Full text of Economic Review (Federal Reserve Bank of Atlanta) : May/June 1993, Volume 78, Number 3

View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Jcpnpmic
/Review
May/June 1993
Volume 78, Number 3

Federal Reserve
Bank of Atlanta

In This Issue:
i n t e r n a t i o n a l Policy Coordination:
Can We Have Our Cake a n d Eat It Too?
I n f l a t i o n a n d t h e Yield Curve
FYI—Consumer
Prices:
E x a m i n i n g H o u s i n g Rental C o m p o n e n t s
R e v i e w Essay—Capital Ideas: The Improbable
Origins of Modern Wall Street

r
-

-S' -

,, • •

. r-- \

•

.
-

M
. .
v -

' .

>- ...

.:
.

r '

. r

- *

—

• ;

.
' 'v

•

...

. ••;

-i

. :

. ' f\

.
V

-r

- " V

e"
c

--'-.i

(

• -.V v

>.••-../

. —- • ' • 4I '
/.
.
/ •
r- V -i»
"
ir-L
• / -; S y . t .

•

'"/

- <

^.... . . ...
/.

• •"v. " ,

;

•

, v^

f
• ì

•:
A.

•

• '

r
•

>
'

V • -

-V •

• "

•

, -

' -

•

" •

- A

—

-/-.• .

^ V

<C\\
.

. -Y

http://fraser.stlouisfed.org/
Federal Reserve
•
' Bank
^ of St., Louis -

' " .

—

- '' -

; '

. t/

•

Eepnpmic
Review
May/June 1993, Volume 78, Number 3

Federal Reserve
Bank of Atlanta

President

Robert P. Forrestal
Senior Vice President and
Director of Research

Sheila L. Tschinkel
Vice President and
Associate Director of Research

B. Frank King

Research Department
William Curt Hunter, Vice President, Basic Research
Mary Susan Rosenbaum, Vice President, Macropolicy
Thomas J. Cunningham, Research Officer, Regional
William Roberds, Research Officer, Macropolicy
Larry D. Wall, Research Officer, Financial

Public Affairs
Bobbie H. McCrackin, Vice President
Joycelyn T. Woolfolk, Editor
Lynn H. Foley, Managing Editor
Carole L. Starkey, Graphics
Ellen Arth, Circulation

The Economic Review of the Federal Reserve Bank of Atlanta presents analysis of economic
and financial topics relevant to Federal Reserve policy. In a format accessible to the nonspecialist, the publication reflects the work o f the Research Department. It is edited, designed, produced, and distributed through the Public Affairs Department.
Views expressed in the Economic
eral Reserve System.

Review are not necessarily those of this Bank or of the Fed-

Material may be reprinted or abstracted if the Review and author are credited. Please provide the
Bank's Public Affairs Department with a copy of any publication containing reprinted material.
Free subscriptions and limited additional copies are available from the Public Affairs Department, Federal Reserve Bank of Atlanta, 104 Marietta Street, N.W., Atlanta, Georgia 30303-2713
(404/521-8020). Change-of-addrcss notices and subscription cancellations should be sent directly to the Public Affairs Department. Please include the current mailing label as well as any new
information. ISSN 0732-1813

ontents
May/June 1993, Volume 78. Number 3

/nternational Policy
Coordination: Can We Have
Our Cake and Eat It Too?
Marco Espinosa and
Chong K. Yip

inflation and the
Yield Curve
Peter A. Abken

Along with the increasing development of international goods and
capital markets, countries have shown growing interest in crafting
joint macroeconomic policies. This article examines some of these
policy coordination efforts, particularly those the authors classify as
"incomplete," which involve countries agreeing to use policies to
achieve some common objective but do not require relinquishing all
autonomy to a central authority in the design of macropolicies.
The authors also review some of the theoretical research on the
subject, which suggests that economies linked by capital and service
markets—open economies—generate externalities for each other.
Countries choosing policies independently cannot fully incorporate
into their decision-making processes the effects of their choices on
the world community. According to the literature reviewed, only full
coordination of macropolicies results in the internalization of policy
effects and leads to improved welfare worldwide.
The authors emphasize that currently, however, most coordination efforts are incomplete. At the same time, only full coordination
is able to eliminate the transmission of negative externalities, and
decisionmakers are therefore tempted to try to reconcile mutually
exclusive policy approaches.

The determinants of interest rates across the spectrum of bond
maturities are of keen interest to borrowers and lenders as well as to
economists. The inflation forecasts implicit in interest rates carry
strong credibility with most market observers because interest rates
in part represent bets backed by wealth rather than casual forecasts
with little at stake. The trouble is that interest rates are influenced
by more than just market expectations of inflation; other factors
cloud the inflation signal perceived in the term structure or relationship between interest rates of various maturities.
Despite the complexity of interest rates, people who want to
gauge inflation expectations turn to the yield curve. This article discusses key research studying the information on inflation contained
in the nominal term structure of interest rates. The current evidence
suggests that the yield curve does indeed give useful forecasts of inflation, especially at longer-term horizons, but much still needs to
be learned about the various factors that influence nominal rates.

FYI— Consumer Prices:
Examining Housing
Rental Components
R. Mark Rogers, Steven W.
Henderson, and Daniel H.
Ginsburg

iteview Essay—Capital Ideas:
The Improbable Origins of
Modern Wall Street
by Peter L. Bernstein
Thomas J. Cunningham

The Consumer Price Index (CPI) is one of the most closely followed indicators of conditions in the U.S. economy. Recently, however, questions were raised about the validity of several rental
components in the CPI. For the 1990-91 recession and subsequent
recovery, these components were stronger than expected, seemingly
out of line with other CPI housing components. Because the rental
components make up more than one-fourth of the CPI, the concern
over their accuracy is legitimate.
This article analyzes how well the CPI rental components reflected actual conditions from 1990 to 1992 and examines the
sources of apparent divergence. The authors conclude that after taking into account appropriate lag times, the CPI for the residential
rent component was correlated strongly with housing prices and
multifamily vacancy rates over the same period. Owners' equivalent
rent was at best weakly correlated with these variables, but analysis
of the index's methodology suggests that one should not expect
such relationships to be tight. For the hotel/motel component,
changes in methodology can account for the apparent discrepancy
with overall prices for hotels during the 1989-91 period.

Bernstein's personalized, often anecdotal history of finance during the past thirty years reflects his experience with both the theoretical and applied sides of finance theory. In this essay, the reviewer,
who finds Bernstein's book entertaining and informative if slightly
broad for the academic and finance professionals who will be drawn
to it, sketches this era of "spectacular transformation" in finance
theory and practice.

international Policy
Coordination:
Can We Have Our
Cake and Eat It Too?

Marco Espinosa and Chong K. Yip

The authors are an
economist in the macropolicy
section of the Atlanta Fed's
Research Department and an
assistant professor of
economics at Georgia State
University, respectively.
They
thank Frank King, Eric Leeper,
and Mary
Rosenbaum
for helpful comments
on earlier drafts.

Federal
Reserve B a n k of Atlanta

ecent events in European financial markets—in particular, Britain's
and Italy's (at least temporary) departure from the exchange rate
mechanism of the European monetary system, the French and
Danish referendum votes on European union, as well as the emphasis on coordination stressed at the recent G-7 meetings—have
generated a wave of headlines and have focused attention on the various
ways in which macroeconomic policies are coordinated internationally.
When a country's economy has no links with other economies—when an
economy is "closed"—the fiscal or monetary policy behavior of other countries is irrelevant. However, when economies are linked by capital and services markets—that is, are "open"—policy decisions at home may have an
impact abroad and vice versa. As the world is becoming increasingly integrated in trade and financial markets, economic events in one country inevitably have a bigger impact on other countries. Under global capital markets,
for example, most countries have access to the same pool of world savings,
and individual governments' borrowing and lending activities affect interest
rates—and consequently, economic activity—worldwide. For this reason,
the U.S. government budget deficit has been broadly blamed for the high
worldwide interest rates that characterized the 1980s, and the "high" interest
rates set by Germany's Bundesbank and the recent "caution" of the Bank of
Japan have recently been held responsible for retarding economic growth in
Europe, and perhaps worldwide.
As the development of international goods and capital markets has progressed to a degree unseen during the 1960s or 1970s, countries have shown
increasing interest in crafting and adopting joint macropolicies or participating in international coordination efforts—at least implicitly as a means

EconoinicRevieu'9

of ameliorating the transmission of negative policyspillover effects—and a significant body of research
has focused on designing models to analyze the impacts of such efforts. Perhaps the first question to be
addressed is that of why countries do not simply move
toward a world consisting exclusively of closed economies that would not be susceptible to the negative eff e c t s that might be transmitted across countries.
Implicit in international coordination efforts is the notion that the positive effects of market integration outweigh its potential negative effects across countries
and that most of the potentially undesirable effects can
in fact be eliminated or at least drastically reduced
with the right coordination scheme. Although, as the
discussion makes clear later, there is no such thing as
the "right" coordination scheme, the point of departure
for discussion is the assumption that integration is
worth pursuing.
A number of theoretical perspectives, each with its
own policy implications, have found favor with policymakers over time. The purpose of this article is to
examine some of the various macroeconomic policy efforts, particularly for the ways in which they attempt
to minimize the intercountry transmission of negative
spillover effects of policy decisions. The discussion also considers some of the theoretical research on this
subject.

i?ecent Coordination Efforts
Efforts to coordinate macroeconomic policies are
not new. For example, during the Bretton Woods era
(1944-73), a degree of policy coordination was embodied in the system of fixed exchange rates against
the U.S. dollar, which was then tied to gold. In the beginning the Bretton Woods agreement allowed currency exchange rates of member countries to vary 1 percent
against the U.S. dollar, which in turn was pegged to
gold at a fixed rate—$35 per ounce. By agreeing to fix
exchange rates, the monetary authorities involved
committed their countries, unconditionally, to trade
foreign and domestic currencies at certain exchange
rates in the foreign exchange market and thus to modify their domestic money supplies to achieve these
rates. Partly because the agreement restricted to some
degree the monetary policies of the forty-four countries participating, the Bretton Woods system broke
down in the early 1970s.
Efforts to coordinate policies have continued, however. A more recent example of exchange rate coordi-

2
Econom ic Review

nation is the Plaza Accord of 1985, established between the G-5 countries (the group of five leading industrial countries—France, Japan, the United States,
the United Kingdom, and West Germany). Through the
early 1980s, the dollar had appreciated steadily against
major currencies, and by December 1984 the dollar
exchange rates against the German mark and Japanese
yen were 3.1 and 247.96, respectively. Along with the
strong dollar came large U.S. trade deficits as Americans imported more and exported less. Policymakers
viewed these developments as alarming and in need of
reversal. The Plaza Accord provided multilateral support for reducing the foreign exchange value of the
dollar, with the purpose of further reducing the U.S.
current account deficit. 1
Policy Coordination in Europe. Because half of
European trade occurs between European countries,
reducing exchange rate volatility has been a major
consideration in coordination efforts. The assumption
has been that high volatility of exchange rates makes it
hard to predict the terms of trade (the relative prices of
imports over exports), with consequent costly efficiency losses. 2 Some of the first attempts at coordinating
exchange rates in Europe date back to the Joint Float
agreement of 1972. This agreement called for its members to hold their currencies' bilateral exchange rates
to a 2.5 percent variation. Although the origins of the
EMS proposal can be traced to the 1957 Treaty of
Rome, which founded the European Community, the
Joint Float Agreement provided the working foundations of the European Monetary System (EMS) as it
originated in 1979.
Since 1979, EMS exchange rate coordination efforts have been overseen by the exchange rate mechanism (ERM). The ERM calls for the exchange rates in
each member country to vary no more than 2.25 percent from its bilateral central rate. To achieve this
goal, central banks are willing to intervene in the exchange rate market buying and selling their currencies. 3
Coordination efforts such as those discussed above
can be classified as "incomplete," in the sense that the
countries involved in these efforts agree to use their
macroeconomic policies to achieve a common objective—a fixed exchange rate or an exchange rate band.
If the goal is achieved, the member countries may realize efficiency gains as a result of the reduced exchange
rate volatility and increased volume of trade. However,
other than taking actions necessary to achieve the exchange rate objective, the participating countries are
free to use their macroeconomic policies as they see
fit. Clearly, countries involved in such exchange rate

May/June 1993

coordination efforts do not relinquish to a central authority all autonomy in macropolicy decisions. At the
same time—and in feet because countries retain considerable autonomy—these coordination arrangements
do not eliminate negative spillovers across countries.
Some countries have therefore pursued more comprehensive coordination commitments that would ameliorate the transmission of negative macroeconomic
policy spillovers. As evidence of this trend, the stated
goals of the EMS have evolved into an economic convergence of its members as a means of eventually
achieving monetary union, and the ERM is seen as an
intermediate step in that direction. Unlike the coordination efforts discussed earlier, this proposed union
supposedly would have an independent European Central Bank (ECB) and an associated European System
of Central Banks (ESCB) overseeing implementation of the common monetary policy. Article 7 of the
Maastricht Treaty protocol (1992), for example, states
that "neither the ECB nor a national Central Bank, nor
any member of their decision-making bodies shall
seek nor take instructions from EC institutions or
member governments." In fact, member governments
would relinquish their autonomy in setting monetary
policy to the independent ECB and ESCB. This type
of coordination is referred to here as "complete."
The Maastricht Treaty, however, does not stop at the
prospect of a unified monetary policy with a single currency. Some of the stated goals of the economic community allude to a "deeper integration into a virtual
federal Europe." Such statements have inspired several
studies that try to establish a parallel between the envisioned European economic union and the United States'
complete de facto coordination agreement among its
member states (see, for example, Paul Van Rompuy,
Filip Abraham, and Dirk Heremans 1991). As envisioned by these authors, EMU members would function much like states of the United States, not only
giving up sovereignty over seignorage (revenue raised
by printing money) as a means to finance government
deficits but allowing for complete mobility of all resources across member countries and the creation of a
supernational government actively involved in all aspects of the community's economy, with individual
country governments playing a subsidiary role. As
Ralph C. Bryant has commented, such federalism issues are "well beyond the domains of 'coordination'"
(1993). This discussion concentrates instead on issues
related to coordination as Bryant has defined them:
"Coordination goes further than mutual recognition in
focusing on cross-border spillovers and 'arbitrage
pressures' eroding the differences among national

Federal
Reserve Bank of Atlanta

economies and policies. And coordination is more ambitious in promoting intergovernmental cooperation to
deal with them. Coordination involves jointly designed
mutual adjustments of national policies (commitments
about time paths of policy instruments, not merely aspirations about time paths . . .)" (1993, 11).
As will be discussed in more detail later, only complete coordination agreements effectively eliminate the
transmission of policy effects across countries, but
they hinge on a country's relinquishment of autonomy
in choosing policy. The distinction between complete
and incomplete coordination is a crucial one because
some policymakers and analysts often express the belief that countries can reap the benefits of a complete
coordination arrangement while committing only to
incomplete coordination—that it is possible to have
one's cake and eat it too. 4
This trade-off between the loss of policy autonomy
and gains from eliminating negative spillovers may
help explain why a complete coordination effort among
countries seems so elusive and may ultimately be impossible to attain. In fact, the European "currency crisis" of September 1992 is a reminder of other failed
coordination efforts and of the apparent fragility that
surrounds the process. However, the emergence of a
two-tier approach to coordination in Europe, with
some countries firmly committed to EMU while others
lag, illustrates the belief that coordination of macropolicies is considered beneficial and should not be
abandoned. 5

An Economic Modeling Perspective
To help understand the benefits and possible costs of
policy coordination, economists have developed coordination models. While they are in general too simple
to capture fully the complex features of coordination,
these models shed valuable light on some of the major
issues that may inhibit joint policy schemes.
Externalities. The "invisible-hand" principle of
Adam Smith, familiar to any beginning economics student, states that the best way to promote social wellbeing is to allow everyone to pursue his or her own
interest. It is also well known, however, that the principle may fail under the presence of what economists
call "externalities." The most often cited example of
negative externalities is that of a plant discharging polluted water into a river that is a town's only water
source. Clearly, the sole pursuit of what the plant considers in its own best interest does not promote the

EconoinicRevieu'

welfare of the whole society; there are quite different
private and social costs at stake.
In general in such situations, certain policies may
restore efficiency by creating appropriate incentives for
firms or other economic actors to achieve, on balance,
what is best for the society as a whole. For example, in
the case involving water pollution a pollution-control
policy could be established: having to pay a fee every
time waste is dumped would help polluters realize
that natural resources they use are not "free." The idea
is that they would then "internalize" this cost in their
decision-making process and have incentives to look
for alternatives to dumping, such as investing in cleaner
technologies. Such a policy can be achieved, however,
only if there is a central authority to enforce it.

The Mundell-Fleming models suggest that a government could manipulate monetary and fiscal policies
in such a way as to attain internal and external "balance" simultaneously, namely, to achieve current account balance and full employment. This paradigm
still motivates many of the discussions on international
policy coordination. 8 For example, some analysts suggest that the U.S. proposal at the 1978 summit meeting
of the G-7—that the United States, Japan, and Germany coordinate their policies in an expansionary effort—was motivated by the threat of trade deficits. If
the incomes of a country and its trading partners grow
simultaneously, other things being equal, the possibility of external imbalance (large trade deficits) is reduced.

It may be helpful to think of individual countries as
analogous to the residents of the town affected by water pollution. To avoid negative spillovers across countries, coordination of macroeconomic policies may be
desirable for the well-being of the world economy. 6
This line of thinking characterizes several of the formal attempts to study international interdependencies.

These models assume that changes in domestic
government expenditures influence domestic output,
which in turn has an impact on the foreign country's
current account and thus affects the level of foreign
output. Likewise, a drop in the foreign country's output has an effect on the domestic current account and
output.

Open-Economy Macroeconomics
John M. Fleming (1962) and Robert A. Mundell
(1963) set the foundation for the formal analysis of international macroeconomic policy coordination. 7 They
analyzed the feedback of monetary and fiscal policies
between two countries. Their models consist of goods
and asset demand functions (including money) in the
Keynesian tradition. National expenditures—consumption and investment—are assumed to depend on domestic output and the real (inflation-adjusted) rate of
interest while net exports depend on income at home
(imports) and abroad (exports) as well as the real exchange rate. For instance, a rise in German income increases Britain's exports because Germans can afford
to buy more goods, including imported goods. Another assumption is that public demand for money is a
function of income and interest rates at home—that is,
as income rises and people want to buy more goods,
they increase their demand for money. At the same
time, if the opportunity cost of holding money goes up,
the public moves away from money and into higheryield assets. A third assumption is that changes in private domestic holdings of foreign securities—capital
outflows or inflows—are a function of interest rates at
home and abroad. Countries in the economy are assumed to be similar in these aspects.

4
Reo no in ic Review

The Mundell-Fleming models also illustrate how
the effectiveness of aggregate demand policies in an
open economy may be affected. For example, a policy
increasing government expenditures, intended to lead
to greater domestic output by stimulating purchases of
domestic commodities, may be diminished in its effectiveness by a concurrent increase in purchases of import goods.
These models constitute a first step toward understanding the transmission of macroeconomic policy
effects between countries. They demonstrate how the
consequences of adopting certain policies at home depend not only on those policies themselves but also on
policies implemented abroad.

To Coordinate or Not to Coordinate
Recognizing countries' interdependence and the
potential for negative spillover raises the question of
whether a country would be better off choosing policy
on its own or in cooperation with other countries. The
work of Koichi Hamada (1976) pioneered the analysis
of international monetary policy coordination. 9
Hamada's models are based on a game theoretical
approach that views governments as solving their economic choices after considering a series of strategies,
just as card players do (playing individually or as a
team). 10 There is evidence that studying macropolicy

May/Kmc 1993

coordination in the context of strategic considerations
is a sound approach. For example, one could argue that
it was not a coincidence that the North American Free
Trade Agreement was put on the table at about the
time that General Agreement on Tariffs and Trade
(GATT) negotiations appeared to be stalling. A trade
alliance within North America would show the European Community how the United States was ready to
move forward with free trade, with or without the European Community. In fact, one could argue that it is
progress toward NAFTA that has led to recent concessions in agricultural policy from the EC in the GATT
negotiations.
Hamada's analysis was performed for a world in
which exchange rates are fixed, which in fact they had
been in the then-recent Bretton Woods era. However,
Hamada's insights do not hold only for fixed exchange
rates. In the framework of game theory, coordination
of monetary policies would also be beneficial in a
flexible exchange rate environment like those under
which many countries operate today.
As an illustration, consider a world with free trade
of goods and capital, and assume that there is a tradeoff between unemployment and inflation. Countries
choosing monetary policies independently may have a
bias to choose an expansionary policy in order to
achieve a target rate of unemployment. For instance, a
particular country tries to lower its unemployment rate
by running an expansionary monetary policy. Such a
policy may trigger the following sequence of events:
At least in the short run, there may be a depreciation
of the country's currency with respect to other countries, which would in turn lead to a reduction of its
current account deficit against those countries. The affected countries may react by pursuing expansionary
policies themselves, in order to counter their currency's appreciation. In doing so, each country weighs
only the inflationary consequences of its policies for
its own economy (in the same way in which the factory owner causing water pollution in the example above
does not internalize the negative impact of its actions).
However, in the absence of any trade barriers the simultaneous expansionary policies would result in a
higher rate of inflation worldwide. The fact of decentralized actions by countries precludes a country's
ability to internalize the costs its actions impose on the
world community. If, on the other hand, each of these
countries were to give up its autonomy in choosing
monetary policy, agree on a common objective, and
have a central authority directing a single, common
monetary policy, all would be forced to consider the
worldwide inflationary impact of their policies.

DigitizedFederal
for FRASER
Reserve Bank of Atlanta

Hamada determined that whenever free trade of
goods existed, countries coordinating their actions generally achieved more desirable outcomes than resulted
when countries acted independently. Hamada recognized, however, that coordination of monetary policies
is difficult to achieve. A country faces trade-offs when
setting economic goals, and coordination further restricts each country's already limited policy options to
achieve the best possible developments at home.
Consequently, each country may assign different
priorities to various goals. For example, assume that
Germany and Italy have monetary policy at their disposal and that each recognizes the trade-off between
inflation and unemployment. Germany may decide that
low inflation is its priority while Italy may decide that
it is willing to suffer higher inflation rates to achieve
lower unemployment rates. Now consider that Germany and Italy decide to coordinate their efforts and
relinquish their domestic monetary policies for a common central monetary policy. The question becomes
how to assign weights to each country's objectives.
That decision is a political one, about which economic
theory is silent.
What is a policymaker to sift from these abstract
concepts underlying coordination? The discussion so
far seems to say that countries may be better off under
coordinated policy than acting independently. More precisely, under free trade countries seem to benefit from
choosing in unison a single, common policy. That observation is a potent policy recommendation. It is also,
however, a typical example of theory having little to
say about the actual process—the logistics and negotiations—required for its implementation.

Complete versus Incomplete
Coordination
It is important to note that fixed exchange rate
agreements and " m a n a g e d " floating exchange rate
agreements like the ERM are not the type of coordination Hamada studied. The nature of the policy coordination implicit in fixed exchange rate agreements may
reduce exchange rate volatility and thereby improve
efficiency. However, as was discussed earlier, under
such incomplete coordination agreements the member
countries continue to enjoy some degree of discretion
over their macroeconomic policies and are not able to
internalize fully the consequences of their policies. For
example, Germany is still committed to the EMS, the
intermediate goals that the EMS represents, and the

EconoinicRevieu'

eventual creation of a European Central Bank. But
Germany's policy choices about financing its unification costs have prevented interest rates from coming
clown in Europe, indirectly creating negative externalities, or costs, for its neighbors. The type of policy
coordination Hamada suggested above involves complete coordination—countries' total relinquishment of
autonomous policies such as would characterize the
EMU in its third stage, when a European central bank
is in place.
It is important, however, to recognize that so far
the theory on which a recommendation of complete
macropolicy coordination is built has several shortcomings. For example, Hamada's models do not specify the
mechanisms that affect consumption and investment
decisions, making it hard to analyze the ways in which
alternative policies may affect these behaviors. In addition, time plays no role in these models although interest rates are the price of intertemporal allocation of
consumption and as such should require that temporal
dynamics be a key consideration in analysis. Asset demands depend on current and future interest rates—for
example, individuals choose assets to acquire on the
basis of prevailing interest rates as well as expectations
regarding future interest rates. Models such as Hamada's, however, are silent as to how agents build their
expectations regarding future interest rates. Because
they are static, they do not address the question of
whether multicountry interdependence is only temporary. In spile of their shortcomings, these kinds of models constitute a solid first step toward understanding the
international transmission of macroeconomic policies,
and some of the models' insights apply for more complicated and realistic environments.

D y n a m i c General Equilibrium Models
Dynamic general equilibrium models were developed
to address some of the problems identified above.
These models reproduce the simultaneous nature of
economic variables and assume that interest rates and
prices are determined by the underlying economic
structure rather than by exogenous behavior patterns
that may not be related to the fundamentals of an
economy. As their name implies, dynamic general
equilibrium models are also able to incoiporate the dynamic nature of economic systems. 11
The models discussed earlier implicitly recommend
international m a c r o e c o n o m i c policy coordination.
Given these models' shortcomings, however, it is im-

Econom ic Review
6

portant to investigate the robustness of that implication. Docs the same policy prescription hold in a dynamic general equilibrium framework? The discussion
that follows concentrates on models that address monetary and fiscal policy.
Monetary Policy. Neil Wallace (1984) set out to
dispel the notion that there is a single "best" monetary
policy for a nation. His model, which can be thought
of as providing long-term recommendations, emphasized the notion (perhaps obvious but rarely acknowledged) that monetary policy has different effects across
different economic groups. (Few would argue, for example, with the fact that U.S. monetary policy in 1992
generated a surge of mortgage refinancing at home but
also displeased holders of three-month certificates of
deposit, which experienced record low returns).
Preston J. Miller and Wallace's (1985) analysis of
international coordination of monetary policies is an
open-economy extension of Wallace's model. Miller
and Wallace start with the observation that although
countries appeared to have gained more discretion
over their policies under a flexible exchange rate
regime, flexible exchange rates did not ameliorate the
transmission of negative spillover effects across countries. In this context, coordination of monetary policies
has been abdicated as a means of improving the workings of flexible exchange rates.
Miller and Wallace's work is designed to contrast
two approaches to choosing monetary policy. In the
first, each country chooses its own monetary policy,
recognizing the fact that other countries' independent
decisions will influence the ultimate effects of its policy and that those decisions are beyond its control. The
second alternative, akin to some of the G-5 or G-7
coordination attempts, involves all countries jointly organizing a central authority to choose a common monetary policy.
At the core of whether or not monetary policy coordination makes a difference is the fact that when countries choose policy independently of each other, their
policies have asymmetric effects at home and abroad.
Miller and Wallace illustrate this asymmetry with the
following example. In their model, when the monetary
authority in the United States engineers higher real interest rates at home, in a well-integrated international
capital market long-term real interest rates will also
tend to increase around the world. However, the impact of such policy on the purchasing power of economic agents trying to cash in their savings at the time
the policy goes into effect will be different at home
than abroad. A tighter monetary policy in the United
States will increase the purchasing power of the people

May/June 1993

at home cashing in their savings, whereas it will reduce the purchasing power of the same group of savers
abroad. Without themselves tightening, foreign countries suffer from higher rates, which translate into a
heavier burden for servicing outstanding government
bonds that would, in turn, require the central bank to
monetize part of the government deficits. By acting independently of other countries in the Miller-Wallace
model, the United States has failed to internalize the
consequences of their actions abroad. If on the other
hand all countries were to choose policy in unison,
each would be able to internalize their policies' consequences for other countries and thus choose policy accordingly.
Miller and Wallace point out that although their
analysis suggests that cooperation is desirable, it does
not determine whether the common policy to be adopted should be a loose or tight monetary policy. And, as
emphasized in Wallace (1984), the choice is important
given that monetary policy does not have a uniform
impact across economic groups.
In the Miller-Wallace model the final form of a
common monetary policy evolves as a consensus that
takes into account the weights assigned to different
economic groups in the different countries. Unfortunately, economic theory suggests nothing in terms of
how to distribute such weights; as in Hamada's analysis, such decisions are political ones that have to be
hammered out in political negotiations. For this reason, Miller and Wallace are reluctant to view their
analysis as prescribing a specific policy recommendation. They acknowledge, however, that their analysis
suggests that adopting some common monetary policies could improve welfare worldwide.
A flavor of the negotiations necessary for resolving
differences among countries attempting coordination
was recently provided by the Maastricht Treaty. Danish
failure to ratify the treaty in early 1992 and the nervousness over the French referendum vote later in the
year may partially explain why the proposed third
stage of EMU—the creation of a European Central
Bank—seems sometimes so elusive. 12
Fiscal Policies. During the 1980s the U.S. government deficit escalated to magnitudes on the order of 5
percent of the gross domestic product. As mentioned
earlier, the size of the U.S. deficit has been seen as
contributing to high interest rates worldwide. At the
same time, the type of coordination among EC countries has gone beyond monetary policy to involve tax
and industrial policies. In light of these developments,
some economists have turned attention to analyzing
the coordination of fiscal policies.

Federal Reserve Bank of Atlanta

Patrick J. Kehoe (1987) investigated whether countries benefit more by coordinating their tax policies,
particularly income taxes. Kehoe's analysis is developed in a dynamic general equilibrium model in which
there is a household sector, a production sector, and a
government sector. The government finances its expenditures via income taxes. Government-produced
goods are turned over to households. Higher taxes lead
to an increase in government-produced goods, which
are valued by the current generation of households.
Capital purchases, however, are a function of disposable
income so that higher taxes lead to lower quantities
of capital and, consequently, lower output and consumption levels for all future generations. Less capital
also leads to higher real rates of interest (the interest
rate goes up as capital becomes more scarce). These
are the trade-offs that a policymaker has to consider in
evaluating a tax policy.
As discussed earlier, when countries choose tax
policies independently of each other but in the context
of integrated capital markets, higher taxes will represent higher interest rates not only at home but also
abroad; countries choosing policy independently do
not internalize the consequences of their actions on interest rates worldwide. Under perfect integration of
capital markets, the incentive to increase taxes at home
increases with the number of countries in the world: A
government acting on its own experiences the positive
short-term effects of taxes regardless of the number of
countries in the world; on the other hand, the government suffers only in a diluted way the negative cumulative effect of higher interest rates because their
marginal contribution to higher interest rates worldwide is smaller the larger the number of countries in
the world. Clearly, coordination of macropolicies
again could allow all countries to internalize the negative externalities that they impose on each other, and,
therefore, coordination is better than noncoordination
of fiscal policies for all countries involved.
A natural question to ask at this point is whether
cooperative outcomes of the sort described above will
necessarily amount to giving up autonomy in choosing
policy. Roberto Chang (1990), who has provided another model of fiscal coordination, also found that the
world economy would be better served by having coordinated fiscal policies among countries. Chang has
shown that under some circumstances countries would
be able to internalize the consequences of their independent macroeconomic policies and thus avoid the transmission of negative externalities across countries without giving up autonomy in choosing policy. If, for example, countries pledged to impose fiscal self-discipline .

Econo in ic Revieu'

provided that other countries did the same, and if each
threatened to abandon this discipline if others did, internalization would be possible. The policy implications of this research make its application promising:
if credible threats could be implemented and the rules
of the game clearly specified, there could be considerable savings of resources in achieving an outcome
comparable to that of official coordination.

Can Coordination Be Undesirable?
While it is true that much of the economics literature
suggests that under free trade coordinating macropolicies—both monetary and fiscal policies—is desirable,
there may conceivably be situations in which coordination could be counterproductive. For example, Kenneth Rogoff (1985) has shown that cooperation among
policymakers can in certain circumstances lead to a
lower level of social well-being than independently set
policy.
R o g o f f s contribution hinges on the concept of
"time inconsistency." 13 At the core of the idea of time
inconsistency is the fact that there are incentives for
governments and economic agents to renege on past
commitments. Economic agents care about the sequence of their lifetime returns, not just consumption
at a particular time. Individuals look ahead, and their
expectations play a crucial role in their decision-making.
If governments fail to recognize the importance of this
process, they may pursue policies that seem optimal
but involve reversing past commitments. For example,
a government may announce that it will reduce capital
taxes permanently to encourage investment. Once capital investment has taken place, the government may
renege on its promise and raise capital taxes, promising never to do it again. Such a tax increase would not
distort private investment decisions because capital investment has already taken place, and the government
would have greater expenditure capabilities. However,
economic agents observe the government's action and
will discount future announced policies. In this scenario, the incentive for the government to renege on its promises may be recognized ahead of time, and the tax cut
originally announced may be of little or no relevance.
Such time-inconsistent policies seem likely to prevent
society from achieving the best possible outcome. 14
Rogoff found that when time inconsistencies exist,
cooperation among countries can be counterproductive.
For example, in the absence of monetary policy coordination any unilateral effort by a central bank to inflate

8
Econom ic Review

(to engage in time-inconsistent policies) will cause its
real exchange rate to depreciate, with the consequent
price increase on imported goods driving up inflation at
home. Such a concomitant depreciation and fears of
accelerating inflation help check the incentives for a
central bank to expand its money. However, in a cooperative arrangement the real exchange rate has no such
tempering influence because each central bank can
count on the others to match any money supply increase without any impact on the real exchange rate.
Cooperation thus forces wage setters to set a higher
rale of nominal wage growth in order to ensure that the
central banks will ratify their target real wage. A cooperative regime may then be characterized by systematically higher inflation rates.
One criticism of Rogoff's findings is that the time inconsistency in his model exists because the governments' economic goals differ from those of their citizens
(see note 14). It has been argued that if the governments and their citizens share economic goals, Rogoff's
results may not occur. However, authors like Kehoe
(1989) have shown that time inconsistencies can arise
in settings in which governments and citizens share the
same economic goals, thus leaving an environment
where cooperation can be counterproductive.
Kehoe found that in the context of an open economy
with capital mobility, a country's desire to renege on a
promise not to tax existing capital is removed by the
threat of capital flight; savings will flee the country with
the highest income tax. Competition across countries
and capital mobility act as an enforcer to eliminate time
inconsistency. However, with fiscal cooperation between countries, this enforcement falls away because
each government can count on the other to match any capital tax increase and thus eliminate the potential for capital flight. According to Kehoe's findings, international
fiscal cooperation exacerbates the time-inconsistent problem and is therefore counteiproductive. To summarize,
if countries coordinate policies but cannot precommit to
never deviating from the agreed-upon policy, such coordination efforts may be counterproductive.

71ie EMU: A Closer Look
Some economists argue that the goal of EMU is to
establish a central authority that would set a common
monetary policy and have member countries agree not
to deviate from the policy. If so, coordination established under the auspices of the EMU should not be
counterproductive.

May/June 1993

It is a matter of some concern that the requirements
established by the EMS for economic convergence are
somewhat vague. 15 Further, each member country may
fulfill the EMS requirements and still maintain dissimilar policies, and adherence to the guidelines may not
be sufficient to guarantee a smooth transition to a monetary union. Even if a smooth transition is achieved
and a common monetary policy adopted, there is no
mechanism to ensure adherence to a common fiscal
policy. For example, suppose that Germany met the
guidelines to become a member of the EMU, that the
union took place, and that all countries agreed to adhere to a common monetary policy. If Germany discovered at a later date that the burden of unifying the
two Germanies implied larger-than-anticipated deficits,
the country could de facto renege on the agreed-upon
fiscal policies, creating the problem of time inconsistencies. According to the research reviewed above, only
if EMU would lead to each member's precommitment
to fiscal and monetary policy rules could a union be
productive.

Partial Coordination
The bulk of the literature on international coordination concludes that coordination of macroeconomic
policies that is complete and undertaken by all the
countries in the world (at least all the countries that
have a major impact on goods and asset prices worldwide)—in other words, "full coordination"—is preferable to having decentralized setting of policies.
As was discussed above, the rationale behind this
dictum is that complete coordination allows countries
to internalize the negative externalities that might otherwise be transmitted across countries. In reality, although there have been several initiatives to engage in
full coordination, it has not been accomplished. Full
coordination may, therefore, be thought of as a benchmark. More commonly, a small group of countries
agrees to coordinate policies, and the result is "partial"
coordination. 1 6 EMU is actually an example of this
type of coordination. 17 In fact, the original goal for
which the European Community was founded was to
achieve free trade among its six members while adopting a common policy with respect to nonmembers. In
view of that stated objective, recent research has concentrated on the issue of partial coordination, attempting
to address the question of whether partial coordination
is in fact superior to having each country make independent policy choices. If, as the discussion above in-

Reserve Bank of Atlanta
Digitized Federal
for FRASER

dicates, full coordination is superior to independent
policymaking, can it be assumed that the same is true
of partial coordination?
When a coalition of some but not all countries is
formed and policymaking is centralized, additional
considerations will surface. For example, if the EMU
eventually materialized and agreed to fiscal and monetary discipline that implied lower interest rates, it would
be possible for nonmembers or outsiders to "free-ride"
on the EMU member countries' efforts. An outsider
would not internalize the impact of, for instance, its
setting a loose monetary policy but would still be in a
position to enjoy the positive impact of a complete coordination effort by the m e m b e r s of the EMU felt
worldwide.
Stephen J. Turnovsky (1988) analyzes partial coordination using an economic model that emphasizes the
negative spillovers on the terms of trade. He assesses
the effects of a subset of countries in the world forming a union. In his analysis government spending is
valued by private economic agents, such as spending
for roads and infrastructure. Because domestic governments purchase import and export goods, they affect
the terms of world trade. With no barriers to trade, one
country's increase in government expenditure on export goods could, for example, raise the goods' relative price worldwide, thus affecting the purchasing
power of other countries. As before, without policy
coordination individual countries seem unable to internalize negative spillovers across countries. Turnovsky
showed that countries that coordinate policy will in
general have less government spending and, accordingly, less influence on the terms of trade. In this context, his model indicates that some coordination is
better than no coordination at all. This conclusion rests
on the rationale that as long as some fiscal discipline is
imposed, its positive impact will be felt worldwide.
However, the countries not bound by a coordination
agreement can enjoy both improved terms of trade and
larger government expenditures than those countries
committed to a common policy. In that sense, countries not involved in the coordination agreement would
be free-riders, raising the question of the frailty of partial coordination agreements. Given that free-riding is
a possibility, countries would have an incentive to
break away from the union. What are the implications
for the future of agreements like the EMU?
Marco Espinosa and Chong K. Yip (forthcoming),
using a dynamic general equilibrium approach, have
found that the number of countries involved in a partial coordination agreement relative to the number of
all countries is crucial in determining the sustainability
¡iconam ic Review1

of the agreement. Given groups of coordinating and
noncoordinating countries, each country balances
gains from coordination against gains from free-riding
on others' coordination. If the number of countries involved in a coordination scheme is too small or too
large, the gains to the coordinating members are such
that the possibility of free-riding on the fiscal discipline of others does not outweigh the advantages of
sticking to their coordinated policy. Espinosa and Yip
show that there is, however, a "right-sized" coordination coalition of relatively homogenous countries.
With the appropriate number of coordinating countries, the incentives to break away from the coalition
disappear. Therefore, this research suggests that agreements like the EMU can be both beneficial and lasting.

Conclusion
Economic theory suggests that in the presence of
open economies, free trade, and efficient capital markets, a country's policies will generate some externalities affecting the rest of the world community. When
countries choose policies independently of each other,

1. See Kahn (1987) for a review of some of these coordination
efforts.
2. See, for example, K u m a r and Whitt (1992) for a crosscountry analysis of exchange rate variability on international trade.
3. See Chriszt (1991) for a review of the ERM.
4. One should not confuse these concepts with the distinction
between full and partial coordination established below.
The terms full and partial coordination refer to the number
of countries relinquishing their policy autonomy—in other
words, the number of countries involved in a coordination
scheme.
5. President Mitterrand, for example, has been quoted as saying that countries that have already ratified the Maastricht
Treaty could continue to move forward.
6. Some authors have taken pains to clarify the different degrees of international interdependence and coordination and
the difference between coordination and cooperation. In
this article, for simplicity, there is no distinction between
coordination and cooperation. Several current studies—for
example, Feldstein (1988)—similarly make no distinction.
Cody (1989) and Humpage (1990), for instance, emphasize that the term international cooperation refers to the
sharing of macroeconomic data and economic goals across

10
Econom ic Review

the impact of their choices on the world community
cannot be fully internalized. Several of the coordination initiatives have implicitly acknowledged this fact.
The discussion in this article establishes that under
these circumstances coordination of macropolicies
seems better for the world than having each country
making independent policy choices.
The description of basic models of full coordination
illustrates the complexity of the nature of international
coordination and points to several issues that require
further studied consideration. For example, given the
desirability of a complete coordination effort, can it be
inferred that incomplete coordination is better than
none at all? Given the difficulties of implementing a
complete coordination agreement, should countries instead spend energy trying to implement incomplete
agreements? If countries agree to coordinate policy,
how does an economic union choose a common objective? What are the difficulties faced by countries that
precommit to coordinated policy adoptions and by the
body of coordinating countries in ensuring adherence
to agreed-upon policy? These are just a few of the areas of concern calling for attention from policymakers
and researchers alike.

countries. International coordination, on the other hand,
refers to the joint setting of macroeconomic policies. By
these definitions international cooperation includes the International Monetary Fund (IMF) and the OECD; some of
the joint actions of the G-5 and G-7 are examples of international coordination.
7. For an excellent survey on the Mundell-Fleming approach
to open economy macroeconomics, the reader is referred to
Frenkel and Razin (1987).
8. See Caves, Frankel, and Jones (1993, chap. 22-23) for a
textbook-level illustration.
9. For an updated survey see Cooper (1985).
10. See chapter 19 of Krugman and Obstfeld (1991) for a simple application of game theory to international coordination.
11. A common thread found in some of the literature on dynamic general equilibrium modeling is that the analysis of
fiscal and monetary policies is based on the public finance
approach originated by Ramsey (1928) and popularized by
Lucas (1986). Lucas suggests viewing fiscal and monetary
polices as trying to allocate distortions from taxes or subsidies in such a way as to maximize society's well-being over
time. To be specific, this approach says that governments
should choose taxes or subsidies so as to maximize the

M a y / J u n e 1993

well-being of all generations in a society. The choice of a
tax-subsidy scheme explicitly incorporates a government
sector that as a player in the international capital market can
influence interest rates worldwide. This methodology represents a holistic approach to choosing "optimal" fiscal and
monetary policies and as such is superior to a methodology
that views them as independent of each other and assigns
them different objectives throughout time.
12. For further analysis of the Maastricht Treaty see Kenen
(1992) and Fratianni, von Hagan, and Waller (1992).
13. For a intuitive description of the concept of time inconsistency, see Mankiw (1992, chap. 12).
14. Rogoff (1985) studies international cooperation in setting
monetary policies in a context in which such policies influence economic activity and the objectives of the private
sector (wage setters) and governments arc at odds with each
other. This conflict arises from the assumption that the society's target employment rate is higher than the wage setter's target employment rate. Several factors can give rise
to this situation—labor unions, for example. Labor unions
are usually willing to accept a lower level of employment in
order to bargain for higher wages for their members.
Once wage setters set their nominal wages, the central
bank can affect the rale of employment by inflating the
economy. An increase in the price level would reduce the
unemployment rate because a lower real wage raises the demand for labor. The monetary authorities, however, cannot
systematically raise the level of employment through infla-

tion. The private sector will eventually recognize this pattern and set nominal wage rates high enough to reflect their
inflation forecast. As this process goes on and the central
bank loses its credibility, the rate of inflation would increase consistently. This result does not imply that the central bank is irresponsible; it simply says that given the lack
of precommitmenl to adhere to a policy's rules, a central
bank's optimal choice each period may be to try to increase
employment, at the expense of higher inflation.
15. The requirements for economic convergence of the E M U
m e m b e r s include specific low-inflation targets, similar
long-term nominal interest rates, q u a s i - f i x e d e x c h a n g e
rates, and fiscal discipline as evidenced by a low maximum
ratio (3 percent) of a government's deficit to GDP.
16. A partial coordination scheme refers to a situation in which
at least one country with influence in world economic activity chooses policy independently. This section considers
only complete policy coordination, under which countries
that are members of a coalition adopt common macroeconomic policies.
17. It is not a coincidence that this discussion has focused on
the EMU and has only tangentially dealt with GATT. The
type of analyses that deal with coordination of macroeconomic policies emphasizes intertemporal trade whereas the
studies that deal with custom unions emphasize contemporaneous trade of different commodities. For an analysis of
free trading zones, sec Krugman (1991).

References
Bryant, Ralph C. " I n t e r n a t i o n a l C o o p e r a t i o n for N a t i o n a l
Macroeconomic Policies: Where Do We Stand?" Paper presented at the Conference Celebrating the Fiftieth Anniversary of Princeton Essays in International Finance, Princeton
University, Princeton, N.J., April 1993.
Caves, Richard E., Jeffrey A. Frankel, and Ronald W. Jones.
World Trade and Payments. 6th ed. New York: Harper
Collins College Publishers, 1993.
Chang, Roberto. "International Coordination of Fiscal Deficits."
of Monetary Economics 25 (June 1990): 34766.
Chriszt, Michael J. " F Y I — E u r o p e a n Monetary Union: How
Close Is It?" Federal Reserve Bank of Atlanta Economic Review 76 (September/October 1991): 21-27.
Cody, Brian J. "International Policy Cooperation: Building a
Sound Foundation." Federal Reserve Bank of Philadelphia
Business Review (March/April 1989): 3-12.
Cooper, Richard. "Economic Interdependence and Coordination of Economic Policies." In Handbook of Internationa!
Economics, vol. 2, edited by Ronald Jones and Peter Kenen,
1195-1234. Amsterdam: North Holland, 1985.
Espinosa, Marco, and Chong K. Yip. "On the Sustainability of
International Coordination." International Economic Review
(forthcoming, 1994).

Federal Reserve B a n k of Atlanta

Fleming, John M. "Domestic Financial Policies under Fixed
and under Floating Exchange Rates." IMF Staff Papers 9,
no. 3 (1962): 369-79.
Feldstcin, Martin, cd. Internationa!
Economic
Cooperation.
National Bureau of Economic Research Conference Report
Series. Chicago: University of Chicago Press, 1988.
Fratianni, Michele, Jurgen von Hagan, and Christopher Waller.
The Maastricht Way to EMU. Princeton, N.J.: Princeton
University, 1992.
Frcnkcl, Jacob, and Assaf Razin. "The Mundell-Fleming Model:
A Quarter-Century Later." IMF Staff Papers (1987): 567-620.
Hamada, Koichi. "A Strategic Analysis of Monetary Interdependence." Journal of Political Economy (1976): 677-700.
H u m p a g e , O w e n F. "A H i t c h h i k e r ' s Guide to International
Macroeconomic Policy Coordination." Federal Reserve Bank
of Cleveland Economic Review 26 (Quarter 1, 1990): 2-14.
Kahn, George A. "Dollar Depreciation and Inflation." Federal
Reserve Bank of Kansas City Economic Review 72 (November 1987): 32-49.
Kehoe, Patrick J. "Coordination of Fiscal Policies in a World
Economy." Journal of Monetary Economics (1987): 349-76.
. "Policy Cooperation among Benevolent Governments
May Be U n d e s i r a b l e . " Review of Economic Studies 56
(1989): 289-96.

EconoinicRevieu'

Kenen, Peter B. EMU after Maastricht.
Washington, D.C.:
Group of Thirty, 1992.
Krugman, Paul R. "Is Bilateralism Bad?" In International Trade
and Trade Policy, edited by E. Helpman and A. Razin, 9-23.
London: The MIT Press, 1991.
Krugman, Paul R., and Maurice Obstfeld. International
Economics: Theory and Policy. 2d ed. New Y o r k : H a r p e r
Collins College Publishers, 1991.
Kumar, Vikram, and Joseph A. Whitt. "Exchange Rate Variability and International Trade." Federal Reserve Bank of
Atlanta Economic Review 77 (May/June 1992): 17-32.
Lucas, Robert. "Principles of Fiscal Policy and Monetary Coordination." Journal of Monetary Economics 17 (1986): 117-34.
M a n k i w , N. Gregory. Macroeconomics.
N e w York: Worth
Publishers, 1992.
Miller, Preston J., and Neil Wallace. "International Coordination of Macroeconomic Policies: A Welfare Analysis." Federal Reserve Bank of Minneapolis Quarterly Review (Spring
1985): 14-21.

12
Econom ic Review

Mundell, Robert A. "Capital Mobility and Stabilization Policy
under Fixed and Flexible Exchange Rates." Canadian Journal of Economics
and Political Science 2 9 ( N o v e m b e r
1963): 475-85.
Ramsey, F.P. "A Mathematical Theory of Saving." Economic
Journal 38 (December 1928): 543-59.
Rogoff, Kenneth. "Can International Monetary Policy Cooperation Be Counterproductive?" Journal of International
Economics 18 (1985): 199-217.
Rompuy, Paul Van, Filip Abraham, and Dirk Heremans. "Economic Federalism and the EMU." European Economy, Special Edition No. 1, 1991, 109-33.
Turnovsky, Stephen J. "The Gains from Fiscal Cooperation in
the Two-Commodity Real Trade Model." Journal of International Economics 25 (August 1988): 111-27.
Wallace, Neil. " S o m e of the Choices for Monetary Policy."
Federal Reserve Bank of Minneapolis Quarterly
Review
(Winter 1984): 15-24.

M a y / J u n e 1993

nflation and the
Yield Curve

Peter A. Abken

he determinants of interest rates across the spectrum of bond maturities are of keen interest to the general public—borrowers and
lenders—as well as to economists. Particularly during 1992, the
spread between the shortest and longest maturity rates grew unusually wide, leaving many observers wondering what to infer
from this gap (see Chart 1). Many, including some policymakers, see the
current steep yield curve as an ominous sign of future inflation. The inflation forecasts implicit in interest rates carry strong credibility with most
market observers because interest rates in part represent bets backed by
wealth rather than casual forecasts with little at stake. The trouble is that interest rates are influenced by more than just market expectations of inflation; other factors cloud the inflation signal perceived in the term structure
or relationship between interest rates.

The author is a senior
economist in the financial
section of the Atlanta Fed's
research
department.
He thanks Will Roberds,
Ellis Tall man, and Steve Smith
for very helpful comments
and Chris Holder for
research
assistance.

Federal
Reserve Bank of Atlanta

This article discusses key past and current research studying the information on inflation contained in the nominal (face-value) term structure of interest rates.1 The literature consists of two distinct but complementary strands.
One analyzes the impact of inflation on interest rates; the other, the implicit
inflation forecast in interest rates. The current evidence suggests that the
yield curve does indeed give useful forecasts of inflation, especially at
longer-term horizons, but much still needs to be learned about the various
factors that influence nominal rates. The first section provides a background
for the discussion in subsequent sections of more recent research. The second section considers research that has examined the short end of the yield
curve, extending to bond maturities of one year. The final section looks
at the longer-term yield curve. One pragmatic reason for this short-term/longterm dichotomy is that Treasury bill yields, which run to maturities of up to

EconoinicRevieu'

one year, are more readily analyzed than note and
bond yields. Another reason is that, according to recent studies (discussed below), the term structure behaves very differently in these two maturity segments.

background
Numerous economists have viewed interest rates in
terms of a decomposition into real (inflation-adjusted)
and expected inflation components. In fact, the notion
that interest rates reflect inflation expectations goes
back centuries. Thomas M. Humphrey surveyed the
historical development of the conceptual division of
nominal rates into the real interest rate and expected
inflation. According to Humphrey, the first record of
such a distinction was made in the 1740s by William
Douglass, a "Scottish-born physician, pamphleteer,
controversialist, and student of American colonial currencies" (1983, 3). Douglass noted how lenders included an inflation premium in loans that were expected to
depreciate in value because of the free printing of unbacked currency. Others to offer the same hypothesis
before the twentieth century include Henry Thornton,
John Stuart Mill, Alfred Marshall, and J.B. Clark. The

fullest articulation of the distinction between nominal
and real rates was by Irving Fisher in his 1896 treatise,
Appreciation and Interest, and was elaborated further
in his 1930 classic, The Theory of Interest.
Fisher viewed the nominal rate of interest in terms
of a goods rate (or real rate) and a rate of appreciation
(or inflation): "The theoretical relation [now called the
Fisher relation] between interest and appreciation implies, then, that the rate of interest is always relative to
the standard in which it is expressed. The fact that interest expressed in money is high, say 15 percent,
might conceivably indicate merely that general prices
are expected to rise (i.e., money depreciate) at the rate
of 10 per cent, and that the rate of interest expressed in
terms of goods is not high, but only about 5 percent"
(1930, 41-42). Fisher expressed the future value of a
dollar of principal (one plus the nominal interest rate)
as the following product:
1 + it = (1 + 7tp( 1 + r), or it = r ; + rt + n%
where K'f is the expected rate of inflation over the
coming period (for example, one year) formed at time
t and rt is the real rate. 2 This is the equation that Fisher derived in his 1896 monograph. For small real rates
and expected inflation rates, the cross-product term is

Chart 1
Treasury Yield Spreads
(Weekly,

January

4, 1989, to April 14,

1993)

Basis Points

14
Econom ic Review

M a y / J u n e 1993

negligible. The logic of this equation is that if a loan
were paid in a basket of goods, the interest charged
would be r , also representing payment in goods, not
money. Conversion into money after one period would
be at the rate K' r which gives the rate of exchange
of money for goods. Inflation increases the dollar
amount paid, and deflation diminishes it. In other
words, an anticipated inflation is predicted to have no
impact on the real return of assets denominated in
fixed monetary value—for example, the real returns
of bonds. This view was in accord with the prevailing
classical monetary theory, which held that monetary
phenomena, like inflation, only affect the price level
and have no "real" effects. Although the Fisher relation is a straightforward concept, it has been very difficult to verify empirically because neither anticipated
inflation nor the real rate of interest is directly observable.
In writing The Theory of Interest, Fisher in part
sought an explanation for the so-called Gibson paradox. In a series of articles written in the 1920s, A.H.
Gibson documented a strong positive correlation between the yields on long-term British bonds (Consols)
and the level of a commodity price index over 130
years, which contradicted classical theory's prediction
that the nominal rate and price level are independent.
Fisher put the problem this way: "If perfect foresight
existed [that is, if the Fisher relation held], continuously rising prices would be associated not with a continuously rising rate of interest but with a continuing
high rate of interest, and falling prices would be associated not with a continuously falling rate of interest
but with a continuing low rate of interest, . . . assuming, in each case, that other influences than price
change remained the same [the real rate of interest is
stable or constant]" (1930, 411-12).
The Fisher relation implies that the nominal interest
rate and the change in the price level are positively
correlated. Fisher attributed the Gibson paradox,
which he confirmed in his own empirical studies of the
con-elation, to the sluggish adjustment of inflation expectations.
Thomas J. Sargent (1973) explained how Fisher's
econometric work supposedly resolved the Gibson
paradox by finding that expectations of inflation were
formed adaptively on the basis of long lags of past inflation experience—ten to thirty years' worth. Fisher's
proxy for unobscrvable expected inflation, a weighted
summation over a long series of past price changes, is
highly correlated with the price level (because the sum
of long series of price changes will be close in value to
the current level). Because of the implausibility of

Federal Reserve Bank of Atlanta

such long lag lengths, many later economists rejected
Fisher's view that sluggish expectations formation explained the Gibson paradox.
Sargent found that the assumed direction of causation running from inflation to nominal rates was erroneous in Fisher's work and many subsequent studies.
Fisher and other later researchers regressed interest
rates (the dependent variable) on current and lagged
inflation rates (the explanatory variables). Sargent
showed statistically that inflation and nominal interest
rates simultaneously cause each other through a feedback relationship. 3 He thus provides a link between
the two distinct strands of the interest rate/inflation literature identified earlier. His analysis of the time-series
properties of inflation and interest rate data led him to
conclude that "the interest rate contains information,
over and above that contained in lagged rates of inflation, that is useful in predicting the rate of inflation"
(1973, 447). Much subsequent research has been devoted to evaluating the forecast power of interest rates
for inflation.

inflation and the Shorter-Maturity
Term Structure
Eugene F. Fama is one of the founders of the efficient markets school of finance, which predicates its
analysis of markets on market participants' rationality
and on their efficient use of all relevant information in
determining market prices. In Fama (1975) he took the
view that, if the real rate of interest is approximately
constant and markets are efficient, the nominal return
on a Treasury bill would be correlated with the subsequently observed rate of inflation over the term of the
Treasury bill. That is, nominal rates would move over
time only because the expected inflation component of
those rates changed. Furthermore, if inflation is predictable to some extent, rationality of inflation forecasts implies that nominal interest rate changes would
move one-for-one with subsequently realized inflation
on average. To test this hypothesis, Fama ran ordinary
least squares (OLS) regressions of ex post (realized)
inflation on interest rates, which according to Fisher
contain the ex ante (expected) inflation rate. His examination of the very short end of the yield curve for the
period from 1953 to 1971 confirmed his hypotheses
that the market is efficient, the expected real rate of return was constant over the sample period, and variations in one- to six-month Treasury bill returns were
statistically significantly correlated with subsequently

EconoinicRevieu'

observed inflation. 4 These results stand in sharp contrast to the previous empirical results that imply market
inefficiency because of sluggish expectations formation.
Fama's article was influential and controversial. His
conclusion that the expected real rate of interest is constant (at least over the sample period he examined) was
criticized by a number of researchers (P.J. Hess and
J.L. Bicksler 1975; C.R. Nelson and G.W. Schwert
1977; D. Joines 1977; K. Garbade and P. Wachtel
1978). They all detected some variation in expected real returns using different methodologies. Nevertheless,
their results supported Fama's main point that interest
rate fluctuations are driven predominantly by variations
in expected inflation. Motivated by these critics' findings, Fama and Michael R. Gibbons (1982) hypothesized that the expected real rate evolves as a "random
walk"—that is, changes in the expected real rate from
one period to another are unpredictable and permanent.
They estimated an inflation-interest rate regression
similar to Fama (1975) but including a time-varying intercept term, representing the random walk of the real
rate, instead of a constant. They found that, during the
1953-77 period, expected inflation over horizons of
one month or one quarter also behaved like a random
walk. Most important, using a more flexible econometric approach they reaffirmed Fama's earlier finding that
fluctuations in expected inflation primarily account for
variations in short-term interest rates. (The ratio of the
variance of monthly expected real returns to the variance of the monthly expected inflation rate was .2, and
the ratio using quarterly data was .25.)
Fama and Gibbons estimated and analyzed a negative correlation between expected inflation and expected real interest rates, which had previously been noted
by a number of researchers. 5 A complete consideration
of why a negative correlation exists is beyond the
scope of this article, but for the sake of providing
some economic intuition one prominent theory is
sketched. The Mundell-Tobin theory (Robert Mundell
1963; James Tobin 1965) holds that an increase in expected inflation and associated higher nominal interest
rates raises the opportunity cost of holding money,
which earns no interest. People attempt to reduce
money holdings for interest-earning assets, namely
bonds, and doing so depresses their expected real return. In turn, the lower expected real rates stimulate
capital investment, lower the return to capital (the real
rate), and increase output. In fact, Fama and Gibbons
advance a different explanation than Mundell-Tobin
for the negative correlation. What is important for the
following discussion of the Fisher relation is simply

16
Econom ic Review

the existence of a correlation between expected inflation and real rates because of its statistical complications in drawing inferences from the term structure.
Frederic S. Mishkin (1981 b) developed a new econometric approach for decomposing nominal rates into
real rate and expected inflation components. Like
Fama, Mishkin took the view that the bond market is efficient and employed the relatively new rational expectations methodology. Operationally, this approach
simply provided a rationale f o r assuming that the
market's inflation forecast errors are entirely unsystematic and unforecastable. This maintained hypothesis
about market behavior allowed Mishkin to substitute
the measurable ex post real interest rate for the unobservable ex ante rate in doing his econometrics, a
well-established procedure in empirical rational expectations studies.
Symbolically, according to the Fisher relation, the
nominal interest rate is if = nc¡ + r (the sum of the ex
ante inflation rate and ex ante real interest rate). The
ex post real rate is measured as it - 7t = (n* + r) -71,=
r, - (71, - 71 p, where Kr is the ex post inflation rate and
the term (nr - n'¡) is the market's inflation forecast error over some time interval like a month or a year.
Mishkin assumed that the average inflation forecast
error over a long period of time is zero, that is, unbiased, and furthermore that it is uncorrelated with itself over time and contemporaneously uncorrelated
with other variables, like the real interest rate or the
rate of inflation. In other words, the ex post real rate
differs from the ex ante real rate only by an unforecastable "noise" term. These assumptions allow predicting the ex ante real rate on the basis of simple
linear regressions of the ex post real interest rate on
observable variables, such as past inflation rates, money growth, and unemployment rates, correlated with
the ex post real rate.
Mishkin's approach was evidently partly motivated
to counter the evidence that had accumulated against
the Fisher relation, and bond market rationality generally, by researchers relying on survey data that he felt
were seriously flawed (see Mishkin 1981b, especially
footnote 7). His method did indeed find strong support
for the Fisher relation. Though influential and widely
cited, Mishkin's procedure has its shortcomings and
its critics (see Kenneth J. Singleton 1981).
Mishkin (1990a) examined the shorter-maturity term
structure and found that, contrary to Fama and Gibbons, the term structure communicates almost nothing
about future inflation. Mishkin's regression analysis is
very similar to Fama and Gibbons's. Mishkin estimates what he terms an inflation-change forecasting

May/June 1993

Statistical rejection of the hypothesis that fimn = 0
indicates a significant correlation between the spread
and future inflation. The hypothesis that f3mn = 1 means
that the spread gives an unbiased forecast of future inflation (at least during the sample period). There are a
number of explanations for the rejection of unbiasedness. One is that the ex ante real interest rate spread
may also predict the change in the inflation rate, but
this variable does not appear in the equation. In fact, it
is implicitly a component of the disturbance term 7fJ",
which because of the real rate spread may be correlated with the nominal interest rate spread, biasing the
slope coefficient away from one (see Box 1).

equation, which is reproduced here to aid the following discussion:
1
Kmi-K" i = OL
mn+B"mn(/x 1 — iI''') + T)"'".
I

This equation expresses the ex post (realized) change in
the inflation rate, measured as the rate over a longer future period m minus the rate over a shorter future period
n, as a linear function of the difference or spread between two corresponding nominal interest rates, observed at time t. For example, the ex post difference in
the rate of inflation over a six-month horizon versus a
three-month horizon, both intervals starting at time t, is
regressed on the time t spread between a six-month
Treasury bill yield and a three-month Treasury bill yield.
The regression assesses the implicit forecasted change in
the inflation rate contained in the yield spread or "slope"
of the term structure. The details of this equation's
derivation from the Fisher relation are given in Box 1.

The regression can be recast in terms of the ex post real
interest rate, with ex post inflation omitted. Mishkin
shows the simple algebra that converts the inflation
change regression into an ex post real interest rate change
regression (ex post real rate regressed on interest rate
spread), in which the slope coefficient is 1 - fimn. Thus,

Boxi
The inflation-change regression derives f r o m the
Fisher relation. T h e expected rate of inflation over m f u ture periods is ne™ = i™- r™ (and nmt = i"t- r" over n per i o d s ) . E x p o s t i n f l a t i o n is e x p e c t e d i n f l a t i o n p l u s a
forecast error: tz"'= Ke™ + s™. W i t h the F i s h e r relation,
n em _ ¡m_ ¡.m^ ^
u n o b s e r v a b l e expected inflation is substituted o u t of t h e e q u a t i o n f o r ex post i n f l a t i o n . T h e
equation for ex post inflation b e c o m e s K"/ = /"' - r m t +
e"'. T h e final step in deriving M i s h k i n ' s equation is to
take the d i f f e r e n c e between equations for w-period and
«-period ex post inflation rates. Because both the inflation rate (see Robert B. Barsky 1987) and n o m i n a l and
real interest rates are close to being random walks, differencing is a s s u m e d to induce stationarity, which is necessary to s a t i s f y O L S regression r e q u i r e m e n t s f o r statistical inference. T h e resulting equation is

k"' - tu'; = (¿7 - /?) - (r? - r") + (s? - <).
The real rate is further d e c o m p o s e d into a d i f f e r e n c e in
the m e a n real rate over the s a m p l e period and deviation
f r o m that m e a n ,
rm = j m +

(rm _ p » ) = Jm

and similarly for the n-period horizon. T h e c o m p o n e n t s
of the spread in real interest rates,
rm

are rearranged to facilitate the analysis. T h e d i f f e r e n c e in
m e a n s over the two forecast horizons is s u b s u m e d in a
constant term, a mn = 7" -Tm, and the real rate deviations
are c o m b i n e d with the inflation forecast errors in a c o m posite disturbance term, r f | " = s"/ - s" -(«"'u"t). T h i s
last equation is then converted into the final f o r m of the
inflation-change regression,

K - K = a,nn + PrJi",' ~ ' P

w h e r e (3mn = 1 if t h e d i s t u r b a n c e term is u n c o r r e l a t e d
with the interest rate spread.
T h e d i s t u r b a n c e t e r m in the a b o v e r e g r e s s i o n , y]"'",
will of course consist only of inflation forecast errors if
the real rate of interest is constant. T h e real world turns
out to b e m o r e c o m p l i c a t e d . By O L S regression theory,
c o n s i s t e n t e s t i m a t i o n of f3mn with its " t r u e " v a l u e requires that the d i s t u r b a n c e term b e uncorrelated with the
interest rate spread. T h e rational e x p e c t a t i o n s a s s u m p tion e n s u r e s that the inflation forecast errors c o m p o n e n t
of the d i s t u r b a n c e is u n c o r r e l a t e d with the spread, but
the real interest rate deviations can b e and in fact are related to the spread, as d e m o n s t r a t e d in n u m e r o u s earlier
papers, including M i s h k i n ( 1 9 8 1 b ) and F a m a and Gibbons (1982). R e j e c t i n g unbiasedness ( p m n = 1) therefore
indicates that the slope of t h e real term structure fluctuates.

_ ,.n _ ( j m _Jn ) + („>» _ M ») ;

Federal
Reserve B a n k of Atlanta

EconoinicRevieu'

with two equations, forecast power of the spread
becomes tautologically divided between explaining
changes in real interest rates and changes in inflation.
For the period from February 1964 to December
1986, Mishkin found that the term structure for Treasury bills with less than six months to maturity has
slope coefficients that are insignificantly different from
zero. For inflation-change regressions using bills with
nine- and twelve-month maturities, some forecast power was evident. Charts 2-4 show scatter plots of the
variables that enter the inflation change regression, using data from July 1964 through September 1991 that
were sampled quarterly to reduce the density of plotted
points. 6 The values of (m, n), in months, are (3, 1), (6,
3), and (9, 6) for Charts 2, 3, and 4, respectively. The
interest rate spread appears on the horizontal axis, and
the ex post inflation change, on the vertical axis. In addition, monthly inflation/interest rate spread pairs are
shown using two symbols: squares represent observations before October 1979, when the Federal Reserve
changed its operating procedure to deemphasize targeting the Federal funds rate; diamonds represent postOctober 1979 observations. As discussed in Mishkin's
work, there is much evidence that nominal interest

rates became more volatile after the change in operating procedure.
If the real interest rate were constant and the market
had perfect foresight, all points would fall along a 45degree line running from the southwest quadrant to the
northeast quadrant in a plot with equal vertical and
horizontal scales. Forecast and actual outcome would
coincide exactly. Because the variability of actual inflation rates is so great compared with the variability
of the nominal interest rate spread, the horizontal and
vertical axes are drawn with different scales; consequently, the "45-degree" line is much less than 45 degrees in these plots. The inflation rate is notoriously
difficult to forecast by any method, and thus large expectation errors are not surprising. The pre-October
1979 observations tend to cluster more tightly because
of the lower volatility that characterized both yields
and inflation during this period. Because interest-rate
spreads at the short end of the yield curve tended to be
positive during the sample period, most observations
fall in the northeast and southeast quadrants.
The monthly inflation-interest rate spread pairs fall
evenly above and below the horizontal axis. The inflation change regression lines in Charts 2 and 3 are flatter

Chart 2
Two-Month Inflation Change versus Three-Month-One-Month Yield Spread,
Treasury Bills
Inflation Change

• nÀ
A
'

tip.! lîAAU k
A

k
S

Line

• July 1964-Oct. 1979
A Nov. 1979-Sept. 1991

•
-1

-1.5

0.5

1.0

1.5

|
2.0

2.5

Yield Spread

Econom ic Review

May/June 1993

Chart 3
Three-Month Inflation Change versus Six-Month-Three-Month Yield Spread,
Treasury Bills
Inflation Change
• July 1964-Oct. 1979

A Nov. 1979-June 1991

m
•

•

[Th Ä

45° Line,

Li ^

A
A [p

-1.5

-1.0

-0.5

A
0

0.5

1.0

1.5

Yield Spread

Chart 4
Three-Month Inflation Change versus Nine-Month-Six-Month Yield Spread,
Treasury Bills
Inflation Change

Federal Reserve Bank of Atlanta

Yield Spread

EconoinicRevieu'

than the "45-degree" lines. Especially for Treasury bill
maturities of less than nine months, as in Mishkin's
work, the small slope coefficients and cloud-like dispersion of observations above and below the regression
line indicate only a weak correlation between yield
change and inflation change.
The flip-side of the inflation change regressions is
the ex post real interest rate regressions. Mishkin found
that the real interest rate slope coefficient, 1 - /3imi, is
highly statistically significant and not significantly different from a value of one for three-month/one-month
interest rate spreads. However, spreads using longer
maturities up to twelve months proved to be almost always statistically insignificant. Under the maintained
hypothesis that expectations are rational, the implication of these results is that changes in nominal interest rates at the very short end of the term structure are
based on variations in the real rate of interest, not inflation expectations.
Fama (1984), N. Gregory Mankiw and Lawrence H.
Summers (1984), Gikas A. Hardouvelis (1988), and
Mishkin (1990a) have all analyzed the determinants of
the slope coefficient (3mii. Theoretically, the value of the
slope coefficient depends on the correlation between
the expected change in the inflation rate and the spread
between real interest rates for the corresponding time
horizon as well as on the ratio of the standard deviation
of the expected inflation rate change to the standard deviation of the spread in real interest rates. Based on a
fixed negative inflation-real rate correlation, the theoretical pattern for the slope coefficient is the following.
For maturities of less than six months, the standard deviation of the expected inflation change is dominated
by the standard deviation of the slope of the real rate.
The slope coefficient takes negative values in this range
but rises toward unity for twelve-month bills and can
greatly exceed unity (approaching 2, depending on the
correlation) as maturity lengthens. As the horizon
moves to long-term maturities, the slope gradually falls
back to unity, implying that the standard deviation of
the real term structure slope goes to zero and all variation in the term structure at long horizons stems from
changing inflation expectations.
In Mishkin (1990a), ex ante inflation rates and ex
ante real rates were estimated using Mishkin's (1981b)
technique, which relies on the rational expectations assumption that inflation forecast errors are uncorrected
with any variables at the time a forecast is made (that
is, they are unsystematic and unpredictable). Using
these estimates, the measured inflation-real rate correlation is - . 8 on average, ranging from - . 5 to - . 9 7 ,
depending on the sample period and maturity of un-

20
Econom ic Review

derlying bonds being analyzed. Mishkin's regression
results are therefore qualitatively consistent with his
theoretical analysis. The pattern of slope coefficients as
the forecast horizon lengthens roughly conforms to the
theoretical shape given a negative correlation.
While this finding is suggestive, it is open to question because neither expected inflation nor the real interest rate is directly observable. Mishkin's analysis
hinges crucially on the validity of the rational expectations assumption that inflation forecast errors are uncorrelatcd with expected inflation and ex ante real
interest rates. Nonzero correlation of inflation forecast
errors with expected inflation or ex ante real interest
rates would also bias the slope coefficient away from a
value of one. (See Singleton 1981 for other concerns
about Mishkin's estimates.) Further consideration of
the long-term interest rate analysis is given later.

Real Interest Rates and Risk Premia
What factors shift the slope of the real term structure? Mishkin does not address this point directly. More
theoretical structure is needed to gain insights into the
behavior of real interest rates, such as that in John C.
Cox, Jonathan E. Ingersoll, Jr., and Stephen A. Ross
(1985a) and Douglas T. Breeden (1986), which give the
most comprehensive development of the contemporary
theory of real interest rates. The real rate of interest is
fundamentally related to the productivity of capital,
which fluctuates over time because of unpredictable
shocks to technology and the production process (for example, technological innovations and oil price shocks).
At the same time, the real rate varies with changes in investors' attitude toward risk and willingness to save and
defer current consumption (which may be induced by
changes in productivity and wealth). 7
According to the expectations theory of the term
structure, longer-term real interest rates are averages
of current and future expected short-term real interest
rates (see Peter A. Abken 1990). Thus, real rates anywhere along the maturity spectrum are interrelated via
the expectations mechanism. However, the term structure of real interest rates is only indirectly observable
in the nominal yield curve; shocks to expected inflation may obscure movements in the real yield curve.
For example, a rise in expected inflation may cancel
the effect of a fall in the real rate on the nominal rate.
Mishkin claims that at the short end of the yield curve,
variation in expected inflation is small compared with
variation in real interest rates.

May/June 1993

The expectations theory is usually amended to allow for possibly time-varying term premia. Risk arises
from the uncertainty surrounding both the nominal
and real returns of Treasury securities. Because of unpredictable movements in real and nominal interest
rates, a bond's real return (price appreciation plus interest earnings) over its life is always uncertain, as is its
nominal return (unless the nominal bond is held until
maturity). Term premia comprise risk premia and inflation uncertainty components. These are discussed in
the next section. Subsequently, recent models of the
nominal term structure are considered that explicitly
treat the dynamics and interactions of real and nominal interest rates and risk premia.
Real and Nominal Risk Premia. Simon Benninga
and Aris Protopapadakis (1983), Breeden (1986), and
Martin D.D. Evans (1989) (among others) have considered risk and risk premia in the context of the Fisher relation. The upshot of their research is that under
conditions of uncertainty, when real and nominal rates
are explicitly modeled as evolving stochastically (randomly) through time, the Fisher relation fails to hold
theoretically. In addition to the real rate and expected
inflation, risk premia and the effect of inflation uncertainty are also components of the decomposition of the
nominal rate.
Theory predicts two sources of risk, nominal and
real, that require modifying the Fisher relation, which
was derived long before the technical tools for modeling uncertainty were developed. The real risk premium
of a nominal bond derives from the bond's usefulness
in hedging against adverse changes in consumption.
Bonds are stores of wealth; they transfer wealth from
one time period to another. What investors ultimately
value is the flow of goods and services that they buy
through current income and savings. Risk-averse investors (who are also consumers) seek to smooth the
flow of consumption over time. This is the basic conclusion from theories of optimal consumption and investment (Breeden 1986). Smooth consumption is
preferred to low consumption one period followed by
high consumption the next because the "disutility" of
deficient consumption outweighs the added utility
from surplus consumption, and bonds can help even
out the consumption flow. Bonds that offer nominal
returns that tend to be high when consumption is high
(or, more technically, marginal utility of consumption is low) and low returns when consumption is low
(marginal utility is high) must compensate investors
for being relatively poor hedges against the risk of
variations in the flow of consumption over time. If the
risk premium were insufficient, risk-averse investors

Federal
Reserve Bank of Atlanta

would avoid buying such bonds or would sell them
and consequently drive down their current price and
increase their expected return, other factors being the
same.
The standard Fisher relation juxtaposes nominal and
real (inflation-indexed) bonds of the same maturity. 8
Benninga and Protopapadakis show an alternative
(but equivalent) formulation of the Fisher relation that
highlights term structure relationships in the decomposition of nominal rates into real rates, expected inflation, and risk and uncertainty components. This approach equates longer-term bond prices with expected
prices for equivalent roll-over investments in shorterterm bonds for the same holding period. The real term
premium—the expected return on longer-term real bonds
over and above that on shorter-term real bonds for the
same holding period—reflects a longer-term real bond's
ability to hedge consumption risk. The hedging characteristics will differ across real bonds of varying maturity, and consequently each will require a different
level of risk premium. Intuitively, a real bond will include a positive risk premium if the sale of that bond
before maturity generates a capital loss at the same
time that consumption is low (and conversely capital
gains when consumption is high). 9 The real risk premium may vary over time, for example, as investor
wealth changes. Thus, in addition to fluctuations in
short-term real rates over time, the real risk premium's
own variation may contribute to shifts in the slope of
the real term structure.
The nominal term premium is analogous to the real
term premium. Nominal term premia may be informally viewed as the sum of two parts. One component
depends on the nominal bond's usefulness as a hedge
against shifts in future consumption, and thus the component is a real risk premium for a nominal bond. The
other component reflects the inflation-hedging characteristics of a nominal bond. If longer-term bonds, for
example, are more susceptible than shorter-term bonds
to having their realized nominal returns over a given
holding period wiped out to some extent by unexpected inflation (because of capital losses on the bonds),
then they will be less desirable to own and will require
extra return as compensation. Even if investors do not
require a risk premium to bear consumption risk, the
nominal term premium would still exist if the inflationhedging characteristics of bonds differed across bond
maturities. 10
Benninga and Protopapadakis point out that only if
nominal and real bond prices are highly correlated will
nominal and real term premia have the same sign. In the
extreme, if expected inflation were constant, real and

EconoinicRevieu'

nominal term premia would be identical, but in general
they will be different and each may vary over time.
Variability in real and nominal term premia contributes
to shifts in the slope of the nominal term structure.11
Uncertainty about the future price level drives another wedge between nominal and real interest rates
and is also another component of nominal term premia. This type of uncertainty is called the Jensen's inequality effect, identified by Stanley Fischer (1975),
among others. 12 As the volatility of the price level increases (or equivalently, as the inflation rate becomes
more volatile), the nominal interest rate falls relative
to the real rate, other things being equal. The magnitude of the Jensen inequality effect depends only on
the degree of variability of the future price level (technically, on its variance and possibly higher moments
as well), not on the risk aversion of the consumers/
investors in the economy. The effect is a mathematical
phenomenon and does represent compensation for
bearing risk.
Box 2 gives some simple examples of Jensen's inequality as applied to the expected rate of inflation as
well as the economic intuition underlying the examples. Dilip K. Shome, Stephen D. Smith, and John M.
Pinkerton (1988) found empirically that the Jensen inequality effect is not statistically significant in the
Fisher relation, whereas their measure of the nominal
risk premium is highly significant. 13
Model-Based Studies of the Term Structure. Two
recent studies, by George G. Pennacchi (1991) and
Tong-sheng Sun (1992), use explicit models of the
nominal and real term structures to measure the real
rate of interest and expected rate of inflation. These
studies employ a class of models that restrict interest
rate movements along the term structure in such a way
that there are no arbitrage opportunities possible (see
Abken 1990). In other words, the models imply that
the ex ante returns to investing in any bond or combination of bonds is the same for a given holding period,
except for the possible addition of a risk premium

(which augments the return but does not represent an
arbitrage opportunity). Although these model-based
studies are not about inflation forecasting per se, they
do shed additional light on the results of regressionbased studies.
Unlike previous research based on continuous-time
term structure models, Pennacchi (1991) and Sun (1992)
worked out tractable models that allow for the nonneutrality of money (specifically, an inverse correlation of
real interest rates and expected inflation). The most
important aspect of these models for the current discussion is their basic setup. Both assume that all term
structure movements are driven by a small set of socalled state variables, which are the sources of unpredictability in interest rates. Both researchers use data
sets on Treasury bill yields of various maturities to infer the relationship between the real rate and expected
inflation. This approach contrasts with regressionbased methods that examine the relationship at each
maturity in isolation. The model-based approach captures the dynamics of the interest variables and is not
intended to assess the forecasting performance of interest rates. Pennacchi's and Sun's work may be useful
in gauging how expected inflation and real rates move
through time.
Pennacchi estimated the parameters for two theoretical processes (that is, equations that describe the dynamics of variables) for the expected rate of inflation
and for the real interest rate. A determinant of the real
interest rate is the rate of return on physical capital,
which represents the real wealth in the economy. The
rate of return on capital is affected by the expected rate
of inflation; consequently, the real rate of interest is also partly determined by expected inflation. The reverse
is also assumed to be true. The expected rate of inflation is partly driven by the contemporaneous return on
capital and by shocks to that return. Thus, like Sargent's earlier work on the Fisher relation, expected inflation and real interest rates are mutually determined
in this model.

Box 2
T h e J e n s e n ' s inequality e f f e c t is illustrated u s i n g a
simple e x a m p l e based on t h e mathematical derivation of
t h e F i s h e r t h e o r e m in B e n n i n g a a n d P r o t o p a p a d a k i s
( 1 9 8 3 ) . E v e n if investors d o n o t require c o m p e n s a t i o n
for bearing risk (that is, if they are risk neutral), the real
interest rate will not b e equal to the nominal rate less the

22
Econom ic Review

rate of e x p e c t e d inflation w h e n inflation is stochastic.
T h e basic point of the e x a m p l e below is to show that the
expected inflation rate is not equal to the reciprocal of
the expected c h a n g e in the purchasing p o w e r of m o n e y
and that the g a p between these quantities widens as uncertainty increases.

M a y / J u n e 1993

T h e inflation rate measures the c h a n g e in the nominal
(dollar) value of a given basket of g o o d s and services.
T h e related concept of c h a n g e in purchasing p o w e r assesses the change in the real value (in terms of the basket
of g o o d s and services) that a fixed n o m i n a l value will
buy, that is, t h e n u m b e r of units of the basket. Symbolically, for current price level pQ (measured by an index)
and future price level p, E(p/p0) represents o n e plus the
expected inflation rate, and Eipjp) denotes one plus the
e x p e c t e d c h a n g e in p u r c h a s i n g p o w e r . A l t h o u g h t h e
Fisher relation is typically stated using expected inflation, purchasing power is m o r e closely connected to the
pricing of nominal assets. 1
S u p p o s e the initial price level p{) equals 1 and the f u ture price level p takes values 1.05 with probability .5 or
1.15 with probability .5. (That is, f u t u r e inflation rates
are 5 percent or 15 percent. T h i s will be called the low
inflation uncertainty case.) T h e (mathematically) expected price level is

T h e J e n s e n ' s inequality e f f e c t is s h o w n in Chart A,
which graphs the reciprocal of the price level against the
price level itself. T h e low inflation uncertainty case, denoted by E, has an expected change in purchasing p o w e r
represented by the midpoint of the chord within the hyperbola of the reciprocal price level. T h i s point lies a b o v e
1 /E(p) on the graph. T h e high inflation uncertainty case,
denoted by / / , has a chord that lies above the previous one
and thus it has a higher midpoint, representing a greater
increase in purchasing power. Note that the expected inflation rate of 10 percent is the same in both cases.
The economic intuition behind this e x a m p l e is that a
m o r e variable inflation rate results in an increase in the
expected purchasing p o w e r of m o n e y , m a k i n g n o m i n a l
bonds relatively m o r e attractive. Other things being
equal, greater variability of inflation drives nominal bond
prices u p and their interest rates down relative to the real
interest rate. Conceptually, the adjustment occurs so that
in equilibrium investors are indifferent concerning holding nominal or real bonds.

E(p) = (.5 • 1.05) + ( . 5 - 1.15) = 1.10,
and the expected inflation rate is [E(p)/p0 - 1J) • 100 =
10%. O n the other hand, calculating the expected c h a n g e
in purchasing p o w e r entails taking the expected value of
the reciprocals of the f u t u r e price levels:

Chart A
The Effect of Inflation Uncertainty on the
Purchasing Power of Money

E{Mp) = .5(1/1.05) + .5(1/1.15) = .91097,
which is not the same as \JE(p) = .90909. In other words,
money's purchasing power is expected to drop by \E(\/p) 1] • 100 = - 8 . 9 % (after rounding) as compared with the case of
constant 10 percent inflation in which the loss is - 9 . 1 percent.
T h e Jensen's inequality term is therefore approximately
E(l/p)

- [1 /E(p)] = .001882 or .19%.

In this e x a m p l e , an uncertain inflation rate slightly reduces the loss of expected purchasing power of m o n e y .
N o w suppose t h e price level b e c o m e s m o r e variable.
T h e future price may g o to 1.2 with probability .5 or stay
unchanged at 1 with probability .5. (That is, future inflation is n o w 0 percent or 2 0 percent.) Repeating the earlier
calculation, the expected inflation rate is still 10 percent,
leaving 1 /E(p) unchanged. H o w e v e r ,

Future Price Level

Note: The convexity of the graph of 1/P is greatly exaggerated
for clarity. The vertical axis shows \E (1/P) - 1] X 100.

E(l/p)

= .5(1/1.2) + (.5 • 1) = . 9 1 6 6 6 , o r - 8 . 3 % ,

making

Note
E(\/p)

- \ l/E(p)] = .007575, or .76%.

T h u s , a m o r e variable or uncertain inflation rate increases the expected purchasing p o w e r of m o n e y ,
E(pjp).

Federal Reserve B a n k of Atlanta

1. Fama (1975,1976) formulates the Fisher relation in terms
of purchasing power, n o t expected inflation. S e e also
Benninga and Protopapadakis (1983) and Shome, Smith,
and Pinkerton (1988).

EconoinicRevieu'

To estimate the behavior of expected inflation, Pennacchi relied on the National Bureau of EconomicResearch (NBER) and the American Statistical Association (ASA) surveys of inflation predications by professional forecasters. Michael P. Keane and David E.
Runkle (1990) found that these forecasts are unbiased
and rational, consistent with the rational expectations
assumptions underlying Pennacchi's model.
Pennacchi is not testing any particular theory about
why a mutual dependence might exist between expected inflation and the real rate of interest. His purpose is
to estimate econometrically the correlation between the
real interest rate and expected inflation and how both
variables react when one or the other is "shocked"
away from its equilibrium level. The structure imposed
on estimation by a term structure model enables him
to make detailed predictions of the mutual dynamics
of these two variables.
From 1968 to 1988, the "instantaneous" correlation
between the innovation (unpredicted component) of
expected inflation and that of the real interest rate was
a statistically significant -.376. Consistent with Mishkin's results using regression-based methods for the
shorter-maturity yield curve, Pennacchi found that the
real interest rate is much more volatile than the expected
rate of inflation. Unlike Mishkin, Pennacchi has modeled real and nominal risk premia and thus separately
identifies movements in the real interest rate, rather
than movements in a composite of real interest rate
and risk premia. The model also reveals that the real
rate of interest is much slower to return to its equilibrium level than expected inflation is to return to its corresponding equilibrium rate.
Sun (1992) achieves modeling objectives similar to
Pennacchi's in statistically characterizing the dynamics of the real interest rate and expected inflation. Sun
extends the celebrated Cox-lngersoll-Ross (CIR) term
structure model (see Cox, Ingersoll, and Ross 1985b
and Abken 1990) to allow for money nonneutrality.
One important feature of the CIR model is that the
volatility of nominal and real interest rate processes is
proportional to the level of the shortest-term interest
rate. This characteristic accords with empirical evidence in a number of studies (see K.C. Chan and others 1992). That is, nominal and real interest rates tend
to be more volatile when rates are high than when
rates are low. Pennacchi's model assumes that nominal and real rates have constant volatility, an assumption that potentially biases his estimates because his
model may be misspecified. (Of course, the CIR model may also misspecify the volatility.) Sun posits a
process generating expected inflation, but he does not

24
Econom ic Review

rely on survey data to estimate the parameters of that
process. Instead, he uses data on the Consumer Price
Index and statistically models the joint conditional distribution of his (unobservable) state variables and the
rate of inflation. The prices of two Treasury bills of
different maturity serve to substitute (instrument) for
the unobservable variables. The bond prices are complicated nonlinear functions of the state variables and
price index; this is the point at which the CIR model,
modified for money nonneutrality, comes into play.
Like Pennacchi, Sun rejects the hypothesis of money neutrality. The correlation coefficient between the
unobservable state variable summarizing uncertainty
in the real economy and the inflation rate is significantly positive (not zero). Sun does not estimate a correlation between real interest rates and expected inflation, but the implication of his result is the same,
namely, that the strict Fisher relation does not hold and
expected real rates are not independent of expected inflation. He estimates a positive nominal risk premium,
which means that the nominal interest rate is greater
than the real rate plus expected inflation. Furthermore,
because of the CIR specification, the magnitude of the
risk premium varies over time in proportion to the level of the state variable, whereas Pennacchi's risk premia are constant over time. Nevertheless, both models
produce qualitatively similar graphs of the real interest
rate and expected inflation.
The work of Pennacchi and Sun reveals that real
short-term interest rates undergo significant variation.
The real interest rate was more volatile than expected
inflation during the 1970s and 1980s. 14 These findings help to clarify Mishkin's regression results, which
indicate that at the short end of the yield curve variations in nominal yields reflect (and forecast) changes
in the slope of the real term structure, not changes in
expected inflation. Regression-based methods alone
cannot sort out the factors moving the real term structure.

inflation and the Longer-Maturity
Term Structure
Mishkin (1990b) extended his investigation of the
term structure to maturities of longer than one year.
He reestimated his inflation-change regressions for
zero-coupon bonds, derived f r o m actually traded
coupon-bearing bonds. These bonds were constructed
with maturities of one, two, three, four, and five years;
all yield spreads were formed relative to the one-year

May/June 1993

bond. The corresponding changes in the CPI were
similarly computed.
Mishkin generally found that the estimated p mn are
greater than unity, sometimes near 2.0, and are all statistically significant. As noted earlier in the discussion
of shorter-maturity term structure, the slope coefficient
from the inflation-change regression depends on the
correlation between the expected change in the inflation rate and the spread in the real interest rate as well
as on the relative volatility of the expected change in
the inflation rate and the real interest rate spread. During his 1953-87 sample period, the correlation ranged
from -.7 to -.95, depending on the underlying discountbond maturity. According to Mishkin's interpretation,
this correlation, combined with high volatility of expected inflation, pushes the slope coefficient toward
values near 2, particularly in regressions using longerterm discount bonds.
The regression /? 2 's (the ratio of explained to total
variation in the change in inflation) was 20 percent at
the two-year maturity (m = 2) and more than 40 percent for four- and five-year maturities. These results
compare with R2,s for the shorter-maturity inflation
change regressions of less than 10 percent and with
those of less than 5 percent for maturities less than
nine m o n t h s . T h u s , l o n g e r - m a t u r i t y interest rate
spreads have statistically significant power to forecast
changes in future inflation.
Charts 5-8 illustrate Mishkin's results for the longermaturity term structure by means of scatter plots using
quarterly sampled monthly observations on yield
changes and inflation changes for the 1964-91 sample
period. These are parallel to Charts 2-4, although all
changes are now expressed in terms of the one-year
bond yield and one-year rate of inflation. The values of
(m, n), in years, are (2,1), (3, 1), (4,1), (5, 1) for Charts 5-8,
respectively. As before, squares represent pre-October
1979 observations, and diamonds stand for post-October
1979 observations. The scatter plots show a distinct tendency for points to cluster in the northeast and southwest quadrants, which indicates spread predictions of
subsequent inflation in the right direction.
Fama (1990) also studied one- to five-year maturity
discount bonds constructed from coupon bonds, and his
work is very similar to Mishkin's (1990b) in methodology and results. Fama focused on the forecast power of
the spread between five- and one-year discount bond
yields because other intermediate spreads are almost
perfectly correlated with this spread. The spread serves
as a single explanatory variable in several related regressions. One equation is an inflation-change regression identical to Mishkin's. Fama also regresses the ex

Reserve Bank of Atlanta
Digitized Federal
for FRASER

post change in the real interest rate, defined as the
change in the one-year (spot) rate minus the change in
the CPI, over fixed periods from one to five years long.
A third equation regresses the change in the one-year
rate on the spread. As Fama shows, by definition the ex
post change in the real rate plus the change in inflation
equals the change in the spot rate. Thus, the regression
coefficients—that is, constant term and slope—in the
estimated real rate and inflation equations sum across
regressions to their estimated values in the spot rate
equation. (Mishkin makes the same point in his research.) Finally, Fama also regressed the term premium, measured by the ex post holding period return on
discount bonds of two to five years in maturity minus
the one-year rate, on the spread. In assessing the regression results, Fama observed that "the yield spread
is the jack-of-all-trades. It responds to information
about term premiums and future spot rates, inflation
rates, and real returns" (73). He emphasizes that variation in expected term premiums, which appear to be
correlated with the business cycle, obscure the forecasts of the yield spread for the other variables. The regression evidence indicates that this variation grows
stronger as maturity lengthens.
A consequence of the spread's containing information about a number of variables is that it forecasts
each imperfectly if the variables are not themselves
perfectly correlated with one another. Particularly for
one-year bonds, the spread predicts changes in the real
rate that are of opposite sign and almost equal magnitude to changes in expected inflation. Again for the
one-year bond, the R2 is 23 percent for the inflation
equation but only about half that amount for the real
interest rate equation. As the horizon lengthens to a
maximum of five years, the spread continues to have
significant forecast power for inflation but virtually
none for the real rate. Another way to view Fama's regression results is to examine the spot rate equation.
The offsetting coefficients for the inflation and real
rate equations necessarily imply that the spread has no
value in predicting changes in the spot rate. Only at
longer horizons does the spread have utility in forecasting the change in the spot rate, precisely because
changes in that rate are predominantly determined by
changes in expected inflation.
A recent study, by Werner F.M. De Bondt and Mary
M. Bange (1992), has challenged the view that inflation
forecast errors are rational in the sense that Mishkin
and Fama assume. As discussed above, Mishkin's and
Fama's interpretation of the slope coefficient in the
inflation-change regressions hinges on inflation forecast errors not exhibiting systematic errors. There is no

EconoinicRevieu'

Chart 5
One-Year Inflation Change versus Two-Year-One-Year Yield Spread,
Treasury Bonds
Inflation Change

Yield Spread

Chart 6
Two-Year Inflation Change versus Three-Year-One-Year Yield Spread,
Treasury Bonds
Inflation Change
5
4
3

• July 1%4-Oct. 1979
A Nov. 1979-Dec. 1988

45° Line

2
1
0
- 1

- 2

-3
-4

Yield Spread

Econom ic Review

May/June 1993

Chart 7
Three-Year Inflation Change versus Four-Year-One-Year Yield Spread,
Treasury Bonds

Inflation Change

a July 1964-Oct. 1979
• Nov. 1979-Dec. 1987

1 :

45° Line

•

A
/
-6

-4

A
-2

Yield Spread

Chart 8
Four-Year Inflation Change versus Five-Year-One-Year Yield Spread,
Treasury Bonds
Inflation Change

0 July 1964-Oct. 1979
A Nov. 1979-Dec. 1986
450 Line

/ /

/
-7

-6

-5

Federal Reserve Bank of Atlanta

-4

/
-3

A
-2

D•

-1

Yield Spread

EconoinicRevieu'

way to know whether systematic inflation forecast errors
influence the regression results without additional information on expectcd inflation. (Fama and Mishkin
simply assume that the forecast errors arc unsystematic.) De Bondt and Bange try to address this issue directly
by using survey information on inflation expectations. 15
De Bondt and Bange use the Livingston forecasts of
the CPI, which are given semiannually by a large group
of economists. The mean forecast at each survey date is
taken as the market expectation. They conclude from
their tests of these forecasts that "expectations are insufficiently adaptive: if the economists paid more attention to recent inflation, and interpreted the prevailing
rate as less of a surprise, they would not make the same
error repeatedly" (485). Contrary to the rational expectations assumption, the survey forecast errors are systematically biased.
Using another set of regression tests, De Bondt and
Bange determine that the inflation forecasts implicit in
interest rates are also biased, essentially in the same
way Fisher had noticed more than sixty years ago:
nominal interest rates rise too slowly as inflation rates
accelerate and decline too sluggishly as inflation rates
abate. The Livingston survey data and nominal interest
rate spreads predicting inflation over the same horizon
were found to be highly con-elated, implying that the
two are biased in the same way.
De Bondt and Bange's strongest results concern
the ability of past inflation forecast errors to predict
ex post term premia. These term premia were computed as the excess returns on discount bonds with
maturities of one, two, three, four, five, and ten years
over the six-month Treasury bill return. If investors
are risk averse, term premia—consisting of nominal
and real risk premia as well as Jensen's inequality
components—should be predictable by variables that
measure risk and volatility. However, as De Bondt
and Bange point out, no researchers have convincingly identified ex ante observable economic variables
that predict variations in term premia. The only predictor has been interest rate spreads (as in Fama
1990). Dc Bondt and Bange found that past survey inflation forecast errors (known at the current date) predict ex post term premia (computed for holding periods
starting at the current date). Overpredictions of inflation arc correlated with higher ex post term premia
and underpredictions with lower ex post term premia.
This finding runs counter to efficient markets theory
because past inflation forecast errors are not plausibly
regarded as measures of risk. Such easily obtained information as past inflation forecast errors should have
no value in predicting future interest rates in an effi-

28
Economic Review

cient, "rational" market. De Bondt and Bange conclude that "the inertia in expectations may be rational
if we consider the costs and benefits of more accurate
forecasts and/or possible regime changes [for example,
changes in Federal Reserve operating procedures]
(with the implication that rational expectations resemble adaptive expectations)" (495). Their research casts
doubt on Mishkin's analysis and interpretation of
yield spread forecasts of changes in inflation and real
interest rates. Nevertheless, they agree with Mishkin
and Fama that, over longer horizons, interest rate
spreads are reliable, though biased, predictors of future inflation.
One line of criticism of De Bondt and Bange's work
is that all of their results derive from the Livingston
survey data, the quality of which is very much open to
question. Mishkin (1981a, 1981b) was an early critic
of the biases in the Livingston data that were then imputed to market expectations generally. More recently,
Keane and Runkle (1990) argued that not all economists polled in compiling the Livingston forecasts are
professional forecasters and therefore some do not
have an economic incentive to be informed and accurate
in their projections. Keane and Runkle also stressed
that averaging across forecasts to get a "consensus"
forecast when individual respondents have different
private information can lead to severe biases detected in
rationality tests. In light of Keane and Runkle's study,
Pennacchi (1991) relied on the NBER-ASA forecasts,
which represent p r o f e s s i o n a l f o r e c a s t e r s ' predictions. 16 One should be concerned that on the one hand
Keane and R u n k l e d e t e r m i n e d that the quarterly
NBER-ASA forecasts of the Gross National Product
deflator pass properly constructed rationality tests, yet
on the other the monthly Livingston CPI survey forecasts appear to be so biased in De Bondt and Bange's
tests. Further investigation of this issue is needed.

Conclusion
Nominal interest rates are determined by a variety of
factors that limit their accuracy in predicting the future
course of any single factor. The expected rate of inflation has long been considered an important influence
on interest rates. However, other factors that obscure
the information in the yield curve about future inflation may vary substantially over time. Fluctuations in
real interest rates, real and nominal risk premia, and
inflation uncertainty components potentially cloud the
term structure's information on inflation.

May/June 1993

that the yield curve does give a relatively reliable forecast of inflation, particularly at longer horizons. A fruitful area for future research is assessing the comparative
value of implicit yield-curve forecasts versus alternative methods for making longer-term inflation forecasts.

Despile the complexity of interest rates, most people
who want to gauge inflation expectations turn to the
yield curve. There are clearly many unresolved issues
concerning the behavior and dynamics of nominal interest rates and their relation to expected inflation.
Nevertheless, recent empirical research demonstrates

Notes
1. All discussion is in terms of the Treasury yield curve because firm- or agency-specific credit risk complicates the
analysis for non-Treasury bonds.
2. See Humphrey (1983, 7) for the arbitrage argument Fisher
used in deriving this formula.
3. Sargent tests for mutual feedback statistically and argues that
the apparent mutual dependence works indirectly through
other omitted variables that influence inflation and interest
rates. He developed a simple macroeconomic model that
gives rise to the Gibson paradox in artificial data simulated
by the model. Unlike Fisher's explanation, long lags in
forming expectations of inflation play no role in Sargent's
model. He concludes that simple regressions of interest
rates on current and lagged inflation rates do not necessarily
provide any information on the process of expectations formation.
4. Because Fama computed Treasury bill returns assuming
that bills were held to maturity, the return is the same as the
yield—that is, a bill matures with fixed face value, which
means that its rate of return is known at the time of purchase. The future return would be uncertain if the bill were
sold before maturity.
5. See Mishkin (1981b. 164) for references as well as F a m a ' s
critics cited above.
6. The inflation rate is computed as the annualized difference
in logarithms of the seasonally unadjusted Consumer Price
Index. The inflation change is then the difference in inflation rates over an /?7-month horizon versus an //-month horizon, where m > //. The interest rate data are from the Center
for Research in Securities Prices (CRSP) and are end-ofmonth annualized, continuously compounded yields. Endof-month yields are aligned with the following month's CPI
in estimating the regressions. The first month's observations each quarter were included in the scatter plots, and the
regressions were run on these sampled data points.
7. Cox, Ingersoll, and Ross (1985a, 372) make the point that
the real interest rate can be different from the expected rate
of return of direct investments in physical capital (or equities) because of the differences in the hedging characteristics of investment in real bonds or physical capital. The real
rate of interest can in principle be greater than the rate of return on physical capital if equities hedge against future
shocks to consumption. In effect, investments in real or purchasing power bonds in this ease would require a risk prem i u m — e x t r a c o m p e n s a t i o n — t o induce investors to buy

Federal Reserve Bank of Atlanta

these bonds. Further discussion of the real risk premium is
given later in this section.
8. This discussion of real (inflation-indexed) bonds should be
viewed as a thought experiment because real bonds do not
trade in the United States.
9. Consider the following two-period (three-date) example using discount bond prices. The forward price of a real bond
to be issued one period from now and maturing one period
later is by definition equal to the two-period bond price divided by the one-period bond price. In a risk-neutral world,
this forward price is the expected price of the one-period
bond to be issued one period from now. In a world with
risk-averse investors, however, the forward price equals the
expected one-period price plus a risk premium. Technically, this risk premium is proportional to the covariance of the
investors' discounted marginal rate of substitution (ratio of
future to current marginal utilities of consumption) with
next p e r i o d ' s one-period bond price. In plain English, a
two-period bond will include a positive risk premium if the
sale of that bond after the first period generates a capital
loss at the same time that consumption is low (and conversely capital gains when consumption is high). On the
other hand, the maturing one-period bond pays off its lace
amount of the commodity-services basket (see Box 2). In
other words, the covariance of return with consumption for
the longer-term bond makes it a poorer consumption hedge
than the shorter-term bond.
10. This inflation-hedge component would be present even if
investors were risk neutral. Technically, the nominal risk
premium is proportional to the covariance of the product of
the marginal rate of substitution and change in purchasing
power with the price of the future one-period nominal bond
(in the context of the two-period example in note 9). Risk
neutrality implies a constant MRS, making the covariance
simply between purchasing power and the nominal bond
price, which is not a risk premium per se.
11. Empirical researchers often lump any variation in nominal
rates arising from nominal and real risk premia (and uncertainty components) into their measure of the real interest
rate, as is the case in Mishkin's application of the Fisher relation in Box 1. See also Mishkin (1981b, 167) and De
Bondt and Bange (1992, 490).
12. See especially Benninga and Protopapadakis (1983, 859)
for a derivation of Fisher's theorem under conditions of uncertainty.

EconoinicRevieu'

13. In particular, their test of the Fisher relation could not discriminate between measures of expected price change in
terms of inflation or changes in purchasing power. Both
were estimated with coefficients insignificantly different
from one, the theoretical value as implied by the Fisher relation.
In a slightly different context, Campbell (1986) found
that a Jensen inequality effect that distinguishes several versions of the expectations hypothesis of the term structure is
of second-order importance and can be ignored in empirical
work.
14. Sun does not discuss the derived time series for expected
inflation and real interest rate in any detail, except to note
that expected inflation is much "smoother" than actual inflation. Both series are plotted in his Figure 1. Pennacchi's

are shown in his Figure 7. Sun observes that the nominal
risk premium is positive but does not analyze any other
characteristics of the derived premium, which would have
been an interesting exercise.
15. Pennacchi also used survey information, although his purpose was different from Dc Bondt and Bange's. Based on
the work of Keane and Runkle that showed how the quarterly NBER-ASA survey data satisfy the rational expectations assumptions, he used the survey data as an input in
estimating a nominal term structure model.
16. Pennacchi addressed the aggregation bias issue identified
by Keane and Runkle by using the median instead of the
mean forecast error, a choice that puts less weight on extreme individual forecasts.

References
Abken, Peter A. "Innovations in Modeling the Term Structure
of Interest Rates." Federal Reserve Bank of Atlanta Economic Review 4 (July/August 1990): 2-27.
Barsky, Robert B. "The Fisher Hypothesis and the Forecastability and Persistence of Inflation." Journal of Monetary Economics 19(1987): 3-24.
Benninga, Simon, and Aris Protopapadakis. "Real and Nominal
Interest Rates under Uncertainty: The Fisher Theorem and the
Term Structure." Journal of Political Economy 91 (1983):
856-67.
Breeden, Douglas T. " C o n s u m p t i o n , P r o d u c t i o n , Inflation
and Interest Rates." Journal of Financial Economics
16
(1986): 3-39.
Campbell, John Y. "A Defense of Traditional Hypotheses about
the Term Structure of Interest Rates." Journal of Finance 41
(1986): 183-93.
Chan, K.C., G. Andrew Karolyi, Francis A. Longstaff, and Anthony B. Sanders. "An Empirical Comparison of Alternative
Models of the Short-Term Interest Rate." Journal of Finance 41 (1992): 1209-27.
Cox, John C., Jonathan E. Ingersoll, Jr., and Stephen A. Ross.
"An Intertemporal General Equilibrium Model of Asset
Prices." Econometrica 53 (1985a): 363-84.
• "A Theory of the Term Structure of Interest Rates."
Econometrica 53 (1985b): 385-407.
De Bondt, Werner F.M., and Mary M. Bange. "Inflation Forecast Errors and Time Variation in Term Premia "Journal of
Financial and Quantitative Analysis 27 (December (1992):
479-96.
Evans, Martin D.D. " W h a t Can the Term Structure Tell Us
about Expected Inflation? A Theoretical Analysis." New
York University, Graduate School of Business Administration, Salomon Brothers Center for the Study of Financial Institutions, Working Paper 510, March 1989.
Fama, Eugene F. "Short Term Interest Rates as Predictors of
Inflation." American Economic Review 65 (1975): 269-82.
• "Inflation Uncertainty and Expected Returns on Treasury Bills." Journal of Political Economy 84 (1976): 427-48.

30
Econom ic Review

• "The Information in the Term Structure." Journal of Financial Economics 13 (1984): 509-28.
• "Term Structure Forecasts of Interest Rates, Inflation,
and Real R e t u r n s . " Journal of Monetary Economics
25
(1990): 59-76.
Fama, Eugene F., and Michael R. Gibbons. "Inflation, Real Returns and Capital Investment." Journal of Monetary
Economics 9 (1982): 297-323.
Fischer, Stanley. "The Demand for Index Bonds." Journal of
Political Economy 6 (June 1975): 509-34.
Fisher, Irving. "Appreciation and Interest." Publications of the
American Economic Association 11 (1896): 1-100.
• The Theory of Interest. 1930. Reprint. New York: Augustus M. Kelly, 1970.
Garbade, K., and P. Wachtel. "Time Variation in the Relationship between Inflation and Interest Rates." Journal of Monetary Economics 4 (1978): 755-65.
Hardouvelis, Gikas A. " T h e Predictive P o w e r of the T e r m
Structure during Recent Monetary Regimes." Journal of Finance 43 (June 1988): 339-56.
Hess, P.J., and J.L. Bicksler. "Capital Asset Prices versus Time
Series Models as Predictors of Inflation: The Expected Real
Rate of Interest and Market Efficiency." Journal of Financial Economics 2 (1975): 341 -60.
Humphrey, Thomas M. "The Early History of the Real/Nominal Interest Rate Relationship." Federal Reserve Bank of
Richmond Economic Review 69 (May/June 1983): 2-10.
Joines, D. "Short-Term Interest Rates as Predictors of Inflation:
Comment." American Economic Review 67 (1977): 476-77.
Keane, Michael P., and David E. Runkle. "Testing the Rationality of Price Forecasts: New Evidence from Panel Data."
American Economic Review 80 (September 1990): 714-35.
Mankiw, N. Gregory, and Lawrence H. Summers. "Do LongTerm Interest Rates Oveireact to Short-Term Interest Rates?"
Brookings Papers on Economic Activity 1 (Spring 1984):
223-42.
Mishkin, Frederic S. "Are Market Forecasts Rational?" American Economic Review (June 1981a): 295-306.

M a y / J u n e 1993

- "The Real Interest Rate: An Empirical Investigation."
Carnegie-Rochester
Conference Series on Public Policy 15
(1981b): 151-200.
. "What Does the Term Structure Tell Us about Future
Inflation?" Journal of Monetary Economics 25 (1990a): 7795.
- "The Information in the Longer Maturity Term Structure about Future Inflation." Quarterly Journal of Economics 105 (August 1990b): 815-28.
Mundell, Robert. "Inflation and Real Interest." Jownal of Political Economy 11 (1963): 280-83.
Nelson, C.R., and G.W. Schwert. "Short-Term Interest Rates as
Predictors of Inflation: On Testing the Hypothesis That the
Real Rate of Interest Is Constant." American Economic Review 67 (\977): 478-86.

Sargent, T h o m a s J. "Interest Rates and Prices in the Long
Run." Journal of Money, Credit and Banking 5 (February
1973): 385-449.
Shome, Dilip K., Stephen D. Smith, and John M. Pinkerton.
" T h e Purchasing Power of Money and Nominal Interest
Rates: A Re-Examination." Journal of Finance 43 (1988):
1113-25.
Singleton, Kenneth J. "Extracting Measures of Ex Ante Real
Interest Rates from Ex Post R a t e s . "
Carnegie-Rochester
Conference Series on Public Policy 15 (1981): 201-12.
Sun, Tong-sheng. "Real and Nominal Interest Rates: A DiscreteTime Model and Its Continuous-Time Limit." Review of Financial Studies 5 (1992): 581-611.
Tobin, James. "Money and Economic Growth." Econometrica
33 (1965): 671-84.

Pennacchi, George G. "Identifying the Dynamics of Real Interest Rates and Inflation: Evidence Using Survey Data." Review of Financial Studies 4 (1991): 53-86.

Federal Reserve B a n k of Atlanta

EconoinicRevieu'

FYI

Consumer Prices:
Examining Housing
Rental Components

R. Mark Rogers, Steven W . Henderson, and Daniel H. Ginsburg

r
R. Mark Rogers is forecast coordinator on the macropolicy
team of the Atlanta Fed's research department. Steven W.
Henderson is a supervisory
economist and head of the
housing team in the Office of
Prices and Living
Conditions
(OPLC), Bureau of Labor
Statistics. Daniel H. Ginsburg
is a supervisory
economist
and chief of the Section of
Services, also with OPLC.

32
Econom ic Review

he Consumer Price Index (CPI) is one of the most closely followed economic indicators of conditions in the U.S. economy. It
is one of a number of broad price indexes that attempt to measure
inflation—that is, the rate of change in prices. For many reasons,
a sound measure of price changes, especially at the consumer level, is important. CPI data are used by monetary and fiscal policymakers, by
participants in financial markets, by economic forecasters, by people writing
contracts with cost escalation clauses (cost-of-living adjustments, or COLAs),
and by average consumers who worry about whether pay raises are keeping
up with prices for goods and services purchased. In addition, the federal
government uses the CPI to adjust Social Security payments and income tax
brackets and standard deductions for personal income taxes.

For all these users of CPI data, it is important that the data accurately measure what is intended. However, a number of economy watchers have questioned the accuracy of the CPI and several of its components. 1 One important
set of components that has been called into question on several occasions
measures housing rental costs.2 Most recently this set of measures has raised
questions when during the recession of 1990-91 and subsequent recovery—
while housing prices were depressed, apartment vacancy rates remained high,
and hotel vacancies were rising—rental CPI components were stronger than
expected. Because these rental components make up more than one-fouith
of the CPI, including figures for residential rent (tenant-occupied housing),

M a y / J u n e 1993

owners' equivalent rent (owner-occupied housing),
and lodging while out of town, the concern over rental
components is legitimate. 3 Possible flaws in such a
considerable portion of the C P I ' s c o m p o n e n t s by
weight would certainly call into question the CPI's
overall accuracy as a measure of inflation.
This article attempts to address the issue of how
well the CPI rental components reflected actual conditions from 1990 to 1992 and examines the sources of
apparent divergence. The discussion first explains
BLS methodologies for the housing rental series, including changes in methodologies within the time
frame studied that affected movement in the data. The
various CPI rental components are then compared
with corroborating independent data series to assess
the validity of the CPI components. Two alternative
hypotheses are considered that might help explain the
divergence from expectations for the residential rent,
owners' equivalent rent, and lodging-while-out-oftown CPI subcomponents. The first is that the BLS
methodologies are flawed or no longer relevant. The
second explanation investigated is the possibility that
popular concepts of what is measured differ from what
the CPI methodology actually covers.
The CPI is a measure of the cost of goods or services currently being consumed. It therefore seeks to
measure the cost of housing services or, in other
words, the cost of the use of shelter. Investment considerations reflected in the cost of obtaining a house
are not part of the consumption cost concept that the
CPI attempts to measure.

relative importance is essentially its original base-year
weight multiplied by its price growth relative to other
components. It is these share figures that reveal the
significance of housing in the CPI. 6 Table 1 shows the
sizable shares that both the housing expenditure group
and the rental components have in the CPI.

T h e Significance of Housing
Why are the rental components—owner-occupied,
tenant-occupied, and lodging—critical for understanding trends in the CPI? As was explained earlier, in simple terms housing rental components matter because
they form a large portion of the CPI. The broader housing component makes up a little more than 40 percent
of the overall CPI. The major share of shelter costs is
for rent-related components, the two largest series of
which are owners' equivalent rent and residential rent;
both are measures of cost for the use of shelter. Owners' equivalent rent (OER), discussed in greater detail
below and in Robert F. Gillingham (1980), is essentially the amount a homeowner would pay to rent or
would earn from renting his or her home in a competitive market. Residential rent (RR) measures the cost of
tenant-occupied housing in a broad residential market.
Pure rental components make up 20 percent of the CPI,
and lodging components—hotels, motels, and school
dormitories—add another 2 percent. It will be helpful
in examining recent criticisms of these CPI series to
understand the definitions and methodologies for calculating residential rent, owners' equivalent rent, and
lodging-while-out-of-town data.

What Is the CPI?
Precisely defined, the CPI is a measure of the average change in prices paid by urban consumers for a
fixed market basket of goods and services. 4 Currently, the CPI includes approximately 360 categories of
items, with the weight of each component series determined by actual consumer expenditure patterns as
measured by the Bureau of Labor Statistics' Consumer
Expenditure Survey (CES). The market basket currently used as a benchmark reflects the spending habits of
urban consumers from 1982 to 1984, with categories
generally assigned weights according to the average
consumer expenditure in each category relative to total
expenditures for items covered in the CPI."1
It is significant that over time each component's relative importance in the index changes as component
prices do not move together uniformly. A component's

Reserve Bank of Atlanta
DigitizedFederal
for FRASER

Rent of Shelter
Although the two largest subcomponents of the
housing category, owners' equivalent rent and residential rent, each measure costs of what generally are considered different types of housing, they are more closely
related than it might seem initially. OER is designed to
measure the cost of renting the same type and quality
housing as that occupied by homeowners, excluding
utilities but including maintenance. Such housing is
almost entirely single-family homes but also includes
town homes. Residential rent measures the cost of
rental housing and includes rents for many types of
multifamily and single-family units. The unifying factor
is that both measure the cost or use of a shelter service.

EconoinicRevieu'

While some variety and range in the types of housing are represented in costs for owners' equivalent
rent, the residential rent series covers a far broader
spectrum of housing, and within this category rental
housing having owner-occupied characteristics makes
up only a portion of the tenant-occupied housing units.
This overlap provides the key link between the estimation of residential rent and OER.
Data for both rent series come from the BLS housing survey, started in its current form in 1983 (with
revisions in 1987). 7 The housing unit sample is stratified. That is, the sample is chosen to represent the various types of rental and owner-occupied housing in use
at the survey date. It consists of approximately 40,000
rental units and 20,000 owner units. Each month only
one-sixth of both the rental units and owner units are
used to derive the CPI components. Each of the six
panels is designed to be representative of the full sample. Either in person or by telephone, sample individual
rental units are surveyed every six months and owneroccupied units are contacted once every two years.
However, as discussed further below, characteristics of
owner-occupied housing units—the panels—are used
every six months.

Table 1
Expenditure Groups and Relative Importance
December 1992
Expenditure Group

Relative Importance

CPI, Total
Housing
Shelter
Renters' costs
Residential rent
Other renters' costs
Lodging while out of town
Lodging while at school
Tenants' insurance
Homeowners'costs
Owners'equivalent rent
Household insurance
Maintenance and repairs

100.000
41.404
27.880
7.993
5.801
2.192
1.938
0.220
0.033
19.683
19.303
0.380
0.204

Fuel and other utilities

7.280

Household furnishings and operations

6.243

Note: Official relative importance figures are published by the Bureau of
Labor Statistics only for December of each year.

34
Econom ic Review

The BLS tracks rents for individual units as separate time series. BLS field agents gather from tenants,
managers, or owners of rental units information on the
current month's rent and the previous month's, and also on services provided. From owner-occupied units,
they obtain an estimated or implicit rent, the amount
for which the owner says the unit would rent on the
market. Field agents also collect information on the
characteristics of the housing units, including structure
type, number of rooms, and the age of the unit. The
sample is stratified, based on census information about
neighborhood and average rent level (classified as high,
medium, or low). Since a respondent's estimate of the
implicit rent of owner-occupied units may not appear
to be reasonable in comparison with similar housing,
field agents are allowed to enter their own estimates, and
the field agent's estimate is used.
A weighted-average process has been used to calculate the residential rent index since 1978. On the basis
of housing survey data, changes are estimated (after adjustments for changes in quality) by calculating onemonth-ago and six-month-ago percent changes for
residential rent units surveyed in the latest month. The
rental index numbers are moved by a weighted average
of the one-month and six-month percent changes, with
the one-month change getting a weight of 65 percent.
Percentage changes are applied using stratified components. Rents are quality-adjusted for each month's aging
of the units surveyed. The only changes in the process
have been the selection of new samples, new sampleselection criteria, improvements in noninterview imputations, and updates in automation procedures for
adjustments for quality change. Prior to 1978, the residential rent index was calculated by comparing the current month's rents with earlier rents over different time
periods. Quality adjustments have always been made.
For owners' equivalent rent, the answers to the survey for implicit rent (including any adjustments by
field agents) are important only for the initial estimate
for each unit. After that is established, subsequent values of implicit rent for a given unit are derived by using changes in rent that occur in a specific subsample
of the residential rent units used for the residential rent
index. Essentially, the BLS computes changes in OER
by using rent changes for housing units in the residential rent survey that have characteristics similar to
owner units. A computerized selection process matche s — b y location, structure type, and other general
characteristics—individual owner-occupied units with
similar units in the residential rent sample. The BLS
first looks for perfect matches and then, as necessary,
relaxes constraints one at a time until a satisfactory set

M a y / J u n e 1993

of renters' rental units is found for all owners. After
the initial interview, owners' responses on housing
characteristics become the important information because it is the match of housing characteristics to corresponding rental units that identifies which rental
units will effect a change in OER.
Once the matching process is completed, OER is estimated in the same technical manner as residential rent.
Even though owner units are contacted only every two
years to check for quality and tenure changes, they are
used in the matching process and index calculation
steps every six months. In essence, a subset of the rental
sample that best matches owner-occupied housing is
used to estimate the owners' equivalent rent index.
There was a subtle change in this OER methodology in 1987. In the period from 1983 to 1986, when the
BLS began using rental units in neighborhoods that
were owner-occupied to estimate changes in the OER
component, the rental units were not always of owneroccupied quality or characteristics; some units were
apartments and other types of multifamily units, and
the percentage they accounted for is unknown. In 1987
the BLS began sorting units specifically and confirming owner-occupied characteristics.
Definitional Differences in Rent. Measured rent
covers different things for the OER and RR components of the CPI. Rent figures collected in the housing
survey are on a contract basis for residential rent, and
therefore might include utilities or other additional
costs, but are on a pure rent basis for owners' equivalent rent. For residential rent, the contract covering specific housing units defines the changes in collected
rents and in services provided. These collected rents include any labor involved as part of payment and all services and facilities provided on a contract basis, such
as furniture or utilities. The pure rent basis for owners'
equivalent rent focuses solely on the cost of renting
shelter, excluding other items covered elsewhere in the
CPI like utilities, insurance, and furniture. Owners'
equivalent rent subtracts payment for the furniture and
utilities if they are included in the rent estimates.8
Both rental components are adjusted for changes in
quality. If there has been a major structural change in
an owner unit, an appropriate adjustment is made in the
level of rent. Adjustments are also made for changes in
services and facilities provided by the landlord—for
example, eliminating the inclusion of utilities in the
rent or adding a room to an apartment. Beginning in
1988 the BLS also began to adjust rent for aging,
which is viewed as reducing the quality of housing
units. Table 2 summarizes the historical approaches to
measuring home ownership costs in the CPI.

Reserve Bank of Atlanta
DigitizedFederal
for FRASER

Residential rent and owners' equivalent rent are
closely related in terms of how they are estimated.
Each is payment for a service, and both are dependent
upon the same survey data for their estimates. A later
section will review how well each of the series corroborates related data series.

Z o d g i n g While Out of T o w n
The lodging-while-out-of-town component of the
CPI is a very different rent measure from the OER and
RR. For the purposes of this article it is, however, important to examine it because it is a rent component—
measuring the cost of use of shelter—and its movement over the 1990-91 period diverged markedly from
general expectations of what was appropriate for thenexisting economic conditions.
The lodging-while-out-of-town housing subcomponent has an interesting history; recent methodological
improvements have had a sizable impact on its patterns.
This component essentially measures changes in lodging costs at hotels and motels and was first introduced
into the CPI in January 1964. Movement in the index is
based on rate changes for hotel and motel rooms rented
for pleasure or family activities and excludes business,
institutional, and convention use. Several specific characteristics are priced: the number of occupants, room
location and number, type and number of beds, and
amenities available (for example, telephone, television,
and transportation). Taxes charged to the consumer are
included as part of the total fees. These can include
sales taxes, hotel/motel taxes, special facility and services taxes, and any taxes levied by special authorities
for specific purposes, such as constructing and operating convention centers and tourist facilities.
The choice of geographic area covered by the series
has been an important part of its methodology and has
been a critical factor behind apparent changes in behavior. The recent acceleration in the component's rate
of increase can be accounted for largely by the relaxation in geographic constraints for selection of hotels and motels in the sample. Originally and through
1986, quotes were limited to hotels and motels in the
CPI's geographically defined pricing areas, the urban
areas covered by the CPI. Before 1978, outlets (the establishments being priced) were selected from national
travel directories for lodging and included national
chain hotels/motels and individual, local facilities. 9
Starting in 1978, unemployment insurance files, which

EconoinicRevieu'

Table 2
Historical Approaches to Measuring Homeownership
In the Consumer Price Index
Home
Purchase
Used?

Other Homeowners'
Costs in Index?

Index Used for
Homeowners' Costs

Time Period

Index Title

1. Pre-1953

CPI-W

Yes

Rent index

Cost and weights of home purchase, mortgage principal,
capital improvements considered as investments and
excluded. Weights for home maintenance, interest,
taxes, and insurance moved by rent.

2. 1953-63

CPI-W

Yes

Separate items priced
for homeownership

Home purchase added using weight of homes purchased
during expenditure survey. Also added indexes for
mortgage interest, incidental expenses, ground rent,
taxes, repairs, and improvements.

3. 1964-77

CPI-W

Yes

Same as 2

New expenditure weights used.

4. 1978-82

CPI-U and
CPI-W

Yes

Same as 2

New expenditure weights used, mortgage interest and
FHA sample problems.

5. 1967-82

CPI-U X1

Rent index

CPI-U rerun excluding home purchase, mortgage
interest, property taxes, property insurance,
maintenance and repairs indexes. Rent index
reweighted and substituted. X-1 started in 1980,
now extends back through 1967.

6. 1983-86

CPI-U (and
CPI-W for
1985-86)

No*

Owners' Equivalent
Rent

Individual rental units reweighted by homeowner/
renter ratio. Home purchase price, mortgage
property taxes excluded.

7. 1983-84

CPI-W

Yes

Same as 2

Conversion to O E R for the CPI-W occurred two
years after CPI-U change.

8. 1987-present

CPI-U and
CPI-W

No*

Owners' Equivalent
Rent

Homeowners' implicit rents moved by change in
matched pure rents. New weights and sample.

* Except for homeowner purchases of insurance, maintenance and repairs, and appliances,

Details

which are included and reweighted equal to renters.

contain the names, locations, and number of employees for hotels and motels across the country, were used
as the universe for selecting sample outlets to be
priced, with the probability of being selected proportional to an establishment's number of employees.
While the intent was to capture rates for travel and
pleasure for consumers in the covered urban areas,
lodging outlets were still restricted to those within the
CPI urban areas and therefore did not necessarily reflect consumer rates in locations where consumers
tended to vacation. In 1983 the method for selecting
hotel and motel outlets began to change and when fully
implemented would affect index movement. The new
method changed from the CPI independent sourcing of
hotel and motel outlets for pricing to that of the pointof-purchase survey (POPS)—the primary procedure for
selecting sample stores and other retail establishments
for the CPI program. 10 This survey is conducted by the
Bureau of the Census for BLS.
To implement the new methodology for this lodging component, a hotel/motel category was added to
the point-of-purchase survey. Many of the outlets
identified by POPS were not located in the officially
defined priced primary sampling units (PSUs) or urban areas, however, and their inclusion necessitated
greater data collection costs. Nonetheless, because
such areas are where the POPS households actually
had lodging expenses, BLS shifted to POPS in order
to access a more representative and more accurate
sample for the lodging-while-out-of-town component.
With the 1987 revision of the overall CPI, the BLS
gradually began using the POPS data for lodging while
out of town by increasing yearly the number of pricing
areas using only POPS data. Thus, no longer were hotels and motels limited to within twenty-five miles of
the urban areas priced for the CPI, and outlets became
eligible for the lodging sample regardless of their location in the United States. (The cost of lodging in foreign destinations by U.S. citizens is not covered in the
figures gathered.) Pricing is done by telephone if the
facility is too far away for field agents to visit.
Using POPS as an outlet source means that hotels/motels in the sample selected now fall primarily in
tourist areas rather than in large cities, the focus being
on the cost of lodging at the consumer destination.
With more of the sample in beach areas, ski resorts,
m a j o r a t t r a c t i o n s , and along interstate h i g h w a y s
throughout the country's key tourist areas, the sample
better represents actual consumer travel patterns.
Collection of Hotel/Motel Prices. The lodgingwhile-out-of-town component is priced monthly in the
CPI. The rate sought is the lowest room rate available

Federal
Reserve Bank of Atlanta

to such individuals, based on current occupancy levels
on the day of collection. Thus, when occupancy is
high, the "rack" or published rate may be obtained;
when business is slow, below-rack rates are collected.
Substantial seasonal price fluctuations often occur because many areas are frequented by vacationers primarily during summer and school vacation times.
Until May 1992 the sample included 152 outlets priced
for lodging while out of town. These were spread over
numerous locations reflecting the selected sample for
each of the eighty-five pricing areas." Necessarily, in
any given area the sample size was somewhat limited.
During the period from January 1987 to May 1992 the
index therefore reflected price movement of a rather
small sample, subject to volatile seasonal price movements, with prices reflecting the actual occupancy rates
of each priced outlet. While monthly price changes
(not seasonally adjusted) would be significantly affected by the volatile seasonal price movement, the yearover-year increase would largely reflect the underlying
price movement in the sample outlets. As of June 1992,
a new, substantially expanded sample, containing over
1,000 outlets, was put in place for all areas and will remain as the sample of priced outlets for five years. Outlets in this sample are also concentrated in vacation
areas. Nearly 60 percent of the sample outlets are located in ten states. 12

Re cent Trends and Explanatory Factors
Given the behavior of the rental components in the
1990-92 period, there is some reason for concern about
whether or not CPI housing components accurately represent changes in consumer prices, especially considering the large shares that housing components, and even
the rent subcomponents alone, have in the CPI. One
important question is whether these rent components
have diverged significantly from the prices they are intended to measure, and if so, why? Divergence may
occur for several reasons. For example, there are times
(such as during recession) when uncertainty about income gains dramatically reduces h o m e purchases
while maintaining or even boosting rental demand. In
addition to income concerns, expectations concerning
home prices can make home prices and rent diverge: if
home prices are expected to rise sharply in the future,
renters may buy now, thereby weakening rents.
Another question to consider is that of whether the
divergences are consistent with other economic data.
In other words, do various economic factors validate

EconoinicRevieu'

the methodology that produces these CPI series? In
addressing this issue the relative behavior of the overall CPI and the residential rent and owners' equivalent
rent components needs to be examined.
Residential Rent. Over the 1970s, increases in residential rent steadily accelerated from a 3 percent pace
early in the decade to a greater than 9 percent rate by
1980, following much the same path as the overall
CPI. Similarly, residential rent inflation rates declined
sharply over the 1980-82 period before undertaking a
roller-coaster downtrend throughout the rest of the
1980s and into 1992. During the 1990-91 period, both
the overall CPI and residential rent component decelerated. However, the growth rate in the overall CPI
slowed much more sharply than the residential rent
component, as shown in Chart 1. Through mid-1992,
while the CPI inflation rate leveled off, the residential
rent rate continued to decelerate.
At first glance, this pattern is somewhat puzzling
because other economic indicators suggested that rents
should have been their weakest in the 1990-91 time
frame. One might argue that rent should have been affected by the downtrend in inflation expectations as
much as the overall CPI was. Furthermore, purchased
homes and rental shelters may be viewed as close substitutes, so that home prices and rent were expected to

move together. However, existing home prices were
weak from 1990 into 1991. New single-family home
prices were especially low and remained so over a
longer period, from 1989 to 1992. During the 1990-91
period, the multifamily sector was still overbuilt, with
multifamily vacancy rates remaining high. Given the
recession and the depressed real estate sector, the CPI
residential rent component and owners' equivalent rent
series appeared to be out of line with housing market
conditions.
From mid-1990 through mid-1991, the overall CPI
inflation rate fell from over 6 percent to under 4 percent
(and decelerated even further into 1992). For that period, residential rent's inflation pace slowed from 4 1/2
percent to 3 percent. While the rate of inflation was
lower, the magnitude of deceleration was not as great
as for the overall CPI.
There are a number of explanations for this divergence in CPI component trends that do not focus on
rental pricing. For example, oil price weakness was a
factor in the overall CPl's greater disinflation. Nevertheless, the divergence of housing costs from general
index behavior warrants explanation.
Because owner-occupied housing and rental housing are substitutes, simple economic theory suggests
that their prices should move together. An important

Chart 1
Total CPI, Residential Rent, and Owners' Equivalent Rent
Year-Ago
Percent Change

38
Econom ic Review

May/June 1993

factor in home and rental prices, however, is that a
home purchase represents more than a purchase of services. Rent and housing prices are indeed affected by
some common elements, but there are also a number
of factors that affect each independently.

home, thereby reducing demand for rental housing and
weakening rents. On the other hand, if housing prices
were expected to decline, renters would have incentive
to defer a home purchase, and such a decision would
help support the level of rents.

Rent is generally considered payment for housing
services (shelter) provided. In contrast, when purchased by an owner-occupant, a home represents both
a financial investment and the buying of shelter services. Home prices are determined by a mix of supply
and demand for both investment opportunities and for
the housing services. However, home prices in the
short run are dominated by relatively fixed supply and
large swings in demand that are interest rate related.
Basically, home prices reflect a small percentage of the
housing stock that is priced infrequently, and rent reflects a larger pool of housing units being priced often.

Finally, another important reason that rent changes
occur less frequently and with a built-in lag compared
with changes in prices for houses on the market is that
rents primarily change when occupancy changes or longterm leases expire. These differing functions of rental
and owner-occupied housing suggest that while rent
and housing prices should be related through their purchase of housing services, it is reasonable that they diverge at times. The delay in the rental market's adjustment to other forces is understandable given the
restraints on charging new rents.

Because rent and housing prices are not determined
by exactly the same factors, there is really no reason to
expect that these prices would always move together in
lockstep. The major factor tying the two together is the
housing services portion of the value of a house, separate from its investment value. For a number of reasons, this factor does not keep rent changes completely
in tandem with changes in housing prices. For one
thing, the consideration of housing as an investment
cannot be completely separated. The most important
factor behind the differences, though, is that demand is
very different for each, yet near-term supply for both is
relatively fixed. As mentioned above, in times of economic uncertainty and high interest rates, buyers tend
to withdraw from the home market while the market
for rental housing remains more stable. It is also significant that there are constraints on consumers in terms
of freely moving back and forth between renting and
purchasing homes—the cost considerations of moving,
for example. An additional factor is that not all renters
can qualify to purchase a house or have adequate savings for down payments.
Rational shifts in households' demand for rental
housing and purchased housing, caused largely by expected changes in housing prices, can partly'explain
near-term divergence in rents and housing prices if
rental housing is recognized as having a purchase option. Just as homeowners have a prepayment option
and can profit when interest rates decline, a renter can
exercise the option to break a lease and buy a home
when expected, discounted future home prices significantly exceed their current market value. In other
words, if housing prices were expected to rise more
rapidly than other asset prices, then renters would be
more motivated to get out of their lease and purchase a

Federal Reserve Bank of Atlanta

What about the data's indications in terms of how
closely CPI rental components are related to various
other economic factors? Many factors affect residential
rent, home prices being only one; vacancy rates as a
measure of supply, wages as a measure of demand, and
the small percentage of turnover in rental markets also
cause average rents along with the CPI rent series to
change slowly. The focus here is on whether the movements in residential rent have been consistent with the
economic factors. A simple statistical model is used to
see whether these variables explain rent movement and
whether they confirm or refute the allegation that this
CPI component is flawed. Charts 2 and 3 show the behavior of home prices and vacancy rates relative to
changes in residential rent.
The model uses monthly year-ago percent changes
for residential rent, the dependent variable, as well as
for median sales prices for existing family homes, one
of the explanatory variables. The multifamily vacancy
rate, left unchanged in its simple percentage form, is
estimated over the January 1970 through December
1989 period (subsequent months are the ex post forecast period) using the Almon distributed lag technique.
This technique allows the explanatory variables to affect residential rent CPI gradually rather than all at
once. With the Almon technique, an independent variable can be entered into the model as a variable spread
over several periods in the past. Each period adds up to
form a cumulative impact on the dependent variable. 13
The regression results shown in Table 3 indicate that
existing home prices and multifamily vacancy rates do
a fairly good job of explaining changes in the growth
rate of residential rent CPI. The adjusted R2 is 0.77, and
the fitted values (the ex post forecast) track this CPI
component reasonably well. The coefficients of the independent variables are as expected, with the housing

EconoinicRevieu'

price variables having a positive sign for the sum of the
lag coefficients while that same sum for the vacancy
rate is negative. Essentially, this result means that as
housing prices accelerate, residential rent inflation goes
up, and as vacancy rates rise, rent inflation declines.
What does this model suggest about the expectations for the rate of residential rent increase over the
1990-92 period given the other variables? As shown in
Chart 4, the ex post forecast closely tracks actual data
until late 1991; the actual data are only slightly higher
than projected over much of the year. But at the end of
1991, the forecast (the implicit "appropriate" rate of
increases based on the independent variables) exceeds
and diverges sharply from the actual. The jump in the
forecast can be accounted for by the lagged effects of
sharp increases in housing prices. The residential rent
CPI was in line with these explanatory variables over
most of 1990 and 1991 but according to this model was
too low in early 1992. In early 1992, according to this
model the lagged effects of 1991 's strong housing prices
caused by lower interest rates should have kept residential rent higher than BLS reported. However, while
the economy was strong enough in the housing market
to boost home prices temporarily, it was too weak to create large rental gains. These higher housing prices raised
the ex post forecast relative to actual values.

Owners' Equivalent Rent. Owners' equivalent
rent decelerated along the same path as the overall CPI
from mid-1990 through mid-1991, as seen in Chart 1.
The OER inflation rate slowed from a 6 percent pace
to 3 percent within twelve months. In 1992, however,
the OER inflation rate rose back up to 3 1/2 percent
and even briefly to almost 4 percent even though housing price increases slowed in the second half. The key
question is whether or not the 1992 surge was appropriate given the economic conditions, yet the possibility of problems with the OER index in 1992 cannot be
separated from the 1990-91 deceleration. Were the OER
data consistent with economic conditions over both periods?
It is inevitable that price index data temporarily diverge from fundamentals. Conceptually, implicit rents
are measured by the movement in fair market rents,
which can temporarily diverge from changes in prices
for homes as an asset. Rents can continue to increase
while house prices fall until vacancy rates and cheaper
owner-occupied housing restore the equilibrium. Because OER tracked the overall CPI during the 1990-91
period, because the overall CPI rate was pulled down
by the energy component, and because OER should
not have been as affected by oil price declines given
that utility bills are separated from rent payments, it is

Chart 2
Single-Family Home Prices versus Residential Rent
Year-over-Year
Percent Change

8
6
4

2
0

1969

1971

40
Econom ic Review

1973

1975

1977

1979

1981

1983

1985

1987

1989

1991

1993

May/June 1993

Chart 3
C P I Residential Rent versus Multifamily V a c a n c y Rates
CPI (Year-over-Year
Percent Change)

Vacancy Rate
(Percent)

9 -

Residential Renk

8
7 -•

-- 9

6
5
4
3 -•

2 •i

Vacancy Rate

0
1970

1 1 1 1 1 1 1 1 1 1 1 1
1972

1974

1976

1978

1980

1982

1984

1986

1990

1992

Chart 4
C P I Residential Rent, Actual versus Forecast
Monthly Year-over-Year
Percent Change

Federal
Reserve B a n k of Atlanta

In-Sample
"Fitted" Forecast

Ex Post
Forecast

EconoinicRevieu'

Table 3
Regression Model for CPI Rent
Dependent variable: CPI, residential rent, year-ago percent changes
Regression technique: Almon distributed lag
Estimation period: 1970M1 -1989M12
Variable

Coefficient

Standard Error

f-statistic

Constant
Dummy Variable

10.1754
-2.97664

0.404999
0.154941

25.1246
-19.2115

R- squared = 0.778510

Number of observations = 240

/?-bar-squared (Adj for df)
Log of likelihood function
Durbin-Watson
Durbin-Watson (4) (0 gaps)
Sum of squared residuals
Standard error of regression
Sum of residuals
Mean of dependent variable
F-statistic (5, 234)
Significance level

=
=
=

=
=
-

=
-

0.773777
-270.221
0.204517
0.687114
133.567
0.755512
-0.295053E-11
5.61154
1 64.496

Root mean squared error
Mean absolute error
Mean error

=
=
=

0.000000
0.742230
0.571414
-0.1 53197E-01

Lagged, distributed variable: Existing housing prices, median, year-ago percent changes
Distributed lag interpretation
Coefficient
0
-1
-2
-3
-4
-5
-6
-7
-8
-9
-10
-11
-12
-13
-14
-15
-16
-17

0.109842E-01
0.158062E-01
0.199891 E-01
0.235329E-01
0.264376E-01
0.287032E-01
0.303298E-01
0.3131 73E-01
0.316657E-01
0.313750E-01
0.304452E-01
0.288763E-01
0.266684E-01
0.238214E-01
0.203353E-01
0.162101 E-01
0.114458E-01
0.604244E-02

Mean lag = 8.15979
Sum of lag coefficients = 0.413986

Economic Review

Standard Error
0.797078E-02
0.631 781 E-02
0.483425E-02
0.353751 E-02
0.246890E-02
0.1 73120E-02
0.149504E-02
0.1 72046E-02
0.211084E-02
0.247792E-02
0.275045E-02
0.290104E-02
0.291 764E-02
0.279428E-02
0.252771 E-02
0.211601 E-02
0.155797E-02
0.852811E-03

f-statistic
1.37806
2.50185
4.13489
6.65239
10.7082
16.5799
20.2869
18.2029
15.0015
12.6618
11.0692
9.95379
9.14040
8.52504
8.04493
7.66070
7.34662
7.08532
Standard error = 0.602371
Standard error = 0.209571 E-01

May/June 1993

Lagged, distributed variable: Multifamily vacancy rates
Distributed lag interpretation
Coefficient

Standard Error

f-statistic

0
0.104941
-1
0.371062E-01
-2 -0.199847E-01
-3 -0.663321 E-01
-4 -0.101936
-5 -0.126796
-6 -0.140913
-7 -0.144286
-8 -0.136916
-9 -0.118802
-10 -0.899452E-01
-11 -0.503443E-01

0.568049E-01
0.381561 E-01
0.222058E-01
0.946030E-02
0.595388E-02
0.126735E-01
0.1 82749E-01
0.215356E-01
0.223071 E-01
0.205505E-01
0.1 6251 0E-01
0.940205E-02

1.84739
0.972484
-0.899977
-7.01163
-17.1209
-10.0048
-7.71073
-6.69992
-6.13779
-5.78100
-5.53473
-5.35461

Mean absolute lag = 5.96596
Sum of lag coefficients = -0.854210

appropriate to question whether the OER deceleration
in 1991 was excessive. Or perhaps the 1992 acceleration
was an offset to excessive weakness in 1991. Were the
1992 data simply flawed, or was there an unexpected lag in the economic conditions further influencing
OER? In examining these questions, a very basic problem surfaces in terms of validating the OER series: the
index has existed in its present form only since 1983,
and there are fewer observations than for residential
rent. Importantly, over the 1984-92 period little variation occurred in the year-ago percent changes until
1990. With so little variation, any number of possible
independent variables in the same model would work
equally well—or equally poorly. 14 Essentially, there are
too few observations to properly model and evaluate
OER against other variables.
Interestingly, both residential rent and OER forecasts derived from the models predicted inflation rates
higher than actual over the period of concern. This result occurred despite some expectation that the explanatory variables would forecast lower inflation rates than
BLS reported.
Lodging While Out of Town. For the most part,
the lodging-while-out-of-town component has closely
tracked the overall CPI. Over the 1980s, there were a
number of spikes in the data, but these generally represented rate increases being phased in during the offseason. Seasonal factors, which anticipated seasonal
declines when rates might have actually been flat,

Federal Reserve B a n k of Atlanta

Standard error = Undefined
Standard error = 0.476799E-01

boosted the seasonally adjusted numbers. However,
the cause behind the very large increases for this CPI
component in 1990 and into 1992 is different from the
demand factors causing the rate of increases in previous years (although the large gains probably were also
exaggerated by seasonal factors).
Starting in August 1990, the twelve-month change
of the lodging-while-out-of-town component jumped
from 7.3 percent in July to 13.4 percent in August.
The twelve-month changes remained in double digits
through October 1991, reaching a high of 21.4 percent
for January 1991. These unexpectedly large increases
resulted from the shift by U.S. vacationers from international to domestic travel because of crisis abroad associated with the Gulf War. This shift in demand to
domestic vacation areas, particularly to popular U.S.
vacation areas where many of the POPs outlets are located, led to higher prices in these locations while
leaving hotel rates little changed in business district
locations.
Importantly, as pointed out earlier, the BLS shifted
the geographic coverage for the hotel/motel series in
1987 toward measuring rates at resort and vacationlocation hotels and motels. Also, over the 1990-91 offseason, rates were raised and the increases were magnified by seasonal factors. However, the hotel/motel
CPI was out of line with some industry statistics.
For comparison purposes, industry data for this study
were obtained from Pannell Kerr Forster Consulting

EconoinicRevieu'

in San Francisco, California. Annual averages were
available for both the overall hotel industry and for the
narrower resort segment of the industry. 15 Indeed, hotel rates for the overall hotel industry were weak in
1989 and even declining in 1990 (see Chart 5). These
figures include corporate rates, however, as well as
pleasure rates and rates for hotels away from vacation
areas and resort areas. In contrast, fees at resort hotels
posted sizable increases in 1989 and in 1990, with a
more moderate rise in 1991. These figures corroborate
the surge in the hotel/motel CPI as being caused primarily by the change in geographic coverage. Taking
into account that the Pannell Kerr Forster data for resort hotels captured rate increases a year earlier than
the BLS data, probably because of differing calculation methods, this CPI component has been consistent
with the appropriate, narrowly defined, industry data. 16
Essentially, this CPI series never measured rates for
the overall hotel industry and currently should not be
construed as doing so.

Conclusion
Though there have been serious attempts to make
them straightforward measures of the costs of housing
services, both OER and residential rent in the CPI implicitly contain prices of services other than shelter.
Each series is affected by consumer behavior in response to current income, expected income, and current and expected prices f o r shelter alternatives—
owned and rented. This consumer behavior limits the
tightness of the relationship of OER and residential
rent over time and causes each to behave in ways that
are inconsistent with analysis based purely on their
shelter functions.
Nevertheless, despite concerns to the contrary, movement in CPI rental components f r o m 1990 to 1992
were not inconsistent with shelter-related economic data. After taking into account appropriate lag times, the
CPI for residential rent (tenant-occupied) was strongly

Chart 5
Hotel and Motel, CPI versus Industry Estimates
Annual Average
Percent Change

f ¡Total Hotels [ J Resort Hotels

Q CPI Hotels and Motels

-5
1985

1986

1987

1988

1989

1990

1991

1992

Source: Data on "total hotels" and "resort hotels" are from Pannell Kerr Forster Consulting, San Francisco.

Econom ic Review

May/June 1993

correlated with housing prices and multifamily vacancy rates over the same period. Although owners' equivalent rent was at best weakly correlated with these variables, an understanding of the index's methodology
suggests that one should not expect such relationships
to be tight. Owners' equivalent rent did lag residential
rent over the 1990-92 period in the same manner in
which it had over the 1983-1989 period. Given that the
current OER series is relatively new, it is too soon to conclude on a statistical basis whether or not its methodology is consistent with shelter-related economic data.
Finally, OER and RR cannot be expected always to
move together because each series is based on different
types of structures.

The hotel/motel CPI component no longer has earlier geographic restrictions, with the result that this
component primarily tracks consumer hotel/motel
rates in key resort areas. This newly defined hotel/motel component has been consistent with narrowly (but
appropriately) defined industry data on rates at resort
areas. It is with overall rates for hotels during the 198991 period that there is an apparent discrepancy. Such a
broad comparison, however, is not appropriate for critiquing the methodology of this price index for consumers. Taking into account changes in methodology,
the movement in residential rent and in the hotel/motel
CPI is readily explainable by economic conditions.

Notes
l . S e e , for example, Wallich and Corcoran (1989), Davies
(1992), and Madigan (1991).
2. Pollin and Stone (1991) and Madigan (1991) discuss the basis for these concerns.
3. Lodging while at school is the final rental component in the
CPI, but it has a very small weight in the CPI and is not discussed in this article.
4. This article examines data from the all-urban index (CPIU), which covers a broader population base than the CPI for
urban wage earners and clerical workers (CPI-W). The allurban data are more closely followed.
5. The 1982-84 period is used as the base period because of
timing factors involved in meeting the deadline for a December 1986 revision for the CPI. Three years of CES data are
needed to derive the component weights, and the 1982-84
data were the most recently processed CES data available.
6. Component weights and relative importance are different
concepts. Weights refer to base-year shares of components
in the CPI—essentially the fixed real (inflation-adjusted)
amounts of a good or service in the market basket in the
base year. Relative importance reflects the share by expenditures for a component of the CPI in any given period as
component price changes diverge. The relative importance
of components with inflation rates that are higher than the
average rises over time. Their share in the index expands
because of relative price changes over time, not because of
any change in the base-year weight. Weights are changed
only when the CPI is redefined, but relative importance
changes whenever prices for components change at different rates.
7. From 1953 to 1983 the shelter component was based on the
cost of purchasing housing and included a mortgage rale
component. The shelter component—and, it can even be argued, the overall CPI—is essentially a new series starting in
1983 because of these changes in definition.
8. Price changes for these costs appear in separate CPI series.

Federal Reserve B a n k of Atlanta

9. Selection was based on an outlet's probability of being selected, which is proportional to its number of rooms.
10. It may be helpful to provide an overview of the various surveys used as inputs into the CPI. (1) The Consumer Expenditure Survey is used to determine expenditure weights for CPI
components. (2) The Continuing Point-of-Purchase Survey
(CPOPS, or POPS) is used to determine which outlets are to
be priced. (3) The CPI survey actually selects and prices the
goods and services. (4) Housing units are selected from information provided by the 1980 Census, combined with an
on-going new construction permit sample.
The point-of-purchase survey is a household interview
survey conducted to obtain the names and addresses of outlets by defined categories of purchase. The survey also obtains the amount of the h o u s e h o l d ' s expenditure at each
outlet for each purchase category. By category, outlets are
selected for pricing using probability proportional to the reported expenditure.
11. Samples reflect priced outlets as of February 1992. Actual
outlets priced vary, as one-fifth of the pricing areas have
their samples reselected each year. The states accounting for
the largest number of outlets for the samples used to measure price movement through May 1992 and their percentage of the sample are as follows: Florida, 17.3; California,
7.7; South Carolina, 5.6; Virginia, 5.2: Hawaii, 4.8; Nevada,
4.8; Illinois, 3.6; New York, 3.2; North Carolina, 3.2; and
Washington, 2.8.
12. The ten states and their percentages are Florida, 19.2; California, 10.7; Nevada, 5.4; Hawaii, 4.6; New York, 3.8;
Texas, 3.7; Virginia, 3.2; North Carolina, 3.1; South Carolina, 3.1; and Tennessee, 3.0. This expanded sample obviously differs somewhat from the pre-June 1992 sample and
could be responsible for a portion of any change in behavior
in this CPI series.
13. After some experimentation with the lag structure, the price
variable coefficients were estimated with a second-order

EconoinicRevieu'

polynomial distribution for the lag coefficients and a lag period of eighteen months, including the contemporaneous
month. The order of polynomial distribution refers to the
shape of the constraint on the coefficients of the lagged
variable. An n order polynomial distribution has n - 1 inflection points, and so the second-order constraint means
that the coefficients follow a simple rising and falling curve.
The vacancy rate variable also was assigned a second-order
polynomial structure but with a lag length of twelve months.
Also, because of an apparent shift in the relationship between CPI residential rent and home prices, a dummy variable was introduced into the model.
The d u m m y variable is set to 1 for the January 1970
through December 1980 period and is set to 0 for all subsequent months. Over the earlier period, growth in home
prices was consistently higher than for residential rent.
Thereafter, the two inflation rates are much closer. It is unknown what causes this apparent and sharp shift in home
price trends relative to residential rent CPI. The discontinuity could reflect any type of factor, ranging from demographics to changes in financial regulations. The d u m m y
variable is added to the model solely for the purpose of letting the more recent data (post-1980) have more impact on
the forecast period. Adding the dummy variable to the model reduces the importance (magnitude) of the constant term
and allows the coefficient of the other variables to play larger roles in the forecast. However, leaving the dummy variable out does not affect the implications of the model. The
goodness of fit is lowered only modestly.

14. In fact, over the 1984-89 period, there is such limited variation in the dependent variable that the independent variables
in a regression (not shown) using housing prices and vacancy rates have the wrong sign (also indicating some multicollinearity problems). Extending the regression estimation
period through 1991 increases the variation in the dependent
variable and gives more reasonable results for the coefficients.
However, these variables do little to explain movement in
owners' equivalent rent. The adjusted R2 is low at 0.17.
15. Pannell Kerr Forster Consulting surveys hotels in the United States for a variety of information, including hotel characteristics, a c c o m m o d a t i o n s , revenues, and room rates.
Questionnaires are sent to management companies, owners,
and operators. Pannell Kerr Forster regional offices compile
the statistics and forward them to the San Francisco office.
Respondents classify themselves according to a number of
hotel categories: conference center, resort hotel, full-service
hotel, limited-service hotel, suite hotel, and convention hotel. Pannell Kerr Forster defines a resort hotel as a hotel,
usually in a suburban or isolated rural location, with special
recreational facilities to attract pleasure-seeking guests. The
survey s a m p l e covers roughly 1,000-1,200 hotels, with
about eighty classified as resort hotels. The quotes for hotels overall and for resort hotels are average daily rates per
occupicd room. Pannell Kerr Forster does not hold constant
the type of room.
16. Pannell Kerr Forster data are based on yearly revenues divided by rooms. BLS figures are derived from averages of
separate monthly estimates.

Bibliography
Davies, Stephen A. "Policymakers Take the Measure of Inflation, An Old Enemy Now Stretched Out in the Dust." The
Bond Buyer, October 30, 1992, 1, 64.
Gillingham, Robert F. "Estimating the User Cost of OwnerOccupied Housing." Monthly Labor Review (February 1980):
31-35.
Henderson, Steven. "Measuring Home-Ownership in the CPI:
The Flow of Services Using Rental Equivalence Has Replaced the Asset Approach." Paper delivered at the sixtyfifth annual Western Economic Association International
Conference, San Diego, California, July 1990.
Lane, Walter F. " O w n e r s ' Equivalent Rent in the American
Consumer Price Index." U.S. Department of Labor, Bureau

Econom ic Review
4 6

of Labor Statistics, Prices Seminar Series Working Paper
8801-1, January 1988.
Madigan, Kathleen. "How Reliable is the Consumer Price Index?" Business Week, April 29, 1991, 70-71.
Pollin, Robert, and Michael Stone. "The Illusion of an Improved CPI." Challenge (January/February 1991): 53-57.
U.S. Department of Labor. Bureau of Labor Statistics. BLS
Handbook of Methods. Bulletin 2285. April 1988.
. Relative Importance of Components in the Consumer
Price Indexes: U.S. City Average. Table 1. December 1992.
Wallich, Paul, and Elizabeth Corcoran. "The Analytical Economist." Scientific American (July 1989): 76-77.

May/June 1993

J^eview Essay
Capital Ideas:
The Improbable Origins of
Modern Wall Street
by Peter L. Bernstein.
New York: The Free Press, 1992.
340 pages. $24.95.

Thomas J. Cunningham

M
K

The reviewer is the research
officer in charge of the
regional section of the Atlanta
Fed's research
department.

Federal Reserve Bank of Atlanta

o you beat the market? William Sharpe, who would eventually
• win the Nobel prize in economics for work centered on this
t
M very question, asked the question in the mid-1960s over lunch
m
with Peter Bernstein, a mutual-fund manager. Bernstein talked
•
about his performance relative to other fund managers and the
Dow. Sharpe interrupted, asking again, "But do you beat the market?"
At the time, it was tough to tell. No one had ever really asked the question before. "The market" was an ill-defined term, and its return was not
well measured. Bernstein may have beaten the Dow, but Sharpe was more
interested in the total return from a very broad market, not just the price appreciation of an index of a small, albeit well-known, number of stocks.
Things have changed. Finance, in both theory and practice, has undergone a spectacular transformation in a relatively short and recent period.
Peter Bernstein traces the history of finance and finance theory—essentially
the last thirty years—in Capital Ideas: The Improbable Origins of Modern
Wall Street.
Although any historical account, particularly one of ideas, is bound to reflect the personal experiences and prejudices of its author, the chronicle of
financial theory and innovation in Capital Ideas is, in large measure, a recounting of the author's personal experience. Bernstein, who is acquainted

Econom ic Review

with both the theoretical and applied sides of finance theory, has been in and around financial markets throughout his adult life. He did not need to second-guess the
intent of the historical figures in his narrative; he simply talked to them. Despite this personal perspective,
his prose at times seems aimed a bit too high for the
general public. Those with an interest or experience in
finance, however, are likely to find a large part of the
book informative and interesting.
The text is divided into six parts: The single-chapter
introduction and conclusion bracket four multichapter
sections that develop specific topics. Within each section and from section to section, topics are treated
more or less chronologically.

Cowles and the Efficient Markets Idea
Bernstein begins his history of finance theory with
a simple explanation of the efficient markets idea—
that is, why excess stock returns are unforecastable.
While this theory now seems innocuous, early (and
even not so early) in this century the proposition annoyed many employed in financial markets, especially
those who made their living as investment advisors.
The first discrete character in Bernstein's historical
narrative is Alfred Cowles, who set out in the 1930s to
see how well investment advisors actually predicted
market performance over time, something that had
never been done before. The effort proved problematic: tracking several thousand recommendations made
by selected investment firms and advisors and comparing their performance against that of the market as a
whole was a straightforward but mammoth task.
Two things came of Cowles's efforts: first, the Cowles
Foundation, which he established to promote the efforts of scholars interested in combining economics
and statistics; second, his results, which showed that,
over the period examined, the best investment advisor
selected stocks whose yields were about half as much
as the market as a whole. Not surprisingly, these results were not warmly received by the investment advisor community.

7Tie Risk Reward Trade-Off
Bernstein next focuses on the idea of the trade-off
between risk (which Cowles did not consider) and reward and the appropriate way to measure each. He al-

4 8
Econom ic Review

so offers some personal insight into how these ideas
were accepted—or, early in the process, vigorously resisted—by portfolio managers.
Even though Cowles's work was widely available, little came of it for two decades, in part, Bernstein argues,
because finance had not developed as an academic
discipline. Bernstein illustrates this point with a story
about the work of Harry Markowitz. When Markowitz
(an eventual Nobel laureate) defended his doctoral dissertation on portfolio selection at the University of
Chicago in 1952, faculty committee member Milton
Friedman (another future laureate), said, "'Harry, I
don't see anything wrong with the math here, but I
have a problem. This isn't a dissertation in economics,
and we can't give you a Ph.D. in economics for a dissertation that's not economics. It's not math, it's not
economics, it's not even business administration.'" The
committee did, however, come around.
Markowitz's contribution to finance, in a gross simplification, was to show that risk in a portfolio can be
contained through diversification: though each individual asset in the portfolio may be risky, the risks
need not be correlated among the assets. Thus, a diversified set of assets may jointly be less risky than the
assets in the portfolio taken individually.
The idea of portfolio diversification was further developed by James Tobin, who took over the Cowles
Foundation and moved it to Yale. In a 1958 article that
was ultimately central to his winning the Nobel prize,
Tobin emphasized portfolio diversification across a variety of different types of assets as the appropriate response to risk. That is, diversifying with a variety of
risky assets may reduce a portfolio's combined risk.
Tobin's work still serves as a standard of the finance
literature. Bernstein shows how far removed the theoretical side of finance was from the practical application when he reports Tobin as saying, " 'I am unique in
that no real world financial enterprise has ever asked
me for any advice whatsoever.'"
This section of Capital Ideas concludes with a
chapter focusing on William Sharpe, whose question
at the beginning of this essay originally challenged
Bernstein. In 1964 Sharpe developed the empirics behind the capital asset pricing model (CAPM). The
model formalizes the risk-reward trade-off by claiming that returns to assets should be a function of risk as
measured by, in Sharpe's work, their relative volatility.
Of course, recently, single-explanatory-variable models of almost anything, asset returns in particular, have
fallen out of favor. Sharpe's work, however, would enable investment professionals like Bernstein eventually
to answer the question, Do you beat the market? in a

May/June 1993

more systematic (risk-adjusted) way than did the simpler work of Cowles.
To measure just what "the m a r k e t " was doing,
Sharpe undertook a comprehensive accounting of returns to broad m e a s u r e s of securities m a r k e t s , a
formidable task in the 1960s environment of limited
data bases and slow, expensive computers. Sharpe's
work was similar to Cowles's not only in the size of
the task but also in its results: Professional investment
managers systematically underperformed the market
as a whole. One could outperform investment professionals simply by buying "the market" and holding on.
Bernstein argues that such results partly account for
the problem in having academic work in finance accepted in the professional investment management
community.

Technical Analysts versus
Efficient Markets Gurus
In the book's third section Bernstein develops the
efficient markets ideas in the area of stocks' price behavior and the particular notions of their apparently
random character. He starts with a discussion of the initial questions raised about the foundations of technical
analysis, which identifies patterns emerging from price
behavior (typically of stocks or commodities) and tries
to match the current patterns with those of the past to
predict the future. What are now called efficient markets theorists, however, see a kind of catch-22 problem
in technical as well as fundamental analysis; if one
were truly able to discern tomorrow's price from past
patterns, then, in hopes of making a profit, one would
today bid the price of a stock up or down to tomorrow's
price. That is, if tomorrow's price is known for certain,
today's price will move to match, eliminating the potential certain profits. In short, technical market analysis and the efficient markets theory do not mix.
The initial attacks on technical analysis, however,
came not from the financial theorists but from applied
statisticians. In any random behavior, they argued, one
can, after the fact, discern certain patterns. Inevitably,
these patterns will repeat, but any apparent trend is illusory. Such repetition is no different from any random process observed closely and over a long enough
period. In a truly random process, a "technical analyst" will be able to call the next day correctly about
half the time; so, over time, might a coin.
Paul Samuelson, the first American Nobel laureate
in economics, practiced his beliefs in a buy-and-hold

Federal
Reserve Bank of Atlanta

strategy during his many years as a trustee and member of the finance committee of the College Retirement Equities Fund, a large pension fund for college
teachers. Samuelson, firmly in the efficient-markets
camp, did not think portfolio managers performed any
real service. In an article in the inaugural, fall 1974 issue of the Journal of Portfolio Management (of which
Bernstein was the founder and first editor) Samuelson
expressed his scorn for portfolio managers:
T h e y also serve w h o only sit and hold; but I suppose
the f e e s to be earned by such sensible and prosaic b e havior are less than f r o m essaying to give it that old
post-college t r y . . . .
But a respect for evidence c o m p e l s m e to incline
t o w a r d the h y p o t h e s i s that most p o r t f o l i o d e c i s i o n m a k e r s should go out of b u s i n e s s — t a k e u p p l u m b i n g ,
teach Greek, or help produce the annual G N P by serving as c o r p o r a t e e x e c u t i v e s . E v e n if this a d v i c e to
drop dead is good advice, it obviously is not counsel
that will b e eagerly followed. Few people will c o m mit suicide without a push.

The section concludes with a chapter on Eugene
Fama, whom I think of as the current dean of the efficient markets theorists. Fama's work beginning in the
mid-1960s led to the now-prevalent view of the efficient markets literature that news—surprising news—
is the primary factor responsible for moving securities
prices. Securities analysis may have a role in gathering
and disseminating information, and without this activity markets may not be as efficient as they otherwise
would be. The possibility of gaining anything other than
a short-term advantage from such analysis, though, either in information or some clever trading strategy, is
remote. 1

A Class in Corporate Finance
The fourth section of Capital Ideas discusses the
development of what are now some of the basics of
standard corporate finance classes. While most of the
section looks at finance from the buyer's side (how to
structure portfolios) the first chapter of the section
considers the security seller's perspective—how should
a corporation finance itself? On the corporate side,
probably the most startling theoretical innovation to
come from finance is the Modigliani-Miller theorem,
developed in 1958. Modigliani and Miller (Nobel laureates both, though separately) asked whether the
structuring of a company's balance sheet—that is,

EconoinicRevieu'

whether the company finances itself by selling debt
(borrowing) or selling equity (stock)—matters. The
Modigliani-Miller theorem answers the question negatively; whether financing comes from debt or equity is
irrelevant for the firm's owners. The project to be financed will have some payoff to the current stockholders whether they have split that payoff with additional
stockholders or additional bondholders. Institutional
complications (for example, tax considerations) may
lead current equity owners to prefer one means of financing to another, but without these frictions the
choice of issuing equity or debt is a toss-up. If the
overall project is worthwhile, the market will increase
its valuation of the company.
The remaining chapters in the section, in a return to
the financial market perspective, discuss a generalization of the capital asset pricing model and options
pricing models. The chief problem with the CAPM, as
noted above, is that it relies on a single-factor risk to
explain asset returns. Arbitrage pricing theory, developed in 1976 by Stephen Ross, extends the CAPM to
include a variety of a given security's attributes as factors that can explain its return. Finally, Bernstein describes the development, in 1973, of the Black-Scholes
model—which explains an option's price as a function
of the underlying stock's price, volatility, the rate of
interest, and the time to expiration of the option—and
the growth of its acceptance.

F r o m Academe to Application
By this point in the book, the reader is well up on current finance and events. Black-Scholes calculations are
routinely performed in real time on exchange floors.
Buy-and-hold is a recognized and valued investment
strategy. Indexes and technical data services that show
how various market measures are performing over time
have proliferated. Financial theory has found its way
into the practice of finance, and now it is time for the
financial theorists to follow. Bernstein describes this
process in the book's final topical section, titled "From
Gown to Town."
The section's first chapter discusses Wells Fargo
Bank's entry into index funds in 1971. (Banks are not
now permitted to offer this sort of security.) The bank's
strategy was to combine the "best," most diverse portfolio to be had (the market as a whole) with the "best"
management strategy (buy-and-hold). By any standard
of fund performance, Wells Fargo did quite well with
this plan. Strategists at Wells Fargo brought together

50
Econom ic Review

financial academics and combined them with computers
and money to create index funds through which investors could participate in the returns associated with
financial markets indexes. This venture, detailed by
Bernstein, was a triumph for Wells Fargo, which at the
time took a considerable risk, and for financial theory
and theorists. Whatever resistance the traditional finance
community retained toward employing academics largely came to an end.
Nowadays theoretical and practical innovation is
the rule. Two of the more visible recent innovations
are program trading—that is, doing extensive market
price data analysis to find small price "discrepancies"
and trading on them—and portfolio insurance. Bernstein discusses these developments in the second and
third chapters in the section.
The chapter on portfolio insurance is the most interesting in the book. Bernstein's personal experience allows him to discuss the development of portfolio
insurance from both a practitioner's and a theoretician's viewpoint. Portfolio insurance involves trading
in both the actual assets and their futures in a specified
manner to protect ("insure") the portfolio from large
changes in value. The opening of futures markets for
major stock indexes went a long way toward making
portfolio insurance-type trading practical. The downside of such trading was the possibility of a disruption
like the market crash of October 1987. Without assigning blame to portfolio insurance, Bernstein gives a
thoughtful discussion of the resulting Brady Commission report. Whether or not it caused the crash, portfolio insurance certainly suffered because of that event.
The breakdown in the relationship between equity
market prices and futures market prices led portfolio
insurance trading to fail under exactly the sort of conditions that many portfolio insurance clients had hoped
to protect themselves against.

Conclusion
The final chapter contains an overview and conjectures about the future. Bernstein concludes that the
revolution in finance, while dramatic, is not complete.
Ever more efficient, competitive, and innovative markets will make everyone better off.
Bernstein does not discuss current developments affiliated with the ideas he has traced throughout the book.
Efficient markets theory is being questioned from a
number of different perspectives. New financial instruments (such as interest rate swaps and derivatives), all

May/June 1993

with unique and sometimes not well understood risk
characteristics, are proliferating. Bernstein is probably
wise, in a book about the origins of finance theory, to
ignore current problems and developments that he had
not explored earlier in the text. A conclusion with too
many open-ended issues would have detracted from the
historical tone the author established.

however, indicate that the recounting of the theory's
development or implications is uninteresting. Quite the
contrary, Bernstein has produced a readable and entertaining text. Finance professors who devised radically
new trading strategies because "lifestyles were in danger and it was time for an invention" are bound to
make interesting subjects.

Overall, Capital Ideas is an interesting history of finance, but the subtitle, The Improbable Origins of
Modern Wall Street, hints at more than the text delivers. Applying statistical techniques to financial data
strikes me as more of a probable than improbable path
of intellectual development. Acknowledging that the
origins of financial theory are not startling does not,

My one quibble is that Bernstein perhaps tries to
reach too broad an audience. It is unlikely that many,
outside of academics and finance professionals, will
be drawn to the subject. Within the finance community, however, many will find Capital Ideas interesting,
but they may have benefited more from a technically
oriented presentation.

Note
1. Interestingly, Fama has also done some work leading the
questioning of the efficient markets hypothesis; see David N.
DeJong and Charles H. Whiteman, "More Unsettling Evi-

Federal
Reserve B a n k of Atlanta

dence on the Perfect Markets Hypothesis," Federal Reserve
Bank of Atlanta Economic Review 77 (November/December
1992): 1-13.

EconoinicRevieu'

Federal Reserve Bank of Atlanta Working Papers

he Research Department of the Federal Reserve Bank of Atlanta publishes a working paper series to convey the research of
staff economists and visiting scholars and stimulate professional discussion and exploration of economic and financial subjects. To subscribe to the working paper series or to receive single copies of recently published papers (sent automatically to current subscribers), please check the appropriate box(es) below and return the form as indicated.

93-1

Has the Romers' Narrative Approach Identified Monetary Policy Shocks?
Eric M. Leeper

93-2

Risk Taking and Failure in De Novo Savings and Loans in the 1980s
William C. Hunter, James A. Verbrugge, and David A. Whidbee

93-3

Core Deposits and Physical Capital: A Reexamination of Bank Scale Economies and
Efficiency with Quasi-Fixed Inputs
William C. Hunter and Stephen G. Timme

93-4

Some Implications of Risk Neutrality for Time Variation in Stock Returns
Ronald W. Best and Stephen D. Smith

93-5

The Dynamic Impacts of Monetary Policy: An Exercise in Tentative Identification
David B. Gordon and Eric M. Leeper

93-6

Managerial Rents and Optimal Regulatory Intervention in Troubled Banks
Thomas H. Noe, Michael J. Rebello, and Larry D. Wall

93-7

Generalized Method of Moments Tests of Forward Rate Processes
Peter A. Abken

93-8

Private Investment and Sovereign Debt Negotiations
Roberto Chang

•

Please start my subscription to the Working Paper Series.

•

I currently receive other Federal Reserve Bank of Atlanta publications. (To expedite your order, please provide the subscriber number that appears above your mailing address: R
).

•

Please send me a copy of the following working papers:

• 93-1

• 93-2

• 93-3

• 93-4

• 93-5

• 93-6

• 93-7

3 93-8

Name
Address

City

State

ZIP

Return this form to the Public Affairs Department, Federal Reserve Bank of Atlanta, 104 Marietta Street, N.W., Atlanta,
Georgia 30303-2713. Some previously published working papers are also available. For more information write to this address or call 404/521-8020.

: -

í
' ,
-,;>\ - . -

- '

;

^'
. -

•

;

xT,

-•

;
r

y - y....
i

-•

V
,

,
•

-i

•- V ,

- .

c
r

:•:•••

- . _

•'•y '

•

V
'•:: y
-

-í '
y

• ' -

V- •
\

•

•
- ..

. )
- V ^ ...
pi
-y
••
• ^ - y.

- •

•

, ,

" "
. ^

.
, y !"

" •
•
•- , • •-

y
,

'.
í

•
,1

•

y"
.. N
V

, ,

---A

-, ^

' .'

,
v-

•

i '
;

•

•>,- .• -

i ,-s."

- V -

- y

• V

•

y i

: ^
, ..

, • -

•

. .

•

. .
'
.
- •-•; ^ . . y
• ;
„.-^

I-

. ...

;

\ - H, .

• - V

<:
V

¡I

V
: "

Vie"

y .
" - '
- •

;

/
'y

• - -c -

: V •
:

" •

;

>
/ '

http://fraser.stlouisfed.org/
Federal
Bank of, St.
^ ' SReserve
, yLouis r

-N.

- ,

'
V''

• -.,r

• .

.. --

y
"

- ''

•

• '
"

-: -N •
'

y •

Á' - ;

« -

—'
V

c
t'

-- -

, •

,. y

• -

y, "s

•-•• /.

...

..
"

•

• .

Bulk Rate
U.S. Postage

PAID

RESERVE

Atlanta, GA
Permit 292

B A N K OF
Public Affairs

Department

104 M a r i e t t a S t r e e t ,

N.W.

Atlanta, Georgia 30303-2713
( 4 0 4 )

5 2 1 - 8 0 2 0

i«.,Ui.i..M.iltl..Mi!i..,it.!

R000Q000Ü02950 - 0
MS. C A R O L ALDRIDGE
FED RES BANK OF PHILADELPHIA
10 INDEPENDENCE
PHILADELPHIA, PA 19106

printed on recycled paper