modest crosscountry correlation of the innovations. McCallum [157] or Helliwell [118]. we described the puzzles that arose when trying to match the international RBC model to the data. ﬁrst. which is not. in a world of integrated capital markets. and exports. variable capital utilization takes the place of investment ﬂows across countries. though they also relate their discussion to the international RBC literature. In our earlier discussion. for example.4.90 Business Cycles: Fact. which in turn increase employment. Speciﬁcally. this is accompanied by increases in investment and employment in the other country. Obstfeld and Rogoff [168] provide a broader view of the empirical puzzles in the international macroeconomics literature. capital utilization rates increase. The most important impact of variable factor utilization is on the crosscorrelations of the factor inputs. • . and also those of hours and employment. introducing variable capital utilization reduces the volatilities of consumption. PUZZLES REVISITED The puzzles in the international business cycle literature have been the topic of much research (see. the crosscorrelation of hours and investment become positive. the volatility of the shocks necessary to match output volatility is substantially reduced. 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. In their review. Intuitively. Fallacy and Fantasy of the US economy. 4. we would expect capital to ﬂow to regions where the rates of return are highest. The Feldstein–Horioka puzzle: According to this puzzle. See. which is desirable. One of the key ﬁndings regarding the role of variable capital utilization is that. Second. Obstfeld and Rogoff [168] argue that many of the puzzles in international macroeconomics may just be speciﬁc to the models researchers are using. capital. Yet. the survey by Baxter [33]).
the international consumption correlation puzzle tends to be speciﬁc to a model’s assumptions. In contrast to some of the other contributions that we have discussed. ﬁrst. Backus and Smith [24] show that in an economy with traded and nontraded goods. The purchasing power parity (PPP) puzzle: The PPP puzzle arises from the fact that shocks to the real exchange rate are very persistent. whereas the other four refer to the behavior of quantities. They argue that. see Tesar and Werner [203]). They also note that the pricing puzzles require such features as nominal rigidities of the type that we will consider in the next chapter. Meese and Rogoff [161] further show that most exchange rate models forecast exchange rates no better than a naive random walk. say. the literature that we have described has revealed a variety of promising directions for future research. Surprisingly. as we described above. The exchange rate disconnect puzzle: This puzzle captures the notion that there exists a relationship between exchange rates and any macroeconomic variable. perfect risk sharing across countries implies that countries which experience declines in the relative price of consumption should receive large transfers of goods. In Section 4. Obstfeld and Rogoff [168] ﬁnd that adding transport costs goes a long way towards accounting for the ﬁrst ﬁve puzzles. the international portfolio diversiﬁcation puzzle. Second. the analysis of Obstfeld and Rogoff [168] is an attempt to unify the explanations of a related but disparate set of ﬁndings. 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 crosscountry consumption correlations exceed the crosscountry output correlations in a typical international RBC model. .2. and does not have the same weight as. the last two puzzles are pricing puzzles. whereas the opposite is true in the data.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. Clearly.
This page intentionally left blank .
The RBC conclusions regarding the response of hours. output. Basu. productivity. taken issue with the notion that prices can adjust costlessly to clear markets.There also exist openeconomy versions of the New Keynesian model that have been used for policy analysis 1 For a review and discussion of these models. from the onset. 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. the response of hours to productivity shocks is negative. and Kimball [32] also present evidence that hours worked and other variables fall in response to technology improvements in the short run. Fernald. see Christiano. and other variables to a technology shock have been questioned by empirical results obtained along several different lines. 93 . Gali [96] has argued that in a suitably restricted vector autoregression (VAR) including measures of hours. This is in contrast to the RBC model. Limited participation models also allow alternative mechanisms for the propagation of real and monetary shocks. which predicts that hours rise on impact to a positive technology shock.Chapter 5 New Keynesian Models The New Keynesian approach has gained signiﬁcance in the modern macroeconomics literature. such as variable factor utilization. and other variables. output. 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. Eichenbaum. More recently. and Evans [66]. Their approach involves purging the standard Solow residual of factors that might lead to procyclicality.1 The behavior of hours and productivity has been a topic of debate. Critics of the original real business cycle (RBC) approach had. On the one hand.
Ct is a composite consumption good deﬁned as Ct = 1 0 Cit ( −1)/ /( −1) di . hours worked Nt . (1.1. Ut ) is given by H (Nt . and effort levels Ut to maximize ∞ E0 t=0 βt ln (Ct ) + λm ln Mt Pt − H (Nt . Fallacy and Fantasy and for understanding the role of monetary shocks. and > 1 is the elasticity of substitution among consumption goods. Suppose that a representative household chooses consumption Ct .3) where Cit is the quantity of good i ∈ [0. THE BASIC MODEL Consider a simple New Keynesian model with monopolistic competition. respectively. and variable labor effort due to Gali [96]. money holdings Mt .1) subject to 1 0 Pit Cit di + Mt = Wt Nt + Vt Ut + Mt−1 + ϒt + t (1. . 2.5) ϒt and t denote monetary transfers and proﬁts. Ut ) (1. price rigidities. . In this expression.2) for t = 0. Wt and Vt denote the nominal prices of an hour of work and effort. and then discuss the empirical ﬁndings in more detail.94 Business Cycles: Fact.2 We ﬁrst describe a simple New Keynesian framework that can be used to rationalize the observations. respectively. 5. see Bowman and Doyle [45]. The functional form for H (Nt . 1] consumed in period t. 1 + σn 1 + σu t (1. and Pt = 1 0 1/(1− ) 1− Pit di (1. . 1. Ut ) = λn λu Nt1+σn + U 1+σu . .. The price of good i is given by Pit .4) is the aggregate price index. 2 For a further discussion.
New Keynesian Models
95
The ﬁrstorder 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 periodbyperiod budget constraint. We can solve for µt from the ﬁrstorder 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 ﬁrstorder 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 ﬁrstorder 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 ﬁrstorder 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 marketclearing conditions, implies that Pt Yt = Mt . Using the ﬁrstorder 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 pricesetting rule in (1.22) together with (1.12) and the expression for equilibrium consumption
and hence a decline in employment will occur. However. but the sign depends on whether ϕ < (>)1. and xt depend only on the current ξt . ϕ ϕ 1 ϕ ξt + 1−γ + γ ηt ϕ (1. Hence. Finally. the price level responds oneforone to an increase in ξt . producing the same output will require less labor input.29) xt = 1 − + (1 − γ) 1 − 1 ηt−1 . given a constant money supply and predetermined prices. A positive monetary shock deﬁned by ξt > 0 has a temporary impact on output. This can be observed by noting that the levels of yt .30) can be used to describe the impact of monetary versus technology shocks. Hence. 1 1−γ ξt − ηt . though with a oneperiod lag. More interestingly. The impact of ξt on labor productivity is also transitory. employment. which corresponds to the situation of shortrun increasing returns to labor. ϕ where x = y − n is the log of labor productivity. . real balances remain unchanged in the face of a positive technology shock.28) (1.98 Business Cycles: Fact. and productivity. nt .30) (1. demand remains unchanged so that ﬁrms will be able to meet demand by producing an unchanged level of output. yt = nt = ξt + γηt + (1 − γ)ηt−1 .27) (1. we can show that pt = ξt−1 − (1 − γ)ηt−1 . In such a case.27)–(1. Fallacy and Fantasy and output derived above. with a positive shock to technology. A positive technology shock deﬁned by ηt > 0 has a permanent positive oneforone impact on output and productivity and a permanent negative impact on the price level if γ < 1. We note that measured labor productivity responds positively whenever ϕ > 1. that is. an increase in ξt causes output and employment to go up for one period and then to revert back to their initial values. when there is no accommodating response in the money supply to real shocks. This result can be best understood by considering the case of γ = 0. The conditions in (1. a positive technology shock has a negative shortrun impact on employment.
Second. This is similar to the approach in Blanchard and Quah [43] for identifying demand versus supply shocks using longrun restrictions on estimated VARs. The identifying assumption is that only technology shocks have a permanent effect on productivity. they argue that if the simulated data from a simple RBC model are subjected to a SVARbased test. 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.New Keynesian Models 99 5. 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. i. First.3 Gali [96] estimates this model using postwar US data. 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. implies that E ( t t ) = I . Chari. The orthogonality assumption. the differencestationary version of the model implies that the response 3The value of the matrix C (L) evaluated at L = 1 gives the longrun multipliers for the model. By contrast. respectively. Eichenbaum. Furthermore. the longrun impact of a given shock. . technology shocks explain only a small fraction of employment and output ﬂuctuations. Kehoe. Christiano. together with a normalization.2.. and Vigfusson [68] suggest that the standard RBC results hold if per capita hours are measured in loglevels as opposed to differences. he ﬁnds that alternative measures of labor input decline in response to a positive technology shock while GDP adjusts only gradually to its longrun level. where z and m denote the sequences of technology and nontechnology t t shocks. Surprisingly.e. The SVAR model interprets the behavior of (log) hours nt and (log) productivity xt in terms of two types of exogenous disturbances — technology and nontechnology 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 . which can be expressed as the restriction C 12 (1) = 0. and McGrattan [63] have challenged the ﬁndings derived from socalled SVARs.
Their ﬁndings corroborate the ﬁndings from the SVAR approach regarding the negative response of hours to technology improvements in the short run. They use both standard regression analysis and simple bivariate VARs for this purpose. Fallacy and Fantasy of hours to a shock to productivity is negative as in Gali and Rabanal [98]. nonresidential investment. They ﬁnd that purging the standard Solow residual of these effects eliminates the phenomenon of “procyclical productivity”. [32] present evidence that support the SVAR ﬁndings by generating a modiﬁed Solow residual that accounts for imperfect competition. 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. . and various prices and interest rates to changes in the puriﬁed Solow residual. the evidence obtained from this approach is robust to longrun identifying assumptions or to the inclusion of new variables in the estimated dynamic system. nondurables and services. they advance price rigidity as the major reason for these deviations from the RBC predictions in the short run. leads to an erroneous conclusion of the negative response of hours worked to a productivity shock. and sectoral reallocation and aggregation effects. Basu et al. variable factor utilization. hours worked. 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. Following Gali [96] and Gali and Rabanal [98]. However. 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. employment. 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. generated substantial controversy and. These ﬁndings have. if hours worked is considered to be stationary in levels. on the other.100 Business Cycles: Fact. Unlike the SVAR approach. They also examine the response of a key set of variables such as output. utilization. This misspeciﬁcation. in their view. durables and residential investment. on the one hand. nonconstant returns to scale. They also uncover further evidence for the negative response of nonresidential investment to such shocks.
On the other hand. On the one hand. More recently. namely. For instance. The question then arises whether a small open economytype RBC model can account for both developed and emerging market business cycles. Emerging market business cycles also display a different set of stylized facts relative to developed economies. consumption varies more than output. and Rebelo [77]. Earlier RBC models for small open economies were developed by Mendoza [162] and Correia. Business cycles in emerging markets exhibit different characteristics compared to those in developed economies.Chapter 6 Business Cycles in Emerging Market Economies In recent years. Argentina. Aguiar and Gopinath 101 . but their analysis is based on the onesector optimal growth model. for example. and income and exports are typically highly volatile. for example. 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. Arellano and Mendoza [17]). the real business cycle (RBC) agenda has been increasingly applied in emerging market contexts. they have been examining the efﬁcacy of RBCtype models in accounting for these facts. Kydland and Zaragaza [144] also study the behavior of an emerging market economy. Neves. the trade balance is strongly countercyclical. Ratfai and Benczur [175. 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. researchers have begun generating business cycle facts for such economies (see. 176]).
and µg denotes average longrun productivity growth.i. The shock zt represents the transitory component of productivity.3) ρg  < 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. with E ( z ) = 0 and Var( z ) = σz . gt denotes the shocks to the growth rate of productivity. g t) ln (gt ) = (1 − ρg ) ln (µg ) + ρg ln (gt−1 ) + g ∞ t }t=0 g t) (1.2) 2 where { z }∞ is distributed as i. In their view. A SMALL OPENECONOMY MODEL OF EMERGING MARKET BUSINESS CYCLES Aguiar and Gopinath [1] present a small openeconomy model of business cycles that allows for permanent and transitory changes to productivity. and evolves as a stationary AR(1) process: zt = ρz zt−1 + z t. t t=0 t t The permanent shock to productivity evolves as t t = gt t−1 = s=0 gs . and Uribe [101] have examined versions of a small openeconomy business cycle model with permanent and transitory shocks to account for emerging versus developed economy experiences. Pancrazi.102 Business Cycles: Fact.1. 6. (1. Thus. here modeled as changes in trend productivity growth. (1.d. ρz  < 1. t−1 .1) where {zt } and { t } represent two alternative productivity processes. (1. g t.i. Fallacy and Fantasy [1] and GarciaCicco. By contrast.d. a stationarityinducing transformation is used to remove the stochastic trend from all variables as xt xt = ˆ .4) 2 where { is distributed as i. Since productivity and output depend on the cumulation of the shocks gt . with E ( = 0 and Var( = σg . To describe the model. let output be produced according to the Cobb– Douglas production technology as Yt = exp (zt )Kt1−α ( t Lt )α . 0 < α < 1.
1 where the last line is obtained by substituting recursively and making use of an iterated expectations argument. .}}. .} 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 . the elasticity of labor supply is given by 1/(ν − 1). The GHH preferences are deﬁned as ut = (Ct − τ t−1 Ltν )1−σ . τ > 0. The economywide resource constraint is given by Ct + Kt+1 = Yt + (1 − δ)Kt + φ 2 Kt+1 − µg Kt 2 Kt − Bt + qt Bt+1 . Thus. . and that the country can borrow using oneperiod debt that sells for the price .Business Cycles in Emerging Market Economies 103 Aguiar and Gopinath [1] consider two types of preferences over consumption and leisure. 1−σ γ 0 < γ < 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. 1−σ ν > 1. (1. The ﬁrst is from the class of socalled GHH (after Greenwood. and the intertemporal elasticity of substitution is given by 1/σ. . (1. In the Cobb–Douglas case.7) This shows that investment is subject to quadratic costs of adjustment. and Huffman [105]) and the second are standard Cobb–Douglas preferences. expected discounted utility remains bounded provided βE (gt ) = βµg < 1.6) 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 . σ ≥ 0. . . Hercowitz. preferences are given by ut = (Ct (1 − Lt )1−γ )1−σ .5) Here.
f (β. Lt ) is deﬁned by (1. 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. ∂V . Furthermore.104 Business Cycles: Fact. the value function deﬁned in terms of the transformed variables is given by ˆ ˆ V (Kt . Lt ) + uc (Ct .9) (1. gt+1 )}. Lt ) ˆ ∂Yt = 0.8) where r ∗ is the world interest rate. Lt ) gt + φ gt ˆ Kt+1 − µg gt ˆ Kt ∂V ˆ ∂Bt+1 = f (β. Bt+1 . ˆ ∂Kt+1 (1.10) (1. gt )Et = 0.Lt . and Lt are ˆ uc (Ct . b represents the steadystate level of debt. Bt+1 . gt )Et ˆ ˆ uL (Ct .11) ˆ uc (Ct . when choosing how much debt to take on. 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 . Fallacy and Fantasy qt . ˆ where u(Ct . Lt ) + f (β.Kt+1 . the ﬁrstorder conditions ˆ ˆ with respect to Kt+1 . ∂Lt . Bt . Likewise. Assuming it exists. gt ) = ˆ ˆ ˆ Ct . gt ) is given by γ(1−σ) βgt1−σ in the former case and by βgt in the latter. 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 .Bt+1 max ˆ ˆ ˆ {u(Ct . qt t (1. Lt )gt qt + f (β. and ψ > 0 governs the elasticity of the interest rate to changes in indebtedness. ˆ Substituting for Ct using the resource constraint. gt )Et V (Kt+1 . The model expressed in terms of the transformed or “hatted” variables is solved by using standard recursive methods. zt .5) in the case of GHH preferences and by (1. zt+1 .6) in the case of Cobb–Douglas preferences.
4). are estimated using a generalized method of moments (GMM) approach that we describe in detail in the following chapter. zt .Business Cycles in Emerging Market Economies 105 The envelope conditions are given by ˆ ˆ ˆ ∂Yt ∂V (Kt . zt . Lt ). gtσ exp (zt )αgtα ˆ Kt Lt 1−α (1.14) subject to the production function (1. gt ) ˆ = −uc (Ct . consider only the case with GHH preferences. The remainder of the parameters. and the equation describing the real interest rate (1. a subset of the parameters is set at values determined a priori. ˆ ˆ ∂V (Kt .7).13) .12) (1. gt ) ˆ = uc (Ct . The solution for the model is obtained by implementing a loglinear approximation to the equilibrium conditions described above. the laws of motion for the shocks (1. . including the parameters of the shock processes. 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 .8).2)–(1. the resource constraint (1. 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 . 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 .1). In the recent applications. (1. Bt . Bt . ˆ ∂ Bt For simplicity.
5 or 5. This leads to a large trade deﬁcit which tends to persist for a considerable period of time (16 quarters). as well as the correlations of the latter three with output.4 for Mexico. consumption.25 or 0. Fallacy and Fantasy 6. They also consider the mean and standard deviation of (unﬁltered) income growth as well as the autocorrelations of (ﬁltered) income and (unﬁltered) income growth. investment. DO SHOCKS TO TREND PRODUCTIVITY EXPLAIN BUSINESS CYCLES IN EMERGING MARKET ECONOMIES? Aguiar and Gopinath [1] consider two versions of their model. ρg . The moments that they use for estimation include the standard deviations of log (ﬁltered) income. These ﬁndings have been challenged by GarciaCicco et al. σg . φ). net exportstoGDP. 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.2. ρz . the autocorrelations of transitory shocks are roughly similar. The parameters to be estimated are given by (µg . In particular. the shock to trend productivity can generate consumption booms that tend to appear sidebyside with current account deﬁcits in emerging market economies.41 for Canada depending on the speciﬁcation for preferences. 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. σg /σz . The most noteworthy ﬁnding that emerges from the estimation is that the ratio of shock standard deviations. Since the number of parameters is less than the number of moment conditions.106 Business Cycles: Fact. 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. and 2. who argue that the results of Aguiar and Gopinath [1] are due to their use of . an emerging market economy version estimated using quarterly data for Mexico and a developed country version estimated for Canada between 1980 and 2003. as is the capital adjustment parameter φ. [101]. By contrast. This ratio captures the importance of shocks to trend productivity. there are also overidentifying conditions which can be used to test the model’s implications. Thus. Furthermore. is 0. This yields 11 moment conditions. they show that a positive transitory shock to productivity reduces consumption in anticipation of lower income in the future. σz .
and secondorder autocorrelations of output growth. [101] argue that a similar ﬁnding is not true for an emerging market economy such as Argentina. In particular. whereas in the data these quantities are roughly equal. the permanent shock is estimated to be more volatile and persistent than the transitory shock. They consider a model that is very similar to the one in Aguiar and Gopinath [1]. and less volatile than output growth if the opposite is true. However. this choice of sample period does not seem out of line. and the trade balancetooutput ratio. unlike Aguiar and Gopinath [1]. [101] ﬁnd that the overidentifying restrictions of the model are rejected. they show that output ﬂuctuations in the postWorld War II period are as large as those in the preWorld War II period. GarciaCicco et al. and. and ﬁnd that the consumptionsmoothing motive in response to a transitory shock dominates the anticipatory effect of consumption to a permanent productivity shock. As we described above. Second. which is not signiﬁcantly different from zero. in terms of the parameter estimates. The standard deviations of the shocks and the autoregressive parameter for the permanent shock are estimated precisely. By contrast. They determine this empirically. First. consumption growth. Third. Whereas the ﬁrst result suggests that the estimated model does not emphasize the role of permanent shocks sufﬁciently. the correlations of output growth with consumption growth. As we described earlier.Business Cycles in Emerging Market Economies 107 a short sample to estimate lowfrequency movements in productivity. GarciaCicco et al. and the trade balancetooutput ratio. and estimate the parameters of the stochastic processes for the permanent and transitory shocks using 16 moment conditions. investment growth is found . this is not the case for the autoregressive coefﬁcient for the transitory shock. they consider the variances and ﬁrst. and the unconditional mean of output growth. In particular. in contrast to what is observed in the data. consumption growth in the estimated model is calculated to be less volatile than output growth. [101] use annual data for Argentina over the period 1913–2005. investment growth. the second result suggests that it overemphasizes them. consumption growth will be more volatile than output growth if permanent shocks are more important than transitory shocks. GarciaCicco et al. investment growth. much of the business cycle literature for developed countries has concentrated on the postWorld War II period. The authors also ﬁnd that the trade balancetooutput ratio is estimated to be around four times as volatile as output growth. Given the evidence that business cycles have moderated during this period.
108 Business Cycles: Fact. their ﬁndings cannot be used to disentangle which of these factors is responsible for the failure of the model to replicate the observations. The authors show that the estimated model also fails to replicate the autocorrelations of output growth and the trade balancetooutput ratio. with a constant world interest rate. suggesting in this case that neither source of shocks is sufﬁciently volatile. regardless of whether shocks to productivity are permanent or transitory. Fallacy and Fantasy to be insufﬁciently volatile. consumption increases in response to an innovation to output in roughly the same magnitude as output. the trade balance is unaffected and the trade balancetooutput ratio inherits the behavior of output. If the shocks are permanent. The authors also show that the random walk behavior of the trade balancetooutput ratio remains. . the trade balancetooutput ratio displays a near random walk behavior even though the estimated autocorrelation function in the data is downwardsloping. if the shocks are transitory. consumption in the model follows a near random walk even though output and investment become stationary variables. they argue that some of their ﬁndings that we described above point to the importance of shocks that are different from productivity shocks. In particular. GarciaCicco et al. 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. In particular. the behavior of the trade balancetooutput ratio again follows a near random walk because in this case. By contrast. Hence. as households perceive the income increase to be permanent. However. [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.
The recent business cycle literature has witnessed the use of a variety of techniques regarding the empirics of business cycles. In this chapter. In her original contribution. which assesses the adequacy of the model based on the behavior of the relative variability and comovement of 109 . 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] extended this approach to derive measures of ﬁt for an underlying economic model. An alternative approach was proposed by Christiano and Eichenbaum [65]. we described some criticisms leveled against the assumptions of the real business cycle (RBC) framework. We also discussed variations of the model that could be used to account for speciﬁc correlations in the data. 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. 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. Amongst the most prominent of these assumptions is the assumption of perfect price ﬂexibility. we will study alternative approaches for matching the implications of a theoretical model with the data. One of the most popular approaches has been based on the dynamic factor model. and also provide an overview of the estimation versus calibration debate. Their approach may be viewed as an extension of the standard RBC approach. This model was initially proposed by Sargent and Sims [185] for describing cyclical phenomena. 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.Chapter 7 Matching the Model to the Data In the previous chapters.
t−s ) = 0 for all ν ˜ i. Finally.3) Both the common factors and the idiosyncratic factors may be serially correlated. . More precisely. E (f˜t f˜t−s ) = 0 for t = s and E (˜ i.t ) = 0 ν ˜ for i = 1. . t = s. that is. 7. let {wt }∞ denote an ndimensional mean zero.t ) = 0 E (˜ i. (1. To describe how to formulate unobservable index models. the variances and autocovariances of the observed series {wt } can be decomposed in terms of the variances and ˜ autocovariances of a lowdimensional set of unobserved common factors and . According to this model. 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 lowdimensional vector of unobservable factors and a set of idiosyncratic shocks that are mutually uncorrelated and uncorrelated with the factors. and νt is an n × 1 vector of idiosyncratic shocks ˜ that are mutually uncorrelated and uncorrelated with common factors. n for i = j. (1.t νj. . Bayesian estimation of the socalled 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. j. . 57] discusses alternative approaches for conducting statistical inference in calibrated models. Under these assumptions. covariance stationary stochastic ˜ t=0 process used to describe observations on the (possibly detrended) values of a set of variables.1) ˜ ˜ where {H (s)}∞ s=−∞ is a sequence of (n × k)dimensional matrices.t νi.2) (1. we require that ˜ E (f˜t νi. Canova [56. 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. A kfactor unobservable index model for wt is given by ˜ ∞ wt = ˜ s=−∞ ˜ ˜ H (s)f˜t−s + νt . ft is a k × 1 vector of common factors.1.110 Business Cycles: Fact.
∞ 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). the diagonal elements of Sw (ω) are real. . Under the assumptions ˜ underlying (1. complex numbers. ω ≤ π. notice that offdiagonal elements of Sw (ω) are. n. in general. ˜ ˜ r = . since Rxl xh (r) = Rxl xh (−r). Substituting for Rw (r) into (1. . Let Rw (r) = E (wt (wt+r ) ). . . be the autocovariance function of {wt }.Matching the Model to the Data 111 the idiosyncratic shocks. .1 (1.1). = s=−∞ An alternative representation of the autocovariance function is provided by the spectral density function ∞ Sw (ω) = r=∞ Rw (r)e −irω . . .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 . h = 1. . . 0. However. 1This function is welldeﬁned as long as .4) Assuming that Sw (ω) is nonsingular at each frequency ω. −1. . 1.
which describe how the common factors affect the behavior of the elements of wt at all leads and lags. this model must be estimated jointly across . He discusses a classical measurement error case. ˜ ˜ where H (ω) denotes the Fourier transform of H (s). error terms. measurement errors or idiosyncratic components not captured by the underlying RBC model.i. as in Altug [5]. The common factor is identiﬁed as the innovation to the technology shock. as well as a case with orthogonal prediction errors. Unlike the unrestricted factor model which can be estimated frequency by frequency.d. 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. 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). Also. The unrestricted version of the dynamic factor model does not place restrictions on ˜ the matrices H (s).112 ∞ Business Cycles: Fact. As these authors observe. Hence. 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. Sargent [184] also employs the device of augmenting a singular model with additional idiosyncratic shocks. and the idiosyncratic shocks are interpreted as i. The dynamic factor model can be estimated and its restrictions tested across alternative frequencies using a frequency domain approach to time series analysis. The use of the dynamic factor model in business cycle analysis dates back to the work of Sargent and Sims [185]. The restricted factor model makes use of the crossequation restrictions across the linear decision rules implied by the original model.i. Another wellknown application of this approach is due to Altug [5].d. the dynamic factor model provides decomposition at each frequency that is analogous to the decomposition of variance in the conventional factor model. Altug also uses this representation to estimate the model using maximum likelihood (ML) estimation in the frequency domain. it is not possible to identify ˜ the common factors with different types of shocks to the economy.
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. In the data. model.1. Watson’s analysis does not depend on assuming that the unobserved factors and the idiosyncratic shocks are uncorrelated. The ACGF for yt is similarly denoted Ay (z). Also. let xt denote an n × 1 vector of covariance stationary random variables. Deﬁne the autocovariance generating function (ACGF) for xt by ∞ Ax (z) = r=−∞ E (xt xt−r )z r . the joint process for the data and the error is introduced to motivate goodnessofﬁt measures. and aggregate output. the model cannot explain the cyclical variation of the observed variables. when the restrictions of the underlying model are imposed. .1. the vector of variables yt corresponds to the empirical counterpart of xt .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. see Hansen and {xt }∞ t=0 Sargent [114]. not to describe a statistical model that can be used to conduct statistical tests. investment in equipment. To describe Watson’s approach. and Ax summarizes the unconditional secondmoment properties of this model. The question at hand is whether the data generated by the model are able to reproduce the behavior of the observed series. deﬁne the n × 1 vector of 2 For further discussion of maximum likelihood estimation in the frequency domain. denotes the stochastic process generated by some underlying Here. 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. For this purpose. investment in structures. 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. 7. Unlike the approach adopted by Altug and Sargent.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. Instead.
5 In the ﬁrst case. Fallacy and Fantasy errors ut that are required to reconcile the ACGF implied by the model with that of the data. it is the size of the sampling error that is used to judge whether . or σxy  ≤ σx σy : 2 2 2 min σu = σy + σx − 2σxy s. Then. σxy (1. notice that Ay (z) can be determined from the data. which implies that the ACGF for ut can be expressed as Au (z) = Ay (z) + Ax (z) − Axy (z) + Ayx (z). the assumption regarding Axy (z) is that it is zero. 4This is similar to the interpretation provided by Altug [5].3 To continue. To illustrate this approach. yt . and ut are assumed to be serially uncorrelated scalar random variables. a lower bound may be deduced for the variance of ut without imposing any restrictions on Axy (z). since Axy (z) is unknown. 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). in Watson’s framework. Hence.t. If. the assumption of pure measurement error which is uncorrelated with the true variable is not appropriate.6) (1. Axy (z) = 0. σxy  ≤ σx σy . 5 We omit the third case that Watson considers because it requires results for the Cramer representation of stationary time series. let us consider two cases. To calculate Au (z).114 Business Cycles: Fact.5) where Axy (z) is the joint ACGF for xt and yt . This implies a version of a classical errorsinvariables approach. (1. some additional assumptions are needed to operationalize this approach.7) ˆ a given model ﬁts that data. Nevertheless. Ay (z) = Ax (z). 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. 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. 3 In a standard goodnessofﬁt approach. xt .4 However. the difference between Ay (z) and A error. let Ay (z) denote the population autocovariance function and Ay (z) ˆy (z) is ascribed to sampling denote its estimated counterpart. deﬁne ut by the relation ut = yt − xt . and the size of the sampling error can be determined from the datagenerating process for yt . and Ax (z) from the model. In the standard dynamic factor model. in addition. the error term ut is the approximation error in describing the observed data with the underlying economic model. In this case. However.
10) where = Cx UVCy−1 . given the properties of the U and V matrices. where ij. we cannot minimize the variance of ut directly.u denotes the i. Furthermore. yt ) is positive semideﬁnite. the joint covariance matrix of (xt . 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. where W is an n × n matrix. as in the scalar case. we consider a transformation that allows us to determine the size of ut . xt can be represented as xt = yt . 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 . An alternative approach is to minimize the weighted sum of the variances as tr(W u ). he considers the model in King. An implication of this result is that. Proposition 1 in Watson [209] demonstrates that the unique matrix xy .u . 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 ). tr( u ) = n i=1 ij. The problem in this case is to choose xy to minimize tr(W u ) subject to the constraint that the covariance matrix of (xt . 132]. For k = 1. and S is a k × k diagonal matrix. As a 2 consequence. (1. is given by xy = Cx VUCy . (1.Matching the Model to the Data 115 It is easy to see that the solution for this problem is to set σxy = σx σy . Watson [209] uses this methodology to generate goodnessofﬁt measures for a standard RBC model with a stochastic trend in technology. this result corresponds to the one in the scalar case. Plosser. yt ) is singular. In this case. Now suppose that xt and yt are serially uncorrelated random vectors with covariance matrices x and y .9) In this expression. One convenient transformation is the trace of u . and Rebelo [131. xt and yt are perfectly correlated so that xt = γyt . As a consequence. (1. j element of . Instead. V is a k × k orthonormal matrix (with VV = Ik ). The covariance matrix of ut is given by u = y + x − xy + yx . which minimizes the weighted sum of the variances of the elements of ut subject to the constraint that this matrix is positive semideﬁnite.8) where γ = σx /σy . Speciﬁcally. He loglinearizes . σu = (σy − σx )2 at the minimum.
see also Altug [4]. They also describe how to derive impulse response functions for time series models which have reduced rank. Other Applications Forni and Reichlin [93] use the dynamic factor model to describe business cycle dynamics for large crosssections. and Sala [103] show how more general classes of equilibrium business cycle models can be cast in terms of the dynamic factor representation. and the error in reconciling the model with the data. consumption. 6 A comparison of the spectra implied by the model and those generated by actual data also ﬁgured in early versions of Altug [5]. 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. Fallacy and Fantasy the ﬁrstorder conditions to obtain the solution for logdifferences of output. models for which the number of exogenous shocks is less than the number of series. that is. suggesting that the stochastic trend properties of the model do not hold in the data. 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. the economywide shocks can be identiﬁed using structural vector autoregression (SVAR) techniques.2. Giannone. and investment occur at frequencies corresponding to the business cycle periodicities of 6–32 quarters. the variance explained by the model. They examine the behavior of fourdigit industrial output and productivity for the US economy for the period 1958–1986 and ﬁnd evidence in favor of at least two economywide shocks. both having a longrun effect on sectoral output. they show that the number of common factors can be determined using the method of principal components. consumption. . and the number of hours worked. However. 7.6 He ﬁnds that the biggest differences between the spectra for output. Based on a law of large numbers argument.116 Business Cycles: Fact. His approach allows for a decomposition by frequency of the variance in each observed series. and the unobserved factor model can be estimated by using equationbyequation ordinary least squares (OLS). investment.1. their results also indicate that sectorspeciﬁc shocks are needed to explain the variance of the series. Reichlin.
δ. σλ ) . the elements of restrictions E H1t ( 1) satisfy the unconditional moment = 0. θ. 1 = (δ. θ. 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. the depreciation rate δ is set to reproduce the average depreciation on capital as E δ− 1− kt+1 it − kt kt = 0.Matching the Model to the Data 117 7. γ.2. the parameters in 1 are estimated using simple ﬁrstmoment restrictions implied by the model. To describe how their approach is implemented. given data on gross investment it and the capital stock kt+1 . For example. (2. ρ. Christiano and Eichenbaum [65] consider the hoursproductivity 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 estimation strategy is based on a subset of the ﬁrst and second moments implied by their model. Their approach may be viewed as an extension of the standard RBC approach. σµ . λ.11) . Likewise. and the exogenous stochastic processes. which assesses the adequacy of the model based on the behavior of the relative variability and comovement of a small set of time series. and the share of capital in the neoclassical ¯ production function. let 1 denote a vector of parameters determining preferences. g . the share of capital satisﬁes the Euler equation E β−1 θ yt+1 kt+1 +1−δ 1 ct ct+1 = 0. the deterministic steady states are derived for the transformed variables. technology. Some of the parameters included in 1 may be the depreciation rate of capital. As in the standard RBC approach. Since there is a stochastic trend in this economy arising from the nature of the technology shock process. GMM ESTIMATION APPROACHES Other papers have employed nonlinear estimation and inference techniques to match equilibrium business models with the data. More generally. Proceeding in this way.
let f( 1) = [f1 ( 1 ). it . they use a Waldtype test based on the orthogonality conditions implied by the relevant unconditional moments.12) (2. n) and σn /σy/n . F ( ) = 0. it . Also. corr(y/n. where ct denotes private consumption. n) − (y/n)t nt = 0.19) Under the null hypothesis that the model is correctly speciﬁed. σn /σy/n . let A be a 2 × k matrix of zeros and ones such that A and F( ) = f ( 1) − A = [corr(y/n. σn /σy/n ] (2. f2 ( 1 )] (2. (2. n).17) represent the model’s restrictions for corr(y/n. nt . 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. 2 x = ct . (2.20) . (2. labor hours. gt .13) (2.14) (2.18) .16) Christiano and Eichenbaum [65] are interested in testing restrictions for the correlation between hours and productivity. and the relative variability of hours versus average productivity. The restrictions of the model can be summarized as E H2t ( 2) = 0. n). gt . 2 σn /(σn /σy/n ) corr(y/n. 2 ] as the k × 1 vector containing the true values of the parameters and second moments for the model. Fallacy and Fantasy 2 consist of the standard RBC secondmoment restrictions as The elements of E yt2 (σx /σy )2 − xt2 = 0. and (y/n)t .15) E (y/n)2 σn /σy/n t E 2 − nt = 0. public consumption. Deﬁne = [ 1 . (2. For any parameter vector 1 . the average productivity of labor. private investment. To do this. 2 2 E nt − σn = 0.118 Business Cycles: Fact.
. in fact. constituted “wisdom or whimsy”. As we described earlier. investment.3. Letting test statistic for the secondmoment restrictions is based on the distribution of F ( ˆ T ) under the null hypothesis. there have been ongoing discussions on both sides of the issue. the authors show that the statistic J = F ( ˆ T ) Var[F ( ˆ T )]−1 F ( ˆ T ) (2.Matching the Model to the Data 119 In practice. 7. the containing T observations. government expenditures. Since then. Using this distribution. For example. In his words: Indeed. Using data on private consumption. 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. the GMM approach has been used in other recent applications. In a highly suggestive analysis. calibrated or otherwise. Eichenbaum [82] asked whether RBC analysis. Even if 7 See Kydland and Prescott [142. Aguiar and Gopinath [1] and GarciaCicco. 143] for a further discussion and defense of their approach. Christiano and Eichenbaum [65] estimate the parameters of the model using GMM and examine the various unconditional second moments implied by the model.21) is asymptotically distributed as a χ2 random variable with two degrees of freedom. and Uribe [101] employ this approach in their analysis of business cycles in emerging markets. Pancrazi. there is sampling error in estimating from a ﬁnite data set ˆ T denote the estimated value of . aggregate hours. and average productivity. Kydland and Prescott [141] initiated this debate in the modern business cycle literature.7 In this section. once we quantify the uncertainty in model predictions arising from uncertainty about model parameter values. our view of what the data is [sic] telling us is affected in a ﬁrstorder way. we will deal with a variety of criticisms aimed directly at the calibration approach and some suggested alternatives to it. 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.
they use the twoshock version of the RBC model proposed by Christiano and Eichenbaum [65]. Eichenbaum [82] also presented evidence to show that the performance of the standard model as well as modiﬁcations that allow for labor hoarding. To estimate impulse response functions from the data.1. 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. Such a break accounts for the slowdown in productivity growth in the US that occurred in the late 1960s. but the data contain almost no evidence against either the view that the answer is really 5% or that the answer is really 200%. and impulse response functions in response to shocks. The answer could be 70% as Kydland and Prescott [142] claim. its power spectrum. there is simply an enormous amount of uncertainty of just what percent of aggregate ﬂuctuations they actually do account for. they consider the autocorrelation function for output. They assume that the technology shocks are differencestationary. and identify the technology shock under the assumption that it has a permanent effect on output. the authors use the SVAR technique developed by Blanchard and Quah [43].120 Business Cycles: Fact. What the data are actually telling us is that. Since the single technology shock model is singular. for example. The striking results in Cogley and Nason [70] show that the autocorrelation function .3. 7. As part of their diagnostics. 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. The autocorrelation function from the model is generated by simulating the model 1000 times. whereas government shocks evolve as a persistent AR(1) process. which includes shocks to productivity and government consumption as well as the indivisible labor feature of Hansen [112] and Rogerson [179]. They consider a twovariable VAR with output and hours worked (or consumption). while technology shocks almost certainly play some role in generating the business cycle. deteriorate considerably when a break in the sample is allowed for. They compare the autocorrelation function and the impulse response function in the data with those generated by the model using generalized Q statistics.
In particular. let Xt = f (Zt . f ) = H (Xt . capital accumulation. I ) denote the joint density for the data and the parameters. 57] proposes an alternative approach for providing statistical inference in calibrated models. Cogley and Nason [70] also argue that the model displays weak propagation mechanisms which arise from intertemporal substitution. the data display a humpshaped response to transitory shocks whereas the model implies a much smaller monotonic decay. 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. Also. conditional on the function f and the parameters β. as is the power spectrum.Matching the Model to the Data 121 for output implied by the model is nearly ﬂat. let Xt denote a vector stochastic process with a known distribution that describes the evolution of a set of observed variables. and let H (Xt . say. βf .3. the model has some success in matching the estimated response functions in the data for the permanent component of GDP. and the spectrum for output displays signiﬁcant power at the business cycle frequencies of 2.33–7 years per cycle. conditional on the information set available to the researcher I and the function f . in some sense. the autocorrelations of output at lags 1 and 2 are signiﬁcant and positive. In terms of the impulse response functions. Let G(Xt f . investment. Deﬁne expectations of the functions of the simulated data 8These results are. Calibration as Estimation Canova [56. GDP. and adjustment costs. but it is unable to match the impulse response function for the transitory component. . let π(βI . Denote the predictive density p(Xt I . To describe his approach. and interest rates. The test statistic also shows that the implications of the model are rejected at signiﬁcance levels of 1% or less. complementary to those regarding the role of ﬁltering on the cyclical properties generated from a standard RBC model. See Cogley and Nason [69]. βI . f )d β.2. 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. f ) denote the density for the parameters β.8 By contrast. 7. β) denote the density of the vector Xt .
• Draw vectors β from π(βI . In general. f ) and zt from k(Zt ). generate {xt }T and compute µ(xt ) using t=1 the model xt = f (zt . Simulation exercises conducted after a subset of the parameters has been estimated (using GMM. For example. f ) by E (µ(Xt )f . β) or its approximation. I ) = = µ(Xt )p(Xt I . f )dXt H (Xt . βI . one could count estimates of elements of β obtained from alternative studies and smooth the resulting histogram to obtain the density π(βI . Canova [56] describes how to account for such an approximation error when computing the moments of the simulated data. This information may be derived from alternative data sets. for . (3. or estimation techniques. Then. 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. • For each drawing of β and zt . The calibration approach involves putting a point mass on a particular value of β.122 Business Cycles: Fact. the model can be simulated repeatedly for different values of β and Zt to compute sample paths for Xt . • Repeat the above two steps N times. f )d βdXt . • Construct the frequency distribution of µ(xt ) and other statistics of interest that can be used to evaluate the performance of the model. If such information is hard to obtain. More precisely. model speciﬁcations.22) This approach allows for a probability statement regarding the behavior of various moments implied by the model. f ) for the unknown parameters. given the information I available to the researcher and a density k(Zt ) for the exogenous processes. then one can use a uniform distribution. deﬁne µ(Xt ) = 1 if ν(Xt ) ∈ D and zero otherwise. f ). Notice also that the function f is typically unknown and must be obtained using some approximation method. While the densities H and p are unknown. where ν(Xt ) is some moment of interest and D is a bounded set. suppose that we are interested in evaluating Pr(ν(Xt ) ∈ D). the approach that Canova advocates for conducting statistical inference in calibrated models is as follows: • Select a density π(βI . Fallacy and Fantasy (denoted by µ(Xt )) under the predictive distribution p(Xt I .
. markups. Alternatively. In a novel analysis. and Patterson [9] examine the implied behavior of output. Such nonlinearities may take the form of conditional heteroscedasticity such as ARCH or GARCH effects. and while they could not ﬁnd evidence of chaotic behavior. They conclude that the nonlinearity must lie in the transmission mechanism. GMM. the factor inputs. Using an array of diagnostic tests. either in the form of nonlinear stochastic dynamics or deterministic chaos. 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. Ashley. Brock and Sayers [47] tested real macroeconomic variables displayed by deterministic chaotic dynamics.Matching the Model to the Data 123 example) involve putting a point mass on β∗ estimated according to a particular method.3. we discussed the role of alternative factors that could give rise to endogenous changes in productivity such as labor hoarding. Nonlinearity in Macroeconomic Time Series Another criticism of the RBC model literature is that it has typically been concerned with examining the ﬁrst. Neftci [166] was among the ﬁrst to demonstrate that unemployment ﬂuctuations were asymmetrical along the business cycle. They found strong evidence of nonlinearity in industrial production. 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. they nevertheless showed that postwar employment. there may exist asymmetries in various economic variables. Altug. and timevarying markups.3. increasing returns. they do not ﬁnd evidence of nonlinearity in the Solow residuals measured after allowing for increasing returns to scale. they note that observed measures of output and factor inputs such as the capital stock and hours worked display marked nonlinear behavior.and secondmoment properties of aggregate economic variables for the purpose of matching a model to the data. Speciﬁcally. unemployment. and industrial production could be described as nonlinear stochastic processes. In our discussion in Chapter 3. say. and argued that any reasonable macroeconomic model should display some form of nonlinear dynamics (see Potter [172] for a review). Ashley and Patterson [22] developed a test to test for deviation from linear stochastic processes. or variable capacity utilization. variable capacity utilization.
The SNP approach nests standard VARs. The approach to matching the model to the data is through the efﬁcient method of moments (EMM) developed by Gallant and Tauchen [99]. In the ﬁrst step. The last feature is achieved by taking a transformation of the normal density using Hermite polynomials (see Gallant and Tauchen [100]). and for nonnormal disturbances. The SNP approach uses a standard VAR to model the conditional mean and an ARCHGARCH structure to model the conditional variance. However.e. the economic model is simulated for a given set of parameters. and the UK.. but they are unable to reproduce nonlinearities in these time series. which is a twostep procedure. 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. and Huffman [105]) and adjustment costs in investment. skewness). 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. Italy. it also allows for periods of high volatility followed by low volatility (conditional heteroscedasticity). The ﬁrst model selected by the SNP is a VAR(3) with conditional heteroscedasticity and a Hermite . and shows that nonlinearities such as skewness. This ensures that the nonlinearities in the underlying model are not eliminated through a linearization procedure. and excess volatility (i. Valderrama [207] considers a simple Brock–Mirmantype growth model with GHH preferences (see Greenwood. and conditional heteroscedasticity are common for many of these aggregates. Japan. He argues that standard general equilibrium models are able to replicate the ﬁrst. A value iteration approach is used as the solution procedure. GARCH). kurtosis. Valderrama [207] selects three statistical models to describe the properties of the data. Hercowitz.. Mexico.124 Business Cycles: Fact. a seminonparametric (SNP) model is estimated to characterize the statistical properties of the data. and it allows for a nonGaussian error term. for conditional heteroscedasticity (ARCH. excess kurtosis). Then.e.and secondmoment properties of such variables. asymmetric business cycles (i. The statistical models are ﬂexible enough to allow for increasing time dependence in the mean of the process (captured through a VAR). The objective function is distributed as a X 2 statistic. as in the GMM approach. He poses this as a “canonical” challenge to the RBC approach. In the second step. 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.
but not the nonlinearity in consumption. it can capture the nonlinearity in investment. The second model is a VAR(3) with ARCH(1) errors. The linear VAR(3) model also implies a much lower volatility for the consumption and investment series than the other two models. 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. [9] who show that the nonlinearities in the observed series must lie in the propagation mechanism. Two other models are selected for the purpose of describing the data. the other two statistical models allow for these features and hence do not require such a high adjustment cost parameter. Combined with the ﬁndings of Altug et al. Valderrama [207] ﬁnds that the irreversibility constraint does not bind during the solution of the model. Surprisingly. Valderrama ﬁnds that the RBC model is not successful at generating the conditional variance of investment. which allows for nonlinearity in terms of the ARCH effects but continues to assume that the errors are Gaussian. 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. 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 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. By contrast. 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. and arrive at results that indicate little or no signiﬁcant impact of irreversibility or lumpiness on aggregate investment dynamics. 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. this parameter is estimated to be around 10. its estimates are around 3. whereas in the other two models. When the statistical model is forced to be a linear VAR(3). 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.
irreversibility may play a more important role when ﬁnancial constraints at the plant level and the household level are taken into account. Fallacy and Fantasy multisector growth models considered in the literature (see. one may consider the impact of another type of aggregate shock. 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. is due to their assumption of a very large variance for these idiosyncratic shocks. Gertler.126 Business Cycles: Fact. the reason why aggregate investment displays lower variability in the presence of irreversibility when ﬁrms are subject to idiosyncratic shocks. the aggregate impact of irreversible investment disappears in a general equilibrium framework. Moreover. Veracierto [208] argues that when the size of sectoral shocks is consistent with that observed in the data. and Gilchrist [41]). At the same time. These arguments suggest that nonlinearities are another . this is a puzzling ﬁnding. 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. Given the importance of nonlinearity in macroeconomic time series documented in a number of studies. 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. our analysis of the simple RBC model with ﬂexible prices and wages suggests that it cannot successfully match many features of the data. for otherwise they would not be identiﬁed. However. Both Veracierto [208] and Thomas [205] assume that labor is perfectly mobile across plants. monetary shocks. More generally. alternatively. and may be responsible for softening the impact of the inﬂexibility faced by plants in terms of their capital adjustments. 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. Thus. for example. as Valderrama [207] states. Coleman [73] or Ramey and Shapiro [174]) is due to their assumption of unrealistically large sectoral shocks.” Yet. Similarly. as in Bertola and Caballero [42]. 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 pricesetting behavior on the part of ﬁrms or. As Caballero [53] emphasizes: “An important point to note is that since only aggregate data were used. namely. these microeconomic nonlinearities must matter at the aggregate level.
we note that Altug’s [4] results did not lead to support for the model. Yet. it is unlikely that RBC models would be subjected to the types of diagnostic tests proposed by Watson [209]. If the simple model is counterfactual. [103] in a VAR framework. and argues against the estimation and comparison of a “heavily restricted linear time series model (for example. highlighting problems in ﬁnding appropriate 9 See Altug [4]. for example. one could argue that the GMM approach in Christiano and Eichenbaum [65] involves choosing a subset of moments of an equally restrictive nonlinear model.Matching the Model to the Data 127 area where the assumptions of the simple RBC model fail to hold. King claims to be on the quantitative theory side of the debate.4. may be preferable. One can also examine the claim that a less restrictive approach to estimation and model evaluation following the approach in Christiano and Eichenbaum [65]. 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. . 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. but in what sense does this make the model a better interpretation of reality? King also discusses the shortcomings of this approach. Hoover. and a potential direction for improving the model’s ﬁt.9 In the absence of the initial frequencydomain estimation based on the factor representation. 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. 7. 17). The Debate Reconsidered King [128] discusses the arguments on different sides of the debate under the heading “Quantitative Theory and Econometrics”. p. Yet.3. and Salyer [116]. an RBC model with some or all of its parameters estimated) to an unrestricted time series model. Many have argued that this was to be expected (see Hartley. The factor representation underlying Altug’s analysis has also been resuscitated by Giannone et al. For some time. 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 approach in Christiano and Eichenbaum [65] is too restrictive. In this sense. The initial criticisms of . 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. far from this occurring. 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. 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. Fallacy and Fantasy instruments and in the appropriate selection of moment conditions for model evaluation. In fact. 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. This interpretation precludes the notion that quantitative theorizing can take the place of formal econometric methods. and the New Keynesian challenge has shown that the model fails in generating the observed negative response of hours to productivity improvements. 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. The simple propagation mechanisms inherent in the model have been deemed incapable of reproducing business cycle dynamics of key variables such as output.128 Business Cycles: Fact. 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. As Geweke [102] notes. These arguments show that there is no clearcut dichotomy between these approaches. Conversely. Another criticism of the RBC approach is that it fails to capture nonlinearities in key macroeconomic time series. These developments show that the contribution of the more formal econometric techniques need not be dismissed as cursorily as is sometimes the case. Perhaps initially the RBC theorists were concerned that the estimation versus calibration debate would discredit the use of wellspeciﬁed dynamic general equilibrium models for the purpose of capturing the quantitative behavior of economic variables.
To capture the ﬁrst feature. Eichenbaum. One can ask whether this class of models overcomes some of the criticisms leveled against the original RBC approach. According to them: [This] model is stark in its assumptions. consider the model proposed by Christiano. adjustment costs in investment. There are no market frictions and distortions. and Scott [123] for policy analysis of a small open economy. Finally.Matching the Model to the Data 129 the RBC approach have led to a wide set of extensions of the original RBC approach. Above all. There is also an interestratesetting rule that deﬁnes . 7. and animal spirits. The aim of this model is to reproduce the inertial behavior of inﬂation and persistence in real quantities. As we discuss in the next chapter. ﬁrms are assumed to borrow working capital to ﬁnance their wage bill. the model incorporates wage and price contracts following the approach in Calvo [54]. and variable capital utilization. The model also incorporates a variety of “real” frictions such as habit persistence in consumption. the original RBC approach has even instigated a new generation of Keynesian models that offer features such as credit accelerators. The model assumes a representative household and a symmetric equilibrium for ﬁrms. no ﬁxed costs or discontinuities. the model recently proposed by Kapetanios. and a variety of techniques have been developed for the purpose of matching the model to the data. for example. there is no banking sector and no speciﬁc role for money and credit in the monetary transmission mechanism. In sum. especially in the short run. The recent class of models developed for this purpose has vastly more features than any simple RBC model. sunspot equilibria. For example. There are no market frictions and no locational speciﬁcity. Consider. 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.4. 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. Yet this model has a core set of 26 equations! As another example. 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. Pagan. the model contains assumptions that are almost guaranteed to be violated by the data. and Evans [67] for the analysis of the effects of monetary policy shocks on real and nominal variables.
Demers. [67] derive much of their evidence regarding the types of real rigidities on the results of calibrationtype exercises of the current generation of macroeconomic models such as habit persistence or adjustment costs. [67] assume that monetary policy shocks are drawn from a given and known distribution. with the latter result derived from the existence of statecontingent securities for consumption. For example. the assumption of a constant adjustment cost parameter would fail to hold. Christiano et al. there are shortcomings in their approach. provide some evidence on the role of price and wage rigidities. in many ways. Yet. For applications. Demers. Furthermore. However. the irreversible investment model studied by Demers [78] and others has been shown to generate a timevarying adjustment cost that varies in response to changes in objective and subjective forms of risk and uncertainty. one could also question the implications of the model should there be changes in the distribution of the exogenous processes. and Altug [79]. but homogeneous in their consumption and asset holdings. estimated models of adjustment costs have been shown to imply implausible degrees of costs of adjustment. [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 socalled method of indirect inference.130 Business Cycles: Fact. This raises 10 For a recent example of the method of indirect inference applied to a model of the EU economy. Christiano et al. an exploratory analysis of the role of nominal and real rigidities in propagating shocks. the model assumes a continuum of households which are heterogeneous in their wage and the hours that they work. see Meenagh. Fallacy and Fantasy monetary policy. 11]. see also Demers. By contrast. implying that the propagation mechanisms would vary with changes in the distribution of exogenous variables facing agents. Minford. and Demers [10.10 The authors’ analysis is. In such cases. . the impulse responses to a given monetary shock. 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. For example. 11 For a review and discussion. see Altug. 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. Christiano et al. but in a highly restrictive environment. and Wickens [160].11 Christiano et al.
and . the use of Bayesian methods and. Second. Similar to Canova [56]. in fact. the production technology. suppose that the state space representation for the model’s solution consists of the following: • a transition equation St = g (St−1 . and θ is a vector of parameters characterizing preferences. There are a variety of approaches for solving nonlinear dynamic stochastic models. which is typically a highdimensional object. the method presupposes that a solution can be found for the underlying economic model of interest. amongst others. To illustrate this approach. As in the analysis of Altug [5]. structural or invariant to interventions. vt is a vector of innovations. and information. which is obtained from the prior distribution and the likelihood function. First. Second. 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. they augment the approximate linear laws of motion with additional shocks. Judd [122] provides a textbook treatment of many currently used solution methods. Third. The recent research on DSGE models has also employed Bayesian analysis as a way of incorporating prior uncertainty about the parameters of interest (see. However. vt . McGrattan.Matching the Model to the Data 131 the issue of whether DSGE models are. and Wright [159] consider an equilibrium business cycle model with household production and distortionary taxation. where St is a vector of state variables that describe the evolution of the model over time. Rogerson. must be examined. Following the initial contribution of Altug [5]. in particular. θ). The Bayesian estimation of DSGE models involves several steps. Schorfhiede [186] and An and Schorfhiede [15]). 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. a variety of other papers have implemented maximum likelihood estimation of dynamic equilibrium models. More recent applications include Canova [60] and Ireland [121]. the likelihood function for the observations must be formed. for example. the posterior distribution for the parameters.
p(y T θ)π(θ)d θ (4. Likewise. 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. the state space representation allows us to compute the likelihood of the observations y T = (y1 . θ)dSt . where Yt are the observables and wt are the shocks to the observables (which may take the form of measurement errors. . . θ)p(St y t .132 Business Cycles: Fact. among others). we can compute p(Yt St−1 . The excellent survey by FernandezVillaverde [90] describes in detail how to implement Bayesian estimation of DSGE models. θ)}T . θ). we can also compute the probabilities p(St St−1 . θ)dSt 12This is known as the Chapman–Kolmogorov equation. . θ) T = p(S1 S0 . θ). Hence. Fallacy and Fantasy • a measurement equation Yt = h(St . . θ). the posterior distribution can be expressed as π(θy T ) = p(y T θ)π(θ) . θ) = p(y1 θ) t=2 T p(yt y t−1 . . θ). θ). this sequence is computed by making use of the relation12 p(St+1 y t . θ)p(St y t−1 . θ)dS0 t=2 p(Yt St . wt . yT ) at the parameter values θ as T p(y .23) The remaining task is to compute the sequence {p(St y t−1 . Notice that Yt = h(g (St−1 . Given the probability density functions for the shocks fv ( · ) and fw ( · ). Following his approach. which shows t=1 the conditional distribution of the states given the observations. θ) = p(St+1 St . wt . θ) and p(Yt St . Given a prior distribution π(θ) for the parameters θ. vt .
a temporary contractionary monetary shock has a temporary negative effect on output and inﬂation. their ﬁndings attribute the largest role 13 See Schorfhiede [186] or FernandezVillaverde [90]. p(yt St . Otherwise. θ) = p(yt St . even if we can evaluate the posterior distribution. Smets and Wouters [194. θ)dSt Finding the posterior distribution of the parameters displayed in (4. Third. a method known as the particle ﬁlter13 is used to evaluate the likelihood function. the GDP deﬂator. on the whole. they ﬁnd that the estimates of the structural parameters of the model are plausible and. then the Kalman ﬁlter is used to obtain the likelihood function. they ﬁnd that. a shock to the household’s discount factor. In contrast to the approach in Christiano et al. consumption. First. if the transition and measurement equations are linear and if the shocks are normally distributed. First. . employment. Second. Fourth. similar to those for the US economy. 195] provide two important examples of this approach. However. Their estimates imply more price stickiness than the estimates of Christiano et al. the estimated DSGE model performs as well as standard and Bayesian VARs. For example. however. an investmentspeciﬁc shock.23) involves a few remaining steps.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. θ)p(St y t−1 . [67]. θ) . They consider the behavior of seven key macroeconomic time series in the euro area — real GDP. θ)p(St y t−1 . Smets and Wouters [194] argue that their approach offers several important advances. a labor supply shock. These include a productivity shock. based on Bayesian model evaluation criteria. investment. [67] to estimate the parameters of a DSGE model with real and nominal frictions for the euro area using Bayesian methods. exploring it for different values of θ typically involves the use of sampling through Markov chain Monte Carlo (MCMC) methods. the real wage. 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]. and a government consumption shock plus three “costpush” shocks. and the nominal shortterm interest rate — by minimizing the posterior distribution model’s parameters based on a linearized state space representation for the model. They extend the approach in Christiano et al.
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 pricesetting 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 wellarticulated 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 policymakers 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. FernandezVillaverde and RubioRamirez [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 Calvotype 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 tstatistics, 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
one needs to consider multisector models since the onesector growth model does not allow for a consideration of these issues. it may be that shocks are concentrated in a particular sector and are propagated to the rest of the economy from that sector. In a recent analysis. I have omitted a discussion of multisector models in generating business cycle behavior. Benhabib and Farmer [38]. at times contentious. Many recent works assume that there is a search technology such that ﬁrms and workers are randomly matched. Speciﬁcally. More generally.Chapter 8 Future Areas for Research The literature on business cycles is vast. For example.1 In such cases. There are many other topics of interest that we could have discussed regarding business cycles. In this book. 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”. Nevertheless. multisector models may also have shortcomings because they typically imply that overall expansions in the economy are associated with contractions in one sector. the wage is then set by a 1 See. but always thoughtprovoking. and contrasts it with the standard real business cycle (RBC) approach. for example. 137 . even the experience of reviewing some of this literature shows that it is lively. For one. I have offered a perspective based on my own perceptions and mindsets. However. Farmer [87] provides an eloquent enunciation of this approach. Farmer [88] generates a model of selffulﬁlling equilibria which arises from market failure associated with the labor market. I have not provided a discussion of models with selffulﬁlling expectations and multiple equilibria.
This leads to a model with a continuum of steadystate equilibria. In these equilibria. For a recent application in the context of a standard RBC model. Undoubtedly. The policy implications of such a model differ from those of the standard textbook Keynesian model. which is transmitted to real variables through investors’ beliefs about the stock market. 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. new techniques. he assumes that ﬁrms offer the same wage in advance. all ﬁrms earn zero proﬁts but not all equilibria have the same welfare properties. However. 3 For a recent analysis of models with ﬁnancial frictions in a dynamic general equilibrium setting. 2This approach owes its origins to the analysis in Mortensen and Pissarides [165]. This feature of the model allows for the role of “conﬁdence”. Fallacy and Fantasy Nash bargaining process.138 Business Cycles: Fact. and Rouabah [171]. irrespective of what the new models or techniques will look like.2 In contrast to this literature.3 The above considerations show that business cycles will continue to be a topic of discussion among economists for many years to come. Farmer [88] drops the assumption that wages are determined as a result of a Nash bargaining process. . the model with a search externality emphasizes the role of conﬁdence in restoring fullemployment output. see Pierrard. This leads to a demandconstrained equilibrium. Instead. 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. see Cooley and Quadrini [76]. we may conclude that the debate surrounding their development will not cease to arouse interest or to generate future avenues for research. and new controversies regarding both. de Walque. By contrast. which seeks to alleviate demandinduced shortfalls in output through large ﬁscal stimuli. this discussion will be accompanied by the development of new models.
Aguiar. pp.. 7. forthcoming. Demers. M. and R. B. Altug. 3. S. 68. and G. and Real Business Cycles in a Small Open Economy.” International Economic Review.. 4. pp. Patterson (1999). “Gestation Lags and the Business Cycle. “Household Choices in Equilibrium. F. Cambridge: Cambridge University Press. “International Transmission of the Business Cycle in a Multisector Model. 8. S. R. pp. E. pp. 14. and C. A’Hearna.” Journal of Monetary Economics. and M. Zimmermann (2002). “The Effect of Past Experience on Female Wages and Labor Supply. 10. 273–300. S. and R.. and D. F. 46.” Canadian Journal of Economics. “Are Technology Shocks Nonlinear?” Macroeconomic Dynamics. (1989). and R. Demers (2007). pp. “Efﬁciency. 889–920. Asset Pricing for Dynamic Economies. Altug. Jermann (2000). S. pp.” Journal of Macroeconomics. Zimmermann (2004). S. Alvarez. 139 . and C. 6.” Econometrica. 775–797. pp. 982–993. Output. S. F.Bibliography 1. “Money. 2. Ambler. S.. S. Altug. 45–85. pp. Demers. Ashley. S. 3. 257–276. 5. Demers (2009). (1985). Cardia. 321–346. 11. 58. Equilibrium. and U. Labadie (2008). pp. “Political Risk and Irreversible Investment. 51. and M.” CESifo Economic Studies. 506–533. “Emerging Market Business Cycles: The Trend Is the Cycle. S.” Econometrica. S. 69–102. Miller (1990). 65. 12. “More International Evidence on the Historical Properties of Business Cycles. Altug. 13. 47. Woitek (2001).. Altug. Cardia.” Journal of Political Economy. 430–465. 115. “The Investment Tax Credit and Irreversible Investment. Altug. 543–570. pp. 53.” Unpublished manuscript. Miller (1998). Ahmed.” European Economic Review. 9. and P. Altug. E. Gopinath (2004). Altug. “International Business Cycles: What Are the Facts?” Journal of Monetary Economics. Murthy (1994). 27. Ambler. “TimetoBuild and Aggregate Fluctuations: Some New Evidence. and Asset Pricing with Risk of Default. 30. and U. pp.” Review of Economic Studies.
18. E. and D. and M.” Journal of Economic Perspectives. 30. Handbook of International Economics. 18. and E. 13. “Procyclical Productivity: Increasing Returns or Cyclical Utilization?” Quarterly Journal of Economics. D. Prices and Budget Deﬁcits in the United Kingdom. pp. 113–172. “Oil and the Macroeconomy Since the 1970s. Backus.” Journal of International Economics. Proietti (2003). 335–405.. 297–316. 96. and P. and W. Kontolemis. and G. pp. 29. “International Real Business Cycles. 2. Fernald. and T. Rust (eds. pp. 26. Backus. (1996). 33. S. M.” In T. “Bayesian Analysis of DSGE Models. Hansen. J. Fallacy and Fantasy 15. and L. Artis. pp. 1418–1448. Handbook of Computational Economics Vol. Basu. 19. Basu. Kehoe. 5915. Smith (1993). 20. An. 745–775. 1701–1918. and T. Kilian (2004). 28. 171–252..). P. Kydland (1992). Basu. E.140 Business Cycles: Fact. 30. M. Kehoe (1992). D. Arellano. Artis. (1995). Kendrick. 115–134. “Credit Frictions and ‘Sudden Stops’ in Small Open Economies: An Equilibrium Business Cycle Framework for Emerging Markets Crises.” Econometric Reviews. R. S. Barro. pp. Chadha. 237. pp.” Journal of Economic Perspectives. Kimball (2006). 719–751. Cambridge: Cambridge University Press. R. 51. C. M. 32. pp.” Innocenzo Gasparini Institute for Economic Research (IGIER) Working Paper No. and F. 13. “Government Spending. Grossman and K. “Are Technology Improvements Contractionary?” American Economic Review. . 864–888. S. Z.” American Economic Review. Handbooks in Economics. “Dating the Euro Area Business Cycle. Zhang (1999). McGrattan. pp. 26. D. 24. 120–132. Baxter. R. “International Evidence on the Historical Perspective of Business Cycles. Basu.” Journal of Political Economy. 45–68. (1987). 82. 31.” NBER Working Paper No.” In G. Altug. Artis. M.). Cooley (ed.” International Economic Review. and W. “International Business Cycles and the ERM: Is There a European Business Cycle?” International Journal of Finance and Economics. 100. Dynamic Macroeconomic Analysis: Theory and Policy in General Equilibrium.” Journal of Monetary Economics. 17.. 25. D. D. 20. Nolan (eds. 70. 21.). pp. Backus. J.. and A. Zhang (1997).). Ashley. Interest Rates. “Further Evidence on the International Business Cycle and the ERM: Is There a European Business Cycle?” Oxford Economic Papers. pp. 27. S. S. “Business Cycles in International Historical Perspective. 16. Amsterdam: NorthHolland. Rogoff (eds. Taylor (1999). and F. 1–16. 331–356. and F. “Mechanics of Forming and Estimating Dynamic Linear Economies. pp. Schorfhiede (2007). Vol. 111. Marcellino. pp. P. 22. Anderson. and M. L. pp. pp. and C.. 35. pp.. Mendoza (2003). Patterson (1989). Osborn (1997). “Business Cycles for G7 and European Countries. Artis. pp. “International Business Cycles: Theory and Evidence. 1801–1864. Frontiers of Business Cycle Research. M. Sargent (1996). Backus. 23.” Journal of Business. Kimball (1997). 249–279. Barsky. M. 221–247. 685–704. Amman. “International Trade and Business Cycles. “Linear versus Nonlinear Macroeconomics: A Statistical Test. “Consumption and Real Exchange Rates in Dynamic Economies with Nontraded Goods. Amsterdam: Elsevier Science/NorthHolland. Princeton: Princeton University Press. 1. Kydland (1995). Kehoe. “Cyclical Productivity with Unobserved Input Variation.” In S.” In H. and D. pp. and J.
“Micro Data and General Equilibrium Models. Eichenbaum (1996). “New Keynesian. R. “Homework in Macroeconomics: Household Production and Aggregate Fluctuations.” American Economic Review. M. “Staggered Prices in a UtilityMaximizing Framework. L. Amsterdam: Elsevier Science. and R. pp. 49. “Factor Hoarding and the Propagation of Business Cycle Shocks.. Wright (1991). Caballero. A. and M.). King (1999). 170–180. “The Financial Accelerator and the Flight to Quality. pp. Bernanke. 78. 71–90.” Journal of Monetary Economics. “Variable Factor Utilization and International Business Cycles.” Journal of Monetary Economics. 1B. North Holland: Elsevier Science. 99. and S. Brock. “Agency Costs. M. and M. and T. New York: Columbia University Press for the NBER. pp. pp. O. and D. Baxter. and U. 575–593. Cyclical Analysis of Time Series: Selected Procedures and Computer Programs. 383–398. B. Jermann (1997).. and M. M. M. Gertler. 35. and J. “Irreversibility and Aggregate Investment. “Tax Disturbances and Real Activity in the Postwar United States. pp. Crucini (1995). C. 37.). Hansen. Benhabib. J. 245–274. 47. 12. Baxter. B.” The Review of Economics and Statistics. 79. Measuring Business Cycles. pp. M. and C. Burnside. and W. Bernanke. “Aggregate Investment. Quah (1989). Gilchrist (1996). Gertler (1989). 36. pp. 441–462. A. Burnside. (1994). pp.” Journal of International Economics. and C.” Journal of Monetary Economics. 22. 101. D. 421–443.Bibliography 141 34. “Is the Business Cycle Characterized by Deterministic Chaos?” Journal of Monetary Economics. 50. 52. and S. 40. Farr (2005). Doyle (2003). R. 54. J. pp. 42. Baxter. 48. 223–246. 41. pp. Vol.” Journal of Political Economy.” American Economic Review. 335–347. and R. Heckman (1999). OpenEconomy Models and Their Implications for Monetary Policy. G. Caballero (1994). 1–15. Eichenbaum. 38.” Review of Economic Studies. Baxter. Handbook of Macroeconomics. Benhabib.. Woodford (eds. M. 79. 8. Boschan (1971). Rogerson. W10130. 36. Heibling (2003). pp. 45. 81. and B. Net Worth and Business Fluctuations. 2003762. “Measuring Business Cycles: Approximate Bandpass Filters for Economic Time Series. 87(1). W. pp. Mitchell (1946). Farmer (1996). C. Taylor and M. 14–31. Browning. G.” In J. Vol. Ch. 654–673. M..” Federal Reserve Board (FRB) International Finance Discussion Paper No. (1983). 53. “The Dynamic Effects of Demand versus Supply Disturbances.” International Economic Review.” American Economic Review. Handbook of Macroeconomics. and R. “Have National Business Cycles Become More Synchronized?” NBER Working Paper No. 37.” Review of Economics and Statistics. “Indeterminacy and SectorSpeciﬁc Externalities.” Journal of Political Economy. pp. 44. “Labor Hoarding and the Business Cycle. 46. 39. Calvo. pp. 1154–1174. 61. Rebelo (1993). 43. Braun. (1999). 86. Bowman. Woodford (eds. 33.” American Economic Review. 51. Blanchard. Ch. G. 1166–1187. Bertola. 1A. Burns. 821–854. and D. Bry. pp. New York: NBER.” In J. 12. M. Taylor and M. “The International Diversiﬁcation Puzzle Is Worse Than You Think. . Bordo. “Business Cycles and the Asset Structure of Foreign Trade. 65. and R. Sayers (1988).
56. pp. 6791. 28. Nolan (2002). and M. 64. and J. T. 65. pp. (2008). Kehoe. Campbell. M. pp. Canova. 82.” Journal of the European Economic Association.” Macroeconomic Dynamics. 60.” Journal of Political Economy. Methods for Applied Macroeconomic Analysis.” Journal of Monetary Economics. F. 68. McGrattan (2004). Eichenbaum. and R. “Commodity Trade and International Risk Sharing. (2009). 85. Christiano. 1201–1249. 463–506. and C. Nason (1995). Canova. 61. (1988). 75. (ed. 3–24. “Sensitivity Analysis and Model Evaluation in Simulated Dynamic General Equilibrium Economies.. pp. Obstfeld (1991). F. 583. 67. 29. 113. pp. P.. Nason (1995). “Are Structural VARs Useful Guides for Developing Business Cycle Theories?” Federal Reserve Bank of Minneapolis Research Department Working Paper No. T. (2006). F. 182. 63.” Journal of Monetary Economics.” Journal of Applied Econometrics. L. 123–144. 697–721. “Statistical Inference in Calibrated Models. Princeton: Princeton University Press. pp. Princeton: Princeton University Press. Cole. 71. Chadha. 36.) (1995). Frontiers of Business Cycle Research. Christiano. pp. “What Explains the Great Moderation in the US? A Structural Analysis. “Effects of the Hodrick–Prescott Filter on Trend and Difference Stationary Time Series: Implications for Business Cycle Research. 41. F. 72. T. 247–280. Vigfusson (2003). Cogley. H. F. and C. Christiano. 492–511. T. “Why Does Inventory Investment Fluctuate So Much?” Journal of Monetary Economics. J.” International Economic Review. 477–501. 66. and M. J. Canova. “Current Real Business Cycle Theories and Aggregate Labor Market Fluctuations. 1–38. 1. and C. 7. “How Much Structure in Empirical Models?” CEPR Discussion Paper No. “Financial Structure and International Trade. Cooley. Cooley (ed. H. “Sticky Price and Limited Participation Models of Money: A Comparison. pp. “Economic Growth and Business Cycles. pp. M. pp. 59. Canova. Canova. .” International Economic Review. Chari. 430–450.” American Economic Review. 237–259. “Output Dynamics in Real Business Cycle Models. 74. (1988). 19. pp. L. M. L. 21. L. and E. 9819. Eichenbaum. 58. V. “The Behavior of Interest Rates in a General Equilibrium Multisector Model with Irreversible Investment. Evans (2005). “Inspecting the Mechanism: An Analytical Approach to the Stochastic Growth Model.. (1997). 9. Cogley. 69. Christiano. pp. “Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy. Princeton: Princeton University Press. 72–89. 631. (1994). 62. Canova. and L. Prescott (1995).” ECB Working Paper No. and E. and J.” Journal of Economic Dynamics and Control. pp. Fallacy and Fantasy 55. Cooley. Christiano. 73. Sala (2006). 70. “What Happens After a Technology Shock?” NBER Working Paper No. F. Eichenbaum (1992). (1995).” In T. W. Eichenbaum.). 253–278. pp. 57. Frontiers of Business Cycle Research. “Back to Square One: Identiﬁcation Issues in DSGE Models. Evans (1997). 206–227. 33. Cole. pp. L. (1994). Coleman.” European Economic Review. “A Long View of the UK Business Cycle. 1–45.” American Economic Review.” National Institute Economic Review..142 Business Cycles: Fact.
Cambridge. “Investment Dynamics. “An AreaWide Model (AWM) for the Euro Area. and Banking. Horioka (1980). (1985)... W8716. 2nd ed. pp. “Perfect Competition and the Effects of Energy Price Increases on Economic Activity. F. J.).. L. Gregory.” Review of Economic Studies. pp. “Business Cycles in Small Open Economies. Dynamic Macroeconomic Analysis: Theory and Policy in General Equilibrium. Rebelo (1995). A. Real Business Cycle Theories. and the Business Cycle: Do Technology Shocks Explain Aggregate Fluctuations?” American Economic Review. 15. 77. Gali.” Journal of International Money and Finance. Hansen.. M. M. 79. pp. M. DC: Brookings. 83. Fagan. 1089–1113. J. Finn. 86. Credit. Princeton: Princeton University Press. 1867– 1960.Bibliography 143 76. M. and C.” NBER Working Paper No. MA: MIT Press. 95. J. 1929–1982. 90. 1004. RubioRamirez (2007). “Realistic CrossCountry Consumption Correlations in a TwoCountry. 94. 90. and Animal Spirits. R. Cooley. FernandezVillaverde. “A Neoclassical Model of the Phillips Curve Relation. Altug (2003). 89. 453–473.” Review of Economic Studies. Trends in American Economic Growth. “Real Business Cycles: Wisdom or Whimsy?” Journal of Economic Dynamics and Control. pp. Quadrini (1999). J. 34–154. Washington. Demers. 92. pp. “Irreversibility and the Arrival of Information. 82. “Technology. (1999). 51–78. pp. I. 14677. 13166. Eichengreen. J. pp. Singleton (1988).” ECB Working Paper No. . “A Time Series Analysis of Representative Agent Models of Consumption and Leisure Choice Under Uncertainty. Fair (1992). 314–329. 65. 44. Farmer. 58. The Macroeconomics of SelfFulﬁlling Prophecies. G.” Cowles Foundation Discussion Paper No. M. “Let’s Get Real: A Factor Analytical Approach to Disaggregated Business Cycle Dynamics. 88. J. Frisch. and K. 91.” In S. and G. Eichenbaum. and L. 42. pp.” Journal of Money.” Economic Journal. Denison. “Crises Now and Then: What Lessons from the Last Era of Financial Globalization?” NBER Working Paper No. R. Correia. “The Econometrics of DSGE Models. 607–626. 400–416. and V. Smith (1992). pp. M. 96. 249–271. pp.” Quarterly Journal of Economics. Demers. Devereux. and New Keynesian Economics. 14846. and C.” In Economic Essays in Honor of Gustav Cassel. Altug. (2009). Feldstein. B.” Journal of Monetary Economics. A Monetary History of the United States. 3–16. (1991). Chadha.” European Economic Review. Schwartz (1963). “The Cowles Commission Approach. Equilibrium Business Cycle Model. 85. (1999). 165–193. Forni. and R. Farmer. 39. “Domestic Saving and International Capital Flows. Crashes. 93. Nolan (eds. and A. and R. 78. 11. pp. Demers. (1991). Henry. E. Reichlin (1998). “Conﬁdence. (1933). Eichenbaum. 87. T. 32. Neves. “How Structural Are Structural Parameters?” NBER Working Paper No. R. Employment. M. (2009). 80. M. 89.. and M. Bordo (2002). 333–350. and S. FernandezVillaverde. 81. M. 103. London: George Allen & Unwin. Friedman. Mestre (2001). “Propagation Problems and Impulse Problems in Dynamic Economies. 84.” NBER Working Paper No. Cambridge: Cambridge University Press. (2000). and S.
Hercowitz. 91–115. pp. .S. 115. Princeton: Princeton University Press. pp. Gallant. J. Capacity Utilization. 117. 101. Head. and J. “Calibration as Estimation. London: Routledge.” European Economic Review. and P. 116. Cambridge. “Financial Autarky and International Business Cycles. pp. 96. Hartley. Fallacy and Fantasy 97. Hansen. pp. Gregory.” International Economic Review.” Journal of Political Economy. 102. D. Rabanal (2005).” University of North Carolina Working Paper. 9. 601–627. pp. Common Factors and the Empirical Validation of Equilibrium Business Cycle Models. and the Business Cycle. 309–327. Version 8. pp. “Large Sample Properties of Generalized Method of Moments Estimators. 104. 57–89. NBER Macroeconomics Annual 2004. Gordon. GarciaCicco. 402–417.. Z. G. and P. J.” SCE Computing in Economics and Finance Working Paper No. Tauchen (1996). Pancrazi. 657–681. Z. (1985).7. Growth/Productivity/Unemployment. Chicago: University of Chicago Press. pp.) (1998). 110. 132. Heathcote. 27. J. R.144 Business Cycles: Fact. pp. pp. Huffman (1988).).” Journal of Monetary Economics. (1990). and the Real Business Cycle. 401. L. A. User’s Guide. and L. “SNP: A Program for Nonparametric Time Series Analysis.. 677–702. “LongRun Implications of InvestmentSpeciﬁc Technological Change. (2008). 107. R. Hansen. 257–279. Smith (1989). 99. The Measurement of Durable Goods Prices. (1990). 12629. 49. Sargent (1980). L. Salyer (eds. 2. 109. Inﬂation. “Which Moments to Match?” Econometric Theory. “Formulating and Estimating Dynamic Linear Rational Expectations Models.. and G. 87. 114. Greenwood. J. MA: MIT Press. 342–362. Gertler and K.. and T.” In P. 16.” Journal of Economic Dynamics and Control. Giannone. “Computational Experiments and Reality. “Invariance Properties of Solow’s Productivity Residual.” Econometrica. “Investment. “The Role of InvestmentSpeciﬁc Technological Change in the Business Cycle. Geweke. 12. Hoover. K. pp. Z. Greenwood. A. Reichlin. and F. R.).” Econometric Reviews. Rogoff (eds. A. J. Tauchen (1998). “A Comparison of Business Cycle Dating Methods. and K. R. D. Perri (2002). Raynauld (1997). “The Relation Between Price and Marginal Cost in U. 44. 113. 78. 1029–1054. MA: MIT Press. pp. 921–947. pp.” American Economic Review. Pagan (2003). Gali. Harding.” Journal of Economic Dynamics and Control. (1999). 50. J.. 7–46.” American Economic Review. 38. and A. J. Gregory. Real Business Cycles: A Reader. 106. R. “Real Business Cycles in Emerging Countries?” NBER Working Paper No. pp. (1988). Gallant. Gali.” Journal of Monetary Economics. 225–318. 112. 103. Cambridge. and A. pp. 111. (1982). Monetary Policy. Hall.. R. Hercowitz. and P. J. Uribe (2006). Krusell (2000). and G. 98. Industry. 71–112. and G. pp. Krusell (1997). Hansen. Greenwood. Sala (2006). “Technology Shocks and Aggregate Fluctuations: How Well Does the RBC Model Fit Postwar US Data?” In M. 1681–1690. Hercowitz. 105. L. “VARs. Hall. and M. J. “Measuring World Business Cycles. 108.. Diamond (ed. 100. “Indivisible Labor and the Business Cycle.” Journal of Econometrics.
pp. Kapetanios. A. 135. and S.” Review of Economic Studies. 134. Region.Bibliography 145 118. and F. MA: MIT Press. and the Structure of International Asset Markets. “Debt Constrained Asset Markets. and S. pp.. and J. 86. (1936). A.” Journal of Econometrics. 191–211. G. 14. (1926). Perri (2002). (1995). Hess. 21. 139. 124. Terrones (2003). Real Exchange Rates. pp. 21. 13.” American Economic Review. and Prices in a Real Business Cycle Model. King. Prescott (1997).S. (1998). (2004). pp. R. pp. Growth. T. “Credit Cycles. 123. C. 138.” Journal of Money. and Business Cycles: II. 309–341. 923–936. Whiteman (2003).” Review of Economic Studies. 702. “Production. 121. 56.” Econometrica. DC: Brookings Institution. R. K. Levine (1993). Stock. Köse. R. Numerical Methods in Economics. and D. King. P. and C. Plosser. 207–231. and S. 573–606. (1998). “Stochastic Trends and Economic Fluctuations. Pagan. Scott (2007). 81. Rebelo (1988). Kehoe. Judd. Moore (1997). 865–888. Prasad. and K. and M. 130. C. and E. Köse.” Review of Economics and Statistics.” Oxford Review of Economic Policy. “Money. 70. and Banking. “International and Intranational Business Cycles. J. pp. pp. King.” Archiv für Sozialwissenschaft und Sozialpolitik. and C. R. Plosser (1984).” American Economic Review. Shin (1997). 132. Credit. King. E.” American Economic Review. R. 363–380. Watson (1991). “International Business Cycles with Endogenous Incomplete Markets. R. pp. and A. 119. London: Macmillan. Kondratiev. Credit. How Much Do National Borders Matter? Washington.. Rebelo (1988). Plosser. R. “Production. Keynes. 81. The Basic Neoclassical Model. and CountrySpeciﬁc Factors. C. 60. pp. “Low Frequency Filtering and Real Business Cycles. 195–232.” Economic Quarterly. Plosser. 211–248. Levine (2001). pp. 565–594. Interest. 1216–1239.” Journal of Economic Dynamics and Control. “Making a Match: Combining Theory and Evidence in PolicyOriented Macroeconomic Modeling. pp. R. J. Growth. pp. P. “Quantitative Theory and Econometrics. “Consumption. Helliwell. Otrok. 17. (1996). 126. and D. A. pp. 136. King. and Business Cycles: I. Business Cycles. 120. 69(3). 122. 125.. “Technology Shocks in the New Keynesian Model. 93–109. “International Business Cycles: World. pp. 137. “Postwar U. 595–609..” Econometrica.” Journal of Monetary Economics.. pp. Kiyotaki.” Journal of Political Economy. 127. King. C. The General Theory of Employment. 128.. 907–928. . “Implications of Efﬁcient Risk Sharing without Commitment. and Money. Ireland. N. 1–16. 133. pp. Kehoe. 819–840. Kehoe. “Liquidity Constrained Markets versus Debt Constrained Markets. N. (1995). 53–105. 63. 131. Kocherlakota.” Journal of Monetary Economics.” Journal of International Money and Finance. 575–598. 74. Cambridge.T. New Directions. 136. pp. Rebelo (1993). J. N. Kollman. “Die Langen Wellen der Konjunktur [The Long Waves of the Business Cycle]. 105. 129. G. 29. M. pp. “How Does Globalization Affect the Synchronization of Business Cycles?” IZA Discussion Paper No. 93. Hodrick. and M.
F. “Recursive Contracts. Kydland. and E. 91. J. and R. R. Lucas.. 397–414. 159. Cambridge. D. “Interest Rates and Currency Prices in a TwoCountry World. “An Equilibrium Model of the Business Cycle.). “Business Cycles and Aggregate Labor Market Fluctuations. pp. pp. 93. 33. (1996). Prescott (1990). Lewis. McGrattan. 85. Fallacy and Fantasy 140. Employment and Inﬂation. K. R. E. 145. 161–178. 153. and C. and C. “What Can Explain the Apparent Lack of International Consumption Risk Sharing?” Journal of Political Economy. “Identifying the Common Component in International Economic Fluctuations. (1995). 151. Prescott (1982). 14. (1986).” Journal of Political Economy. 157.” Universitat Pompeu Fabra (UPF) Economics Working Paper No. pp. “On ‘Real’ and ‘StickyPrice’ Theories of the Business Cycle. Wright (1997). and L. Long. “An Equilibrium Model of the Business Cycle with Household Production and Fiscal Policy. A. pp. 77.” Scandinavian Journal of Econometrics.” In T. R. E2008/11. Wickens (2008). pp. 39–69. Minford. (1972). 158. 1345–1370. McGrattan.” Journal of Political Economy. McCallum. (1996). 721–754. 148. pp.. 160. “Real Business Cycles. “Argentina’s Lost Decade. 10. 7–29. “Understanding Business Cycles. Plosser (1983). MA: MIT Press. 4. Princeton: Princeton University Press. 3–18. pp. F. 149. M. “National Borders Matter: CanadaUS Regional Trade Patterns. J. F. 267–290. Zaragaza (2002).” Journal of Monetary Economics. Rapping (1969). R. . Kydland.” Journal of Political Economy. pp. Stabilization of the Domestic and International Economy. 126–156. and T. 155. Cooley (ed. (1976). (1977). Prasad (2003). Frontiers of Business Cycle Research. pp. “Real Wages. Credit. 146. Amsterdam: NorthHolland. 19–46.” Review of Economic Dynamics. 144. pp. B. “Testing a DSGE Model of the EU Using Indirect Inference. Lucas. 18. R. pp. pp. “The Econometrics of the General Equilibrium Approach to Business Cycles. 154. pp. E.” Journal of Economic Theory. Lucas. (1994). 152. 267–297. 141.” Cardiff Business School Economics Working Paper No. Lucas. Lucas. 335–359. (1982). 142. “Business Cycles: Real Facts and a Monetary Myth. (1975).” Econometrica. 150. 573–601. Kydland. Recursive Macroeconomic Theory. pp. Kydland.” Economic Journal. Ljungqvist. Rogerson. “Econometric Policy Evaluation: A Critique.). 337. pp. “The Macroeconomic Effects of Distortionary Taxation.” International Economic Review. R. pp. Brunner and A. Meenagh. 83. R.” Journal of Monetary Economics.” Journal of Money. 615–623. Lucas. Kydland. and E. 103–123. 1113–1144. 38. pp. 156. F. Marimon (1999).146 Business Cycles: Fact. L.” In K. 152–165. R.” Journal of Monetary Economics. 147. Meltzer (eds. and Banking. 104(2). and R. and E. F.” American Economic Review. 101–127. E. pp. 5. 50. 143. McCallum. 1(2) (Supplementary Series). J. “Expectations and the Neutrality of Money. and M. Sargent (2000). “TimetoBuild and Aggregate Fluctuations. Marcet. Lumsdaine.” Federal Reserve Bank of Minneapolis Quarterly Review. P. 113. and E. Prescott (1991).
169. pp. S. 1869–1908. and P.” American Economic Review. NBER Macroeconomics Annual 2000.” Scandinavian Journal of Economics. “Real Business Cycles in a Small Open Economy.). 181.Bibliography 147 161.” Journal of Monetary Economics. 162. 307–328. V. Ramey. 171. O. (1991). Statistical Indicators of Cyclical Revivals (NBER Bulletin No. “Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample?” Journal of International Economics.” Journal of Monetary Economics.” CEPR Discussion Paper No. 97. 1–37. 11–44. “Displaced Capital: A Study of Aerospace Plant Closings. pp. pp. S. Rodrik. Rouabah (2008). C.” Unpublished manuscript. Bernanke and K. 167. Meese. “Financial (In)stability. “Economic Fluctuations in Central and Eastern Europe: The Facts. 21.” Review of Economic Studies. E. Neftci. Obstfeld. and Future. Plosser (1982). Business Cycles: The Problem and Its Setting. (1984).” Journal of Economic Surveys. “Trends and Random Walks in Macroeconomic Time Series. pp. (1997). and P. Benczur (2005). Romer. 175. “Policy Uncertainty and Private Investment in Developing Countries.” Paper presented at the IEA 15th World Congress. 107. S. “Indivisible Labor. and C. Present. Turkey. 316–334. Nelson. A. pp. New York: W.. Rogerson. Ohanian. and A. Microeconomic Foundations of Employment and Inﬂation Theory. Mitchell. C. (2005). (1970). Phelps. Mortensen. (1988). Ratfai. R. 164. “Nonlinear Time Series Modelling: An Introduction. M. 797–818. New York: National Bureau of Economic Research. 163. Norton & Co. and M. (1986). Cambridge. Mendoza. 13. (1989). Mitchell. 81. 10. A. D. (1999). 166. Lotteries and Equilibrium. and K. Burns (1938). Rogoff (2001). Rogoff (eds. “Facts of Business Cycles Around the Globe. D. and K. “Are Economic Time Series Symmetric over the Business Cycle?” Journal of Political Economy. pp. 173. “The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?” In B. 3–24. “The Macroeconomic Effects of War Finance in the United States: World War II and the Korean War.” CarnegieRochester Conference Series on Public Policy. E. 36. 339–390.W. “The Prewar Business Cycle Reconsidered: New Estimates of GNP. Pissarides (1994). 391–415. 23–40. pp. 61. 170. C. 139–162. 76. (1986).” American Economic Review. de Walque. Benczur (2008). “Is the Stabilization of the Postwar Economy a Figment of the Data?” American Economic Review. (1927). pp. Istanbul. Prescott. 3–16. Shapiro (2001). E. 179. 168. Rebelo. 174. 87. pp. 165. G. pp. Supervision and Liquidity Injections: A Dynamic General Equilibrium Approach. L. “Job Creation and Job Destruction in the Theory of Unemployment. Romer. (1991). 172.” Journal of Political Economy. Potter. 229–242. pp. pp. pp. . Rogoff (1983). 25. 505–528. “Real Business Models: Past. 109. pp. 177. New York: National Bureau of Economic Research. 176. W.” Journal of Political Economy. R. and C. 69). 14. 217–238. 958–992. pp. 180. MA: MIT Press. 4846. and A. Pierrard. “Theory Ahead of Business Cycle Measurement. 178.” Journal of Development Economics. Ratfai. 92. W.
” Journal of the European Economic Association. “Tastes and Technology in a TwoCountry Model of the Business Cycle: Explaining International Comovements. Stock. 184. Rotemberg. 361–387.” Review of Economics and Statistics. MA: MIT Press. “Imperfect Competition and the Effects of Energy Price Increases on Economic Activity. 1A. “Econometric Issues in the Analysis of Equilibrium Business Cycle Models.” Econometrica. pp. Striving for Growth After Adjustment: The Role of Capital Formation. 196. 190. 28.” In T.” Journal of the European Economic Association.” In J. Woodford (1996).” Journal of Applied Econometrics. F. 5. NBER Macroeconomics Annual 2002. Credit. pp. Fallacy and Fantasy 182. 20. 189. 192. Sargent. 645–670. and M. and M. and L. pp. 195. pp. pp. 243–293. T. Rotemberg. 968–1006. Handbook of Macroeconomics. “Technical Change and the Aggregate Production Function. Cambridge. Cooley (ed.” Journal of Applied Econometrics. and M. 62. North Holland: Elsevier Science. Sims. E. 105–146. J. 187. Smets. Woodford (eds. 251–287. . 21. Rogoff (eds. 97. “Dynamic General Equilibrium Models of Imperfectly Competitive Product Markets. Sims (1977). and M. 183.” American Economic Review. and C. pp. pp. Macroeconomic Time Series. (1934).). “An Estimated Dynamic Stochastic General Equilibrium Model of the Euro Area. Schorfhiede. Smets.” Econometrica. 3–64. 198. 197. 185. 11–30. Watson (2003). and R. 191. 45–109. 1–48. J. “Business Cycle Modeling Without Pretending to Have Too Much A Priori Economic Theory. and A. F.” American Economic Review. (2000). pp. J. pp.” Journal of Monetary Economics. pp.). Gertler and K. “Private Investment and Macroeconomic Adjustment: A Survey. 194. Income and Causality. “Business Cycle Fluctuations in U. Sargent. pp.” In L. Vol. 159–230. R. Schumpeter. Wouters (2005). 540–553. “The Summation of Random Causes as the Source of Cyclic Processes. (1972). “Macroeconomics and Reality. Frontiers of Business Cycle Research. MA: Harvard University Press. and R. pp. A. 200. pp. Woodford (1995). Solimano (1993). J. 193. J. 85. Washington. J. DC: The World Bank. L. 161–183. “Has the Business Cycle Changed and Why?” In M. 312–320.” In New Methods in Business Cycle Research: Proceedings from a Conference. and M.). T. “Money. 1.” Journal of Political Economy. (1957). (1937). and Banking. “Understanding Changes in International Business Cycle Dynamics. Serven and A. Princeton: Princeton University Press. 1123–1175. 39. pp. 199. F. Schumpeter. 5. Solimano (eds.S. pp. Stock. Solow. 549–577. J. 15. MN: Federal Reserve Bank of Minneapolis.). “Loss Functionbased Evaluation of DSGE Models. 168–185. C. (1939). “Comparing Shocks and Frictions in US and Euro Area Business Cycles: A Bayesian DSGE Approach. (1988). Sims. K. Watson (2005). Stock. The Theory of Economic Growth. Slutsky. Business Cycles. Serven. Watson (1999). Taylor and M.” Journal of Money. New York: McGrawHill. Minneapolis. Cambridge. “Two Models of Measurement Error and the Investment Accelerator. 186. Stockman. Singleton. C. Wouters (2003).148 Business Cycles: Fact. 188. 48. (1989). pp. (1980). Tesar (1995). pp.
(2007). 202. Vol. L. 31(9). and I. “Statistical Nonlinearities in the Business Cycle: A Challenge for the Canonical RBC Model. Chicago: The University of Chicago Press. 2957–2983. The Royal Swedish Academy of Sciences (2004). MA: Harvard University Press. Tesar. “PlantLevel Irreversible Investment and Equilibrium Business Cycles. . J. 10. pp. 110. Veracierto. Santomero (eds. (2007). “Measures of Fit for Calibrated Models. 97. 208. and Forecasting. pp.” American Economic Review.Bibliography 149 201.). V. “Is Lumpy Investment Relevant for the Business Cycle?” Journal of Political Economy.de/institute/wpol/html/toolkit. 181–197. Zarnowitz. “The Internationalization of Securities Markets Since the 1987 Crash. 203. Prescott (1989). (1993). Watson. Stokey. (1986). 206. 209. N. pp. Uhlig.” Available at http://www2. Werner (1998). 281–372. Finn Kydland and Edward Prescott’s Contribution to Dynamic Macroeconomics: The Time Consistency of Economic Policy and the Driving Forces Behind Business Cycles.” Federal Reserve Bank of Minneapolis Quarterly Review. Summers.” In R. 207. 101. Thomas. 1. (2002). Lucas. DC: The Brookings Institution. Business Cycles: Theory. (2002). 210. pp. 23–27. Washington. 204. pp. “A Toolkit for Analyzing Nonlinear Dynamic Stochastic Models Easily. M. M. and R. BrookingsWharton Papers on Financial Services. Litan and A. D. Cambridge. Indicators. 1102–1130. L. 508–534.” Journal of Political Economy. pp. 1011–1041. “Some Skeptical Observations on Real Business Cycle Theory. H. Advance Report on the Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel. pp.” American Economic Review.” Journal of Economic Dynamics and Control. Weitzman. History. (1992). 205.wiwi.htm/. M.huberlin. Recursive Methods in Economic Dynamics. (1997). “Subjective Expectations and AssetReturn Puzzles. 92. with E. Valderrama. 211.
This page intentionally left blank .
124 credit market frictions. 94 animal spirits. 49 inside money. 16 leading indicator. 65 lagging indicator. 19. 130 indivisible labor. 125 Kalman ﬁlter. 117 goodnessofﬁt measures. 16 calibration. 83 generalized method of moments (GMM). 113 home production. 138 deterministic steady state.Index accommodative monetary policy. 59 irreversible investment. 68 investmentspeciﬁc technological change. 45 indirect inference. 42 business cycles coincident indicator. 110 for large crosssections. 58 international portfolio diversiﬁcation. 101 international. 33 Bellman’s equation. 120 enforcement constraints. 66 current account. 131 kurtosis. 109. 124 . 110 complete contingent claims. 73 common factors. 135 dynamics of output. 54 impulse response functions. 39 dynamic factor analysis. 34 crosscountry correlations. 43 capital account. 129 identiﬁability of. 73 demandconstrained equilibrium. 58 aggregate price index. 16 in emerging markets. 54 homeproduced good. 71 conditional heteroscedasticity. 116 151 dynamic stochastic general equilibrium (DSGE) models. 80 international risk sharing.
132 monopolistic competition. 2 state space representation. 53 industryspeciﬁc. 94 multisector models. 49 Business Cycles: Fact. 135 Sudden Stops. 137 New Keynesian model. 137 selffulﬁlling expectations. 50 Solow residual. 78 optimal linear regulator problem. 56 transition equation. 137 seminonparametric (SNP) model. 38 spectral density function. 97 total factor productivity (TFP). 63 government consumption. 54 market incompleteness. 48 quantity anomaly. 131 unconditional moments. Fallacy and Fantasy Riccati equation. 123 nonnormal disturbances. 132 structural models. 82 maximum likelihood estimation frequency domain. 124 social planner’s problem. 42 posterior distribution. 88 vector autoregression (VAR). 55 limited risk sharing. 137 multiple equilibria. 124 shocks. 78 monetary. 98 technology. 67 puzzles. 78 energy. 33 countryspeciﬁc. 97 real business cycle (RBC) theory numerical solution. 33 nonconvexities in labor supply. 124 nonspecialization in endowments. 77 nontraded good. 99 market good. 33 skewness. 66 real balances. 111 stabilization policy. 132 predictive density.152 labor hoarding. 112 measurement equation. 110 variable capacity utilization. 49 nonlinear time series models. 56 . 58 structural. 44 unobservable index models. 101 transactions services from money and banking. 121 price anomaly. 38 trade balance. 101 symmetric equilibrium. 94 productivity spillovers. 33 reverse causality. 81 lottery models. 66 price rigidities. 43 search models.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.