World Scientific
Publishing Co.
Pte. Ltd.
F a c
t ,
F
a
l
l
a
c
y
a
n
d
F
a
n
t
a
s
y
B
u
s
i
n
e
s
s
C
y
c
l
e
s
This page intentionally left blank This page intentionally left blank
NE W J E RSE Y • L ONDON • SI NGAP ORE • BE I J I NG • SHANGHAI • HONG KONG • TAI P E I • CHE NNAI
World Scientifc
F a
c
t
,
F
a
l
l
a
c
y
a
n
d
F
a
n
t
a
s
y
B
u
s
i
n
e
s
s
C
y
c
l
e
s
S u m r u G A l t u g
Koç University, Turkey &
Centre for Economic Policy Research, UK
Library of Congress 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
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
. Turkey & Centre for Economic Policy Research. UK
ess sin Bu les Cyc
and y allac Fact.
HB3711.A424 2009 338. This book. recording or any information storage and retrieval system now known or to be invented.
For photocopying of material in this volume. ISBN13: 9789812832764 (hardcover) ISBN10: 9812832769 (hardcover) 1.com
Printed in Singapore.. please pay a copying fee through the Copyright Clearance Center. cm. and fantasy / by Sumru G. USA. MA 01923. 5 Toh Tuck Link. I. or parts thereof.Published by World Scientific Publishing Co. London WC2H 9HE
Library of Congress CataloginginPublication Data Altug. Covent Garden. Includes bibliographical references and index. Danvers. In this case permission to photocopy is not required from the publisher. without written permission from the Publisher. Inc. may not be reproduced in any form or by any means. fallacy. Ltd.
Typeset by Stallion Press Email: enquiries@stallionpress. Business cycles.
. Business cycles : fact. Suite 401402. Pte. 222 Rosewood Drive.5'42dc22 2009034620
British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library. Ltd. including photocopying. Title. Hackensack. Singapore 596224 USA office: 27 Warren Street. electronic or mechanical. Altug. Sumru. NJ 07601 UK office: 57 Shelton Street. All rights reserved.
Copyright © 2010 by World Scientific Publishing Co. p. Pte.
beginning with Keynes. Largescale computers in the postWorld War II era played an important role in the development of simultaneous equation models. This book draws upon Kydland and Prescott’s original contribution. 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.Preface
How do intellectual disciplines progress? Undoubtedly. 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]. This debate continues to this day. real business cycle (RBC) analysis has come to provide a ﬂexible and popular approach for examining macroeconomic phenomena. In 2004. the discipline of economics — and macroeconomics. The model was rejected. I was a Ph. the development of new techniques or new ways of modeling can also affect the course that a discipline takes. Finn Kydland and Edward Prescott received the Nobel Prize in Economics. The Great Depression greatly inﬂuenced the perceptions of a generation of economists. The oil shocks of the 1970s and the 1980s affected economists’ views regarding the sources of macroeconomic ﬂuctuations. in particular — is affected by major changes in economic conditions.D.
v
. The ﬁeld of macroeconomics has changed signiﬁcantly due to Kydland and Prescott’s contributions. 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]. Sometimes. In recent years. My thesis was on estimating the model in the same article.
Kevin Hoover.vi
Business Cycles: Fact. Jordi Gali’s [97] recent text articulates an alternative New Keynesian framework for describing aggregate ﬂuctuations. researchers at central banks have begun using socalled dynamic stochastic general equilibrium (DSGE) models for policy analysis. This book takes a more eclectic approach.
. The volume by Thomas Cooley [74] can be considered a primer of RBC analysis and its applications. In recent years. James Hartley. This book attempts to provide an overview of the burgeoning business cycle literature that. asking some basic questions about RBC analysis and summarizing the ongoing controversies surrounding it. in many ways. Not content with the initial rejection of the model. Talented and creative individuals extended the initial Kydland–Prescott research in different ways. This literature evolved in different ways. and Kevin Salyer’s [116] collection of articles provides a critique of the calibration approach. There have been a number of excellent publications that have examined different facets of this literature. reﬂects my own interests. many researchers also pursued the econometric analysis of RBC models. Fallacy and Fantasy
Kydland and Prescott’s seminal article initiated the school of RBC analysis.
. . . .2. .2. . . . . . . .3. . . 4. . . .2. . . . . . . .1. . . 3.5. Facts . 4. . . .1. . . . . . . A Numerical Solution . . . . . . . . . InvestmentSpeciﬁc Technological Shocks 3. . . Historical Business Cycles . 65 68 69 71
. . . . . . . . . . .5. . . . . . . . . .3. . . . . . . .2. . . 3. . . . . “Puzzles” . . Energy Shocks . . . . . . . . . . . . . . . . . . 3. . . . .4. . . The Role of International Risk Sharing 4. . . . 3. . . . . . . . .4. . . . . .4. . . . . . . The Productivity Puzzle . . . . .2. . 2. 3. . . . . . . . . . . . . . . . . . . Complete Contingent Claims
vii
v 1 7 . .1. . . . . . . . The Euro Area Business Cycle . . . 2. . . . . 4. .5.Contents
Preface 1. . . . . . . . . . . An RBC Model . . . . . . . . . . . . . . 3. . . . . . 2. . . . . . 2.1. 3. . .5. . . . .2. . Stylized Facts . . . .1. . . . . . . . . .4. . . . . . . 7 15 19 25 28 33 34 39 46 48 49 52 56 58 59 63 65 . . . Is There a World Business Cycle? 2. . . . . . . . Deﬁning a Business Cycle . The Source of the Shocks . . Pareto Optimal Allocations . . . . 3. Models of Business Cycles 3. . . Reverse Causality . .1. . . Introduction Facts 2. . A Model with Indivisible Labor Supply . . . . . International Business Cycles 4. .2. . . . . . . . . . . . . .4. . . . . . 3. . . . . . . . . . Initial Criticisms . . . . .2.3. . . . . . . . . . . .
. . . . . . . . . .2. 5. . 4. . The Debate Reconsidered . .
. .
. .
7.
. . .3. . . . . . . . . . . . .2. . . . . . . . . . 7. The Basic Model . Calibration as Estimation . .
. . . .
4. . .2. . . . GMM Estimation Approaches . 5. . . . . . . . The Calibration versus Estimation Debate . . . . . .1. .2. . . . .
. . . . . . . . . . . . . . . Puzzles Revisited . . . 7. . . . Limited Risk Sharing . . Fallacy and Fantasy
4. . . . . 7. . . . . . . Measures of Fit for Calibrated Models . . .1. . . . . .
. .4. . . . . .2. . . . . . 7. . . . . . The Dynamics of Output . . .3. . .1. . . . . . . . . .2. . . . . . . 7. . . . . .
Matching the Model to the Data 7. . . . . . . . . . . . Other Applications . . . . . . .
Future Areas for Research
Bibliography Index
. . . Nontraded Goods .
6. Other Extensions . . . . . . No Asset Trading . . . 6.2. . . .
. . . . .
. . . . . . . . . . . .3. . . .3. . . . . . Nonspecialization in Endowments 4.viii
Business Cycles: Fact. . .4. . . . . . . . . . . .3. .2.1. A Small OpenEconomy Model of Emerging Market Business Cycles . . . . 4. . . . . . . . . . . . .3. . . . .
Business Cycles in Emerging Market Economies 6. . Dynamic Factor Analysis . . . . . . . .2. . . . . Nonlinearity in Macroeconomic Time Series 7. . Do Shocks to Trend Productivity Explain Business Cycles in Emerging Market Economies? . . . 4. . . 4. . . . . 7. . . .3. . . . Empirical Evidence .1. . .4. . . . 7. . .2.1. . . 7. . DSGE Modeling .7. . . . . . . . .1. . Trade in Equity Shares . .
8.6.
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. .5. . . . .4. .3. . . . . .
and crises experienced by market economies in the 19th and early 20th centuries.” Ever since. the notion of business cycles originated from various types of panics. there were periodic movements or “long waves” in economic variables. 1840–1890. both growth and business cycles could be ascribed to the process of innovation. corresponding to the introduction of steel and steam engines. “crises” are an endemic feature of capitalist economies. more effort is required to master it. and motor engines. other scholars investigated methods for the systematic
1
. when one examined the history of commercial cycles for the capitalist economies of the time. corresponding to the invention of electricity. He identiﬁed three long waves: 1780–1840. one of the most prominent thinkers of the time. Kondratiev [137] argued that in addition to shorter economic cycles. other economists observed that the alternating phases of prosperity and depression seemed to follow each other on a regular basis. and 1890–1950. depressions.Chapter 1
Introduction
In the opening page of the book Business Cycles published in 1927. However. Wesley Mitchell [163] comments as thus: “As knowledge of business cycles grows. much effort was devoted to understanding the causes of what many viewed as “abnormal” phenomena. 188] sought to explain the existence of such long waves as an outcome of technological innovations. chemical processing. corresponding to the Industrial Revolution. there have been many developments in the ﬁeld of business cycles. Historically. This book describes these new developments. Schumpeter [187. As Mitchell [163] recounts. In his framework. While some scholars proposed different theories to explain ﬂuctuations in economic activity. According to Karl Marx. In the 1920s.
After the early work of Burns and Mitchell. or stagﬂation as it is popularly known. led researchers to account for the observations using new mechanisms for the effects of money on output. post World War II. The Norwegian economist Ragnar Frisch [95] developed the notions of impulse and propagation mechanisms for describing business cycles. The oil shocks of the 1970s and the experience of high inﬂation and high unemployment.
. a business cycle is the simultaneous downturns and upturns of a large number of economic series. Keynes’ General Theory [127] laid the foundations for the analysis of shortrun economic ﬂuctuations. which continue to form the main vehicle for empirically studying business cycles. According to them. Writers such as Frisch and Slutsky embedded the notion of business cycles into simple dynamic systems driven by stochastic shocks. the focus shifted to stabilization policy. Another important channel which affected the study of business cycles was the development of statistical and time series methods. Their inﬂuence on thinking about business cycles has persisted to this day. The Great Depression and World War II were two major events in the development of business cycle analysis. and it continues to this day in the business cycle dating methodology employed by the NBER.2
Business Cycles: Fact. Phelps [170] and Lucas [148. the Keynesian framework was interpreted as a model of output determination at a point in time. the lessons of the Great Depression were vivid in the minds of many policymakers. Their ideas were formulated in terms of linear time series models. Nevertheless. and modeled business cycles as the response of a secondorder dynamic system to random shocks. Burns and Mitchell [50] and researchers at the National Bureau of Economic Research (NBER) began to identify the phenomena of business cycles. During this time. Their work involved the dating of business cycles and the development of leading indicators for the US economy.1 We discuss the NBER methodology and the stylized facts of business cycles in Chapter 2. 149] developed monetary models of the business cycle as a way
1 See also Zarnowitz [211]. In their seminal contributions. Post World War II. Slutsky [193] argued that the sum of a number of uncorrelated shocks could produce serially correlated or smooth movements in the generated series. The period following World War II was an era of high and sustained growth in many countries. Fallacy and Fantasy
measurement and identiﬁcation of business cycles.
In his article “Understanding Business Cycles”. this involves matching a small set of moments implied by the model with those in the data. namely.Introduction
3
of providing a consistent theoretical foundation for describing the impact of changes in money on output. was identifying technology shocks that could generate cyclical ﬂuctuations of magnitudes observed in the data (see Summers [202]). 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. In some sense. Widely known as the calibration approach. timetobuild in investment. 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. Though this approach is cited extensively. 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. They also proposed a methodology for confronting their theory with the data. as identiﬁed by skeptics. 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. One of the major issues with Kydland and Prescott’s contribution. the debate on the empirical validation of business cycle models. we cannot ignore the calibration approach or. the durability of leisure. In Phelps’ and Lucas’ framework. there was a revival of interest more generally in examining aggregate economic activity as recurrent phenomena characterizing the functioning of economies with optimizing agents. During this period. Since this book purports to discuss business cycles. unanticipated shocks to money. and inventories. the speciﬁc mechanisms postulated in this literature as leading to business cycles. more generally. We will discuss RBC models further in Chapter 3. the merits of this approach have been a topic of debate. this feature of Kydland and Prescott’s analysis
. The literature on real business cycle (RBC) theory owes its existence to Kydland and Prescott [141]. failed to garner sufﬁcient empirical support. Despite providing great theoretical advances in the analysis of aggregative phenomena. namely. the emphasis was on generating the Phillipscurve type of phenomena between inﬂation and unemployment based on informational frictions.
A more general critique to RBC analysis was mounted by the New Keynesian challenge. endogenous changes in efﬁciency due to increasing returns to scale. Instead. models have been developed that study the role of market completeness/incompleteness on cyclical ﬂuctuations (see. For example. The New Keynesian viewpoint breaks with the RBC approach by contesting the view that prices adjust frictionlessly to clear markets. the observed procyclical movements in productivity should merely be a response to exogenous technology shocks (see Prescott [173]). Hall [110. and they have constituted the topic for much further study. is known to be procyclical. Many of these issues have taken on the character of “puzzles” in the RBC literature. More recently. Fallacy and Fantasy
was viewed as “fantastical” by many. The original RBC framework has also been extended in recent years to examine the business cycle phenomena in emerging market economies. 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. Productivity. A deeper problem lay in modeling the movement of economywide averages (possibly fallaciously) in terms of the behavior of a representative or standin household. the original Kydland–Prescott model could not explain the relative variability of hours and productivity or real wages. hence.4
Business Cycles: Fact. On the one hand. and international capital ﬂows. ﬁnancial diversiﬁcation. Subsequent research that evolved from the original Kydland–Prescott exercise has proceeded along several different dimensions. for example. or the Solow residual. 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. The model lacked money. The RBC literature has also generated international business cycle models to replicate ﬁndings on current account dynamics. markups. international risk sharing. According to the RBC approach. We will examine a few of these directions in later chapters. Other skeptics contested the implications of the RBC approach for the observed behavior of productivity. In a series of papers. Heathcote and Perri [117]). It also failed to account for the correlation between hours and productivity. and variable factor
. it faced the problem of reconciling observations on money–output correlations within a model where the main driving force was real productivity shocks. it introduces alternative mechanisms for generating price stickiness such as imperfect competition among ﬁrms.
This facilitates the analysis of the different effects of government versus technology shocks. Smets and Wouters [194. or monetary versus technology shocks. More recently.
2 Canova [58] is an excellent reference source on the quantitative and empirical analysis of dynamic stochastic general equilibrium models. 195]). We will discuss New Keynesian models in detail in Chapter 5.
. The New Keynesian challenge has proceeded along theoretical lines (see Rotemberg and Woodford [182]) and empirical considerations (see Gali [96] or Basu. for example. vector autoregression (VAR) and structural VAR (SVAR) have also proven to be popular in empirical macroeconomic research. 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].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. SVAR also permits an identiﬁcation of shocks. and Kimball [32]). Fernald. The controversy over the calibration approach also resulted in work on alternative methods for empirically analyzing business cycle phenomena.2 We will discuss the issues involved in matching the model with the data in Chapter 7. dynamic stochastic general equilibrium (DSGE) models have been developed to identify shocks and propagation mechanisms of business cycles models (see.
This page intentionally left blank
.
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. or lagging indicators of the business cycle. This framework is based on the principle of identifying turning points in economic activity and determining which series constitute leading. Mitchell [163]. Mitchell and Burns [164]. we describe the National Bureau of Economic Research (NBER) methodology for dating business cycles and present the basic facts regarding business cycles. coincident. and Osborn [20]. and Stock and Watson [199].1.Chapter 2
Facts
The vast literature on business cycles has focused on generating the stylized facts regarding cyclical ﬂuctuations. among others). the European or euro area business cycle has also been the topic of much recent study (see Artis. Kontolemis. We will also discuss some of the empirical ﬁndings in these regards.
2. and Burns and Mitchell [50] provide a framework to describe the main features of business cycles. 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]. In this chapter. Much of the early work on business cycles was implemented for the US economy. a cycle consists of expansions occurring at about the same time in many
7
. However. Basu and Taylor [31] have examined business cycles in an international historical context.
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. this sequence of changes is recurrent but not periodic.
This deﬁnition has formed the basis of modern thinking about business cycles. and second. ﬁnd the cyclical peaks and troughs in a given set of economic variables. they are not divisible into shorter cycles of similar cycles with amplitudes approximating their own. Some of these questions are precisely the ones that we seek to answer in this book. and also lay down rules for excluding irregular movements and other similar changes. followed by similarly general recessions. and amplitude of each speciﬁc cycle are compared with that of the reference cycle. in duration business cycles vary from one year to ten or twelve years. or secular trends? Finally. the cyclical behavior of each series is then examined relative to the reference cycle. If one talks about “ﬂuctuations in the aggregate economic activity of nations”. Once the reference dates are found. when one considers the statement regarding expansions occurring in “many economic activities”. and Burns and Mitchell [50] is comprised of two mutually reinforcing acts: ﬁrst. Burns and Mitchell [50] themselves noted that this deﬁnition raised as many questions as it sought to answer. timing. As part of this analysis. Mitchell and Burns [164]. determine whether these turning points are sufﬁciently common across the series. and revivals which merge into the expansion phase of the next cycle. In their words (Burns and Mitchell [50]. Fallacy and Fantasy
economic activities. random ﬂuctuations. then an aggregate business cycle or a reference cycle is identiﬁed. the comments regarding the duration and amplitude of business cycles are based on actual observations of cyclical phenomena. contractions. Burns and Mitchell [50] stress that the notion of a reference cycle should not be equated with an observable construct. Ch.8
Business Cycles: Fact. how should we deal with seasonal changes. the duration. 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. p. In seeking to identify “recurrent changes”. If the answer to the latter question is in the afﬁrmative. The NBER approach to identifying business cycles as outlined by Mitchell [163]. 2. 14):
. whether it pertains to the measurement of business cycles or the construction of models of cyclical ﬂuctuations.
business cycles can be seen only ‘in the mind’s eye’.
. a trough occurs at point C for a growth cycle while point D deﬁnes a peak.Facts
9
When we speak of ‘observing’ business cycles we use ﬁgurative language. For. Figure 2. 2. like other concepts. we say that it is
Fig.1 displays the difference between classical and growth cycles. What we literally observe is not a congeries of economic activities rising and falling in unison. When the economy is moving from a trough to a peak.1. but changes in readings taken from many recording instruments of varying reliability. Following the terminology in Zarnowitz [211]. There is an alternative approach which considers the decline in the series measured as a deviation from its longrun trend.
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. One advantage of using growth cycles is that they have expansions and contractions that are approximately of the same duration. Point A deﬁnes a trough for a classical cycle while point B deﬁnes a peak. This is known as a classical business cycle. By contrast.
Classical and Growth Cycles. By contrast. classical cycles typically have recessions that are shorter than expansions because of the growth effect.
at most. The duration of the business cycle is the length of time (in months. An alternative approach is to assume a stochastic trend modeled as a unit root in the series at hand. Nevertheless. One approach is to use a linear detrending procedure which assumes that the underlying series possesses deterministic time trends. in their survey of empirical business cycles.10
Business Cycles: Fact.1 gives the dates of business cycles or the socalled “reference dates” for the US economy since 1857. This period was dubbed as the period of the “Great Moderation”. The amplitude of a business cycle is the deviation from trend. The contribution of Nelson and Plosser [167] was to show that economic time series such as real GDP typically possess unit roots. the average length of an expansion has increased to 57 months from. and a recession is said to occur when the economy is moving from a peak to a trough. and Watson [133] argue. 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. or years) that the economy spends between two troughs or. national income. quarters. equivalently. This committee uses data on real output. However. 38 months preWWII. although a computer algorithm exists that can approximate the results (see Bry and Boschan [49]). employment. Similarly. Stock. As an example. real business cycle (RBC) models which allow for trends in the technology shock imply that growth and business cycles are jointly determined. Table 2. the practice of separating the trend and cyclical component using linear time series methods is well established. Plosser. Post WWII. One of the longest expansions to have occurred postWWII is between March 1991 and November 2001 — a period of 120 months. the average length of a contraction has decreased to ten months from 17 months or more in the years preceding 1945. As King. The turning points are determined judgmentally. and trade at the sectoral and aggregate levels to identify and date business cycles. The dating of business cycles for the US is done formally by the NBER Business Cycle Dating Committee. If one chooses to use growth cycles. two peaks. Stock and
. Fallacy and Fantasy
in an expansion. There are several approaches to detrending economic time series. The shortest cycle is 17 months or nearly six quarters. and the longest cycle is 128 months or more than ten years. there is an issue of how to identify the cyclical component of a given series. this committee recently announced that the US economy had formally been in a recession since December 2007. There are 32 complete cycles in the entire sample period. We can also observe expansionary and contractionary phases of the business cycle from this table.
Table 2. 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
.1. Business Cycle Reference Dates Peak Trough
US Business Cycle Expansions and Contractions.
Fallacy and Fantasy
17 22 18 10
55 48 53 67
. (Continued) Business Cycle Reference Dates Peak Trough Contraction Peak to Trough May 1937(II) February 1945(I) November 1948(IV) July 1953(II) August 1957(III) April 1960(II) December 1969(IV) November 1973(IV) January 1980(I) July 1981(III) July 1990(III) March 2001(I) December 2007 (IV) Average. all cycles: 1854–2001 (32 cycles) 1854–1919 (16 cycles) 1919–1945 (6 cycles) 1945–2001 (10 cycles)
∗ 13 cycles. Source: NBER.12
Table 2. ∗∗ 15 cycles.
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.1.
The parameter µ controls the smoothness of the trend component. (1. in particular.1). If these are trendstationary. µ is typically recorded as 1600. King and Rebelo [130]. For quarterly data. The Hodrick–Prescott ﬁlter deﬁnes gt to solve
∞ ∞
min
gt t=−∞
(yt − gt ) + µ
t=−∞
2
(gt+1 − gt ) − (gt − gt−1 ) . the cyclical component can be represented as ytc = yt − gt = µ(1 − L−1 )2 (1 − L)2 yt . 1 + µ(1 − L−1 )2 (1 − L)2 (1. The ﬁrst approach tends to generate spurious business cycle effects in the detrended series. Eq. is modiﬁed at the beginning and end of the sample. including Singleton [192]. let yt denote the series in question and gt denote the unknown trend component. and Cogley and Nason [69]. if the underlying data are
. that business cycle dynamics obtained by using a HP detrending procedure depend on the properties of the underlying data. The properties of the HP ﬁlter have been studied by many authors. The general equation characterizing the ﬁlter. Several alternative approaches have been proposed in the recent business cycle literature to separate the trend versus cyclical component of a time series. the ﬁlter is applied using a ﬁnite sample. Taking the derivative with respect to gt yields
∞ ∞
µgt+2 − 4µgt+1 + (1 + 6µ)gt − 4µgt−1 + µgt−2 =
t=−∞ t=−∞
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. 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. Cogley and Nason have argued. then the detrending procedure has favorable properties.2)
In practice. Speciﬁcally.1)
Using the lag operator notation and simplifying. whereas the second exacerbates the role of shortterm noise.
2
The cyclical component of the series is deﬁned as ytc = yt − gt .
periodicities between six quarters and eight years. . 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. K . . It turns out that the timedomain representation of the bandpass ﬁlter is an inﬁnite moving average. i.5s)). application of the HP ﬁlter induces spurious business cycle ﬂuctuations. 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. once the ideal ﬁlter’s weights are found in the time domain.
(1. the optimal
. . I (2) or less.14
Business Cycles: Fact. Let s denote the number of observations in a year. In particular. however. where I (·) is the indicator function. However. (1. ak = a−k for k = 1. The ﬁlter is designed to have a number of other properties. Fallacy and Fantasy
differencestationary.5s and 8s. then it has trend elimination properties. The bandpass ﬁlter retains the movements in a series for the frequencies associated with the periodicities of 0. A second approach is based on the spectral analysis of economic time series. The frequency response function of the ideal bandpass ﬁlter is deﬁned as βbp (ω) = I (2π/(8s) ≤ ω ≤ 2π/(0. Hence.e.3)
where the moving average coefﬁcients are chosen to be symmetric. Baxter and King derive the bandpass ﬁlter by considering lowpass and highpass ﬁlters with the required properties. 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). the Baxter–King ﬁlter will eliminate deterministic quadratic trends or render stationary series that are integrated up to order two.4)
where ψ(L) is a (K − 1)order symmetric moving average polynomial. They show that if the sum of the moving average coefﬁcients is zero. . K ak = 0.
01 8. In what follows.
2.01 0. We also observe the cyclical troughs and peaks of economic activity associated with the business cycle. These facts constitute important benchmarks by which to judge the performance of alternative models of business cycles.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.2 50 55 60 65 70 75 80 85 90 95 00 05 BPTEST USA 0.2 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). STYLIZED FACTS
In this section. We consider a measure of the business cycle as the comovement of the cyclical behavior of individual series with
. We can observe the cyclical downturns and upturns corresponding to the NBER business cycle reference dates from this graph. 2.0 7.6 9.00 0. the notion of a “business cycle” is also concerned with the comovement of a large number of economic variables.03
15
50 55 60 65 70 75 80 85 90 95 00 05 BPCYCLEUSA
Fig.02 0. Trend and Cyclical Components in USA GDP.Facts
9.02 8.03 0. we use this approach to generate the cyclical components of the different series. we provide some stylized facts of business cycles. Figure 2. showing the large gains in productivity and growth attained over this period.2.8 8. Nevertheless. Section 2.04 0.04 7.05 0.
approximating ﬁlter with maximum lag K is obtained by truncating the ideal ﬁlter’s weights at lag K . We observe that real GDP displays a marked positive trend over the sample period. The cyclical component tracks very well the US business cycle as identiﬁed by the NBER.4 0.6 0.2 discusses these properties of business cycles.
2 The salient facts of a business cycle can be described as follows: 1. variables that move in the same direction over the cycle as real output are procyclical. Exceptions are production of agricultural goods and natural resources. investment. a leading indicator reaches a peak before real GDP reaches its peak and bottoms out (reaches a trough) before real GDP. In other words. Consumption of
1 Stock and Watson [197] use data on 71 variables to characterize US business cycle phenomena over the period 1953–1996. the interwar era between World War I and II. inventories. Consumption of durables is much more volatile than consumption of nondurable goods and services. 2 Early work on developing a composite index of leading indicators is due to Mitchell and Burns [164]. productivity. We can also examine whether different time series are out of phase with real GDP. the contemporaneous correlation of output in different sectors of the economy is large and positive. Variables that move in the opposite direction (rise during recessions and fall in expansions) are countercyclical. and monetary aggregates. real wages. 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. The stylized facts of business cycles are typically deﬁned in terms of the behavior of the main components of GDP. According to this approach. and imports are all strongly procyclical. Leading indicators are useful for predicting subsequent changes in real GDP. lagging indicators reach a peak or trough after real GDP.16
Business Cycles: Fact. Coincident indicators reach a peak or a trough at roughly the same time as real GDP. asset returns and prices. and separate their sample period into the preWorld War I era. and the crosscorrelations of the cyclical component of GDP with the cyclical component of each series for lags and leads up to six periods. Finally. These have been described in a number of economic studies. hours. which are not especially procyclical. For example. Fallacy and Fantasy
the cyclical component of real output. Consumption. and passed over to the Conference Board in 1995. Variables that display little correlation with output over the cycle are called acyclical. They make use of the Baxter–King bandpass ﬁlter to identify the trend versus cyclical components of each series.1 Information on leading indicators is also collected by the Conference Board. Such a composite index was made ofﬁcial by the US Department of Commerce in 1968. Real output across virtually all sectors of the economy moves together. and the postWorld War II era. Among other measures. 2.
. they consider the comovement of the cyclical component of GDP with the cyclical component of each series.
whereas nondurables ﬂuctuate considerably less. Productivity is slightly procyclical. but both real wages and productivity vary considerably less than output. an inverted yield curve.
12. 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. while capacity utilization tends to be coincident. or lagging indicators. while average weekly hours ﬂuctuate much less. tends to shrink during expansions and increase during recessions. 13. They have not displayed a steady pattern in terms of variability to GDP or in terms of leading. Net exports are countercyclical. The risk premium for holding private debt. coincident. Since imports are more strongly procyclical than exports.
. Government spending tends to be acyclical. 10. employee hours. 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. Proﬁts are highly volatile. Real wages are procyclical or acyclical. That is. The employment series lags the business cycle by a quarter. or the yield spread between corporate paper and Treasury bills with six months’ maturity. and capacity utilization are all strongly procyclical. 11. Employment ﬂuctuates almost as much as output and total hours of work. the trade balance tends to be countercyclical. Nominal interest rates tend to be procyclical. The reason for this countercyclical behavior is likely to be changes in default risk. namely. Velocity and the money supply are procyclical. Total employment. Investment in equipment and nonresidential structures is procyclical with a lag.
durable goods ﬂuctuates more than GDP. 5.
6.
9. The correlation between government expenditures and output is nearly zero.Facts
17
3. 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. The correlation with output is generally negative.
8. Investment in residential structures is procyclical and highly volatile. 4. but weakly so.
7.
In this sense. Witness the impact of the large oil shocks in the 1970s. Since the early 1980s. hence.18
Business Cycles: Fact. although it is important to emphasize that the market has fallen without a subsequent recession. The standard deviation of inﬂation is lower than that of real GDP. contemporaneous correlations between output ﬂuctuations in different countries were highest in the interwar period. In the preWWI period and interwar period. inﬂation appears to be countercyclical. By contrast. the procyclicality of M2 has diminished since the 1980s. a topic to which we will return. the stock market fell in the quarter before each of the eight recessions. reﬂecting the common experience of the Great Depression. In the postWWII era. This set of facts constitutes a benchmark which a successful business cycle model should meet. 16. Inﬂation is a coincident indicator. 19. for example. prices and inﬂation were procyclical in the preWWII era. Other papers have generated business cycle facts using data over long samples. 17. Finally. The stock market is positively related to the subsequent growth rate of real GDP. A similar change appears to have characterized the behavior of pricelevel ﬂuctuations. Fallacy and Fantasy
14. 18. employment. and productivity have proved among the most difﬁcult to reconcile by current business cycle theories. changes in stock prices have been taken as providing information about the future course of the real economy. However. Chadha and Nolan [62] identify the stylized facts
. One way to rationalize this phenomenon may be in terms of monetary policy that is more accommodative — under a gold standard. the increased persistence of inﬂation and the countercyclical behavior of prices observed in this era. real wages. inﬂation was procyclical with a very low mean. The changes in the cyclical behavior of prices and inﬂation also have implications for the socalled impulses and propagation mechanisms of business cycles. 15. There is also a marked increase in the persistence of inﬂation after WWII. The correlation is typically larger in the postwar period than in the prewar period. with the exception of Germany and Japan. Money (M2) is procyclical and tends to be a leading indicator of output. there is more evidence of supplysidedriven ﬂuctuations in output. Between 1945 and 1980. 20. The behavior of prices and inﬂation appears to have changed over time. The ﬁndings concerning the behavior of hours.
The countries selected are
. Artis and Zhang [18. 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. While her ﬁndings have garnered much interest.3. 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. Stock and Watson [198] seek to avoid the problem of poor data quality in the preWWII 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. and another 15% to reduced shocks to food and commodity prices. Kontolemis. If the higherquality data in the postWWII period are subjected to a transformation that makes the pre. 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. and Osborn [20] propose business cycle turning points for a number of countries based on industrial production. 15% to smaller shocks to productivity. Their ﬁndings suggest the emergence of a groupspeciﬁc European business cycle since the formation of the ERM. Christina Romer [180.and postWWII data of comparable quality. Their ﬁndings also provide evidence on the factors behind the “Great Moderation”. They ﬁnd that the standard deviations of a typical variable in their sample declined by about 20–40% since the 1980s. They suggest that if one considers additional evidence based on a larger sample of countries over the different periods. then one discerns no change in output ﬂuctuations across the two periods. which is independent of the US cycle.Facts
19
of business cycles for the UK economy over a long sample period. the variability of output appears to have diminished in the postWWII period. 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. They attribute around 20–30% of the reduction in output volatility to better monetary policy. 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.
2.
Whereas in the postWWII era the euro area countries exhibited high rates of growth. Pt peak RCt recession continuation . and Proietti [21] discuss alternative approaches to dating euro area business cycles.3 To describe this algorithm. suppose that the economy can be in one of two mutually exclusive states. Tt trough
Let pEP = Pr(Pt+1 ECt ) denote the probability of transiting to a peak.20
Business Cycles: Fact. Artis. The Markov process distinguishes between turning points within the two states by assuming that Et = Rt = ECt expansion continuation . and a trough terminates a recession. Suppose that a peak terminates an expansion. conditional on being in an expansion. Let St denote a discrete random variable that follows a ﬁrstorder Markov process
3This algorithm follows Harding and Pagan [115]. implying that classical cycles should also be considered. making growth cycles the relevant concept. which implies that the probability of continuing an expansion pEE = Pr(ECt+1 ECt ) is given by pEE = 1 − pEP . They also consider the lead/lag relationships between countries at peaks and troughs. expansion Et or recession Rt . The euro area experience makes consideration of both types of business cycles relevant. deﬁne pRT = Pr(Tt+1 RCt ) as the probability of transiting to a trough.
. in recent years growth has slowed and even absolute declines in real GDP have been observed. Marcellino. who extended the Bry and Boschan [49] algorithm to a quarterly setting. Fallacy and Fantasy
the G7 countries along with the prominent European countries. As in the Stock and Watson [197] approach. the probability of continuing a contraction pRR = Pr(RCt+1 RCt ) is pRR = 1 − pRT . These authors also articulate a formal model of turning points based on restrictions imposed on a Markov process. Likewise. and to determine whether cyclical movements are similar across different countries. conditional on being in a contraction. They use this information to examine the international nature of cyclical movements. they distinguish between classical and growth cycles. Then.
and a recession termination sequence. (St−4 . as follows: ETSt = {( yt+1 < 0) ∪ ( RTSt = {( yt+1 > 0) ∪ (
2 2
yt+2 < 0)} yt+2 > 0)}. Pt+1 . we can deﬁne an expansion termination sequence. The duration of a phase is required to be two quarters. St = RCt . St−1 .
Thus. features a contractionary state at time t. Tt ) belong to a contraction. St−3 . which deﬁnes a peak. pEP . St−3 . Likewise. A similar condition exists for troughs. The algorithm is completed by ﬁnding the probability that the economy will transit to a peak. conditional on having been in a contractionary phase.
. Else. peaktopeak or troughtotrough. Pt ) both belong to an expansion and (RCt . which deﬁnes a trough. St ). Pt−4 . Now consider applying this algorithm to dating classical cycles. Then. can only be followed by a peak at t + 1. yt+1 = yt+1 − yt > 0 and 2 yt+2 = yt+2 − yt > 0 represent the candidate sequence for terminating a contraction. conditional on having been in an expansionary phase.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. if ETSt is false. respectively. Hence. the expansion continues. St ). and that ETSt is true. features an expansionary state at time t. Likewise. St = ECt . yt+1 = yt+1 − yt < 0 and 2 yt+2 = yt+2 − yt < 0 represent the candidate sequence for terminating an expansion. St−1 . and on complete cycles. The minimum duration of a complete cycle is given as ﬁve quarters. St−2 . This is the joint event that the history at time t. St−2 . which is automatically satisﬁed since the states (ECt . This is the joint event that the history at time t. expansion or contraction. the probability that the economy will transit to a trough. RTSt . Let yt denote the underlying series. ETSt . a peak at date t − 4. (St−4 . is pRT .
Fallacy and Fantasy
and that RTSt is true. 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. The turning points determined by the CEPR Business Cycle Dating Committee are also indicated in this table. the contraction continues.22
Business Cycles: Fact. Euro Area Business Cycle Turning Points. 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. 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. and Proietti [21]. [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. if RTSt is false.2.2 shows the business cycle turning points for both classical and deviation or growth cycles obtained using this approach. Finally. The Centre for Economic Policy Research (CEPR) has recently formed a committee to set the dates of the euro area business cycle. Artis et al.
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. Henry. and Mestre [85].
. Marcellino. Otherwise.
Table 2.4 Table 2. [21]. and a trough cannot occur if output is above it. 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. We note that these are similar to the classical cycles identiﬁed by Artis et al.
They ﬁnd that the volatility of business cycles has moderated in most G7 countries over the past 40 years.00 13.6 0. France. and Canada) over the period 1960– 2002. UK. which is speciﬁed in terms of the growth rates of quarterly GDP for the G7 countries. they use a socalled factorstructural vector autoregression (FSVAR). and Terrones [139] and others. Stock and Watson [199] provide a comprehensive analysis of the volatility and persistence of business cycles in G7 countries (deﬁned to include the US. [20].Facts
14. Germany. 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.
Fig.03 0.03 1970
23
14.01 14. in sync with Artis et al. Prasad.01 13.3. Let Yt denote the n × 1 vector of
.2
13. Japan. the eurozone countries and Englishspeaking countries (including Canada. However. they ﬁnd evidence on the emergence of two cyclically coherent groups.4 1970
1975
1980
1985
1990
1995
2000
2005
1975
1980
1985
1990
1995
2000
2005
BPTRENDEURO
EUROAREA
BPCYCLEEURO
Fig.8 0.02 0. they ﬁnd no evidence for closer international synchronization over their period of study. Italy. 2. This is similar to the ﬁndings of Köse. 2. Stock and Watson [199] also seek to provide evidence on the sources of the changes. Trend and Cyclical Components in Euro Area GDP. First. They base their results on various measures of correlation of GDP growth across countries. namely. the UK. and the US).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).02 0.0 0. They also provide evidence on the synchronization of international business cycles. This is a standard structural vector autoregression (VAR) with an unobserved factor structure imposed on the VAR innovations.4 0.
domestic shocks explain almost all of the variance for Japan. reﬂecting the impact of the tenyearlong deﬂationary episode for this country. . The standard VAR model is given by Yt = A(L)Yt−1 + νt . These may arise from the role of international trade. . and the least for Italy and Japan. and (iii) the spillover effects of the domestic shocks on the hstepahead forecast error for each country.5)
where ft is a k × 1 vector with k ≤ n. . their relative variance varies by country and by time period. Stock and Watson [199] ﬁnd evidence for two common shocks. Fallacy and Fantasy
real GDP growth for the G7 countries considered in this study. σνk ). France. . The model allows for spillover effects to occur through the lagged effect of an idiosyncratic shock. Second. Notice that the common shocks affect the output of multiple countries contemporaneously. They consider two sample periods: 1960–1983 and 1984–2002. σfk ) and E (νt νt ) = diag (σν1 . and Germany. and the elements of ft denote the common international shocks. This provides the identiﬁcation scheme to help identify the common international shocks. they ﬁnd that most of the variance of GDP growth can be attributed to common and idiosyncratic domestic shocks. The covariance structure of the shocks is given by E (ft ft ) = diag (σf 1 . They also provide a variance decomposition for the impact of (i) the common international shocks. whereas the idiosyncratic shocks affect them with a lag. . for example. the elements of t denote the countryspeciﬁc idiosyncratic shocks. where the vector of reducedform errors vt has the factor structure ν t = ft + t . In this representation. the role of international sources of ﬂuctuations arising from common shocks or from
. . (3. First. However. In the ﬁrst period.6) (3. the impact of international shocks is estimated to be greatest for countries such as Canada. . (ii) the domestic shocks. where the elements of ft and t are assumed to be mutually independent.24
Business Cycles: Fact. . In the second period.
. region . Let yi.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. Developed Asia. they also show that the decline in overall volatility of GDP growth for countries such as the US. t = 1. 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.t = ai + biworld ft world + bi
region region fr.t country
+ bi
country country fn. see. Finally. and Raynauld [109] or Lumsdaine and Prasad [154].t denote the observable variables for i = 1.4. They argue that many recent studies of international business cycles focus on a subset of countries — not the world. and also of the failure of business cycles to become more synchronized despite the great increase in trade ﬂows over this period. . • R regional factors ( fr Africa. 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. . • the single world factor ( f world ). and the UK. . IS THERE A WORLD BUSINESS CYCLE?
Köse. However. . Canada. they also ﬁnd that shocks have become more persistent in countries such as Canada. consumption. M the number of time series per country. Let N denote the number of countries. and investment) across all countries and regions as well as across countries and aggregates separately. Latin America. for example. N × M . Germany. and T the number of time periods. France.
. one per each country). .t . . Otrok. Gregory.7)
5 For recent studies. Europe.
(4. or Oceania). T . The behavior of each observable variable is related to the factors as follows: yi.t
+
i. and Italy. . This is the source of the moderation of individual business cycles. and the UK is due to the decline in the variance arising from international shocks.Facts
25
spillovers appears to have increased for the US. Head.
2. There are three types of factors: • N countryspeciﬁc factors ( fn . and Whiteman [138] consider the issue of a world business cycle. where r stands for North America. Developing Asia.
t−2
+ · · · + φi.t = Then.
(4. one approach is to use a Kalman ﬁltering algorithm together with classical statistical techniques to estimate the model’s parameters. k = 1. K .t−2
+ · · · + φfk . an alternative is to use Bayesian estimation techniques. and show how much of the variation in yi. i. In what follows.t ) = σf2 .2
i. Finally.
. scales are identiﬁed by assuming that each σf2 is equal to a constant. Thus. There are k several ways to estimate this model.t . Likewise. Since the common factors are unobserved.t ufk . One identiﬁcation device is to assume that the factor loading on one of the factors is positive.
(4. conditional on the data.t . and defer a discussion of the estimation techniques to Chapter 7. .t is due to the common factors. all the covariation among the observable series yi. .10)
where E (ufk .pi
i. .9)
= φfk . and the country factors are identiﬁed by positive factor loadings on the output of each country. are mean zero.t
= φi.
(4. Let fk. and zero otherwise. The innovations k ui.t . the model assumes that the behavior of M × N time series can be explained by N + R + 1 factors. which themselves may be serially correlated. where N + R + 1 < MN . . s. This procedure yields a joint posterior distribution for the unobserved factors and the unknown parameters.t−1
+ φi. For the world factor. for the regional factor. i = 1.t . the sign of the factor loading for US output is considered positive. . .t fk .t ) = 0 for i = j.t ui.t−s ) = 0 for all k. we cannot identify uniquely the sign or scale of the factors or the factor loadings.t−s ) = σi2 for i = j and s = 0. we report results based on the median of the posterior distribution for the factors. and ufk . Thus. respectively.
fk .t can be explained by each factor.2
fk . and to follow autoregressions of orders pi and qk .8)
where E (ui. Fallacy and Fantasy
j
where E ( i.1
i. contemporaneously uncorrelated random variables. The coefﬁcients bi denote the factor loadings.qk
fk .t−pi
+ ui. Both the idiosyncratic errors and the factors are allowed to be serially correlated. the sign of the factor loading corresponding to North America is considered positive. . M × N .t−qk
+ ufk . E (ufk .t j. .t ui. as
i.1
fk . Since the factors are unobservable.26
Business Cycles: Fact.t−1
+ φfk .t .
the recession of the 1980s associated with the debt crisis in developing countries and the tight monetary policies in the major developed countries. if anything. The paper also allows for a simple decomposition of variance attributed to the different factors. Var(yi. The authors also document some noteworthy ﬁndings regarding the sources of volatility of investment growth.11)
country 2 country ) Var(fn. they ﬁnd that country and idiosyncratic factors account for a much larger share of variability in investment growth than the world and regional factors. idiosyncratic
. 9% of consumption growth. country factors account for a much larger fraction of consumption growth than output growth. They interpret this ﬁnding as evidence for a world business cycle. ﬁrst. which is discussed in detail in Chapter 4.t ). Recall that the world. we have that Var(yi. By contrast.t ) + Var( i. the world factor based on a large set of developed and developing countries implies that the recession of the 1980s was. + (bi
Thus. This ﬁnding has been documented further in the international business cycle literature. the regional and country factors explain covariation or comovement that is less persistent. as severe as the recession of the mid1970s.t ) They ﬁnd evidence that. Using the representation of the model. Second. the world factor explains nearly 15% of variation in output growth.Facts
27
Köse et al. However. the world factor is more successful in explaining economic activity in developed relative to developing countries. A second important ﬁnding from the paper is that most of the persistent comovement across countries and aggregates is captured by the world factor. the fraction of variance attributed to the world factor can be expressed as (biworld )2 Var(ft world ) .t ) = (biworld )2 Var(ft world ) + (bi
region 2
) Var(fr.t
region
) (4. Third. and the downturn and recession of the early 1990s. and 7% of investment growth. Second. Speciﬁcally. regional. and country factors are modeled as autoregressive processes which may display substantial persistence depending on the estimated values of the parameters.’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. in contrast to models estimated using a smaller sample of countries.
prices. Paralleling this ﬁnding. Demers. and other European countries plus Argentina for the period since 1870.
6 For recent reviews. the dynamic factor model suffers from the shortcoming that all covariation is attributed to the common factors. and Altug [79]. the UK. It featured ﬁxed exchange rates and worldwide capital market integration. The last two ﬁndings of the paper imply that regional factors play a minor role in aggregate output ﬂuctuations.5. much work on investment behavior shows that irreversible investment decisions are particularly susceptible to risk and uncertainty. there is little evidence that the volatility of European aggregates can be attributed to the European regional factor. Yet this ﬁnding could also be due to model misspeciﬁcation. The authors pose this as a puzzle. They consider the time series behavior of output. 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. Demers. Fallacy and Fantasy
shocks explain a much larger fraction of the volatility in investment growth in developing countries relative to developed countries. disruptions to market access. They divide this period into four periods that also reﬂect the monetary and capital account regimes prevailing in them.
2. See.
.6 Altug. HISTORICAL BUSINESS CYCLES
Basu and Taylor [31] examine business cycles within an international historical perspective. exchange rates. especially in an era of increased trade ﬂows and ﬁnancial integration.28
Business Cycles: Fact. • 1870–1941: This era corresponded to the classic gold standard. and the current account for 15 countries including the US. Rodrik [178] emphasizes the importance of political risk (in his case. total consumption.7 We conjecture that variability in investment due to such factors is a key channel for inducing idiosyncratic volatility in developing economies’ investment behavior. Yet. This last result is interpreted as evidence against the existence of a European business cycle. see Caballero [53] or Demers. 7 In a related literature. the risk of policy reversal) for irreversible investment decisions. Serven and Solimano [189]. investment. there is also the possibility that some observed covariation could result from the spillover effects of idiosyncratic or country shocks. for example. and debt and currency crises. However. As Stock and Watson [199] note. real wages. 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.
the UK. Sweden. Bordo and Heibling [44] use real GDP or industrial production series to determine synchronization. Germany. For example.
. France. and the US. Canada. including the era of the Great Depression. they use a statistical measure of synchronization developed by Harding and Pagan [115]. They examine business cycles in 16 countries during the period 1880–2001.versus supplyside factors or shocks and the role of alternative propagation mechanisms such as price rigidity. Portugal. The countries that they examine are Australia. which examines whether the turning points in the different series occur at similar dates. most explanations of the Great Depression attribute the source of this massive downturn. It also provides a way to identify the importance of demand. Spain. which corresponded to the postWorld War II era of reconstruction and a resumption of global trade and capital ﬂows. Consideration of alternative historical periods. Finland. Basu and Taylor [31] also argue that their periodization allows for an analysis of international capital ﬂows and capital mobility in affecting cyclical ﬂuctuations. Instead of using information on NBERtype reference dates. Early 1970s–present: This period features a ﬂoating exchange regime and a period of increasing globalization.” Among other concepts. Switzerland. Bordo and Heibling [44] ask whether national business cycles have become more synchronized. Japan. Norway. the world economy shifted from a globalized regime to an autarkic regime. the Netherlands. They also use standard output correlations and factorbased measures.
Basu and Taylor [31] argue that considering such a breakdown allows for the analysis of the impact of different regimes on cyclical phenomena. thus allows for an examination of such mechanisms as nominal rigidities in propagating shocks. Denmark. the socalled concordance correlation. 1945–1971: The third regime is the Bretton Woods era. and consider the four exchange rate regimes also examined by Basu and Taylor [31].Facts
29
•
•
•
1919–1939: During this period. Synchronization refers to the notion that “the timing and magnitudes of major changes in economic activity appear increasingly similar. which simultaneously occurred in a number of countries. to monetary phenomena. Italy. This period also corresponded to the Great Depression.
12)
¯ Let Si = (1/T ) T Sit denote the estimated probability of being in an t=1 expansion. Fallacy and Fantasy
To brieﬂy describe the concordance correlations. A simple measure of concordance is deﬁned by the variable 1 Iij = T
T
[Sit Sjt − (1 − Sit )(1 − Sjt )]. In the interwar and postBretton Woods era. They argue that Sit = 1 for most countries during the Bretton Woods era of 1948–1972 and hence leave this period out. The key ﬁnding is that during the classical gold standard. we divide Iij by a consistent estimate of its standard error under the null hypothesis of independence.30
Business Cycles: Fact. then the standardized index equals −1. ¯ ¯ ¯ ¯ the expected value of the concordance index is 2Si Sj + 1 − Si − Sj . and if the two cycles are unrelated. Note ∗ that the variance of Iij is given by ∗ Var(Iij )
4 = 2E T
T t=1
¯ ¯ (Sit − Si )(Sjt − Sj ) . Subtracting this from Iij yields the meancorrected concordance index:
∗ Iij
1 =2 T
T t=1
¯ ¯ (Sit − Si )(Sjt − Sj ). on average. then under the assumption that Sit and Sjt are independent.
(5. They deﬁne a recession as one or more consecutive years of real GDP growth. Bordo and Heibling [44] calculate the business cycle indicators Sit for each of the four exchange rate regimes. let Sit and Sjt denote binarycycle indicator variables which assume a value of 1 if the economy is in an expansion. 0 otherwise.13)
∗ To derive the concordance correlation coefﬁcient.
t=1
(5. then the standardized index equals 1. cycles were.
If the two cycles are perfectly synchronized (so that Sit = Sjt for all t). For the gold standard era. say. if they are in different states in each period (so that Sit = 1 − Sjt ). uncorrelated
. 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. while an expansion is deﬁned as one or more years of positive GDP growth. the standardized index equals 0. more than half of all national business cycles become positively related to each other. for countries i and j.
and spillover effects in generating this result. there are no dynamics in the relation between output growth rates and the factors. and that the increasing importance of global shocks reﬂects the forces of globalization. but this was not so for ﬁnancial or twin crises. In the former. In this model.
. Typically. However. before 1914 and after 1971.Facts
31
with each other. which measure the magnitude as well as the direction of output changes. They consider a socalled center country version of this model. whereas beginning with the interwar period. In the trade linkages version. lagged GDP growth of the major trading partner is included in place of the center country’s. They also ﬁnd higher output correlations for the core European countries (the EEC) and the Continental European countries. more positive bilateral output correlations by era. They ﬁnd that both global (common) shocks and transmission have become more important. Eichengreen and Bordo [84] examine the nature of crises in two periods of globalization. The authors also estimate a socalled static approximate factor model for the growth rates of GDP. especially the integration of goods and services through international trade and the integration of ﬁnancial markets. idiosyncratic shocks.” 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 an increase in correlations for the AngloSaxon countries. they started becoming synchronized with each other. 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. as in Stock and Watson [199]. and a trade linkages version. suggesting that it is not transmission from the center country that accounts for the increased role of transmission. We discuss emerging market business cycles in detail in Chapter 6. and the idiosyncratic shocks are allowed to be serially correlated and heteroscedastic. They ﬁnd that there has been a tendency for higher. the importance of transmission for peripheral countries arises only in the trade model. The authors also examine standard output correlations. Bordo and Heibling [44] conclude by noting that global shocks are the dominant inﬂuence across all regimes. Finally. lagged GDP growth of the center country is included alongside the lagged value of the own country’s GDP growth in the VAR representation. They argue that banking crises were less severe in the period before 1915. such crises have ﬁgured importantly in emerging market output ﬂuctuations.
This page intentionally left blank
.
perfectly competitive. The Great
33
. as Keynes [127] did. one could argue. Equivalently. wars. the result is unemployment and greater declines in output relative to a situation with ﬂexible prices. have ﬁgured prominently in the recent literature. This approach emphasizes the role of intertemporal substitution motives in propagating shocks. and other disruptions in market activity have also been acknowledged to play a role. more precisely. if real wages do not adjust downward when there is a negative shock to demand. Keynesian and New Keynesian models stress the role of frictions such as price stickiness. complete markets in which there are real or technology shocks. By contrast. In a simple labor market model. More controversially. 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. Weather shocks have always had a place among the impulses thought to trigger ﬂuctuations in economic activity. Technology shocks which alter a society’s production possibilities frontier. whether they are permanent or transitory such as oil shocks.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. that investors’ “animal spirits” or. one could also assign a role to taste shocks or changes in preferences of consumers. Current real business cycle (RBC) theory argues that business cycles can arise in frictionless. Political shocks. The shocks or impulses thought to instigate business cycles have typically been varied and diverse. changes in their subjective beliefs could help to trigger business cycles.
173). and international trade.2
3. home production. In this case. The standard RBC model also assumes that there is no government. There is a representative consumer who derives utility from consumption and leisure. Bernanke and Gertler [40] or Kiyotaki and Moore [134]. All markets are perfectly competitive. we discuss an international RBC model to describe how the transmission mechanisms proposed by this approach translate into an openeconomy setting. 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.34
Business Cycles: Fact. We will discuss these extensions in the following sections. The prototypical RBC model has the structure of a standard neoclassical growth model with a labor/leisure choice incorporated. and a constantreturnstoscale (CRTS) production technology for producing the single good using labor and capital. 2 Rebelo [177] provides a recent review of alternative models and directions associated with the RBC agenda. Some of the extensions of the RBC approach have modiﬁed the basic framework to incorporate the role of government. we discuss variants of the RBC model. all factors are fully employed. 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. 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].1. 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. for example. A number of these assumptions have come under criticism in the recent literature.
1 See.1 In this chapter. p.
. that households consume only goods produced in the market. but his discussion is abbreviated to satisfy the constraints imposed by a journal article. In the following chapter. it seems imperative to discuss the main propagation mechanisms of the RBC model. and that there is no trade with the rest of the world. Since the debate has revolved around the efﬁcacy of technology shocks and intertemporal substitution effects in replicating business cycles. and all prices adjust instantaneously to clear markets.
2)
Hence.3 Assuming a solution exists.
(1.3)
Assume that the technology shock follows an AR(1) process:
t+1 .1)
where 0 < β < 1 is the discount factor. The existence of a solution can be shown using standard recursive methods for analyzing such problems. zt+1 = µ+ρzt + 0 < α < 1.
t
∼ i. 0 < θ < 1. we will suppose that the utility function has the form U (ct .5)
where it denotes economywide investment and 0 < δ < 1 is the depreciation rate. lt ) . (1.
(1.6)
Notice that the only shock in this model is the technology shock. For ease of exposition.Models of Business Cycles
35
Turning to the basic model.
3 See. which is assumed to be stationary but persistent as long as ρ = 1.4)
Capital evolves according to kt+1 = (1 − δ)kt + it .
.d.
0 < ρ < 1. We will discuss the propagation mechanism when describing the solution for the model.
The time constraint requires that the sum of leisure and labor hours equals 1: lt + ht = 1. and N (0.. The aggregate feasibility constraint is deﬁned as ct + it = yt . The problem described above is an inﬁnitehorizon dynamic stochastic optimization problem. There is a representative ﬁrm with CRTS production that is affected by a stochastic technology shock each period: yt = exp (zt )ktα ht1−α .i. lt ) = ctθ lt1−θ . (1. for example. time spent not working is taken as leisure. Altug and Labadie [6] or Stokey and Lucas with Prescott [201]. σ 2 ). (1. the representative agent has timeseparable preferences over consumption and leisure choices given by
∞
U = E0
t=0
βt u(ct . (1.
the ﬁrstorder conditions with respect to ct . There is no unemployment in the model as time not spent working is (optimally) taken as leisure. respectively.t+1 [exp (zt+1 )α(kt+1 /ht+1 )α−1 + (1 − δ)] = 1. zt+1 ) = λt . Notice that this is just a simple Robinson Crusoe economy in which the Crusoe produces output using capital and labor.10) (1.12) (1. lt + ht = 1. ¯ it = 0. zt+1 )}
ct .36
Business Cycles: Fact. kt /ht . which he then consumes. Suppose for the moment that there is no investment in the model. U1t (1.lt . Substituting for lt = 1−ht using the time constraint. lt ) = λt .kt+1
(1. Let λt denote the Lagrange multiplier on the resource constraint.8) (1.13)
The ﬁrst equation shows that the marginal rate of substitution between consumption and leisure is equal to the marginal product of labor.7)
subject to ct + kt+1 = exp (zt )ktα ht1−α + (1 − δ)kt . βEt Vk (kt+1 . These conditions can be rewritten as U2t = exp (zt )(1 − α)(kt /ht )α U1t βEt U1. To analyze the optimal consumptionleisure
. kt = k. lt ) = λt exp (zt )(1 − α)ktα ht−α . his choice is between how much to work and how much time to spend in leisure. that is. and ct = yt . lt ) + βEt V (kt+1 . The second equation is the intertemporal Euler equation. Fallacy and Fantasy
let the value function associated with the problem be expressed as V (kt . Vk (kt . Observe that the ratio U2 /U1 is a function of the capitallabor ratio.9) (1. Thus. zt ) = λt exp (zt )αktα−1 ht1−α + (1 − δ) . zt ) = max {U (ct .11)
where U1t and U2t denote the marginal utility of consumption and leisure. U2 (ct . This means that the capital accumulation process will also be a function of the capitallabor ratio. kt+1 and lt and the envelope condition are U1 (ct . (1.
the income and substitution effects of an increase in productivity on hours worked cancel each other out.
(1. Figure 3. lt+i ) = θ(ct+i /lt+i )θ−1 for i ≥ 0 and using the result in equation (1. we have the intertemporal Euler equation characterizing the behavior of the optimal capital stock over time. 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
.14). However.16)
Now a temporary negative technology shock also affects investment or saving. then Crusoe will work less in response to a temporary negative technology shock and will also consume and produce less. then Crusoe spends 50% of his time working and 50% taking leisure. Now suppose output can be consumed or invested so that it = 0. which under our preference speciﬁcation becomes (1 − θ)ct = exp (zt )(1 − α)(kt /ht )α . for example. we consider equation (1.Models of Business Cycles
37
choice. Crusoe will typically respond to a temporary negative shock by lowering savings and also by working less.14). (1.14) 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 .12). for this preference speciﬁcation. we note that if the substitution effect of an increase in productivity dominates the income effect. Using our preference speciﬁcation.15)
Thus. Notice that the technology shock does not affect optimal hours worked because. we can rewrite the intertemporal Euler equation by noting that U1 (ct+i .
. Then. θ(1 − ht ) We can solve this equation for the optimal labor hours ht∗ as ht∗ = θ(1 − α) . 1 − θ + θα (1. if θ = 2/3 and α = 1/2. Thus.1 describes the representative agent’s optimum. the impact of these changes is to induce ﬂuctuations in labor supply and employment as well as investment. in addition to equation (1.
3) and then take the ﬁrst differences: ln (yt+1 ) = zt+1 + α ln (kt+1 ) + (1 − α) ln (ht+1 ). 3. it will have a lower capital stock in the future. 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 ). implying that the effect of the shock persists. 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 .
.
Robinson Crusoe’s Optimum. if the economy experiences a period with lower investment. Fallacy and Fantasy
Fig.1. skt + sht = 1. 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.
Furthermore. where sht = wt ht /pt yt denotes the share of labor in national income.
To derive an observable measure of the Solow residual. Under constant returns to scale.38
Business Cycles: Fact. take the logarithm of both sides of equation (1. Under these assumptions.
Similarly. Step 1: Compute the deterministic steady state for the model. Using this fact. according to Denison [80]. Since the utility function is strictly increasing.19)
0 ≤ ht ≤ 1. Letting u(ct . We assume that capital depreciates at the rate 0 < δ < 1 and the technology shock follows the process in (1.2. 0 < θ < 1. A NUMERICAL SOLUTION
We now provide a numerical solution to further illustrate the properties of the standard RBC model. 1 − h).17) (2. the problem now becomes
∞ {it . zt+1 = µ + ρzt + ct ≥ 0.
t+1 . and convert the original nonlinear dynamic optimization procedure to one which has a quadratic objective and linear constraints (see also Christiano [64]). 1 − ht ) = θ ln (ct ) + (1 − θ) ln (1 − ht ). This is obtained by setting the exogenous technology shock equal to its mean value and evaluating the conditions (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 . Notice that in the RBC framework the Solow residual is typically treated as exogenous. 40% of GNP growth in the USA between 1929 and 1957 resulted from technical development. consumption plus investment equals output at the 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. Critics of the RBC approach argue that there are endogenous components to the ﬂuctuations in the Solow residual. 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]. we can substitute for consumption in the utility function to obtain u(ct ) = u( exp (zt )ktα ht1−α − it .18) (2.
The quadratic approximation procedure is implemented as follows.
3.16) together with the feasibility
. rendering the RBC interpretation invalid.14) and (1.
(2.4).
h (1 − θ) + θ(1 − α) − (1 − θ)δ r +δ or ¯ h= θ(1 − α)(r + δ) . we obtain ¯ ¯ ¯ k ¯ (1 − θ)h 1 − δ y ¯ Now note that ¯ ¯ ¯ ¯ k (k/h)1−α k α = = = .
(2. ¯ θ 1−h c ¯ (2.20)
Next. To ﬁnd an explicit expression for h parameters. and ht . investment is equal to depreciation: ¯ ¯ i = δk. we obtain ¯ y ¯ 1−θ h = (1 − α) .22) as ¯c ¯ y (1 − θ)h¯ = (1 − α)θ(1 − h)¯ .22) (2. Rearranging.40
Business Cycles: Fact. Using (1. ¯ ¯ 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).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.16). note that in the deterministic steady state. we can solve for the steadystate capitallabor ratio as ¯ exp (¯ ) z k = α ¯ r +δ h
1/(1−α)
. use the relation in (2.23) ¯ + θ(1 − α)h = θ(1 − α).14). substituting for c ﬁrst and then dividing through by y . Let z = µ/(1 − ρ) denote the ¯ unconditional mean for the (log of ) technology process. kt . Fallacy and Fantasy
conditions at constant values for ct . (r + δ)(1 − θα) − (1 − θ)δα (2.
. Simplifying the result in (1.
4 For this purpose. h). Thus. h) and ¯ ≡ (0. ht − h) . s ¯ ¯ Deﬁne the vector of state variables as xt = (1. zt − z ) and the ¯ ¯ vector of control variables as ut = (it − i. z . ∂s∂s
where all partial derivatives are evaluated at ¯. Notice that the utility function can be written as s u(s) = (s − ¯) T (s − ¯). See Campbell [55]. k. let s ≡ (1. We can write the law of motion for the state variables as 0 1 0 0 000 1 1 0 kt+1 = 0 (1 − δ) 0 kt + 0 1 0 it + 0 . s + (s − ¯) s ∂s 2 ∂s∂s
Deﬁne e as the 5 × 1 vector with a 1 in the ﬁrst row and zeros elsewhere. zt+1 zt 0 µ 0 ρ 000 t+1 s s Notice that s − ¯ = (x. u) . i. 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 .Models of Business Cycles
41
Step 2: Approximate the original utility function by a quadratic function around the deterministic steady state.
. 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 − ¯). u
4 An alternative approach is to loglinearize the model. kt − k. z. 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) . i. k.
Such problems can be solved using the methods for solving dynamic optimization problems that satisfy a certainty equivalence property. This is now an optimal control problem with a quadratic objective and linear constraints. (2. Fallacy and Fantasy
Step 3: Convert the original dynamic optimization problem into a problem with linear constraints and a quadratic objective function.26)
. Ljungqvist and Sargent [146] and Anderson. where d and P are quantities to be determined. Using the deﬁnition of T . 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 ). the dynamic optimization problem can now be written as
∞ {ut }t=0
max E0 ∞
t=0
βt [xt Rxt + ut Qut + 2xt Wut ]
(2. xt . and ut . and Sargent [16] provide further discussion of the formulation and estimation of linear dynamic economic models. in other words.25)
where E (εt+1 ) = 0 and E (εt εt ) = .24)
subject to the linear law of motion xt+1 = Axt + But + εt+1 . This class of problems is known as optimal linear regulator problems.42
Business Cycles: Fact. Given the structure of the problem. Bellman’s equation becomes x Px + d = max x Rx + u Qu + 2x Wu
u
+ βE (Ax + Bu + ε) P(Ax + Bu + ε) + d
. t ≥ 0. 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 . McGrattan. notice that the value function will be a quadratic function in the state variables: V (x) = x Px + d .
(2. Substituting for next period’s state variables and using this expression for the value function.
29) (2. d = (1 − β)−1 [βE (ε Pε)]. 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 ). The equation for P is known as the algebraic matrix Riccati equation for socalled optimal linear regulator problems. A stochastic process for zt is also speciﬁed and a random number generator is used to simulate the TFP time series.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 ). Calibration refers to the practice of determining the parameters of the model based on its steady state properties and the results of other studies.5 Solving for u yields u = −(Q + βB PB)−1 (W x + βB PAx) = −Fx. = T x and = Tx.
. The simulated series are used to generate time series for consumption. where F = (Q + βB PB)−1 (βB PA + W ).
5To derive this result. The policy function associated with Pn is Fn+1 = (Q + βB Pn B)−1 (βB Pn A + W ).27)
Notice that the shocks εt+1 do not affect the optimal choice of ut . the model is then calibrated with the data. Q . ∂u ∂u ∂u 6 Notice that the matrices R. output. A. and B must be further restricted to ensure the existence of a solution to the matrix difference equation deﬁning Pn .28) (2. (2. In the RBC literature. given particular functional forms for the production function and the utility function (see Cooley and Prescott [75]).Models of Business Cycles
43
The ﬁrstorder conditions with respect to u are Qu + W x + β[B PAx + B PBu] = 0. we have used the rules for differentiating quadratic forms as
∂u Qu ∂x Tu ∂u Tx = [Q + Q ]u = 2Qu. One condition that sufﬁces is that the eigenvalues of A are bounded in modulus below unity. starting from P0 = 0.
3655. Cyclical Properties of Key Variables. By contrast. The RBC theorists have interpreted these results as implying that.7496 Correlation with Output 1 0. These are obtained by simulating the behavior of the different series across a given history of the shock sequence. We ﬁnd that consumption ﬂuctuates slightly less than output and investment ﬂuctuates signiﬁcantly more. we draw a sequence of shocks {ˆ t+1 }2999 that are normally distributed t=0 with mean zero and standard deviation 0. y = 0. θ = 0. In terms of the correlation of each series with output. δ = 0. Consider the parameter values β = 0. variances.1.36. Standard Deviation (%) Output Consumption Investment Hours Productivity 8. The means.44
Business Cycles: Fact.9163 0.10. ρ = 0. Finally. so the approach differs signiﬁcantly from standard econometrics. we ﬁnd that all series are procyclical.6684 7.9238 1. with hours showing the least procyclicality. we calculate unconditional moments for the different series after dropping the ﬁrst 1000 observations. in the ﬁrst instance. covariances. We ﬁrst calculate a i ¯ ¯ set of unconditional moments that have been used in the RBC literature to match the model with the data. c = 0. and generate a sequence of ˆ technology shocks beginning from some initial value z0 as zt+1 = ρzt + ˆ t+1 . the model is capable of delivering some of the salient features
Table 3. ¯ ¯ = 0.028. We then use the linear decision rules for all variables to simulate for the endogenous variables based on the same history of shocks. The typical set of unconditional moments used in the RBC literature is displayed in Table 3.1134 12. productivity is almost as variable as output.1281. and h = 0. Likewise. The model is then judged to be close or not close based on this comparison.4783.95. The steady values for the variables are given by k = 1. and ¯ σ = 0.9807 0.1. and labor. Speciﬁcally. Fallacy and Fantasy
investment. we ﬁnd that the variation in hours is typically quite low.028.8616 8. α = 1/3. We now illustrate this solution method for the standard RBC model with a laborleisure choice presented at the beginning of this section.95.4934 0. and crosscorrelations for the simulated time series are compared to the comparable statistics in the data. Trends in the data are removed using a given ﬁlter.9840
.1128.2952.
4
0
5
10 Years after shock
15
20
Fig. These responses have been widely
Hours Output Consumption Capital Technology
1. investment.6 0.5 0. the model predicts that consumption ﬂuctuates less than output whereas investment ﬂuctuates more.Models of Business Cycles
45
of the data. Impulse Responses to a Shock in Technology. capital and consumption show a humpedshaped response.9 0. In terms of the variation accounted by the technology shocks. the RBC model has been viewed as being able to generate cycles endogenously and having economic signiﬁcance even if the calibration method is controversial.8 0. 3. and productivity are all strongly procyclical. Consumption.2 illustrates the response of all the variables to a 1% shock in technology starting from the steadystate capital stock.2. the technology shock explains much but not all of the variation in output. 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.7 0. Figure 3. We note that hours and output both increase and then fall back to a lower level. This is discussed further in the following sections. Speciﬁcally. as we observe in the data. Consumption shows a gradual positive response because investment is initially high.
. Thus.1 1 % deviations form steady state 0. By contrast. rising ﬁrst and then declining back to a lower level.
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. 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. 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. This is accomplished by assuming that individuals can work all the time or not at all. 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. A lottery then determines whether an individual actually works. this is inconsistent with the notion that TFP shocks are the driving force behind business cycle activity. where ˆ 0 < h < 1.1.30)
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.Models of Business Cycles
49
arises in matching the model and the data is that in the data. πt . whereas in the RBC model labor’s share is ﬁxed. • Reverse causality: The stylized facts of business cycles state that money. The conventional wisdom regarding the moneyoutput correlation. ˆ Suppose that workers are constrained to work either zero or h hours. especially M2. This economy is one in which individuals and a ﬁrm trade a contract that commits the household to work h0 hours with probability πt . summarized by Friedman and Schwartz [94] in their study of monetary history. 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. is that the causality goes from money to output but with “long and variable lags”. appears to be a leading indicator of aggregate output.
3. it is assumed that individuals choose the probability of working.31)
. labor’s share of income moves countercyclically. However. A Model with Indivisible Labor Supply
The indivisible labor. (4. the individual gets paid regardless of whether he works or not.4. To account for the nonconvexities introduced by the work/nonwork decision.
33) (4.t .t = c1.t denote the consumption of an unemployed worker and c1.34)
It follows that c0. With complete insurance and identical preferences that are separable with respect to consumption and leisure. lt )] s. (4. c1. but ex post they differ because some work while others enjoy leisure. When agents do work.t ) are 1 − πt = (1 − πt )λt .35)
.t denote the consumption of an employed agent. Fallacy and Fantasy
Let c0. Then the expected utility of the representative consumer. c1. is given by ˆ E [u(ct . lt )] = πt u(c1. all individuals have the same consumption but the unemployed are better off.t πt = πt λt .c0.t .t = ct so that the agent consumes the same amount whether or not he is working. expected utility (where the expectation is over whether or not you work) is ˆ ln (ct ) + πt A ln (1 − h) + (1 − πt )A ln (1).t = ct .t. Notice that the agent will consume ct whether or not he is working. This is a feature that is counterfactual to the working of actual labor markets. ex ante all individuals are alike. c0. the ﬁrstorder conditions with respect to (c0.t (4.
Notice that the social planner chooses the consumption allocations of each agent plus the probability of their working. Let λt denote the Lagrange multiplier on the feasibility constraint for consumption.t
(4. 1 − h) + (1 − πt )u(c0. Omitting the work/leisure decision for the moment. they ˆ must supply h hours of work so that there is no choice over hours of work directly. Hence. Assume that the individual utility function has the form u(c.50
Business Cycles: Fact. Hence. The social planner solves the problem
πt . taking into account the work versus nonwork decision.c1.t . the unemployed worker enjoys higher utility since working causes disutility. πt c1. l ) = ln (c) + A ln (l ). 1). In this model.t + (1 − πt )c0.32)
max E [u(ct .t .
35) and use ln (1) = 0 to obtain ˆ E [u(ct . πt = Ht /h. We now solve the model subject to the time constraint. Ht ) = ln (ct ) − BHt .Ht .
θ−1 1−θ λt = βEt λt+1 exp (zt+1 )θkt+1 Ht+1 + (1 − δ) . lt )] = ln (ct ) + πt A ln (1 − h) = ln (ct ) − BHt .32) with equation (4. Now substitute for πt in (4.38) (4. the latter speciﬁcation — which is linear in total hours Ht — implies a high intertemporal elasticity at the aggregate level.36)
Comparing equation (4.37) (4. By contrast. The preferences can now be written as
∞
U = E0
t=0
βt u(ct . Ht ) .kt+1 }
max
E0
t=0
βt ln (ct ) − BHt + λt [exp (zt )ktθ Ht1−θ +(1 − δ)kt − ct − kt+1 ]
. resource constraint. where B= ˆ −A ln (1 − h) . production function.
where u(ct . Using the deﬁnition of Ht . and law of motion for the capital stock described earlier.39)
. The former speciﬁcation for preferences implies a low intertemporal elasticity of substitution in labor supply. ˆ h (4.36) shows the effect of the lottery assumption.
The ﬁrstorder conditions are 1 = λt . ct B = λt exp (zt )(1 − θ)ktθ Ht−θ . which is consistent with assumptions about individual behavior. The problem is
∞ {ct .
(4.Models of Business Cycles
51
ˆ where A is a positive constant.
7 compared to the model without indivisibilities which implies a value near unity.52
Business Cycles: Fact.2. β.41)
where the term in square brackets shows the rate of return to investing in the aggregate production technology.8
3. 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 }.4. and Heckman [48] for further discussion regarding the reconciliation of micro evidence with dynamic general equilibrium models. and it is not intended to incorporate the microeconomic foundations of the labor market.
(4. and the stochastic process for the technology shock using the approach in Kydland and Prescott [141]. δ. With Ht determined in (4.40)
Equations (4.37) and (4. A. Hansen argues that the model with indivisibilities can generate a variability of hours relative to productivity around 2. 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. 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. Fallacy and Fantasy
Notice that (4.
(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 .
.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)] . Hansen [112] calibrates this model by specifying values for the unknown parameters θ.40). Hansen.
the standard model cannot deliver the strong procyclical response of hours without procyclical behavior in productivity. Per capita output is produced according to the Cobb–Douglas production function yt = (zt nt )1−θ ktθ .42)
¯ where ct denotes consumption and N −nt denotes leisure of the representative household. They consider a prototypical RBC model with a labor/leisure choice. one which is based on a timeseparable logarithmic speciﬁcation and a second which incorporates the Hansen indivisible labor assumption. They generate the negative correlation between hours worked and real wages or productivity by introducing government consumption shocks. They consider two different speciﬁcations for the labor/leisure choice.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.
(4.46)
.45)
where zt is a technology shock which evolves according to the process zt = zt−1 exp (λt ).44) V ( N − nt ) = ¯ N − nt
for all t. Let N denote the time endowment of the representative household per period. (4. (4. and α is a parameter that governs the impact of government consumption on the marginal utility of private consumption.
p where ct is private consumption. 0 < θ < 1. The social planner ranks alternative consumption/leisure streams according to
∞
E0
t=0
¯ βt [ln (ct ) + γV (N − nt )] .43)
is public consumption. ¯ ln (N − nt ) ¯ (4. Consumption services ct are related to private and public consumption as follows: ct = ct + αgt . namely. and utilize the solution of the social planner’s problem to derive the ¯ competitive equilibrium allocations. gt p
(4.
47)
Since the technology shock follows a logarithmic random walk. The key assumption is that private and government consumption are not perfect substitutes. u3 < 0. ct = ct /zt .54
Business Cycles: Fact. consider a decisionmaker who has preferences
∞
βt u(cmt . ρ < 1. hmt . hours of work can increase along a downwardsloping labor demand curve. Following Benhabib. the solution for the model can be more fruitfully expressed in terms of the transformed variables ¯ ¯ ¯ ¯ kt+1 = kt+1 /zt . cmt is the consumption of a market good. an increase in government consumption leads to a negative wealth effect for consumers through the economywide resource constraint. yt = yt /zt . In this model. and µt is the innovation to ln (¯t ) g g g with standard deviation σµ .
. In this expression. Hence. government consumption shocks lead to shifts in the labor supply curve so that.
t=0
where 0 < β < 1. u2 > 0. and hht is labor time spent in home work. Rogerson. and leisure is deﬁned as time not spent working in the market or at home: lt = 1 − hmt − hht . cht . hht ). and u4 < 0. hours increase and average productivity declines in response to a positive government consumption shock. Assume that u1 > 0. Fallacy and Fantasy
In this equation. g (4. The aggregate resource constraint stipulates that consumption plus investment cannot exceed output in each period: ct + gt + kt+1 − (1 − δ)kt ≤ yt . in the absence of technology shocks.48) where ln (¯ ) is the mean of ln (¯t ). λt is an independent and identically distributed process with mean λ and standard deviation σλ .
p
(4. and Wright [39]. and gt = gt /zt . The total amount of time available to the household is normalized as unity.
Home Production
Another way of improving the model’s ability to match the data is to introduce home production. 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 . hmt is labor time spent in market work. cht is consumption of the homeproduced good. If leisure is a normal good.
However.Models of Business Cycles
55
At each date.t+1 + kh. kht . labor effort is likely to be adjusted ﬁrst in response to a productivity shock. but market capital goods are rented to ﬁrms at the competitive rental rate rt .t+1 ≤ wt hmt + rt kmt + (1 − δm )kmt + (1 − δh )kht . This gives us another margin to substitute market labor and improves the model’s predictions. Household capital goods are used in home production.
Labor Hoarding
A third way to resolve the productivity puzzle is through a labor hoarding argument. 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). In the model. Alternative measures put home production to be in the range of 20–50% of GDP. However. thereby leading to greater variability in labor. Letting wt denote the wage rate and δm and δh denote the depreciation rates on market and home capital. where zht is a shock to home production and g is increasing and concave in labor and capital. and household capital goods kht . zht ). Home goods are produced according to the home production function: cht = g (hht . respectively. A rise in market productivity may induce households to substitute away from home production towards market production. one criticism of the home production theory is that it suggests that all movements out of the labor force (toward home production) are voluntary. then ﬁrms may retain workers even though they are not exerting much effort. home production is used to produce a nontradeable consumption good. This says that the effective labor input can be altered even though the total number of workers is ﬁxed. Eventually more workers may be hired or ﬁred. Unlike the standard model. the household’s budget constraint is cmt + km. if there are costs to hiring or laying workers off. market capital goods kmt . the household can purchase market goods cmt . Hence.
. the labor supply curve also shifts in response to a good productivity shock. but only after longer periods of time.
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]. Eichenbaum. 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 ). Ht = fNt . Yt = At Ktα [fNt Wt ]1−α .56
Business Cycles: Fact. the total number of workers. capital Kt . and Rebelo [52]. Thus. and a labor input that reﬂects variations in work effort following Burnside. hence. Nt .
In the standard model. This can lead to variation in the Solow residual that is not related to changes in productivity or technology. To brieﬂy describe their framework. suppose
.
3. The comments made regarding ﬂuctuations in the effort level of labor also apply to capital.3. so that the movements in the Solow residual are not entirely exogenous. and Wt . The measured capital in the Solow residual does not take into account optimal ﬂuctuations in capital utilization rates. consider a production function for aggregate output Yt which depends on an exogenous technology shock At . Speciﬁcally.4. If labor hoarding is included in the model. Decisions to alter effort levels will be the outcome of a maximizing decision. let f denote a ﬁxed shift length. where Ht denotes the total hours worked. Fallacy and Fantasy
To illustrate the role of labor hoarding. 0 < α < 1. then the productivity/hours correlation is reduced and more closely matches the data. the work effort of each individual. can be treated as a direct input into the aggregate production function. In the absence of variations in work effort. 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. the production function can be written as Yt = St Ktα [Ht ]1−α .
(4. and dft is the amount of transactions services used in the ﬁnal goods industry.53)
∗ as dht = h(ρt /wt )(ct + it ). where h = (τ )−1 . In this expression.50)
where ndt and kdt denote the amounts of labor and capital allocated to the ﬁnancial sector. By contrast. (4. The time required for this activity is nτt = τ(dht /(ct + it ))(ct + it ). wt nτt + ρt dht .Models of Business Cycles
57
the ﬁnal good is produced according to the production technology yt = f (kft . Households maximize the expected discounted value of utility from consumption ct and leisure lt as
∞
E0
t=0
βt U (ct . Likewise. φt is a shock to production of the ﬁnal good at time t.52)
where τ < 0. Finally. τ < 0.49)
where kft is the amount of capital. Households are also assumed to combine time and transactions services to accomplish consumption and investment purchases. nft is the amount of labor services. The household chooses an amount of transactions services dht so as to minimize the total transactions costs. where wt is the real wage and ρt is the rental price of transactions services. (4. dft )φt . The production of the intermediate good is given by dt = g (ndt .
0 < β < 1. respectively. where 0 < δ < 1 is the depreciation rate on capital. capital. nft . kdt )λt . and λt captures technological innovations to the ﬁnancial services industry. 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. (4. ﬁrms rent labor.51)
Households are assumed to own the capital stock and to make investment decisions it subject to the resource constraint ct + it ≤ yt + (1 − δ)kt . lt ) . hours allocated to ∗ producing transactions services is nτt = τ ∗ (h(ρt /wt ))(ct + it ). we
.
Consistent with the RBC view. 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. Speciﬁcally. a positive shock to productivity occurs. if interest rates are rising and ﬁrms wish to invest more in anticipation of higher expected proﬁts. investment demand rises and output will also increase in response to the additional hours worked.5. If the substitution effect of an increase in the marginal product of labor outweighs the wealth effect of the productivity shock. As a result. Then.9 During an expansion. Suppose that a shock to the production of ﬁnal goods or. or a broad measure of money that includes commercial credit. responds to the positive productivity shock. commercial credit. there will be an increase in hours of work. As a consequence. nt = nft + ndt + nτt . One
9 See Sims [190]. The increase in output will also stimulate the demand for transactions services by households and ﬁrms. consumption and leisure of the representative consumer will rise but so will investment demand as consumers seek to spread the extra wealth over time. Fallacy and Fantasy
require that the household’s time allocated to the different activities sums to 1. Hence. Their results are derived from a structural vector autoregression (VAR) with longrun restrictions to identify the alternative structural shocks.
. Another possibility is that the central bank uses the accommodative monetary policy. inside money. equivalently.58
Business Cycles: Fact. the central bank may expand the money supply to keep interest rates from rising too rapidly. This framework can be used to rationalize the observations regarding money and output. the causality runs from the productivity shock to money even though the increase in money occurs before the higher productivity shock. so ﬁrms will wish to ﬁnance a greater volume of goods in process. is more closely related to output than outside money. they ﬁnd that an important source of the moneyoutput correlation is output shocks affecting inside money in the short run.
3. 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. or inside money.
The notion behind this type of change is that production of capital goods becomes increasingly efﬁcient over time.
10 Altug.
. Hercowitz. Demers. the relative price of new equipment declined by 3% at an average annual rate while the equipmentGNP ratio increased substantially. Furthermore. and Krusell [106]. 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
∞
E0
t=0
βt [θ ln (ct ) + (1 − θ) ln (1 − lt )] .
0 < θ < 1.5. This has been examined by Greenwood. War shocks can also be modeled as a source of cyclical ﬂuctuations (see Barro [27] and Ohanian [169]). thereby stimulating output growth. [107]. Although their analysis is not general equilibrium.1. during the period 1950–1990. 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. Hercowitz.11 In Greenwood et al. they nevertheless show that political events can have nonnegligible effects on real economic variables. 11 In Greenwood.10
3. 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. In the previous section. Other papers that have employed the RBC framework with ﬁscal shocks include Braun [46] and McGrattan [158]. there was a negative correlation (−0. Following Gordon [104]. In this vein. These authors consider the impact of distortionary taxes on real outcomes.Models of Business Cycles
59
way of dealing with this criticism is to introduce alternative sources of shocks into RBC models.46) between a detrended measure of the relative price of capital and investment expenditures. InvestmentSpeciﬁc Technological Shocks
An important source of shocks is the investmentspeciﬁc technological change. 107]. 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. these authors motivate their analysis by noting that. we discussed the role of government or ﬁscal shocks in generating observed comovements of the data.
where ζt+1 and ηt+1 are innovations to productivity and investmentspeciﬁc technological change. αs . 0 < δs < 1. and γz and γq denote the average gross growth rates of the two processes. φe . respectively. (5. there is a government which levies taxes on labor and capital income at the rates τl and τk . φs . Equipment is utilized at the timevarying rate h.
where z is a measure of total factor productivity.57) (5. equipment denoted ke and structures denoted ks .56)
Thus. Investment in both structures and equipment is also subject to convex costs of adjustment denoted as a = ae + as . z) = z(hke )αe ksαs l 1−αe −αs . The shocks to technology and to the efﬁciency of equipment capital constitute the drivers of the t+1 t+1 model. Structures evolve according to the standard law of motion: ks = (1 − δs )ks + ii .54)
The law of motion for equipment is given by ke = (1 − δe (h))ke + ie q.58)
Finally. the service ﬂow is denoted hke . (5.55)
where the variable q shows the investmentspeciﬁc technological change in equipment and δe (h) = b h. κe > 0. κs > 0. The revenue raised by the government is
. (5.60
Business Cycles: Fact. respectively. ks . (5. The production function is given by y = F (hke . They evolve as zt+1 = γz exp (ζt+1 ) and qt+1 = γq exp (ηt+1 ). as = φs (ks − κs ks )2 . ω ω > 1. investment in equipment is subject to variable rates of utilization and depreciation which are convex in the utilization rate. that follow ﬁrstorder Markov processes. There is a production technology that depends on labor and the services from two types of capital goods. αe + αs < 1. 0 < αe . hence. l . Fallacy and Fantasy
where ct denotes consumption and lt denotes labor supply. where ae = exp (η)φe (ke /q − κe ke /q)2 /(ke q).
investmentspeciﬁc technological change increases the services from old equipment at the same time as it increases the quantity of equipment available for next period. However. we have that g = γ (γ g )αe g αs . and it is set at
12 From the production function. however. grows faster. The resource constraint and the accumulation equation for structures imply that y. ae . this quantity is determined as the ratio of Gordon’s [104] price index of qualityadjusted equipment and the price index for consumption. respectively. (5. Since the inverse of q denotes the relative price of equipment in terms of consumption goods. as . Equipment. 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 . it also affects the utilization cost of existing equipment this period by lowering the replacement cost of old capital goods. The ﬁrst parameter is determined using data on the inverse of the relative price of investment goods following Gordon [104]. 1/(1−αe −αs ) (1−αs )/(1−αe −αs ) γq . the investmentspeciﬁc technological shock plays a key role. and ks all have to grow at the same rate. The parameters that are determined using a priori information include the average growth rate of the investmentspeciﬁc shock γq . the depreciation rate on structures δs . Hence. In the balanced growth path equilibrium. denote this rate by g . A higher value of q directly affects the quantity of equipment that will be available to produce output next period.12 A key parameter to be estimated from the data is γq .
. respectively.60)
In this analysis. 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 . (5. and w denotes the real wage rate. c. and the tax rate on wage income τl . is . or z q
g = γz ge = γz
1/(1−αe −αs ) αe /(1−αe −αs ) γq . at the rate ge = γq g . and hence increases its utilization.Models of Business Cycles
61
returned to the private sector through lumpsum transfers τ. This yields the government budget constraint: τ = τk (re hke + rs ks ) + τl wl . ie .59)
where re and rs denote the rental rate on equipment and structures.
Furthermore. which raises utilization of equipment today. and bh ω = 0. Hence. the investmentspeciﬁc shock affects uses of factors such as labor and capital through changed investment opportunities.40. Greenwood et al. ω = 1. 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.53.032.18. the impact of the investmentspeciﬁc technological shock depends on the quantitative importance of the transmission mechanism. They conclude that though investmentspeciﬁc shocks contribute to business cycle ﬂuctuations. The second set of parameters is calculated using information on the longrun behavior of a set of the model’s variables.0124. the investmentspeciﬁc technological shocks explain around 28% of business cycle ﬂuctuations as captured by the standard deviation of output. According to the ﬁrst.59.056 and τl = 0. Fallacy and Fantasy
1.62
Business Cycles: Fact. They ﬁnd that under a high value of φ.40. τk = 0. αe = 0. they are not the main factor. For a low value of φ. there is a decline in the equipment’s replacement value. This constitutes an upper value for the contribution of the shocks q. Hence. Given that equipment investment is only 7% of GNP and 18% of the value of output being derived from the use of equipment.20. The model is simulated to understand the contribution of the investmentspeciﬁc technological shocks in generating cyclical ﬂuctuations. The restriction that adjustment costs are zero along the balanced growth path yields the values κe = κs = g .12. αs = 0. This leads to increased employment and output. the parameter φ is set to equalize the consumption/output correlation implied by the model to that in the data. where the parameters b and h cannot be identiﬁed separately. the symmetry condition φe = (g /ge )2 φs ≡ φ leaves only the parameter φ to be determined. only the q shock is assumed to be operative. which tends to raise the stock of equipment next period.
. 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. the fraction of output directly affected by the shock will typically be small. g = 1. By contrast. this ﬁgure is 32%. This standard calibration exercise yields θ = 0. [107] calibrate the adjustment cost parameter under two alternative assumptions. Unlike the standard technology shock in RBC models. For this purpose. δs = 0. β = 0. Simultaneously. A positive shock to this variable works by increasing the rate of return to equipment investment.97. Likewise.
Rotemberg and Woodford [183] provide a rationale for the role of energy price shocks in a model with imperfect competition. Barsky and Kilian [28] analyze the relationship between oil shocks and changes in economic activity.5.
where zt is an exogenous technology shock. Energy Shocks
From the beginning. As in the standard RBC framework. By contrast. and ut is the rate of utilization of capital. 0 < θ < 1. and ω1 > 1.
The production function is given by yt = (zt lt )θ (kt ut )1−θ .
0 < β < 1. kt is the stock of capital. 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 . lt ) . The new feature in Finn’s analysis is that capital utilization requires energy: et ν0 utν1 = . ω0 > 0. ν1 > 1. capital utilization is the channel through which energy shocks enter the model. Finn [92] argues that it is possible to rationalize the role of energy shocks in models with perfect competition.
. the source of large supplyside shocks for the RBC agenda has been sought in oil shocks. σ > 0.2. lt ) = t 1−σ
1−σ
−1
. δ(ut ) = ω0 utω1 . ω1
where 0 < δ(ut ) < 1.Models of Business Cycles
63
3. kt ν1 ν0 > 0. We brieﬂy describe her model before concluding this chapter. In her model.
with c α (1 − lt )1−α u(ct .
0 < α < 1. As in the previous model.
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
. First.64
Business Cycles: Fact. This shows the direct effect of energy. an increase in pt reduces et and ut through the production function.
Thus. a positive energy price shock operates in a similar way as a negative technology shock. Thus.
. there are two channels for the transmission of energy shocks in the model. capital. Finally. where pt is the price of energy. Finn [92] describes qualitatively and quantitatively the impact of increases in the price of energy. Hence. and energy. but there is also an indirect effect which works through the effect of the utilization rate on the depreciation of capital. ut . then the decrease in the values of et . This tends to reduce the investment it and the future capital stock kt . If the price shock is persistent. and hence the real wage wt and work effort lt . 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. and lt continue into the future and reduce the future marginal productivity of capital. 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. The decrease in et further reduces the marginal product of labor. the economywide resource constraint is given by ct + it + pt et ≤ yt . the production of output depends on labor.
In this chapter. Kehoe.
4. They study the properties of a baseline model against the actual features in the data. 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. They can be used to model the impact of the ability to trade in goods and to borrow and lend internationally. [26] study
65
.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. The closedeconomy model cannot be used to understand the propagation of shocks across countries or regions.1. 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. openeconomy business cycle models allow for the role of technology processes in different countries in generating cyclical ﬂuctuations. Backus et al. and Kydland [25] consider data on 12 developed countries over a sample period dating from 1960 to 1990. Backus. we review some of the facts of international business cycles and discuss models that have been used to rationalize the ﬁndings. including correlations of macroeconomic aggregates across countries and movements in the balance of trade. FACTS
One of the important areas of research has been to derive the socalled stylized facts of international business cycles. Unlike the closedeconomy model. nor can it be used to discuss notions of international risk sharing.
Italy. and t . Japan. By contrast. Canada. [25. and the United States. Austria. The price anomaly: The volatility of the terms of trade. zt∗ denote the technology shocks to the domestic and foreign countries. France. these authors use a symmetric speciﬁcation with A= 0. Based on several estimated speciﬁcations for the US and the European aggregate. the theoretical model implies the reverse ranking. and the crosscorrelations of productivity are greater than those of consumption. measured as the Solow residual. 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. ∗ denote the respective innovations to these t shocks.258. Germany. the United Kingdom. The large autoregressive coefﬁcients displayed on the diagonal capture the persistence in observed productivity. In this regard. Backus et al. The ﬁndings can be summarized under two headings: • The quantity anomaly: In the data. deﬁned as the relative price of imports to exports. Among the important features of their analysis is the estimation of the correlation properties of international productivity shocks using data on estimated Solow residuals.
where zt .906 0. Thus.088 . The individual countries they consider are Australia.
•
. 26] estimate a bivariate AR(1) process for the domestic and foreign shocks as zt zt∗ =A zt−1 + ∗ zt−1
t ∗ t
. the exercise in this literature is to replicate the correlation properties among the different series given the variability and correlation properties of the shocks. 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.906
with Std ( t ) = Std ( ∗ ) = 0. 0. the crosscountry correlations of output are greater than those of productivity.66
Business Cycles: Fact. Switzerland. respectively. is much higher in the data than it is in the theoretical model. ∗ ) = 0. The data are ﬁltered using the Hodrick–Prescott ﬁlter.00852 and Corr( t .088 0.
In other words. and Zimmermann [14] extend the Backus et al. With perfect risk sharing. the increase in investment plus consumption is greater than the increase in output. The twocountry frictionless international business cycle model also implies that consumers in different countries can share risk perfectly. The ﬁndings of Ambler et al. This leads to a strong investment response in the US. This changes the outputinvestment correlation from positive to negative in the domestic country. this ratio displays little volatility. we thus observe a ﬂow of factors from other countries or regions to the US. whereas in the model. First. suppose a positive technology shock occurs in one country. The international RBC model produces these results because it implies that the incentive is to use inputs where they are most productive. they are negative. 26]. Cardia. Since investment responds strongly to a positive productivity shock. This fact leads to negative crosscorrelations between output. This feature of the model leads to the crosscountry correlations of consumption being much higher in the model than in the data. In a model with perfect capital mobility. In their original analysis. investment. [25] also examine a variety of perturbations of their original setup. say. what is typically observed in the data are positive comovements in these series across countries. [14] conﬁrm the ﬁndings of Backus et al. 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. Backus et al. though with lower magnitudes. By contrast. the US. investment. investment and the trade balance are more volatile in the model than they are in the data. implying that the real exchange rate is also less volatile in the model than it is in the data. 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. but the variability of investment and the trade balance continue to remain
.International Business Cycles
67
Ambler. and employment are positive and generally high in the data. which is accompanied by output and employment increases. which leads to a trade deﬁcit in the domestic country. Furthermore. and employment in the theoretical model. They estimate the moments of interest using the generalized method of moments (GMM) estimation. [25. which allows them to obtain standard errors. They also ﬁnd that the crosscountry correlations of output. [26] sample to 20 industrialized countries over the period 1960–2004.
but stems from the permanent income hypothesis: when domestic and foreign productivity shocks are correlated. [25.
4. [25] argue that this result is not due to international risk sharing considerations. In this case. Hence. They ﬁnd that the results are very similar to the case with a small transport cost. the crosscountry correlation of consumption increases further. THE ROLE OF INTERNATIONAL RISK SHARING
Before we continue with other extensions that have been proposed in the international RBC literature. the foreign agent increases consumption and reduces investment in anticipation of future income increases. Fallacy and Fantasy
counterfactually high. Backus et al. 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. However. Next. The standard international RBC model studied by Backus et al. Second. as does the positive correlation of consumption across countries. 26] considers the case with complete contingent claims and perfect risk sharing across different countries. they ﬁnd that investment and the trade balance become much less volatile. but has little effect on the crosscountry correlations of consumption and output.5. They also examine various sorts of trading frictions. and the crosscountry correlation of output goes from being negative to positive. With this change. a rise in productivity in the domestic country signals a rise in productivity in the foreign country through the spillover effects. The ﬁrst of these involves introducing a small trading friction in the form of a transport cost.2.2 and the correlation of the innovations from 0. Surprisingly. they eliminate all trade in goods and assets by considering an autarky situation. They ﬁnd that this friction has the effect of reducing the volatility of investment and the trade balance. elimination of trade in statecontingent claims does not help to lower the crosscountry correlations of consumption. it is the correlation between technology shocks that leads to any connection between countries.088 to 0.258 to 0. Cole [71] and Cole and Obstfeld [72] examine a twocountry real version of the Lucas asset pricing model for this purpose (see Lucas [152]). they consider large spillovers across countries by changing the offdiagonal elements of the A matrix from 0. it is important to understand the role of alternative ﬁnancial arrangements in generating the crosscountry correlations. Agents from both
.68
Business Cycles: Fact.
The functions y1 : S → Y and y2 : S → Y are continuous functions that are bounded away from zero. we can examine the solution to a central planning problem which maximizes the weighted utility of agents in the domestic and foreign countries subject to the
. In the ˆ ˆ beginning. We assume that endowments are stationary in levels. Pareto Optimal Allocations
We begin by characterizing the Pareto optimal allocations.j .t }∞ deﬁned by t=0
∞
E0
t=0
βt U (c1. Neither good is storable.t . c2 ) lim
This requirement on the utility function ensures that both goods are consumed in equilibrium.2. strictly increasing.t ) .2. c2 ) = 0.
j
j
0 < β < 1. We assume that endowment is a timeinvariant function of the exogenous shock. Deﬁne Y ≡ [y. Let π denote the vector of stationary probabilities which satisfy π = P π.
(2. where y > 0 and y < ∞. To do this. c2 →0 U2 (c1 . Country 1 has a random endowment of good Y1 and country 2 has a random endowment of good Y2 . 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. j) element pi.j = p(st = sj st−1 = si ). c1 →0 U2 (c1 . There are two goods. where pi. ¯ ¯ Assumption 2.
4. c2 ) = ∞.International Business Cycles
69
countries are identical in terms of preferences and differ only in terms of endowments. Y1 and Y2 .1)
We have the following assumption. c2 ) lim
U1 (c1 . The representative consumer in country j has preferences over random j j sequences {c1.1. y ]. The utility function U : R+ → R is continuously differentiable.
U1 (c1 . For all c1 . c2. c2. and strictly concave. c2 > 0. agents observe the current realization.
2 Assumption 2.1.t .
. Let µi (st ) denote the Lagrange multiplier for the resource constraint for good i. st−1 .t + ci. The ﬁrstorder conditions imply that φj U1 (c1.t .t ) + φ2 U (c1. . φj U2 (c1. Deﬁne ˆ π(s t ) = π(st st−1 )π(st−1 st−2 ) · · · π(s1 s0 )π(s0 ).5)
. s0 ) denotes the history of the shocks at time t.t ) 2 2 U2 (c1.t .3)
subject to
1 2 ci. c2. Fallacy and Fantasy
resource constraints. c2. 2. c2. the endowment is nonstorable.2)
subject to ci1 (s t ) + ci2 (s t ) = yi (st ). i = 1.
Since preferences are timeseparable. the social planner’s problem at time t involves solving the static problem
1 1 2 2 max[φ1 U (c1.t .4) (2.t .t .t ) 2 2 U1 (c1. The social planner chooses sequences {ci1 (s t ).t . c2. c2 (s t )) t=0 st
(2. 2.t .t . φ1
(2. c2. φ1 φ2 .t ) 1 1 U2 (c1. c2 (s t )) + φ2 U (c1 (s t ). Let φj denote the Pareto weight for country j. c2. or
1 1 U1 (c1. c2.t denotes the consumption of good i by residents of country j.t .
where ci. c2. and there is no investment process. .t ) = µ1 (st ). 2.t ) = µ2 (st ) for j = 1.70
Business Cycles: Fact.t )]
(2. 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 ). .t ) j j j j
= =
φ2 .t = yi. where s t = (st . j
i = 1.
2.9) (2.6)
Assuming that preferences are of Cobb–Douglas variety.t 2 c2.t ) =
α 1−α (c1.t 2 c1.t . Complete Contingent Claims
Consider ﬁrst the behavior of allocations in a competitive equilibrium in which agents can trade claims that pay off contingent on all possible realizations of the income processes in each country. will cause consumption in both countries to increase.t = δy1.2. It is straightforward to demonstrate
.t = (1 − δ)y2.
4.International Business Cycles
71
Hence. which also causes the output of country 2 to increase through positive spillover effects. In what follows.t 1 c2.
(2. It is convenient to assume that preferences are of the Cobb–Douglas variety: U (c1.t c2.t = (1 − δ)y1. 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. This is an implication of perfect risk sharing that arises in a complete contingent claims equilibrium. These expressions show that consumers in each country consume a constant fraction of output in each period. we will examine the role of alternative ﬁnancial arrangements that can be used to rationalize the observations. the Pareto optimal consumption allocations satisfy
1 c1.t = δy2. c2.t . we ﬁnd that the ratio of marginal utilities across agents in different countries is equalized for all states and goods.7) (2.t )1−ρ
1−ρ
.10)
where δ= 1+ φ2 φ1
σ
1 and σ = ρ .8) (2.
(2. Also note that a shock to the output of country 1.
2 are given by
1 1 1 1 y1.t + p(st )c2.t + z1.t =
p(st )
2 + c2.
CA = y2.
1 For further discussion. is given by
1 1 CA1 = y1.t + p(st )c2.t
p(st )
2 + c2. 2. 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. it is without loss of generality to consider only oneperiod contingent claims. Fallacy and Fantasy
that there exists a contingent claims equilibrium for this economy.t
+
st+1 ∈S 2 z1.
.t + p(st )z2.t = c1. p(st )
(2. Thus.t .11)
y2. The current account for each country.1 Agents can also trade in the goods of the other country. 7.t −
2
2 c1. st ) z1 (st+1 ) + p(st )z2 (st+1 ) 2 c1. st )
st+1 ∈S
2 z1 (st+1 ) 2 + z2 (st+1 ) . 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 .t
(2.t + +
p(st )
2 + z2.12)
where q(st+1 .72
Business Cycles: Fact. Given the Markov nature of uncertainty. j = 1. 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.t
q(st+1 .t
1 1 q(st+1 . denoted as CAj . and that the competitive equilibrium allocations are Pareto optimal. see Altug and Labadie [6]. The sequential budget constraints for residents of country j for j = 1.t − c1. Ch. Let p(st ) denote the relative price of good y2 in terms of the numéraire good y1 .
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 .
Notice that the noassettrading allocation is identical to the central planning allocation if α=δ= 1+ or φ = 1+
1
φ2 φ1
σ
.t is positively transmitted to the residents of country 2 by the increase in demand (and hence the relative price) for good y2.t .t . 1 2 c2.t = y1.t = αy2. despite the absence of trade in ﬁnancial assets. αy2.
(2. A large negative shock in the amount of a good is similarly transmitted across borders.76
Business Cycles: Fact. The international ratio of marginal utilities across countries is identical across goods and states. Hence.t = (1 − α)y1.t .t
This price can be substituted into the consumption functions above to show that
1 c2.t .t . A large and positive shock in the amount of good y1.17) (2.t + c2. 2 c1.t .
ρ −1
1−α α
and φ2 = 1 − φ1 .t = y2.t + c1. Fallacy and Fantasy
Equilibrium in the two goods markets requires that
1 2 c1.18)
Substitute the consumption functions into the marketclearing conditions and solve for the relative price to show that pt = (1 − α)y1.
•
. efﬁcient risk sharing occurs through changes in the relative price of goods. Efﬁcient risk sharing can occur despite the lack of ﬁnancial assets and insurance. This exercise illustrates several points: • The absence of international capital mobility does not necessarily imply that the allocation is not Pareto optimal.
respectively.4.t in terms of wt and p2.t .t ≤ p2.t in units of wt .2.t c2. under the assumption that the current account is zero.t + wta .t + p2. despite the absence of ﬁnancial capital mobility. The total consumption of countries 1 and 2 is
1 1 c1.t y1.t c2. Cole and Obstfeld [72] introduce a third good. 2 2 c1. and can show that real interest rates and real asset returns will be equal for the two countries.t + pt c2.t denotes the relative price of good y2.t α2 [wt1 + wt2 ] .
4.t + αy2.
The equilibrium relative prices satisfy p1.t = (1 − α)y1.t + (1 − α)y2. As Cole and Obstfeld [72] point out.t .t + wtb .
Notice that correlation of the total value of consumption of country 1 and country 2 is positive and equal to one. 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. the positive transmission of shocks occurs because countries specialize in the production of goods.t
.t = p2. 2 2 2 p1.t + p2. αw y2.t + cw.t = α1 [wt1 + wt2 ] . αw y1. Let α1 .t denotes the relative price of good y1.t = αy1. so that p1.t c1. Call this third good w. Agents in countries 1 and 2 have budget constraints given. αw denote the expenditure shares under the assumption of Cobb–Douglas preferences and let w be the numéraire good.t + pt c2.t ≤ p1.t c1. Nonspecialization in Endowments
To illustrate the impact of nonspecialization. by
1 1 1 p1.t + cw. α2 .International Business Cycles
77
•
•
Notice that we can price ﬁnancial assets in the model.t y2.
whereas shocks to wt1 . αn denote the expenditure shares under the assumption of Cobb–Douglas preferences and let y1 be the numéraire good. a condition for Pareto optimality.t . 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. α2 . then it is beneﬁcial to trade equity shares or other forms of ﬁnancial assets.t or y2. Let α1 .78
Business Cycles: Fact. Call this good n and assume that nj : S → N = [n.
Similar expressions can be derived for the consumption of goods y2 and w. + α2 y1.t = αw
wt1 wt1 + wt2 wt2 wt1 + wt2
+ α1 y1. The shocks to wt1 and wt2 must be perfectly correlated for the allocation with no trade in ﬁnancial assets to be Pareto optimal. by deﬁnition. where wt1 . are sectorspeciﬁc shocks. Shocks to y1.
t t share of the endowment of wt . is constant as the total t t t t
w1
w2
4. 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. If these shocks are not perfectly correlated.t are. This helps to clearly distinguish between countryspeciﬁc shocks. ( w 1 +w 2 ) and ( w 1 +w 2 ). Hence. in the absence of trade in ﬁnancial assets. under the assumption of Cobb–Douglas preferences. Since the current account must always be balanced.t = αw 2 c1. only if the wt varies with st . Fallacy and Fantasy
The consumption of good 1 by agents in countries 1 and 2. when there are sectorspeciﬁc shocks. countryspeciﬁc shocks. the country must export more of the good in which it specializes in production to ﬁnance the import of good wt . Under the
. Nontraded Goods
Suppose now that the third good is a nontraded good.t . wt2 . which affect all sectors within a country. can be shown to satisfy
1 c1. wt2 are not perfectly correlated. there are gains to asset trading that improve risk sharing and allow diversiﬁcation. n] denotes the realization of good n in ¯ country j. The ratio of marginal utilities of both countries for each good will be equal to a constant across all states.2. and industryspeciﬁc shocks.5. The intuition is that.
the correlation of national consumption levels may be close to zero.t+1 + q2.t .t [p2.t = 1 c2. 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).t + p2. Trade in Equity Shares
We now introduce trade in ﬁnancial assets.t 2 c1. 1 − αn α2 = y1. The ratio of marginal utility across countries for a traded good will now depend 1 2 on the ratio nt2 .t + q1.t .t + q2.t [y1. then even if consumption of traded goods is perfectly correlated. Notice that the correlation of consumption across countries will now depend on the proportion of a country’s consumption that is nontradeable.t . nt . If this sector constitutes a large fraction of consumption. An agent’s budget constraint is z1. 2 and i = 1. j Let zi.t
α1 y1.t for j = 1. 1 − αn α1 = y2. nontraded goods provide one vehicle for reducing the large crosscountry correlations implied by the baseline model.International Business Cycles
79
assumption of balanced trade and Cobb–Douglas preferences. nt are perfectly correlated.t ] + z2. There are gains from international risk sharing through asset trade.t z1. 1 − αn α2 = y2.t+1 .t + q1. then the resulting t allocations will not be Pareto optimal. Thus.t z2. the demands for the goods satisfy
1 c1.t y2. 2 denote the shares of good i held by an agent in country j at the beginning of period t.2.t .t c2. 1 − αn
1 2 Each country consumes its endowment of the nontraded good. We now discuss the impact of asset trading and relate it to the model without asset trade discussed earlier.t ] ≥ c1.
j j j j j j j
(2.6. Unless nt .19)
. nt .t 2 c2. n1
4.
In the initial model described earlier. There has been substantial literature on the lack of international portfolio diversiﬁcation and the degree of international consumption risk sharing. c2. This simple example. 2 and i = 1.t . c2. The
. The agent maximizes his objective function subject to the constraint. International risk sharing can be achieved through ﬂuctuations in relative prices in the current account and by trade in ﬁnancial assets.t . there was no trade in ﬁnancial assets and specialization in endowments and the allocation was Pareto optimal.t )q1. it does not require the existence of a full set of contingent claims.t .t+1 ].20) (2.t+1 .t+1 + p2. U1 (c1. c2. so that zi.t+1 . We commented that we could price ﬁnancial assets even 1 1 if these assets were not traded.t+1 )[q1. illustrates an important point. when combined with our discussion of the equilibrium with no trade in ﬁnancial assets.t = z2. we can achieve at least two stationary allocations.t = 1/2 for j = 1. In particular. If we assume that z1.21) (2.t+1 ].
j j j j j j j j j j j j
(2.t = α and 2 2 z1.t . Lucas [152] assumes that agents hold identical j portfolios. 2 so that φj = 1/2 and δ = 1/2. all equity shares are held and the endowment of each good is completely consumed. 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.t = 1 − α. unlike the economy in which there is no trade in ﬁnancial assets.t )p2.t = z2. Agents across countries have identical consumption in this case. depending on the initial distribution of the claims. c2. U1 (c1. then the equilibrium allocation with no trade in ﬁnancial assets can be achieved. then portfolio diversiﬁcation may be achieved by specializing in the holding of equity shares of one country only.t = βEt U1 (c1.t . If these shocks are perfectly correlated. If we introduced trade in equity shares when there is a third good w that is produced by both countries. c2. The ﬁrstorder conditions are U1 (c1.t )q2.80
j
Business Cycles: Fact.t+1 y2.t ) = U2 (c1.t+1 )[q2.t+1 + y1.22)
In equilibrium. c2. In such a world. national wealth is equal across countries and agents have perfectly diversiﬁed portfolios.t = βEt U1 (c1. 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. Hence. Fallacy and Fantasy
Let µt denote the Lagrange multiplier.
Baxter and Jermann [36] argue that the failure of international diversiﬁcation is substantial.2. one could ask whether limited risk sharing opportunities among consumers in different countries could help to better reconcile the data with the model. Lewis [145] also documents that there is insufﬁcient intertemporal risk sharing in consumption. Nevertheless. By contrast. They ﬁnd. that consumption insurance is available through relative price ﬂuctuations and that these price ﬂuctuations are capable of achieving efﬁcient risk sharing. 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. combined with the assumption that utility is nonseparable in traded and nontraded goods. [25]. Devereux.
.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. A more recent paper by Heathcote and Perri [117] extends the Baxter and Jermann model to include more than one traded good. she cannot reject the hypothesis that there is risk sharing. the existence of nontraded goods. Gregory. whether there exists perfect risk sharing in the data appears to depend on the model speciﬁcation adopted by the researcher. as we have noted above. within the context of their model. Empirical evidence on international consumption risk sharing is provided in Backus et al. Their model incorporates human and physical capital and. We have also shown above that relative price ﬂuctuations can be a substitute for trade in ﬁnancial assets in achieving consumption insurance. Clearly. who show that the data are inconsistent with the implications for consumption for the baseline international RBC model. Limited Risk Sharing
As described earlier. 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. As we have noted above. the conclusion on whether there is sufﬁcient or insufﬁcient risk sharing is very sensitive to model speciﬁcations. 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.7. makes it more difﬁcult to determine the optimal degree of consumption risk sharing. but that when nonseparability and asset trade restrictions are combined.
4.
Kollman [136] and Baxter and Crucini [34] examine models where only noncontingent bonds can be traded internationally. li (s t )). 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. 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.82
Business Cycles: Fact. Their analysis involves extending the analysis in Cole and Obstfeld [72] to incorporate an explicit production side. Kocherlakota [135]. where Ai (s t ) is a random shock. and hours worked remain counterfactually negative. In their framework. given a trade structure. 125].
t=0 s t
. Ai (s t )li (s t )). Alvarez and Jermann [12]. 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 (ki (s t−1 ). 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. The preferences of the representative consumer in each country are given by
∞
π(s t )U (ci (s t ). In these models. credit market frictions and restrictions on international capital ﬂows constitute a proposed channel for the propagation of international shocks. Kehoe and Perri [126] consider a model where market incompleteness is obtained endogenously through the introduction of imperfectly enforceable international loans. and others on debtconstrained asset markets. Fallacy and Fantasy
In this vein. They show that the model matches the correlations in the data under a ﬁnancial autarky assumption. Moreover. but the crosscountry correlations of output. These authors ﬁnd that incomplete asset markets help to reduce the crosscountry correlation of consumption. investment. This analysis builds on the earlier works of Kehoe and Levine [124. Heathcote and Perri [117] examine the implications of a twocountry model under alternative assumptions about the ﬁnancial structure.
r≥t
given ki (s t−1 ). s ) = max
r=t sr
t
βr π(s r s t )U (ci (s r ). Ai (s t )li (s t )) + (1 − δ)ki (s t−1 ) . s t ) denotes the value of autarky from s t onwards. ci (s r ) for r ≥ t to solve
∞
Vi (ki (s subject to
t−1
).
i=1
(2.2 2
F (ki (s t−1 ). the competitive equilibrium allocations solve the social planner’s problem deﬁned as
∞
π(s t )βt φ1 U (c1 (s t ). s t ) involves choosing ki (s r ). s t ). l2 (s t )) . An
. Under complete markets. These require that each country prefers the allocation it receives to the one that it could attain if it were in (ﬁnancial) autarky from then onwards.
r=t sr
(2.25)
where π(s r s t ) denotes the conditional probability of the history s r given s t and Vi (ki (s t−1 ). where Vi (ki (s t−1 ). the social planner’s problem with enforcement constraints cannot be formulated recursively as a dynamic programming problem.23)
subject to the resource constraints ci (s t ) + ki (s t ) =
i=1. li (s r )) ≥ Vi (ki (s t−1 ). The reason is that future decision variables such as ci (s r ) and li (s r ) for r > t enter the current enforcement constraints. As Kehoe and Perri [126] explain.
t=0 s t
(2. li (s r ))
ci (s r ) + ki (s r ) ≤ F (ki (s r−1 ). li (s r ).24)
The innovation in Kehoe and Perri [126] is to formulate a version of this problem that allows for enforcement constraints. l1 (s t )) + φ2 U (c2 (s t ). The enforcement constraints are expressed as
∞
βr−t π(s r s t )U (ci (s r ). Ai (s r )li (s r )) + (1 − δ)ki (s r−1 ).International Business Cycles
83
where ci (s t ) denotes consumption of residents of country i.
li (s t )) − Vi (ki (s t−1 ). Mi (s t−1 ) satisﬁes Mi (s t ) = Mi (s t−1 ) + µi (s t ). plus the term
∞
βt π(s t )µi (s t )
r=t sr
βr−t π(s r s t )U (ci (s r ).
2 For other applications. Therefore. Thus.
. the Lagrangian function is written as
∞
βt π(s t ) Mi (s t−1 )U (ci (s t ). and an economy with enforcement constraints. s t ) . li (s r )) − Vi (ki (s t−1 ). It is beyond the scope of this book to derive a solution for the model with enforcement constraints. In this expression.23). s t )] plus the terms relating to the resource constraints.24). see Marcet and Marimon [155].2 This approach is implemented by deﬁning βt π(s t )µi (s t ) as the Lagrange multiplier on the enforcement constraints.
Since π(s r ) = π(s r s t )π(s t ). 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 ). t ≥0
with Mi (s t−1 ) = φi . The bond economy stipulates that all intertemporal trades must be implemented with an uncontingent bond. we will focus on the results of simulating this model and compare it with alternatives that have been obtained in the literature.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. Kehoe and Perri [126] generate solutions for three different economies: complete markets. the standard resource constraints deﬁned by (2. The Lagrangian function becomes comprised of three terms: the weighted sum of utilities deﬁned in (2. Mi (s t ) are just the original planning weights plus the sum of the multipliers µi (s t ) along the path s t . a bond economy. li (s t ))
t=0 s t i
+ µi (s t )[U (ci (s t ). Thus.
σ = 2. the crosscountry correlations of employment and investment are negative in the model. the consumption correlations are signiﬁcantly higher than the output correlations in the model.15. whereas they are positive in the data.27)
1−σ
. the three discrepancies between the model and the data remain.9 and the offdiagonal elements are set at zero.025. γ = 0. (2. The coefﬁcients of the A matrix. whereas these correlations are the opposite in the data.25. [25]. Third. According to one parameterization intended to capture substantial persistence.36. l ) = 1−σ and the production function as F (k.26)
The parameter values that are considered are standard. and δ = 0. are parameterized in different ways.36. A second parameterization allows for high persistence with the diagonal elements equal to 0.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. where 0 < b < ∞.99. Under the complete markets speciﬁcation. α = 0. q(s t ) is the period t price of an uncontingent that pays off one unit in period t + 1 regardless of the state.007 and the correlation of the shocks is set at 0. In the bond economy. albeit with some minor improvements in the various statistics. 0 < α < 1. which governs the persistence properties of the shocks and the spillovers between them. net exports and investment are much more volatile in the model than they are in the data. the variance of the innovations to the productivity shocks in each country is set at Var( i. AL) = k α (Al )1−α .99 and the offdiagonal elements equal to zero. Second.85 and the offdiagonal elements equal to 0. There is also a borrowing constraint that requires that ¯ ¯ borrowing cannot exceed some ﬁnite bound. In the economy with enforcement constraints. σ ≥ 0. the output and
. The ﬁndings from the model echo many of the earlier ﬁndings.
0 < γ < 1. and bi (s t ) is the quantity of the bond by consumers in country i. First.
(2.t ) = 0. the model displays many of the anomalies reported in the literature. b(s t ) ≥ −b. and a third one allows for high spillover with the diagonal elements equal to 0. Similar to Backus et al. These values are consistent with those assumed by Kollman [136] and Baxter and Crucini [34]. the autoregressive coefﬁcients of the shocks for each country captured by the diagonal elements of the A matrix are set at 0. Preferences in each country are taken to be of the form c γ (1 − l )1−γ U (c. and given by β = 0.
so resources are optimally shifted there. the responses are similar to those for the complete markets economy. Suppose that the social planner tries to implement
. As discussed earlier. but that the anomalies regarding the crosscorrelations of output. the volatility of net exports and investment has declined dramatically. the frictionless international RBC model implies that investment ﬂows to the country with the higher productivity shock. Thus. In the bond economy. Considering economies with exogenous adjustment costs to investment. consumption. we notice declines in employment and investment. Since residents of each country can acquire contingent claims that allow them to smooth consumption across all possible history of shocks. the crosscountry correlations of employment and investment have now switched from being negative to positive. The only remaining discrepancy is that net exports are procyclical in the model whereas they are countercyclical in the data. Hence. Finally. they ﬁnd that the volatilities of investment and net exports are diminished. but because of restrictions on asset trading. the authors examine the response of the variables in domestic and foreign countries to a positive shock to productivity in the domestic country. Consumption increases substantially more in the domestic country because consumption and labor complement each other. exogenous adjustment costs are typically added to inhibit the ﬂow of investment. let us consider the impact of a positive productivity shock in the enforcement economy. Finally. we observe the high crosscountry correlations between domestic and foreign consumption. risk sharing across countries implies that consumption increases in the foreign country. This increases the productivity of both capital and labor in the domestic country. 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. Kehoe and Perri [126] note that the enforcement constraint acts as an endogenous inhibiting factor. In the foreign country. Second. we see that the differing responses of the domestic versus the foreign country lead to the negative crosscountry correlations of employment and investment. and employment remain.86
Business Cycles: Fact. as domestic residents save more and also as investment from abroad ﬂows there. they are somewhat dampened. leading to very volatile investment and net exports. The net ﬂow of investment increases the trade deﬁcit in the domestic country. investment. and third. In the literature. The capital stock in the domestic country increases.
4. Hence. The increased investment ﬂows to the domestic country further increase the value of autarky. increasing the incentives for default. and the presence of additional shocks have been developed. Furthermore. but also for the behavior of output. Ambler et al. Hess and Shin [119] reexplore the two international business cycle anomalies emphasized by Backus et al. We describe a few of these extensions in what follows. The high and persistent productivity shock in the domestic country increases the value of autarky. investment. the value of autarky diminishes and the relative weight on the home country also falls. the production side. Hence. the complete markets allocation implies that an increase in output in the domestic country will lead to increases in consumption in both the domestic and foreign countries. the 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). models that relax assumptions regarding preferences.International Business Cycles
87
the complete markets allocation for this economy.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. as the productivity shock dies out. 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.3. and the real exchange rate. In terms of the behavior of consumption. [13] modify the supply side of a twocountry model by adding multiple
. In the presence of an enforcement constraint. to ensure that domestic country residents consume more than foreign country residents. OTHER EXTENSIONS
The international RBC literature has sought to reconcile the ﬁndings in the data not only in terms of the crosscountry consumption correlations. this is not possible. 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”. the planner must restrict investment to the domestic country to prevent the default option from being exercised. Over time. As a result. [25] as well as establish the pattern of productivity growth between industries and countries. however.
They conclude that an international business cycle model driven solely by productivity shocks cannot account for the ﬁndings. They argue that variable capacity utilization has the potential to account for the comovement of factor inputs across countries. overstates the crosscountry correlation of aggregate consumption. The model generates a higher crosscountry correlation of output than standard onesector models.88
Business Cycles: Fact. Basu [29]. In their model. It also predicts crosscountry correlations of employment and investment that are closer to the data. and understates the variability of the trade balance. 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]. The authors ﬁnd that the model predicts the correlation between home and foreign output. output in the traded and nontraded goods sector of each country is produced using sectorspeciﬁc capital and labor which is mobile between sectors. Fallacy and Fantasy
sectors and trade in intermediate goods. Hence. Preferences are assumed to be nonseparable with respect to a composite good. they argue that what is needed is a source of nationspeciﬁc shocks that shift demand. However. a nontradable good. Similar to our discussion above. Likewise. and a nontradable good. 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. and leisure. Instead. there is no international capital or labor mobility. Baxter and Farr [35] develop a model with variable capital utilization as a way of accounting for the international correlations. Stockman and Tesar [200] introduce a nontraded goods sector in each country. 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. They employ preferences that depend on consumption of the tradable goods of countries 1 and 2. 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. consisting of the tradable goods of countries 1 and 2. they ﬁnd that introducing nontradable goods does not sufﬁce to reduce the crosscountry correlation of consumption. The notion is that a positive shock to productivity in the domestic country will tend to reduce the investment
. or Basu and Kimball [30]).
many of the puzzles of the international business cycle literature can be addressed successfully by devising a mechanism to limit investment ﬂows across borders. Baxter and Farr [35] assume that preferences are given by the speciﬁcation in equation (2. Kt .258. The covariance properties of the technology processes for the domestic and foreign technology shocks are parameterized to be near unit roots with zero spillover effects. 0 < α < 1. δ > 0. the budget constraint for the representative household is given by Ct + It + qt Bt+1 ≤ Yt + Bt . Zt : St = Zt Kt .26). The correlation of the innovations to the technology shocks is set at 0. which is consistent with earlier studies. and φ < 0. Assuming that the household in each country owns the capital stock and makes real investment decisions. real pure discount bonds which have the price qt = [1 + rt ]−1 . φ > 0. International trade in assets is made solely with oneperiod. 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. and the utilization rate. the variance of the innovations is set so that the variance of output in the model exactly matches the volatility
. Similar functions characterize the foreign country. Output in the domestic country is produced using capital services St and labor Lt as Yt = At St1−α (Xt Lt )α . where Bt+1 is the quantity of bonds carried over into the next period. where δ > 0.International Business Cycles
89
ﬂow across countries by leading to an increase in capital utilization rather than investment. The capital accumulation process displays both costly utilization of capital and adjustment costs: Kt+1 = [1 − δ(Zt )] + φ(It /Kt )Kt . Finally.
where capital services are the product of the capital stock. The parameterization of the capital utilization function and the adjustment cost function are key aspects of the quantitative analysis. As we described above.
See. In their review. variable capital utilization takes the place of investment ﬂows across countries. the crosscorrelation of hours and investment become positive. 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. Obstfeld and Rogoff [168] provide a broader view of the empirical puzzles in the international macroeconomics literature. this is accompanied by increases in investment and employment in the other country. long averages of saving rates and investment rates tend to be correlated for the OECD countries (see Feldstein and Horioka [89]). which is not. In our earlier discussion. the survey by Baxter [33]). Intuitively. in a world of integrated capital markets. which is desirable. Yet. and exports. Speciﬁcally. Second. One of the key ﬁndings regarding the role of variable capital utilization is that.
4. for example. though they also relate their discussion to the international RBC literature.90
Business Cycles: Fact. introducing variable capital utilization reduces the volatilities of consumption. Fallacy and Fantasy
of the US economy.4. Even with a modest crosscountry correlation of the innovations. the volatility of the shocks necessary to match output volatility is substantially reduced. When a positive productivity shock occurs in one country. which in turn increase employment. and also those of hours and employment. for example. The most important impact of variable factor utilization is on the crosscorrelations of the factor inputs. capital. PUZZLES REVISITED
The puzzles in the international business cycle literature have been the topic of much research (see. Obstfeld and Rogoff [168] argue that many of the puzzles in international macroeconomics may just be speciﬁc to the models researchers are using. capital utilization rates increase. ﬁrst. we would expect capital to ﬂow to regions where the rates of return are highest.
•
. The Feldstein–Horioka puzzle: According to this puzzle. McCallum [157] or Helliwell [118]. we described the puzzles that arose when trying to match the international RBC model to the data.
the analysis of Obstfeld and Rogoff [168] is an attempt to unify the explanations of a related but disparate set of ﬁndings. They argue that.2. Clearly. The exchange rate disconnect puzzle: This puzzle captures the notion that there exists a relationship between exchange rates and any macroeconomic variable.
. say. perfect risk sharing across countries implies that countries which experience declines in the relative price of consumption should receive large transfers of goods. the literature that we have described has revealed a variety of promising directions for future research. Backus and Smith [24] show that in an economy with traded and nontraded goods.
Surprisingly. Second. we discussed various explanations that account for this puzzle — the presence of human capital as discussed by Baxter and Jermann [36] or nontraded goods. In Section 4. as we described above. the international portfolio diversiﬁcation puzzle. 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. whereas the other four refer to the behavior of quantities. the last two puzzles are pricing puzzles. whereas the opposite is true in the data. Obstfeld and Rogoff [168] ﬁnd that adding transport costs goes a long way towards accounting for the ﬁrst ﬁve puzzles. The purchasing power parity (PPP) puzzle: The PPP puzzle arises from the fact that shocks to the real exchange rate are very persistent.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. the international consumption correlation puzzle tends to be speciﬁc to a model’s assumptions. They also note that the pricing puzzles require such features as nominal rigidities of the type that we will consider in the next chapter. see Tesar and Werner [203]). and does not have the same weight as. Meese and Rogoff [161] further show that most exchange rate models forecast exchange rates no better than a naive random walk. In contrast to some of the other contributions that we have discussed. ﬁrst.
This page intentionally left blank
.
1 The behavior of hours and productivity has been a topic of debate. 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. and Kimball [32] also present evidence that hours worked and other variables fall in response to technology improvements in the short run. see Christiano.
93
. productivity. output. such as variable factor utilization. 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. Basu. Eichenbaum. from the onset. and other variables.Chapter 5
New Keynesian Models
The New Keynesian approach has gained signiﬁcance in the modern macroeconomics literature. The RBC conclusions regarding the response of hours. output. Gali [96] has argued that in a suitably restricted vector autoregression (VAR) including measures of hours. Their approach involves purging the standard Solow residual of factors that might lead to procyclicality. Fernald. and other variables to a technology shock have been questioned by empirical results obtained along several different lines. Critics of the original real business cycle (RBC) approach had. On the one hand. More recently. taken issue with the notion that prices can adjust costlessly to clear markets. and Evans [66].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. 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. the response of hours to productivity shocks is negative.
respectively. and effort levels Ut to maximize
∞
E0
t=0
βt ln (Ct ) + λm ln
Mt Pt
− H (Nt . hours worked Nt . and then discuss the empirical ﬁndings in more detail. 1 + σn 1 + σu t (1. THE BASIC MODEL
Consider a simple New Keynesian model with monopolistic competition. respectively.1. and variable labor effort due to Gali [96]. 1.2 We ﬁrst describe a simple New Keynesian framework that can be used to rationalize the observations. and Pt =
1 0 1/(1− ) 1− Pit
di
(1.
5. Ut ) is given by H (Nt . Suppose that a representative household chooses consumption Ct . The price of good i is given by Pit .
2 For a further discussion.4)
is the aggregate price index.5)
ϒt and t denote monetary transfers and proﬁts. Wt and Vt denote the nominal prices of an hour of work and effort. . In this expression.
(1. .2)
for t = 0. Ut ) = λn λu Nt1+σn + U 1+σu .1)
subject to
1 0
Pit Cit di + Mt = Wt Nt + Vt Ut + Mt−1 + ϒt +
t
(1. 1] consumed in period t. see Bowman and Doyle [45]. The functional form for H (Nt . and > 1 is the elasticity of substitution among consumption goods.. price rigidities. Ct is a composite consumption good deﬁned as Ct =
1 0
Cit
( −1)/
/( −1)
di
.3)
where Cit is the quantity of good i ∈ [0. . money holdings Mt .
.94
Business Cycles: Fact. Fallacy and Fantasy
and for understanding the role of monetary shocks. Ut )
(1. 2.
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
demand remains unchanged so that ﬁrms will be able to meet demand by producing an unchanged level of output. Fallacy and Fantasy
and output derived above. yt = nt = ξt + γηt + (1 − γ)ηt−1 . and hence a decline in employment will occur. We note that measured labor productivity responds positively whenever ϕ > 1. with a positive shock to technology. producing the same output will require less labor input. The impact of ξt on labor productivity is also transitory.28) (1. However. which corresponds to the situation of shortrun increasing returns to labor. real balances remain unchanged in the face of a positive technology shock.27) (1. given a constant money supply and predetermined prices.98
Business Cycles: Fact.29)
xt = 1 −
+ (1 − γ) 1 −
1 ηt−1 . In such a case. 1 1−γ ξt − ηt . nt . This result can be best understood by considering the case of γ = 0. Finally. Hence.27)–(1.30) (1. 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. though with a oneperiod lag. a positive technology shock has a negative shortrun impact on employment. the price level responds oneforone to an increase in ξt . employment. an increase in ξt causes output and employment to go up for one period and then to revert back to their initial values. that is. This can be observed by noting that the levels of yt . ϕ
where x = y − n is the log of labor productivity. Hence.
. and xt depend only on the current ξt . More interestingly. A positive monetary shock deﬁned by ξt > 0 has a temporary impact on output. The conditions in (1.30) can be used to describe the impact of monetary versus technology shocks. we can show that pt = ξt−1 − (1 − γ)ηt−1 . when there is no accommodating response in the money supply to real shocks. and productivity. ϕ ϕ 1 ϕ ξt + 1−γ + γ ηt ϕ (1. but the sign depends on whether ϕ < (>)1.
By contrast. 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. The identifying assumption is that only technology shocks have a permanent effect on productivity. Second.2. and Vigfusson [68] suggest that the standard RBC results hold if per capita hours are measured in loglevels as opposed to differences. Eichenbaum.
where z and m denote the sequences of technology and nontechnology t t shocks. 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. Christiano. First. This is similar to the approach in Blanchard and Quah [43] for identifying demand versus supply shocks using longrun restrictions on estimated VARs. 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. and McGrattan [63] have challenged the ﬁndings derived from socalled SVARs. Kehoe. The orthogonality assumption.e. Chari. the longrun impact of a given shock. 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. i. Surprisingly.
. they argue that if the simulated data from a simple RBC model are subjected to a SVARbased test. together with a normalization. Furthermore.. technology shocks explain only a small fraction of employment and output ﬂuctuations. implies that E ( t t ) = I .New Keynesian Models
99
5. which can be expressed as the restriction C 12 (1) = 0. respectively. 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 . 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.3 Gali [96] estimates this model using postwar US data.
100
Business Cycles: Fact. They ﬁnd that purging the standard Solow residual of these effects eliminates the phenomenon of “procyclical productivity”. These ﬁndings have. on the one hand. nondurables and services. Basu et al. if hours worked is considered to be stationary in levels. However. and various prices and interest rates to changes in the puriﬁed Solow residual. generated substantial controversy and. 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. utilization. They also examine the response of a key set of variables such as output. in their view. 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. They use both standard regression analysis and simple bivariate VARs for this purpose. They also uncover further evidence for the negative response of nonresidential investment to such shocks. [32] present evidence that support the SVAR ﬁndings by generating a modiﬁed Solow residual that accounts for imperfect competition. hours worked. leads to an erroneous conclusion of the negative response of hours worked to a productivity shock. variable factor utilization.
. This misspeciﬁcation. durables and residential investment. Their ﬁndings corroborate the ﬁndings from the SVAR approach regarding the negative response of hours to technology improvements in the short run. and sectoral reallocation and aggregation effects. they advance price rigidity as the major reason for these deviations from the RBC predictions in the short run. on the other. Unlike the SVAR approach. 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. nonresidential investment. Following Gali [96] and Gali and Rabanal [98]. 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. nonconstant returns to scale. Fallacy and Fantasy
of hours to a shock to productivity is negative as in Gali and Rabanal [98]. the evidence obtained from this approach is robust to longrun identifying assumptions or to the inclusion of new variables in the estimated dynamic system. employment.
consumption varies more than output. Arellano and Mendoza [17]). Aguiar and Gopinath
101
. and income and exports are typically highly volatile. Kydland and Zaragaza [144] also study the behavior of an emerging market economy. On the one hand. More recently. 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. 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. namely. Business cycles in emerging markets exhibit different characteristics compared to those in developed economies. and Rebelo [77]. for example. but their analysis is based on the onesector optimal growth model. For instance. Argentina. Neves. Earlier RBC models for small open economies were developed by Mendoza [162] and Correia. for example. On the other hand. Emerging market business cycles also display a different set of stylized facts relative to developed economies. Ratfai and Benczur [175.Chapter 6
Business Cycles in Emerging Market Economies
In recent years. researchers have begun generating business cycle facts for such economies (see. The question then arises whether a small open economytype RBC model can account for both developed and emerging market business cycles. they have been examining the efﬁcacy of RBCtype models in accounting for these facts. the real business cycle (RBC) agenda has been increasingly applied in emerging market contexts. 176]). the trade balance is strongly countercyclical.
g t. and µg denotes average longrun productivity growth. 0 < α < 1. with E ( z ) = 0 and Var( z ) = σz . 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.2)
2 where { z }∞ is distributed as i.
(1. The shock zt represents the transitory component of productivity.
ρz  < 1. 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.i. By contrast. here modeled as changes in trend productivity growth. t t=0 t t The permanent shock to productivity evolves as t t
= gt
t−1
=
s=0
gs .d. developed economies typically face stable political and economic policy regimes so that changes to productivity are transitory.102
Business Cycles: Fact. Since productivity and output depend on the cumulation of the shocks gt . and evolves as a stationary AR(1) process: zt = ρz zt−1 +
z t. Pancrazi. In their view. a stationarityinducing transformation is used to remove the stochastic trend from all variables as xt xt = ˆ .
g t)
ln (gt ) = (1 − ρg ) ln (µg ) + ρg ln (gt−1 ) +
g ∞ t }t=0 g t)
(1. (1. with E ( = 0 and Var( = σg .3) ρg  < 1. Thus. let output be produced according to the Cobb– Douglas production technology as Yt = exp (zt )Kt1−α ( t Lt )α . emerging market economies’ experience can be distinguished by the large number of regime shifts.d.1)
where {zt } and { t } represent two alternative productivity processes. gt denotes the shocks to the growth rate of productivity. To describe the model.
(1.
6. Fallacy and Fantasy
[1] and GarciaCicco.1.4)
2 where { is distributed as i.i. t−1
.
.5)
Here.} 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 . σ ≥ 0. . The ﬁrst is from the class of socalled GHH (after Greenwood. (1. . 1−σ ν > 1.}}
1
1 1−σ ˆ ν ν = E0 {(C0 − τL0 )1−σ + βg0 (C1 − τL1 )1−σ 1−σ ν ˆ + βg0 E1 {βg 1−σ (C2 − τL2 )1−σ .}}. and the intertemporal elasticity of substitution is given by 1/σ.6)
The expected discounted utility of the representative consumer can be written as 1 ν ν ˆ E0 {(C0 − τL0 )1−σ + β 1−σ (C1 − τL1 )1−σ 0 1−σ ν ˆ + β2 1−σ (C2 − τL2 )1−σ .
1
where the last line is obtained by substituting recursively and making use of an iterated expectations argument. .
(1. . 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 . preferences are given by ut = (Ct (1 − Lt )1−γ )1−σ . Hercowitz.7)
This shows that investment is subject to quadratic costs of adjustment. . Thus. In the Cobb–Douglas case. the elasticity of labor supply is given by 1/(ν − 1).Business Cycles in Emerging Market Economies
103
Aguiar and Gopinath [1] consider two types of preferences over consumption and leisure. (1. The GHH preferences are deﬁned as ut = (Ct − τ t−1 Ltν )1−σ . 1−σ
γ
0 < γ < 1. expected discounted utility remains bounded provided βE (gt ) = βµg < 1. τ > 0. and that the country can borrow using oneperiod debt that sells for the price
. and Huffman [105]) and the second are standard Cobb–Douglas preferences.
b represents the steadystate level of debt.104
Business Cycles: Fact. 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. gt+1 )}. gt ) is given by γ(1−σ) βgt1−σ in the former case and by βgt in the latter. Lt ) + f (β. f (β. ˆ ∂Kt+1 (1. Lt ) is deﬁned by (1.6) in the case of Cobb–Douglas preferences. the value function deﬁned in terms of the transformed variables is given by ˆ ˆ V (Kt . and Lt are ˆ uc (Ct . Lt ) gt + φ gt ˆ Kt+1 − µg gt ˆ Kt ∂V ˆ ∂Bt+1 = f (β. Bt+1 . gt )Et
ˆ ˆ uL (Ct . gt )Et = 0. Lt )gt qt + f (β. the ﬁrstorder conditions ˆ ˆ with respect to Kt+1 . 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 . The model expressed in terms of the transformed or “hatted” variables is solved by using standard recursive methods. Bt . 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 .9) (1. Furthermore. Lt ) + uc (Ct . Bt+1 .Kt+1 . and ψ > 0 governs the elasticity of the interest rate to changes in indebtedness. zt . gt )Et V (Kt+1 .Lt . Lt )
ˆ ∂Yt = 0.5) in the case of GHH preferences and by (1.
ˆ where u(Ct . ∂V . when choosing how much debt to take on. Likewise. ∂Lt
.
ˆ Substituting for Ct using the resource constraint. qt t (1. Assuming it exists.11)
ˆ uc (Ct .10) (1. zt+1 . gt ) =
ˆ ˆ ˆ Ct . Fallacy and Fantasy
qt .Bt+1
max
ˆ ˆ ˆ {u(Ct .8)
where r ∗ is the world interest rate.
are estimated using a generalized method of moments (GMM) approach that we describe in detail in the following chapter.4). ˆ ∂ Bt For simplicity.2)–(1.
ˆ ˆ ∂V (Kt . 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 . a subset of the parameters is set at values determined a priori. and the equation describing the real interest rate (1. zt . Lt ) + (1 − δ) ˆ ˆ ∂ Kt ∂ Kt + φ gt ˆ ˆ ˆ Kt+1 Kt+1 φ Kt+1 − µg gt − − µg gt ˆ ˆ ˆ 2 Kt Kt Kt
2
.
. consider only the case with GHH preferences. Lt ).
(1. Bt .8). In the recent applications. zt .1).13)
. the resource constraint (1.7).14)
subject to the production function (1. The solution for the model is obtained by implementing a loglinear approximation to the equilibrium conditions described above.Business Cycles in Emerging Market Economies
105
The envelope conditions are given by ˆ ˆ ˆ ∂Yt ∂V (Kt . the laws of motion for the shocks (1. including the parameters of the shock processes. gt ) ˆ = uc (Ct . gt ) ˆ = −uc (Ct . 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 . gtσ exp (zt )αgtα ˆ Kt Lt
1−α
(1. Bt .12) (1. The remainder of the parameters.
the autocorrelations of transitory shocks are roughly similar. This leads to a large trade deﬁcit which tends to persist for a considerable period of time (16 quarters).41 for Canada depending on the speciﬁcation for preferences. [101]. In particular.5 or 5. Thus. as well as the correlations of the latter three with output.25 or 0. These ﬁndings have been challenged by GarciaCicco et al.2. net exportstoGDP. By contrast. By contrast. They also consider the mean and standard deviation of (unﬁltered) income growth as well as the autocorrelations of (ﬁltered) income and (unﬁltered) income growth. 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. φ). investment. is 0. The most noteworthy ﬁnding that emerges from the estimation is that the ratio of shock standard deviations. who argue that the results of Aguiar and Gopinath [1] are due to their use of
. a positive shock to trend productivity causes consumption to respond more than output in the expectation that income will be higher in the future. ρg .106
Business Cycles: Fact. they show that a positive transitory shock to productivity reduces consumption in anticipation of lower income in the future.4 for Mexico. ρz . The parameters to be estimated are given by (µg . and 2. This yields 11 moment conditions. 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 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. Fallacy and Fantasy
6. σg . This ratio captures the importance of shocks to trend productivity. σg /σz . as is the capital adjustment parameter φ. consumption. the shock to trend productivity can generate consumption booms that tend to appear sidebyside with current account deﬁcits in emerging market economies. σz . 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. DO SHOCKS TO TREND PRODUCTIVITY EXPLAIN BUSINESS CYCLES IN EMERGING MARKET ECONOMIES?
Aguiar and Gopinath [1] consider two versions of their model.
As we described earlier. they consider the variances and ﬁrst. and less volatile than output growth if the opposite is true. this is not the case for the autoregressive coefﬁcient for the transitory shock.Business Cycles in Emerging Market Economies
107
a short sample to estimate lowfrequency movements in productivity. The standard deviations of the shocks and the autoregressive parameter for the permanent shock are estimated precisely. Whereas the ﬁrst result suggests that the estimated model does not emphasize the role of permanent shocks sufﬁciently. Second. which is not signiﬁcantly different from zero. the correlations of output growth with consumption growth. consumption growth. and estimate the parameters of the stochastic processes for the permanent and transitory shocks using 16 moment conditions. The authors also ﬁnd that the trade balancetooutput ratio is estimated to be around four times as volatile as output growth. In particular. and the unconditional mean of output growth. investment growth. and the trade balancetooutput ratio. investment growth is found
. However. 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]. GarciaCicco et al. First. They determine this empirically. and the trade balancetooutput ratio. [101] use annual data for Argentina over the period 1913–2005. [101] argue that a similar ﬁnding is not true for an emerging market economy such as Argentina. consumption growth in the estimated model is calculated to be less volatile than output growth. GarciaCicco et al. [101] ﬁnd that the overidentifying restrictions of the model are rejected. GarciaCicco et al. in terms of the parameter estimates.and secondorder autocorrelations of output growth. unlike Aguiar and Gopinath [1]. much of the business cycle literature for developed countries has concentrated on the postWorld War II period. and. whereas in the data these quantities are roughly equal. consumption growth will be more volatile than output growth if permanent shocks are more important than transitory shocks. the second result suggests that it overemphasizes them. Given the evidence that business cycles have moderated during this period. As we described above. in contrast to what is observed in the data. they show that output ﬂuctuations in the postWorld War II period are as large as those in the preWorld War II period. By contrast. this choice of sample period does not seem out of line. and ﬁnd that the consumptionsmoothing motive in response to a transitory shock dominates the anticipatory effect of consumption to a permanent productivity shock. Third. investment growth. In particular.
108
Business Cycles: Fact. In particular. If the shocks are permanent. consumption increases in response to an innovation to output in roughly the same magnitude as output. regardless of whether shocks to productivity are permanent or transitory. The authors show that the estimated model also fails to replicate the autocorrelations of output growth and the trade balancetooutput ratio. GarciaCicco et al. 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. as households perceive the income increase to be permanent. 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.
. with a constant world interest rate. they argue that some of their ﬁndings that we described above point to the importance of shocks that are different from productivity shocks. However. if the shocks are transitory. Fallacy and Fantasy
to be insufﬁciently volatile. the trade balance is unaffected and the trade balancetooutput ratio inherits the behavior of output. Hence. suggesting in this case that neither source of shocks is sufﬁciently volatile. the behavior of the trade balancetooutput ratio again follows a near random walk because in this case. In particular. consumption in the model follows a near random walk even though output and investment become stationary variables. [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. By contrast. their ﬁndings cannot be used to disentangle which of these factors is responsible for the failure of the model to replicate the observations. Hence.
In this chapter. 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. which assesses the adequacy of the model based on the behavior of the relative variability and comovement of
109
. Watson [209] extended this approach to derive measures of ﬁt for an underlying economic model. 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. 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. and also provide an overview of the estimation versus calibration debate. Their approach may be viewed as an extension of the standard RBC approach. 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. 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. 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. Amongst the most prominent of these assumptions is the assumption of perfect price ﬂexibility. In her original contribution. We also discussed variations of the model that could be used to account for speciﬁc correlations in the data.Chapter 7
Matching the Model to the Data
In the previous chapters. we will study alternative approaches for matching the implications of a theoretical model with the data.
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. n for i = j.t ) = 0 E (˜ i. 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. covariance stationary stochastic ˜ t=0 process used to describe observations on the (possibly detrended) values of a set of variables. .t−s ) = 0 for all ν ˜ i. let {wt }∞ denote an ndimensional mean zero.t νi. Finally.
(1.
7. Fallacy and Fantasy
a small set of moments. Under these assumptions. . 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. that is.3)
Both the common factors and the idiosyncratic factors may be serially correlated. To describe how to formulate unobservable index models. A kfactor unobservable index model for wt is given by ˜
∞
wt = ˜
s=−∞
˜ ˜ H (s)f˜t−s + νt .1. t = s.t ) = 0 ν ˜ for i = 1. 57] discusses alternative approaches for conducting statistical inference in calibrated models. (1.2) (1.110
Business Cycles: Fact. . . E (f˜t f˜t−s ) = 0 for t = s and E (˜ i. we require that ˜ E (f˜t νi. According to this model. j. Canova [56.1)
˜ ˜ where {H (s)}∞ s=−∞ is a sequence of (n × k)dimensional matrices.t νj. More precisely. 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
. and νt is an n × 1 vector of idiosyncratic shocks ˜ that are mutually uncorrelated and uncorrelated with common factors. ft is a k × 1 vector of common factors.
notice that offdiagonal elements of Sw (ω) are. n.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 . 1.
=
s=−∞
An alternative representation of the autocovariance function is provided by the spectral density function
∞
Sw (ω) =
r=∞
Rw (r)e −irω .4)
Assuming that Sw (ω) is nonsingular at each frequency ω. . ˜ ˜ r = . . . complex numbers. Under the assumptions ˜ underlying (1. However.
be the autocovariance function of {wt }. in general. . −1.
ω ≤ π.Matching the Model to the Data
111
the idiosyncratic shocks.
∞
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). . . since Rxl xh (r) = Rxl xh (−r). . .1). h = 1. the diagonal elements of Sw (ω) are real.1
(1. Substituting for Rw (r) into (1.
1This function is welldeﬁned as long as
. . 0. . Let Rw (r) = E (wt (wt+r ) ).
He discusses a classical measurement error case.i.i. Another wellknown application of this approach is due to Altug [5]. The unrestricted version of the dynamic factor model does not place restrictions on ˜ the matrices H (s).d. and the idiosyncratic shocks are interpreted as i. this model must be estimated jointly across
. it is not possible to identify ˜ the common factors with different types of shocks to the economy. as in Altug [5]. Hence. Also. The dynamic factor model can be estimated and its restrictions tested across alternative frequencies using a frequency domain approach to time series analysis. measurement errors or idiosyncratic components not captured by the underlying RBC model. Altug also uses this representation to estimate the model using maximum likelihood (ML) estimation in the frequency domain. as well as a case with orthogonal prediction errors. error terms. 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). 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.112
∞
Business Cycles: Fact. The common factor is identiﬁed as the innovation to the technology shock. ˜ ˜ where H (ω) denotes the Fourier transform of H (s). The restricted factor model makes use of the crossequation restrictions across the linear decision rules implied by the original model. which describe how the common factors affect the behavior of the elements of wt at all leads and lags.d. Unlike the unrestricted factor model which can be estimated frequency by frequency. Sargent [184] also employs the device of augmenting a singular model with additional idiosyncratic shocks. The use of the dynamic factor model in business cycle analysis dates back to the work of Sargent and Sims [185]. 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. As these authors observe. the dynamic factor model provides decomposition at each frequency that is analogous to the decomposition of variance in the conventional factor model.
model.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. and aggregate output. when the restrictions of the underlying model are imposed.1. 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.
denotes the stochastic process generated by some underlying Here. Also. the joint process for the data and the error is introduced to motivate goodnessofﬁt measures. let xt denote an n × 1 vector of covariance stationary random variables.
7. not to describe a statistical model that can be used to conduct statistical tests. To describe Watson’s approach.
. Instead. investment in structures. and Ax summarizes the unconditional secondmoment properties of this model. 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.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. Watson’s analysis does not depend on assuming that the unobserved factors and the idiosyncratic shocks are uncorrelated. However. The ACGF for yt is similarly denoted Ay (z). In the data. 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. the model cannot explain the cyclical variation of the observed variables. see Hansen and
{xt }∞ t=0
Sargent [114]. investment in equipment.1. the vector of variables yt corresponds to the empirical counterpart of xt . Deﬁne the autocovariance generating function (ACGF) for xt by
∞
Ax (z) =
r=−∞
E (xt xt−r )z r . deﬁne the n × 1 vector of
2 For further discussion of maximum likelihood estimation in the frequency domain. The question at hand is whether the data generated by the model are able to reproduce the behavior of the observed series. Unlike the approach adopted by Altug and Sargent.
3 To continue. in addition. it is the size of the sampling error that is used to judge whether
. Nevertheless.t. or σxy  ≤ σx σy :
2 2 2 min σu = σy + σx − 2σxy s. In the standard dynamic factor model.4 However. the error term ut is the approximation error in describing the observed data with the underlying economic model.5 In the ﬁrst case. However. Ay (z) = Ax (z). To calculate Au (z). let us consider two cases. If. This implies a version of a classical errorsinvariables approach. Fallacy and Fantasy
errors ut that are required to reconcile the ACGF implied by the model with that of the data. and Ax (z) from the model. notice that Ay (z) can be determined from the data. Axy (z) = 0. yt . σxy  ≤ σx σy . and the size of the sampling error can be determined from the datagenerating process for yt . and ut are assumed to be serially uncorrelated scalar random variables. the assumption regarding Axy (z) is that it is zero. a lower bound may be deduced for the variance of ut without imposing any restrictions on Axy (z). σxy
(1. Then.6) (1.114
Business Cycles: Fact.7)
ˆ a given model ﬁts that data. 5 We omit the third case that Watson considers because it requires results for the Cramer representation of stationary time series. which implies that the ACGF for ut can be expressed as Au (z) = Ay (z) + Ax (z) − Axy (z) + Ayx (z). 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). 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. To illustrate this approach.5)
where Axy (z) is the joint ACGF for xt and yt . Hence. since Axy (z) is unknown. the difference between Ay (z) and A error. 4This is similar to the interpretation provided by Altug [5]. 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. In this case. some additional assumptions are needed to operationalize this approach. in Watson’s framework.
3 In a standard goodnessofﬁt approach. xt . let Ay (z) denote the population autocovariance function and Ay (z) ˆy (z) is ascribed to sampling denote its estimated counterpart. the assumption of pure measurement error which is uncorrelated with the true variable is not appropriate. (1. deﬁne ut by the relation ut = yt − xt .
9)
In this expression. yt ) is positive semideﬁnite. which minimizes the weighted sum of the variances of the elements of ut subject to the constraint that this matrix is positive semideﬁnite. Furthermore. we cannot minimize the variance of ut directly. given the properties of the U and V matrices. Instead. 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 ). The covariance matrix of ut is given by u = y + x − xy + yx . is given by
xy
= Cx VUCy . In this case. where ij. An implication of this result is that. yt ) is singular. 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.u denotes the i. the joint covariance matrix of (xt . (1. Watson [209] uses this methodology to generate goodnessofﬁt measures for a standard RBC model with a stochastic trend in technology. we consider a transformation that allows us to determine the size of ut . One convenient transformation is the trace of u . Now suppose that xt and yt are serially uncorrelated random vectors with covariance matrices x and y . xt can be represented as xt = yt . Speciﬁcally. The problem in this case is to choose xy to minimize tr(W u ) subject to the constraint that the covariance matrix of (xt .Matching the Model to the Data
115
It is easy to see that the solution for this problem is to set σxy = σx σy . xt and yt are perfectly correlated so that xt = γyt . For k = 1. j element of . As a consequence. 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 . Proposition 1 in Watson [209] demonstrates that the unique matrix xy . and Rebelo [131. He loglinearizes
.8)
where γ = σx /σy . where W is an n × n matrix. as in the scalar case. this result corresponds to the one in the scalar case.
(1. tr( u ) = n i=1 ij. An alternative approach is to minimize the weighted sum of the variances as tr(W u ). V is a k × k orthonormal matrix (with VV = Ik ).10) where = Cx UVCy−1 . As a 2 consequence. and S is a k × k diagonal matrix. σu = (σy − σx )2 at the minimum. 132].u . Plosser. he considers the model in King. (1.
1. and investment occur at frequencies corresponding to the business cycle periodicities of 6–32 quarters.6 He ﬁnds that the biggest differences between the spectra for output. Giannone.
6 A comparison of the spectra implied by the model and those generated by actual data also ﬁgured in early versions of Altug [5]. His approach allows for a decomposition by frequency of the variance in each observed series. see also Altug [4]. the economywide shocks can be identiﬁed using structural vector autoregression (SVAR) techniques. their results also indicate that sectorspeciﬁc shocks are needed to explain the variance of the series. and the error in reconciling the model with the data.
. investment. 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. 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. consumption. that is. models for which the number of exogenous shocks is less than the number of series. Based on a law of large numbers argument. Reichlin.
7. they show that the number of common factors can be determined using the method of principal components. consumption. Fallacy and Fantasy
the ﬁrstorder conditions to obtain the solution for logdifferences of output. However. the variance explained by the model.116
Business Cycles: Fact. 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. 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 number of hours worked. and the unobserved factor model can be estimated by using equationbyequation ordinary least squares (OLS). suggesting that the stochastic trend properties of the model do not hold in the data. They also describe how to derive impulse response functions for time series models which have reduced rank.2. both having a longrun effect on sectoral output. Other Applications
Forni and Reichlin [93] use the dynamic factor model to describe business cycle dynamics for large crosssections.
(2.2. σµ . Likewise. and the share of capital in the neoclassical ¯ production function.
Proceeding in this way. 1 = (δ. σλ ) . Since there is a stochastic trend in this economy arising from the nature of the technology shock process. θ.
given data on gross investment it and the capital stock kt+1 . Their approach may be viewed as an extension of the standard RBC approach.11)
. technology. which assesses the adequacy of the model based on the behavior of the relative variability and comovement of a small set of time series. the share of capital satisﬁes the Euler equation E β−1 θ yt+1 kt+1 +1−δ
1
ct ct+1
= 0. Their estimation strategy is based on a subset of the ﬁrst and second moments implied by their model. and the exogenous stochastic processes. let 1 denote a vector of parameters determining preferences. More generally. ρ.Matching the Model to the Data
117
7. As in the standard RBC approach. the deterministic steady states are derived for the transformed variables. GMM ESTIMATION APPROACHES
Other papers have employed nonlinear estimation and inference techniques to match equilibrium business models with the data. θ. 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. the parameters in 1 are estimated using simple ﬁrstmoment restrictions implied by the model. γ. g . δ. For example. the elements of restrictions E H1t (
1)
satisfy the unconditional moment = 0. 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. λ. the depreciation rate δ is set to reproduce the average depreciation on capital as E δ− 1− kt+1 it − kt kt = 0. To describe how their approach is implemented. Some of the parameters included in 1 may be the depreciation rate of capital.
f2 (
1 )]
(2. n) − (y/n)t nt = 0. it . gt .
where ct denotes private consumption. For any parameter vector 1 .
2 2 E nt − σn = 0. corr(y/n. nt .
(2. σn /σy/n . the average productivity of labor. Deﬁne = [ 1 .
(2. let f(
1)
= [f1 (
1 ). it .18)
. gt .13) (2. Fallacy and Fantasy
2 consist of the standard RBC secondmoment restrictions as
The elements of
E yt2 (σx /σy )2 − xt2 = 0.
2 σn /(σn /σy/n ) corr(y/n.17)
represent the model’s restrictions for corr(y/n. To do this.12) (2. they use a Waldtype test based on the orthogonality conditions implied by the relevant unconditional moments. and the relative variability of hours versus average productivity.14) (2. labor hours. 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. n). F ( ) = 0.16)
Christiano and Eichenbaum [65] are interested in testing restrictions for the correlation between hours and productivity.118
Business Cycles: Fact. private investment. σn /σy/n ]
(2.15)
E (y/n)2 σn /σy/n t E
2 − nt = 0. n). let A be a 2 × k matrix of zeros and ones such that A and F( ) = f (
1) − A
= [corr(y/n. n) and σn /σy/n .19)
Under the null hypothesis that the model is correctly speciﬁed. The restrictions of the model can be summarized as E H2t (
2)
= 0. 2
x = ct . and (y/n)t .
(2. (2. public consumption.20)
. Also. 2 ] as the k × 1 vector containing the true values of the parameters and second moments for the model.
Eichenbaum [82] asked whether RBC analysis.
7. once we quantify the uncertainty in model predictions arising from uncertainty about model parameter values. Aguiar and Gopinath [1] and GarciaCicco. the containing T observations. Using data on private consumption. in fact. our view of what the data is [sic] telling us is affected in a ﬁrstorder way. the GMM approach has been used in other recent applications. the authors show that the statistic J = F ( ˆ T ) Var[F ( ˆ T )]−1 F ( ˆ T ) (2.7 In this section. Since then. Letting test statistic for the secondmoment restrictions is based on the distribution of F ( ˆ T ) under the null hypothesis.
. In a highly suggestive analysis. Pancrazi. constituted “wisdom or whimsy”. Even if
7 See Kydland and Prescott [142. 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. and Uribe [101] employ this approach in their analysis of business cycles in emerging markets.Matching the Model to the Data
119
In practice. there have been ongoing discussions on both sides of the issue. aggregate hours. government expenditures. 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. 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. calibrated or otherwise. Using this distribution. 143] for a further discussion and defense of their approach. For example. Kydland and Prescott [141] initiated this debate in the modern business cycle literature.3.21)
is asymptotically distributed as a χ2 random variable with two degrees of freedom. investment. we will deal with a variety of criticisms aimed directly at the calibration approach and some suggested alternatives to it. As we described earlier. In his words:
Indeed. there is sampling error in estimating from a ﬁnite data set ˆ T denote the estimated value of .
its power spectrum. They compare the autocorrelation function and the impulse response function in the data with those generated by the model using generalized Q statistics. deteriorate considerably when a break in the sample is allowed for. To estimate impulse response functions from the data. and impulse response functions in response to shocks.
7. As part of their diagnostics. whereas government shocks evolve as a persistent AR(1) process. 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. They consider a twovariable VAR with output and hours worked (or consumption). The answer could be 70% as Kydland and Prescott [142] claim. and identify the technology shock under the assumption that it has a permanent effect on output.
Eichenbaum [82] also presented evidence to show that the performance of the standard model as well as modiﬁcations that allow for labor hoarding. 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
. they use the twoshock version of the RBC model proposed by Christiano and Eichenbaum [65]. What the data are actually telling us is that.120
Business Cycles: Fact. the authors use the SVAR technique developed by Blanchard and Quah [43].3.1. which includes shocks to productivity and government consumption as well as the indivisible labor feature of Hansen [112] and Rogerson [179]. while technology shocks almost certainly play some role in generating the business cycle. 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. Since the single technology shock model is singular. but the data contain almost no evidence against either the view that the answer is really 5% or that the answer is really 200%. for example. Such a break accounts for the slowdown in productivity growth in the US that occurred in the late 1960s. they consider the autocorrelation function for output. there is simply an enormous amount of uncertainty of just what percent of aggregate ﬂuctuations they actually do account for. They assume that the technology shocks are differencestationary.
β) denote the density of the vector Xt .33–7 years per cycle. complementary to those regarding the role of ﬁltering on the cyclical
properties generated from a standard RBC model. let Xt = f (Zt . f ) denote the density for the parameters β. Cogley and Nason [70] also argue that the model displays weak propagation mechanisms which arise from intertemporal substitution. but it is unable to match the impulse response function for the transitory component. conditional on the information set available to the researcher I and the function f . as is the power spectrum. βI .3. and the spectrum for output displays signiﬁcant power at the business cycle frequencies of 2. To describe his approach.8 By contrast.2. f ) = H (Xt . f )d β. and interest rates. Deﬁne expectations of the functions of the simulated data
8These results are. Denote the predictive density p(Xt I . say. Also. 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. 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.
7. Calibration as Estimation
Canova [56. in some sense. GDP. The test statistic also shows that the implications of the model are rejected at signiﬁcance levels of 1% or less. 57] proposes an alternative approach for providing statistical inference in calibrated models. and adjustment costs. let Xt denote a vector stochastic process with a known distribution that describes the evolution of a set of observed variables. investment. and let H (Xt .
. See Cogley and Nason [69]. the autocorrelations of output at lags 1 and 2 are signiﬁcant and positive. βf . 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. capital accumulation.Matching the Model to the Data
121
for output implied by the model is nearly ﬂat. let π(βI . In particular. I ) denote the joint density for the data and the parameters. conditional on the function f and the parameters β. Let G(Xt f .
For example. β) or its approximation. for
. • For each drawing of β and zt . While the densities H and p are unknown.22)
This approach allows for a probability statement regarding the behavior of various moments implied by the model. βI . where ν(Xt ) is some moment of interest and D is a bounded set. generate {xt }T and compute µ(xt ) using t=1 the model xt = f (zt . or estimation techniques. Notice also that the function f is typically unknown and must be obtained using some approximation method. 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. given the information I available to the researcher and a density k(Zt ) for the exogenous processes. If such information is hard to obtain. one could count estimates of elements of β obtained from alternative studies and smooth the resulting histogram to obtain the density π(βI . deﬁne µ(Xt ) = 1 if ν(Xt ) ∈ D and zero otherwise. Simulation exercises conducted after a subset of the parameters has been estimated (using GMM. This information may be derived from alternative data sets. the model can be simulated repeatedly for different values of β and Zt to compute sample paths for Xt . • Construct the frequency distribution of µ(xt ) and other statistics of interest that can be used to evaluate the performance of the model. I ) = = µ(Xt )p(Xt I . the approach that Canova advocates for conducting statistical inference in calibrated models is as follows: • Select a density π(βI . f ) and zt from k(Zt ). f )dXt H (Xt . In general. The calibration approach involves putting a point mass on a particular value of β. More precisely. model speciﬁcations. f )d βdXt . Then. f ) for the unknown parameters.122
Business Cycles: Fact. then one can use a uniform distribution. Canova [56] describes how to account for such an approximation error when computing the moments of the simulated data. (3. • Repeat the above two steps N times. f ). f ) by E (µ(Xt )f . Fallacy and Fantasy
(denoted by µ(Xt )) under the predictive distribution p(Xt I . • Draw vectors β from π(βI . suppose that we are interested in evaluating Pr(ν(Xt ) ∈ D).
Ashley. In a novel analysis. or variable capacity utilization. Such nonlinearities may take the form of conditional heteroscedasticity such as ARCH or GARCH effects. Using an array of diagnostic tests. Neftci [166] was among the ﬁrst to demonstrate that unemployment ﬂuctuations were asymmetrical along the business cycle. say. variable capacity utilization. and while they could not ﬁnd evidence of chaotic behavior.
. Yet there is a new literature that shows that macroeconomic time series may exhibit marked nonlinear behavior. there may exist asymmetries in various economic variables.3. Nonlinearity in Macroeconomic Time Series
Another criticism of the RBC model literature is that it has typically been concerned with examining the ﬁrst. they nevertheless showed that postwar employment. and industrial production could be described as nonlinear stochastic processes. and timevarying markups. and argued that any reasonable macroeconomic model should display some form of nonlinear dynamics (see Potter [172] for a review). Altug. 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. They conclude that the nonlinearity must lie in the transmission mechanism.
7. They found strong evidence of nonlinearity in industrial production. we discussed the role of alternative factors that could give rise to endogenous changes in productivity such as labor hoarding.3. Ashley and Patterson [22] developed a test to test for deviation from linear stochastic processes. In our discussion in Chapter 3. increasing returns. unemployment. GMM. and Patterson [9] examine the implied behavior of output. markups. they note that observed measures of output and factor inputs such as the capital stock and hours worked display marked nonlinear behavior. they do not ﬁnd evidence of nonlinearity in the Solow residuals measured after allowing for increasing returns to scale. either in the form of nonlinear stochastic dynamics or deterministic chaos. Brock and Sayers [47] tested real macroeconomic variables displayed by deterministic chaotic dynamics. the factor inputs.Matching the Model to the Data
123
example) involve putting a point mass on β∗ estimated according to a particular method. and the underlying productivity shocks using a simple production function framework.and secondmoment properties of aggregate economic variables for the purpose of matching a model to the data. Alternatively. Speciﬁcally.
However. The statistical models are ﬂexible enough to allow for increasing time dependence in the mean of the process (captured through a VAR). 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.. Valderrama [207] selects three statistical models to describe the properties of the data. kurtosis. Hercowitz. 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. skewness). Valderrama [207] considers a simple Brock–Mirmantype growth model with GHH preferences (see Greenwood. 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. and the UK. 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. and excess volatility (i. GARCH). The objective function is distributed as a X 2 statistic. He argues that standard general equilibrium models are able to replicate the ﬁrst.and secondmoment properties of such variables. a seminonparametric (SNP) model is estimated to characterize the statistical properties of the data.. excess kurtosis).124
Business Cycles: Fact. The SNP approach nests standard VARs. and it allows for a nonGaussian error term. The SNP approach uses a standard VAR to model the conditional mean and an ARCHGARCH structure to model the conditional variance. He poses this as a “canonical” challenge to the RBC approach. as in the GMM approach.e. the economic model is simulated for a given set of parameters. Then. In the second step. A value iteration approach is used as the solution procedure. and shows that nonlinearities such as skewness. and conditional heteroscedasticity are common for many of these aggregates. The ﬁrst model selected by the SNP is a VAR(3) with conditional heteroscedasticity and a Hermite
. Italy. and Huffman [105]) and adjustment costs in investment.e. it also allows for periods of high volatility followed by low volatility (conditional heteroscedasticity). 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]. In the ﬁrst step. Mexico. for conditional heteroscedasticity (ARCH. Japan. but they are unable to reproduce nonlinearities in these time series. asymmetric business cycles (i.
Matching the Model to the Data
125
polynomial of degree 4. Two other models are selected for the purpose of describing the data. and arrive at results that indicate little or no signiﬁcant impact of irreversibility or lumpiness on aggregate investment dynamics. Valderrama [207] ﬁnds that the biggest difference among the three statistical models is in terms of the adjustment cost parameter. Veracierto [208] argues that the reason why irreversibility has an impact on aggregate investment in various
. the other two statistical models allow for these features and hence do not require such a high adjustment cost parameter. it can capture the nonlinearity in investment. 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. When the statistical model is forced to be a linear VAR(3). implying that the irreversibility feature is not important for matching the model to the data. By contrast. this parameter is estimated to be around 10. This is similar to some earlier results in the literature. Surprisingly. Valderrama [207] ﬁnds that the irreversibility constraint does not bind during the solution of the model. but not the nonlinearity in consumption. 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. which allows for nonlinearity in terms of the ARCH effects but continues to assume that the errors are Gaussian. The second model is a VAR(3) with ARCH(1) errors. its estimates are around 3. Combined with the ﬁndings of Altug et al. Valderrama ﬁnds that the RBC model is not successful at generating the conditional variance of investment. [9] who show that the nonlinearities in the observed series must lie in the propagation mechanism. The linear VAR(3) model also implies a much lower volatility for the consumption and investment series than the other two models. φ. 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. whereas in the other two models. these results suggest that such features as ﬁnancial frictions or irreversibility in investment are required to match the nonlinearities of consumption and investment. The ﬁrst of these is a linear VAR(3) so that the statistical procedure is similar to the RBC approach of matching the impulse responses from a standard VAR with those generated from the underlying economic model.
as Valderrama [207] states. 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. as in Bertola and Caballero [42]. However. Both Veracierto [208] and Thomas [205] assume that labor is perfectly mobile across plants. for example. Thus. 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. These arguments suggest that nonlinearities are another
. Fallacy and Fantasy
multisector growth models considered in the literature (see. namely. This latter feature may compensate for the rigidity faced by plants given that capital and labor are substitutable according to the Cobb–Douglas speciﬁcation of technology. these microeconomic nonlinearities must matter at the aggregate level.126
Business Cycles: Fact. and Gilchrist [41]). Moreover. the aggregate impact of irreversible investment disappears in a general equilibrium framework.” Yet. alternatively. As Caballero [53] emphasizes: “An important point to note is that since only aggregate data were used. is due to their assumption of a very large variance for these idiosyncratic shocks. Gertler. the reason why aggregate investment displays lower variability in the presence of irreversibility when ﬁrms are subject to idiosyncratic shocks. Similarly. 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. irreversibility may play a more important role when ﬁnancial constraints at the plant level and the household level are taken into account. for example. this is a puzzling ﬁnding. Coleman [73] or Ramey and Shapiro [174]) is due to their assumption of unrealistically large sectoral shocks. our analysis of the simple RBC model with ﬂexible prices and wages suggests that it cannot successfully match many features of the data. At the same time. More generally. and may be responsible for softening the impact of the inﬂexibility faced by plants in terms of their capital adjustments. monetary shocks. 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. one may consider the impact of another type of aggregate shock. Veracierto [208] argues that when the size of sectoral shocks is consistent with that observed in the data. for otherwise they would not be identiﬁed.
” Returning to the inception of the estimation versus calibration debate. 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.4. Hoover. for example. [103] in a VAR framework. one could argue that the GMM approach in Christiano and Eichenbaum [65] involves choosing a subset of moments of an equally restrictive nonlinear model.3. but in what sense does this make the model a better interpretation of reality? King also discusses the shortcomings of this approach.
7. 17). Yet. Yet. it is unlikely that RBC models would be subjected to the types of diagnostic tests proposed by Watson [209]. p. 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. and a potential direction for improving the model’s ﬁt. Many have argued that this was to be expected (see Hartley. we note that Altug’s [4] results did not lead to support for the model. One can also examine the claim that a less restrictive approach to estimation and model evaluation following the approach in Christiano and Eichenbaum [65].Matching the Model to the Data
127
area where the assumptions of the simple RBC model fail to hold. If the simple model is counterfactual.9 In the absence of the initial frequencydomain estimation based on the factor representation. and Salyer [116]. King [128] advocates such an approach as an alternative to calibration that does not face the problems of “testing highly restrictive linear models”. King claims to be on the quantitative theory side of the debate.
. 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. may be preferable. The Debate Reconsidered
King [128] discusses the arguments on different sides of the debate under the heading “Quantitative Theory and Econometrics”. an RBC model with some or all of its parameters estimated) to an unrestricted time series model. highlighting problems in ﬁnding appropriate
9 See Altug [4]. The factor representation underlying Altug’s analysis has also been resuscitated by Giannone et al. For some time. and argues against the estimation and comparison of a “heavily restricted linear time series model (for example.
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. As Geweke [102] notes. and the New Keynesian challenge has shown that the model fails in generating the observed negative response of hours to productivity improvements. Conversely.128
Business Cycles: Fact. These arguments show that there is no clearcut dichotomy between these approaches. 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. far from this occurring. Fallacy and Fantasy
instruments and in the appropriate selection of moment conditions for model evaluation. The initial criticisms of
. These developments show that the contribution of the more formal econometric techniques need not be dismissed as cursorily as is sometimes the case. one could argue that the approach in Christiano and Eichenbaum [65] is too restrictive. 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. This interpretation precludes the notion that quantitative theorizing can take the place of formal econometric methods. 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. 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. 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. 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. 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. The simple propagation mechanisms inherent in the model have been deemed incapable of reproducing business cycle dynamics of key variables such as output. In fact. Another criticism of the RBC approach is that it fails to capture nonlinearities in key macroeconomic time series.
To capture the ﬁrst feature. For example. According to them:
[This] model is stark in its assumptions. The recent class of models developed for this purpose has vastly more features than any simple RBC model. the model recently proposed by Kapetanios. Finally. consider the model proposed by Christiano. Above all. the original RBC approach has even instigated a new generation of Keynesian models that offer features such as credit accelerators. There are no market frictions and distortions. 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. As we discuss in the next chapter. and Evans [67] for the analysis of the effects of monetary policy shocks on real and nominal variables. for example. One can ask whether this class of models overcomes some of the criticisms leveled against the original RBC approach. Eichenbaum. Consider. The aim of this model is to reproduce the inertial behavior of inﬂation and persistence in real quantities. There is also an interestratesetting rule that deﬁnes
.4. There are no market frictions and no locational speciﬁcity. and animal spirits. there is no banking sector and no speciﬁc role for money and credit in the monetary transmission mechanism. the model contains assumptions that are almost guaranteed to be violated by the data.
7. markets are assumed to clear at all times. a topic to which we turn next.Matching the Model to the Data
129
the RBC approach have led to a wide set of extensions of the original RBC approach. adjustment costs in investment. and variable capital utilization. and a variety of techniques have been developed for the purpose of matching the model to the data. no ﬁxed costs or discontinuities.
Yet this model has a core set of 26 equations! As another example. In sum. the model incorporates wage and price contracts following the approach in Calvo [54]. The model assumes a representative household and a symmetric equilibrium for ﬁrms. The model also incorporates a variety of “real” frictions such as habit persistence in consumption. and Scott [123] for policy analysis of a small open economy. as all agents have complete knowledge of the economy and complete understanding of shocks when they hit. especially in the short run. Another extension of the original approach has been the development of applications that are useful for policy analysis. sunspot equilibria. Pagan. ﬁrms are assumed to borrow working capital to ﬁnance their wage bill.
For example. Furthermore. and Demers [10. but homogeneous in their consumption and asset holdings. provide some evidence on the role of price and wage rigidities. [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. the assumption of a constant adjustment cost parameter would fail to hold. 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. 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 model assumes a continuum of households which are heterogeneous in their wage and the hours that they work. Christiano et al. see also Demers. but in a highly restrictive environment. For example. the impulse responses to a given monetary shock. Fallacy and Fantasy
monetary policy. 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 For a review and discussion. see Altug. in many ways. there are shortcomings in their approach. Minford. [67] assume that monetary policy shocks are drawn from a given and known distribution. one could also question the implications of the model should there be changes in the distribution of the exogenous processes. and Wickens [160]. estimated models of adjustment costs have been shown to imply implausible degrees of costs of adjustment. with the latter result derived from the existence of statecontingent securities for consumption. [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. This raises
10 For a recent example of the method of indirect inference applied to a model of the EU economy. an exploratory analysis of the role of nominal and real rigidities in propagating shocks.11 Christiano et al. Christiano et al. and Altug [79]. 11].10 The authors’ analysis is. By contrast. implying that the propagation mechanisms would vary with changes in the distribution of exogenous variables facing agents. In such cases. Yet. Demers. Demers. Christiano et al. For applications. see Meenagh.130
Business Cycles: Fact.
where St is a vector of state variables that describe the evolution of the model over time. Similar to Canova [56]. for example. they augment the approximate linear laws of motion with additional shocks. vt is a vector of innovations. First. examining the posterior distribution calculated as the prior distribution times the likelihood function is often more straightforward computationally than maximizing the likelihood function. and θ is a vector of parameters characterizing preferences. they employ time domain methods by making use of the Kalman ﬁlter with a linear law of motion for the state variables and a linear measurement equation for a key set of observables. θ). Second. and Wright [159] consider an equilibrium business cycle model with household production and distortionary taxation. Second. and
. Third. which is typically a highdimensional object. and information. As in the analysis of Altug [5]. suppose that the state space representation for the model’s solution consists of the following: • a transition equation St = g (St−1 . the production technology. 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. Rogerson. the likelihood function for the observations must be formed. a variety of other papers have implemented maximum likelihood estimation of dynamic equilibrium models. structural or invariant to interventions. in particular. amongst others.Matching the Model to the Data
131
the issue of whether DSGE models are. McGrattan. More recent applications include Canova [60] and Ireland [121]. Following the initial contribution of Altug [5]. this approach ﬁrst provides a link with the calibration approach by allowing the researcher to use information from microeconomic studies or macroeconometric estimates. the posterior distribution for the parameters. the use of Bayesian methods and. in fact. the method presupposes that a solution can be found for the underlying economic model of interest. must be examined. To illustrate this approach. vt . There are a variety of approaches for solving nonlinear dynamic stochastic models. The Bayesian estimation of DSGE models involves several steps. Judd [122] provides a textbook treatment of many currently used solution methods. which is obtained from the prior distribution and the likelihood function. Schorfhiede [186] and An and Schorfhiede [15]).
23)
The remaining task is to compute the sequence {p(St y t−1 . we can also compute the probabilities p(St St−1 . Given the probability density functions for the shocks fv ( · ) and fw ( · ). θ)dSt
12This is known as the Chapman–Kolmogorov equation. which shows t=1 the conditional distribution of the states given the observations. Hence. θ) = p(St+1 St . . θ). θ)dSt . The excellent survey by FernandezVillaverde [90] describes in detail how to implement Bayesian estimation of DSGE models. θ)
T
=
p(S1 S0 . θ). θ)p(St y t−1 .
Given a prior distribution π(θ) for the parameters θ. among others).132
Business Cycles: Fact. the state space representation allows us to compute the likelihood of the observations y T = (y1 . Following his approach. θ)p(St y t . the posterior distribution can be expressed as π(θy T ) = p(y T θ)π(θ) . . θ). this sequence is computed by making use of the relation12 p(St+1 y t . Fallacy and Fantasy
• a measurement equation Yt = h(St . θ) = p(y1 θ)
t=2
T
p(yt y t−1 . . θ)}T . vt . 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. Likewise. we can compute p(Yt St−1 . yT ) at the parameter values θ as
T
p(y . wt . wt .
. Notice that Yt = h(g (St−1 . θ). where Yt are the observables and wt are the shocks to the observables (which may take the form of measurement errors. θ) and p(Yt St . θ). θ)dS0
t=2
p(Yt St . . p(y T θ)π(θ)d θ (4.
θ)dSt
Finding the posterior distribution of the parameters displayed in (4. they consider a full set of structural shocks. 195] provide two important examples of this approach. consumption. a temporary contractionary monetary shock has a temporary negative effect on output and inﬂation. Smets and Wouters [194] argue that their approach offers several important advances. a shock to the household’s discount factor. the GDP deﬂator. the real wage. then the Kalman ﬁlter is used to obtain the likelihood function. if the transition and measurement equations are linear and if the shocks are normally distributed. They extend the approach in Christiano et al. p(yt St . based on Bayesian model evaluation criteria. For example. the estimated DSGE model performs as well as standard and Bayesian VARs. Smets and Wouters [194. on the whole. θ) . investment. and the nominal shortterm interest rate — by minimizing the posterior distribution model’s parameters based on a linearized state space representation for the model. [67]. Their estimates imply more price stickiness than the estimates of Christiano et al. they ﬁnd that. θ)p(St y t−1 . [67]. similar to those for the US economy. θ)p(St y t−1 . These include a productivity shock.
.23) involves a few remaining steps. a method known as the particle ﬁlter13 is used to evaluate the likelihood function. Otherwise. Third. Fourth. However. [67] to estimate the parameters of a DSGE model with real and nominal frictions for the euro area using Bayesian methods. employment. a labor supply shock. They consider the behavior of seven key macroeconomic time series in the euro area — real GDP. Second. In contrast to the approach in Christiano et al. First. 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.Matching the Model to the Data
133
and an application of Bayes’ theorem as p(St y t . however. First. an investmentspeciﬁc shock. their ﬁndings attribute the largest role
13 See Schorfhiede [186] or FernandezVillaverde [90]. and a government consumption shock plus three “costpush” shocks. they ﬁnd that the estimates of the structural parameters of the model are plausible and. exploring it for different values of θ typically involves the use of sampling through Markov chain Monte Carlo (MCMC) methods. even if we can evaluate the posterior distribution. θ) = p(yt St .
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
and contrasts it with the standard real business cycle (RBC) approach. Many recent works assume that there is a search technology such that ﬁrms and workers are randomly matched.Chapter 8
Future Areas for Research
The literature on business cycles is vast. For one. he considers the role of a search externality and a lemons problem in the market for search inputs. There are many other topics of interest that we could have discussed regarding business cycles. multisector models may also have shortcomings because they typically imply that overall expansions in the economy are associated with contractions in one sector. I have not provided a discussion of models with selffulﬁlling expectations and multiple equilibria. Speciﬁcally. However. but always thoughtprovoking. In a recent analysis. Farmer [88] generates a model of selffulﬁlling equilibria which arises from market failure associated with the labor market. Benhabib and Farmer [38].
137
. I have offered a perspective based on my own perceptions and mindsets. I have omitted a discussion of multisector models in generating business cycle behavior. For example. Many have argued that such models are more in the spirit of true Keynesianism because they stress the role of “animal spirits”. In this book. it may be that shocks are concentrated in a particular sector and are propagated to the rest of the economy from that sector. the wage is then set by a
1 See. one needs to consider multisector models since the onesector growth model does not allow for a consideration of these issues.1 In such cases. at times contentious. even the experience of reviewing some of this literature shows that it is lively. Nevertheless. Farmer [87] provides an eloquent enunciation of this approach. for example. More generally.
This feature of the model allows for the role of “conﬁdence”. By contrast. The policy implications of such a model differ from those of the standard textbook Keynesian model. and new controversies regarding both. Farmer [88] drops the assumption that wages are determined as a result of a Nash bargaining process.2 In contrast to this literature. this discussion will be accompanied by the development of new models. he assumes that ﬁrms offer the same wage in advance. we may conclude that the debate surrounding their development will not cease to arouse interest or to generate future avenues for research. all ﬁrms earn zero proﬁts but not all equilibria have the same welfare properties. irrespective of what the new models or techniques will look like. For a recent application in the context of a standard RBC model. Fallacy and Fantasy
Nash bargaining process. and Rouabah [171]. which is transmitted to real variables through investors’ beliefs about the stock market.
.138
Business Cycles: Fact. However. This book has also omitted a discussion of models with credit and collateral constraints. the model with a search externality emphasizes the role of conﬁdence in restoring fullemployment output. In these equilibria. 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 For a recent analysis of models with ﬁnancial frictions in a dynamic general equilibrium setting. 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 Pierrard. This leads to a model with a continuum of steadystate equilibria. Instead.
2This approach owes its origins to the analysis in Mortensen and Pissarides [165]. which seeks to alleviate demandinduced shortfalls in output through large ﬁscal stimuli. see Cooley and Quadrini [76]. This leads to a demandconstrained equilibrium. new techniques.3 The above considerations show that business cycles will continue to be a topic of discussion among economists for many years to come. de Walque. Undoubtedly.
45–85. and Asset Pricing with Risk of Default. S.” Unpublished manuscript. and R. “Efﬁciency.” Econometrica. S. pp. “Political Risk and Irreversible Investment.” European Economic Review. 68. 321–346. Demers (2007). Ashley. and C. (1989). Demers. “Emerging Market Business Cycles: The Trend Is the Cycle. Altug.” Canadian Journal of Economics. 3. pp. pp. pp. and R. 10. Altug. Altug. Ambler. “Money. M.” International Economic Review. pp. 14. Altug. and D.” Journal of Monetary Economics. and Real Business Cycles in a Small Open Economy. 27. 139
. 13. “The Investment Tax Credit and Irreversible Investment. S. 982–993. and U. “The Effect of Past Experience on Female Wages and Labor Supply. 889–920. 273–300. 5. R. 11. and P. “Gestation Lags and the Business Cycle. Miller (1990). Altug. S.. and U. 2. S. 430–465. F. Altug. 4. 8. 543–570. Alvarez. pp. S. Output. 51.” Journal of Macroeconomics. S. F. S. 6. 115. and M. pp. 47. B. and R. F. and C. 69–102. 775–797. and M. 506–533.. Ahmed. pp. 30. (1985). 257–276. 9. forthcoming. 3. Equilibrium. pp.” Econometrica.Bibliography
1. Patterson (1999).” Review of Economic Studies. Asset Pricing for Dynamic Economies.. Demers.. Labadie (2008). Jermann (2000).” Journal of Political Economy. “More International Evidence on the Historical Properties of Business Cycles. Cambridge: Cambridge University Press. Aguiar. “TimetoBuild and Aggregate Fluctuations: Some New Evidence. Cardia. 53. 7. Cardia. pp. “International Transmission of the Business Cycle in a Multisector Model. “International Business Cycles: What Are the Facts?” Journal of Monetary Economics. and G. “Household Choices in Equilibrium. S. 12.” CESifo Economic Studies. 46. Altug. 58. S. Altug. Gopinath (2004). “Are Technology Shocks Nonlinear?” Macroeconomic Dynamics. E.. Murthy (1994). 65. Demers (2009). E. Ambler. S. Zimmermann (2004). pp. Miller (1998). Zimmermann (2002). Woitek (2001). A’Hearna.
S. and W. 120–132. P. 100. Handbooks in Economics. Arellano. J. and T.. Anderson. R. 5915. 13. 745–775. An. Vol. 113–172. and T. 1418–1448.. “Linear versus Nonlinear Macroeconomics: A Statistical Test. 115–134.” Journal of Economic Perspectives. Z. 70.).
. pp. pp. 18. Kontolemis. 19. “International Trade and Business Cycles.” Innocenzo Gasparini Institute for Economic Research (IGIER) Working Paper No. Amman. 96. 864–888. 249–279. Artis. 23. pp. pp.” Journal of Monetary Economics. 13. Ashley. Kydland (1995). pp.” In T. “Procyclical Productivity: Increasing Returns or Cyclical Utilization?” Quarterly Journal of Economics. C. Basu. R. L. Kydland (1992). “Business Cycles for G7 and European Countries. M. 171–252.. 685–704.” International Economic Review. pp. Sargent (1996). D. Barro. 237. 1801–1864. Baxter. P. Basu. “Credit Frictions and ‘Sudden Stops’ in Small Open Economies: An Equilibrium Business Cycle Framework for Emerging Markets Crises. Dynamic Macroeconomic Analysis: Theory and Policy in General Equilibrium. M. 16. Handbook of Computational Economics Vol. and F.). D. pp. Kimball (1997). pp.140
Business Cycles: Fact. “Dating the Euro Area Business Cycle. 26. Backus. 30. Kimball (2006). 331–356. and J. pp. and M. Cooley (ed. Artis.” Journal of Economic Perspectives. and D. 24. “International Evidence on the Historical Perspective of Business Cycles. “Mechanics of Forming and Estimating Dynamic Linear Economies. and D. 82... (1995). S. Amsterdam: NorthHolland. 297–316. S. Backus.” Journal of International Economics. 25. Taylor (1999). M. Barsky. Proietti (2003). pp. 31. E.” American Economic Review. Marcellino. (1987). D. Princeton: Princeton University Press. and L. Kehoe. pp. “International Real Business Cycles. “Cyclical Productivity with Unobserved Input Variation.. S. 221–247. 28. pp. Zhang (1999). Chadha. Altug. Cambridge: Cambridge University Press. 20. 1–16. “Government Spending. Rogoff (eds. pp. and F. M. Handbook of International Economics. 51. (1996). Basu. “Further Evidence on the International Business Cycle and the ERM: Is There a European Business Cycle?” Oxford Economic Papers. S. McGrattan. Frontiers of Business Cycle Research. and A. 1. J. pp. Kendrick. 35.” Journal of Business. 20. Fernald. 27.” Journal of Political Economy. “Are Technology Improvements Contractionary?” American Economic Review. 2.” NBER Working Paper No. and M. Artis. Fallacy and Fantasy
15. pp. 45–68. Kilian (2004). “International Business Cycles and the ERM: Is There a European Business Cycle?” International Journal of Finance and Economics. D. Backus.” Econometric Reviews. 22. M. and F.” In S. D. Backus. and P. Prices and Budget Deﬁcits in the United Kingdom. Mendoza (2003). pp.). Kehoe. Interest Rates. 335–405. 30. and E. R. 719–751. 1701–1918. and G. 18. “Business Cycles in International Historical Perspective. and W. Patterson (1989). pp. Nolan (eds. Artis. Osborn (1997).” In G.” In H. Hansen. Basu. 17. 21. and C. Grossman and K. Zhang (1997). “Bayesian Analysis of DSGE Models. “Consumption and Real Exchange Rates in Dynamic Economies with Nontraded Goods. Amsterdam: Elsevier Science/NorthHolland. 29. “Oil and the Macroeconomy Since the 1970s. E. 26. 111. Schorfhiede (2007). “International Business Cycles: Theory and Evidence. M. Kehoe (1992). 32. Smith (1993). 33. Rust (eds.).
pp. M. W10130.” Journal of Monetary Economics. Rebelo (1993). Net Worth and Business Fluctuations. King (1999). “The Financial Accelerator and the Flight to Quality. Quah (1989). 1–15. and W. 87(1). J. G. 2003762. and D. M. Taylor and M. pp. (1999). Amsterdam: Elsevier Science. Farmer (1996).” American Economic Review. “Have National Business Cycles Become More Synchronized?” NBER Working Paper No. “The Dynamic Effects of Demand versus Supply Disturbances.” Federal Reserve Board (FRB) International Finance Discussion Paper No. Heckman (1999). and R. Measuring Business Cycles. Burns. pp. Hansen. and M. Handbook of Macroeconomics. “The International Diversiﬁcation Puzzle Is Worse Than You Think. “Aggregate Investment. C. pp. 170–180. “Is the Business Cycle Characterized by Deterministic Chaos?” Journal of Monetary Economics. Bernanke. Mitchell (1946). and R. and B. 54. 1154–1174. Brock. Gertler (1989). A. 33. 421–443. 36. Baxter. 53. 78. 1A. pp. and R. 821–854. “Labor Hoarding and the Business Cycle. (1994). Bordo. North Holland: Elsevier Science.” Journal of Monetary Economics. M. Woodford (eds. 36. pp. pp. pp. 101. 37. 43. 79. 35. pp. M.
. pp. pp. 335–347.” Journal of International Economics. Baxter. “Business Cycles and the Asset Structure of Foreign Trade. “Staggered Prices in a UtilityMaximizing Framework. M. 22. C. Handbook of Macroeconomics. 81. and S. “Tax Disturbances and Real Activity in the Postwar United States. 245–274. 49. and J. Bry. Woodford (eds. Braun.” In J. pp. 51. M. “Measuring Business Cycles: Approximate Bandpass Filters for Economic Time Series. B. “Homework in Macroeconomics: Household Production and Aggregate Fluctuations. Caballero. 12. “Micro Data and General Equilibrium Models. Eichenbaum. Gertler. 44. 48. Vol. B. Browning. Calvo. OpenEconomy Models and Their Implications for Monetary Policy.). Farr (2005). Cyclical Analysis of Time Series: Selected Procedures and Computer Programs. 71–90. 441–462.” Review of Economics and Statistics.” Journal of Political Economy. W. and C. and C. Rogerson. L.. and M. Ch. 383–398. 99. 223–246. “Irreversibility and Aggregate Investment. “Agency Costs. 1166–1187. 40. Burnside. Heibling (2003). A. Ch.” Journal of Monetary Economics. “Factor Hoarding and the Propagation of Business Cycle Shocks. Wright (1991). 12. 41. Sayers (1988). R. “Indeterminacy and SectorSpeciﬁc Externalities. Eichenbaum (1996). 46. 8. Caballero (1994). 79.” In J. D. J. Boschan (1971). 575–593. 14–31. and S.” Review of Economic Studies..” The Review of Economics and Statistics. Baxter. R. 39. 38. G. Blanchard. and M. 42. New York: Columbia University Press for the NBER. Benhabib. and D. Crucini (1995). pp. Bertola. “Variable Factor Utilization and International Business Cycles. “New Keynesian. 1B.. Gilchrist (1996). Burnside. Bowman. 65..Bibliography
141
34. pp. M. and U.). 61. G. and R. Vol. 86. 47.” American Economic Review. O. Baxter.” American Economic Review.” American Economic Review. 654–673. and T. New York: NBER. Taylor and M. 52. Benhabib. pp. 50.” International Economic Review. M.” Journal of Political Economy. Jermann (1997). Bernanke. Doyle (2003). 45. 37. (1983).
247–280. Christiano. 56. 41. Canova. “Inspecting the Mechanism: An Analytical Approach to the Stochastic Growth Model. pp.” Macroeconomic Dynamics. Sala (2006).142
Business Cycles: Fact. Nason (1995). 75. 697–721. Obstfeld (1991).) (1995). “Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy.” Journal of Applied Econometrics. 69.” ECB Working Paper No. “Sticky Price and Limited Participation Models of Money: A Comparison. 113. 123–144. Eichenbaum. 19.” International Economic Review. 70. Cooley. pp.” European Economic Review. L. 430–450.. 21. pp. pp. F. W.” American Economic Review. Coleman. 72–89. Nason (1995). “The Behavior of Interest Rates in a General Equilibrium Multisector Model with Irreversible Investment. 72. 1. Canova. M. and M. Christiano. “Current Real Business Cycle Theories and Aggregate Labor Market Fluctuations. and C.” Journal of Monetary Economics. Frontiers of Business Cycle Research. Princeton: Princeton University Press.” Journal of Economic Dynamics and Control. 67. M. (2006). and E. Fallacy and Fantasy
55. McGrattan (2004). 62. “Output Dynamics in Real Business Cycle Models. 59. (1988). “Effects of the Hodrick–Prescott Filter on Trend and Difference Stationary Time Series: Implications for Business Cycle Research. Christiano. “Back to Square One: Identiﬁcation Issues in DSGE Models. T. pp. (2008). and M. and C. F. Kehoe. pp.. L. Canova. T. 74. 237–259.” American Economic Review. Canova. “Statistical Inference in Calibrated Models. Princeton: Princeton University Press.). H. and R. Canova. Evans (1997). 1–45. Eichenbaum. “Are Structural VARs Useful Guides for Developing Business Cycle Theories?” Federal Reserve Bank of Minneapolis Research Department Working Paper No. “What Explains the Great Moderation in the US? A Structural Analysis. F.
. Christiano. “Why Does Inventory Investment Fluctuate So Much?” Journal of Monetary Economics. 1–38. Christiano. Cogley. and C. and J.” National Institute Economic Review. 631. pp. 82. 9. Nolan (2002). “How Much Structure in Empirical Models?” CEPR Discussion Paper No. 1201–1249. Evans (2005). 64.” Journal of Monetary Economics. Eichenbaum. 206–227. and L.” In T. Vigfusson (2003).” Journal of Political Economy. 33. 29. 9819. pp. Cole. pp. 71. 65. Princeton: Princeton University Press. T. 73. M. Canova. (1988). F. 57. “What Happens After a Technology Shock?” NBER Working Paper No. 6791. Cooley (ed.. 477–501. P. (2009). Prescott (1995). Eichenbaum (1992). 36. J. H. 68. pp. (1995). “Commodity Trade and International Risk Sharing. V. Frontiers of Business Cycle Research. 85. Cole. F. “Economic Growth and Business Cycles. “Sensitivity Analysis and Model Evaluation in Simulated Dynamic General Equilibrium Economies. 60. L. 7. Cooley. Methods for Applied Macroeconomic Analysis. 28. 492–511. Chadha. L. pp. (1994). 583. (ed. J. 463–506. Cogley. pp.” International Economic Review. Chari. “Financial Structure and International Trade.” Journal of the European Economic Association. pp. (1994). (1997). F. and J. T. L. 66. pp. “A Long View of the UK Business Cycle.. Campbell. 61. 253–278. and E. 63. 3–24. 182. pp. 58.
Singleton (1988).” Review of Economic Studies. 51–78. “Crises Now and Then: What Lessons from the Last Era of Financial Globalization?” NBER Working Paper No. 400–416. 90. pp. 453–473. Nolan (eds. pp. Devereux. and K. M. Denison. pp. M. R. Equilibrium Business Cycle Model. “Propagation Problems and Impulse Problems in Dynamic Economies. Mestre (2001). (1985). (2009).” Economic Journal. “An AreaWide Model (AWM) for the Euro Area. Altug (2003). “Technology. Finn. T. I. 84.” In S. Henry. MA: MIT Press. 90. F. M. Demers. M. Horioka (1980). Fagan. pp.” Quarterly Journal of Economics.” European Economic Review. J. 80. Smith (1992). pp. Feldstein. and the Business Cycle: Do Technology Shocks Explain Aggregate Fluctuations?” American Economic Review. Employment. and S. Dynamic Macroeconomic Analysis: Theory and Policy in General Equilibrium. 42. Crashes. (2000). M. pp. 95. and G. Reichlin (1998). 88. Schwartz (1963). 14846. 1004. (1933). 81. 44. Frisch. E. Gregory.” In Economic Essays in Honor of Gustav Cassel. 607–626. 89. Eichenbaum.” Cowles Foundation Discussion Paper No. Forni. A Monetary History of the United States. Neves. Chadha.. 32. and C. J. and Banking. pp. 3–16. FernandezVillaverde. and S. (1991). pp. Eichenbaum. W8716. 89. FernandezVillaverde.” Journal of International Money and Finance. “Irreversibility and the Arrival of Information. Credit. The Macroeconomics of SelfFulﬁlling Prophecies. Quadrini (1999). Cambridge. “Conﬁdence. 93. 87. “Domestic Saving and International Capital Flows. 1089–1113. “Let’s Get Real: A Factor Analytical Approach to Disaggregated Business Cycle Dynamics. and Animal Spirits. R. 85. J. Real Business Cycle Theories. “Investment Dynamics. (2009). Demers. R. B. Eichengreen.” Journal of Money. and A. J. 91. “Business Cycles in Small Open Economies. Trends in American Economic Growth. Correia. Cambridge: Cambridge University Press. M. J. 86.. M. Altug. (1999). (1991). 15. 65. 314–329..” NBER Working Paper No. Farmer. Bordo (2002). 249–271. “Realistic CrossCountry Consumption Correlations in a TwoCountry. Farmer. 39. (1999).” NBER Working Paper No. RubioRamirez (2007). Princeton: Princeton University Press. 2nd ed. “How Structural Are Structural Parameters?” NBER Working Paper No. 103. 14677. 34–154. 333–350. 94.” Journal of Monetary Economics. 1929–1982.Bibliography
143
76. and L.). and V.. 83. and C. 92. 11. and M.. L. London: George Allen & Unwin. 78. and R.” Review of Economic Studies. Washington. Hansen. 58. M. 165–193. “Real Business Cycles: Wisdom or Whimsy?” Journal of Economic Dynamics and Control. pp.” ECB Working Paper No.
. Rebelo (1995). Cooley. 1867– 1960. Gali. 96. DC: Brookings. M. 82. “The Cowles Commission Approach. Friedman. pp. Demers. “A Time Series Analysis of Representative Agent Models of Consumption and Leisure Choice Under Uncertainty. A. “Perfect Competition and the Effects of Energy Price Increases on Economic Activity. and New Keynesian Economics. and R. 79. G. pp. Fair (1992). 13166. J. “The Econometrics of DSGE Models. “A Neoclassical Model of the Phillips Curve Relation. 77.
). 110. Uribe (2006). Cambridge. “The Role of InvestmentSpeciﬁc Technological Change in the Business Cycle. R.. Gallant. 921–947. Pancrazi. 402–417. and A. Z. “Real Business Cycles in Emerging Countries?” NBER Working Paper No. Hansen. 225–318. (1990). Greenwood. J. MA: MIT Press. 677–702. 9. Monetary Policy.” International Economic Review. K. Gregory. “A Comparison of Business Cycle Dating Methods. Hall. 98. Princeton: Princeton University Press. and P.” Journal of Econometrics. R. G. and M. Industry. London: Routledge. 12629. (1999). The Measurement of Durable Goods Prices. Smith (1989). Salyer (eds. “Computational Experiments and Reality. J. “Technology Shocks and Aggregate Fluctuations: How Well Does the RBC Model Fit Postwar US Data?” In M. J. 106. Gali. (2008). and the Real Business Cycle. and P. and L. Gali. pp. and K. pp.” University of North Carolina Working Paper. 50. pp. L. 113. 342–362. Greenwood. Hansen. 44. J. “Which Moments to Match?” Econometric Theory. 108. Z. J. Capacity Utilization. Hoover. 99. pp. “SNP: A Program for Nonparametric Time Series Analysis. 109. 27. 111. 104. Harding. Krusell (2000). “Investment. Growth/Productivity/Unemployment. and G. A. D. 87..) (1998). Perri (2002).” SCE Computing in Economics and Finance Working Paper No. 107. Tauchen (1996).” In P. 601–627. Cambridge. Huffman (1988). Rabanal (2005). 49. pp. 57–89. (1990). pp.. and G. Hercowitz.” American Economic Review. 2. pp. 100. pp. pp. pp. J.” Journal of Economic Dynamics and Control.. Inﬂation. pp.” Econometric Reviews.
. Common Factors and the Empirical Validation of Equilibrium Business Cycle Models. Greenwood. A. pp. J. 1029–1054. J..” American Economic Review.). 1681–1690.7. “Invariance Properties of Solow’s Productivity Residual. and A. User’s Guide. (1982). “Calibration as Estimation. Hercowitz.144
Business Cycles: Fact. pp. 12. NBER Macroeconomics Annual 2004. “VARs. 116. Real Business Cycles: A Reader. and F. and the Business Cycle.. Hartley. Heathcote. “The Relation Between Price and Marginal Cost in U. 103. “Indivisible Labor and the Business Cycle. R. “Formulating and Estimating Dynamic Linear Rational Expectations Models. Fallacy and Fantasy
97. “Financial Autarky and International Business Cycles. 71–112. 78. Gordon. 105. 112. Geweke. Giannone. 101.” European Economic Review. 657–681. Reichlin. Version 8. Krusell (1997). 115. Head. and P. MA: MIT Press. Pagan (2003). Rogoff (eds. L. Hall. and T.” Journal of Economic Dynamics and Control. Hansen. (1988). D.. L. (1985). Raynauld (1997). 7–46. “Measuring World Business Cycles. and J. 117. Z. A. 257–279. Diamond (ed. R. 401. 38. R. 16. “LongRun Implications of InvestmentSpeciﬁc Technological Change. Tauchen (1998). J.” Journal of Monetary Economics. 114. Gertler and K. pp. 102. 309–327. GarciaCicco. Gregory. Sargent (1980). pp.” Econometrica. Sala (2006).” Journal of Monetary Economics. and G. 91–115. Gallant. R.” Journal of Political Economy. Chicago: University of Chicago Press. 132.S. 96. “Large Sample Properties of Generalized Method of Moments Estimators. Hercowitz.
A. and Money. pp. N. 126.” Oxford Review of Economic Policy.Bibliography
145
118.” American Economic Review. Growth. DC: Brookings Institution. 819–840. Whiteman (2003). Rebelo (1988). and K. Moore (1997). “International Business Cycles: World. 363–380. MA: MIT Press. “Consumption. 138. 923–936. C.” Journal of Econometrics. 131. pp. Stock. Prescott (1997). pp. “Liquidity Constrained Markets versus Debt Constrained Markets. “Stochastic Trends and Economic Fluctuations. “Postwar U. 907–928. R. Kondratiev. 128. Kehoe. “How Does Globalization Affect the Synchronization of Business Cycles?” IZA Discussion Paper No. 139. 136.” American Economic Review. Helliwell.S. A. G.” Journal of Political Economy. 575–598.” Econometrica. and M. 86. P. Credit. 309–341. pp. and A. 120. Levine (2001). 211–248. Plosser. J. Kocherlakota. Watson (1991). Rebelo (1993). 63. (1998). 119. Ireland.” Econometrica.. and E.” Archiv für Sozialwissenschaft und Sozialpolitik. 29. and Business Cycles: II. King. and CountrySpeciﬁc Factors. King. and C. (1995). King. Kollman. 14. C. “Money. Keynes. Hodrick. 134. Growth. and S. Kiyotaki.. and the Structure of International Asset Markets.” Journal of Monetary Economics. 121. K. R. 132. 122. Köse. 136. 123. 105. Scott (2007). The Basic Neoclassical Model. and S. and D. 207–231. Kapetanios.” Review of Economics and Statistics. Plosser. 702. 21. 135. Plosser. pp. Cambridge. Credit. R. Region. 137.” Journal of Economic Dynamics and Control. and M. 125. 130.
. 133. 81. 93. 93–109. 21. 74. and J. King.” Review of Economic Studies. 865–888. 70. “Technology Shocks in the New Keynesian Model. Otrok. Plosser (1984). pp. 595–609. 53–105. Perri (2002). Interest. Köse. and Banking.. J. and D. A. pp. T. C. C. New Directions. 69(3). pp.. “International and Intranational Business Cycles. “Die Langen Wellen der Konjunktur [The Long Waves of the Business Cycle]. pp. Kehoe... pp. pp. Hess. How Much Do National Borders Matter? Washington. R. (1995). Prasad. 565–594. “Production. 1216–1239. 60. and F. 13. Shin (1997). R. P. “International Business Cycles with Endogenous Incomplete Markets. (1926). “Production. N. “Quantitative Theory and Econometrics. Real Exchange Rates. pp. Business Cycles.” Journal of International Money and Finance. (1936).” Review of Economic Studies. pp. and Business Cycles: I. “Debt Constrained Asset Markets. pp. pp.” American Economic Review. J. pp. and Prices in a Real Business Cycle Model. “Credit Cycles.” Journal of Money. M. London: Macmillan. and C. Rebelo (1988). Pagan. 195–232. N. (1996). 124. King.T. 129. 127. 17. “Making a Match: Combining Theory and Evidence in PolicyOriented Macroeconomic Modeling.” Journal of Monetary Economics. Judd. pp. King. 191–211. G. Levine (1993). (1998). 81. 573–606. 56. “Low Frequency Filtering and Real Business Cycles. R. R. 1–16. Kehoe. R. and S.” Economic Quarterly. (2004). E. The General Theory of Employment. Numerical Methods in Economics. Terrones (2003). “Implications of Efﬁcient Risk Sharing without Commitment. pp.
146
Business Cycles: Fact. and E. pp. 156.” In K. Lucas. Credit. R. R. E. R. M. 158. 103–123. 101–127.). 5. E2008/11. pp. pp. 83. 160. pp. pp. Lumsdaine. pp. and R. F. McCallum. Rogerson. Recursive Macroeconomic Theory. 149. 615–623. pp.” Scandinavian Journal of Econometrics. “Expectations and the Neutrality of Money. R. pp. Brunner and A. R.” Journal of Monetary Economics. “Testing a DSGE Model of the EU Using Indirect Inference. Sargent (2000). 148. pp. Ljungqvist. (1994). 18. Kydland.” Journal of Political Economy. (1996). 145. F. and E. Marcet. 126–156. 143. 157. Prescott (1982). pp. “The Econometrics of the General Equilibrium Approach to Business Cycles. (1972). R. 573–601. “An Equilibrium Model of the Business Cycle with Household Production and Fiscal Policy. McCallum. Lewis. 151. Kydland.” American Economic Review. Stabilization of the Domestic and International Economy.. Employment and Inﬂation. F. J.” Journal of Political Economy.” In T. “TimetoBuild and Aggregate Fluctuations. 91. “Argentina’s Lost Decade. “An Equilibrium Model of the Business Cycle. Meltzer (eds. 142. “National Borders Matter: CanadaUS Regional Trade Patterns. 159. Plosser (1983). and Banking. 3–18.). 337. Long. F. 154. Prescott (1990). and T. and C. 113. E. and M. (1977). J.” Universitat Pompeu Fabra (UPF) Economics Working Paper No. Cambridge. Wickens (2008). McGrattan. Prescott (1991). and R. (1996). Wright (1997).” Journal of Political Economy. 267–297. B. Kydland.” Federal Reserve Bank of Minneapolis Quarterly Review.” International Economic Review. 39–69. Lucas.” Econometrica. Lucas. pp. 397–414. 33. A. (1995)..” Review of Economic Dynamics. and E. pp. R. pp. McGrattan. Kydland. (1982). 721–754. 77. pp. and E. 147. 144. Kydland. 7–29. pp.” Journal of Monetary Economics. 155.” Cardiff Business School Economics Working Paper No. Amsterdam: NorthHolland. 1(2) (Supplementary Series). 4. 150. L. “Real Business Cycles. Princeton: Princeton University Press. 335–359. 141. “What Can Explain the Apparent Lack of International Consumption Risk Sharing?” Journal of Political Economy. 10. (1986). 161–178. Meenagh. 93. P. pp. “Real Wages. Prasad (2003). MA: MIT Press. Marimon (1999).” Economic Journal. “The Macroeconomic Effects of Distortionary Taxation. pp. 1345–1370. 38. and L. (1976). Cooley (ed. J. pp. 19–46.” Journal of Monetary Economics. Fallacy and Fantasy
140. Frontiers of Business Cycle Research. 146. F. 152. Minford. 152–165. (1975). and C. “Identifying the Common Component in International Economic Fluctuations. 85.” Journal of Money.
. 153. “Business Cycles: Real Facts and a Monetary Myth. “Interest Rates and Currency Prices in a TwoCountry World. “Understanding Business Cycles. Lucas. R. 267–290. “Recursive Contracts. 14. D. E. 1113–1144. Rapping (1969). Lucas. 50. K. “Econometric Policy Evaluation: A Critique.” Journal of Economic Theory. Zaragaza (2002). 104(2). “Business Cycles and Aggregate Labor Market Fluctuations. “On ‘Real’ and ‘StickyPrice’ Theories of the Business Cycle. Lucas.
167. Romer. (1989). (1997). 97.” Unpublished manuscript. Ratfai. New York: National Bureau of Economic Research. New York: National Bureau of Economic Research. “The Prewar Business Cycle Reconsidered: New Estimates of GNP. and P. C. 162. (1991). Microeconomic Foundations of Employment and Inﬂation Theory. V. “Job Creation and Job Destruction in the Theory of Unemployment. Potter. Mitchell. Shapiro (2001). 1–37. “Facts of Business Cycles Around the Globe. Benczur (2008).W. Plosser (1982). pp. “Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample?” Journal of International Economics. 173. Rogerson. 174. 177. W.” CarnegieRochester Conference Series on Public Policy. Pissarides (1994).” Journal of Monetary Economics. A.” Journal of Development Economics. Statistical Indicators of Cyclical Revivals (NBER Bulletin No. and P. 11–44. pp. 3–24. de Walque. Rodrik. (1991).” Paper presented at the IEA 15th World Congress. Phelps. S. G.” Review of Economic Studies. Rouabah (2008). 61. 81. 178. Rogoff (1983).
. (1970). 163. Mortensen. “Indivisible Labor. 168. 165. 109.. Obstfeld.” Scandinavian Journal of Economics. pp. 92. 181. pp. Ramey. Cambridge. Neftci. Present. 505–528. Mendoza. and K. 1869–1908. pp. and A. MA: MIT Press. Istanbul. Ratfai. (2005). “Is the Stabilization of the Postwar Economy a Figment of the Data?” American Economic Review. pp. R. E. (1986). pp. 166. 36. and K. (1988). Rebelo. Romer. New York: W. Supervision and Liquidity Injections: A Dynamic General Equilibrium Approach. pp. D. Benczur (2005). and C. A. 229–242. 107. “Trends and Random Walks in Macroeconomic Time Series. 172. L. O. Mitchell. and Future. Bernanke and K. Meese. 316–334. pp. Prescott. pp. 175. S. NBER Macroeconomics Annual 2000. “Nonlinear Time Series Modelling: An Introduction. 391–415.Bibliography
147
161. S. 3–16. Ohanian.). (1927). Norton & Co.” Journal of Political Economy. (1984). Burns (1938). 339–390. “The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?” In B. 176. Rogoff (2001). 69). 139–162. Rogoff (eds. pp. M. 23–40. “Economic Fluctuations in Central and Eastern Europe: The Facts. Business Cycles: The Problem and Its Setting. 25. D.” Journal of Political Economy. 4846. 171. pp.” CEPR Discussion Paper No. “Policy Uncertainty and Private Investment in Developing Countries. 180. “The Macroeconomic Effects of War Finance in the United States: World War II and the Korean War. 13. (1986). 217–238. and C. “Real Business Cycles in a Small Open Economy.” Journal of Economic Surveys. “Financial (In)stability.” Journal of Monetary Economics. 307–328. E. Turkey. 14. W. 958–992. R. “Displaced Capital: A Study of Aerospace Plant Closings. 797–818. (1999). 76.” American Economic Review. 164. 169. 87. 10. Lotteries and Equilibrium.” American Economic Review. 170. 179. pp. pp. and A. and M. 21. Nelson. Pierrard. C. “Are Economic Time Series Symmetric over the Business Cycle?” Journal of Political Economy. “Theory Ahead of Business Cycle Measurement. E. pp. C. “Real Business Models: Past.
and R. 197. Sims. 1–48. Smets. 62. J.” American Economic Review. 200. 45–109. Credit. Solimano (1993). pp. Wouters (2005). Stockman. F. and A. Solimano (eds. Rotemberg. 251–287. pp. Business Cycles. North Holland: Elsevier Science. Watson (2005).).” Journal of Monetary Economics.” Journal of the European Economic Association. “Technical Change and the Aggregate Production Function. (1937). F. 21. “Business Cycle Modeling Without Pretending to Have Too Much A Priori Economic Theory. “Understanding Changes in International Business Cycle Dynamics. “Dynamic General Equilibrium Models of Imperfectly Competitive Product Markets. 28. Solow. and Banking. “Has the Business Cycle Changed and Why?” In M.” Journal of Applied Econometrics. F. pp. Sims. J. “Business Cycle Fluctuations in U. 540–553. MA: Harvard University Press. pp. Watson (1999). 185. K. 361–387. Schorfhiede.” Journal of the European Economic Association. A. Stock. Stock. “Loss Functionbased Evaluation of DSGE Models.
. Woodford (1995). (1980). Woodford (eds. Minneapolis. Handbook of Macroeconomics. Fallacy and Fantasy
182.” Review of Economics and Statistics. Cambridge. pp.” Econometrica. pp. “The Summation of Random Causes as the Source of Cyclic Processes. Frontiers of Business Cycle Research. and R. 184. 159–230. 85.).148
Business Cycles: Fact. 645–670. 243–293. 193. 196. Serven and A. and M. pp. “Private Investment and Macroeconomic Adjustment: A Survey. Princeton: Princeton University Press. DC: The World Bank. 191.” Journal of Political Economy. and M. pp.” Journal of Applied Econometrics. pp. pp. 1. J. 5. Slutsky. The Theory of Economic Growth. 195. 312–320. (2000). and M. 15. (1972). 188. Cooley (ed. (1988). pp. 186. 3–64. Sargent. 11–30. 1123–1175. 549–577. Smets. J. New York: McGrawHill. Sims (1977). 5.” In L. T. C. Schumpeter. Tesar (1995). Gertler and K. 48. 187. (1989). 194. pp. and M.” In T. Taylor and M. 190. MN: Federal Reserve Bank of Minneapolis. 199. 968–1006. “Macroeconomics and Reality. “An Estimated Dynamic Stochastic General Equilibrium Model of the Euro Area. pp. “Comparing Shocks and Frictions in US and Euro Area Business Cycles: A Bayesian DSGE Approach. “Money. J. L. Macroeconomic Time Series. 198. 161–183. Serven. 105–146. Cambridge. Sargent. E. Watson (2003). MA: MIT Press.” Journal of Money. Rotemberg.). Washington. 168–185.). and L. C. Striving for Growth After Adjustment: The Role of Capital Formation. “Two Models of Measurement Error and the Investment Accelerator. J. Woodford (1996). R. NBER Macroeconomics Annual 2002. 97.” Econometrica. 20. 183. pp. Stock. pp. “Imperfect Competition and the Effects of Energy Price Increases on Economic Activity. 189. (1934). pp. Vol. Rogoff (eds. Wouters (2003). (1957). 39. pp. Income and Causality.” In New Methods in Business Cycle Research: Proceedings from a Conference. and C.S. and M. Singleton. “Tastes and Technology in a TwoCountry Model of the Business Cycle: Explaining International Comovements. 192. “Econometric Issues in the Analysis of Equilibrium Business Cycle Models. T.” American Economic Review. 1A. J.” In J. (1939). Schumpeter.
206.” Journal of Economic Dynamics and Control. Finn Kydland and Edward Prescott’s Contribution to Dynamic Macroeconomics: The Time Consistency of Economic Policy and the Driving Forces Behind Business Cycles. with E. 210. 508–534. (1993). pp. 208. 281–372. pp. 97. 202. Thomas. V. Prescott (1989).wiwi. Advance Report on the Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel. Weitzman. (1992). (2007). “Statistical Nonlinearities in the Business Cycle: A Challenge for the Canonical RBC Model. (2002).” Available at http://www2. 209. Cambridge. 92. pp. M. DC: The Brookings Institution. 101. (2002). Valderrama. 1.de/institute/wpol/html/toolkit. Summers. pp. 2957–2983. History. 1102–1130. D. BrookingsWharton Papers on Financial Services. Recursive Methods in Economic Dynamics.” American Economic Review.). Tesar.” In R. pp. Zarnowitz. Litan and A. The Royal Swedish Academy of Sciences (2004). 23–27. Washington. Business Cycles: Theory. L. 31(9). and R.Bibliography
149
201. pp. 10. (1997). Watson.” Federal Reserve Bank of Minneapolis Quarterly Review.” Journal of Political Economy. N. 110. Werner (1998). “A Toolkit for Analyzing Nonlinear Dynamic Stochastic Models Easily. 211. Santomero (eds. 204. MA: Harvard University Press. Vol. 1011–1041. 205. Uhlig. M. “The Internationalization of Securities Markets Since the 1987 Crash. (1986). 181–197. Indicators. “PlantLevel Irreversible Investment and Equilibrium Business Cycles.” American Economic Review. M. and I. “Some Skeptical Observations on Real Business Cycle Theory. J. Chicago: The University of Chicago Press. 203. Veracierto. and Forecasting. pp.htm/. L. Lucas.huberlin. (2007). H.
. Stokey. “Is Lumpy Investment Relevant for the Business Cycle?” Journal of Political Economy. “Measures of Fit for Calibrated Models. 207. “Subjective Expectations and AssetReturn Puzzles.
This page intentionally left blank
.
34 crosscountry correlations. 83 generalized method of moments (GMM). 129 identiﬁability of. 45 indirect inference. 66 current account. 94 animal spirits. 73 demandconstrained equilibrium. 19. 73 common factors. 124
. 16 in emerging markets. 117 goodnessofﬁt measures. 120 enforcement constraints. 33 Bellman’s equation. 110 complete contingent claims. 58 international portfolio diversiﬁcation. 65 lagging indicator. 131 kurtosis. 71 conditional heteroscedasticity. 68 investmentspeciﬁc technological change. 124 credit market frictions. 113 home production. 116 151
dynamic stochastic general equilibrium (DSGE) models. 58 aggregate price index. 135 dynamics of output. 43 capital account. 125 Kalman ﬁlter. 138 deterministic steady state. 109. 130 indivisible labor. 59 irreversible investment. 16 leading indicator. 42 business cycles coincident indicator. 39 dynamic factor analysis. 16 calibration. 54 homeproduced good.Index
accommodative monetary policy. 49 inside money. 80 international risk sharing. 54 impulse response functions. 110 for large crosssections. 101 international.
131 unconditional moments. 49 nonlinear time series models. 101 transactions services from money and banking. 66 price rigidities. 33 countryspeciﬁc. 101 symmetric equilibrium. 137 New Keynesian model. 33 reverse causality. 56 transition equation. 67 puzzles. 48 quantity anomaly. 97 real business cycle (RBC) theory numerical solution. 137 multiple equilibria. 77 nontraded good. 54 market incompleteness. 123 nonnormal disturbances. 50 Solow residual. 121 price anomaly. 58 structural. 137 selffulﬁlling expectations. 82 maximum likelihood estimation frequency domain. 56
. 137 seminonparametric (SNP) model. 42 posterior distribution. 124 shocks. 97 total factor productivity (TFP). 53 industryspeciﬁc. 132 monopolistic competition. 94 multisector models. 88 vector autoregression (VAR). 78 energy. 38 spectral density function. 135 Sudden Stops. 33 skewness. 49
Business Cycles: Fact. 43 search models. 124 social planner’s problem.152 labor hoarding. 111 stabilization policy. Fallacy and Fantasy Riccati equation. 94 productivity spillovers. 33 nonconvexities in labor supply. 112 measurement equation. 63 government consumption. 132 structural models. 55 limited risk sharing. 44 unobservable index models. 78 optimal linear regulator problem. 81 lottery models. 78 monetary. 132 predictive density. 2 state space representation. 66 real balances. 99
market good. 98 technology. 38 trade balance. 124 nonspecialization in endowments. 110 variable capacity utilization.