View original document

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

Data Dependence
William Poole
This article was originally presented as a speech at the Middle Tennessee State University Annual
Economic Outlook Conference, Murfreesboro, Tennessee, September 29, 2006.
Federal Reserve Bank of St. Louis Review, March/April 2007, 89(2), pp. 77-83.

I

am very pleased to participate in the
Annual Outlook Conference here at Middle
Tennessee State University. However, perhaps strangely, I’ll not say much about the
outlook. Others are better qualified than I to discuss that subject. My topic is how the Fed adjusts
policy when the economy departs from the central tendency outlook. Of course, forecasters commonly have somewhat different views, but each
forecaster’s central tendency, or baseline, forecast
provides his or her best guess as to how the economy will evolve. However, forecasters also need
to be able to say something about probabilities
of other outcomes. The probability distribution
of possible outcomes is substantially affected by
policy responses to deviations from the baseline
outlook if and when those deviations occur. And,
although I say “if and when,” everyone in the
forecasting business knows that our knowledge
of forecast errors requires that we put much more
weight on the “when” than the “if.”
The views I express here are mine and do not
necessarily reflect official positions of the Federal
Reserve System. I thank my colleagues at the
Federal Reserve Bank of St. Louis for their comments. Bill Gavin, vice president in the Research
Division, provided special assistance.
Let me also note at the outset that this speech
is something of a companion to another speech I

gave recently, “Understanding the Fed,” which
was published in the St. Louis Fed’s Review and
is available on our web site.1

SOME BACKGROUND
More than three years ago now, in June 2003,
the Federal Open Market Committee (FOMC) set
its federal funds rate target at a 40-year low of 1
percent, completing, as it turned out, a series of
reductions from a rate of 6½ percent in 2000. The
policy statement accompanying the change in the
policy target concluded with a concern about an
“unwelcome substantial fall in inflation.” The
decline in the inflation rate was only one of a
string of surprises to which the FOMC reacted as
it brought its target rate down. The most shocking
of the surprises, of course, was the terrorist attack
on the United States on September 11, 2001. It
would be time consuming, but not difficult, to
recount this history, pointing to the data releases
and events that led the FOMC to reduce its target
rate between early 2001 and June 2003; such an
account would provide a clear illustration of what
is meant by “data dependence.”
1

William Poole “Understanding the Fed,” Federal Reserve Bank of
St. Louis Review, January/February 2007, 89(1), pp. 3-13;
http://research.stlouisfed.org/publications/review/past/2007/.

William Poole is the president of the Federal Reserve Bank of St. Louis. The author appreciates comments provided by colleagues at the
Federal Reserve Bank of St. Louis. William T. Gavin, vice president in the Research Division, provided special assistance. The views expressed
are the author’s and do not necessarily reflect official positions of the Federal Reserve System.

© 2007, The Federal Reserve Bank of St. Louis. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in
their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made
only with prior written permission of the Federal Reserve Bank of St. Louis.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL

2007

77

Poole

The roughly two-year period after June 2003
was quite different in the sense that monetary
policy does not appear to have been very data
dependent. Following its meeting on August 12,
2003, the FOMC issued a statement that said,
among other things, that “the Committee believes
that policy accommodation can be maintained
for a considerable period.” The funds rate target
remained at 1 percent for a full year. The era of a
1 percent target ended when the FOMC raised the
target to 1¼ percent on June 30, 2004, a policy
adjustment the FOMC had signaled at its previous
meeting in May. By then, as the economy’s recovery continued, there was no doubt that the FOMC
would have to raise its policy target by a substantial amount to support its long-term inflation
objective.
In each of the next 16 consecutive meetings,
the FOMC voted to raise the target for the federal
funds rate by 25 basis points, finally pausing at 5¼
percent in August of 2006. It appeared to some
that policy was on autopilot, as the FOMC raised
the target by 25 basis points meeting after meeting,
apparently independent of incoming information.
That view, I believe, was mistaken. When the
FOMC began the series of rate increases, in June
2004, the statement included this sentence:
“Nonetheless, the Committee will respond to
changes in economic prospects as needed to fulfill
its obligation to maintain price stability.” Similar
language has appeared in every statement since,
and the minutes of the meetings have emphasized
the same point. What happened over the 18
months after June 2004 was, basically, that incoming data indicated that the economy was so close
to the track expected earlier that there was no reason to depart from the “measured pace” of rate
increases of 25 basis points at every meeting.
My purpose today is to discuss in a systematic
fashion the dependence of policy on new information. I can give you a feel, though not a formula,
for why policy decisions are sometimes more data
dependent than at other times. When the target
rate was at 1 percent, or only modestly above, it
was clear that rates had to rise, but a sufficiently
large surprise would have led the FOMC to stop,
slow, or accelerate the increase. In the event, data
surprises were minimal and the FOMC raised the
78

MARCH/APRIL

2007

target by 25 basis points 17 times in a row. Increasingly, though, as the FOMC raised the target funds
rate, policymakers became more sensitive to the
possibility that data surprises could alter the
policy course. As it turned out, the decision to
stop raising rates was determined, in my mind,
less by data surprises than by the economy’s slowing more or less as had been expected many
months before. The August FOMC meeting turned
out to be a good time to pause to take stock of
where the economy stood and the likely course
of events going forward. Whether the August decision to hold the target funds rate unchanged will
turn out to be a pause in the process of raising
rates, a longer-lasting stop, or even the peak, will
depend on the economy’s evolution in coming
months.

THE MODEL
To operate monetary policy effectively and
to understand how policy actions affect the economy, the Federal Reserve relies heavily on economic theory developed over the span of many
decades. The theoretical framework is complicated in its technical form and implementation
but quite straightforward in its bare-bones abstract
framework. The real economy evolves along a
trend that is buffeted by a variety of economic
shocks. Inflation evolves along a trend that is
determined by monetary policy and also buffeted
by these same economic shocks. Although these
shocks drive the business cycle and make the
near-term uncertain, expectations about longerterm trends in both real output growth and inflation have become quite stable.
Long-run output growth has almost always
been fairly predictable because its trend is determined by the trends in the growth of real factors
such as the labor force, the capital stock, and the
level of technology in science, industry, and
management. These trends evolve slowly; since
World War II, real growth has fluctuated around
a 3½ percent average and forecasts of future
growth tend to be centered on that number or
perhaps somewhat lower because labor force
growth is slowing as baby boomers retire.
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Poole

Inflation, on the other hand, has not always
been so predictable. Before 1987, there were wide
swings in the inflation trend and, unlike the case
for real gross domestic product (GDP), longhorizon forecasts of inflation were actually more
uncertain than short-horizon forecasts.2 Today,
after a quarter century of effort by the Fed to
actively contain inflation, inflation has also
become more predictable over all horizons; and
forecasts over longer horizons are now much more
accurate than those over shorter horizons.3 Evidence that long-term inflation has become more
predictable is important, because it means that
the Fed has found a way to anchor the inflation
trend.
Thus, our basic model is of an economy in
which both real growth and the inflation rate are
buffeted by economic shocks in the short run but
then tend to return to predictable long-term trends.
The fluctuations of both output and inflation
around trends have moderated a great deal over
the past 25 years, partly and importantly because
of better monetary policy. This better policy is
due to the Fed concentrating on its objective for
long-run price stability through a more systematic
reaction to incoming information about the economic shocks.
At one time, many economists believed that
there was an inherent tension between stabilizing
inflation and stabilizing the real economy. Over
the past 25 years, we have learned that a condition for stabilizing the real economy is stabilizing
long-run inflation expectations. Thus, one of the
most important things to understand about the
dependence of monetary policy actions on arriving
information is that the Federal Reserve has a deep
commitment to achieving a long-run outcome for
inflation that is in accord with its price stability
objective. Put another way, short-run policy is
strongly motivated by long-run considerations.
2

See Stephen K. McNees, “How Accurate Are Macroeconomic
Forecasts?” Federal Reserve Bank of Boston New England Economic
Review, July/August 1988, pp. 15-36.

3

See evidence on forecast errors over 3-, 12-, and 24-month intervals from 1997 through 2006 in William T. Gavin and Kevin L.
Kliesen, “Forecasting Inflation and Output: Comparing Data-Rich
Models with Simple Rules,” Federal Reserve Bank of St. Louis
Working Paper 2006-054A, September 2006.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MONETARY POLICY
A fundamental component of monetary policy
is the decision about the long-run policy objective
for inflation. This aspect of policy should not be
data dependent. It is possible that an advance in
economic knowledge will teach that we should
have a different long-run inflation objective. No
such advance is on the horizon; but even if it were,
it would not be an exception to the rule that the
policy objective should be independent of incoming information about the current state of the
economy. The policy objective determines the
long-run inflation trend in our model and, more
importantly, the nominal anchor for the economy.
The reaction of policy to incoming news
depends on the state of the economy relative to
the trends. The private sector needs to know the
Federal Reserve’s inflation objective so that it
knows how to view fluctuations around the trend.
Recently, several individual FOMC members
have characterized the long-run inflation goal as
a “comfort zone of 1-2 percent inflation” as measured by inflation in the chain price index for personal consumption expenditures. Although the
FOMC itself has not adopted a formal, quantitative inflation objective, several members, including me, have said that they believe that greater
clarity about the long-run objective would help
both the Committee and the markets to make more
informed decisions.
It is much easier to agree on a long-run inflation objective than on short-run policy actions
consistent with the objective. There is agreement
on two conflicting principles. First, it is all too
easy to overreact to short-run developments.
Agreement on that principle is reflected in the
FOMC’s emphasis on core inflation—inflation
measures excluding volatile food and energy
prices—as a guide to short-run policy. Moreover,
above-trend inflation may be acceptable under
some circumstances, provided we are confident
that past policy actions have been sufficient to
slow inflation in the future. Nevertheless, there
is also agreement on a second principle: It is all
too easy to allow wishful thinking on inflation
to delay necessary tough policy decisions. The
FOMC does its best to make the right choices
MARCH/APRIL

2007

79

Poole

when, as is often the case, “all too easy to overreact” collides with “all too easy to allow wishful
thinking on inflation.”
In one sense, long-run policy is the accumulation of individual short-run policy decisions.
However, if individual decisions reflect only reactions to short-run developments in the economy,
then there is no telling where long-run policy
will go. The right way for the Fed to think about
short-run policy decisions is that they have to be
part of, or fit into, a coherent long-term plan. The
market’s understanding of this plan is central to
the determination of long-term interest rates. In
general, the rate on any bond depends on expected
short rates over the horizon of the bond. Thus, the
10-year Treasury bond rate depends on expectations of short-term interest rates over the 10-year
horizon.
Market expectations about future interest rates
depend on the interaction of two interrelated
sources of influence. One, obviously, concerns
Federal Reserve decisions on the intended federal
funds rate. Also important are expectations as to
the demands for and supplies of funds in the
private market. For example, with simultaneous
investment and housing booms, credit demands
will be high and interest rates will tend to be bid
up. In pursuing its policy goals, the FOMC will
be adjusting the federal funds rate as needed to
keep the inflation rate low and stable. Thus, the
market forms expectations about the underlying state of the economy that will bear on Fed
decisions.
The Federal Reserve is constantly evaluating
the situation in the markets and trying to adjust
the intended federal funds rate to produce a satisfactory equilibrium in the economy. When we put
the Federal Reserve’s and the market’s decisions
and expectations together, we have a macroeconomic equilibrium.
The interaction between the Federal Reserve
and the markets may be confusing at first sight,
and indeed was confusing to economists for generations until conceptual breakthroughs in the
1960s and 1970s clarified the issue. Market behavior depends on expectations as to what the Federal
Reserve is going to do, and what the Federal
Reserve is going to do depends on what the market
80

MARCH/APRIL

2007

and the economy are anticipated to do. The full
rational expectations macroeconomic equilibrium
occurs when the market behaves as the Federal
Reserve expects and the Federal Reserve behaves
as the market expects. In both cases we assume
that the expectations are fully rational, by which
we mean that the expectations are fully informed
on the basis of all available information. The
abstraction of a full rational expectations macroeconomic equilibrium provides a powerful starting point for analysis of a data-dependent policy.

CAN THE MARKET PREDICT
DATA DEPENDENCE?
The “Taylor rule” is a stylized view of the
Fed’s reaction to incoming information. In 1993,
Stanford economist John Taylor proposed a simple
formula relating the federal funds rate to (i) a
long-run inflation target and (ii) short-run deviations of inflation from that target and short-run
deviations of real GDP from a measure of “potential real GDP.”4 Taylor suggested that his simple
relationship characterized in broad outline the
actual behavior of the federal funds rate in the
early years of the Greenspan FOMC. The essence
of this relationship is that in the long-run the
FOMC seeks to keep the federal funds rate roughly
consistent with a level that is believed to produce
a target level of inflation. Taylor assumed a target
rate of inflation of 2 percent per year measured
by the total consumer price index (CPI). In the
short run, the relationship implies that the FOMC
adjusts the target federal funds rate up as either
the observed inflation rate exceeds its target or
real GDP exceeds potential real GDP. Conversely,
under the Taylor rule, the FOMC reduces the target federal funds rate when inflation falls below
its target and/or real GDP falls short of potential
real GDP.
The Taylor rule reflects the primacy of a longrun inflation objective while incorporating short4

John B. Taylor, “Discretion versus Policy Rules in Practice,”
Carnegie-Rochester Conference Series on Public Policy, 39,
December 1993, pp. 195-214. Taylor compared the values of his
formula against the observed history of the funds rate from 1987
through 1992.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Poole

run stabilization efforts. The rule provides a formula for computing a baseline, or reference,
interest rate that is consistent with policy achieving the Fed’s objectives for both output stabilization and price stability. I discussed the Taylor rule
in some detail in the speech I mentioned earlier,
“Understanding the Fed,” and refer you to its
published version in this Review if you want to
dig into the subject more deeply.
Now I’ll turn to some comments on future
Fed policy, but I want to remind you that I am
speaking for myself—other FOMC participants
may have different views about how future policy
adjustments will depend on arriving information.
All economic indicators may have implications
for the evolution of the real economy and inflation.
I emphasize “may” because we have to filter out
as best we can possible data errors and inconsistencies across various indicators.
Before I discuss future Fed policy in any detail,
I begin with a warning. New information drives
both market adjustments and policy changes, but
new information is inherently unpredictable. To
gain a sense of the impact of new information on
interest rates, I’ve analyzed data from the eurodollar futures market and discussed the results in
some detail in “Understanding the Fed.” The
bottom line of that analysis is that forecasts embedded in the eurodollar futures market explain 42
percent of the variance of fluctuations in the actual
eurodollar yield three months ahead. Thus, unpredictable events even over a three-month horizon
are responsible for 58 percent of the variance of
the eurodollar yield. Over a six-month horizon,
unpredictable events are responsible for more than
70 percent of the variance. Thus, I can discuss
various scenarios but have no way of knowing
which scenario will come to pass.
Let’s start with the outlook for the rest of
2006. Forecasts made by FOMC members and
transmitted to Congress in July were 3¼ to 3½
percent growth for real GDP and an increase for
the core personal consumption expenditures
(PCE) chain price index of 2¼ to 2½ percent. As
for 2007, the central tendency of the FOMC members’ GDP forecasts is 3 to 3½ percent. This growth
outlook should be consistent with keeping the
economy close to full employment, based on the
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Congressional Budget Office forecast of potential
GDP growth of 3.24 percent in 2007. As for inflation, the central tendency forecast of FOMC participants for 2007 is 2 to 2¼ percent. Thus,
inflation is expected to recede only very slowly
from its current level.
There are two cases in which the economic
news will pretty clearly predict a change in the
Fed’s policy stance. If incoming economic indicators show that both output and inflation are
rising above these forecasts, then, in the absence
of any other information, we can expect that the
FOMC will increase its target federal funds rate.
On the other hand, if both output and inflation
come in weaker than expected, we are unlikely
to see further increases in the federal funds target; indeed, if economic weakness is pervasive
enough, the FOMC will at some point reduce the
target funds rate.
The most interesting—not to mention controversial and difficult—cases are those in which the
outlook for inflation and output move in opposite directions. In such cases, the FOMC has to
call on all its experience and judgment to reach a
decision. It is very difficult for me to be precise
about the judgments I am likely to reach based
on incoming information because a host of considerations, some of which I cannot foresee, may
enter the calculus. But I’ll make a stab at how
things could play out to illustrate my thought
process.
A critically important consideration in my
mind concerns the inflation process and the
importance of the Fed’s commitment to low and
stable inflation. It is my conviction that temporizing on actions to control inflation is an invitation
to trouble. Accepting higher inflation, or even a
continuation of the current rate of inflation, in
an effort to sustain current employment levels
will only lead to more grief later. Once inflation
and inflation expectations rise, the economy will
become less stable and reducing inflation from
an elevated rate will be more costly than taking
the medicine now. Having said that, if inflation
pressures are easing, even if only gradually, and
there is a genuine prospect that inflation will
return to the comfort zone, then I see no reason
to accelerate the decline in inflation by maintainMARCH/APRIL

2007

81

Poole

ing a restrictive policy in the face of declining
employment. Policy needs to be as disciplined
as necessary to get the job done, but not more so.
The long-run inflation goal and the attitude
I’ve expressed about what risks to take suggest
that I will have a bias in the way I interpret incoming information. If data on the real economy come
in weaker than expected—if it appears that the
economy is falling below the baseline forecast
path—then my bias will be in the direction of
wanting to be sure that the data paint a consistent
picture before I’ll advocate a policy easing. But
if the picture is consistent, and inflation risk is
receding, then I’ll not hesitate to advocate policy
easing.
What I hope the FOMC can accomplish is to
retain full market confidence that the long-run
rate of inflation will remain in the comfort zone.
I hope that forecasters assign very low probability to inflation outcomes over the medium term
of 3 to 5 years outside the comfort zone no matter
what the incoming data look like. Although I am
talking about inflation over a horizon well beyond
the usual forecast horizon of 1 to 2 years, the longrun inflation outlook has a direct bearing on the
forecast. The long bond rate today depends critically on expected inflation over the maturity of
the bond. Thus, rates that enter importantly into
any economic forecast, such as mortgage and
corporate bond rates, depend on the long-run
inflation outlook. This outlook has been quite
stable in recent years, and that fact is evidence of
a major monetary policy success.
With long-run inflation contained, the FOMC
has flexibility to respond, vigorously if necessary,
to economic weakness should it arise. The FOMC
brought the target federal funds rate down aggressively in 2001 in response to incoming information. Aggressive easing kept the recession mild.
If the economy comes in below the baseline forecast in coming quarters, the FOMC will have room
to act as aggressively as required. I have no idea
what scale of easing might be appropriate, for
that will depend on the nature of the incoming
information. Still, I believe forecasters should
assign a relatively low probability to deep recession precisely because of the FOMC’s demonstrated willingness to act aggressively as necessary.
82

MARCH/APRIL

2007

I’ve given you my take on what data dependence means and the attitudes that underlie my
likely responses. I’ve also emphasized that an
efficient rational expectations equilibrium requires
that the market behave as the policymakers expect
and policymakers behave as the market expects.
The market’s evaluation of the prospects for policy is revealed in the futures markets for federal
funds and eurodollar deposits. Current futures
prices predict that the federal funds target is
expected to begin moving down. Because these
market quotes change day by day in response to
new information, I do not want to attempt to be
particularly precise as to the timing—anything I
write as I draft these remarks may be out of date
by the time I deliver them or within a few weeks,
anyway. What I can safely note is that the market’s
expectation of future policy easing has been taking
hold gradually since late June, say, in response
to data on the real economy suggesting that real
growth is slowing and inflation data suggesting
that the worst may be over on that front.
Although expectations about future policy
actions are revealed transparently in the futures
market for short-term interest rates, I want to
underscore my earlier point about the limited
accuracy of those forecasts. Some of the forecast
misses have been pretty dramatic. For example,
in December 2000, the futures market forecasts
were for a decline in the eurodollar yield of 35
basis points over the following three months
and a total of 67 basis points over the six-month
period. Instead, the FOMC acted aggressively to
lower the funds rate target starting in January and
continuing through May 2001 by a total of 250
basis points. The FOMC acted aggressively as
incoming information pointed to growing weakness in economic activity. Both the FOMC and
the markets were surprised by incoming information indicating that the economy was weakening
quickly and significantly.
Although I cannot predict unpredictable new
information, I’ve tried to provide a sense of how
I might respond to new information as it arrives.
I note, however, that it is rare that a single data
report is decisive. The economic outlook is determined by numerous pieces of information. Important data such as the inflation and the employment
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Poole

reports are cross-checked against other information. The FOMC is aware of the possibility of data
revisions and short-run anomalies. Sometimes
data ought to be discounted because of anomalous
behavior.
An example was the increase in tobacco
prices in late 1998. Tobacco prices had a transitory impact on measured inflation, both total and
core indices, during December 1998 and January
1999, but produced no lasting effect on trend
inflation. Similarly, information about real activity
sometimes arrives that indicates transitory shocks
to aggregate output and employment. An example
of such a transitory shock is the strike against
General Motors in June and July 1998. Similarly,
the September 2005 employment report reflected
the impact of Hurricane Katrina, which was
expected to be, and turned out to be, temporary
from a national perspective.
Transitory and anomalous shocks to the data
are ordinarily rather easy to identify. Both Fed and
market economists develop estimates of these
aberrations in the data shortly after they occur.
The principle of looking through aberrations is
easy to state but probably impossible to formalize
with any precision. We know these shocks when
we see them, but could never construct a completely comprehensive list of such shocks ex ante.
Policymakers piece together a picture of the
economy from a variety of data, including anecdotal observations. When the various observations fit together to provide a coherent picture,
the Fed can adjust the intended rate with some
confidence. The market generally understands
this process, as it draws similar conclusions from
the same data.
So, given policy objectives, and given a view
about how policy decisions affect the economy,
the central bank can in principle specify a policy
rule, or response function, that guides policy
adjustments in response to incoming information.
To achieve a good result, the general public and
market participants need to understand the objectives and the response function so that the private
economy can determine its activities with full
knowledge of how the central bank will act. Of
course, uncertainty is an inherent characteristic
of the economic world. What should be predictF E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

able are the central bank’s responses to the neverending sequence of surprises that characterize
the economic environment.
Market commentary often indicates frustration that the FOMC does not lay out a clearer path
for policy, arguing that the FOMC is unpredictable. That view, I believe, is off base. Typically
the FOMC cannot be predictable with regard to
the path of the target federal funds rate because
new information driving policy adjustments is
not predictable. All of us would like to be able to
predict the future. We in the Fed do the best we
can, but the markets should not complain that
the FOMC lacks clairvoyance! What the FOMC
strives to do is to respond systematically to the
new information. There is considerable evidence
that the market does successfully predict FOMC
responses to the available information at the
time of regularly scheduled meetings.5

CONCLUDING COMMENT
To say that policy is data dependent means
that policy changes will depend on the incoming
news about the state of the economy, both real
growth and inflation. That the policy setting is
data dependent is a good sign. It means that policy
is in a range than can be considered neutral—
that is, thought to be consistent with the Fed’s
longer-run policy objectives. It is important to
remember that the long-run inflation objective
should not be data dependent. If the objective is
well understood, people will know whether the
current inflation rate is above or below the desired
trend. They will know how to interpret incoming
information to gauge what it means for the policy
stance. I believe that is just about exactly where
we are today.

5

See, for example, William Poole, “How Predictable Is Fed
Policy?” Federal Reserve Bank of St. Louis Review,
November/December 2005, 87(6), pp. 659-68;
http://research.stlouisfed.org/publications/review/past/2005/.

MARCH/APRIL

2007

83

84

MARCH/APRIL

2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Data, Data, and Yet More Data
William Poole
This article was originally presented as a speech at the Association for University Business and
Economic Research (AUBER) Annual Meeting, University of Memphis, Memphis, Tennessee,
October 16, 2006.
Federal Reserve Bank of St. Louis Review, March/April 2007, 89(2), pp. 85-89.

I

’ve long had an interest in data, and I think
that this topic is a good one for this conference. The topic is also one I’ve not
addressed in a speech.
A personal recollection might be a good place
to begin. In the early 1960s, in my Ph.D. studies
at the University of Chicago, I was fortunate to be
a member of Milton Friedman’s Money Workshop.
Friedman stoked my interest in flexible exchange
rates, in an era when mainstream thinking was
focused on the advantages of fixed exchange rates
and central banks everywhere were committed
to maintaining the gold standard. Well, I should
say central banks almost everywhere, given that
Canada had a floating-rate system from 1950 to
1962. Friedman got me interested in doing my
Ph.D. dissertation on the Canadian experience
with a floating exchange rate, and later I did a
paper on nine other floating rate regimes in the
1920s. For this paper I collected daily data on
exchange rates from musty paper records at the
Board of Governors in Washington.
What was striking about the debates over
floating rates in the 1950s is that economists
were so willing to speculate about how currency
speculators would destabilize foreign exchange
markets without presenting any evidence to support those views. In this and many other areas,

careful empirical research has resolved many
disputes. Our profession has come a long way in
institutionalizing empirical approaches to resolving empirical disputes. The enterprise requires
data, and what I will discuss is some of the history of the role of the Federal Reserve Bank of
St. Louis in providing the data.
Before proceeding, I want to emphasize that
the views I express here are mine and do not
necessarily reflect official positions of the Federal
Reserve System. I thank my colleagues at the
Federal Reserve Bank of St. Louis for their comments. Robert H. Rasche, senior vice president and
director of research, provided special assistance.

ORIGINS
The distribution of economic data by the
Research Division of the Federal Reserve Bank
of St. Louis can be traced back at least to May
1961. At that time, Homer Jones, then director
of research, sent out a memo with three tables
attached showing rates of change of the money
supply (M1), money supply plus time deposits,
and money supply plus time deposits plus shortterm government securities. His memo indicated
that he “would be glad to hear from anyone who

William Poole is the president of the Federal Reserve Bank of St. Louis. The author thanks colleagues at the Federal Reserve Bank of St. Louis.
Robert H. Rasche, senior vice president and director of research, provided special assistance. The views expressed are the author’s and do
not necessarily reflect official positions of the Federal Reserve System.

© 2007, The Federal Reserve Bank of St. Louis. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in
their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made
only with prior written permission of the Federal Reserve Bank of St. Louis.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL

2007

85

Poole

thinks such time series have value, concerning
promising applications or interpretations.”
Recollections of department employees from
that time were that the mailing list was about
100 addressees.
Apparently Homer received significant positive feedback, since various statistical releases
emerged from this initial effort. Among these
were Weekly Financial Data, subsequently U.S.
Financial Data; Bank Reserves and Money, subsequently Monetary Trends; National Economic
Trends (1967) and International Economic Trends
(1978), all of which continue to this date. In April
1989, before a subscription price was imposed, the
circulation of U.S. Financial Data had reached
almost 45,000. A Business Week article published
in 1967 commented about Homer that “while most
leading monetary economists don’t buy his theories, they eagerly subscribe to his numbers.” As
an aside, as a Chicago Ph.D., I both bought the
theories and subscribed to the data publications.
By the late 1980s, according to Beryl Sprinkel
(1987, p. 6), a prominent business economist of
the time, “weekly and monthly publications of
the Research Division, which have now become
standard references for everyone from undergraduates to White House officials, were initially
Homer’s products.”
Why should a central bank distribute data as
a public service? Legend has it that Homer Jones
viewed as an important part of his mission providing the general public with timely information
about the stance of monetary policy. In this sense
he was an early proponent, perhaps the earliest
proponent, of central bank accountability and
transparency. While Homer was a dedicated
monetarist, and data on monetary aggregates
have always figured prominently in St. Louis
Fed data publications, data on other variables
prominent in the monetary policy debates at the
time, including short-term interest rates, excess
reserves, and borrowings, were included in the
data releases.
Early on, the various St. Louis Fed data publications incorporated “growth triangles,” which
tracked growth rates of monetary aggregates over
varying horizons. Accompanying graphs of the
aggregates included broken trend lines that illus86

MARCH/APRIL

2007

trated rises and falls in growth rates. This information featured prominently in monetarist critiques
of “stop-go” and procyclical characteristics of
monetary policy during the Great Inflation period.
Does the tradition of data distribution initiated
by Homer Jones remain a valuable public service?
I certainly believe so. But I will also note that the
St. Louis Fed’s data resources are widely used
within the Federal Reserve System. This information is required for Fed research and policy
analysis; the extra cost of making the information
available also to the general public is modest.

RATIONAL EXPECTATIONS
MACROECONOMIC EQUILIBRIUM
The case for making data readily available is
simple. Most macroeconomists today adhere to a
model based on the idea of a rational expectations
equilibrium. Policymakers are assumed to have
a set of goals, a conception of how the economy
works, and information about the current state
and history of the economy. The private sector
understands, to the extent possible, policymakers’
views and has access to the same information
about the state and history of the economy as
policymakers have.
An equilibrium requires a situation in which
(i) the private sector has a clear understanding of
policy goals and the policymakers’ model of the
economy and (ii) the policy model of the economy
is as accurate as economic science permits. Based
on this understanding, market behavior depends
centrally on expectations concerning monetary
policy and the effects of monetary policy on the
economy, including effects on inflation, employment, and financial stability. If the policymakers
and private market participants do not have views
that converge, no stable equilibrium is possible
because expectations as to the behavior of others
will be constantly changing.
The economy evolves in response to stochastic disturbances of all sorts. The continuous flow
of new information includes everything that
happens—weather disturbances, technological
developments, routine economic data reports,
and the like. The core of my policy model is that
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Poole

market responses and policy responses to new
information are both maximizing—households
maximize utility, firms maximize profits, and
policymakers maximize their policy welfare
function.
A critical assumption in this model is the symmetry of the information that is available to both
policymakers and private market participants. In
cases where the policymakers have an informational advantage over market participants, policy
likely will not unfold in the way that markets
expect, and the equilibrium that I have characterized here will not emerge. Hence, public access
to current information on the economy at low
cost is a prerequisite to good policy outcomes.

THE EVOLUTION OF ST. LOUIS
FED DATA SERVICES
Data services provided by the Federal Reserve
Bank of St. Louis have evolved significantly from
the paper publications initiated by Homer Jones.
The initial phase of this evolution began in April
1991 when FRED®, Federal Reserve Economic
Data, was introduced as a dial-up electronic bulletin board. This service was not necessarily low
cost. For users in the St. Louis area, access was
available through a local phone call. For everyone
else, long-distance phone charges were incurred.
Nevertheless, within the first month of service,
usage was recorded from places as wide ranging
as Taipei, London, and Vancouver.1 FRED was
relatively small scale. The initial implementation
included only the data published in U.S. Financial
Data and a few other time series. Subsequently,
it was expanded to include the data published in
Monetary Trends, National Economic Trends, and
International Economic Trends. At the end of
1995, the print versions of these four statistical
publications contained short histories on approximately 200 national and international variables;
initially FRED was of comparable scope.
The next step occurred in 1996 when FRED
migrated to the World Wide Web. At that point,
403 national time series became available instan1

Eighth Note (1991, p. 1).

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

taneously to anyone who had a personal computer
with a Web browser. An additional 70 series for
the Eighth Federal Reserve District were also available. The data series were in text format and had
to be copied and pasted into the user’s PC. In July
2002, FRED became a true database and the user
was offered a wider range of options. Data can be
downloaded in either text or Excel format. Shortly
thereafter, user accounts were introduced so that
multiple data series can be downloaded into a
single Excel workbook, and data lists can be stored
for repeated downloads of updated information.
In the first six months after this version of FRED
was released, 3.8 million hits were recorded to
the web site. In a recent six-month period, FRED
received 21 million hits from over 109 countries
around the world. FRED currently contains 1,175
national time series and 1,881 regional series.
FRED data are updated on a real-time basis as
information is released from various statistical
agencies.
After 45 years, Homer Jones’s modest initiative to distribute data on three variables has developed into a broad-based data resource on the
U.S. economy that is available around the globe
at the click of a mouse. Through this resource,
researchers, students, market participants, and
the general public can reach informed decisions
based on information that is comparable to the
information policymakers have.
In the past year, we have introduced a number
of additional data services. One of these, ALFRED®
(Archival Federal Reserve Economic Data), adds
a vintage (or real-time) dimension to FRED. The
ALFRED database stores revision histories of the
FRED data series. Since 1996, we have maintained
monthly or weekly archives of the FRED database.
All the information in these archives has been
populated to the ALFRED database, and the user
can access point-in-time revisions of these data.2
We have also extended the revision histories of
many series back in time using data that were
2

We do not maintain histories of daily data series in ALFRED.
Interest rates and exchange rates appear at daily frequencies in
FRED. In principle, these data are not revised, though occasional
recording errors do slip into the initial data releases. Such reporting
errors are corrected in subsequent publications, so there is sometimes a vintage dimension to one of these series.

MARCH/APRIL

2007

87

Poole

recorded in U.S. Financial Data, Monetary Trends,
and National Economic Trends. For selected
quarterly national income and product data, we
have complete revision histories back to 1959
for real data and 1947 for nominal data. Revision
histories are available on household and payroll
employment data back to 1960. A similar history
for industrial production is available back to 1927.
Preserving such information is crucial to
understanding historical monetary policy. For
example, Orphanides (2001, p. 964) shows “that
real-time policy recommendations differ considerably from those obtained with ex-post revised
data. Further, estimated policy reaction functions
based on ex-post revised data provide misleading
descriptions of historical policy and obscure the
behavior suggested by information available to the
Federal Reserve in real time.” Orphanides concludes that “reliance on the information actually
available to policy makers in real time is essential for the analysis of monetary policy rules.”
Such vintage information also is essential for
analysis of conditions at subnational levels. For
example, in January 2005 the Bureau of Labor
Statistics estimated that nonfarm employment in
the St. Louis MSA had increased by 38.8 thousand
between December 2003 and December 2004.
This increase was widely cited as evidence that
the MSA had returned to strong employment
growth after four years of negative job growth.
However, these data from the Current Employment
Statistics were not benchmarked to more comprehensive labor market information that is available
only with a lag.3 The current estimate of nonfarm
employment growth in the St. Louis MSA for this
period, after several revisions, is only 11.6 thousand, less than 30 percent of the increase originally
reported.
Another data initiative that we launched several years ago is FRASER®—the Federal Reserve
Archival System for Economic Research. The
objective of this initiative is to digitize and distribute the monetary and economic record of the
U.S. economy. FRASER is a repository of image
files of important historical documents and serial
publications. At present we have posted the entire

history of The Economic Report of the President,
Economic Indicators, and Business Conditions
Digest. We have also posted images of most issues
of the Survey of Current Business from 1925
through 1990 and are working on filling in images
of the remaining volumes. The collection also
includes Banking and Monetary Statistics and the
Annual Statistical Digests published by the Board
of Governors, as well as the Business Statistics
supplements to the Survey of Current Business
published by the Department of Commerce. We
are currently working, in a joint project with the
Board of Governors, to create digital images of
the entire history of the Federal Reserve Bulletin.
Finally, we are posting images of historical statistical releases that we have collected in the process
of extending the vintage histories in ALFRED
back in time. These images should allow scholars,
analysts, and students of economic history to
reconstruct vintage data on many series in addition to those we are maintaining on ALFRED.

3

4

Wall and Wheeler (2005).

88

MARCH/APRIL

2007

TRANSPARENCY,
ACCOUNTABILITY, AND
INFORMATION DISTRIBUTION
As just indicated, the scope of the archival
information in FRASER extends beyond numeric
data. Ready access to a wide variety of information
is essential for transparency and accountability
of monetary authorities and the public’s full
understanding of policy actions. Since 1994, the
Federal Reserve System and the FOMC have
improved the scope and timeliness of information
releases. I have discussed this progress in previous
speeches.4 Currently, the FOMC releases a press
statement at the conclusion of each scheduled
meeting and three weeks later follows up with
the release of minutes of the meeting. The press
release and the minutes of the meetings record
the vote on the policy action. The policy statement and minutes give the public a clear understanding of the action taken and insight into the
rationale for the action.
Contrast the current situation with the one in
1979. At that time, actions by the Board of
See, for example, Poole (2005).

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Poole

Governors on discount rate changes were reported
promptly, but there was no press release subsequent to an FOMC policy action and FOMC meeting minutes were released with a 90-day delay.
On September 19, 1979, the Board of Governors
voted by the narrow margin of four to three to
approve a ½-percentage-point increase in the
discount rate, with all three dissents against the
increase. This information generated the public
perception that Fed officials were sharply divided
and, therefore, that the Fed was not prepared to
act decisively against inflation. John Berry (1979,
p. A1), a knowledgeable reporter at the Washington
Post, observed that “the split vote, with its clear
signal that from the Fed’s own point of view interest rates are at or close to their peak for this business cycle, might forestall any more increases in
market interest rates.” However, the interpretation
of the “clear signal” was erroneous. On that same
day, the FOMC had voted eight to four to raise the
range for the intended funds rate to 11¼ to 11¾
percent. More importantly, three of the four dissents were in favor of a more forceful action to
restrain inflation (see Lindsey, Orphanides, and
Rasche, 2005, pp. 195-96). Neither the FOMC’s
action, the dissents, nor the rationale for the dissents were revealed to the public under the disclosure policies then in effect. The result was to
destabilize markets, with commodity markets, in
particular, exhibiting extreme volatility.

REFERENCES
Berry, John. “Fed Lists Discount Rate to Peak of 11%
on Close Vote.” Washington Post, September 19,
1979, p. A1.
Business Week. “Maverick in the Fed System.”
November 18, 1967.
Eighth Note. “Introducing FRED.” Federal Reserve
Bank of St. Louis, May/June 1991, p. 1.
Orphanides, Athanasios. “Monetary Policy Rules
Based on Real-Time Data.” American Economic
Review, September 2001, 91(4), pp. 964.
Poole, William. “FOMC Transparency.” Federal
Reserve Bank of St. Louis Review, January/February
2005, 87(1), pp. 1-9.
Sprinkel, Beryl W. “Confronting Monetary Policy
Dilemmas: The Legacy of Homer Jones.” Federal
Reserve Bank of St. Louis Review, March 1987,
69(3), p. 6.
Wall, Howard J. and Wheeler, Christopher H.
“St. Louis Employment in 2004: A Tale of Two
Surveys.” CRE8 Occasional Report No. 2005-1,
Federal Reserve Bank of St. Louis, February 2005.

CONCLUSION
The tradition of data services was well established when I arrived in St. Louis in 1998, and I
must say that I am proud that leadership in the
Bank’s Research Division has extended that tradition. Data are the lifeblood of empirical research
in economics and of policy analysis. Our rational
expectations conception of how the macroeconomy works requires that the markets and general
public understand what the Fed is doing and
why. Of all the things on which we spend money
in the Federal Reserve, surely the return on our
data services is among the highest.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL

2007

89

90

MARCH/APRIL

2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Stock Market Booms and Monetary Policy
in the Twentieth Century
Michael D. Bordo and David C. Wheelock
This article examines the association between stock market booms and monetary policy in the
United States and nine other developed countries during the 20th century. The authors find, as
was true of the U.S. stock market boom of 1994-2000, that booms typically arose during periods
of above-average growth of real output and below-average inflation, suggesting that booms reflected
both real macroeconomic phenomena and monetary policy. They find little evidence that booms
were fueled by excessive liquidity. Booms often ended within a few months of an increase in
inflation and consequent monetary policy tightening. They find few differences across the different
monetary policy regimes of the century. (JEL E300, E520, G180, N100, N200)
Federal Reserve Bank of St. Louis Review, March/April 2007, 89(2), pp. 91-122.

E

xtended periods of rapidly appreciating
equity, housing, and other asset prices
in the United States and elsewhere since
the mid-1990s have brought increased
attention to the effects of monetary policy on
asset markets and the appropriate response, if
any, of monetary policy to asset price booms.
Some economists argue that financial markets
are inherently volatile and that market prices
often stray from fundamentals, suggesting that
policymakers could improve welfare by attempting to deflate asset price booms, especially if sudden declines in asset prices are likely to depress
economic activity. Other economists contend that
financial markets process information efficiently.
These economists tend to believe that policymakers usually cannot determine when assets
are mispriced and, hence, that they cannot
enhance aggregate welfare by reacting to asset
price movements.1
The U.S. stock market boom of the late 1990s
arose during a period of increased productivity

growth, which many observers hailed as evidence
of a “New Economy” that justified rapid appreciation of equity prices. The period was also marked
by low and stable inflation, which may have contributed to the boom by foreshadowing strong
growth of economic activity and corporate profits.
Some analysts have argued, however, that the
rapid rise in equity prices was simply a manifestation of loose monetary policy that happened to
generate asset price inflation rather than consumer
price inflation.2 The end of the boom did coincide
with a tightening of monetary policy. This tightening seems to have been in response to rising
consumer price inflation and inflation expecta1

See Kohn (2006) for a recent comparison of alternative monetary
policy strategies in response to asset price booms. See also Bordo
and Wheelock (2004).

2

Rapid growth of asset prices amid low consumer price inflation
renewed interest in the question of whether monetary policy should
target measures of inflation that include asset prices as well as
consumer prices. Proponents of broader inflation measures include
Goodhart and Hofmann (2000) and Bryan, Cecchetti, and O’Sullivan
(2002).

Michael D. Bordo is a professor of economics at Rutgers University and an associate of the National Bureau of Economic Research. David C.
Wheelock is an assistant vice president and economist at the Federal Reserve Bank of St. Louis. The authors thank Edward Nelson and Rajdeep
Sengupta for comments on an earlier draft. Daniel McDonald and Joshua Ulrich provided research assistance.

© 2007, The Federal Reserve Bank of St. Louis. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in
their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made
only with prior written permission of the Federal Reserve Bank of St. Louis.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

91

Bordo and Wheelock

tions, though some studies conclude that the Fed
also sought to contain the booming stock market.3
The purpose of this article is to discern
whether the patterns of output and productivity
growth, inflation, and monetary policy observed
during the U.S. boom of the 1990s were similar
to those of other stock market booms in the United
States and elsewhere during the 20th century.
We are especially interested in whether these
patterns differed across monetary policy regimes.
Monetary neutrality implies that monetary policy
should not affect the price of stocks, which are
claims on real assets, in the long run. Empirical
studies conclude, however, that policy actions
affect stock prices in the short run, and many
researchers contend that the form of policy rule
used by monetary authorities can also affect asset
markets over longer horizons. In particular, some
argue that use of a monetary rule involving a
clearly specified, credible inflation objective
could lessen instability in financial markets,
though others contend that such rules can promote “imbalances” that may lead to financial
instability. This article seeks to identify similarities or differences in the association of monetary
policy and stock market booms across different
monetary policy environments. In so doing, we
hope to gain insight into the role of monetary
policy in supporting or ending asset booms.4
We construct monthly, real (i.e., inflationadjusted) stock price indices for the United States
and nine other countries for which the necessary
data are available over most of the 20th century.
We then identify extended periods of unusually
rapid appreciation in the indices for each country,
which we define as booms. Finally, we use a simple event methodology to examine the behavior
of important macroeconomic and monetary policy
variables during stock market booms, and we
3

For contrasting views on whether the Fed adjusted policy in
response to the stock market during this period, see Cecchetti (2003),
Rigobon and Sack (2003), Hayford and Malliaris (2004), and Meyer
(2004).

4

Bordo and Wheelock (2004) investigate the association of nominal
U.S. stock prices with output, inflation, and money stock growth
over the 19th and 20th centuries. The present article, by contrast,
focuses on periods of rapid appreciation of real stock prices and
compares the U.S. experience with the experiences of other
countries.

92

MARCH/APRIL 2007

compare U.S. experiences with those of the other
countries in our sample. We find that 20th century
stock market booms typically were associated
with the business cycle, arising when output (real
gross domestic product [GDP]) growth was above
average and ending as output growth slowed.
We also find that booms tended to arise when
consumer price inflation was low and end after a
period of monetary policy tightening associated
with an actual or threatened rise in inflation. These
patterns differ little across time and, therefore,
across the different policy regimes in place over
the 20th century. Finally, the patterns we observe
for U.S. stock market booms also appear broadly
similar to those of other countries in our data.
The next section of this article briefly discusses how monetary policy might affect stock
prices. We then present information about the
stock market booms in our data. Subsequent sections examine the macroeconomic conditions
under which 20th century stock market booms
occurred in the United States and other countries.
The final section summarizes our observations
and conclusions.

MONETARY POLICY AND STOCK
PRICES
The basic efficient-markets present-value
model posits that stock prices reflect discounted
expected future dividends and, hence, that price
changes reflect changes in expected dividends
and/or the discount rate (proxied by the real
interest rate). Because stocks are claims on real
assets, monetary neutrality implies that policy
should not affect real stock prices in the long run.
Monetary policy actions might affect stock prices
over shorter horizons, however, by altering the
path of expected dividends, the discount rate, or
the equity premium.5 Early models of the effects
of monetary policy on asset prices focused on
the impact of changes in liquidity on the demand
5

The equity premium is the excess return for holding equities over
short-term debt securities, which in the United States averaged
about 3 percent over the 19th and 20th centuries. The premium
provides compensation for uncertainty about the timing and magnitude of future cash flows associated with ownership of equities
rather than fixed-income securities.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

for various assets that comprise the portfolio of
the private sector. Policy actions that increase
liquidity cause asset prices to rise and returns
to fall as households adjust their portfolios in
response to an increase in central bank liabilities.
Other models focused on the impact of policy on
the cost of capital and, hence, the expected growth
rates of corporate dividends or earnings.6
Several studies have found evidence that
monetary policy actions affect stock prices in
the short run. Bernanke and Kuttner (2005), for
example, estimate that during 1989-2002 an unanticipated 25-basis-point increase in the Federal
Reserve’s target for the federal funds rate produced
a 1 percent decline in equity prices. Further, they
find that the impact of unanticipated monetary
policy actions on stock prices occurred mainly
through their impact on expected future dividends
and excess returns (i.e., the equity premium)
rather than the real interest rate.
Although monetary policy actions appear to
affect stock prices in the short run, many economists contend that the form of policy rule used by
monetary authorities can affect the performance
of asset markets over longer horizons. Some economists argue that monetary policies that result in
persistent or highly variable inflation destabilize
financial markets (e.g., Schwartz, 1995). Rules that
stabilize the price level, however, are commonly
thought to lessen the chance of asset price bubbles
(e.g., Woodford, 2003). Some economists argue,
however, that a commitment to low inflation can
foster imbalances that lead to asset price bubbles
by generating overly optimistic expectations of
future economic growth (e.g., Borio and Lowe,
2002). Federal Reserve Chairman Alan Greenspan
made this claim at a Federal Open Market Committee (FOMC) meeting in 1996:
We have very great difficulty in monetary
policy when we confront stock market bubbles.
That is because, to the extent that we are
successful in keeping product price inflation
down, history tells us that price-earnings ratios
under those conditions go through the roof.
What is really needed to keep stock market
6

See Bordo and Wheelock (2004) for additional discussion and
references.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

bubbles from occurring is a lot of product price
inflation, which historically has tended to
undercut stock markets almost everywhere.
There is a clear tradeoff. If monetary policy
succeeds in one, it fails in the other. (FOMC,
September 24, 1996, pp. 30-31)

Historically, U.S. stock market returns have
been negatively correlated with inflation (Fama
and Schwert, 1977). Goodfriend (2003) argues
that prior to the 1980s, monetary policy was an
important source of both macroeconomic and
financial market instability, which could explain
the negative relationship between stock returns
and inflation. An increase in inflation would tend
to depress stock returns because long-term interest
rates would rise in response to higher expected
inflation and tighter monetary policy and because
tighter policy would also slow economic activity
and thereby reduce current and future corporate
earnings. A reversal of policy in response to a
weak economy and lower inflation would tend
to reduce interest rates and boost stock returns.
Goodfriend (2003) contends that asset price
movements are less likely to be correlated with
policy actions if monetary policymakers are firmly
committed to maintaining price stability. Under
such a regime, he argues, long-term interest rates
will be more firmly anchored and real activity,
corporate profits, and real interest rates will
exhibit less cyclical variability. Hence, under a
policy rule that maintains a stable price level,
movements in asset prices are likely to be less
correlated with specific monetary policy actions.
One objective of this article is to determine
whether stock market booms in the United States
and other countries typically have been associated with low inflation, especially with changes
in monetary policy that foster price stability, and
whether we can observe differences in the relationships over time that might be associated with
differences in policy regimes.

STOCK MARKET BOOMS
There is, of course, no precise definition of
an asset boom, and researchers have imposed a
number of filters to identify specific episodes that
MARCH/APRIL 2007

93

Bordo and Wheelock

Table 1
Stock Market Booms

Country

Boom start:
local market minimum

Boom end:
local market peak

Average annual
% change from
month after trough
to peak

Australia

Dec. 1920

Feb. 1929

10.7

Sept. 1930

Mar. 1937

17.8

Oct. 1934

July 1956

July 1960

15.8

Aug. 1957

Canada

France

Germany

Italy

When prior
25-month peak
surpassed
July 1921

Oct. 1966

Dec. 1969

21.8

Oct. 1967

Aug. 1977

Nov. 1980

21.9

Sept. 1979

July 1982

Sept. 1987

25.2

Mar. 1986

Dec. 1990

Jan. 1994

18.9

Oct. 1993

Aug. 1998

June 2000

13.4

Jan. 1999

Dec. 1920

Sept. 1929

17.4

n/a

June 1932

Mar. 1937

28.0

n/a

Oct. 1953

July 1956

24.6

July 1954

Oct. 1977

Nov. 1980

22.1

n/a

July 1984

July 1987

17.9

July 1985

Jan. 1995

Apr. 1998

19.2

Nov. 1995

Aug. 1998

Aug. 2000

34.7

Dec. 1999

Nov. 1920

July 1924

20.9

n/a

Nov. 1926

Feb. 1929

40.4

Dec. 1927

Dec. 1950

Apr. 1955

28.4

n/a

Aug. 1958

Apr. 1962

20

July 1960

June 1981

Apr. 1987

24.4

Jan. 1985

Feb. 1995

Aug. 2000

23.6

Jan. 1997

June 1957

Sept. 1960

43.6

Aug. 1958

Aug. 1982

Apr. 1986

31.8

July 1983

Mar. 1995

Feb. 2000

23.9

Sept. 1996

May 1932

July 1935

27.5

n/a

July 1950

Sept. 1955

18.5

Aug. 1952

June 1958

Aug. 1960

56.4

Oct. 1958

Dec. 1977

May 1981

35.0

n/a

Dec. 1982

Aug. 1986

38.2

Mar. 1986

Nov. 1995

Feb. 2000

33.6

July 1997

NOTE: *Market decline ended less than 12 months after boom peak; acomparison Jan. 1915–Dec. 1940; bcomparison Jan. 1947–Dec. 2004;
ccomparison Feb. 1920–Dec. 1940; dcomparison Jan. 1920–Dec. 1939; ecomparison Jan. 1950–Dec. 2004; fcomparison Feb. 1921–Dec. 1938;
gcomparison Feb. 1923–Dec. 1940; hcomparison March 1920–Dec. 1939; icomparison Feb. 1917–Dec. 1940; jcomparison Feb. 1916–
Dec. 1939; kcomparison Jan. 1947–Sep. 2004.

94

MARCH/APRIL 2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Months duration
after prior peak
surpassed

Average annual
% change from
month after
prior peak

Comparison
average annual
% change
during period

Percent decline
12 months
after peak

Percent decline
to next minimum

91

9.1

3.6a

–20.1

–41.0

30

13.7

3.6

–12.2

–31.6

35

15.9

2.4b

–11.6

–20.2*

26

17.5

2.4

–24.8

–42.2

14

32.1

2.4

–27.2

–47.2

18

39.3

2.4

–35.8

–46.3*

3

36.9

2.4

–23.8

–23.8

18

6.5

2.4

–0.8

–23.8

n/a

n/a

3.7c

–37.5

–75.1

n/a

n/a

3.7

–35.6

–35.6

25

23.5

3.4b

–9.2

–32.4

n/a

n/a

3.4

–25.2

–52.5

24

15.3

3.4

–18.5

–26.7*

29

20.1

3.4

–10.0

–28.2*

8

42.0

3.4

–36.0

–43.6

n/a

n/a

2.5d

–16.1

–34.7

14

37.9

2.5

–12.0

–57.0

n/a

n/a

2.8b

–17.4

–11.1*

21

14.3

2.8

–18.7

–54.1

25

36.2

2.8

–32.8

–45.0

43

26.8

2.8

–29.5

–60.1

25

54.1

6.0e

–24.0

–49.3

33

28.8

6.0

–18.4

–44.7

41

27.8

6.0

–25.4

–69.9

n/a

n/a

0.4f

–13.4

–20.0*

37

22.7

3.1e

–16.6

–22.2*

23

58.7

3.1

–17.6

–17.6

n/a

n/a

3.1

–46.8

–54.1

5

34.3

3.1

–26.7

–47.9

31

34.9

3.1

–18.8

–56.5

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

95

Bordo and Wheelock

Table 1, cont’d

Country

Boom start:
local market minimum

Boom end:
local market peak

Average annual
% change from
month after trough
to peak

Oct. 1930

Feb. 1934

28.6

Feb. 1932

Jan. 1950

Jan. 1953

54.3

Jan. 1952

Dec. 1957

June 1961

36.3

Oct. 1958

Sept. 1982

Dec. 1989

23.9

Mar. 1983

Japan

Netherlands

Sweden

United Kingdom

United States

96

MARCH/APRIL 2007

When prior
25-month peak
surpassed

July 1924

Feb. 1929

10.9

Jan. 1926

June 1932

Mar. 1937

26.6

n/a

Apr. 1952

June 1957

20.3

Nov. 1954

Dec. 1957

Mar. 1961

22.2

May 1959

Sep. 1981

July 1987

22.0

Mar. 1983

Jan. 1991

Aug. 2000

17.4

June 1993

Mar. 1922

July 1929

16.9

n/a

May 1932

Mar. 1937

23.2

n/a

Mar. 1958

Aug. 1961

15.1

Aug. 1958

Sept. 1980

Mar. 1984

36.8

May 1981

Sept. 1992

Feb. 2000

31.4

Sept. 1995

June 1932

Dec. 1936

15.4

Feb. 1936

June 1952

July 1955

20.0

July 1954

Feb. 1958

Apr. 1961

25.4

Dec. 1958

Sept. 1981

July 1987

21.3

Oct. 1982

June 1994

Dec. 1999

12.6

Apr. 1996

Oct. 1923

Sept. 1929

23.7

Dec. 1924

Mar. 1935

Feb. 1937

39.7

Oct. 1935

Sept. 1953

Apr. 1956

28.8

Mar. 1954

June 1962

Jan. 1966

13.3

Dec. 1963

July 1984

Aug. 1987

22.9

Feb. 1985

Apr. 1994

Aug. 2000

17.1

Mar. 1995

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Months duration
after prior peak
surpassed

Average annual
% change from
month after
prior peak

Comparison
average annual
% change
during period

Percent decline
12 months
after peak

Percent decline
to next minimum

24

26.1

1.8g

–12.0

–16.6

12

93.1

6.9e

–30.0

–36.4

32

38.7

6.9

–23.0

–52.7

81

22.5

6.9

–41.0

–47.6*

36

6.1

–1.8h

–15.7

–71.8

n/a

n/a

–1.8

–18.2

–31.5

31

15.4

4.1b

–19.8

–32.1*

22

15.0

4.1

–10.5

–31.2

52

20.1

4.1

–17.2

–36.7*

74

19.6

4.1

–26.8

–65.4

n/a

n/a

–1.8i

–13.1

–73.3

n/a

n/a

–1.8

–12.2

–12.2

36

13

5.7b

–15.3

–20.2

34

34.7

5.7

–24.6

–29.3

53

30.6

5.7

–31.2

–67.2

10

5.4

–0.4j

–23.6

–44.2

12

16.4

2.8b

–17.3

–31.4

28

19.6

2.8

–17.8

–31.0

57

21.5

2.8

–23.4

–34.8*

44

12.7

2.8

–10.6

–50.2

57

24.4

2.4a

–30.1

–80.6

16

30.2

2.4

–39.0

–45.8

25

29.3

4.4k

–9.6

–20.1

25

10.3

4.4

–12.5

–20.1*

30

21.6

4.4

–22.3

–27.5*

64

18.7

4.4

–22.8

–46.8

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

97

Bordo and Wheelock

they then define as booms. We adapt the methodology of Pagan and Sossounov (2003) to identify
sustained periods of rising real stock prices in the
United States and nine other developed countries (Australia, Canada, France, Germany, Italy,
Japan, the Netherlands, Sweden, and the United
Kingdom).7 First, we calculate a monthly index
of real stock prices for each country by deflating
a nominal stock price index by a consumer price
index. We list our data and sources in the appendix. Next, we identify real stock price index peaks
and troughs within rolling, 25-month windows.
We require that peaks and troughs alternate, and
so eliminate all but the highest maximum that
occurred before a subsequent trough and all but
the lowest minimum that occurred before a subsequent peak. We classify as booms all periods
of at least three years from trough to peak with
an average annual rate of increase in the real stock
price index of at least 10 percent. We also classify
as booms a few episodes of exceptional real stock
price appreciation that were shorter than three
years.8
Table 1 lists the episodes we define as booms
for each country in our sample. For each boom,
we include information about the average annual
percentage increase in the market index from the
market trough to its peak. Because several booms
began as recoveries from market declines, we also
note when the real stock price index surpassed its
prior 25-month peak and report the average annual
percentage increase in the index after that date.
For comparison, Table 1 also reports information about long-run average annual rates of change
in the real stock price index for each country. For
example, the U.S. real stock price index increased
at an average annual rate of 2.4 percent during
1915-40 and 4.4 percent during 1947-2004. Thus,
the periods we define as booms were characterized
7

We selected our sample countries based on the availability of historical data on a stock market index and key macroeconomic series,
which obviously gives rise to possible sample selection bias. We
are unsure of the extent to which our findings would differ if our
sample included recently developed or emerging market economies.

8

Helbling and Terrones (2004) use a similar procedure to identify
booms and busts. Specifically, they identify turning points in the
log-level of real equity prices over five-quarter windows and define
booms (busts) as the largest one-fourth of all price increases
(declines).

98

MARCH/APRIL 2007

by rates of appreciation that were substantially
higher than long-run averages. Finally, Table 1 also
includes information about the extent to which
the real stock price index declined during the 12
months following a market peak and from the
market peak to the next market trough. Almost all
booms were followed by real declines of at least
10 percent within 12 months. Not all booms ended
with a spectacular crash, however, and the lengths
and sizes of market declines after booms varied
widely.
Cross-country comparisons of real stock price
index growth rates are problematic because of
differences in the composition of the stock market
indices of individual countries. For the interwar
period, cross-country comparisons are further
complicated by differences across countries in
(i) the dates when monthly data on a nominal
stock price index and inflation are first available
and (ii) the nature and the availability of stock
price data for the late 1930s associated with when
countries became involved in World War II.
For the post-World War II period, we report
average growth rates for 1947-2004 for all countries in the sample except Germany, Italy, and
Japan, for which we report growth rates over
1950-2004. The real stock price indices for these
three countries exhibit rapid growth during the
1950s compared with average growth rates for subsequent decades. Among the other sample countries, we note considerable variation in average
real stock price growth rates, ranging from 2.4
percent for Australia to 5.7 percent for Sweden.
Again, however, such long-run cross-country
comparisons are problematic because the performance of stock markets varied considerably over
time within countries, as well as because of differences in the coverage of industries and firms in
the stock market indices of individual countries.
Not surprisingly, we find considerable coincidence in the occurrence of stock market booms
across sample countries. For example, most countries experienced a substantial increase in real
stock prices during the 1920s and a market peak
in 1929. Several countries also had booms in the
mid-1930s as their economies climbed out of the
Great Depression. More recently, most countries
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Figure 1
Real GDP Growth Relative to Its Long-Run Average: Post-1970 Booms
Percentage Points
4.0
3.0
U.S. Boom of 2000
2.0
1.0
0.0
–1.0

Median

–2.0
–3.0
–4.0
–20 –18 –16 –14 –12 –10 –8

–6

–4

–2

0

2

4

6

8

10

12

14

16

18

20

Quarters from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

in our sample had booms in the 1980s and again
in the 1990s. Several countries experienced a
market peak within a few months of the U.S. peak
in August 1987; and, among our sample, only
Japan failed to experience a stock market boom
in the 1990s, leading to a peak in 1999 or 2000.9

THE U.S. STOCK MARKET BOOM
OF 1994-2000
This article seeks to discern whether patterns
observed during the U.S. stock market boom of
the 1990s were similar to those observed during
prior booms in the United States and other countries. U.S. stock prices rose rapidly during the
second half of the 1990s, which many analysts
attributed to advances in information-processing
technology and increased productivity growth.
Both current U.S. output (GDP) and productivity
9

See Bordo and Wheelock (2006) for further evidence on the coincidence of stock market booms and correlation of market returns
across countries during the 20th century. See also Goetzmann, Li,
and Rouwenhorst (2001).

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

growth were high during these years, whereas
inflation was low.
Figure 1 plots data on U.S. real GDP growth
relative to its long-run average during the 20 quarters before and after the peak in real stock prices
in the third quarter of 2000 (quarter “0”). Real GDP
growth exceeded its long-run average by approximately 1 percentage point during the 17 quarters
preceding 2000:Q3, then declined sharply as the
U.S. economy entered a recession in 2000:Q4.
Figure 1 also plots the median growth rates of real
GDP (relative to its long-run average) during market peak quarters and in the 20 quarters before and
after market peaks across all post-1970 stock market booms among our sample countries, including
the U.S. boom of 1994-2000.10 Across all booms,
median output growth was much closer to its longrun average than U.S. output growth was during
the 1994-2000 boom. The decline in output after
the “typical” market peak also began later and
was much less steep than was experienced after
10

We define long-run average GDP growth as the average annual rate
during 1960-2001.

MARCH/APRIL 2007

99

Bordo and Wheelock

Figure 2
Labor Productivity Growth Relative to Its Long-Run Average: Post-1970 Booms
Percentage Points
2.5
2.0

U.S. Boom of 2000

1.5
1.0
0.5
0.0
Median

–0.5
–1.0
–1.5
–2.0
–2.5
–5

–4

–3

–2

–1

0

1

2

3

4

5

Years from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

Figure 3
Inflation Relative to Its Long-Run Average: Post-1970 Booms
Percentage Points
6.0

4.0

2.0

0.0

Median

–2.0
U.S. Boom of 2000
–4.0

–6.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

100

MARCH/APRIL 2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

the U.S. stock market peak in 2000:Q3. Figure 1
also displays the mean absolute deviation of real
GDP growth (relative to its long-run average) in
each quarter across all post-1970 booms. Real GDP
growth varied widely during post-1970 booms,
and even more so in the quarters after market
peaks.11
Market analysts frequently attributed the
stock market boom of the late 1990s to advances
in information technology and an increase in
productivity growth that seemed to justify expectations of elevated corporate profits and dividends.
Figure 2 plots data on labor productivity growth
during the U.S. boom of 1994-2000 and the median
across all post-1970 stock market booms.12
Whereas U.S. labor productivity growth exceeded
its long-run average during four of five years
between 1996 and 2000, across all booms, median
productivity growth hovered near its long-run
average.13 Hence, in occurring during a period
of above-average productivity growth, the U.S.
boom of 1994-2000 was somewhat unusual among
recent stock market booms.
While output and productivity growth were
both unusually rapid during the U.S. stock market
boom of the late 1990s, inflation was unusually
low. Consumer price inflation (CPI) hovered
between 2.5 and 3 percent from 1992 to 1996, then
held below 2 percent from late 1997 to early 1999.
Figure 3 plots monthly data on CPI inflation
(relative to its long-run average) during the U.S.
boom of 1994-2000, as well as the median across
all post-1970 booms.14 The figure shows that inflation was below its long-run average throughout
the 60 months preceding the August 2000 peak
in U.S. real stock prices. Further, the figure shows
11

For example, several countries had stock market booms that coincided with the U.S. boom of 1994-2000. Among them, Australia,
Canada, Sweden, and the United Kingdom experienced real GDP
growth rates that were consistently higher than their long-run averages; but France, Germany, and Italy had growth rates that were
near or below average.

12

Throughout the paper, for all figures plotting annual data, we
define the market peak year “0” as the year prior to the actual peak
if the peak occurred in the first half of a year.

13

Here we define the long-run average productivity growth rate as
the average annual growth rate for 1970-2004.

14

Here we define the long-run average inflation rate as the average
rate during 1947-2004.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

a decline in the inflation rate that occurred in 1997
and early 1998 (months “44” to “29”) and an
increase during 1999 and the first half of 2000
(approximately the last 20 months of the boom
period). Across all post-1970 booms, median inflation was below average and declining until some
12 months before a stock market peak month,
when inflation began to rise. Thus, both the U.S.
stock market boom of 1994-2000 and the “typical”
post-1970 boom arose when inflation was below
average and ended after several months of rising
inflation.
The U.S. stock market boom of 1994-2000
attracted considerable attention from Federal
Reserve officials and other policymakers. Fed
officials feared that rapid gains in stock market
wealth would cause rapid growth in spending
and inflation, but officials were perhaps even
more concerned that a sudden decline in the
market could lead to a recession.15 Policymakers
were uncertain about how to respond to the booming stock market, while financial markets were
acutely sensitive to any statements or actions
by the Fed that signaled possible changes in the
direction of policy. Although the Fed was becoming increasingly transparent about its policies,
it neither specified an inflation objective nor
explained how it might react to the booming
stock market.
In December 1996, Federal Reserve Chairman
Greenspan made his famous “irrational exuberance” speech, in which he wondered publicly how
to determine when equity prices are too high in
relation to fundamentals (Greenspan, 1996). Stock
prices fell briefly after the Chairman’s speech on
fears that the Fed would tighten monetary policy
or take other actions to slow the growth of stock
prices. Indeed, at an FOMC meeting in February
1997, Greenspan suggested that the Fed might
want to tighten policy in response to rising stock
prices. He argued that the prevailing level of equity
prices, along with unusually narrow interest rate
credit spreads, “suggest[s] that product prices
alone should not be the sole criterion [for conducting monetary policy] if we are going to maintain
15

See Meyer (2004) for an interesting account of Federal Reserve
policymaking during this period.

MARCH/APRIL 2007

101

Bordo and Wheelock

Figure 4
Short-Term Interest Rate Relative to Its Rate in Peak Month: Post-1970 Booms
Percentage Points
3.0
2.0
1.0
Median

0.0
–1.0
–2.0

U.S. Boom of 2000

–3.0
–4.0
–5.0
–6.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

a stable, viable financial system whose fundamental goal…is the attainment of maximum sustainable economic growth” (FOMC, February
4-5, 1997, p. 103).
The FOMC increased its federal funds rate
target by 25 basis points in March 1997, but then
left the target unchanged over the remainder of
the year. Inflation was falling, which puzzled Fed
officials who struggled to understand the decline
amid rapid economic growth and falling unemployment.16 Chairman Greenspan was an early
proponent of the view that advances in information processing technology had increased the
potential growth rate of output, but most Fed
officials and staff were skeptical.17 Despite their
misgivings about the stock market, however, Fed
officials chose not to raise their funds rate target
as long as inflation continued to fall. Of course,
in not cutting their target, Fed officials permitted
the (ex post) real funds rate to rise as the inflation
rate fell.

Figure 4 plots data on the federal funds rate
during the U.S. stock market boom of 1994-2000,
and the median level of short-term interest rates
across all post-1970 booms in our dataset.18 Both
the funds rate and the median are shown relative
to their levels in the months of stock market peaks
(month “0”). Figure 5 plots the level of the real
interest rate, defined as the nominal short-term
interest rate minus the trailing year-over-year
inflation rate, during the U.S. stock market boom
of 1994-2000 and the median level across all
booms. Finally, Figure 6 plots the spread between
the yield on long-term Treasury securities and
the short-term interest rate for boom periods. The
real interest rate and the term spread are two measures that economists often monitor to gauge the
stance of monetary policy.
Figures 5 and 6 show that the real funds rate
rose and the term spread fell during 1997 (months
“–43” to “–32”), and, hence, by these measures,
18

16

Meyer (2004, pp. 79-80).

17

Meyer (2004, pp. 80-84, 123-25).

102

MARCH/APRIL 2007

For countries for which data are available, we use an overnight
interest rate similar to the U.S. federal funds rate. Otherwise, we use
another short-term money market interest rate. See the appendix
for details.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Figure 5
Real Interest Rate: Post-1970 Booms
Percentage Points
8.0

6.0

4.0
Median
2.0

0.0
U.S. Boom of 2000
–2.0

–4.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

Figure 6
Interest Rate Term Spread: Post-1970 Booms
Percentage Points
4.0

3.0

2.0

1.0
Median
0.0

–1.0

U.S. Boom of 2000

–2.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

103

Bordo and Wheelock

monetary policy tightened. Concern about stock
market speculation and rapid economic growth
kept the Fed from cutting its funds rate target,
even as inflation fell, which effectively allowed
policy to tighten.
The next major move in monetary policy came
in response to a sudden demand for liquidity in
the wake of a Russian government bond default
and spreading financial crisis in Asia during the
summer and fall of 1998. The Fed cut its funds
rate target by 75 basis points between September
and November 1998 (months “–23” and “–21”),
and the real interest rate fell. The spread between
long-term Treasury security yields and short-term
interest rates also rose as the demand for liquidity
abated.
Inflation began to rise in 1999; citing “a significant risk of rising inflation,” the FOMC began to
raise its federal funds rate target in June (Board
of Governors of the Federal Reserve System, 1999,
p. 242). Over the subsequent year, the Committee
increased its target by a total of 175 basis points.19
For the most part, however, increases in the federal funds rate merely kept pace with the rising
inflation rate, which left the real interest rate
essentially unchanged. Fed officials sought to
contain inflation throughout the period, but
resisted the temptation to increase the funds rate
faster than the inflation rate because they desired
to accommodate a perceived increase in the potential growth rate of the economy associated with
higher productivity growth. Fed officials also
worried that aggressive tightening could cause a
sharp decline in the stock market and a substantial slowing of economic activity.20
By mid-2000, Fed officials had decided to
act more aggressively against inflation. At their
May 2000 meeting, FOMC members concluded
that demand growth was continuing to exceed
even the increased rate of potential output growth
and that more aggressive tightening was necessary:
“A more forceful policy move...was desirable in
light of the extraordinary and persisting strength
19

The target had been reduced from 5 percent to 4.75 percent on
November 17, 1998. The target was raised to 5 percent on June 30,
1999, and elevated in five more steps to 6.5 percent as of May 16,
2000.

20

Meyer (2004, pp. 162-63).

104

MARCH/APRIL 2007

of overall demand, exceeding even the increasingly rapid growth of potential supply” (Board of
Governors of the Federal Reserve System, 2000,
p. 224). Thus, rather than increasing their funds
rate target by the usual 25-basis-point increment,
Fed officials voted to raise their target by 50 basis
points to 6.5 percent, where it remained throughout the rest of 2000. As shown in Figures 5 and 6,
the real interest rate rose and the term spread
declined sharply during the six months preceding
the August 2000 peak in real stock prices. Hence,
by conventional measures, the stock market boom
ended after several months of increasingly tighter
monetary policy.21
Figures 5 and 6 also plot the median real
interest rate and term spread levels across all post1970 booms in our dataset. The pattern followed
by the median real rate level is similar to that of
the real federal funds rate during the U.S. boom
of 1994-2000: After peaking near 4 percent some
24 months before a stock market peak, the median
real rate fell approximately 1 percentage point
before rising again during the year preceding the
market peak. The median term spread does not,
however, exhibit the decline observed in the U.S.
term spread during the last months of the U.S.
boom of 1994-2000. Nevertheless, it appears that,
like the U.S. stock market boom of 1994-2000, the
end of the “typical” post-1970 boom followed
some tightening of monetary policy associated
with rising inflation.

MONETARY POLICY AND STOCK
MARKET BOOMS BEFORE 1970
Next we examine the economic and monetary
policy conditions under which stock market
booms occurred earlier in the 20th century. The
historical approach enables us to examine the
association of booms with macroeconomic con21

The specific ending date of the boom is ambiguous. Although we
date the end of the boom as the month that the inflation-adjusted
S&P 500 composite index reached its peak (based on monthly averages of daily closing values), the peak in the nonadjusted index
occurred on March 24. The dates on which other indices peaked
include the Dow Jones industrial average on January 14, the New
York Stock Exchange composite index on September 1, and the
NASDAQ composite index on March 10.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Figure 7
Real GDP Growth Relative to Its Long-Run Average: Prewar Booms
Percentage Points
15.0
10.0

5.0

Median

0.0
–5.0
U.S. Boom of 1929
–10.0

–15.0
–20.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

6

7

8

Years from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

ditions and monetary policy under different
monetary and financial regulatory regimes.

Pre-World War II Booms
The stock market booms of 1923-29 and 19942000 stand out among all U.S. booms in terms of
their length and the extent to which stock prices
rose, and they have often been compared to one
another. Like the 1990s boom, the U.S. boom of
1923-29 arose during a period of above average
economic growth and low inflation. As with the
recent boom, in the 1920s, many analysts attributed the booming stock market to advances in
technology and business management techniques
that promised rapid growth of economic activity
and corporate profits.22 Technological breakthroughs of the late-19th and early-20th centuries,
especially in electric power distribution and
motors, were widely adopted by American industry in the 1920s (David, 1990). Productivity growth
22

See Bordo and Wheelock (2006) and White (2006) and the references therein.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

increased sharply, especially in the manufacturing sector. For the private domestic economy as
a whole, total factor productivity and labor productivity grew at average annual rates of 2.0 percent and 2.2 percent, respectively, during 1919-29,
compared with rates of 1.1 percent and 1.5 percent
during 1909-19 and 1.6 and 1.8 percent during
1929-37 (Kendrick, 1961, p. 72).23
Figure 7 plots U.S. real GDP growth relative
to its long-run average during 1929 (year “0”) and
the 16 surrounding years.24 The figure also plots
median real GDP growth (relative to its long-run
average) across all prewar booms in our dataset,
including the U.S. boom of 1923-29. U.S. output
growth was especially rapid at the start of the
23

Annual data suggest, however, that productivity growth was slower
toward the end of the 1920s when the stock market boom was in
full swing than it had been earlier in the decade. Total factor productivity growth and labor productivity growth averaged 2.6 and
3.0 percent during 1920-24 and 1.3 and 1.7 percent during 1925-29
(Kendrick, 1961, Table A-XXII). Productivity change is, however,
more correctly measured between similar points in the business
cycle.

24

We define the long-run average real GDP growth rate as the average
annual rate for 1871-1939.

MARCH/APRIL 2007

105

Bordo and Wheelock

Figure 8
Inflation Rate Relative to Its Long-Run Average: Prewar Booms
Percentage Points
6.0
4.0
2.0
0.0
–2.0
–4.0
Median
–6.0
U.S. Boom of 1929

–8.0
–10.0
–12.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

6

7

8

Years from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

Figure 9
Commercial Paper Rate: 1920s Booms
Percentage Points
8.0
7.0

U.S. Boom of 1929

6.0
5.0
4.0
3.0
Median
2.0
1.0
0.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

106

MARCH/APRIL 2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

boom in 1923 (year “–6”) and also in 1929 (year
“0”), when output growth exceeded its long-run
average by 2.8 percentage points. Output growth
was below its long-run average, however, in 1927
and 1928.25 By contrast, median output growth
across all booms exceeded its long-run average
during both market peak years and the three years
preceding the peak.
Figure 8 shows the behavior of inflation
during the U.S. boom of 1923-29, as well as the
median inflation rate across all prewar booms in
our dataset. U.S. inflation was below average during 1926-28 and approximately equal to its longrun average during 1929. The modest increase in
the inflation rate during the last year of the boom
is similar to the pattern observed during the U.S.
boom of 1994-2000 and in the median across all
post-1970 booms. Figure 8 also shows an increase
in the median rate of inflation toward the end of
prewar booms in general. Thus, like the typical
post-1970 stock market boom, the typical prewar
boom arose when inflation was below average
and ended within a year or two of higher inflation.
The Federal Reserve System was established
in 1914, and monetary policy was still in its
infancy during the 1920s. World War I disrupted
the international gold standard, but the United
States only briefly suspended gold payments
during the war. The Federal Reserve Act required
the System to maintain a gold reserve, but by the
early 1920s, the Fed’s gold reserves were sufficient
to allow policymakers to pursue discretionary
monetary policy. Fed officials successfully resisted
attempts by Congress to impose an explicit inflation objective on the Fed, and Fed officials made
few public statements about their policy objectives
or tactics. The Fed pursued a strategy aimed at
manipulating bank reserves and market interest
rates to achieve an evolving set of objectives,
which by 1928 included control of the booming
stock market.26
25

Throughout this article we use annual real GDP data from Maddison
(2003), which are adjusted to be comparable across countries.
Quarterly estimates of U.S. real gross national product from Balke
and Gordon (1986) indicate that output growth exceeded its long-run
(i.e., 1875-1939) average by 0.9 percentage points during 1923:Q1–
1929:Q3 and by 3.5 percentage points during 1928:Q1–1929:Q3.

26

See Chandler (1958), Friedman and Schwartz (1963), Meltzer (2003),
Wheelock (1991), and Wicker (1966) for discussion and evidence
on the Fed’s policy objectives and strategy during the 1920s.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

The Federal Reserve tightened monetary
policy aggressively in 1928-29, prompted by the
rapid rise in stock prices and a perception that
Federal Reserve credit was being used to finance
speculative activity. Fed officials viewed speculation in stocks, commodities, and other assets
as a manifestation of inflation that called for a
tightening of credit conditions.
Figure 9 plots the interest rate on commercial
paper of four- to six-month maturity during the
stock market boom of 1923-29, as well as the
median level of short-term interest rates across
all booms of the 1920s in our data set.27 The figure
shows a sharp increase in U.S. short-term interest
rates during the 18 months ending in September
1929, coinciding with the Fed’s tightening. The
median interest rate level across all booms rose
much less.
Figure 10 plots the real commercial paper
interest rate during the stock market boom of
1923-29 and the median level of interest rates
across all booms of the decade. The U.S. real rate
increased from an average of 4.7 percent during
1926-27 to an average of 6.0 percent between
January 1928 and September 1929. The median
across all 1920s booms followed a somewhat
different pattern, first declining some 18 months
before the stock market peak, then rising in the
six months before the market peak.28
The Fed’s monetary policy tightening of 192829 is also evident in the behavior of the money
stock. Figure 11 plots the growth rate of the money
stock relative to its long-run average during the
U.S. boom of 1923-29 and the median growth
rate across all prewar booms.29 U.S. money stock
growth fell below its long-run average in 1926
and trended downward to a low point in 1932.
27

We lack commercial paper rate data for Australia and Canada for
the prewar period.

28

We do not include a figure with the term spread because market
yield data on both a short- and long-term government security are
not available for many countries during the 1920s. U.S. short-term
interest rates rose relative to long-term rates over the 18 months
ending in September 1929, with the commercial paper rate rising
from about 100 basis points above the long-term U.S. Treasury bond
yield to about 250 basis points above the Treasury yield. Hence,
the yield curve became increasingly inverted.

29

Here we define the long-run average money stock growth rate as
the average rate during 1881-1939.

MARCH/APRIL 2007

107

Bordo and Wheelock

Figure 10
Real Commercial Paper Rate: 1920s Booms
Percentage Points
16.0

12.0

U.S. Boom of 1929

8.0

4.0
Median
0.0

–4.0

–8.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

Figure 12 shows that the growth of the real money
stock also declined relative to its long-run average,
especially in 1929. By contrast, median nominal
and real money stock growth rose during the year
of market peaks, as Figures 11 and 12 also show.
Thus, the Fed’s policy tightening over the 18
months before the stock market peak in September
1929 was considerably more aggressive than the
tightening that occurred toward the end of the
typical prewar boom.
Although the Federal Reserve provided substantial liquidity following the October 1929 stock
market crash, monetary conditions tightened
again in 1930 and the U.S. economy plunged
into a depression. Real interest rates soared (see
Figure 10), and both the nominal and real money
stocks collapsed (see Figures 11 and 12). Real GDP
and the price level both fell sharply (see Figures 7
and 8). Although concerned about the economy,
the Federal Reserve failed to mount an aggressive
response to the Depression, in part because policymakers were fearful of reigniting stock market
speculation.30 In the event, the real value of the
Standard and Poor’s composite stock price index
108

MARCH/APRIL 2007

fell some 80 percent from its 1929 peak to its low
in 1932.
Many countries experienced significant
declines in economic activity and stock prices
during the Great Depression. Several, including
the United States, also experienced a stock market
boom as their economies recovered. In many
countries, recovery began when their currency
was devalued or the country abandoned the gold
standard (Eichengreen, 1992). Stock markets
recovered and boom periods were characterized
by rapid output and money stock growth and
moderate inflation.
U.S. stock prices rose rapidly during the mid1930s: The Standard and Poor’s composite index
rose at an inflation-adjusted rate of nearly 40 percent per year between March 1935 and February
1937. The boom ended abruptly in early 1937,
however, and the U.S. economy entered a recession. Once again, the end of the boom coincided
with a tightening of monetary conditions.
30

See Friedman and Schwartz (1963), Meltzer (2003), and Wheelock
(1991 and 1992) and the references therein for explanations of the
Fed’s behavior during the Great Depression.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Figure 11
Money Stock Growth Relative to Its Long-Run Average: Prewar Booms
Percentage Points
15.0
10.0
5.0
0.0
–5.0

Median

–10.0
U.S. Boom of 1929
–15.0
–20.0
–25.0
–30.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

6

7

8

Years from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

Figure 12
Real Money Stock Growth Relative to Its Long-Run Average: Prewar Booms
Percentage Points
10.0

5.0
Median
0.0

–5.0

–10.0
U.S. Boom of 1929

–15.0

–20.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

6

7

8

Years from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

109

Bordo and Wheelock

The Federal Reserve had largely stayed on
the sidelines as the U.S. economy pulled out of
the Depression and the pace of economic activity
accelerated. As the recovery continued, however,
Fed officials became increasingly concerned about
the potential for inflation. Beginning in 1933, gold
inflows caused bank reserves and the money
stock to grow rapidly, and banks built up huge
stocks of reserves in excess of legal requirements.
Fed officials feared that the accumulation of
excess reserves posed an inflationary threat and
took a series of steps to reduce them: The Federal
Reserve Board increased reserve requirements
by 50 percent in August 1936; the Treasury
Department began to neutralize gold inflows in
December 1936; and the Fed hiked reserve requirements again on January 30, 1937. Following these
actions, money stock growth slowed (from 13
percent in 1936 to 4 percent in 1937), interest
rates rose, and the stock market peaked and began
to fall as the boom ended.

Early Postwar Booms
We examine stock market booms of the 1950s
and 1960s separately from those of the 1970s to
1990s because of sharp differences in the regulatory and monetary regimes that prevailed in the
two periods. In addition, for some countries, highfrequency economic and financial data are not
available for the 1950s and 1960s.
Many countries adopted new regulations on
financial markets and international capital flows
in response to the financial disruptions of the
Great Depression. The Great Depression also effectively ended the international gold standard, as
countries either abandoned the standard altogether
or imposed exchange controls that limited its functioning (Eichengreen, 1992). World War II brought
even tighter controls, especially in Europe, that
included restrictions on the issuance of private
securities and the movement of capital across
international borders. The postwar international
monetary system was defined by the Bretton
Woods system of fixed exchange rates and capital
controls. Wartime controls were gradually relaxed
over time, but deregulation was protracted. The
pace of deregulation quickened in the 1970s and
1980s as countries sought to keep their financial
110

MARCH/APRIL 2007

markets competitive in the face of advances in
information-processing technology that encouraged financial innovation and globalization. At
the same time, the international monetary regime
changed dramatically with the collapse of the
Bretton Woods system in 1971.31
The United States experienced two stock market booms during the 1950s and 1960s, one ending
in April 1956 and another ending in January 1966.
Figure 13 plots data on U.S. real GDP growth relative to its long-run average during these two
episodes. The figure also plots the median growth
rates across all booms of the 1950s and 1960s in
our data set, as well as the mean absolute deviation of observations around the median.
U.S. output growth was highly variable during the boom of 1953-56. As shown in Figure 13,
real GDP contracted by almost 1 percent in 1954
(year “–1”), but expanded by nearly 7 percent in
1955 (year “0”).32 Output growth was less variable
during the stock market boom of 1962-66 and
exceeded its long-run average throughout the
period. Output growth exceeded its long-run
average by nearly 3 percentage points in 1965
(year “0”) and continued to grow rapidly after the
peak in the real stock price index in January 1966.
Median real GDP growth exceeded its long-run
average across all booms of the 1950s and 1960s,
and rose during the final two years of booms. Output growth was unusually rapid during the 1950s
and 1960s, especially among European countries
and Japan. Whereas U.S. output growth fell below
its long-run average within a year of the U.S. stock
market peaks in April 1956 and January 1966,
median output growth remained above its longrun average after the “typical” stock market boom
of the 1950s and 1960s had ended.33
31

See Bordo and Wheelock (2006) for additional discussion and
evidence on changes over time in the coincidence of stock market
booms and correlation of stock returns across our sample countries.
See also Goetzmann, Li, and Rouwenhorst (2001) and Obstfeld
and Taylor (1998) for evidence on the international integration of
financial markets throughout the 20th century.

32

As noted previously, in figures that present annual data, we define
year “0” as the year prior to the actual market peak if the peak
occurred during the first half of a calendar year.

33

On the “Golden Age” of European economic growth before 1973,
see Crafts and Toniolo (1996).

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Figure 13
Real GDP Growth Relative to Its Long-Run Average: 1950s-60s Booms
Percentage Points
6.0

4.0

2.0

0.0

–2.0

Median
U.S. Boom of 1956

–4.0

U.S. Boom of 1966
–6.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

6

7

8

6

7

8

Years from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

Figure 14
Inflation Rate Relative to Its Long-Run Average: 1950s-60s Booms
Percentage Points
6.0
Median
U.S. Boom of 1956
U.S. Boom of 1966

4.0

2.0

0.0

–2.0

–4.0

–6.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

Years from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

111

Bordo and Wheelock

Figure 15
Short-Term Interest Rate Relative to Its Rate in Peak Month: 1950s-60s Booms
Percentage Points
5.0
4.0
3.0
2.0
1.0
0.0
–1.0
–2.0

Median
U.S. Boom of 1956

–3.0

U.S. Boom of 1966
–4.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

Figure 14 plots inflation rates (relative to their
long-run averages) during the two U.S. stock
market booms of the 1950s and 1960s, alongside
median inflation across all booms of the period
and the mean absolute deviation of observations
around the median. As during both the prewar
and post-1970 periods, during the 1950s and
1960s stock market booms typically arose when
inflation was below its long-run average and ended
after inflation had risen. The U.S. booms of the
1950s and 1960s were no exception. After a burst
of inflation during the Korean War, the U.S. inflation rate stayed below 1 percent per year throughout 1952-55, before rising to 3 percent in 1956-57.
Inflation then fell back and remained below 2
percent in each year from 1958 to 1965. Inflation
rose again in 1965, however, and reached 3.4
percent in 1966.
Resembling the patterns of both prewar and
post-1970 booms, the increases in inflation before
the U.S. stock market peaks in 1956 and 1966
were accompanied by higher interest rates and
other evidence of monetary tightening. Figure 15
plots data on short-term interest rates during the
two U.S. booms and the median across all booms
112

MARCH/APRIL 2007

of the 1950s and 1960s.34 The figures show the
level of the interest rate in each month relative
to its level in the month when the real stock price
index reached its peak (month “0”).
Short-term interest rates rose some 2 percentage points over the 24 months ending in April
1956, reflecting actions by the Federal Reserve to
ward off inflation and curb the flow of credit to
the stock market. The Fed began a series of tightening steps in 1954 that included open-market
operations and hikes in the discount rate and
margin requirements.35 Fed officials continued
to focus on inflation during 1955 and 1956 and
frequently discussed the importance of preventing
inflation from rising. For example, at an FOMC
34

When available, we use an overnight interest rate, such as the U.S.
federal funds rate, in constructing the interest rate series for this
period. Otherwise, we use a short-term Treasury security yield.
Monthly data on the federal funds rate are not available until July
1954, for example, and so we use the yield on 3-month Treasury
bills for earlier months.

35

The Securities Exchange Act of 1934 empowered the Federal
Reserve Board to regulate margin requirements on loans granted
by banks and securities firms for the purpose of purchasing or
owning stocks. The margin requirement establishes the minimum
percentage of a stock purchase that must be self-financed rather
than financed by borrowing.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Figure 16
Real Short-Term Interest Rate: 1950s-60s Booms
Percentage Points
4.0
3.0
2.0
1.0
0.0
–1.0
Median
U.S. Boom of 1956

–2.0

U.S. Boom of 1966
–3.0

–60 –56 –52 –48 –44 –40 –36 –32 –28 –24 –20 –16 –12 –8 –4

0

4

8 12 16 20 24 28 32 36 40 44 48 52 56 60

Months from Market Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

meeting in January 1955, Chairman William
McChesney Martin Jr. stated that “[w]hat we are
wrestling with at the moment is the possibility
that inflationary seeds may be germinating, and
that when they come to full bloom it will be difficult to restrain them...We want to nip inflation
in the bud” (Federal Open Market Committee,
January 11, 1955, pp. 7-8). Martin also expressed
concern about the booming stock market, arguing
that “we ought to be considering the possibility
of another signal to the stock market either through
a further increase in the margin or, preferably
through the discount rate” (p. 9).
Fed officials expressed concern about the
booming stock market throughout 1955, occasionally associating movements in stock prices
with general price inflation. For example, at an
FOMC meeting in August, Chairman Martin commented that “all danger signals are now flashing
red. Inflation is a thief in the night and if we don’t
act promptly and decisively we will always be
behind...A move such as we had in General
Motors (stock) of fifteen points in one day would
be disastrous if it developed over the whole price
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

level” (Federal Open Market Committee, August 2,
1955, p. 13).
In the event, inflation remained low throughout 1954 and 1955 and, consequently, the increase
in short-term interest rates during these years
resulted in a similar-sized increase in the real
interest rate, as shown in Figure 16. Money stock
growth also slowed, as shown in Figure 17, though
real money stock growth remained slightly above
its long-run average, as shown in Figure 18.36
Monetary conditions did not tighten as
sharply before the U.S. stock market peak in
January 1966 as they had before the April 1956
peak. As shown in Figure 15, the federal funds
rate rose by approximately 0.5 percentage points
36

Calomiris and Wheelock (1998) note a close, negative correlation
between the level of free reserves and inflation during the 1950s.
Further, they note that growth of the money stock (M1) was positively correlated with the level of free reserves during the 1950s,
but not during the 1960s, which could explain the Fed’s apparent
success in maintaining low inflation during the 1950s, but not during the 1960s. Romer and Romer (2002) find that Fed policy was
consistent with the “Taylor principle” during the 1950s, but not
during the 1960s, in that movements in the real interest rate were
sufficient to stabilize inflation during the 1950s, but not during
the 1960s.

MARCH/APRIL 2007

113

Bordo and Wheelock

Figure 17
Money Stock Growth Relative to Its Long-Run Average: 1950s-60s Booms
Percentage Points
6.0

Median
U.S. Boom of 1956
U.S. Boom of 1966

4.0

2.0

0.0

–2.0

–4.0

–6.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

6

7

8

Years from Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

Figure 18
Real Money Stock Growth Relative to Its Long-Run Average: 1950s-60s Booms
Percentage Points
8.0
6.0
4.0
2.0
0.0
–2.0
–4.0
Median
U.S. Boom of 1956

–6.0

U.S. Boom of 1966
–8.0
–8

–7

–6

–5

–4

–3

–2

–1

0

1

2

3

4

5

6

7

8

Years from Peak

NOTE: The shaded area comprises the median  the mean absolute deviation.

114

MARCH/APRIL 2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

in late 1964 (months “–15” to “–13”), but changed
little during 1965. The real funds rate, shown in
Figure 16, peaked in early 1965 (month “–11”),
then declined in the second quarter when inflation
began to rise. The real funds rate then changed
little over the remainder of the year. Further, both
nominal and real money stock growth remained
above their long-run average rates during 1965
(year “0” in Figures 17 and 18).
The January 1966 stock market peak did,
however, occur shortly after a highly publicized
monetary policy action. The increase in inflation
in the second quarter of 1965 persisted through
the remainder of the year, and by the fourth quarter Fed officials were convinced that monetary
policy had to tighten. Memoranda from the FOMC
meeting in November state that Chairman Martin
argued that “the country was in a period of creeping inflation...In short, he thought the economy
was growing too fast at the moment” (FOMC,
November 23, 1965, p. 85). Fed officials then took
steps to tighten, including a highly publicized
discount rate increase in early December that
sparked a sharp rebuke from President Johnson.37
The stock market peak occurred shortly thereafter,
and the boom of 1962-66 was over.
Other countries that had booms during the
1950s and 1960s experienced interest rate and
money stock growth patterns that were similar to
those of the two U.S. booms. The median level of
short-term interest rates across all booms rose by
1 percentage point during the 8 to 15 months
before stock market peaks. The median real interest rate level fluctuated widely, with little trend
during the 24 months before market peaks; but
both nominal and real money stock growth
declined during the last two years of booms,
indicating that some monetary policy tightening
preceded the end of the typical boom.

of output and productivity and low inflation.
Technological advances and higher productivity
growth convinced many observers that corporate
profits would continue to grow rapidly and justify
soaring equity prices. Further, shrinking government budget deficits and low inflation suggested
that interest rates would remain low. Eventually,
however, inflation began to rise, monetary policy
tightened, and the boom ended.
Our review of earlier stock market booms in
the United States and nine other developed
countries during the 20th century indicates that
the patterns observed during the U.S. boom of
1994-2000 were similar to those of earlier booms.
Stock market booms typically arose when output
growth exceeded its long-run average and when
inflation was below its long-run average. There
were, however, exceptions. Notably, we find that
across all post-1970 booms the median growth
rates of real GDP and productivity did not substantially exceed their long-run averages.38 We
find less variation in the association of booms with
low inflation than we do in the association of
booms with rapid output or productivity growth.
Further, we find that both nominal and real money
stock growth were typically below average during booms, suggesting that booms did not result
from excessive liquidity.39
We find that 20th century stock market booms
often ended following an increase in inflation
and a tightening of monetary conditions. All U.S.
booms ended after explicit tightening by the
Federal Reserve in response to actual or threatened inflation. The Fed tightened policy in 192829 because policymakers believed that asset price
appreciation was a form of inflation that required
an aggressive response. During subsequent booms,
38

Bordo and Wheelock (2006) speculate that increased financial
globalization since the 1970s may have weakened the connection
between stock market performance and domestic output growth in
some countries. Further, among European countries, stock market
performance in the 1990s may have been heavily influenced by
steps taken to integrate national economies, especially monetary
union.

39

Some analysts have argued that asset booms reflect excessive
growth of credit. We have been unable to locate data on credit that
are comparable across all of the countries in our sample, especially
for earlier periods. A cursory review of what data we have obtained,
however, shows no consistent association of booms with credit
growth.

OBSERVATIONS AND
CONCLUSIONS
The U.S. stock market boom of 1994-2000
arose during a period of unusually rapid growth
37

See Maisel (1973, pp. 69-77).

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

115

Bordo and Wheelock

the Fed’s principal goal was to halt an incipient
rise in consumer price inflation, though concern
about stock market speculation appears to have
been a secondary reason for tightening in some
cases.
Although the U.S. monetary policy regime
changed substantially over time, we find little
variation in the association of stock market booms
with inflation or in the end of booms with monetary policy actions to control inflation. Even under
the gold standard of the 1920s, however, the Fed
had considerable latitude in pursuing discretionary monetary policy and conceivably the absence
of a clear statement of objectives contributed to
instability in asset markets. Recent research suggests that the form of rule used by monetary
authorities, including its communication strategies, can influence how policy actions affect economic activity.40 Additional research is needed,
however, to determine whether a clear statement
of objectives and strategy to achieve those objectives would alter the association of asset booms
with low inflation or the effect that policy actions
to control inflation have on asset markets.

REFERENCES

Bordo, Michael; Eichengreen, Barry; Klingebiel,
Daniela and Martinez-Peria, Maria Soledad. “Is the
Crisis Problem Growing More Severe?” Economic
Policy, April 2001, 16(32), pp. 51-82.
Bordo, Michael; Erceg, Christopher; Levin, Andrew
and Michaels, Ryan. “Three Great American
Disinflations.” Unpublished manuscript, August
2005.
Bordo, Michael and Wheelock, David C. “Monetary
Policy and Asset Prices: A Look Back at Past U.S.
Stock Market Booms.” Federal Reserve Bank of St.
Louis Review, November/December 2004, 86(6),
pp. 19-44.
Bordo, Michael and Wheelock, David C. “When Do
Stock Market Booms Occur? The Macroeconomic
and Policy Environments of 20th Century Booms.”
Working Paper No. 2006-051A, Federal Reserve
Bank of St. Louis, September 2006.
Borio, Claudio and Lowe, Philip. “Asset Prices,
Financial and Monetary Stability: Exploring the
Nexus.” Working Paper No. 114, Bank for
International Settlements, July 2002.

Balke, Nathan and Gordon, Robert J. “Appendix:
Historical Data,” in Robert J. Gordon, ed., The
American Business Cycle: Continuity and Change.
Chicago: University of Chicago Press, 1986, pp.
781-850.

Bryan, Michael F.; Cecchetti, Stephen G. and
O’Sullivan, Roisin. “Asset Prices in the Measurement of Inflation.” NBER Working Paper No. 8700,
National Bureau of Economic Research, January
2002.

Bernanke, Ben S. and Kuttner, Kenneth N. “What
Explains the Stock Market’s Reaction to Federal
Reserve Policy?” Journal of Finance, June 2005,
60(3), pp. 1221-57.

Calomiris, Charles W. and Wheelock, David C. “Was
the Great Depression a Watershed for American
Monetary Policy?” in Michael D. Bordo, Claudia
Goldin, and Eugene N. White, eds., The Defining
Moment: The Great Depression and the American
Economy in the Twentieth Century. Chicago:
University of Chicago Press, 1998, pp. 23-65.

Board of Governors of the Federal Reserve System.
“Minutes of the Federal Open Market Committee,
May 16, 2000.” 87th Annual Report, 2000.
Board of Governors of the Federal Reserve System.
“Minutes of the Federal Open Market Committee,
June 29-30, 1999.” 86th Annual Report. 1999.
40

Bordo et al. (2005), for example, examine three major U.S. disinflations since the Civil War and conclude that both policy institutions and central bank communication strategies affect expectations
of disinflation and subsequent economic activity.

116

MARCH/APRIL 2007

Cecchetti, Stephen G. “What the FOMC Says and
Does When the Stock Market Booms,” in A. Richards
and T. Robinson, eds., Asset Prices and Monetary
Policy: Proceedings of the Annual Research
Conference of the Reserve Bank of Australia.
Sydney, Australia: Reserve Bank of Australia,
November 2003, pp. 77-96.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

Chandler, Lester V. Benjamin Strong, Central Banker.
Washington, DC: Brookings Institution, 1958.
Crafts, Nicholas, and Toniolo, Gianni. Economic
Growth in Europe Since 1945. Cambridge:
Cambridge University Press, 1996.
David, Paul A. “The Dynamo and the Computer: An
Historical Perspective on the Modern Productivity
Paradox.” American Economic Review, May 1990,
80(2), pp. 355-61.
Eichengreen, Barry. Golden Fetters: The Gold Standard
and the Great Depression, 1919-1939. New York:
Oxford University Press, 1992.
Fama, Eugene F. and Schwert, G. William. “Asset
Returns and Inflation.” Journal of Financial
Economics, November 1977, 5(2), pp. 115-46.
Federal Open Market Committee. Minutes of the
Federal Open Market Committee. Various meetings.
Friedman, Milton and Schwartz, Anna J. A Monetary
History of the United States, 1867-1960. Princeton:
Princeton University Press, 1963.
Goetzmann, William N.; Li, Linfeng and Rouwenhorst,
K. Geert. “Long-Term Global Market Correlations.”
NBER Working Paper No. 8612, National Bureau of
Economic Research, November 2001.
Goodfriend, Marvin. “Interest Rate Policy Should
Not React Directly to Asset Prices,” in William C.
Hunter, George G. Kaufman, and Michael
Pomerleano, eds., Asset Price Bubbles: The
Implications for Monetary, Regulatory, and
International Policies. Cambridge, MA: MIT Press,
2003, pp. 445-57.
Goodhart, Charles A. and Hofmann, Boris. “Do Asset
Prices Help Predict Consumer Price Inflation?”
Manchester School, 2000, 68(Suppl.), pp. 122-40.
Greenspan, Alan. “The Challenge of Central Banking
in a Democratic Society.” Remarks at the Annual
Dinner and Francis Boyer Lecture of the American
Enterprise Institute for Public Policy Research,
Washington, DC, December 5, 1996.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Hayford, Marc D. and Malliaris, A.G. “Monetary
Policy and the U.S. Stock Market.” Economic
Inquiry, July 2004, 42(3), pp. 387-401.
Helbling, Thomas and Terrones, Marco. “Asset Price
Booms and Busts—Stylized Facts from the Last
Three Decades of the 20th Century.” Working paper,
International Monetary Fund, March 2004.
Kendrick, John W. Productivity Trends in the United
States. Princeton: Princeton University Press, 1961.
Kohn, Donald L. “Monetary Policy and Asset Prices.”
Remarks at Monetary Policy: A Journey from Theory
to Practice, a European Central Bank Colloquium
held in honor of Otmar Issing, Frankfurt, Germany,
March 16, 2006.
Maddison, Angus. The World Economy: Historical
Statistics. Paris: Organisation for Economic
Co-operation and Development, 2003.
Maisel, Sherman J. Managing the Dollar. New York:
W.W. Norton, 1973.
Meltzer, Allan H. A History of the Federal Reserve,
Volume 1: 1913-1951. Chicago: University of
Chicago Press, 2003.
Meyer, Laurence H. A Term at the Fed: An Insider’s
View. New York: Harper Business, 2004.
Obstfeld, Maurice and Taylor, Alan M. “The Great
Depression as a Watershed: International Capital
Mobility over the Long Run,” in Michael D. Bordo,
Claudia Goldin, and Eugene N. White, eds., The
Defining Moment: The Great Depression and the
American Economy in the Twentieth Century.
Chicago: University of Chicago Press, 1998, pp.
353-402.
Pagan, Adrian R. and Sossounov, Kirill A. “A Simple
Framework for Analysing Bull and Bear Markets.”
Journal of Applied Econometrics, January/February
2003, 18(1), pp. 23-46.
Rigobon, Roberto and Sack, Brian. “Measuring the
Reaction of Monetary Policy to the Stock Market.”
Quarterly Journal of Economics, May 2003, 118(2),
pp. 639-69.

MARCH/APRIL 2007

117

Bordo and Wheelock

Romer, Christina D. and Romer, David H. “A
Rehabilitation of Monetary Policy in the 1950s.”
American Economic Review, May 2002, 92(2), pp.
121-27.

White, Eugene N. “Bubbles and Busts: The 1990s in
the Mirror of the 1920s.” NBER Working Paper No.
12138, National Bureau of Economic Research,
March 2006.

Schwartz, Anna J. “Why Financial Stability Depends
on Price Stability.” Economic Affairs, Autumn
1995, 15(4), pp. 21-25.

Wicker, Elmus R. Federal Reserve Monetary Policy,
1917-1933. New York: Random House, 1966.

Wheelock, David C. The Strategy and Consistency of
Federal Reserve Monetary Policy, 1923-1933.
Cambridge: Cambridge University Press, 1991.

Woodford, Michael. Interest and Prices: Foundations
of a Theory of Monetary Policy. Princeton: Princeton
University Press, 2003.

Wheelock, David C. “Monetary Policy in the Great
Depression: What the Fed Did and Why.” Federal
Reserve Bank of St. Louis Review, March/April 1992,
74(2), pp. 3-28.

118

MARCH/APRIL 2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

APPENDIX
This appendix provides information about the data and sources used in this article. It describes the
stock price index data that are used and then provides information about the data used in each figure
in the article.

STOCK PRICE INDEX (nominal, monthly data)
For all countries except the United States, the stock price data are from Global Financial Data
(www.globalfinancialdata.com). The following lists the Global Financial Data series identifier and
description for each country:
Australia:
Canada:
France:
Germany:
Italy:
Japan:
Netherlands:
Sweden:
United Kingdom:
United States:

AORDM, Australia ASX All-Ordinaries
GSPTSEM, Canada S&P/TSX 300 Composite
SBF250M, France SBF-250 Index
FWBXXM, Germany CDAX Composite Index
BCIIM, Banca Commerciale Italiana Index
N225M, Japan Nikkei 225 Stock Average
AAXM, Netherlands All-Share Price Index
SWAVM, Sweden Affarsvarlden General Index
FTASM, UK FT-Actuaries All-Share Index
NBER Macro History Database, series m11025a (1871:01–1920:12);
Standard & Poor’s 500 Composite Index (1941-43 = 10), monthly average of
daily data obtained from Haver Analytics (1921:01–2004:12)

REAL STOCK PRICE (monthly)
We use consumer price index data to deflate nominal stock prices to obtain a real stock price. For
all countries except the United States, our consumer price index data are from Global Financial Data.
The following lists the Global Financial Data consumer price index series identifier for each country.
Monthly observations are available beginning from the month listed in parentheses.
Australia:
Canada:
France:
Germany:
Italy:
Japan:
Netherlands:
Sweden:
United Kingdom:
United States:

CPAUSM (1912:01)
CPCANM (1914:01)
CPFRAM (1915:01)
CPDEUM (1923:12)
CPITAM (1920:01)
CPJPNM (1922:01)
CPNLDM (1919:01)
CPSWEM (1916:01)
CPGBRM (1914:01)
BLS, series ID: CUUR0000SA0, CPI–all urban consumers, U.S. city average,
all items, not seasonally adjusted, 1982-84 = 100 (1913:01–2004:12)

NOTES ABOUT THE FIGURES
We compute all growth rates using log first differencing, unless otherwise noted. For all figures
displaying annual data, if the peak month of a boom occurred in the first six months of a year, we
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

119

Bordo and Wheelock

attribute the peak to the prior calendar year. Otherwise, we attribute the peak to the calendar year that
it occurred. For figures displaying monthly or quarterly data, we attribute the peak to the actual month
or quarter that it occurred.

Figure 1
Real GDP: Quarterly data were downloaded from the OECD (Organisation for Economic Co-operation and Development) NEQ database of Haver Analytics. Data are available beginning in the quarter
listed in parentheses: Australia (1960:Q1); Canada (1961:Q1); France (1978:Q1); Germany (1991:Q1);
Italy (1980:Q1); Japan (1980:Q1); Netherlands (1977:Q1); Sweden (1980:Q1); United Kingdom (1960:Q1);
United States (1960:Q1). We compute growth rates as year-over-year growth rates for each quarter. We
define the long-run average growth rate as the average growth rate for 1960-2001, calculated using the
annual data from Maddison (2003, Tables 1B, 2B, and 5B).

Figure 2
Labor Productivity: Annual data on GDP per hour worked obtained from the OECD productivity
database (July 2005). The data for all countries span the years 1970-2004. We define the long-run average
growth rate as the average growth rate for 1970-2004.

Figures 3, 8, and 14
Inflation: The sources for consumer price index data are listed above. We compute annual inflation
rates by averaging annualized monthly growth rates. For booms ending prior to 1940, we define the
long-run average growth rate as the average growth rate from the first available observation through
1939 (first available observations: Australia, 1902; Canada, 1911; France, 1872; Germany, 1924; Italy,
1871; Japan, 1871; Netherlands, 1882; Sweden, 1871; United Kingdom, 1871; United States, 1870). For
booms ending after 1940, we define the long-run average growth rate as the average growth rate for
1947-2004.

Figure 4
Short-Term Interest Rate: Except as noted, monthly data on an overnight interest rate were downloaded from the International Financial Statistics (IFS) database of the International Monetary Fund.
The IFS series identifier and description are listed below. The month of the first available observation
is listed in parentheses.
Australia:
Canada:
France:
Germany:
Italy:
Japan:
Netherlands:
Sweden:
United Kingdom:
United States:

C193IM, short-term, weighted average of loans outstanding (1969:07)
C156IM, money market (MMkt) overnight financing rate (1975:01)
C132IM, MMkt opening rate: day-to-day loans against private bills (1965:01)
C134IM, interbank overnight (1965:01)
C136IM, 3-month interbank deposits, daily average (1971:01)
C158IM, Tokyo overnight call money (1965:01)
C138IM, MMkt rate on bankers’ call loans (1965:01–1998:12)
C144IM, day-to-day interbank loans (1965:12)
C112IM, interbank overnight offer rate (1972:01)
C111IM, interbank overnight federal funds (1965:01)

Figure 5
Real Interest Rate: We compute the real short-term interest rate as the difference between the nominal
short-term interest rate and CPI growth in the given month, and we compute CPI growth as the year-over120

MARCH/APRIL 2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Bordo and Wheelock

year change in the CPI for that month. Monthly data are from sources listed above. Monthly data on
government security yields were downloaded from the IFS database. The IFS series identifier and
description are listed below.

Figure 6
Interest Rate Term Spread: We compute the term spread as the difference between the yields on
long-term government securities and the short-term interest rate used to construct Figure 4. Monthly
data on government security yields were downloaded from the IFS database. The IFS series identifier
and description are listed below.
Australia:
Canada:
France:
Germany:
Italy:
Japan:
Netherlands:
Sweden:
United Kingdom:
United States:

C193IB, 10-year government nonrebate bond yield
C156IB, 10-or-more-year government bond yield to maturity
C132IB, 5-or-more-year government bond yield to maturity
C134IB, 3-or-more-year government & agency bond yield, weighted average
C136IB, 9-to-10-year government bond yield
C158IB, yield to maturity of all ordinary government bonds
C138IB, 10-year government bond yield: most recent bond
C144IB, 9-year government bond yield
C112IB, 20-year government bonds issued at par
C111IB, 10-year government bond yield at constant maturity

Figures 7 and 13
Real GDP: Data are from Maddison (2003, Tables 1B, 2B, and 5B) for 1871-2001 and the OECD for
2001-04. For booms ending prior to 1940, we define the long-run average growth rate as the average
growth rate for 1871-1939. For booms ending after 1940, we define the long-run average growth rate as
the average growth rate for 1960-2001.

Figures 9 and 10
Commercial Paper Interest Rate: Monthly data were obtained from Global Financial Data, except as
noted. There are no data for Australia, Canada, or Germany for the prewar period, and hence the medians
plotted in the figure exclude these countries. The following lists the Global Financial Data series identifier and description for each country. Data availability is listed in parentheses.
France:
Italy:
Japan:
Netherlands:
Sweden:
United Kingdom:
United States:

IPFRAW, private discount rate (1922:01–1940:06)
IPITAW, private discount rate (1922:01–1939:09)
IPJPN3D, private bills 3-month discount rate (1900:01–1945:06)
IPNLDW, private discount rate/advances (1919:01–1940:05)
IPSWEW, private discount rate (1926:01–1941:12)
IPGBR3D, U.K. private discount rate (1900:01–2004:12)
U.S. commercial paper interest rate (4- to 6-month), Federal Reserve Board,
Banking and Monetary Statistics (1943, Table 120, pp. 450-51) (1919:01–1941:12)

Figures 11 and 17
Money Stock: Except as noted below, our data are for a broad money stock measure and come from
Bordo et al. (2001, Appendix A). Recent data, which we obtained from Haver Analytics, are from the
OECD. Data for euro area countries end in 1998. We do not include these countries in the calculation
of median growth rates after 1998. For booms ending prior to 1940, we define the long-run average
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL 2007

121

growth rate as the average growth rate for 1881-1939. For booms ending after 1940, we define the
long-run average growth rate as the average growth rate for 1947-2004.
Australia:
Canada:
France:
Germany:
Italy:
Japan:
Netherlands:
Sweden:
United Kingdom:
United States:

Haver Analytics series C193FM3@OECDMEI, M3 (1998-2004)
Haver Analytics series C156FM1@OECDMEI, M1 (1996-2004)
Haver Analytics series C132FM3@OECDMEI, M3 (1990-1998)
Haver Analytics series C134FM2@OECDMEI, M2 (1990-1998)
Haver Analytics series C136FM2@OECDMEI, M2 (1996-1998)
Haver Analytics series C158FM2@OECDMEI, M2 + CDs (1998-2004)
Haver Analytics series C138FM3N@OECDMEI, M3 (1990-1998)
Haver Analytics series C144FM3N@OECDMEI, M3 (1996-2004)
Haver Analytics series C112FM4@OECDMEI, M4 (1998-2004)
Friedman and Schwartz (1963, Table A1, column 8) (1882-1959);
Haver Analytics series C111FM2@OECDMEI, M2 (1960-2004)

Figures 12 and 18
Real Money Stock: We compute real money stock growth as the difference between the growth rates
of the nominal money stock and consumer price index. For booms ending prior to 1940, we define the
long-run average growth rate as the average growth rate from the first available observation through 1939
(first available observations: Australia, 1902; Canada, 1911; France, 1881; Germany, 1926; Italy, 1881;
Japan, 1881; Netherlands, 1882; Sweden, 1881; United Kingdom, 1881; United States, 1881). For booms
ending after 1940, we define the long-run average growth rate as the average growth rate for 1947-2004.

Figures 15 and 16
Short-Term Interest Rate: Except as noted, monthly data on an overnight interest rate and/or
Treasury bill rate were obtained from Global Financial Data. The following lists the Global Financial
Data series identifier and description for each country. Data availability is listed in parentheses.
Australia:
Canada:

ITAUS3D, 3-month T-bill yield (1928:07–2004:12)
ITCAN3D, 3-month T-bill yield (1934:03–1956:12); IMCAND,
overnight MMkt rate (1957:01–2004:12)
France:
ITFRA3D, 3-month T-bill yield (1931:01–1936:04; 2002:01–2004:12);
IMFRAD, call money rate (1936:05–2001:12)
Germany:
ITDEUM, 3-month T-bill yield (1953:01–1954:02; 2002:01–2004:12);
IMDEUD, call money rate (1954:03–2001:12)
Italy:
ITITA3M, 3-month T-bill yield (1946:04–1978:05; 2002:01–2004:12);
IMITAD, interbank overnight rate (1978:06–2001:12)
Japan:
IMJPND, overnight lending rate (1949:01–2004:12)
Netherlands:
IMNLDD, overnight interbank rate (1929:01–2001:12);
ITNLDD, 3-month T-bill yield (2002:01–2004:12)
Sweden:
ITSWE3D, 3-month T-bill yield (1955:01–1965:07);
IMSWED, overnight interbank rate (1965:08–2004:12)
United Kingdom: IMGBRD, overnight interbank rate (1945:12–2004:12)
United States:
ITUSA3SD, 90-day T-bill secondary market (1920:01–1954:06);
overnight federal funds rate, Federal Reserve Bank of St. Louis FREDII database
(1954:07–2004:12)

122

MARCH/APRIL 2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Trends in Neighborhood-Level Unemployment
in the United States: 1980 to 2000
Christopher H. Wheeler
Although the average rate of unemployment across U.S. metropolitan areas declined between 1980
and 2000, the geographic concentration of the unemployed rose sharply over this period. That is,
residential neighborhoods throughout the nation’s metropolitan areas became increasingly divided
into high- and low-unemployment areas. This paper documents this trend using data on more than
165,000 U.S. Census block groups (neighborhoods) in 361 metropolitan areas over the years 1980,
1990, and 2000; it also examines three potential explanations: (i) urban decentralization, (ii) industrial shifts and declining unionization, and (iii) increasing segregation by income and education.
The results offer little support for either of the first two explanations. Rising residential concentration of the unemployed shows little association with changes in population density, industrial
composition, or union activity. It does, however, show a significant association with both the
degree of segregation according to income as well as education, suggesting that decreases in the
extent to which individuals with different levels of income and education live in the same neighborhood may help account for this trend. (JEL J11, J64, R20, R23)
Federal Reserve Bank of St. Louis Review, March/April 2007, 89(2), pp. 123-41.

T

he rate of unemployment is one of the
most basic indicators used to gauge the
state of the economy. High rates, of
course, tend to occur in recessionary
periods when levels of economic activity decline,
whereas lower rates tend to prevail in times of
expansion when employers typically increase the
size of their payrolls. Over time, as the economy
fluctuates between periods of expansion and
recession, we see corresponding changes in the
rate of unemployment.
Although this temporal variation in unemployment is widely known, there is also a fair amount
of variation geographically. At any point in time,
unemployment can differ substantially across
states, cities, and counties as a result of differences
in industrial compositions, labor market demographics, and region-specific shocks.
Geographic variation even extends down to
extremely small areas: Census tracts and block

groups (i.e., neighborhoods).1 Hence, within the
same metropolitan area, some neighborhoods
have a much higher incidence of unemployment
than others.
To be sure, residential areas in the United
States have long exhibited a tremendous amount
of heterogeneity with respect to the characteristics of the households that inhabit them. Some
neighborhoods, quite simply, tend to be populated
by households with high levels of income and
wealth, whereas others are inhabited by relatively
poor households. It is therefore not at all surprising
that, within any local labor market, there would
be neighborhoods with high levels of unemployment and those with low levels.
1

As noted here, these are extremely small areas. In the year 2000,
tracts encompassed roughly 1.3 square miles and 1,600 households
on average, whereas block groups averaged approximately 0.33
square miles and 500 households.

Christopher H. Wheeler is a senior economist at the Federal Reserve Bank of St. Louis. Elizabeth La Jeunesse provided research assistance.

© 2007, The Federal Reserve Bank of St. Louis. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in
their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made
only with prior written permission of the Federal Reserve Bank of St. Louis.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

MARCH/APRIL

2007

123

Wheeler

However, what is particularly interesting about
the extent to which individuals sort themselves
by characteristics, such as the incidence of unemployment, concerns the potential implications for
various labor market outcomes. In particular, a
large literature examining “social interactions” has
argued that the characteristics of individuals’
residential areas greatly influence their economic
outcomes. Case and Katz (1991), for instance, find
strong peer effects characterizing a variety of
behaviors, including criminal activity, drug and
alcohol use, schooling, and employment status
within a sample of residential areas in Boston.
Similarly, Topa (2001) finds evidence of local
spillovers in unemployment across Census tracts
in Chicago: High levels of unemployment within
a neighborhood tend to have a negative influence
on the employment prospects of individuals residing within or near that neighborhood. Wilson
(1987) suggests that neighborhood effects of this
sort form the basis of the rise in inner city poverty
in the United States in recent decades. As successful workers have gradually left inner cities, those
who remain are surrounded by rising levels of
poverty and joblessness, which makes it increasingly less likely that the residents of these areas
will find work.
Understanding the extent to which individuals
are segregated, therefore, is an important topic.
However, although existing research has looked
at residential segregation based on race (e.g.,
Cutler, Glaeser, and Vigdor, 1999) and income
(e.g., Wheeler, 2006), relatively little work has
studied the segregation of the unemployed from
the employed.2
This paper seeks to do so by examining the
distribution of unemployment across metropolitan area–level neighborhoods, defined by Census
block groups, over the years 1980, 1990, and 2000.
The primary findings indicate that the extent to
which unemployed workers are concentrated
residentially increased dramatically over this
period. For example, in 1980, the 90th percentile
of the distribution of neighborhood unemploy-

ment rates averaged 11 percent over the 361 U.S.
metropolitan areas in the sample, whereas the 10th
percentile averaged 3.7 percent. By 2000, the 90th
percentile had risen to 12.5 percent while the 10th
percentile had dropped to 1.3 percent, suggesting
that neighborhoods in the United States have
become increasingly polarized into high- and
low-unemployment areas.3
What accounts for this trend? Although these
are not intended to be a comprehensive set of
potential explanations, I consider three possibilities. First, the process of urban decentralization
(i.e., the gradual movement of metropolitan populations in the United States from central cities to
suburban locales) may have reduced the employment opportunities of households that continue
to reside in historical city centers. That is, just as
Wilson (1987) has argued, sprawl may have created a steadily rising gap between rates of unemployment in central cities and those in suburbs.
Second, changes in the labor market, such as
declining union activity and the shift of employment away from manufacturing toward other
sectors, may have reduced the employment opportunities for workers in particular neighborhoods
more so than it has for others. For instance, if a
city’s low- to middle-income neighborhoods are
populated primarily by manufacturing workers,
whereas the residents of its high-income neighborhoods are employed in professional services, a
decline in the manufacturing sector (or a rise in
the professional services sector) may result in a
rising differential between neighborhood unemployment rates. Third, there may have been an
increase in the extent to which skilled and
unskilled workers are segregated across residential areas. That is, independent of either urban
decentralization or shifts in union and industrial
activity, the degree to which high- and low-skill
workers live in the same neighborhoods may
have decreased over time, thus leading to rising
concentration of unemployment.
To summarize briefly, the findings offer little
support for either of the first two explanations.
3

2

The studies surveyed above, especially Case and Katz (1991) and
Topa (2001), focus on estimating the strength of peer effects rather
than documenting the evolution of segregation.

124

MARCH/APRIL

2007

These are unweighted statistics. If the percentiles are calculated
by weighting each neighborhood by the size of its labor force, the
average 90th percentile increased from 10.7 to 11.2 percent over this
period while the 10th percentile dropped from 3.8 to 1.5 percent.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

The change in the amount of unemployment
concentration across neighborhoods shows little
association with changes in population density
(a proxy for urban decentralization), changes in
the local rate of union coverage, or changes in
the shares of employment accounted for by nine
broad industrial sectors (including manufacturing).
The results do, however, reveal a strong positive
association between unemployment concentration
and measures of segregation according to income
and (college) education across neighborhoods.
As such, the findings suggest that rising concentration of unemployment is related to an increase
in the extent to which households have sorted
themselves residentially by income and education.

DATA AND MEASUREMENT
The data are taken from the decennial U.S.
Census of Population as compiled by GeoLytics.4
These files identify a variety of characteristics of
the households residing in a host of geographic
units, including counties, tracts, and neighborhoods, throughout the entire country. The primary
advantage of the GeoLytics files is the consistency
of the spatial units for which the data are identified: GeoLytics maintains a constant set of definitions in computing aggregate statistics for
neighborhoods, tracts, counties, and all other
geographic entities. As a result, the statistics
reported for each spatial unit are directly comparable from one year to the next.
From these data, I create a number of variables
at the metropolitan area–level, including population demographics, density (i.e., residents per
square mile), and industrial composition. I also
construct a rate of union coverage for each metropolitan area using the state-level rates reported
by Hirsch, Macpherson, and Vroman (2001).5
These quantities are intended to help identify the
characteristics that are associated with changes
4

More information about these data is available at
www.geolytics.com.

5

These data are available at www.unionstats.com. Metropolitan
area–level unionization rates are calculated as weighted averages
of the state-level rates, where the weights are given by the fraction
of each metro area’s labor force located in each state.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

in the geographic distribution of unemployment
within a city.6
The primary object of interest—the degree to
which unemployment is spatially concentrated—
is measured in two fundamental ways. First, I
compute the differences between three percentiles
(90th, 50th, and 10th) of the distribution of
neighborhood-level unemployment rates.7 Higher
values of these three differentials (90-10, 90-50,
50-10) indicate greater disparity (i.e., higher concentration) among neighborhood-level unemployment rates.
Second, I calculate an index of dissimilarity,
which measures the degree to which the members
of a particular group (in this case, unemployed
individuals) are unevenly distributed throughout
a city’s neighborhoods. This index is given as
follows:
(1)

Dissimilarity =

empi
1 N unempi
−
,
∑
2 i =1 unemptotal emptotal

where unempi is the number of unemployed
individuals in neighborhood i, unemptotal is the
number of unemployed individuals in the metropolitan area, empi is the number of employed
individuals in neighborhood i, emptotal is the number of employed individuals in the metropolitan
area, and N is the total number of neighborhoods
in the metropolitan area.
As described by Cutler, Glaeser, and Vigdor
(1999), the index of dissimilarity ranges between
0 (least concentrated) and 1 (most concentrated)
and is commonly interpreted as the fraction of
unemployed individuals that would need to move
(i.e., change neighborhood of residence) in order
for the unemployed to be uniformly distributed
across a city’s neighborhoods. This particular
metric has been widely used in the literature
studying trends in racial segregation, but it can
be applied readily to the analysis of segregation
based on any binary indicator.
6

Metropolitan areas are the local labor markets examined throughout
the analysis. The terms “city” and “metropolitan area” are used
interchangeably for expositional purposes.

7

The 90th percentile, for example, represents the unemployment
rate that is greater than the unemployment rates of 90 percent of
the neighborhoods.

MARCH/APRIL

2007

125

Wheeler

Table 1
Summary Statistics: Unemployment Concentration
Year

Variable

1980

1990

2000

Mean

Standard deviation

Minimum

Maximum

Dissimilarity

0.18

0.04

0.047

0.3

90-10 Difference

0.073

0.029

0.007

0.18

90-50 Difference

0.046

0.022

0.001

0.126

50-10 Difference

0.027

0.011

0.005

0.082

90th Percentile

0.11

0.038

0.03

0.252

50th Percentile

0.064

0.022

0.019

0.147

10th Percentile

0.037

0.017

0

0.106

Dissimilarity

0.27

0.04

0.16

0.38

90-10 Difference

0.113

0.039

0.051

0.268

90-50 Difference

0.074

0.03

0.025

0.211

50-10 Difference

0.039

0.013

0.016

0.097

90th Percentile

0.131

0.043

0.051

0.303

50th Percentile

0.057

0.018

0.026

0.137

10th Percentile

0.018

0.009

0

0.052

Dissimilarity

0.31

0.05

0.15

0.5

90-10 Difference

0.112

0.037

0.049

0.271

90-50 Difference

0.076

0.029

0.031

0.206

50-10 Difference

0.037

0.012

0.015

0.092

90th Percentile

0.125

0.042

0.054

0.3

50th Percentile

0.049

0.018

0.022

0.132

10th Percentile

0.013

0.009

0

0.047

NOTE: Unweighted statistics calculated from 361 metropolitan areas in each year.

I define neighborhoods as block groups, which
are the smallest geography for which detailed
Census data are publicly available. As noted here
previously, block groups are quite small: In the
year 2000, they averaged approximately 500
households and covered roughly a third of a
square mile. Households within the same neighborhood, then, can reasonably be expected to have
some sort of interaction with one another (e.g.,
passing on the street). Conceptually, this feature of
neighborhoods matches well with the theoretical
literature on neighborhood effects (e.g., Benabou,
1993), which treats neighborhoods as areas over
which economic agents come into contact with
one another.
126

MARCH/APRIL

2007

BASIC TRENDS
Between 1980 and 2000, the unemployed
became increasingly concentrated in relatively few
residential areas. For example, in 1980, the median
unemployed worker lived in a neighborhood with
an unemployment rate of 7.5 percent (i.e., the
unemployment rate within a worker’s own neighborhood of residence was 7.5 percent or greater for
at least 50 percent of all unemployed workers).8
Two decades later, the median unemployed worker
lived in a neighborhood with an unemployment
8

This figure is calculated by taking a weighted median across all
neighborhoods within a metropolitan area, where the weights are
the number of unemployed individuals within each neighborhood.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

Figure 1
Neighborhood Unemployment Percentiles
Unemployment Rate
0.14

0.12

0.1

90th Percentile
50th Percentile

0.08

10th Percentile

0.06

0.04

0.02

0
1980

rate of 7.9 percent. This trend is particularly
striking in light of the fact that the average metropolitan area unemployment rate declined from
6.9 percent to 5.9 percent over this period.
Rising residential concentration of the unemployed is also apparent from the index of dissimilarity (1) and the percentile differentials. Summary
statistics appear in Table 1.9 On average, the dissimilarity index increased from 0.18 in 1980 to
0.31 in 2000. Again, interpreting this index as
the fraction of unemployed workers that would
need to relocate in order for the unemployed to
be uniformly distributed in a metropolitan area,
these results reveal an enormous increase in the
concentration of unemployment. An additional
13 percent of all unemployed workers would
have needed to relocate in 2000 to equalize
unemployment across all neighborhoods.
The percentile differences reveal a qualitatively similar pattern. In 1980, the average difference between the neighborhoods at the 90th and
10th percentiles of the unemployment distribution
was 7.3 percentage points. Two decades later,

1990

2000

the difference was 11.2 percentage points. Based
on the 90-50 and 50-10 differences, it is clear that
this increase occurred at both the top and bottom
of the neighborhood unemployment distribution,
although the majority of the increase in the 90-10
gap was associated with an increase of the 90th
percentile relative to the median. The average 9050 gap increased by 3 percentage points between
1980 and 2000, whereas the mean 50-10 gap
increased by 1 percentage point.
Figure 1 plots the average values of the 90th,
50th, and 10th percentiles between 1980 and 2000.
Much of the widening of neighborhood unemployment distributions within the urban areas of the
United States took place between 1980 and 1990,
when the average 90th percentile increased while
the 50th and 10th percentiles decreased. Between
1990 and 2000, all three percentiles actually
decreased by similar amounts, leaving the three
differentials mostly unchanged between 1990
and 2000.10
10

9

A list of the metropolitan areas in the sample appears in the
appendix.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

The decrease in each percentile is very likely associated with the
general decrease in unemployment during the 1990s. Recall, the
average metropolitan area–level unemployment rate decreased
from 6.4 percent to 5.9 percent between 1990 and 2000.

MARCH/APRIL

2007

127

Wheeler

Figure 2
Normalized Neighborhood Unemployment Percentiles
Unemployment Rate
0.08

0.06
90th Percentile

0.04

50th Percentile
10th Percentile

0.02
1980

0

1990

2000

–0.02

–0.04

–0.06

Although doing so does not influence the
magnitudes of the percentile differences, it is also
worthwhile to examine the evolution of each
unemployment percentile after controlling for
each metropolitan area’s overall unemployment
rate. That is, Figure 1 may be somewhat difficult
to interpret because the percentiles may be higher
(or lower) in one year than another simply because
overall rates of unemployment have risen (or
fallen). As an alternative, I calculate a set of “normalized” percentiles by taking the deviations of
each metro area’s percentiles from its overall rate
of unemployment. That is, instead of reporting
the three raw percentiles, I report each percentile
minus the unemployment rate for the entire metropolitan area. The averages of these normalized
percentiles appear in Figure 2. What they show,
of course, is very much the same pattern: an
increase in the rate of unemployment among
neighborhoods with already high levels of unemployment and a decrease among neighborhoods
with already low levels.11
11

The normalized 90th, 50th, and 10th percentiles were 0.042,
–0.004, and –0.031, respectively, in 1980. In 2000, they were
0.066, –0.009, and –0.046.

128

MARCH/APRIL

2007

SOME POSSIBLE EXPLANATIONS
What might account for the increase in the
geographic concentration of unemployment? This
section considers three straightforward hypotheses
that might help to explain this trend: the movement of city populations toward suburban areas
(sprawl), changes in industrial composition and
union activity, and rising segregation of individuals by income and education.

Sprawl
One of the most prominent theories in urban
economics over the past half century suggests that
the movement of population and employment
away from city centers toward suburban locales
has created an underclass of unemployed workers
in central cities. This idea, known widely as the
spatial mismatch hypothesis, was first studied
by Kain (1968).
The basic rationale behind this theory is
straightforward. As city populations and employers move away from traditional central business
districts, it becomes more difficult for workers
who choose to remain in those central cities to
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

find and secure jobs. Increased spatial isolation
from employment opportunities, presumably,
increases commuting costs and makes the job
search process more difficult. In addition,
increased distance may limit access to information
about available jobs or create negative attitudes
about central city workers among employers.
Thus, as employers move farther away, it becomes
less likely that the residents of historical city
centers will be able to locate and maintain a job.
Although somewhat mixed, the evidence does
provide some support for this idea. Weinberg
(2000) finds that job centralization, measured by
the fraction of jobs located within the central city
of a metropolitan area (relative to the fraction of
residents in the central city), is strongly, positively
associated with the employment rate of black
workers. These workers, on average, represent
large fractions of central-city dwellers. Ihlanfeldt
and Sjoquist (1989) find that the earnings of both
black and white low-skill workers tend to decrease
with job decentralization, which is consistent
with the idea that sprawl has made it more difficult for individuals in certain neighborhoods to
find work.
Quantifying sprawl, however, tends to be
somewhat difficult because the term does not
have a precise definition. There are, of course, a
variety of measures that attempt to capture the
basic concept that individuals and employers
move from dense cores toward less-populated
suburban peripheries. Such measures include the
fraction of a metropolitan area’s population or
employment located in a central city, the fraction
within certain distances of the historical city
center, or overall metropolitan area density. As it
happens, many of these measures turn out to be
positively correlated with one another (see Glaeser
and Kahn, 2004).
In this paper, I quantify urban decentralization within a metropolitan area using population
density, which is constructed as a weighted average of neighborhood-level densities. The weights
in this case are given by each neighborhood’s
share of total metropolitan area population. Hence,
a metropolitan area’s density is taken to be the
density of the neighborhood in which the average
resident lives. Because suburban locales tend to
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

have much lower residential densities than urban
cores, lower levels of population density ought
to be associated with more extensive sprawl.12
Summary statistics describing levels of population density among the 361 metropolitan areas
in the sample in each year appear in Table 2.
Between 1980 and 2000, the average metropolitan
area saw its density decrease from 3,080 to 3,004
residents per square mile. Although average density did increase slightly during the 1980s, it
dropped during the 1990s, leaving the residential
density faced by a typical metropolitan resident
lower in 2000 than in two decades earlier.13
This pattern is generally consistent with the longstanding trend for U.S. populations to spread out
geographically.

Industrial Shifts and Unionization
The past several decades have been characterized by decreasing employment in certain
sectors, but increasing employment in others.
Most notably, manufacturing employment has
decreased while service employment has
increased. In addition, rates of unionization have
fallen substantially.
Some of these changes can be seen in the
summary statistics reported in Table 2. Between
1980 and 2000, the average share of manufacturing
in total employment declined from 22 percent to
14 percent across the 361 metropolitan areas in
the sample, whereas the fractions of workers
employed in education and health services rose
from 17 percent to 20 percent. Rates of unionization decreased from an average of 24 percent in
1980 to 14 percent in 2000.
How might these changes influence the geographic distribution of unemployment within a
metropolitan area? If workers in certain neighborhoods tend to be employed in similar types of
industries, or if unionization is relatively con12

In the year 2000, the average central city population density was
2,716 residents per square mile. Suburban densities that year averaged 208 residents per square mile. See Hobbs and Stoops (2002).

13

Looking at median changes rather than mean changes, metropolitan
area density actually decreased between 1980 and 1990. The median
change was –75 residents per square mile, indicating that density
actually decreased in the majority of metropolitan areas during
the 1980s.

MARCH/APRIL

2007

129

Wheeler

Table 2
Summary Statistics: Unemployment Covariates
Year

Variable

Mean

Standard deviation

Minimum

Maximum

1980

Population density
3,080.4
% Manufacturing
0.22
% Agriculture, forestry, fisheries
0.05
% Construction
0.06
% Wholesale trade
0.04
% Retail trade
0.17
% FIRE
0.05
% Public administration
0.06
% Education services
0.1
% Health services
0.07
Unionization rate
0.24
Education segregation
0.29
Income segregation
0.07

2,508.9
0.1
0.04
0.02
0.01
0.02
0.02
0.04
0.04
0.02
0.08
0.07
0.04

349.4
0.03
0.006
0.03
0.01
0.11
0.02
0.2
0.05
0.03
0.09
0.026
0.003

34,719.7
0.54
0.24
0.15
0.09
0.24
0.14
0.28
0.38
0.22
0.37
0.49
0.24

1990

Population density
3,083.4
% Manufacturing
0.18
% Agriculture, forestry, fisheries
0.04
% Construction
0.06
% Wholesale trade
0.04
% Retail trade
0.18
% FIRE
0.06
% Public administration
0.05
% Education services
0.09
% Health services
0.09
Unionization rate
0.17
Education segregation
0.34
Income segregation
0.135

2,613.2
0.08
0.03
0.02
0.01
0.02
0.02
0.03
0.04
0.02
0.07
0.06
0.05

607.1
0.03
0.008
0.04
0.01
0.12
0.03
0.2
0.05
0.04
0.06
0.19
0.04

35,993.8
0.48
0.19
0.12
0.11
0.26
0.16
0.22
0.38
0.22
0.32
0.51
0.31

2000

Population density
3,004.1
% Manufacturing
0.14
% Agriculture, forestry, fisheries
0.02
% Construction
0.07
% Wholesale trade
0.03
% Retail trade
0.12
% FIRE
0.06
% Public administration
0.05
% Education services
0.09
% Health services
0.11
Unionization rate
0.14
Education segregation
0.33
Income segregation
0.13

2,674.6
0.07
0.02
0.01
0.008
0.01
0.02
0.03
0.04
0.02
0.06
0.056
0.05

641.7
0.02
0.002
0.03
0.01
0.08
0.03
0.02
0.05
0.06
0.04
0.19
0.02

37,377.7
0.44
0.15
0.13
0.08
0.17
0.2
0.19
0.37
0.27
0.27
0.47
0.38

NOTE: Unweighted statistics calculated from 361 metropolitan areas in each year. “FIRE” is the financial, insurance, and real estate sector.

130

MARCH/APRIL

2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

centrated among the residents of certain neighborhoods, these changes may have produced differential rates of unemployment across different areas
within a city. In other words, rather than a change
occurring in the way residents of a metropolitan
area sort themselves across neighborhoods (e.g.,
into areas populated primarily by either highskill or low-skill workers), it may simply be that
changes in the labor market have affected workers
in different neighborhoods in different ways.

Segregation by Income and Education
The increase in concentration of unemployment may, on the other hand, be the product of
greater segregation of individuals by income and
education. If the manner by which individuals
sort themselves into residential areas has created
neighborhoods with concentrations of either highor low-skill individuals, we should see increasing
disparity between the unemployment rates of
different neighborhoods. Low-skill individuals,
after all, tend to experience higher rates of unemployment than high-skill individuals.14
On the surface, this explanation seems related
to the urban decentralization hypothesis sketched
above. Indeed, previous work has suggested that as
city populations spread out, households become
increasingly sorted into high- and low-income
neighborhoods (e.g., Glaeser and Kahn, 2004).
Recent work, however, challenges this view. In
particular, Wheeler (2006) finds little association
between the extent to which urban populations
spread out and the income differentials they
exhibit across either neighborhoods or tracts.
To quantify income segregation, I compute
the extent of variation between neighborhoods
as follows:
N

(2)

Income Variation =

2

∑ ωi ( y i − y )

,

i =1

where y–i is the average household income of
neighborhood i, y– is the average household income
in the city, ω i is the share of the metropolitan
area’s households living in neighborhood i, and
14

For example, the Bureau of Labor Statistics reports that the average
rate of unemployment tends to decrease with education attainment.
See www.bls.gov/news.release/empsit.t04.htm.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

N is the number of neighborhoods in the metropolitan area. This quantity reflects the extent of
heterogeneity in the average income levels of
different residential areas.
To measure educational segregation, I compute
an index of dissimilarity for college graduates.15
Recall, the resulting values represent the fraction
of a city’s population with a bachelor’s degree or
more that would have to relocate for these individuals to be uniformly distributed throughout the
city.
Summary statistics describing the evolution of
these two segregation measures appear in Table 2.
Clearly, both quantities increased between 1980
and 2000. On average, the amount of betweenneighborhood income variation nearly doubled
over this period, although essentially all of the
increase took place during the decade of the 1980s.
The dissimilarity index for college graduates
rose from 0.29 to 0.34 between 1980 and 1990. It
then showed a modest decline during the 1990s,
dropping to 0.33 by 2000.

EMPIRICAL ANALYSIS
Specification and Primary Results
To test the hypotheses outlined here, I consider
the following statistical model in which the degree
of neighborhood unemployment heterogeneity
(or concentration) in city c in year t, sct , is
expressed as follows:
(3)

sct = δc + δt + β X ct + εct ,

where δc is a city-specific effect intended to represent any time-invariant characteristics that may
influence the extent of variation in unemployment
across a city’s neighborhoods (e.g., a long-standing
history of residential segregation), δt is a yearspecific effect designed to pick up time trends
that influence all cities, Xct is a vector of timevarying city-level characteristics, and εct is a statistical residual.
15

Studies of human capital and skills typically define an individual
as having a “high” or “low” level of education based on whether
he or she has a four-year college degree or not. Hence, I define
educational segregation (i.e., the extent to which high- and loweducation individuals do not live with one another) based on college completion.

MARCH/APRIL

2007

131

Wheeler

Table 3
Correlates of Unemployment Concentration
Dependent variable
Regressor

Dissimilarity

% College

0.32* (0.09)

% Female

–0.35 (0.24)

% Black

–0.1 (0.13)

% Under 24

0.43* (0.14)

% Over 65

90-10 Difference

90-50 Difference

–0.17* (0.04)

–0.1* (0.04)

–0.08* (0.02)

–0.006 (0.11)

0.15 (0.1)

–0.16* (0.05)

0.12* (0.06)
–0.03 (0.07)

50-10 Difference

0.14* (0.06)

–0.02 (0.03)

0.004 (0.06)

–0.03 (0.03)

0.44* (0.17)

0.03 (0.08)

0.02 (0.07)

0.009 (0.04)

–0.27* (0.08)

–0.03 (0.04)

–0.05 (0.04)

0.01 (0.02)

% Manufacturing

0.18* (0.09)

–0.03 (0.04)

–0.02 (0.04)

–0.008 (0.02)

% Agriculture, forestry, fisheries

0.27* (0.15)

0.1 (0.07)

0.07 (0.07)

0.03 (0.03)

% Foreign-born

% Construction

0.33* (0.17)

–0.02 (0.08)

–0.005 (0.07)

–0.02 (0.04)

% Wholesale trade

0.09 (0.22)

–0.001 (0.1)

0.05 (0.09)

–0.06 (0.05)

% Retail trade

0.19 (0.13)

0.03 (0.06)

–0.02 (0.03)

% FIRE

0.27 (0.2)

–0.09 (0.1)

–0.06 (0.09)

–0.02 (0.04)

0.25 (0.15)

–0.1 (0.07)

% Public administration

0.02 (0.06)

–0.01 (0.07)

–0.08* (0.03)

–0.4* (0.17)

0.14* (0.08)

0.08 (0.08)

0.06* (0.04)

% Health services

0.07 (0.17)

0.14* (0.08)

0.11 (0.08)

0.03 (0.04)

Unemployment rate

0.23* (0.11)

0.96* (0.05)

0.64* (0.05)

0.33* (0.02)

% Education services

Unionization rate

0.03 (0.07)

0.05 (0.03)

0.04 (0.03)

0.016 (0.014)

Education segregation

0.25* (0.05)

0.1* (0.02)

0.05* (0.02)

0.05* (0.01)

Income segregation

0.42* (0.07)

0.18* (0.03)

0.14* (0.03)

0.04* (0.014)

Log population density

0.016 (0.011)

–0.004 (0.005)

–0.002 (0.005)

–0.002 (0.002)

R2

0.66

0.71

0.58

0.59

NOTE: Standard errors are reported in parentheses. All regressions include time dummies for the years 1980 and 1990 and interactions
of these dummies with three U.S. Census region indicators; * indicates significance at the 10 percent level or better. “FIRE” is the
financial, insurance, and real estate sector.

The vector of characteristics, Xct , includes the
following: log population density; the proportions
of the city’s resident population that are (i) female,
(ii) black, (iii) foreign-born, (iv) under the age of
24, and (v) over the age of 65; the share of total
employment in each of nine broad sectors; the
city’s overall unemployment rate; the proportion
of the city’s labor force that is covered by a union
contract; and measures of segregation of households by income and education across neighborhoods.16 I also include three region dummies
that are interacted with the year indicators, δt .
Many of these variables are intended to
account for some basic economic and demographic
132

MARCH/APRIL

2007

factors that may influence the distribution of
unemployment within a city’s neighborhoods.
Unemployment might, for example, vary significantly across neighborhoods as a result of the
racial, gender, or age composition of the local population. In addition, some neighborhoods may be
more sensitive to changes in the local business
cycle than others. Hence, the unemployment rate
and the six region-year interactions are included
16

The nine industries are manufacturing; agriculture, forestry, fisheries; construction; wholesale trade; retail trade; finance, insurance,
real estate; public administration; education services; and health
services. Because of changes in the industrial classification system
between 1990 and 2000, these were the only broad sectors that
could be constructed on a consistent basis from the GeoLytics data.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

to control for the influence of fluctuations in
local and regional economic activity.
The remaining covariates are included to
assess the hypotheses sketched above. In particular, population density is a rough proxy for urban
decentralization; the industry shares and unionization rate quantify changes in the labor market
facing workers; and the segregation measures
represent the degree of income and educational
sorting across a city’s neighborhoods.
Estimation of equation (3) is accomplished
using the within-estimator, whereby all variables
are expressed as deviations from averages taken
within metropolitan areas. The parameters are
then estimated by ordinary least squares. The
results appear in Table 3. Each column lists the
coefficients for a particular measure of unemployment concentration.
Beginning with the unemployment dissimilarity index in the first column of estimates, it is
evident that a number of the demographic characteristics are significantly associated with the
geographic concentration of unemployment.
Cities with larger fractions of individuals either
under 24 or over 65 years of age tend to have more
unequal distributions of unemployed workers
across neighborhoods. Cities in which these two
groups are heavily represented may, for example,
be strongly segregated by age. College towns, for
instance, have large fractions of relatively young
households clustered in certain neighborhoods.
If these individuals also experience relatively high
rates of unemployment, the dissimilarity index
would be especially high in these cities. The significantly positive coefficient on the college fraction, which tends to be especially high in college
towns, may reflect this same effect. The results
also suggest that a higher fraction of the resident
population that is foreign born corresponds to less
unemployment concentration. This finding may
simply indicate that cities with large numbers of
immigrants have rapidly growing economies and,
hence, a low incidence of unemployment among
all individuals. It could also be a reflection of the
fact that immigrants tend to be more active labor
force participants than domestic workers, at least
among those who have relatively little education
(Aaronson et al., 2006).
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Moving on to the three hypothetical causes
for the rise in unemployment concentration, it
is apparent that sprawl shows little systematic
association with the dissimilarity index. The
coefficient on the logarithm of population density
is statistically negligible. In addition, the union
coverage rate and five of the nine industry shares
are insignificant. Moreover, based on the signs of
the four significant industry share coefficients,
none supports the hypothesis sketched in the
section “Industrial Shifts and Unionization.” In
particular, the decline of manufacturing and rise
of professional services (e.g., education) should
be associated with the displacement of relatively
low-skill workers but rising employment opportunities for high-skill workers. To the extent that
these types of workers reside in different neighborhoods, these changes should generate greater
concentration of unemployment. According to
the results in Table 3, these changes tend to be
associated with decreases in unemployment
concentration.
Changes in the extent of residential segregation by income and education, by contrast, correlate strongly with changes in the geographic
concentration of unemployment. There is, of
course, likely to be some endogeneity associated
with the income segregation variable. After all,
as the distribution of unemployed households
becomes more uneven within a metropolitan area,
the distribution of income will very likely become
more uneven, too, because income tends to be
strongly tied to employment status. As a result,
the coefficient on income segregation likely
exhibits some upward bias. Nevertheless, the
positive association between these two quantities
is at least broadly consistent with the incomesorting hypothesis.
Moreover, the estimates also demonstrate a
significant connection between unemployment
concentration and the segregation of college
graduates, which is less obviously endogenous
with respect to the dependent variable. Unlike
income differentials across neighborhoods, there
is little reason to believe that an increase in the
concentration of unemployed households should
cause highly educated households to become
more segregated residentially. This suggests that
MARCH/APRIL

2007

133

Wheeler

any bias in the estimated coefficient on the education dissimilarity index may be small.
The estimates in the next three columns of
Table 3, where the dependent variables are the
unemployment percentile differences, offer many
of the same conclusions. The greater the change
in the extent of between-neighborhood income
variation or the separation of college graduates
from individuals with less education, the larger
the differentials in the unemployment rates of
different residential areas. Neither the unionization rate nor the log of population density shows
a significant association with any of the differentials, and only a few of the industry shares produce
significant coefficients.
As one might expect, changes in a metropolitan area’s overall unemployment rate are strongly
associated with the dissimilarity index and all
three unemployment rate differentials, suggesting
that the local business cycle is an important determinant of the geographic distribution of unemployment. Again, if economic downturns simply
affect workers in certain neighborhoods (say, lowskill workers in relatively low-income areas) more
than others, then one would expect to see all four
measures of unemployment concentration move
directly with the overall rate of unemployment.
That is precisely what the estimates in Table 3
indicate. Interestingly, however, even after having
accounted for this effect, there remains strong
evidence that rising concentration of unemployment has been driven by changes in the extent to
which households are segregated by income and
education. Thus, although local business cycle
effects are clearly important, they cannot completely account for the trends in neighborhoodlevel unemployment.

Results Using Weighted Percentiles
Because the percentiles used above are computed in an unweighted fashion, it is possible that
they provide misleading inferences about the
extent to which unemployed workers are spatially concentrated. For example, certain neighborhoods may be extremely small, possessing
only a few households, the majority of whom
happen to be unemployed. These neighborhoods
may then help to create extremely large values
134

MARCH/APRIL

2007

for a 90-10 or 50-10 difference. Yet, because they
only contain an extremely small share of a metropolitan area’s total stock of unemployed individuals, unemployment concentration might, in
actuality, be somewhat modest in this metro area.
A similar problem does not influence the
dissimilarity index because, as shown in equation
(1), the index implicitly gives less weight to neighborhoods with smaller numbers of employed and
unemployed individuals. Hence, an extremely
small neighborhood with a very high unemployment rate will contribute relatively little to the
index value because its shares of unemployed
and employed workers will be small.
In this section, I examine weighted percentiles, where the weights are given by the size
of each neighborhood’s labor force. After computing these percentiles, I simply create 90-10, 9050, and 50-10 differences and estimate the same
regressions as those reported in Table 3.
Summary statistics indicate that these
weighted measures of unemployment concentration did rise, although not as sharply as the
unweighted measures. On average, the 90-10,
90-50, and 50-10 differences stood at 0.069, 0.044,
and 0.026, respectively, in 1980. By 2000, they
had risen to 0.096, 0.065, and 0.031.
The regression results for these weighted differentials are presented in Table 4. For the most
part, they generate similar conclusions to those
drawn earlier. There is little evidence of the importance of industrial shifts and changes in union
activity. Population density does, in this case,
show a significant association with the 90-10
and 90-50 differences. However, the coefficients
are positive, indicating that rising sprawl (i.e.,
falling density) is associated with less unemployment concentration rather than more.
On the other hand, there is once again strong
evidence that the rising segregation of individuals
by educational attainment—specifically, the separation of college graduates from those with less
education—and increasing income variation
across neighborhoods are associated with rising
unemployment concentration. Cities characterized
by larger increases in residential sorting along
these two dimensions have seen, on average,
larger increases in their levels of unemployment
concentration.
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

Table 4
Robustness-Weighted Percentile Difference
Dependent variable
Regressor

Weighted 90-10 difference

Weighted 90-50 difference

Weighted 50-10 difference

% College

–0.13* (0.04)

–0.07* (0.04)

–0.06* (0.02)

% Female

0.09 (0.09)

0.14 (0.1)

–0.05 (0.05)

% Black

0.004 (0.05)

0.05 (0.05)

–0.04* (0.03)

% Under 24

0.05 (0.05)

0.07 (0.05)

–0.02 (0.03)

% Over 65
% Foreign-born

0.06 (0.07)

0.03 (0.07)

0.02 (0.03)

–0.02 (0.03)

–0.04 (0.03)

0.02 (0.02)

% Manufacturing

0.007 (0.04)

–0.005 (0.03)

0.01 (0.02)

% Agriculture, forestry, fisheries

0.05 (0.06)

0.009 (0.06)

0.04 (0.03)

0.02 (0.07)

–0.006 (0.07)

% Construction
% Wholesale trade
% Retail trade

–0.015 (0.09)

0.000004 (0.08)

0.02 (0.03)
–0.01 (0.04)

0.06 (0.05)

0.04 (0.05)

0.02 (0.03)

% FIRE

–0.07 (0.08)

–0.01 (0.08)

–0.06 (0.04)

% Public administration

–0.007 (0.06)

0.006 (0.06)

% Education services

0.07 (0.07)

0.001 (0.07)

0.07* (0.03)

% Health services

0.16* (0.07)

0.14* (0.07)

0.02 (0.03)

Unemployment rate

0.94* (0.05)

0.63* (0.04)

0.3* (0.02)

Unionization rate

–0.01 (0.03)

–0.02 (0.03)

–0.01 (0.03)

0.006 (0.01)

Education segregation

0.09* (0.02)

0.04* (0.02)

0.05* (0.01)

Income segregation

0.13* (0.03)

0.11* (0.03)

0.02* (0.01)

Log population density

0.01* (0.005)

0.008* (0.004)

0.003 (0.002)

R2

0.78

0.65

0.55

NOTE: Standard errors are reported in parentheses. All regressions include time dummies for the years 1980 and 1990 and interactions
of these dummies with three U.S. Census region indicators; * indicates significance at the 10 percent level or better. “FIRE” is the
financial, insurance, and real estate sector.

CONCLUSION
This paper has documented a rise in the
extent to which unemployed households throughout 361 U.S. metropolitan areas have become
concentrated residentially. In 1980, the median
unemployed worker resided in a neighborhood
with an unemployment rate of 7.5 percent; by
2000, that rate was 7.9 percent. Again, this is particularly striking in light of the fact that, on average,
unemployment rates were lower in 2000 than in
1980. Other measures of residential concentration
of the unemployed—an index of dissimilarity
and differences between three percentiles (either
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

weighted or unweighted) of the neighborhood
unemployment distribution—show similar qualitative trends. Hence, although the overall rate of
unemployment has not trended upward over
time, there is evidence of an upward trend in the
spatial concentration of the unemployed within
the country’s urban labor markets.
Among three plausible explanations, I find
the greatest support for the idea that increased
segregation of households by income and educational attainment underlies this trend. There is
less consistent evidence that sprawl or structural
changes in the labor market are responsible.
As noted in the introduction, these results
MARCH/APRIL

2007

135

Wheeler

are especially interesting because the literature
on neighborhood effects suggests that a number
of labor market outcomes are tied to the characteristics of one’s place of residence. Indeed, following this general premise, rising unemployment
concentration may help to account for two additional trends that have been observed in the
United States over the past three decades: (i) rising
inequality in both income and earnings and (ii) an
increase in the expected duration of unemployment. Both are well documented.
Between 1971 and 1995, the amount by which
the 90th percentile of the U.S. wage distribution
exceeded the 10th percentile grew from 266 percent to 366 percent (Acemoglu, 2002).17 This
increase has been accompanied by growing dispersion among the earnings of individuals of different “skill” groups (e.g., as defined by education
and experience) as well as those within the same
group. Although there has not been a long-run
trend in the overall rate of unemployment,
Abraham and Shimer (2001) report that the mean
unemployment duration rose by roughly 20 percent (from 10 weeks to 12 weeks) between 1980
and 2000. Much of this rise can be linked to an
increase in what they call “very long-term” unemployment (more than 26 weeks), which has more
than tripled as a share of the labor force since 1969.
As one might expect, research studying these
two patterns has identified some of the most likely
culprits. Rising inequality is very likely related
to skill-biased technological change, changes in
the institutional makeup of the labor market (e.g.,
declining union activity and minimum wage
changes), and growth in international trade and
immigration. Longer spells of unemployment are
probably tied to demographic changes, especially
the aging of the working population and an
increase in the fraction of women participating
in the labor force. Older workers and women
tend to experience somewhat longer periods of
unemployment (Abraham and Shimer, 2001).
Very little work, however, has considered
that there may be a spatial aspect to these phenomena. With rising concentration of the unem17

Similar evidence has been reported in many other studies,
including Levy and Murnane (1992), Katz and Murphy (1992),
and Juhn, Murphy, and Pierce (1993).

136

MARCH/APRIL

2007

ployed, workers in search of a job might find it
increasingly difficult to locate one. Recall that
Topa (2001) finds evidence consistent with local
spillovers in unemployment status across Census
tracts in Chicago. Again, this result may be the
product of an adverse network effect (i.e., if
workers find jobs through neighborhood contacts)
or employers simply avoiding workers from highunemployment neighborhoods due to a social
stigma. Rising concentration of unemployment
in certain neighborhoods may, then, give rise to
growing unemployment durations among workers
living in these neighborhoods and further decrease
their income and labor earnings relative to the
rest of the labor force over time.
It is interesting to note that, over the sample
period studied here, the majority of the increase
in the geographic concentration of unemployment
took place during the 1980s, when much of the
rise in both income inequality and unemployment
duration took place. Although far from conclusive, the fact that the timing of these phenomena
matches closely certainly suggests that there may
be a connection among them.

REFERENCES
Aaronson, Stephanie; Fallick, Bruce; Figura, Andrew;
Pingle, Jonathan and Wascher, William. “The Recent
Decline in the Labor Force Participation Rate and
Its Implications for Potential Labor Supply.”
Brookings Papers on Economic Activity, 2006, I,
pp. 69-134.
Abraham, Katharine G. and Shimer, Robert. “Changes
in Unemployment Duration and Labor Force
Attachment.” NBER Working Paper 8513, National
Bureau of Economic Research, October 2001.
Acemoglu, Daron. “Technical Change, Inequality,
and the Labor Market.” Journal of Economic
Literature, March 2002, 40(1), pp. 7-72.
Benabou, Roland. “Workings of a City: Location,
Education, and Production.” Quarterly Journal of
Economics, August 1993, 108(3), pp. 619-52.
Case, Anne C. and Katz, Lawrence F. “The Company
You Keep: The Effects of Family and Neighborhood

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

on Disadvantaged Youths.” NBER Working Paper
3705, National Bureau of Economic Research, May
1991.
Cutler, David M.; Glaeser, Edward L. and Vigdor,
Jacob L. “The Rise and Decline of the American
Ghetto.” Journal of Political Economy, June 1999,
107(3), pp. 455-506.
Glaeser, Edward and Kahn, Matthew. “Sprawl and
Urban Growth,” in J. Vernon Henderson and JacquesFrancois Thisse, eds., Handbook of Regional and
Urban Economics. Volume 4. Amsterdam: Elsevier,
2004, pp. 2481-527.
Hirsch, Barry T.; Macpherson, David A. and Vroman,
Wayne G. “Estimates of Union Density by State.”
Monthly Labor Review, July 2001, 124(7), pp. 51-55.
Hobbs, Frank and Stoops, Nicole. “Demographic
Trends in the 20th Century.” U.S. Census Bureau,
Census 2000 Special Reports, Series CENSR-4.
Washington, DC: U.S. Government Printing Office,
November 2002.
Ihlanfeldt, Keith R. and Sjoquist, David L. “The Impact
of Job Decentralization on the Economic Welfare of
Central City Blacks.” Journal of Urban Economics,
July 1989, 26(1), pp. 110-30.
Juhn, Chinhui; Murphy, Kevin M. and Pierce, Brooks.
“Wage Inequality and the Rise in Returns to Skill.”
Journal of Political Economy, June 1993, 101(3),
pp. 410-42.

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Kain, John F. “Housing Segregation, Negro
Employment, and Metropolitan Decentralization.”
Quarterly Journal of Economics, May 1968, 82(2),
pp. 175-97.
Katz, Lawrence F. and Murphy, Kevin M. “Changes
in Relative Wages, 1963-1987: Supply and Demand
Factors.” Quarterly Journal of Economics, February
1992, 107(1), pp. 35-78.
Levy, Frank and Murnane, Richard J. “U.S. Earnings
Levels and Earnings Inequality: A Review of Recent
Trends and Proposed Explanations.” Journal of
Economic Literature, September 1992, 30(3), pp.
1333-81.
Topa, Giorgio. “Social Interactions, Local Spillovers
and Unemployment.” Review of Economic Studies,
April 2001, 68(2), pp. 261-95.
Weinberg, Bruce A. “Black Residential Centralization
and the Spatial Mismatch Hypothesis.” Journal of
Urban Economics, July 2000, 48(1), pp. 110-34.
Wheeler, Christopher H. “Urban Decentralization
and Income Inequality: Is Sprawl Associated with
Rising Income Segregation Across Neighborhoods?”
Working Paper No. 2006-037A, Federal Reserve
Bank of St. Louis, 2006.
Wilson, William J. The Truly Disadvantaged: The
Inner City, The Underclass, and Public Policy.
Chicago: University of Chicago Press, 1987.

MARCH/APRIL

2007

137

Wheeler

APPENDIX
Metropolitan Areas
Abilene, TX
Akron, OH
Albany, GA
Albany-Schenectady-Troy, NY
Albuquerque, NM
Alexandria, LA
All MSAs
Allentown-Bethlehem-Easton, PA-NJ
Altoona, PA
Amarillo, TX
Ames, IA
Anchorage, AK
Anderson, IN
Anderson, SC
Ann Arbor, MI
Anniston-Oxford, AL
Appleton, WI
Asheville, NC
Athens-Clarke County, GA
Atlanta–Sandy Springs–Marietta, GA
Atlantic City, NJ
Auburn-Opelika, AL
Augusta-Richmond County, GA-SC
Austin-Round Rock, TX
Bakersfield, CA
Baltimore-Towson, MD
Bangor, ME
Barnstable Town, MA
Baton Rouge, LA
Battle Creek, MI
Bay City, MI
Beaumont–Port Arthur, TX
Bellingham, WA
Bend, OR
Billings, MT
Binghamton, NY
Birmingham-Hoover, AL
Bismarck, ND
Blacksburg-Christiansburg-Radford, VA
Bloomington, IN
Bloomington-Normal, IL
Boise City–Nampa, ID
Boston-Cambridge-Quincy, MA-NH
Boulder, CO
138

MARCH/APRIL

2007

Bowling Green, KY
Bremerton-Silverdale, WA
Bridgeport-Stamford-Norwalk, CT
Brownsville-Harlingen, TX
Brunswick, GA
Buffalo–Niagara Falls, NY
Burlington, NC
Burlington–South Burlington, VT
Canton-Massillon, OH
Cape Coral-Fort Myers, FL
Carson City, NV
Casper, WY
Cedar Rapids, IA
Champaign-Urbana, IL
Charleston, WV
Charleston–North Charleston, SC
Charlotte-Gastonia-Concord, NC-SC
Charlottesville, VA
Chattanooga, TN-GA
Cheyenne, WY
Chicago-Naperville-Joliet, IL-IN-WI
Chico, CA
Cincinnati-Middletown, OH-KY-IN
Clarksville, TN-KY
Cleveland, TN
Cleveland-Elyria-Mentor, OH
Coeur d’Alene, ID
College Station–Bryan, TX
Colorado Springs, CO
Columbia, MO
Columbia, SC
Columbus, GA-AL
Columbus, IN
Columbus, OH
Corpus Christi, TX
Corvallis, OR
Cumberland, MD-WV
Dallas–Fort Worth–Arlington, TX
Dalton, GA
Danville, IL
Danville, VA
Davenport-Moline–Rock Island, IA-IL
Dayton, OH
Decatur, AL
Decatur, IL
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

Deltona–Daytona Beach–Ormond Beach, FL
Denver-Aurora, CO
Des Moines, IA
Detroit-Warren-Livonia, MI
Dothan, AL
Dover, DE
Dubuque, IA
Duluth, MN-WI
Durham, NC
Eau Claire, WI
El Centro, CA
El Paso, TX
Elizabethtown, KY
Elkhart-Goshen, IN
Elmira, NY
Erie, PA
Eugene-Springfield, OR
Evansville, IN-KY
Fairbanks, AK
Fargo, ND-MN
Farmington, NM
Fayetteville, NC
Fayetteville-Springdale-Rogers, AR-MO
Flagstaff, AZ
Flint, MI
Florence, SC
Florence–Muscle Shoals, AL
Fond du Lac, WI
Fort Collins–Loveland, CO
Fort Smith, AR-OK
Fort Walton Beach–Crestview-Destin, FL
Fort Wayne, IN
Fresno, CA
Gadsden, AL
Gainesville, FL
Gainesville, GA
Glens Falls, NY
Goldsboro, NC
Grand Forks, ND-MN
Grand Junction, CO
Grand Rapids–Wyoming, MI
Great Falls, MT
Greeley, CO
Green Bay, WI
Greensboro–High Point, NC
Greenville, NC
Greenville, SC
Gulfport-Biloxi, MS
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Hagerstown-Martinsburg, MD-WV
Hanford-Corcoran, CA
Harrisburg-Carlisle, PA
Harrisonburg, VA
Hartford–West Hartford–East Hartford, CT
Hattiesburg, MS
Hickory-Lenoir-Morganton, NC
Hinesville–Fort Stewart, GA
Holland–Grand Haven, MI
Honolulu, HI
Hot Springs, AR
Houma–Bayou Cane–Thibodaux, LA
Houston–Sugar Land–Baytown, TX
Huntington-Ashland, WV-KY-OH
Huntsville, AL
Idaho Falls, ID
Indianapolis, IN
Iowa City, IA
Ithaca, NY
Jackson, MI
Jackson, MS
Jackson, TN
Jacksonville, FL
Jacksonville, NC
Janesville, WI
Jefferson City, MO
Johnson City, TN
Johnstown, PA
Jonesboro, AR
Joplin, MO
Kalamazoo-Portage, MI
Kankakee-Bradley, IL
Kansas City, MO-KS
Kennewick-Richland-Pasco, WA
Killeen-Temple–Fort Hood, TX
Kingsport-Bristol-Bristol, TN-VA
Kingston, NY
Knoxville, TN
Kokomo, IN
La Crosse, WI-MN
Lafayette, IN
Lafayette, LA
Lake Charles, LA
Lakeland, FL
Lancaster, PA
Lansing–East Lansing, MI
Laredo, TX
Las Cruces, NM
MARCH/APRIL

2007

139

Wheeler

Las Vegas–Paradise, NV
Lawrence, KS
Lawton, OK
Lebanon, PA
Lewiston, ID-WA
Lewiston-Auburn, ME
Lexington-Fayette, KY
Lima, OH
Lincoln, NE
Little Rock–North Little Rock, AR
Logan, UT-ID
Longview, TX
Longview, WA
Los Angeles–Long Beach–Santa Ana, CA
Louisville, KY-IN
Lubbock, TX
Lynchburg, VA
Macon, GA
Madera, CA
Madison, WI
Manchester-Nashua, NH
Mansfield, OH
McAllen-Edinburg-Mission, TX
Medford, OR
Memphis, TN-MS-AR
Merced, CA
Miami–Fort Lauderdale–Miami Beach, FL
Michigan City-La Porte, IN
Midland, TX
Milwaukee-Waukesha–West Allis, WI
Minneapolis–St. Paul–Bloomington, MN-WI
Missoula, MT
Mobile, AL
Modesto, CA
Monroe, LA
Monroe, MI
Montgomery, AL
Morgantown, WV
Morristown, TN
Mount Vernon–Anacortes, WA
Muncie, IN
Muskegon–Norton Shores, MI
Myrtle Beach–Conway–North Myrtle Beach, SC
Napa, CA
Naples-Marco Island, FL
Nashville-Davidson-Murfreesboro, TN
New Haven–Milford, CT
New Orleans–Metairie-Kenner, LA
140

MARCH/APRIL

2007

New York–Northern New Jersey–Long Island,
NY-NJ-PA
Niles–Benton Harbor, MI
Norwich-New London, CT
Ocala, FL
Ocean City, NJ
Odessa, TX
Ogden-Clearfield, UT
Oklahoma City, OK
Olympia, WA
Omaha–Council Bluffs, NE-IA
Orlando-Kissimmee, FL
Oshkosh-Neenah, WI
Owensboro, KY
Oxnard–Thousand Oaks–Ventura, CA
Palm Bay–Melbourne-Titusville, FL
Panama City–Lynn Haven, FL
Parkersburg-Marietta-Vienna, WV-OH
Pascagoula, MS
Pensacola–Ferry Pass–Brent, FL
Peoria, IL
Philadelphia-Camden-Wilmington, PA-NJDE-MD
Phoenix-Mesa-Scottsdale, AZ
Pine Bluff, AR
Pittsburgh, PA
Pittsfield, MA
Pocatello, ID
Port St. Lucie–Fort Pierce, FL
Portland–South Portland–Biddeford, ME
Portland-Vancouver-Beaverton, OR-WA
Poughkeepsie-Newburgh-Middletown, NY
Prescott, AZ
Providence–New Bedford–Fall River, RI-MA
Provo-Orem, UT
Pueblo, CO
Punta Gorda, FL
Racine, WI
Raleigh-Cary, NC
Rapid City, SD
Reading, PA
Redding, CA
Reno-Sparks, NV
Richmond, VA
Riverside–San Bernardino–Ontario, CA
Roanoke, VA
Rochester, MN
Rochester, NY
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Wheeler

Rockford, IL
Rocky Mount, NC
Rome, GA
Sacramento–Arden-Arcade–Roseville, CA
Saginaw-Saginaw Township North, MI
Salem, OR
Salinas, CA
Salisbury, MD
Salt Lake City, UT
San Angelo, TX
San Antonio, TX
San Diego–Carlsbad–San Marcos, CA
San Francisco–Oakland-Fremont, CA
San Jose–Sunnyvale–Santa Clara, CA
San Luis Obispo–Paso Robles, CA
Sandusky, OH
Santa Barbara–Santa Maria, CA
Santa Cruz–Watsonville, CA
Santa Fe, NM
Santa Rosa–Petaluma, CA
Sarasota-Bradenton-Venice, FL
Savannah, GA
Scranton–Wilkes-Barre, PA
Seattle-Tacoma-Bellevue, WA
Sheboygan, WI
Sherman-Denison, TX
Shreveport–Bossier City, LA
Sioux City, IA-NE-SD
Sioux Falls, SD
South Bend–Mishawaka, IN-MI
Spartanburg, SC
Spokane, WA
Springfield, IL
Springfield, MA
Springfield, MO
Springfield, OH
St. Cloud, MN
St. George, UT
St. Joseph, MO-KS
St. Louis, MO-IL
State College, PA
Stockton, CA
Sumter, SC

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W

Syracuse, NY
Tallahassee, FL
Tampa–St. Petersburg–Clearwater, FL
Terre Haute, IN
Texarkana, TX-Texarkana, AR
Toledo, OH
Topeka, KS
Trenton-Ewing, NJ
Tucson, AZ
Tulsa, OK
Tuscaloosa, AL
Tyler, TX
Utica-Rome, NY
Valdosta, GA
Vallejo-Fairfield, CA
Vero Beach, FL
Victoria, TX
Vineland-Millville-Bridgeton, NJ
Virginia Beach–Norfolk–Newport News,
VA-NC
Visalia-Porterville, CA
Waco, TX
Warner Robins, GA
Washington-Arlington-Alexandria, DC-VAMD-WV
Waterloo–Cedar Falls, IA
Wausau, WI
Weirton-Steubenville, WV-OH
Wenatchee, WA
Wheeling, WV-OH
Wichita Falls, TX
Wichita, KS
Williamsport, PA
Wilmington, NC
Winchester, VA-WV
Winston-Salem, NC
Worcester, MA
Yakima, WA
York-Hanover, PA
Youngstown-Warren-Boardman, OH-PA
Yuba City, CA
Yuma, AZ

MARCH/APRIL

2007

141

142

MARCH/APRIL

2007

F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W