View original document

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

Now
Quarterly

FEDERAL RESERVE BANK OF ST. LOUIS

REVIEW

Federal Reserve Bank of St. Louis
P.O. Box 442
St. Louis, MO 63166-0442

FIRST QUARTER 2014
VOLUME 96 | NUMBER 1

The Rise and Fall of Labor Force Participation in the United States
James Bullard

REVIEW

The Great Trade Collapse and Rebound: A State-by-State View
Cletus C. Coughlin

A Guide to Tracking the U.S. Economy
Kevin L. Kliesen

QE: Is There a Portfolio Balance Effect?
Daniel L. Thornton

The Evolution of Federal Reserve Policy and the
Impact of Monetary Policy Surprises on Asset Prices
First Quarter 2014 • Volume 96, Number 1

Brett W. Fawley and Christopher J. Neely

REVIEW
Volume 96 • Number 1
President and CEO
James Bullard

Director of Research
Christopher J. Waller

1
The Rise and Fall of Labor Force Participation
in the United States
James Bullard

Policy Adviser
Cletus C. Coughlin

Deputy Director of Research
David C. Wheelock

Review Editor-in-Chief
William T. Gavin

Research Economists
David Andolfatto
Alejandro Badel
Subhayu Bandyopadhyay
Maria E. Canon
YiLi Chien
Silvio Contessi
Riccardo DiCecio
William Dupor
Carlos Garriga
Rubén Hernández-Murillo
Kevin L. Kliesen
Fernando M. Martin
Michael W. McCracken
Alexander Monge-Naranjo
Christopher J. Neely
Michael T. Owyang
B. Ravikumar
Juan M. Sánchez
Daniel L. Thornton
Yi Wen
David Wiczer
Christian M. Zimmermann

13
The Great Trade Collapse and Rebound:
A State-by-State View
Cletus C. Coughlin

35
A Guide to Tracking the U.S. Economy
Kevin L. Kliesen

55
QE: Is There a Portfolio Balance Effect?
Daniel L. Thornton

73
The Evolution of Federal Reserve Policy and the
Impact of Monetary Policy Surprises on Asset Prices
Brett W. Fawley and Christopher J. Neely

Managing Editor
George E. Fortier

Editors
Judith A. Ahlers
Lydia H. Johnson

Graphic Designer
Donna M. Stiller

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

i

Review Now Published Quarterly
Review is published four times per year by the Research Division of the Federal Reserve Bank of St. Louis. Complimentary print subscriptions are
available to U.S. addresses only. Full online access is available to all, free of charge.

Online Access to Current and Past Issues
The current issue and past issues dating back to 1967 may be accessed through our Research Division website:
http://research.stlouisfed.org/publications/review. All nonproprietary and nonconfidential data and programs for the articles written by
Federal Reserve Bank of St. Louis staff and published in Review also are available to our readers on this website.
Review articles published before 1967 may be accessed through our digital archive, FRASER: http://fraser.stlouisfed.org/publication/?pid=820.
Review is indexed in Fed in Print, the catalog of Federal Reserve publications (http://www.fedinprint.org/) and in IDEAS/RePEc, the free online
bibliography hosted by the Research Division (http://ideas.repec.org/).

Authorship and Disclaimer
The majority of research published in Review is authored by economists on staff at the Federal Reserve Bank of St. Louis. Visiting scholars and
others affiliated with the St. Louis Fed or the Federal Reserve System occasionally provide content as well. Review does not accept unsolicited
manuscripts for publication.
The views expressed in Review are those of the individual authors and do not necessarily reflect official positions of the Federal Reserve Bank of
St. Louis, the Federal Reserve System, or the Board of Governors.

Subscriptions and Alerts
Single-copy subscriptions (U.S. addresses only) are available free of charge. Subscribe here:
https://research.stlouisfed.org/publications/review/subscribe/.
Our monthly email newsletter keeps you informed when new issues of Review, Economic Synopses, Regional Economist, and other publications
become available; it also alerts you to new or enhanced data and information services provided by the St. Louis Fed. Subscribe to the newsletter
here: http://research.stlouisfed.org/newsletter-subscribe.html.

Copyright and Permissions
Articles may be reprinted, reproduced, republished, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s),
and full citation are included. In these cases, there is no need to request written permission or approval. Please send a copy of any reprinted or
republished materials to Review, Research Division of the Federal Reserve Bank of St. Louis, P.O. Box 442, St. Louis, MO 63166-0442;
STLS.Research.Publications@stls.frb.org.
Please note that any abstracts, synopses, translations, or other derivative work based on content published in Review may be made only with
prior written permission of the Federal Reserve Bank of St. Louis. Please contact the Review editor at the above address to request this permission.

Economic Data
General economic data can be obtained through FRED (Federal Reserve Economic Data), our free database with more than 200,000 national,
international, and regional data series, including data for our own Eighth Federal Reserve District. You may access FRED through our website:
http://research.stlouisfed.org/fred2.
© 2014, Federal Reserve Bank of St. Louis.
ISSN 0014-9187

ii

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

In Memoriam: Michael J. Dueker

ur friend and former colleague, Michael J. Dueker,
passed away on Wednesday, January 29, 2014. Mike
was a well-respected economist and policy advisor, a
sought-after colleague and coauthor, and a close friend of many
current and former employees of the St. Louis Fed.
Mike joined the Research Division of the Federal Reserve
Bank of St. Louis as an economist in 1991, following his graduation from the University of Washington, Seattle, where he earned
a Ph.D. in economics. At the Bank, Mike was promoted three
times—to senior economist, research officer, and assistant vice
president—in recognition of his strong performance as a research
economist and policy advisor. Mike left the Bank in 2008 to return
to the Seattle area to work for Russell Investments. He then also became a regular contributor
to the Blue Chip Economic Indicators panel of professional forecasters. Mike remained a good
friend of the St. Louis Fed after leaving for Seattle, returning for several short visits to participate in Bank conferences and to work with coauthors.
Mike was hired by Anatol B. (“Ted”) Balbach, who was the St. Louis Fed’s director of
research at the time. Mike was a well-trained econometrician who had demonstrated strong
technical skills, particularly in econometrics and statistics, in his Ph.D. dissertation. However,
Ted was concerned whether Mike would make a good economist. “I know Mike is a good
‘metrician,’” Ted said, “but I don’t know about the ‘econ’ part—is he a good economist?” Mike
clearly was the most talented of the new Ph.D.’s that the St. Louis Fed interviewed that year,
however, so Ted was willing to take a chance on Mike. Ted was not disappointed.
Mike’s research focused on developing new econometric methods and applying those
methods to important economic policy questions. He was not an economic theorist, but he
had a deep understanding of economics and knowledge of financial markets. Mike shared
generously with his colleagues. In seminars, Mike asked penetrating questions. Frequently, he
offered constructive comments and criticism that helped his colleagues build better economic
models. Mike really was the economist that Ted hoped he would be, and then some.
Mike’s understanding of economics, strong technical skills, and friendly manner made
him a sought-after colleague and coauthor. He published over 50 articles on a broad range of
topics. As of January 2014, Mike ranked among the top 5 percent of economists worldwide in
terms of number of publications and measures of publication impact. Mike had coauthors all
over the world. Early in his career, he spent a year at the Swiss National Bank, where he gained
insights into the implementation of monetary policy in different countries and developed close
working relationships with the Bank’s economists that led to several published articles. He also

O

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

iii

In Memoriam: Michael J. Dueker

wrote with several of his St. Louis Fed colleagues on a wide range of topics, including the
impact of price-level shocks on financial stability, Federal Reserve actions to smooth interest
rates, and excess returns in foreign exchange market. Mike continued to do research and write
with coauthors from the St. Louis Fed and elsewhere after he moved to Seattle in 2008.
Many of Mike’s papers were highly technical, but he always found important applications.
In one of his most highly cited works, Mike found that the difference between the yields on
long- and short-term government securities is a good predictor of recessions.1 Mike built on
this research to develop methods of classifying and forecasting the business cycle and other
variables with discrete outcomes. Discrete outcome models parse the business cycle into
phases—say, expansion and recession. One of Mike’s interests was in forecasting transitions
from expansion to recession using currently available data. He recognized that output growth,
employment, and other economic and financial market variables forecast business cycle transitions, while the behavior of those variables in turn reflected the current phase of the cycle.
Mike developed a model he called the Qual-VAR that combines discrete variables such as the
business cycle phase with the commonly used vector autoregression (VAR). The Qual-VAR
led Mike to construct a business cycle index from which he assessed the probability that the
economy would enter recession at some given date in the future.2
Mike also studied monetary policy and financial markets, first by modeling volatility in
financial markets and then by identifying trends in foreign exchange rates. He also studied
how Federal Reserve discount rate changes affected market rates, the setting of federal funds
rate targets, and how explicit inflation targets might improve economic performance.
Mike had a particular interest in modeling the behavior of variables, such as central bank
policy rates and commercial bank prime lending rates, that change by discrete amounts. Mike
proposed a methodology that both greatly simplified estimation of time-dependent classification models (such as the “dynamic probit” model) and incorporated features common to
financial time-series models (autoregressive conditional heteroskedasticity). He applied the
model to study the behavior of commercial bank prime lending rates and showed that periods
of high volatility in the prime rate coincided with market uncertainty about monetary policy.
Further, Mike showed that models of discrete variables that allow for changes in variance outperform constant variance specifications.3 This creative joining of disconnected models in
the literature is something at which Mike excelled.
Mike had a terrific career. He gave sound advice to policymakers and wrote important
research papers. He was an outstanding, well-liked colleague who offered freely of his time
and talents, and he will be missed. ■

NOTES

iv

1

“Strengthening the Case for the Yield Curve as a Predictor of U.S. Recessions.” Federal Reserve Bank of St. Louis
Review, March/April 1997, 79(2), pp. 41-51; http://research.stlouisfed.org/publications/review/article/3183.

2

“Dynamic Forecasts of Qualitative Variables: A Qual VAR Model of U.S. Recessions.” Journal of Business and
Economic Statistics, January 2005, 23(1), pp. 96-104.

3

“Conditional Heteroscedasticity in Qualitative Response Models of Time Series: A Gibbs Sampling Approach to
the Bank Prime Rate.” Journal of Business and Economic Statistics, October 1999, 17(4), pp. 466-72.

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

The Rise and Fall of Labor Force Participation
in the United States
James Bullard

Monetary policy choices going forward are explicitly tied to labor market performance. Hence, the
sharp decline in the labor force participation rate following the 2007-09 recession has become a salient
topic. Presented here are a summary of labor force participation rate data and projections, a survey
of the literature that studies the recent decline in the participation rate, and a view toward fruitful
paths for future research. (JEL E32, E52, J11, J21, J22)
Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 1-12.

abor force participation in the United States has been a controversial subject in current
macroeconomic discussions. In this article, I try to offer my own perspectives on the
issue.
The participation rate—a measure of the number of people actively involved in labor
markets—has generally been a secondary concern in macroeconomics. However, with recent
sharp declines following the financial crisis and recession of 2007-09, it has suddenly become
a salient topic, and one that gets discussed even in non-economic settings.
At its broadest level, the debate about the labor force participation rate is a debate about
the nature of the U.S. economy over the 4½ years since the end of the recession, in the summer
of 2009. Should we characterize the economy as substantially back to normal after a very severe
recession? Or has little progress really been made, so that the economy remains far from its
potential?
There are clear lines of argument on both sides, sometimes blurring political boundaries.
Some suggest that the extraordinary policy response since the end of the recession has been
largely ineffectual, perhaps citing the very flat employment-to-population ratio since 2009,1
and that their own suggested policy responses would have produced better outcomes. Others
emphasize the risk associated with the extraordinary policy response, perhaps citing the Fed’s
now $4.1 trillion balance sheet and the nation’s relatively high debt-to-gross domestic product
(GDP) ratio. Still others argue that the economy has recovered as well as can be expected in
the wake of a major financial crisis, perhaps citing a recovery in real consumption expendi-

L

James Bullard is president and CEO of the Federal Reserve Bank of St. Louis. An earlier version of this article was delivered as a speech to the
Exchequer Club, Washington DC, on February 19, 2014. The author thanks his staff for helpful comments.
© 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views
of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. 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.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

1

Bullard

Figure 1
Unemployment Rate Before, During, and After the 2007-09 Recession
Percent
12

10

8

6

4

2

0
Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09 Jan-10 Jan-11 Jan-12 Jan-13 Jan-14
NOTE: The shaded bar indicates the 2007-09 recession as determined by the National Bureau of Economic Research.
SOURCE: Bureau of Labor Statistics and National Bureau of Economic Research. Last observation January 2014.

tures, an improved housing market, a recovery in equity price valuations, and substantially
lower unemployment. This last group might point to the euro area as an example of an economy that has suffered through a double-dip recession over the past several years, eventually
leading to unemployment rates exceeding 12 percent, while the United States avoided this fate.
Labor market performance is at the heart of the debate over how to characterize the state
of the U.S. economy. While unemployment in the United States was at 10 percent in the fall
of 2009 (Figure 1), it has now declined to 6.6 percent according to the latest reading and has
generally declined much faster than many forecasters anticipated. In tandem with this rosy
development, however, there has been a substantial decline in labor force participation. Some
say that the decline in labor force participation is a bad omen for U.S. macroeconomic performance, with labor market dropouts reflecting frustration with the state of the economy. I call this
the “bad omen” view. Under this interpretation, the decline in the unemployment rate does not
really reflect an improving labor market, and policymakers should look elsewhere to measure
labor market outcomes. Others, however, argue that the decline in labor force participation
simply reflects changing demographics in the U.S. economy and that different demographic
groups have different propensities to participate in market work. As we have different numbers
of people in these different demographic groups, we should naturally expect the aggregate
labor force participation rate to change. I call this the “demographics” view. Under this inter2

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Bullard

pretation, the unemployment rate remains about as good an indicator of overall labor market
health as ever, and recent sharp declines in the unemployment rate should indeed be taken as
indicative of an improving economy and an improving labor market.
In sum, the bad omen view interprets the recent declines in labor force participation as
suggestive of a very weak labor market and discounts the signal coming from recent fasterthan-expected declines in unemployment. The demographics view interprets recent declines
in labor force participation as more benign and takes the signal coming from recent faster-thanexpected declines in unemployment at face value. Since the Federal Open Market Committee
(FOMC) has explicitly tied monetary policy choices to labor market performance, it is of considerable importance which view is more nearly correct.
I offer three perspectives on these questions. First, I simply summarize the data on labor
force participation and provide some background on why this variable has suffered in relative
obscurity until now. Second, I summarize my views on some of the available literature concerning labor force participation as it exists today. In my opinion, this literature is generally supportive of the demographics view, although there are different strands and many issues are not
satisfactorily resolved. Third, I discuss the future of research in this area, which is to move to
more-sophisticated approaches to labor force participation. The more-sophisticated class of
models might be based on the so-called home production literature. Without going into the
details of this approach, I believe that future progress in this area must become more serious
about the incentives of households to supply labor to market work versus nonmarket work.2

THE LABOR FORCE PARTICIPATION RATE
The concept of labor force participation divides the population into three groups: employed,
unemployed, and out of the labor force. This last group could also be called “nonparticipants”
because they are neither working nor searching for market work.
The labor force participation rate is a ratio. In the numerator is the labor force, the sum
of all persons employed and persons unemployed. We think of this group as participants in
the market workplace. Many have jobs, and the rest are looking for jobs. In the denominator is
the civilian noninstitutional population 16 years of age or older—that is, a generally accepted
collection of everyone who conceivably could be working.
Here are some round numbers to keep in mind. The employed group is currently on the
order of 145 million people. The unemployed group is on the order of 10 million people. And
the nonparticipant group is on the order of 91 million people. The groups vary greatly in size;
in particular, the nonparticipant group is large relative to the unemployed group.3 One quirk
of organizing the data this way is that people routinely report moving from nonparticipation
to market work without reporting themselves as unemployed. In other words—at least officially—they were not working and were not searching for a job but nevertheless ended up
working at a job in the next reporting period. Evidently, they were not really properly categorized as “nonparticipants.” I have always found this to be an unsatisfactory aspect of this
method of data organization.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

3

Bullard

Figure 2
Labor Force Participation Rate
Percent
68

66

64

62

60

58

56
Jan-48 Jan-54 Jan-60 Jan-66 Jan-72 Jan-78 Jan-84 Jan-90 Jan-96 Jan-02 Jan-08 Jan-14
NOTE: The shaded bars indicate recessions as determined by the National Bureau of Economic Research.
SOURCE: Bureau of Labor Statistics and National Bureau of Economic Research. Last observation January 2014.

Many discussions of contemporary unemployment forecasts focus on the extent to which
nonparticipants will rejoin the labor market. During the late 1990s, for example, many workers
seemed to come off the sidelines into the workplace because of an exceptionally strong economy. At the Federal Reserve Bank of St. Louis, we have constructed unemployment forecasts
in recent years assuming that movements from nonparticipation to employment would be
minimal while unemployment was at relatively high levels. This has served us well, as we
have more accurately predicted declines in unemployment in the past year than many other
forecasters.4
If you know only one aspect of the data on labor force participation, it should be this:
Labor force participation used to be relatively low. It rose during the 1970s, 1980s, and 1990s;
peaked in 2000; and has generally been declining since 2000 (Figure 2).
From 1948 to 1966, the labor force participation rate was relatively low and relatively stable,
averaging 59.1 percent—substantially lower than today’s value of 63 percent. It is important
to note that we normally consider the U.S. economy to have performed relatively well during
this period, especially during the long expansion of the 1960s. Evidently, low labor force
participation does not equate with weak economic growth. Surely this is because the factors
driving economic growth differ from those driving labor force participation.
4

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Bullard

After about three decades of trending upward, the labor force participation rate peaked
in the first half of 2000 at 67.3 percent. The rate of increase was slower in the 1990s than in
the 1970s or 1980s. The peak was more than 8 percentage points higher than the average level
during 1948-66. Many labor force participation studies during this period focused on the
increasing participation rates of women. However, whatever effects came from that source,
or any other source, the labor force participation rate could not continue to increase forever.
Households make choices about how much labor to supply given current wages and work
environments, and women newly joining the labor force would find the right level of participation and stop there.
Since 2000, the labor force participation rate has generally been declining. The pace of
decline was particularly sharp during the 2007-09 recession, but the participation rate also
declined steadily in the early 2000s and since the end of the recession in mid-2009.
The general picture, then, is one of a hump shape in U.S. aggregate labor force participation during the postwar era. A satisfactory theory must account for this hump shape. One way
to build such a theory is to appeal to demographics. The nation’s workforce had a younger
profile as the Baby Boom generation came of age, and it will have an older profile as the Baby
Boom generation continues to retire. Since different age groups have different propensities to
participate, this suggests a promising avenue to explain the labor force participation data.
I daresay that the demographic explanation is the gut instinct of many macroeconomists.
This is why labor force participation sits in the backseat of many macroeconomic models.
Many, including me, might reason that a good demographic model combined with more women
in the labor force during the 1970s, 1980s, and 1990s could explain a very large fraction of the
hump-shaped movements in aggregate labor force participation over the postwar era. If such
a model were fitted to the data, only a small amount of variation in the participation rate would
remain to be explained. That small remaining amount of variation might be attributable to
business cycle (cyclical) effects, or it might just be noise about the fundamental hump-shaped
trend. Relatively minor cyclical effects on labor force participation would likely be too small
to have major macroeconomic implications given everything else going on in a macroeconomic
model. Consequently, it might seem that we do not need to worry too much about the labor
force participation rate for business cycle purposes.
But all this is just in the heads of macroeconomists. I now turn to some of the recent
research on labor force participation to determine the extent to which such a theory has
actually been devised.

RECENT RESEARCH ON LABOR FORCE PARTICIPATION
Let’s start with the Bureau of Labor Statistics (BLS). The BLS is, of course, very close to
the data and it routinely projects labor force participation over the medium term. In general,
its medium-term forecasts from the mid-2000s proved to be too high, meaning that its forecast
labor force participation rate was considerably higher than the values actually observed. More
recent medium-term BLS forecasts call for a declining rate of participation over the next decade
or so, all the way down to 61.6 percent in 2022 (Figure 3).5 Recall that today’s participation rate
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

5

Bullard

Figure 3
Labor Force Participation Rate: BLS Data and Projections
Percent
68

Data
Toossi (2013) projection

67
66
65
64
63
62
61
60
1990

1993

1996

1999

2002

2005

2008

2011

2014

2017

2020

SOURCE: Bureau of Labor Statistics and Toossi (2013). Last observation 2013.

is 63.0 percent, so the rate is projected to continue to decline by around 15 basis points per year.
According to BLS projections,6 more than 70 percent of this decline is due to purely demographic factors—that is, changes in population shares by age groups, assuming unchanged
participation rates for each group.7
To the extent that this forecast pans out, the basic direction for the labor force participation rate is down, not up. Those waiting for an upward swing in labor force participation as
the economy continues to expand will be disappointed, on average, if this forecast comes to
pass. I read the BLS work as supportive of the demographics hypothesis I described previously.
Fujita (2013) provides some additional insight concerning the decline in aggregate U.S.
labor force participation since 2000. Fujita’s calculations suggest that about 65 percent of the
decline in the participation rate was due to retirements and disability. Fujita points out that
the empirical evidence suggests members of these groups have only a small probability of
returning to the labor force. If we limit attention only to a period of relatively high economic
stress, such as 2007:Q1 to 2011:Q2, we do see more of the decline in participation attributable
to discouraged workers; but, even then, this is only about 25 percent according to Fujita’s calculations. Over a less stressful period, such as 2012:Q1 to 2013:Q2, the entire decline in the
aggregate labor force participation rate is attributable to retirements, with no effect at all coming from an increase in discouraged workers. I read Fujita’s contribution as also supportive of
the demographics hypothesis.
6

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Bullard

Figure 4
Labor Force Participation Rate: BLS Data and Aaronson et al. (2006) Projections
Percent
68

Data
Aaronson et al. (2006) projection

67
66
65
64
63
62
61
60
1990

1993

1996

1999

2002

2005

2008

2011

2014

SOURCE: Bureau of Labor Statistics and Aaronson et al. (2006). Last observation 2013.

Davig and Mustre-del-Río (2013) provide some analysis of the “shadow” labor supply to
gain insight into whether this group is likely to return to the labor force. The shadow group is
defined as those who want a job but are not actively seeking one. The authors document that
this group is demographically similar to the unemployed. They suggest that any impact on
aggregate labor force participation from this group is likely to be small, because flows from
this group to unemployment are small and less likely to occur as the recovery continues. I read
this as also supportive of the demographics view.
In a somewhat older paper, Aaronson, Fallick, Figura, Pingle, and Wascher (2006) examined the decline in labor force participation following the 2001 recession and tried to ascertain
how much of the decline at that time was cyclical. It is perhaps important to recall that there
was an earlier debate on declining labor force participation, long before the deep recession of
2007-09. Their paper contains as part of the analysis an empirical model of the trend labor force
participation rate that includes demographic factors. If that trend model is projected forward
to today from 2006, it predicts nearly exactly the labor force participation rate observed in 2012
and 2013 (Figure 4). What a great piece of out-of-sample forecasting! I read this as supportive
of the demographics view. This model also projects continued decline in the labor force participation rate in the years ahead.
Kudlyak (2013) follows up on the empirical model proposed by Aaronson et al. (2006).
Again, the model contains key demographic information such as age, gender, and birth-year
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

7

Bullard

cohort effects. The model suggests that current aggregate labor force participation rates are
not far off from the model’s predicted trend participation rate. Again, I read this as supportive
of the demographic view.8
Erceg and Levin (2013) study the intersection between the labor force participation rate
and monetary policy. Their paper is a “thinking outside the box” exercise. In what I have presented so far, there is a certain inevitable logic. I stated that the data on labor force participation
cry out for an explanation based mostly on increasing labor force participation by women and
slowly changing demographics. The existing literature more or less provides such an explanation. Erceg and Levin instead ask whether there are other ways to think about this issue. They
present evidence from U.S. states on prime-age males and suggest that the declines in labor
force participation after the 2007-09 recession for this particular group were mostly cyclical.
The authors then ask how monetary policy might be conducted in a world where labor force
participation has an important cyclical component. They suggest that the participation decision
should receive more attention in monetary policy research, a point on which I will agree below.
I do not find the evidence on cyclical versus structural changes in labor force participation
in Erceg and Levin (2013) as persuasive as the other empirical work I have reviewed.9 Labor
force participation for prime-age males, for instance, has also been on a secular decline for
many years. Nevertheless, Erceg and Levin’s points about how to conduct monetary policy in
a world with important cyclical components in labor force participation are well made.10 In
fact, I will argue that incorporating home production in economic models—as Erceg and Levin
do—is where the future lies.
Some authors report somewhat higher estimates of the fraction of the decline in labor
force participation since 2000 due to cyclical factors. For instance, Aaronson, Davis, and Hu
(2012) use still another empirical model with demographic factors included and conclude that
more than half of the decline in aggregate labor force participation from 2000 to 2011 is due
to cyclical factors. Van Zandweghe (2012) tries an alternative method of decomposing the data
from 2007 to 2011 and concludes that more than half of the decline is cyclical. The approach
used by Hotchkiss and Rios-Avila (2013) emphasizes nonlinear factors following the severe
2007-09 recession; these authors conclude that nearly all of the decline in aggregate labor force
participation following the recession was cyclical. Bengali, Daly, and Valletta (2013) study the
correlation in the changes in employment and labor force participation in state-level data to
gain insight; they conclude that a substantial cyclical component exists in the observed aggregate decline in labor force participation.
I am not necessarily swayed by these alternative approaches or results. But they certainly
do show that there are many ways to cut the data and interpret the findings. This leads me to
my final remarks—namely, where should the literature on labor force participation go next?

HOME PRODUCTION AS THE FUTURE
So far, I have reviewed some interesting economic literature on a topic that has been hot,
not just among economists, but also among politicians, the media, financial markets, and
even others who are not normally close students of macroeconomic developments. Much of
the literature I have reviewed uses the same basic idea: Certain demographic groups have a
8

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Bullard

certain propensity to participate in market work, and one of the main things we need to do as
economists is project the number of people in each of these groups to determine a reasonable
estimate of the expected (or “normal” or “trend”) labor force participation rate in the U.S. economy. Much of the literature concludes that demographics have contributed substantially to
the observed decline in U.S. labor force participation since 2000.
Still, most of the literature I reviewed is a bit hollow. Simply saying that people in certain
demographic groups tend to make the participation decision one way or another does not do
enough to analyze the incentives of household labor supply decisions. The more we know about
the details of the household labor supply choices, including choices to participate in market
work, the better we can predict the impact of policy on labor force participation. Furthermore,
we would like these decisions to be part of the macroeconomic model, as Erceg and Levin
(2013) suggest.11
There is one strand of the literature that does provide a more complete picture of household incentives to supply labor and participate in labor markets: the literature on so-called
home production. We need not go into the details here, but the idea is simple. Think of a household as the owner of capital and labor. The household members combine their home capital—
refrigerators, ovens, dishwashers, cars, houses—with their labor time to produce home goods,
such as a trip, a meal, or some child care.12 These goods are not acquired in the market and
are not counted in GDP, but they matter to the household. The home labor provided does not
count in the aggregate statistics on labor supply. The household then has to make decisions
about how much time to supply to market work versus work at home, including how many
members of the household should participate in market work.13 If we were to add to a household production model more explicit treatment of household retirement decisionmaking, in
addition to decisions by younger households to acquire human capital, we would get to a more
complete model of the labor force participation rate.
This approach is much more detailed regarding household decisionmaking than the
research described above. But the extra complexity comes with a benefit, as the approach also
allows macroeconomists to better understand the factors driving household labor supply
decisions in terms of actual options inside the home, as well as with respect to the informal
labor market. More-detailed models in this direction will likely be necessary in the future if
we want to move the debate on labor force participation forward.14
Some researchers have made progress in this direction. Tripier (2004) analyzes a real business cycle version of the Diamond-Mortensen-Pissarides search and matching model that
incorporates a home production decision. Tripier concludes that such a model has counterfactual implications for the unemployment and participation rates, while it can account for
the behavior of the employment rate over the business cycle. Veracierto (2008) confirms
these counterfactual implications in a richer model with endogenous job-acceptance and
job-separation decisions. Tüzemen (2012) extends Tripier’s analysis to allow for on-the-job
search and, hence, job-to-job transitions. This model performs better in matching the business cycle features of the major labor market variables.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

9

Bullard

CONCLUSION
While the unemployment rate has declined in recent years, labor force participation has
also been declining, perhaps suggesting that unemployment is not as reliable an indicator of
macroeconomic performance as it may have been in the past. Here I have given three perspectives on labor force participation: First, I reviewed the data; second, I reviewed the literature;
and third, I suggested directions for future research.
The post-WWII data on aggregate U.S. labor force participation show a hump-shaped
pattern. Participation rose in the 1970s, 1980s, and 1990s before peaking in 2000 and heading
into decline up until the present day. Current BLS projections suggest that this decline will
continue over the coming decade. The rise in labor force participation is often attributed in
part to the maturing of the Baby Boom generation, as well as to the increase in the number of
women in the workforce. The decline has often been attributed to the aging of the U.S. labor
force. A satisfactory model has to account for the rise and fall over many decades. A demographically based model would seem to have a good chance of success in explaining these data.
I reviewed some of the available literature on this topic. My view of the literature is that
carefully constructed, demographically based empirical models of the hump-shaped trend in
the U.S. labor force participation rate do a good job of explaining the data. These models suggest that the current participation rate is not far from the predicted trend. This means, in turn,
that the cyclical component in labor force participation is likely to be relatively small. To the
extent these models are correct, then, the observed unemployment rate remains as good an
indicator of overall labor market health as it has been historically. In particular, the recent,
relatively rapid declines in unemployment can be understood as representing an improving
labor market. This is the judgment that should inform monetary policy going forward.
The literature is not completely satisfactory, however. I discussed how researchers might
gain additional insight by including more detailed household decisionmaking inside economic
models. This would allow us to better understand what motivates or deters labor force participation. I look forward to seeing future research pushing in this direction. ■

10

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Bullard

NOTES
1

For a recent analysis of the employment-to-population ratio as a labor market indicator, see Kapon and Tracy (2014).

2

Tripier (2004), Veracierto (2008), and Tüzemen (2012) take important steps in this direction.

3

See Canon, Kudlyak, and Reed (2014) for an analysis of the relative magnitude of the flows to employment from
unemployment and nonparticipation.

4

See Bullard (2014).

5

See Toossi (2013).

6

Author’s calculation based on figures from Toossi (2013).

7

For more on this topic, see Canon, Debbaut, and Kudlyak (2013).

8

A forthcoming working paper by Hornstein and Kudlyak (2013) includes a more elaborate version of this model.
Their main finding remains that current labor force participation rates are close to what would be predicted by an
empirical model with carefully constructed demographic factors.

9

For more on this issue, see Canon, Debbaut, and Kudlyak (2013).

10 I also largely agree with the points made by Orphanides (2013) in a comment on the paper, in effect that the new

labor market slack indicators proposed by Erceg and Levin would be subject to considerable uncertainty and
could lead policymakers badly astray.
11 For an example of a detailed macroeconomic model with an explicit participation decision that has an impact on

policy recommendations, see Imrohoroğlu and Kitao (2012).
12 Time use surveys, such as the American Time Use Survey (ATUS) conducted by the BLS, provide a wealth of data to

quantify labor supplied to home production. For example, see Aguiar and Hurst (2007).
13 For an example of the different perspective the home production literature provides on issues in monetary eco-

nomics, see Aruoba, Davis, and Wright (2012).
14 For an example of the interplay between home production and labor force participation, see Greenwood,

Seshadri, and Yorukoglu (2005).

REFERENCES
Aaronson, Daniel; Davis, Jonathan and Hu, Luojia. “Explaining the Decline in the U.S. Labor Force Participation
Rate.” Federal Reserve Bank of Chicago Chicago Fed Letter, No. 296, March 2012;
http://www.chicagofed.org/digital_assets/publications/chicago_fed_letter/2012/cflmarch2012_296.pdf.
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.” Brooking Papers on Economic
Activity, Spring 2006, 37(1), pp. 69-134;
http://www.brookings.edu/~/media/projects/bpea/spring%202006/2006a_bpea_aaronson.pdf.
Aguiar, Mark and Hurst, Erik. “Measuring Trends in Leisure: The Allocation of Time over Five Decades.” Quarterly
Journal of Economics, August 2007, 122(3), pp. 969-1006; http://www.jstor.org/stable/25098866.
Aruoba, S. Boragan; Davis, Morris A. and Wright, Randall. “Homework in Monetary Economics: Inflation, Home
Production, and the Production of Homes.” NBER Working Paper No. 18276, National Bureau of Economic
Research, August 2012; http://www.nber.org/papers/w18276.
Bengali, Leila; Daly, Mary and Valletta, Rob. “Will Labor Force Participation Bounce Back?” Federal Reserve Bank of
San Francisco Economic Letter, 2013-14, May 13, 2013;
http://www.frbsf.org/economic-research/publications/economic-letter/2013/may/will-labor-force-participationbounce-back/.
Bullard, James. “Ghosts of Forecasts Past and Future.” Remarks delivered at the Indiana Bankers Association
Economic Outlook Forum Luncheon, Indianapolis, IN, January 10, 2014;
http://research.stlouisfed.org/econ/bullard/pdf/Bullard-IN-Bankers-Association-January-10-2014-Final.pdf.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

11

Bullard
Canon, Maria; Debbaut, Peter and Kudlyak, Marianna. “A Closer Look at the Decline in the Labor Force Participation
Rate.” Federal Reserve Bank of St. Louis Regional Economist, October 2013, 21(4), pp. 10-11;
http://www.stlouisfed.org/publications/pub_assets/pdf/re/2013/d/labor_force.pdf.
Canon, Maria; Kudlyak, Marianna and Reed, Marisa. “Not Everyone Who Joins the Ranks of the Employed Was
‘Unemployed.’” Federal Reserve Bank of St. Louis Regional Economist, January 2014, 22(1), pp. 14-16;
http://www.stlouisfed.org/publications/pub_assets/pdf/re/2014/a/unemployed.pdf.
Davig, Troy and Mustre-del-Río, José. “The Shadow Labor Supply and Its Implications for the Unemployment Rate.”
Federal Reserve Bank of Kansas City Economic Review, Third Quarter 2013, pp. 5-29;
http://www.kc.frb.org/publicat/econrev/pdf/13q3Davig-Mustre.pdf.
Erceg, Christopher J. and Levin, Andrew T. “Labor Force Participation and Monetary Policy in the Wake of the Great
Recession.” IMF Working Paper No. 13/245, International Monetary Fund, July 2013;
http://www.imf.org/external/pubs/ft/wp/2013/wp13245.pdf.
Fujita, Shigeru. “On the Causes of Declines in the Labor Force Participation Rate.” Federal Reserve Bank of
Philadelphia Research Rap Special Report, November 2013; revised February 6, 2014;
http://philadelphiafed.org/research-and-data/publications/research-rap/2013/on-the-causes-of-declines-in-thelabor-force-participation-rate.pdf.
Greenwood, Jeremy; Seshadri, Ananth and Yorukoglu, Mehmet. “Engines of Liberation.” Review of Economic Studies,
January 2005, 72(1), pp. 109-33; http://restud.oxfordjournals.org/content/72/1/109.full.pdf+html.
Hornstein, Andreas and Kudlyak, Marianna. “Estimating the Trend in the LFP Rate.” Federal Reserve Bank of
Richmond, forthcoming working paper, December 2013.
Hotchkiss, Julie L. and Rios-Avila, Fernando. “Identifying Factors behind the Decline in the U.S. Labor Force
Participation Rate.” Business and Economic Research, June 2013, 3(1), pp. 257-75;
http://www.macrothink.org/journal/index.php/ber/article/view/3370/2921.
Imrohoroğlu, Selahattin and Kitao, Sagiri. “Social Security Reforms: Benefit Claiming, Labor Force Participation, and
Long-Run Sustainability.” American Economic Journal: Macroeconomics, July 2012, 4(3), pp. 96-127;
http://www.aeaweb.org/articles.php?doi=10.1257/mac.4.3.96.
Kapon, Samuel and Tracy, Joseph. “A Mis-Leading Labor Market Indicator.” Federal Reserve Bank of New York Liberty
Street Economics (blog), February 3, 2014;
http://libertystreeteconomics.newyorkfed.org/2014/02/a-mis-leading-labor-market-indicator.html.
Kudlyak, Marianna. “A Cohort Model of Labor Force Participation.” Federal Reserve Bank of Richmond Economic
Quarterly, First Quarter 2013, 99(1), pp. 25-43;
http://www.richmondfed.org/publications/research/economic_quarterly/2013/q1/pdf/kudlyak.pdf.
Orphanides, Athanasios. “Discussion of ‘Labor Force Participation and Monetary Policy in the Wake of the Great
Recession’ by Erceg and Levin.” Delivered at the Federal Reserve Bank of Boston 57th Economic Conference
Fulfilling the Full Employment Mandate, April 2013.
Toossi, Mitra. “Labor Force Projections to 2022: The Labor Force Participation Rate Continues to Fall.” Monthly Labor
Review, December 2013, pp. 1-28;
http://www.bls.gov/opub/mlr/2013/article/pdf/labor-force-projections-to-2022-the-labor-force-participationrate-continues-to-fall.pdf.
Tripier, Fabien. “Can the Labor Search Model Explain the Fluctuations of Allocations of Time?” Economic Modelling,
January 2004, 21(1), pp. 131-46; http://www.sciencedirect.com/science/article/pii/S0264999302000871.
Tüzemen, Didem. “Labor Market Dynamics with Endogenous Labor Force Participation and On-the-Job Search.”
Working Paper No. 12-07, Federal Reserve Bank of Kansas City, October 2012;
http://www.kansascityfed.org/publicat/reswkpap/pdf/rwp12-07.pdf.
Van Zandweghe, Willem. “Interpreting the Recent Decline in Labor Force Participation.” Federal Reserve Bank of
Kansas City Economic Review, First Quarter 2012, pp. 5-34;
http://www.kc.frb.org/publicat/econrev/pdf/12q1VanZandweghe.pdf.
Veracierto, Marcelo. “On the Cyclical Behavior of Employment, Unemployment and Labor Force Participation.”
Journal of Monetary Economics, September 2008, 55(6), pp. 1143-57;
http://www.sciencedirect.com/science/article/pii/S0304393208000998.

12

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

The Great Trade Collapse and Rebound:
A State-by-State View
Cletus C. Coughlin

During the Great Trade Collapse in the United States, which began in late 2008, one concern was that
such a large collapse would transform exporting firms into strictly domestic firms or, worse, drive them
out of business. In either case, it was feared that U.S. exporting might, at best, revive slowly. However,
this fear about long-lived export impacts did not materialize. Clearly there were large export effects,
but the sharp decline was quickly followed by a sharp rebound that began in mid-2009. In contrast to
previous research, this study examines this historic episode from the perspective of individual states.
A comparison of intensive and extensive trade margins reveals that the adjustment occurred to a
greater extent on the intensive than on the extensive trade margin. In other words, the adjustment
process entailed relatively larger changes in average exports per firm than in the number of exporting
firms. It is likely much easier to alter existing export levels than to, first, eliminate all exports by a firm
and, second, either restart exports by this firm or become a new entrant into exporting. The bottom
line is that the U.S. export sector weathered the challenges associated with the global recession and
financial crisis quite well. The fact that relatively large firms dominate U.S. exports likely contributes
to the resiliency of the U.S. export sector. (JEL F10, F14, N72, R12)
Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 13-33.

n 2008, U.S. exports of goods totaled $1,147.4 billion in chained 2005 dollars (Figure 1).
As part of the sudden, steep, and synchronized decline in trade worldwide, termed the
“Great Trade Collapse” (GTC) by Baldwin (2009), U.S. exports declined 13.4 percent to
$993.9 billion in 2009. At the same time, U.S. gross domestic product (GDP) declined 2.8
percent. This collapse was short-lived, as export growth rebounded sharply and exports in
2010 totaled $1,142.2 billion, which is roughly the 2008 level.1 This 14.9 percent growth in
exports substantially exceeded the 2.5 percent increase in GDP during 2010.
This episode in trade history, unprecedented in U.S. post-World War II economic history,
affected countries throughout the world and has attracted the attention of many researchers.2
Most analyses have relied on country-level data and focused on explanations for the GTC and,
to a lesser extent, the subsequent rebound.

I

Cletus C. Coughlin is senior vice president and chief of staff at the Federal Reserve Bank of St. Louis. The author appreciates the research assistance
of Li Li and Diana Cooke.
© 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views
of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. 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.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

13

Coughlin

Figure 1
U.S. Exports of Goods
Billion, SA Chained U.S. Dollars, 2005
1,400
1,300
1,200
1,100
1,000
900
800
2006

2007

2008

2009

2010

2011

2012

NOTE: The export data are Census-based rather than balance of payments-based. SA, seasonally adjusted.

Crowley and Luo (2011) examine three primary hypotheses for the GTC.3 First, and likely
the key explanation, is that trade collapsed as a result of a decline in aggregate demand. The
fact that trade declined proportionately far more than GDP suggests a demand shock explanation that also accounts for different types of goods, vertical specialization, and inventory
adjustments is likely to be important.4,5 For example, Eaton et al. (2011) find that more than
80 percent of the decline in the ratio of trade to GDP resulted from a spending shift away from
manufactured goods, particularly durable goods.
A second explanation stresses the disruption in the supply of traded goods because of the
increased difficulties in securing trade finance during the financial crisis. Generally speaking,
most would agree that trade finance conditions deteriorated but that their impact was much
less important than the decline in demand. Chor and Manova (2012) find evidence that financing difficulties were a contributing factor, but pinning down their quantitative importance is
difficult, partially because of a lack of data. Small- and medium-sized firms were probably
affected more by credit constraints than larger firms, which do the bulk of exporting.
A third explanation focuses on increased trade barriers. Despite a frequent finding that
import restrictions tend to increase during periods of economic weakness, empirical evidence
suggests this explanation is quantitatively unimportant in understanding the GTC.6 The rapid
rebound in trade also casts doubt on the importance of the import restrictions argument.
My focus is not on the causes of the GTC, but rather on the international trade experiences
of U.S. states during the collapse (2008-09) and the rebound (2009-10). Most importantly, the
goal is to provide information on the relative importance of the adjustments on the extensive
(i.e., the number of exporting firms) and intensive (i.e., the average exports per firm) trade
margins. A close look at these margins might provide insights into reasons for the sharp
rebound in export growth. Such a sharp rebound was initially viewed as unlikely by many
14

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

because the very large declines in exports during the GTC raised the possibility of long-lived
adverse effects on trading relationships. Examining the extensive margin shows how the number of exporting firms changed during the collapse and rebound. Examining the intensive
margin shows how the average exports of firms changed. Not surprisingly, I find changes on
both margins; however, changes on the intensive margin are found to be relatively more important than changes on the extensive margin during both the collapse and the rebound.
Prior to presenting the analysis, the next section provides a description of the export data
used and some basic facts about states and their exporting performance. This description is
followed by an analysis of trade margins at the state level. Next, a closer look at state-level
exports and their respective trade margins over time and across countries is undertaken. A
summary of results and conclusion complete the article.

A NATIONAL OVERVIEW USING STATE EXPORT DATA
Data on exports by U.S. states to foreign destinations are from the Origin of Movement
series.7 These data are compiled by the Foreign Trade Division of the U.S. Census Bureau.
The data in this series identify the state from which an export begins its journey to a foreign
country. Beginning in 1987, the Origin of Movement series provides the current-year export
sales, or free-alongside-ship costs if not sold, for all 50 U.S. states to 242 foreign destinations
(generally countries).8 These export sales are for merchandise sales only and do not include
services exports.
Since 1992, the Foreign Trade Division of the U.S. Census Bureau has issued an annual
report providing a profile of exporting companies. These reports are based on economic census
and survey data on file at the Census Bureau, administrative records from other government
agencies, and documents filed for export clearances. Until recently, only exporting companies
were profiled, but both exporters and importers are now profiled.9 My analysis uses data for
2008, 2009, and 2010 and requires the use of state data—most importantly, state-destination
data. The profile contains information on identified companies and their export values. This
linkage generates what are termed “known export values.” As a result, the data in Table 1 contain entries for total exports, some portion of which cannot be linked to individual exporters,
as well as the portion that can be linked.10
In 2008, the year prior to the majority of the trade collapse, identified exporters accounted
for $1,150.9 billion in current-dollar exports, or 89.4 percent of the total value of exported
goods. During 2009, known exports declined 18.3 percent. From 2008 to 2009, the number
of identified exporters declined from 289,711 to 276,643, a decrease of 4.5 percent. Turning
to the rebound, in 2010 identified exporters accounted for $1,140.4 billion in exports or 89.2
percent of the total value of exported goods.11 Relative to 2009, known exports rose 21.3 percent in 2010. Meanwhile, the number of identified exporters rose from 276,643 to 293,988, an
increase of 6.3 percent.
Using these national data and subsequently adjusting the export values for price changes
reveals that adjustments on the intensive margin exceeded adjustments on the extensive margin
during both the trade collapse and trade rebound.12 For example, during the trade collapse,
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

15

Coughlin

Table 1
Exports by Identified Companies
U.S. Exports and Exporters

2008

2009

2010

Total exports ($ bil.)

1,287.4

1,056.0

1,278.3

Known value (Identified exporters) ($ bil.)

1,150.9

940.4

1,140.4

Known value/Total exports (%)
No. of identified exporters

89.4

89.1

89.2

289,711

276,643

293,988

SOURCE: U.S. Census Bureau. A Profile of U.S. Importing and Exporting Companies, 2010-2011 and prior reports in this
series.

known exports declined 13.6 percent. A straightforward decomposition of this total change
shows that the number of exporting firms declined 4.5 percent and average exports per firm
declined 9.6 percent.13 Thus, the decline in the intensive margin was more than double that
in the extensive margin.14 A similar, but not as stark, finding pertains to the trade rebound
when known exports increased 15.1 percent. Exporting firms increased by 6.3 percent and
average exports per firm increased by 8.3 percent. Adjustments clearly occurred on both margins, but these findings suggest that the intensive margin accounts for the majority of the
adjustments during both the collapse and the rebound.
This finding is similar to the lack of destruction of trade relationships during the GTC in
other countries—France (see Fontagné and Gaulier, 2009), Japan (see Wakasugi, 2009), and
Belgium (see Behrens, Corcos, and Mion, 2013). This evidence is also consistent with findings
by Schott (2009) for other U.S. “trade shock” episodes. Schott used detailed, firm-level trade
data to analyze U.S. firms’ responses to the 2001 recession and the 1997 Asian financial crisis
and found that the collapse of trade nationally was driven primarily by changes in the intensive
margin. In other words, trade fell because firms sold less of what they were already selling
rather than eliminating trade altogether.15 By using state-level data, I hope to provide additional insights as to the regularity of this finding across states.
I highlight some well-known facts about exporting firms and export markets to provide
additional background for the analysis. Despite being outnumbered by smaller firms, relatively
large firms dominate U.S. exports. In fact, 287,498 (97.8 percent) of the total 293,988 exporters
in 2010 had fewer than 500 employees, while 6,490 exporters (2.2 percent) had 500 or more
employees. Nonetheless, large firms (>500 employees) accounted for 66.2 percent of known
exports, while relatively small firms (<500 employees) accounted for the remaining 33.8 percent of known exports. In addition, a relatively small number of firms account for the majority
of exports. For 2010, the top 50 exporters (roughly 0.02 percent of all exporters) accounted
for 29.0 percent of all known exports, and the top 2,000 exporters (roughly 0.68 percent of all
exporters) accounted for 76.9 percent of all known exports.
Not surprisingly, relatively large exporters tend to export to more countries than their
relatively smaller counterparts. Slightly more than one in four large exporters (26.1 percent)
shipped to only one country. These exports accounted for 0.4 percent of the exports of large
16

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

exporters. Meanwhile, smaller exporters were more likely to ship to only one country; as a
result, the percentages of these firms and their export shares were larger than those of their
larger counterparts. As the number of export destinations increases, the relative importance
of large exporters increases. For 2010, 69.3 percent of the exports of large firms were shipped
to 50 or more countries. Meanwhile, very few small exporters (roughly 430, or 0.15 percent of
all small exporters) shipped to 50 or more countries.
Finally, the top 25 export destinations account for most U.S. exports. For 2010, the top 25
markets accounted for 82.9 percent of U.S. exports.16 A similar percentage of U.S. exporters,
81.8 percent, ship to these 25 markets. Not surprisingly, large exporters handle the majority
(67.4 percent) of these exports, despite representing a relatively small share (2.6 percent) of
the total number of exporters supplying these 25 markets.
Ideally, for the analyses in this article, individual firm data would be used to allow examination of the trading behavior of individual firms. Such data would allow observation of the
beginning or ending of a firm’s trade involving a specific good or destination and the change
in the value of a firm’s ongoing trade involving a specific good or destination. In the present
case, I have state-level data, so I am limited to using the number of firms in a state that export
to a specific country.
The state-level data limitation highlights the distinction between plants and firms. Because
a firm with plants in different states could export to a given country from plants in different
states, the sum of the number of exporters over all states would exceed the number of exporters
nationally. Also, a firm with plants in multiple states might export to different countries depending on the plant. If the exporters over all countries were summed, this sum would exceed the
number of exporters at the national level. Thus, the analysis here combines single-plant firms
with multiplant firms.
Why might combining single- and multiplant firms matter? The response to a decline (or
an increase) in foreign demand can differ between a single-plant firm and a multiplant firm.
For example, assume a firm has plants in two states and that both plants produce exports for
the same foreign country. In response to a decline in foreign demand, the firm might choose
to serve the foreign country from one plant rather than two. As a result, from the firm’s perspective, the adjustment is completely on the intensive margin, while from a plant’s perspective,
the adjustment differs across states. From the perspective of one state, there is a decline on the
extensive margin and an uncertain change on the intensive margin, while from the perspective
of the other state, there is no change on the extensive margin. Any change is on the intensive
margin. The key point of this example is that the dynamics based on firms can differ from the
dynamics based on plants. Any interpretation of the results in this article must keep this possibility in mind.
Table 2 contains summary information, by state, for the number of foreign destinations
with export activity.17 Virtually every state serves the top 30 export markets. Moreover, even
during the GTC, more often than not the number of export destinations tended to increase.
Relative to 2008, 27 states in 2009 experienced an increase in export destinations, 18 states a
decrease, and 5 states no change. Especially noteworthy is that every state in New England
added 30 or more export destinations. Only one state, North Dakota, suffered a double-digit
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

17

Coughlin

Table 2
Number of Export Destinations by State
State
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
Florida
Georgia
Hawaii
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming

Total level
2008

Change
2008-09

Total level
2009

Change
2009-10

Total level
2010

129
34
143
101
184
128
138
87
178
165
44
84
166
142
120
125
117
137
96
157
156
139
150
108
131
61
101
109
106
166
79
172
157
79
162
118
124
161
95
144
64
144
176
123
86
162
158
76
148
43

–9
2
–7
–7
1
–2
37
–2
–1
2
–2
–2
20
33
0
–1
9
1
53
–8
30
41
1
0
1
–3
6
2
46
24
3
19
–1
–11
27
–4
1
27
41
1
0
0
–1
5
43
–3
–6
–2
32
0

120
36
136
94
185
126
175
85
177
167
42
82
186
175
120
124
126
138
149
149
186
180
151
108
132
58
107
111
152
190
82
191
156
68
189
114
125
188
136
145
64
144
175
128
129
159
152
74
180
43

48
49
43
53
8
49
–2
49
15
22
43
55
2
–5
45
49
39
36
9
30
–2
–5
28
48
37
42
47
37
–2
–5
57
–2
27
48
–4
46
52
4
–1
28
44
38
15
42
1
26
34
36
–2
48

168
85
179
147
193
175
173
134
192
189
85
137
188
170
165
173
165
174
158
179
184
175
179
156
169
100
154
148
150
185
139
189
183
116
185
160
177
192
135
173
108
182
190
170
130
185
186
110
178
91

SOURCE: Author’s calculations using WISER data.

18

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

Figure 2
Export Destinations and State Size (2010)
No. of Countries
210
190
170
150
130
110
90
70
50
9

10

11

12

13

14

15

ln(gross state product)
SOURCE: WISER and Bureau of Economic Analysis.

loss (11) in the number of export destinations. Not surprisingly, in light of the rebound, the
vast majority of states (40) served more export destinations in 2010 than in 2009. No state
suffered a loss of more than five export destinations.
Figures 2 through 4 show some noteworthy differences across states. Figure 2, which uses
2010 data, shows the positive relationship between state size (gross state product in natural
logarithms) and the number of destination countries.18 In other words, larger states tend to
export to more countries. The three largest U.S. states—California, New York, and Texas—
export to many more countries than the five smallest states—South Dakota, Wyoming,
Montana, North Dakota, and Vermont. Figure 2 also shows that the rate of increase in the
number of destination countries decreases as state size increases. For the largest states, the
number of destination countries approaches 200.
Figure 3 shows that larger foreign countries (higher GDPs in natural logarithms) tend to
import from a larger number of U.S. states. Moreover, once a country’s GDP reaches a certain
size, it generally imports from all 50 states.19 All countries with GDPs larger than $401 billion
import from every state, while the 10 countries with the smallest GDPs imported from an
average of slightly more than 21 states.
Figure 4 shows the number of exporters in a state is related positively to state size (gross
state product in natural logarithms).20 For example, the number of exporters in the largest
states—California (72,092 exporters), New York (40,377), and Texas (38,276)—exceeds the
number in the smallest states—South Dakota (965 exporters), Wyoming (421), Montana
(1,539), North Dakota (1,870), and Vermont (1,244)—by a factor of more than 20. Figure 4
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

19

Coughlin

Figure 3
States and Export Destination Size (2010)
No. of States
50
45
40
35
30
25
20
15
10
5
0
15

17

19

21

23

25

27

29

ln(GDP)
SOURCE: WISER and the World Bank.

Figure 4
Exporters and State Size (2010)
No. of Exporting Firms in a Given State
80,000
70,000
60,000
50,000
40,000
30,000
20,000
10,000
0
10

11

12

13

14

15

ln(gross state product)

SOURCE: U.S. Census Bureau and the World Bank.

20

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

also reveals that a given change in gross state product is associated with increases in exporters
at an increasing rate.

TRADE MARGINS DURING THE GREAT TRADE COLLAPSE AND
REBOUND
As I did for the decomposition of the percentage change in exports at the national level in
the previous section, here I undertake the same calculation at the level of individual states.
First, I examine changes during the 2008-09 GTC and, second, I examine changes during the
2009-10 trade rebound.21
During the GTC, known exports declined 13.6 percent nationally; all states but Alaska
and Utah experienced a decline in known exports.22 Exports from Alaska increased 3.2 percent and exports from Utah increased 6.5 percent. Meanwhile, exports from New Mexico
declined the most: –53.7 percent.
As shown in Table 3, decreasing exports resulted from declines on both the extensive and
intensive margins for most states. For example, the number of exporting firms declined in all
but four states—Delaware, Louisiana, Maryland, and Rhode Island. The number of exporting
firms declined even in the two states in which exports increased. Not surprisingly, large states,
such as California and New York, suffered the largest absolute declines in exporters, with each
state losing more than 2,000 exporting firms. Small states, such as Montana, South Dakota,
and North Dakota, suffered the largest percentage declines in exporters. Each of these states
experienced a decline in exporting firms of more than 14 percent.
While the number of exporting firms generally declined across states during the GTC,
exports per firm across states also tended to decline. Exports per firm declined in all but nine
states—Alaska, Arkansas, Connecticut, Kentucky, Nebraska, Nevada, North Dakota, Utah, and
Vermont. Alaska, with a 14.9 percent increase in exports per firm, had the largest percentage
increase, while New Mexico, with a 51.6 percent decline, had the largest decrease.
In comparing the percentage changes in the extensive and intensive margins, declines on
the intensive margin exceed those on the extensive margin in 33 of the 50 states. Thus, not
surprisingly in light of the national numbers, relatively more of the adjustment occurs in terms
of exports per firm than in the numbers of exporters.23
Table 4 restates the results in Table 3 to highlight the ranking of states based on the relative importance of the percentage change in the intensive margin as a share of the percentage
change in exports. During the GTC, seven states show percentage changes in the intensive
margin relative to exports that are 1 or larger. The values for Alaska and Utah reflect the importance of the positive changes in the intensive margin in accounting for increased exports. The
values for Delaware, Maryland, Rhode Island, Louisiana, and Florida reflect the importance
of the negative changes in the intensive margin in accounting for the decrease in exports.
Exports rose in nearly all states during the 2009-10 rebound. Export growth was most
pronounced in Maine and New Hampshire, where growth exceeded 36 percent. The exceptions
to positive export growth were Arkansas, Nevada, and Washington. Exports from Arkansas
declined almost 11 percent.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

21

Coughlin

Table 3
Decomposition of Trade Changes
GTC (2008-09)
State
United States
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
Florida
Georgia
Hawaii
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming

Extensive
margin

–4.5
–5.0
–10.2
–3.7
–8.0
–3.0
–5.6
–6.1
10.2
0.0
–4.0
–4.3
–10.8
–5.2
–5.5
–8.6
–2.3
–5.3
0.3
–8.2
2.4
–4.4
–6.7
–5.2
–4.3
–6.6
–17.4
–7.1
–5.7
–7.3
–4.4
–4.3
–5.6
–2.7
–14.6
–5.5
–5.4
–7.0
–5.5
1.6
–6.5
–16.0
–4.5
–1.4
–2.2
–9.3
–7.2
–4.9
–3.8
–3.9
–8.3

Intensive
margin

–9.6
–14.4
14.9
–22.3
5.8
–8.8
–15.3
5.5
–14.0
–8.4
–4.9
–10.7
–3.4
–13.8
–2.9
–13.4
–27.6
3.1
–18.6
–16.2
–16.1
–7.8
–17.2
–9.8
–10.3
–16.0
–1.6
1.5
5.4
–5.3
–13.4
–51.6
–14.4
–5.9
1.7
–17.9
–2.4
–11.2
–7.7
–23.0
–4.7
–25.7
–1.3
–9.2
8.9
2.5
–7.3
–14.2
–10.2
–9.9
–1.8

Joint
change

0.4
0.7
–1.5
0.8
–0.5
0.3
0.9
–0.3
–1.4
0.0
0.2
0.5
0.4
0.7
0.2
1.2
0.6
–0.2
–0.1
1.3
–0.4
0.3
1.2
0.5
0.4
1.0
0.3
–0.1
–0.3
0.4
0.6
2.2
0.8
0.2
–0.3
1.0
0.1
0.8
0.4
–0.4
0.3
4.1
0.1
0.1
–0.2
–0.2
0.5
0.7
0.4
0.4
0.2

Rebound (2009-10)
Change in Extensive
exports (%) margin

–13.6
–18.6
3.2
–25.2
–2.6
–11.5
–20.0
–0.9
–5.3
–8.4
–8.7
–14.5
–13.8
–18.3
–8.2
–20.8
–29.2
–2.3
–18.3
–23.0
–14.1
–11.9
–22.8
–14.5
–14.2
–21.5
–18.7
–5.7
–0.7
–12.2
–17.2
–53.7
–19.2
–8.4
–13.1
–22.4
–7.7
–17.4
–12.8
–21.8
–10.9
–37.6
–5.7
–10.5
6.5
–7.1
–13.9
–18.4
–13.5
–13.4
–10.0

6.3
1.3
12.4
10.8
12.1
6.1
9.3
6.2
12.7
5.5
5.3
9.3
13.7
5.2
5.5
10.6
6.9
6.9
5.2
5.9
14.1
6.4
5.4
7.2
7.5
6.4
16.4
7.0
8.0
10.7
1.8
6.6
5.9
8.0
17.2
5.4
2.9
7.2
5.8
0.2
5.4
7.1
5.0
6.5
7.8
4.3
3.2
6.7
8.1
4.1
11.7

Intensive
margin

8.3
23.8
5.8
–3.5
–20.4
7.1
0.5
1.7
–6.0
6.8
9.5
1.3
10.8
9.4
12.9
3.2
–1.6
–1.4
13.2
30.9
–7.6
–0.3
24.1
7.7
22.2
22.2
7.8
7.4
–10.0
23.6
9.0
5.1
7.9
1.1
–6.5
14.6
12.2
3.4
10.9
26.9
10.7
9.8
14.9
14.5
12.1
24.0
4.7
–7.8
16.9
7.8
–7.0

Joint
change

0.5
0.3
0.7
–0.4
–2.5
0.4
0.0
0.1
–0.8
0.4
0.5
0.1
1.5
0.5
0.7
0.3
–0.1
–0.1
0.7
1.8
–1.1
0.0
1.3
0.6
1.7
1.4
1.3
0.5
–0.8
2.5
0.2
0.3
0.5
0.1
–1.1
0.8
0.4
0.2
0.6
0.0
0.6
0.7
0.7
0.9
0.9
1.0
0.1
–0.5
1.4
0.3
–0.8

Change in
exports (%)

15.1
25.5
18.9
6.9
–10.8
13.7
9.9
7.9
6.0
12.6
15.3
10.7
26.0
15.2
19.1
14.1
5.2
5.4
19.1
38.7
5.4
6.1
30.8
15.5
31.4
30.0
25.4
15.0
–2.7
36.8
10.9
12.1
14.2
9.2
9.7
20.9
15.5
10.9
17.4
27.1
16.7
17.6
20.7
21.9
20.8
29.3
8.1
–1.6
26.4
12.2
3.8

SOURCE: Author’s calculations using U.S. Census data.

22

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

Table 4
Relative Importance of Intensive Margin Changes
Intensive Margin Relative to Percent Change in Exports
State
Alaska
Delaware
Utah
Maryland
Rhode Island
Louisiana
Florida
New Mexico
Kansas
Arizona
Texas
Ohio
New Jersey
Alabama
Washington
Colorado
California
Illinois
Michigan
West Virginia
New York
Missouri
Wisconsin
Hawaii
Mississippi
United States
Maine
North Carolina
South Dakota
Minnesota
Massachusetts
Iowa
Oregon
Pennsylvania
Georgia
Virginia
New Hampshire
South Carolina
Indiana
Oklahoma
Idaho
Tennessee
Wyoming
Montana
North Dakota
Nebraska
Vermont
Kentucky
Arkansas
Connecticut
Nevada

Federal Reserve Bank of St. Louis REVIEW

GTC (2008-09)
4.70
2.66
1.37
1.14
1.06
1.01
1.00
0.96
0.94
0.89
0.88
0.80
0.78
0.77
0.77
0.76
0.76
0.76
0.76
0.75
0.75
0.74
0.74
0.73
0.73
0.71
0.70
0.70
0.68
0.68
0.66
0.64
0.64
0.60
0.57
0.52
0.43
0.43
0.35
0.32
0.25
0.22
0.18
0.09
–0.13
–0.26
–0.34
–1.35
–2.20
–5.89
–8.08

Origin
Washington
Nevada
Arkansas
Rhode Island
Alabama
New Jersey
Vermont
Maine
Oklahoma
Michigan
Missouri
Tennessee
Mississippi
Ohio
Louisiana
Indiana
Texas
South Carolina
New Hampshire
Wisconsin
West Virginia
Pennsylvania
Illinois
Georgia
Virginia
Utah
South Dakota
New York
United States
Florida
California
Minnesota
Nebraska
New Mexico
Idaho
Oregon
Alaska
Montana
Iowa
Connecticut
Hawaii
North Carolina
Colorado
Massachusetts
Kentucky
Kansas
Arizona
North Dakota
Delaware
Maryland
Wyoming

Rebound (2009-10)
4.97
3.64
1.89
0.99
0.93
0.82
0.82
0.80
0.79
0.78
0.74
0.72
0.71
0.70
0.69
0.68
0.66
0.64
0.64
0.64
0.64
0.63
0.62
0.62
0.59
0.58
0.56
0.55
0.55
0.53
0.52
0.50
0.50
0.42
0.41
0.31
0.31
0.30
0.23
0.21
0.12
0.12
0.05
–0.04
–0.26
–0.31
–0.51
–0.67
–1.00
–1.42
–1.83

First Quarter 2014

23

Coughlin

As shown in Table 3, rising exports resulted from increases on the extensive margin for
all states and increases on the intensive margin for most states. The growth in the number of
exporting firms exceeded 10 percent in 11 states—Alaska, Arizona, Arkansas, Delaware, Idaho,
Iowa, Maryland, Montana, New Hampshire, North Dakota, and Wyoming.
Results for the intensive margin reveal that 39 of the 50 states experienced increases; in 8
states the increases exceeded 22 percent. Maine led the way with an increase of more than 30
percent, which accounts for its overall large percentage increase in exports. For the 11 states
with declines in their intensive margins, the 20.4 percent decline in Arkansas was more than
double the decline in Nevada, which experienced the second-largest decline. Obviously, the
decline in Arkansas is the key factor accounting for the overall decline in its exports.
When the percentage changes in the extensive and intensive margins are compared,
increases on the intensive margin exceeded those on the extensive margin in 29 states during
the rebound. Thus (not surprisingly) in light of the national numbers and similar to the finding during the GTC, relatively more of the adjustment occurred in terms of exports per firm
than in the numbers of exporters.24
Table 4 ranks states based on the relative importance of the percentage change in the intensive margin as a share of the percentage change in exports. Washington, Nevada, and Arkansas,
the exceptions to positive growth in known exports during the rebound, lead the way. Rhode
Island and Alabama are the fourth- and fifth-ranked states. In both cases, positive changes in
the intensive margin account for more than 90 percent of the change in exports.

TRADE DURING THE COLLAPSE AND REBOUND: TIME, SIZE,
GEOGRAPHIC, AND DEMAND PERSPECTIVES
So far, I have explored the trade collapse and rebound separately without regard to possible
connections over time or over space. This section presents an elementary-level exploration of
some of the many possible relationships, beginning with a focus on time and then considering
other perspectives.

Time Perspective
Are there any obvious connections between the two periods with respect to time? For
example, do states with relatively larger trade collapses have relatively larger trade rebounds?
At most, I find an association for such a relationship that is very weak and not statistically
significant. The correlation between the percentage changes in known exports across states
for the two periods is –0.16, which is not statistically significant at the 5 percent level. For
total exports, the correlation coefficient is –0.20, which is also not statistically significant at
the 5 percent level. Similarly, I find no statistically significant association for the percentage
change in the intensive margin across states for the two periods. The correlation coefficient is
–0.16. However, with a correlation coefficient of –0.28, I do find a statistically significant
association for the percentage change in the extensive margin across states for the two periods.
In other words, states with relatively larger declines in their extensive margins experienced
relatively larger rebounds in this margin.
24

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

Size Perspective
This section explores whether there are differences during either period based on state
size. State size does not appear to be associated with the change in exports. For example, the
correlation between state GDP in 2008 and the percentage change in total exports during the
collapse is 0.04, and the correlation between state GDP in 2009 and the percentage change in
total exports during the rebound is –0.14. Neither is statistically significant at the 5 percent
level.
No statistically significant relationship is found between the intensive margin and state
size for either the collapse or the rebound. The correlation coefficient is –0.07 for the collapse
and 0.02 for the rebound, neither of which is statistically significant at the 5 percent level.
Meanwhile, state size is associated with the extensive margin. During the collapse, larger states
tended to have relatively smaller declines in their extensive margins; similarly, during the
rebound, larger states tended to have relatively smaller increases in their extensive margins.
In the former case, the correlation coefficient is 0.24, while in the latter case, the correlation
coefficient is –0.26. While neither case is statistically significant at the 5 percent level, both
are statistically significant at the 10 percent level.

Geographic Perspective
One way to assess the similarity in the changes in trade margins and exports of nearby
states from a geographic perspective is to calculate the Moran’s I spatial statistic. This statistic,
with a range of +1 to –1, indicates clustering when its value is close to +1 and dispersion when
its value is close to –1. If states with high values are located near other states with high values
and states with low values are located near other states with low values, then the associated
Moran’s I value will be close to 1. Meanwhile, if a state with a high value tends to be located
near a state with a low value (and vice versa), then the associated Moran’s I value will be close
to –1. If there is no pattern (i.e., random), then the value will be near zero.25
I use the trade margin measures in Table 3 and the percentage changes in total and known
exports during both the collapse and the rebound to calculate the associated Moran’s I statistic
for a number of cases. For the intensive margin, I find a random geographic distribution across
the 48 contiguous U.S. states.26 In other words, the Moran’s I values are close to zero: –0.10 for
2008-09 and –0.06 for 2009-10, suggesting no statistical association.
The maps in Figures 5 and 6 reflect this lack of geographic association for the intensive
margin. Figure 5 shows the quintile distribution by state on the intensive margin for the trade
collapse, while Figure 6 pertains to the rebound. The lightest color shows the states in the lowest (smallest values) quintile, while the darkest color shows the states in the highest (largest
values) quintile. A positive association would be suggested by a clustering of states with the
same color, while a negative association would be suggested by states with the lightest color
that are contiguous to states with the darkest color. No association would be suggested by a
random distribution of the colors of the states. Both maps reveal a random distribution. For
example, in both Figures 5 and 6, states with the lowest values are scattered throughout the
country; in no instance are more than two of these states contiguous.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

25

Coughlin

For the extensive margin, during both the collapse and the rebound, I find a positive
association at the 0.05 percent statistical significance level. For the collapse, the Moran’s I
value is 0.33; for the rebound, the value is 0.18. The maps in Figures 7 and 8 show this positive association as states of a similar color tend to cluster. For example, in Figure 7, states with
the lowest values—Maine, New Hampshire, and Vermont plus Idaho, Montana, Wyoming,
North Dakota, and South Dakota—tend to cluster. The clustering of Illinois, Indiana, Ohio,
Pennsylvania, Kentucky, Tennessee, and Alabama is also easily observable in the third quintile.
In Figure 8, the clustering of Michigan, Indiana, Ohio, Pennsylvania, and New York is easily
observable in the second quintile. In addition, states in the three highest quintiles tend to be
located west of the Mississippi River, while states in the two lowest quintiles tend to be located
east of the Mississippi River.
Determining whether the percentage changes in exports cluster, regardless of using total
or known exports for either of the periods, reveals values close to zero, indicating no statistically significant association. For example, for the collapse, the Moran’s I value is –0.12 for the
percentage change in total exports and –0.11 for the percentage change in known exports. For
the rebound, the corresponding values are –0.14 and –0.09. Recall that the percentage changes
in the intensive margin were relatively more important than the extensive margin in accounting for the percentage change in known exports, so finding that the pattern for the extensive
margin does not lead to a pattern for known exports is not surprising.

Demand Perspective
As a final topic, I explore the impact of foreign demand on state exports. Consistent with
the explanation that a decline in aggregate demand played the key role in the GTC, state exports
would be expected to be tied to changes in the economic performance of a state’s trading
partners. A simple measure of the change in the economic performance of a state’s trading
partners is the trade-weighted growth of GDP. In other words, weighting each trading partner’s
growth in 2009 (2010) by its share of a state’s trade in 2008 (2009) provides a measure of the
overall performance of these trading partners. A reasonable expectation is that a state’s export
growth in a specific year would be related positively to this summary measure of economic
performance. Using known exports, I find correlation coefficients of 0.11 for the trade collapse
and 0.12 for the trade rebound; however, for neither period do I find statistically significant
relationships.
I also explore the possible connection between both the intensive and extensive margins
and the trade-weighted growth of the state’s trading partners.27 I do not find statistically significant relationships between trade-weighted growth and percentage changes in the intensive
margin for either period. The correlation is –0.08 for the collapse and 0.09 for the rebound.
With respect to the extensive margin, a correlation coefficient of 0.48 suggests the larger
trade-weighted declines in foreign growth were associated with larger percentage declines in
the extensive margin during the collapse, but the correlation only slightly exceeds zero during the rebound.
26

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

Figure 5
Intensive Margin (Distribution of Percentage Change in Average Exports per Firm, 2008-09)

Change (%)
–51.6 to –16.1
–16.1 to –11.2
–11.2 to –7.3
–7.3 to –1.6
–1.6 to 8.9

Figure 6
Intensive Margin (Distribution of Percentage Change in Average Exports per Firm, 2009-10)

Change (%)
–20.4 to –1.4
–1.4 to 7.1
7.1 to 10.7
10.7 to 14.9
14.9 to 30.9

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

27

Coughlin

Figure 7
Extensive Margin (Distribution of Percentage Change in Number of Exporting Firms, 2008-09)

Change (%)
–17.4 to –7.3
–7.3 to –5.6
–5.6 to –4.5
–4.5 to –3.0
–3.0 to 10.2

Figure 8
Extensive Margin (Distribution of Percentage Change in Number of Exporting Firms, 2009-10)

Change (%)
0.2 to 5.2
5.2 to 6.1
6.1 to 7.1
7.1 to 10.6
10.6 to 17.2

28

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

SUMMARY AND CONCLUSION
The GTC and its aftermath provide an excellent opportunity to examine the resiliency of
the U.S. export sector. In contrast to previous studies, I examine this historic episode from the
perspective of individual states. One concern during the GTC was that such a large collapse
would transform firms from exporters into strictly domestic firms or, worse, drive them out of
business. In either case, it was feared that U.S. exporting might, at best, revive slowly. Obviously,
this fear about long-lived export impacts did not materialize. Thus, the U.S. experience during
the GTC was similar to prior trade declines. Clearly, large export effects occurred, but the sharp
decline during the GTC was quickly followed by a sharp rebound. In both periods, the percentage
change in exports far exceeded the contemporaneous percentage change in GDP.
Examining the intensive and extensive trade margins shows that the adjustment occurred
to a greater extent on the intensive than on the extensive trade margin. In other words, the adjustment process entailed relatively larger changes in average exports per firm than in the number
of exporting firms. As indicated by the analysis of individual states, this finding at the national
level is consistent with the results for the majority of states. It is likely much easier to alter existing
export levels than to, first, eliminate all exports by a firm and, second, either restart exports by
this firm or become a new entrant into exporting.
The findings in this article, which rely on state data, are consistent with previous work based
on national data for numerous countries, including the United States, France, Japan, and Belgium.
Moreover, the findings are consistent with current international trade theory that emphasizes
productivity differences across firms and the importance of the additional costs that must be
incurred to engage in international trade.28 Obviously, the more productive the firm, the more
likely it will find exporting to be profitable. The increased costs associated with exporting include
transportation costs, import duties, legal fees, marketing fees, and the development of distribution networks. Some of the increased costs can be viewed as market entry costs. Large and sunk
market entry costs deter firms from exiting a foreign market, especially if a decline in demand is
viewed as temporary. Rather than exiting, firms scale back their operations and wait for better
times. When better times arrive, the firms ramp up their exports.
With respect to additional results, our examination of the time-related adjustment process
revealed little association between the magnitudes of the trade collapses and the trade rebounds.
In other words, I found states with the relatively larger collapses did not experience relatively
larger (or smaller) rebounds. Similarly, I found no significant association between percentage
changes in the intensive margins between the two periods. With respect to the extensive margin,
I did find that relatively larger collapses were associated with relatively larger rebounds.
I found little association between state size and either the trade collapse or rebound experienced by states. I also found state size was not associated with the extent of the changes in the
intensive margin but was associated with the extensive margin. Relatively larger states tended to
have relatively smaller declines in their extensive margin during the trade collapse and smaller
increases in their extensive margin during the trade rebound.
In comparing the similarity of a state’s experience with those of its neighbors, I find no
clustering in terms of percentage changes in exports during either the trade collapse or rebound.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

29

Coughlin

Similar to my previous results, I find no clustering of states with respect to the intensive margin,
but I do find that the extensive margin experiences of nearby states are similar.
Finally, the growth of a state’s trading partners does not systematically affect export growth
during either the collapse or the rebound. This result might be caused by my reliance on annual
data. It is possible that export changes lag the growth changes. Moreover, in addition to income,
there may be other changes that obscure the impact of foreign growth. Similarly, an identified
systematic relationship is lacking at the intensive margin during both periods. With respect
to the extensive margin, I found that larger declines in foreign growth were associated with
larger declines in the extensive margin during the collapse, but larger increases in foreign
growth were not associated with larger increases in the extensive margin during the rebound.
The bottom line is that the U.S. export sector weathered the challenges associated with
the global recession and financial crisis quite well. The fact that relatively large firms dominate
U.S. exports likely contributes to the resiliency of the U.S. export sector. ■

30

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin

NOTES
1

The export data are Census-based rather than balance of payments-based. Annual data are highlighted because
data for the analyses of trade margins at the state level are available only annually. Quarterly data show that U.S.
exports of goods peaked in the third quarter of 2008. Exports declined during each of the next three quarters and
then began to recover during the second half of 2009. By the fourth quarter of 2010, exports were slightly higher
than the previous peak.

2

As noted in Baldwin (2009), the GTC is the steepest decrease in recorded history. Between World War II and the
GTC, global trade declined three times: in 1974-75, 1982-83, and 2001-02. Bussière et al. (2013) state that on a
year-over-year basis global trade declined more than 10 percent in real terms in 2009, while global real GDP fell
0.6 percent.

3

For a more technical examination of these hypotheses, see Bems, Johnson, and Yi (2013). Their conclusions match
those of Crowley and Luo.

4

See Levchenko, Lewis, and Tesar (2010), Eaton et al. (2011), and Bussière et al. (2013) for analysis of compositional
effects (e.g., the decline in demand was likely skewed toward tradable goods) and vertical specialization effects
(e.g., firms spread their production processes across countries, so that the value of trade, entailing mostly intermediate goods, will necessarily exceed value added or GDP). In the latter case, declining demand is multiplied
because of the increasing role of international supply chains. See Allesandria, Kaboski, and Midrigan (2010) for an
analysis of inventory adjustments.

5

See Contessi and de Nicola (2013) for a review of the theoretical and empirical relationships between finance and
international trade.

6

See Crowley and Luo (2011) and Bown and Crowley (2013). Eaton et al. (2011) also conclude that increased trade
barriers had only a minimal effect on global declines in the ratio of trade to GDP.

7

Other studies using the Origin of Movement series include those by Smith (1999), Coughlin and Wall (2003),
Coughlin (2004), Cassey (2009, 2011), and Coughlin and Novy (2013).

8

“Free alongside ship” means that the goods are delivered to a port to the point of loading aboard a vessel for shipment. Thus, the cost of the goods does not include the costs of moving the goods from alongside the vessel to
the foreign buyer.

9

The most recent annual report is titled A Profile of U.S. Importing and Exporting Companies, 2010-2011
(http://www.census.gov/foreign-trade/Press-Release/edb/2011/edbrel.pdf). Two years are reported each year;
the older year contains revised data and the more recent year contains data subject to revision.

10 In other words, if the individual exporter can be identified, then the exports are said to be known. Total exports

are determined by adding known exports and the exports for which no specific exporter can be identified.
11 Because the share of known to total exports is virtually constant over 2008-10, the percentage changes in total

and known exports are very similar: –18.3 versus –18.0 for 2008-09 and 21.3 versus 21.1 for 2009-10.
12 Our definitions of trade margins follow those of Lawless (2010) and Eaton, Kortum, and Kramarz (2004). As dis-

cussed by Lawless (2010), alternative definitions of these terms have appeared in the literature. Moreover, in the
context of multiproduct firms, within-firm margins related to the number of products exported and average
exports across products could be explored.
13 The decomposition of the percentage change in exports is straightforward: %∆X

y1,y0 = [(∆Fy1,y0 * Xy0 /Fy0)/Xy0] *
100 + [(((Xy1/Fy1) – (Xy0 /Fy0)) * Fy0)/Xy0] * 100 + [(∆Fy1,y0 * ((Xy1/Fy1) – (Xy0 /Fy0))/Xy0 ] * 100, where X is the priceadjusted level of known exports, F is the number of identified exporters, and the subscripts identify the years for
the calculation. The first of the three right-hand-side terms is the extensive margin, the second is the intensive
margin, and the third is the joint effect.

14 The joint effect of the changes in the two margins is 0.4 percent.
15 Such a finding is consistent with what is termed “hysteresis in trade”; see Baldwin (1988).
16 Beginning with the destination for the most U.S. exports, the top 25 U.S exports destinations in 2010 were

Canada, Mexico, China, Japan, the United Kingdom, Germany, Korea, Brazil, the Netherlands, Singapore, France,
Hong Kong, Taiwan, Belgium, Australia, Switzerland, India, Italy, Malaysia, Colombia, the United Arab Emirates,
Saudi Arabia, Israel, Chile, and Venezuela.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

31

Coughlin
17 See Cassey (2011) for a more thorough summary of state export patterns.
18 The simple correlation between state size in natural logarithms and the number of destination countries is 0.86.

Using state size in levels, the simple correlation is 0.57. Both correlations are statistically significant at the 5 percent
level.
19 The simple correlation between GDP in natural logarithms and the number of states that export to a given country

is 0.70, which is statistically significant at the 5 percent level. Meanwhile, using GDP in levels, the simple correlation
is 0.26, which is also statistically significant at the 5 percent level.
20 The simple correlation between state size as measured by gross state product in natural logarithms and the num-

ber of exporting firms of plants in the state is 0.76. Using state size in levels, the simple correlation is 0.94. Both
correlations are statistically significant at the 5 percent level.
21 Depending on the situation, I use state-level total and known exports in my analysis. Simple correlations of total

and known exports across states are virtually 1 for 2008, 2009, and 2010. The simple correlations of percentage
changes in total and known exports across states for 2008-09 and 2009-10 are 0.89 and 0.97, respectively. This
suggests that total and known exports are interchangeable for purposes of this article.
22 Utah is the only state that experienced an increase in both total and known exports. Total exports declined in

Alaska during the collapse. All other states experienced declines in both total and known exports.
23 Simple correlations between the percentage changes in trade margins and exports reinforce this conclusion. The

correlation between the intensive margin and exports is 0.93, which is statistically significant at the 5 percent level,
while the correlation between the extensive margin and exports is 0.13, which is not statistically significant at the
5 percent level.
24 Simple correlations between the percentage changes in trade margins and exports reinforce this conclusion. The

correlation between the intensive margin and exports is 0.94, which is statistically significant at the 5 percent level,
while the correlation between the extensive margin and exports is –0.20, which is not statistically significant at
the 5 percent level.
25 A simple illustration is a checkerboard with white and black squares. Normally, the white and black squares are

dispersed yielding a Moran’s I value close to –1. However, if all the white squares were placed together on one
side of the board and the black squares on the other side, then the associated Moran’s I value would be close to 1.
A random arrangement would yield a value close to zero.
26 Alaska and Hawaii are omitted because they have no contiguous states.
27

Increases in a foreign country’s income, holding other things constant, should tend to increase a state’s total
exports to the country. However, as demonstrated by Lawless (2010), the effect in theory on the intensive margin
is ambiguous. By inducing the entry of new exporting firms, average exports per firm may increase, decrease, or
remain unchanged. Because of the induced entry of new exporters, the extensive margin in a state should increase
as a foreign country’s income increases.

28 See Melitz (2003).

REFERENCES
Allesandria, George; Kaboski, Joseph P. and Midrigan, Virgiliu. “The Great Trade Collapse of 2008-09: An Inventory
Adjustment?” IMF Economic Review, 2010, 58(2), pp. 254-94.
Baldwin, Richard. “Hysteresis in Import Prices: The Beachhead Effect.” American Economic Review, September 1988,
78(4), pp. 773-85.
Baldwin, Richard, ed. The Great Trade Collapse: Causes, Consequences, and Prospects. VoxEU.Org eBooks, 2009;
http://www.voxeu.org/epubs/cepr-reports/great-trade-collapse-causes-consequences-and-prospects.
Behrens, Kristian; Corcos, Gregory and Mion, Giordano. “Trade Crisis? What Trade Crisis?” Review of Economics and
Statistics, May 2013, 95(2), pp. 702-09.
Bems, Rudolfs; Johnson, Robert C. and Yi, Kei-Mu. “The Great Trade Collapse.” Annual Review of Economics, 2013, 5,
pp. 375-400.
32

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Coughlin
Bown, Chad P. and Crowley, Meredith A. “Import Protection, Business Cycles, and Exchange Rates: Evidence from
the Great Recession.” Journal of International Economics, May 2013, 90(1), pp. 50-64.
Bussière, Matthieu; Callegari, Giovanni; Ghironi, Fabio; Sestieri, Giulia and Yamano, Norihiko. “Estimating Trade
Elasticities: Demand Composition and the Trade Collapse of 2008-2009.” American Economic Journal:
Macroeconomics, July 2013, 5(3), pp. 118-51.
Cassey, Andrew J. “State Export Data: Origin of Movement vs. Origin of Production.” Journal of Economic and Social
Measurement, 2009, 34(4), pp. 241-68.
Cassey, Andrew J. “State Foreign Export Patterns.” Southern Economic Journal, October 2011, 78(2), pp. 308-29.
Chor, Davin and Manova, Kalina. “Off the Cliff and Back? Credit Conditions and International Trade During the
Global Financial Crisis.” Journal of International Economics, May 2012, 87(1), pp. 117-33.
Contessi, Silvio and de Nicola, Francesca. “What Do We Know About the Relationship Between Access to Finance
and International Trade?” Working Paper No. 2012-054B, Federal Reserve Bank of St. Louis, March 2013;
http://research.stlouisfed.org/wp/2012/2012-054.pdf.
Coughlin, Cletus C. “The Increasing Importance of Proximity for Exports from U.S. States.” Federal Reserve Bank of
St. Louis Review, November/December 2004, 86(6), pp. 1-18;
http://research.stlouisfed.org/publications/review/04/11/Coughlin.pdf.
Coughlin, Cletus C. and Novy, Dennis. “Is the International Border Effect Larger than the Domestic Effect? Evidence
from U.S. Trade.” CESifo Economic Studies, June 2013, 59(2), pp. 249-76; doi:10.1093/cesifo/ifs002.
Coughlin, Cletus C. and Wall, Howard J. “NAFTA and the Changing Pattern of State Exports.” Papers in Regional
Science, October 2003, 82(4), pp. 427-50.
Crowley, Meredith A. and Luo, Xi. “Understanding the Great Trade Collapse of 2008-09 and the Subsequent Trade
Recovery.” Federal Reserve Bank of Chicago Economic Perspectives, Second Quarter 2011, 35(2), pp. 44-70;
http://www.chicagofed.org/digital_assets/publications/economic_perspectives/2011/2qtr2011_part1_crowley_luo.pdf.
Eaton, Jonathan; Kortum, Samuel and Kramarz, Francis. “Dissecting Trade: Firms, Industries, and Export
Destinations.” American Economic Review, May 2004, 94(2), pp. 150-54.
Eaton, Jonathan; Kortum, Samuel; Neiman, Brent and Romalis, John. “Trade and the Global Recession.” NBER
Working Paper No. 16666, National Bureau of Economic Research, January 2011;
http://www.nber.org/papers/w16666.pdf?new_window=1.
Fontagné, Lionel and Gaulier, Guillaume. “French Exporters and the Global Crisis,” in Richard Baldwin, ed., The Great
Trade Collapse: Causes, Consequences and Prospects. Chap. 16. VoxEU.Org eBooks, 2009;
http://www.voxeu.org/article/french-exporters-and-global-crisis.
Lawless, Martina. “Deconstructing Gravity: Trade Costs and Extensive and Intensive Margins.” Canadian Journal of
Economics, November 2010, 43(4), pp. 1149-72.
Levchenko, Andrei A.; Lewis, Logan T. and Tesar, Linda L. “The Collapse of International Trade during the 2008-2009
Crisis: In Search of the Smoking Gun.” IMF Economic Review, 2010, 58(2), pp. 214-53.
Melitz, Marc J. “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity.”
Econometrica, November 2003, 71(6), pp. 1695-725.
Schott, Peter K. “US Trade Margins during the 2008 Crisis,” in Richard Baldwin, ed., The Great Trade Collapse: Causes,
Consequences and Prospects. Chap. 15. VoxEU.Org eBooks, 2009;
http://www.voxeu.org/article/us-trade-margins-during-2008-crisis.
Smith, Pamela J. “Are Weak Patent Rights a Barrier to U.S. Exports?” Journal of International Economics, June 1999,
48(1), pp. 151-77.
Wakasugi, Ryuhei. “Why Was Japan’s Trade Hit So Much Harder?” in Richard Baldwin, ed., The Great Trade Collapse:
Causes, Consequences and Prospects. Chap. 23. VoxEU.Org eBooks, 2009;
http://www.voxeu.org/article/why-was-japan-s-trade-hit-so-much-harder.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

33

34

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

A Guide to Tracking the U.S. Economy
Kevin L. Kliesen

Analyzing and forecasting the performance and direction of a large, complex economy like that of
the United States is a difficult task. The process involves parsing a great deal of data, understanding
key economic relationships, and assessing which events or factors might cause monetary or fiscal
policymakers to change policy. One purpose of this article is to reinforce several key principles that
are useful for tracking the U.S. economy’s performance in real time. Two principles stand out: First,
the economy is regularly hit by unexpected economic disturbances (shocks) that policymakers and
forecasting models cannot predict. Second, most key data used to measure the economy and track its
performance are often revised—and by substantial amounts. (JEL E32, E66)
Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 35-54.

nalyzing and forecasting the performance of the U.S. and global economies is a daunting challenge, even for trained, professional economists. This means the challenge
facing the nonpractitioner is probably much more difficult. For example, suppose a
furniture retailer would like to know the direction of interest rates and the unemployment
rate over the next year or two. The direction of interest rates is important because sales of
durable goods such as furniture tend to be interest rate sensitive. Likewise, if an increasing
percentage of the labor force becomes unemployed, then sales will tend to suffer. But other
economic variables are also important. Household wealth, home sales, and consumer sentiment are often used by forecasters and some monetary policymakers to help predict the
future path of consumer spending on durable goods.
If the retailer guesses wrong and orders too much or too little furniture from the factory,
this may lead to either too much or too little inventory on hand. If too much furniture is
ordered, the retailer’s costs of carrying the extra inventory would increase, whereas if too little
is ordered, the retailer’s sales might suffer. In both instances, the retailer’s profits would probably be reduced relative to what was expected. In short, a furniture retailer has a powerful
incentive to form some assessment of the economy’s future performance.

A

Kevin L. Kliesen is a research officer and economist at the Federal Reserve Bank of St. Louis. Douglas C. Smith and Lowell R. Ricketts provided
research assistance.
© 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views
of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. 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.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

35

Kliesen

This article is not a how-to exercise in building economic models. Rather, it is intended
to assist the noneconomist (nonpractitioner) who wants to analyze and interpret patterns of
economic activity at the macro level. A nonpractitioner can be a businessperson, an investor,
or any individual interested in monitoring the U.S. economy and/or developing an expectation
of its short-term direction. A key conclusion is that the economy’s performance can change
rapidly. Accordingly, the nonpractitioner seeking some clues about the short-term direction
of the economy is advised to monitor a handful of key data and then balance this information
against freely available consensus forecasts of the economy over the next six months or so.
Over time, consensus forecasts, which are simple averages of a group of professional forecasters,
tend to be more accurate than any individual forecast.

A BASIC MODEL OF ECONOMIC FLUCTUATIONS
A practicing forecaster usually needs a model of how the macroeconomy works. For professional forecasters, the “model” is usually a sophisticated system of equations designed to
explain key aspects of the economy—such as growth of real gross domestic product (GDP),
inflation, interest rates, stock prices, and the unemployment rate. Nonpractitioners—those
not actively managing a large econometric forecasting model—tend to be at a distinct disadvantage in this domain. To compensate, the nonpractitioner who needs to make some judgment about the future direction of the economy should adopt a less formal economic model.
Such a model would convey a broad notion of how the economy evolves over the business
cycle. One simplistic model the nonpractitioner can use to organize his or her thoughts would
be the following: U.S. economic activity—or real GDP—revolves around a trend that grows
roughly at a rate determined by the sum of labor productivity growth and population growth.1
This trend is sometimes called the growth rate of the economy’s potential output. Deviations
around this trend—termed “economic fluctuations”—occur because of unexpected disturbances (termed “shocks”), new technologies, and the ever-evolving preferences of consumers,
firms, and government policymakers to save, spend, and regulate.
With this simplistic model, the nonpractitioner can make reasonably accurate assessments
about the likely direction of the economy over the next several months or quarters. For example, if auto and home sales are strengthening, the unemployment rate is falling, job gains are
picking up, and stock prices are rising, then these factors are usually reliable signals that the
economy is on an upswing. The nonpractitioner should thus exploit the fact that many key variables move together, which is known as comovement (see the boxed insert on the next page).
Comovement is important because the economy’s natural state is one of positive growth—
where this growth is dependent on the economy’s fundamentals. At any point in time, then,
the economy will be growing above or below this trend rate of growth, which will then affect
important variables such as, inflation, interest rates, and the unemployment rate.2
A model of inflation also differentiates between short- and long-run movements. Inflation
can vary over shorter periods of time with changes in energy prices or labor costs. However,
over longer periods (several years), actions taken by monetary policymakers will have a significant influence on the economy’s inflation rate. Importantly, this transmission stems from
36

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

Exploit the Principle of Comovement
Private sector professional forecasters and forecasters employed by central banks typically use sophisticated econometric
models based on either economic theory or statistical relationships designed to exploit comovements among key variables. The
table below depicts one way to measure comovement: by noting correlations among key economic variables. The economic
variables listed in the table are real GDP, the unemployment rate, equity prices, the conventional mortgage rate, new orders for
manufactured goods, single-family home sales, and consumer price index (CPI) inflation.

Comovement: Cross-Correlation of Four-Quarter Changes or Growth Rates

Indicator
Real GDP
Unemployment rate
Equity prices
Conventional mortgage rate
New manufacturing orders
Single-family home sales
CPI inflation

Unemployment
Real GDP
rate
1.00
–0.78
0.68
0.46
0.65
0.47
0.34

1.00
–0.54
–0.33
–0.73
–0.39
–0.50

Equity
prices

1.00
0.18
0.62
0.35
0.16

New
Single-family
Conventional
mortgage manufacturing
home
rate
orders
sales
CPI inflation

1.00
0.40
–0.23
0.31

1.00
0.12
0.58

1.00
–0.03

1.00

NOTE: Correlations are based on the sample period 1992:Q3–2013:Q1.
SOURCE: Author’s calculations.

The correlations are based on changes over four quarters; the sample period is 1992 to the present.1 As an example, the growth
of real GDP is highly correlated with the unemployment rate (–0.78), equity prices (0.68), and new manufacturing orders (0.65).
These correlations suggest, therefore, that faster real GDP growth tends to be associated with falling unemployment rates,
higher stock prices, and faster growth of factory orders. Economists then exploit this comovement over time, as well as economic
theory, when building forecasting models or thinking about how the economy evolves in response to changes in some key data.
Economists have also exploited comovement to construct economic indexes designed to measure economic activity using several separate economic series. There are several such indexes designed to mirror broader trends in the economy. Among the more
notable are those published by the Federal Reserve Bank of Chicago (Chicago Fed National Activity Index) and the Federal
Reserve Bank of Philadelphia (Aruoba-Diebold-Scotti Business Conditions Index).2
1 The data are four-quarter growth rates (percent changes), except for the unemployment rate and the conventional mortgage; these are simple
changes over four quarters. For example, the unemployment rate in 2013:Q1 was 7.7 percent and in 2012:Q1 was 8.3 percent. Thus, the fourquarter change was 0.6 percentage points.
2 For the Chicago Fed National Activity Index, see http://www.chicagofed.org/webpages/research/data/cfnai/current_data.cfm. For the Aruoba-

Diebold-Scotti Business Conditions Index, see http://www.phil.frb.org/research-and-data/real-time-center/business-conditions-index/.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

37

Kliesen

the effects of these policy actions on the market’s expectations about future inflation. William
Poole (2005, pp. 303-04), former president of the Federal Reserve Bank of St. Louis, provided
a nice summary of how this might work in practice:
My sense of what I do, which I think is not dissimilar to what most FOMC [Federal Open Market
Committee] members do, is attempt to intuit future inflation pressures from current observed
pressures as they show up in both price changes and resource pressures, or real gaps, in individual
markets. The approach is not totally without theory; for example, wage changes are evaluated in
light of expected productivity trends. I attempt to sort out temporary from more lasting wage
and price changes and attempt informally to construct an appropriately weighted average of disparate experience in various sectors. I look closely at data on inflation expectations, but treat such
data carefully because longer-run expectations are really a vote of confidence on the Fed and not
an independent reading on inflation.
I am extremely uncomfortable with this approach and believe that it is an invitation to future
mistakes. I don’t know what better to do.

The nonpractitioner faces another key disadvantage relative to professional forecasters or
economic policymakers: resource constraints. Thus, returning to our earlier example, small
firms tend to be at a disadvantage compared with large firms in trying to analyze the direction
of the economy. Large firms have the resources to hire economists to sift through the data and
construct their own sophisticated forecasting models, or they can benefit from professional
forecasting services on a contract basis. To help offset this disadvantage, a small business owner
will probably adopt some form of naive forecasting (“what happened last year will happen
again this year”) by reading economic and financial market commentaries from trade associations or perusing “reputable” economic blogs. Some may also use common rules of thumb
purported to gauge the strength and direction of the economy, such as the direction of the
stock market.
The challenge of economic forecasting extends beyond the technical expertise required
to make accurate forecasts. Other factors contributing to this difficult task include the sheer
volume of data, persistent data revisions, and correct interpretation of data that may send
conflicting signals. Other complications are the responses of monetary and fiscal policymakers
and foreign economic developments. Before expounding on how a nonpractitioner might try
to overcome these challenges, the next section offers a brief discussion of the events leading
up to 2008. The so-called Financial Panic of 2008 and the Great Recession offer several examples of the difficulties both nonpractitioners and professional forecasters face as they attempt
to learn about the direction of the economy and forecast its short-term future path.

THE PERILS OF FORECASTING: A LOOK BACK AT 2008
The late economist John Kenneth Galbraith reportedly once remarked that there are
two types of forecasters: those who don’t know and those who don’t know they don’t know.
Galbraith’s aphorism reveals an underappreciated aspect of forecasting: It is inherently difficult. Thus, it was not surprising that the onset of the recent recession was not foreseen by the
majority of the professional forecasting community. According to the Business Cycle Dating
38

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

Figure 1
A Timeline of Blue Chip Forecasts for Real GDP Growth in 2008
Percent (Annual Rate)
4.0

2008 (Real GDP)
December 2007
September 2008

3.0
2.0
1.0

–0.2
Initial Actual

0
–1.0
–2.0

–3.3
Current Estimate

–3.0
–4.0
Jan. 2007

Jul. 2007

Jan. 2008

Jul. 2008

Jan. 2009

Jul. 2009

SOURCE: Blue Chip Economic Indicators, various issues.

Committee of the National Bureau of Economic Research (NBER), the U.S. economic expansion that began in November 2001 ended sometime in December 2007.3 However, by the end
of 2007, very few professional forecasters were predicting a recession in 2008. In fact, in the
December 2007 Blue Chip Economic Indicators, the consensus of the Blue Chip forecasters
was that real GDP would increase by 2.2 percent in 2008. The average of the 10 most pessimistic forecasters was 1.6 percent, while the average of the 10 most optimistic forecasters was
2.7 percent.4
The NBER Business Cycle Dating Committee, like many nonpractitioners, tends to look
at real GDP as a key indicator (among other indicators) of the economy’s performance. For
example, increases (decreases) in expenditures for real final goods and services—such as automobiles, refrigerators, or physician services—are regularly followed by increases (decreases)
in employment and a lower (higher) unemployment rate. As Figure 1 shows, throughout
most of 2007 the Blue Chip Consensus (BCC) of professional forecasters was that real GDP
would increase by about 3 percent in 2008. This figure plots a timeline of BCC forecasts for
real GDP growth in 2008. The first forecast was published in January 2007. Beginning in
September 2007, though, forecasters began to steadily lower their projections for real GDP
growth in 2008. In particular, as discussed below, the forecasts for real GDP growth for 2008
turned sharply lower after the widespread financial turmoil in September 2008. By the end of
November 2008, when the NBER announced that the recession began sometime in December
2007, the BCC forecast for real GDP growth in 2008 had dipped slightly below zero.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

39

Kliesen

Figure 2
A Timeline of Blue Chip Forecasts for CPI Inflation in 2008
Percent (Annual Rate)
5.0
2008 (CPI Inflation)
December 2007
September 2008

4.5
4.0
3.5
3.0
2.5
2.0

1.6
Actual

1.5
1.0
0.5
0
Jan. 2007

Jul. 2007

Jan. 2008

Jul. 2008

Jan. 2009

Jul. 2009

SOURCE: Blue Chip Economic Indicators, various issues.

The direction of inflation is another key indicator of economic performance. First, longterm interest rates such as mortgage rates and corporate bond yields have an inflation premium.5 Accordingly, if inflation or the perceived risk of higher inflation in the future increases,
then interest rates also usually rise. A higher inflation rate may also spur the Fed to raise its
short-term interest rate target, which could also cause long-term rates to rise.6 The direction
of inflation was markedly different over a good portion of this period. As Figure 2 shows, from
January 2007 until March 2008, the BCC forecast was that the CPI would increase by a bit less
than 2.5 percent in 2008.
The relative stability of inflation expectations was somewhat surprising given the behavior
of oil prices and actual inflation over this period. From January 2007 to March 2008, crude
oil prices rose from about $55 per barrel to about $106 per barrel. Over the same period, the
year-to-year percent change in the CPI rose from 2.1 percent to 4 percent. As oil prices and
actual inflation continued to rise over the first half of 2008, forecasters began to dramatically
raise their forecasts for inflation in 2008—from about 2.75 percent in April to 4.5 percent in
September.7 Interestingly, though, forecasts for CPI inflation in 2009 (not shown) rose only
slightly, which suggests that most forecasters tended to believe that the upsurge in inflation in
2008 would be temporary. This forecast proved to be accurate. (See the boxed insert on p. 46.)
A key takeaway message from Figures 1 and 2 is that significant, unexpected economic
shocks can have important effects on the expectations of forecasters—and thus investors and
economic policymakers. The remainder of the article discusses a methodology the nonprac40

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

Table 1
Free Economic Forecasts
Name

Source

Frequency

Survey of Professional
Forecasters

Federal Reserve Bank
of Philadelphia

Quarterly

http://www.phil.frb.org/research-and-data/
real-time-center/survey-of-professionalforecasters/

FOMC Summary of
Economic Projections

Federal Open
Market Committee

Quarterly

http://www.federalreserve.gov/monetarypolicy/
fomccalendars.htm

IMF World Economic
Outlook Reports

International
Monetary Fund

NABE Outlook (partial)

National Association
for Business Economics

Quarterly

http://nabe.com/NABE_Outlook_Summary

Budget and Economic
Outlook

Congressional
Budget Office

Annually

http://www.cbo.gov/publication/43907

Economic Outlook

OECD

Semiannually

Semiannually

URL

http://www.imf.org/external/ns/cs.aspx?id=29

http://www.oecd.org/eco/economicoutlook.htm

NOTE: IMF, International Monetary Fund; NABE, National Association for Business Economics; OECD, Organisation for Economic Co-operation and
Development.

titioner can use to help analyze the current and short-term performance of the U.S. economy.
This approach relies on publicly available data and macroeconomic forecasts. In this framework, “reading the tea leaves” requires an assessment of the following economic conditions:
• the economy’s momentum (slowing or accelerating);
• the headwinds or tailwinds affecting this momentum and how long they are expected
to last; and
• the risks to the outlook—that is, what could produce growth that is either faster or
slower than expected for economic activity and prices.
Endnotes are used for those seeking references or a more in-depth discussion about analyzing general business conditions and the macroeconomy. Although the U.S. economy is obviously affected by events in other countries, the discussion focuses primarily on U.S. data flows
and the decisions adopted by U.S. economic policymakers and their potential economic
consequences.

KEY PRINCIPLES FOR TRACKING THE ECONOMY
Principle #1: Use Freely Available Forecasts
The nonpractitioner should adhere to a set of key economic principles. One such principle is comparative advantage. That is, the nonpractitioner should use forecasts developed by
professional economists with advanced training and experience in modeling and forecasting.
Another principle pertains to the law of demand, which relates the price of a good to its
quantity demanded: Free is usually better. Fortunately, reputable forecasts are freely available
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

41

Kliesen

Figure 3
Forecasts of Four Key Economic Variables
What Are Forecasters Predicting for Real GDP Growth?

What Are Forecasters Predicting for Inflation?

Percent

Percent

5

5
Actual

4

Actual

Forecast

Forecast

4
3
3
2
2

1

1

0
–1
2011:Q4

2012:Q3

2013:Q2

2014:Q1

2014:Q4

0
2011:Q4

2012:Q3

2013:Q2

2014:Q1

2014:Q4

What Are Forecasters Predicting for the
Unemployment Rate?

What Are Forecasters Predicting for the 10-Year
Treasury Yield?

Percent

Percent

10

5
Actual

9

Actual

Forecast

Forecast

4

8
3
7
2

6
5
2011:Q4

2012:Q3

2013:Q2

2014:Q1

2014:Q4

1
2011:Q4

2012:Q3

2013:Q2

2014:Q1

2014:Q4

SOURCE: Survey of Professional Forecasters, November 2013.

to the public (Table 1). The law of large numbers is also a related principle: An average, or
consensus, of many forecasts is usually better than a single forecast by any one forecaster.
Figure 3 shows forecasts of four key economic variables: real GDP growth, inflation, the
unemployment rate, and the 10-year Treasury yield. The forecasts are based on a survey of
professional forecasters and published four times per year by the Federal Reserve Bank of
Philadelphia in its Survey of Professional Forecasters (SPF). In the November 2013 SPF, the
consensus of professional forecasters was that the economy would continue to improve. This
was evident by a modest acceleration in real GDP growth, a modest reduction in the unemployment rate, and a modest upswing in long-term interest rates. Forecasters also expected
inflation to remain relatively low and stable.

42

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

Principle #2: Use Forecast Revisions to Gauge Changes in Economic Expectations
Consensus forecasts are valuable because they provide a “best guess” approach to the
economic outlook. This approach differs sharply from relying on a single forecaster whose
model or theoretical biases may not be readily known. However, a drawback to consensus
forecasts—and, for all practical purposes, all forecasts—is that forecast horizons less than a
year or two (four to eight quarters) ahead can change dramatically because of unexpected
economic events. Still, the nonpractitioner can use this knowledge to help assess whether the
economy is experiencing faster or slower momentum. Just as a car speeds up or slows down,
the economy goes through periods when growth of real GDP, inflation, or employment is faster
or slower than expected. Before discussing how the nonpractitioner can assess changes in the
economy’s momentum, it is crucial to acknowledge some key facts about the U.S. economy.
First, the economy’s normal state of affairs is one of positive growth in real GDP and in
prices (inflation). According to the NBER Business Cycle Dating Committee, from January
1948 to December 2013, the U.S. economy has spent 670 of 792 months (or 85 percent of the
time) in expansion. Second, the growth rate of key indicators, such as employment, retail sales,
real GDP, and inflation, can vary tremendously from month to month, quarter to quarter, or
year to year. Third, actions by the Federal Open Market Committee (FOMC) can influence
the economy in important respects, but generally not immediately. Fourth, unexpected disturbances regularly occur that cause forecasts to go awry.
Figures 1 and 2 show how changes in forecasters’ expectations are reflected in the economy’s momentum. If the economy is exhibiting stable momentum, this generally suggests
that the incoming data flows are in line with expectations. In this case, forecasts for real GDP
growth (and other key indicators) will remain relatively unchanged, as they were over the first
part of 2008. However, faster momentum suggests the incoming data are exceeding expectations (in a good way), and this will be translated into upward revisions in forecasts for real GDP
growth. The opposite holds for slower momentum. An example of the latter situation is the
downgrading of forecasts for real GDP growth that began in 2008 (see Figure 1). In terms of
inflation momentum, Figure 2 shows forecasters continually raised their estimates for CPI
inflation for 2008 over the first eight months of the year in response to rising energy prices.
Changes in momentum, as reflected in data flows, are important because they help forecasters
identify possible shocks to the economy, which can be either positive or negative. These
changes thus feed back into revised forecasts.
One drawback to this approach is that freely available forecasts tend to be published at a
quarterly or annual frequency (see Table 1). However, identifying momentum changes from
the monthly forecasts used in Figures 1 and 2 requires a paid subscription to the Blue Chip
Economic Indicators. As an aside, many professional forecasters tend to update their modelbased forecasts on a daily or weekly basis using the latest available data. But a lot can happen
in three months, so nonpractitioners who use quarterly forecasts need to augment this framework with something else to identify momentum shifts. One relatively easy method is to systematically track key economic data flows to infer future forecast revisions.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

43

Kliesen

Principle #3: Follow the Data to Help Identify Momentum Swings
Momentum swings during the course of the business cycle can be measured by tracking
the evolution of the data relative to the expectations of forecasters and/or financial market
participants. These data flows or other economic news may alter the expectations of investors,
professional forecasters, and policymakers regarding the strength or weakness of the economy.
A recent example of this principle was cited by former Federal Reserve Chairman Ben Bernanke
(2013, p. 4) in his press conference following the June 19, 2013, FOMC meeting:
Although the Committee left the pace of purchases unchanged at today’s meeting, it has stated
that it may vary the pace of purchases as economic conditions evolve. Any such change would
reflect the incoming data and their implications for the outlook.

In the context of measuring economic momentum, if key data repeatedly surprise on the
upside (downside), then this is a signal that forecasters have been underestimating (overestimating) the strength of the economy. To successfully use this framework, the nonpractitioner
must first decide which economic data to focus on.8 This step is crucial for two reasons. First,
some data are more important than others. And second, some data directly influence forecasts
for real GDP and inflation, but most do not. In this section, the discussion focuses on key
nonfinancial variables.9 The importance of financial market conditions is discussed later.
Table 2 provides a list of key data that the noneconomist should monitor on a regular
basis.10 In particular, key series released early in the monthly data cycle include
• the manufacturing and nonmanufacturing purchasing managers indexes (PMIs), which
provide a broad-based overview of economic activity;
• the nonfarm payroll employment and unemployment rate series published by the
Bureau of Labor Statistics in “The Employment Situation”; and
• reports on manufacturing activity (durable goods orders and industrial production),
consumer spending (retail sales and auto sales), and housing activity (housing starts
and new and existing home sales).
A weekly series—initial claims for unemployment insurance benefits—is also included. Initial
claims is an important indicator because (i) it is released each week and (ii) the series tends to
have some predictive power for the number of individuals moving into and out of jobs. For
example, Kliesen, McCracken, and Zheng (2011) show that job growth tends to weaken or
strengthen when the number of initial claims rises above or falls below 400,000.11
Table 2 also includes a market-based forecast for each of the indicators released on a
recurring basis. For each series, economists and market analysts are surveyed and asked to
provide their forecast, or best guess estimate, for the key economic data to be released that
week. These market-based expectations for key upcoming data releases are found on many
freely available economic calendars.12
Table 2 shows how a practitioner can use these expectations to construct a systematic,
simple approach to gauge potential changes in economic momentum in real time based on
data surprises. This method is depicted in the last three columns of the table. First, for each
44

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

Table 2
A Time Horizon of Key Data Flows (April 2013–May 2013)
Date

Indicator

Market
Period expectations

3/31/2013
4/1/2013
4/1/2013
4/2/2013
4/2/2013
4/3/2013
4/4/2013
4/5/2013
4/5/2013
4/5/2013
4/5/2013
4/9/2013
4/10/2013
4/11/2013
4/12/2013
4/12/2013
4/12/2013
4/12/2013
4/12/2013
4/16/2013
4/16/2013
4/16/2013
4/16/2013
4/16/2013
4/16/2013
4/18/2013
4/22/2013
4/23/2013
4/24/2013
4/24/2013
4/26/2013
4/29/2013
4/29/2013
4/29/2013
4/30/2013
5/1/2013
5/2/2013
5/2/2013
5/2/2013
5/7/2013

ISM Manufacturing PMI
Construction spending
Factory orders
Total vehicle sales
ISM Non-Manufacturing PMI
Initial claims
Total nonfarm payrolls
Private payrolls
Unemployment rate
International trade balance
Wholesale inventories
Federal budget balance
Import prices
PPI
Core PPI
Retail sales
Retail sales excluding autos
Business inventories
Housing starts
Building permits
CPI
Core CPI
Industrial production
CU rate
Index of Leading Economic Indicators
Existing home sales (total)
New home sales
Durable goods
Durable goods excluding transportation
Real GDP
Personal income
PCE (expenditures)
Core PCE (prices)
Employment Cost Index
Construction spending
International trade
Productivity
Unit labor costs
Consumer credit ($)

March
February
February
March
March
March 29
March
March
March
February
February
March
March
March
March
March
March
February
March
March
March
March
March
March
March
March
March
March
March
Q1 Advance
March
March
March
Q1 Advance
March
March
Q1 Advance
Q1 Advance
March

5/9/2013

Initial claims

May 4

Actual

54.2
1.0
2.9
15.3
55.8
347
200
209
7.7
–44.6
0.5
–156.0
–0.5
–0.1
0.2
0.0
0.1
0.4
0.930
0.940
0.0
0.2
0.2
78.4
0.1
5.010
0.420
–2.8
0.5
3.0
0.4
0.0
0.1
0.5
0.7
–42.2
1.5
0.6
15.0

51.3
1.2
3.0
15.2
54.4
385
88
95
7.6
–43.0
–0.3
–106.5
–0.5
–0.6
0.2
–0.4
-0.4
0.1
1.036
0.902
–0.2
0.1
0.4
78.5
–0.1
4.920
0.417
–5.7
–1.4
2.5
0.2
0.2
0.0
0.3
–1.7
–38.8
0.7
0.5
8.0

335

323

Better than
expected?

Sign

NTI

No
Yes
Yes
No
No
No
No
No
Yes
Yes
No
No
Same
Yes
Same
No
No
No
Yes
No
Yes
Yes
Yes
Yes
No
No
No
No
No
No
No
Yes
Yes
Yes
No
Yes
No
Yes
No

0
–1
1
1
–1
–1
–1
–1
–1
1
1
–1
–1
0
1
0
–1
–1
–1
1
–1
1
1
1
1
–1
–1
–1
–1
–1
–1
–1
1
1
1
–1
1
–1
1
–1

0
–1
0
1
0
–1
–2
–2
–3
–2
–1
–2
–3
–3
–2
–2
–3
–4
–5
–4
–5
–4
–3
–2
–1
–2
–3
–4
–5
–6
–7
–8
–7
–6
–5
–6
–5
–6
–5
–6

Yes

1

–5

NOTE: CU, capacity utilization; ISM, Institute for Supply Management; PCE, personal consumption expenditures; PPI, producer price index.
SOURCE: Thomson Reuters and author’s calculations.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

45

Kliesen

Assessing Risks to the Economic Outlook
When economic conditions are perceived as “normal,” the risk to the outlook is viewed as balanced. This means that forecasters
generally see little reason to alter their assessment of the short-term outlook. However, if there are developments in the domestic
or global economy that are judged as “abnormal,” then the risks are considered skewed to the upside or downside.
The 2007-09 recession came on the heels of two shocks that have historically proved damaging to the economy’s short-term performance: an unexpectedly large increase in crude oil prices and an epic decline in house prices and housing activity.1 Oil prices
are important because, historically, unexpected increases in oil prices have preceded nearly every post-World War II U.S. recession.
Increases in oil prices tend to (i) raise headline inflation rates and (ii) slow the growth of economic activity. Thus, the analyst or
investor who concludes that rising oil prices increase the probability of weaker growth and higher inflation is usually on safe
terrain.2
In a similar vein, housing is important because construction ripples through the economy—for example, affecting sales of consumer durables such as furniture and appliances—and house prices can change household wealth and thus perhaps consumer
expenditures (the largest component of GDP).3 As this episode demonstrates, nonpractitioners would be wise to pay attention to
developments in energy and housing markets to gauge unfolding risks to the outlook.4
1 Stock and Watson (2012) argue that the sharp rise in oil prices and the financial turmoil were key factors that caused the 2007-09 recession.
2 Hamilton (2008) has documented that 9 of the 10 recessions between 1948 and 2001 were preceded by a rise in oil prices. Rising oil prices also
appear to be an important factor explaining business cycles in other advanced economies. See Engemann, Kliesen, and Owyang (2011).
3 Boldrin et al. (2013) discuss and estimate these housing spillover effects; also see Leamer (2008). Some economists believe that changes in equity

prices also have a significant effect on household consumption.
4 At the same time, many forecasts, including those by the staff members who advise the FOMC, consider numerous alternative scenarios to the

so-called baseline forecast. Some forecasts, for example, attach a probability to a recession scenario, but the probability is generally much smaller
than the baseline “no recession” scenario.

release, determine whether the data were better than expected. Second, if so, arbitrarily assign
an indicator value of +1; if not, assign a –1 (worse than expected). If the data met expectations,
then assign a value of 0. Third, sum the indicator values (+1, –1, and 0) to obtain a net tracking index (NTI). Using the first indicator in Table 2 (the Institute for Supply Management
[ISM] Manufacturing PMI), the market’s expectation for March 2013 was 54.2 but the actual
estimate was 51.3, which was worse than expected, so we assign a value of –1. By the end of
the list on May 9, the cumulative series—which is the NTI—has a value of –5. A negative value
thus indicates that, on net, the data have come in worse than expected and, by assumption,
this implies some weaker economic momentum over this period of data flows. Figure 4 plots
the NTI for data flows that measure economic activity in the fourth quarter of 2012 and the
first quarter of 2013.
Interpreting the NTI is relatively straightforward since it is conditional on one’s assumption about the direction of the expected change in economic activity. And since much of the
data feed directly into estimates of real GDP or are indicators of economic activity more
broadly, the NTI is thus one proxy for the expected change in real GDP in a given quarter—
actual or forecasted. Though not shown here, the nonpractitioner could also construct an
NTI for inflation pressures.
It should be noted that the NTI date listed in Figures 4A and 4B is not the same as the
period of economic measurement. For example, total nonfarm payrolls for March 2013 were
released on April 5, 2013. Figure 4A shows that beginning in the second week of December
46

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

Figure 4
Net Economic Tracking Index
A. 2012:Q4

B. 2013:Q1

+1 = Better Than Expected –1 = Worse Than Expected

+1 = Better Than Expected –1 = Worse Than Expected

30

30

25

25

20

20

15

15

10

10

5

5

0

0

–5
10/8/12 11/7/12 12/7/12 1/6/13

2/5/13

3/7/13

–5
1/9/13

2/8/13

3/10/13

4/9/13

5/9/13

SOURCE: Author’s calculations.

2012, the data flows began to be better than expected, on net. All else equal, this was a signal
that in the fourth quarter the economy was strengthening by more than expected. However,
when the advance estimate was released at the end of January 2013, real GDP for the fourth
quarter of 2012 instead was shown to have declined at a 0.1 percent annual rate. This was far
below the consensus forecast. Consistent with the NTI, though, subsequent revisions by the
BEA slightly raised the advance estimate for real GDP growth in the fourth quarter from –0.1
percent to 0.1 percent.
By contrast, Figure 4B shows that the NTI performed modestly better in the first quarter
of 2013. Beginning in late February and early March 2013, the data began to come in consistently better than expected. In response, forecasters began raising their first-quarter estimate
for real GDP growth. When the advance estimate was released in late April 2013, the economy
was shown to have grown at a 2.5 percent annual rate.
Although the NTI is a very simple metric for measuring momentum, there are a few drawbacks to consider. First, the NTI does not discriminate between the value added of expenditures (e.g., retail sales), employment, survey-based measures, or prices. A second criticism is
that the NTI assigns each series the same weight (equal importance). Thus, key data such as
payroll employment or housing starts should probably be assigned larger weights than series
such as wholesale inventories. One problem confronting all practitioners and nonpractitioners
is the inevitability of data revisions. Although the NTI, by design, cannot account for subsequent revisions, it does help minimize this problem because it includes survey data and other
types of data that are not revised (e.g., consumer confidence or weekly initial claims). But if
the goal is to use the latest data to get a reading on real GDP growth, then revisions are an
issue that the nonpractitioner will need to confront.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

47

Kliesen

Most Key Data Are Based on Sample Estimates
At any point in time, those interested in reading the economy’s tea leaves have a morass of data to follow. The layperson needs to
recognize, however, the limitations of many of these data series. For example, most key government statistics reported by the
Census Bureau, the Bureau of Economic Analysis (BEA), and other agencies are based on sample-based estimates rather than
universe-based counts. For example, to measure the monthly number of jobs created in the United States each month, the Bureau
of Labor Statistics (BLS) does not count every new job at every single business and government entity each month. Instead, the
BLS surveys approximately 145,000 private nonagricultural businesses and government agencies each month.1 These businesses
and agencies represent approximately 557,000 individual worksites. This sample-based estimate is then used to construct an estimate of total nonfarm payroll employment for the nation.2
Despite what many people may believe, a separate survey is used to estimate the nation’s unemployment rate. Each month, the
Census Bureau surveys about 60,000 households about their labor force status. These data are then analyzed and published by
the BLS in its monthly employment report. In addition to providing an estimate of the unemployment rate, the BLS publishes an
estimate of the nation’s population, labor force, and the number of people employed, among other metrics. Many other key data
series, such as factory orders, housing starts, retail sales, and industrial production, are also based on survey data collected by
government and private sector entities. An additional consideration is that most of the key data produced by the government
statistical agencies and private sector firms are seasonally adjusted. This is another potential source of measurement error.
1 These entities employ approximately 9 million nonfarm workers. Technically, the establishment survey counts the number of jobs rather than
workers because some people have more than one job.
2 More detailed information on the methodology used by the BLS to construct labor market measures from the Current Employment Statistics or
the Current Population Survey can be found in the BLS Handbook of Methods; see http://www.bls.gov/opub/hom/.

Principle #4: Beware of Data Revisions
As noted in the description of Principle 3, revisions to data compound the difficulty of
correctly identifying shocks and their significance in real time. These revisions occur largely
because much of the source data collected by the U.S. government statistical agencies are based
on surveys of a sample of economic entities (firms, households, and government offices and
agencies), rather than a survey of the universe of all economic entities. (This process is discussed in the boxed insert above.) As an example of this process, consider the quarterly estimate for the growth of real GDP, which is subject to numerous revisions. These revisions
generally reflect updates in the underlying source data or new data based on more complete
surveys or income tax records. Sometimes, revisions are made to prices or the underlying
statistical methodology used by the government agencies to construct the estimate.
To see how revisions can dramatically change the portrait of the economy’s performance,
consider the estimate of real GDP growth for the fourth quarter of 2007. According to the
NBER, this quarter was the peak of the 2001-07 business expansion. As shown in Figure 5, the
BEA released nine estimates of the annual rate of change for real GDP growth in the fourth
quarter of 2007. In the advance estimate released in late January 2008, the BEA reported that
real GDP rose at a 0.6 percent annual rate. However, when the annual National Income and
Product Accounts (NIPA) revision was released in late July 2008, the estimate for real GDP
growth in the fourth quarter of 2007 was changed to –0.2 percent. This estimate was subsequently changed to 2.9 percent per year in the 2010 annual NIPA revision but was subsequently
marked back down by nearly 1.5 percentage points with the release of the July 2013 NIPA
revision.
48

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

Figure 5
Real-Time Estimates of Real GDP Growth During 2007:Q4
Percent Change (Annual Rate)
3.5
2.9

3.0
2.5
2.1
2.0

1.7

1.5

1.7
1.5

1.0
0.6

0.6

0.6

0.5
0
–0.2

–0.5

Jan. 2008 Feb. 2008 Mar. 2008 Jul. 2008 Jul. 2009 Jul. 2010 Jul. 2011 Jul. 2012 Jul. 2013

NOTE: Dates reflect the month in which the estimates were published.
SOURCE: Bureau of Economic Analysis and Haver Analytics.

What should the nonpractitioner take away from this discussion? First, the data may not
correctly portray the economy’s momentum. This possibility suggests that the nonpractitioner
should take the monthly data flows and the consensus forecasts with a grain of salt. But what
is the alternative? After all, policymakers, FOMC members, businesspersons, and investors
have little choice but to react to the incoming data flows.13
One way a nonpractitioner can minimize the potential havoc caused by data revisions is
to avoid point estimates. Thus, instead of becoming enamored with a forecast for real GDP
growth of 3.5 percent (the point estimate), an interest rate of 3 percent, or an unemployment
rate of 6.5 percent, the nonpractitioner would attempt to assess whether the economic momentum revealed by forecast revisions and the NTI suggest something more likely on either side
of the point estimate. Another method of minimizing the impact of data revisions is to track
financial market conditions. Although there is the possibility of a chicken-versus-egg problem
since financial markets also react to incoming data flows that are subsequently revised, some
financial market series have long been recognized for their leading indicator properties.

Principle #5: Track Financial Market Conditions14
Economic historians have long known that disturbances in the financial sector can have
significant effects on the economy.15 Moreover, stabilizing the real economy through its interventions in financial markets is one of the key reasons central banks exist.16 The Financial
Panic of 2008 provides another example of the financial sector’s far-reaching effects on the
macroeconomy when asset prices and other key financial market indicators are changing significantly.17 Thus, for a more-complete portrait of the economy and potential changes in shortFederal Reserve Bank of St. Louis REVIEW

First Quarter 2014

49

Kliesen

Figure 6
The St. Louis Fed Financial Stress Index
Weekly Data
6

Week of August 5, 2011:
S&P Downgrade and FOMC

5

European Turmoil,
May/June 2010

4
3

September/October 2008

2
1
0
–1
–2
12/31/93

12/31/97

12/31/01

12/31/05

12/31/09

12/31/13

NOTE: The last observation is for the week ending January 31, 2014.

term economic momentum, it is important for the nonpractitioner to understand and track
financial market conditions.
One of the key lessons policymakers learned from the 2007-08 experience is an old one:
It is extraordinarily difficult to predict financial crises with any degree of confidence. But it
can also be difficult to monitor financial conditions because there are literally thousands of
different types of financial indicators—ranging from stock price indexes, to interest rates on
government and corporate debt, to foreign exchange rates, to more elaborate indicators such
as credit default swaps and mortgage-backed securities.18 Fortunately, the nonpractitioner
can overcome much of this difficulty by focusing on a handful of key indicators. Three come
to mind.
The first key financial indicator is a financial stress index (FSI). FSIs are designed to
measure changes in financial conditions. For example, when financial market conditions are
viewed as stable, then financial stresses tend to be relatively normal. In this situation, lenders
are no more risk averse than normal and the volatility of asset prices, such as stock and bond
prices, exhibits no unusual movements. By and large, financial market participants have a
rather sanguine view of the economy. By contrast, if lenders are becoming more risk averse,
asset prices are falling, and volatility is increasing, then financial stresses are on the rise. In
this instance, uncertainty about the health of the economy is increasing.
Rising levels of financial stress tend to weaken the real economy through a variety of
transmission mechanisms. These include reduced wealth, a reduction in bank lending, and
balance-sheet effects that reduce the value of a firm’s collateral. The key innovation of FSIs is
that they combine different types of financial market indicators into one index—much as the
CPI is one measure of the economy’s price level constructed from tens of thousands of different prices on goods and services. For example, economic research has convincingly shown
there is significant information content, and thus predictive power, in the U.S. Treasury yield
50

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen

curve, commonly calculated as the difference between yields on 10-year Treasury securities and
3-month Treasury bills. The yield curve tends to be upward sloping during times of positive
growth in economic activity and tends to narrow as the pace of economic activity slows and—
importantly—to invert before recessions.19
Another measure of financial market stress is the spread between 10-year Treasury securities and Baa-rated corporate bonds. This interest rate spread, called the credit risk spread,
accounts for default risks in private credit markets. Thus, if interest rates are rising and the pace
of real GDP growth is slowing, firms tend to experience slowing sales, which adversely affects
their revenues and thus their financial condition. In response, the risk associated with lending
to firms increases.
One well-known financial stress index that accounts for the information content in these
two interest rate spreads is the St. Louis Fed Financial Stress Index (STLFSI). Figure 6 plots the
STLFSI and shows that financial market stresses rose sharply prior to several recent economic
upheavals that were transmitted to financial markets. These included recent developments in
Europe, the downgrade of U.S. sovereign debt by Standard and Poor’s, and the unexpected downward revision in U.S. real GDP growth in July 2011 discussed previously. Empirical evidence
suggests that rising levels of financial stress are associated with weak or negative growth of industrial production and other measures of economic activity going forward.20

CONCLUSION
Analyzing and forecasting the performance and direction of a large, complex economy like
that of the United States is exceptionally difficult. The process involves parsing a great deal of
data, understanding key economic relationships, and assessing which events or factors might
cause monetary or fiscal policymakers to change policy. One purpose of this article is to reinforce
several key principles that a nonpractitioner should use to analyze U.S. economic and financial
market conditions. The nonpractitioner can do a reasonably good job of tracking changes in the
economy’s momentum by taking advantage of freely available macroeconomic forecasts and
tracking key data. ■

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

51

Kliesen

NOTES
1

Labor productivity is output per hour. Broadly speaking, output is the total value of real GDP. Real GDP measures
the inflation-adjusted dollar value of goods and services produced by labor and property located in the United
States; these are known as factors of production. A “real” series has been adjusted to remove the effects of changes
in prices over time. This adjustment better accounts for increases in the production and consumption of goods
and services (i.e., volumes) in response to underlying factors that most affect supply and demand—such as
changes in population, productivity, and technological innovations. These real factors are the ones that lead to
increases (or decreases) in living standards, what economists term “economic growth.” See Gutierrez et al. (2007)
for a primer on GDP and its construction.

2

In periods of economic recession, the level of real GDP is declining from one quarter to the next (at a negative
growth rate).

3

A rough rule of thumb is that the NBER assumes that the inflection point occurred at the middle of the month.

4

The Blue Chip Consensus is a survey of roughly 50 private sector forecasters. For example, each forecaster submits
his or her forecast for real GDP growth and other key macroeconomic and financial variables for the current and
upcoming year. The consensus is the simple average (mean) of these forecasts. These forecasts are published in
the Blue Chip Economic Indicators on or about the 10th of each month.

5

Economists use the Fisher equation to help analyze changes in interest rates. According to this equation, the
nominal interest rate on, for example, a 30-year U.S. Treasury bond is the sum of (i) the real rate of interest earned
over this period and (ii) the average inflation rate expected over this period (the inflation premium, a premium
demanded by lenders to compensate them for the expected inflation rate over the maturity of the bond). This
simple formulation ignores other risk premiums embedded within these nominal interest rates and the complication of selling a security before it matures, which can significantly affect the holding period rate of return on the
bond.

6

Many sophisticated forecasting models, including those used by the staff economists at the Board of Governors
of the Federal Reserve, assume that the yield on, say, the 10-year Treasury security, is simply an average of a series
of 1-year expected future interest rates. Thus, by changing the level of the overnight federal funds rate, the model
assumes that the Fed can affect the long-term interest rate.

7

Crude oil prices (measured by West Texas Intermediate) rose from a little more than $105 per barrel in March 2008
to about $134 per barrel in July 2008 (monthly averages). Over the same period, the year-to-year percent change
in the CPI increased from 4 percent to 5.6 percent.

8

The St. Louis Fed’s FRED (Federal Reserve Economic Data) database contains one of the world’s largest collections
of freely available economic and financial data. FRED can be accessed at http://research.stlouisfed.org/fred2/.

9

From a GDP-accounting perspective, data that flow directly into real GDP—such as housing starts, retail sales,
and factory shipments—help economists estimate whether the growth of real GDP is likely to change from its
previous-quarter estimate. However, other data flows such as employment, initial unemployment claims, inflation,
and consumer confidence might also be signals of changes in aggregate demand or supply and thus elicit reactions
from policymakers and financial market participants.

10 All these data, and more, are available on FRED, which currently contains more than 60,000 economic series.
11 Gavin and Kliesen (2002) provide a description and overview of the initial claims data. They also show that the ini-

tial claims indicator has some statistically significant ability to predict monthly changes in payroll employment.
12 Many of these forecasts can be found on the “Calendar of Releases”

(http://research.stlouisfed.org/publications/ usfd/cover.pdf ) published each week in the St. Louis Fed’s U.S.
Financial Data publication (http://research.stlouisfed.org/publications/usfd/). Yahoo! provides an economic calendar with market forecasts of these and other variables (http://biz.yahoo.com/c/ec/201315.html).
13 Orphanides and Van Nordren (2002) have shown that the timeliness of the data and subsequent data revisions

make it extremely difficult for monetary policymakers to identify the strength of real GDP relative to potential real
GDP (the output gap) in real time.
14 This section draws from Kliesen, Owyang, and Vermann (2012).
15 See Kindleberger and Aliber (2005) or Reinhart and Rogoff (2009).

52

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Kliesen
16 This principle is known as the lender of last resort. Walter Bagehot’s Lombard Street (1873) is a classic text on the

role of central banks during financial crises.
17 Several other events around this period contributed to the rise of financial market instability. For example, see

“The Financial Crisis: A Timeline of Events and Policy Actions” on the St. Louis Fed’s website
(http://timeline.stlouisfed.org/).
18 Zheng (2012) provides a good summary of key financial indicators (such as the yield curve) for the noneconomist.
19 See Wheelock and Wohar (2009) for a summary of this research.
20 See Kliesen, Owyang, and Vermann (2012).

REFERENCES
Bagehot, Walter. Lombard Street: A Description of the Money Market. London: H.S. King, 1873.
Bernanke, Ben. “Transcript of Chairman Bernanke’s Press Conference.” June 19, 2013;
http://www.federalreserve.gov/mediacenter/files/FOMCpresconf20130619.pdf.
Boldrin, Michele; Garriga, Carlos; Peralta-Alva, Adrian and Sánchez, Juan M. “Reconstructing the Great Recession.”
Federal Reserve Bank of St. Louis Working Paper 2013-006B, February 2013, revised June 2013;
http://research.stlouisfed.org/wp/2013/2013-006.pdf.
Engemann, Kristie M.; Kliesen, Kevin L. and Owyang, Michael T. “Do Oil Shocks Drive Business Cycles? Some U.S.
and International Evidence.” Macroeconomic Dynamics, November 2011 (Suppl. S3), 15(3), pp. 498-517.
Gavin, William T. and Kliesen, Kevin L. “Unemployment Insurance Claims and Economic Activity.” Federal Reserve
Bank of St. Louis Review, May/June 2002, 84(3), pp. 15-28;
http://research.stlouisfed.org/publications/review/02/05/15-28GavinKliesen.pdf.
Gutierrez, Carlos M.; Glassman, Cynthia A.; Landefeld, J. Steven and Marcuss, Rosemary D. “Measuring the Economy:
A Primer on GDP and the National Income and Product Accounts.” Bureau of Economic Analysis, September 2007.
Hamilton, James D. “Oil and the Macroeconomy,” in Steven N. Durlauf and Lawrence E. Blume (eds.), The New
Palgrave Dictionary of Economics. Second Edition. New York: Palgrave Macmillan, 2008.
Kindleberger, Charles P. and Aliber, Robert. Manias, Panics, and Crashes: A History of Financial Crises. Fifth Edition.
Hoboken, NJ: John Wiley & Sons, 2005.
Kliesen, Kevin L.; McCracken, Michael W. and Zheng, Linpeng. “Initial Claims and Employment Growth: Are We at
the Threshold?” Federal Reserve Bank of St. Louis Economic Synopses, 2011, No. 41, December 14, 2011;
http://research.stlouisfed.org/publications/es/11/ES1141.pdf.
Kliesen, Kevin L.; Owyang, Michael T. and Vermann, E. Katarina. “Disentangling Diverse Measures: A Survey of
Financial Stress Indexes.” Federal Reserve Bank of St. Louis Review, September/October 2012, 94(5), pp. 369-97;
http://research.stlouisfed.org/publications/review/12/09/369-398Kliesen.pdf.
Leamer, Edward E. “Housing Is the Business Cycle,” in Housing, Housing Finance, and Monetary Policy. Proceedings of
the 2008 Jackson Hole Economic Policy Symposium, Jackson Hole, Wyoming, August 30-September 1, 2007.
Kansas City, MO: Federal Reserve Bank of Kansas City, 2008, pp. 149-233;
http://www.kansascityfed.org/publicat/sympos/2007/PDF/Leamer_0415.pdf.
Orphanides, Athanasios and Van Norden, Simon. “The Unreliability of Output-Gap Estimates in Real Time.” Review
of Economics and Statistics, November 2002, 84(4), pp. 569-83.
Poole, William. “Safeguarding Good Policy Practice.” Federal Reserve Bank of St. Louis Review, March/April 2005,
87(2, Part 2), pp. 303-06; http://research.stlouisfed.org/publications/review/05/03/part2/PanelDiscussion2.pdf.
Reinhart, Carmen N. and Rogoff, Kenneth S. This Time Is Different: Eight Centuries of Financial Folly. Princeton, NJ:
Princeton University Press, 2009.
Stock, James H. and Watson, Mark W. “Disentangling the Channels of the 2007-2009 Recession.” Brookings Papers
on Economic Activity, Spring 2012, pp. 81-135.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

53

Kliesen
Wheelock, David C. and Wohar, Mark E. “Can the Term Spread Predict Output Growth and Recessions? A Survey of
the Literature.” Federal Reserve Bank of St. Louis Review, September/October 2009, 91(5, Part 1), pp. 419-40;
http://research.stlouisfed.org/publications/review/09/09/part1/Wheelock.pdf.
Zheng, Linpeng. “What Do Financial Market Indicators Tell Us?” Federal Reserve Bank of St. Louis Liber8 Economic
Information Newsletter. January 2012;
http://research.stlouisfed.org/pageone-economics/uploads/newsletter/2012/Lib0112.pdf.

54

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

QE: Is There a Portfolio Balance Effect?
Daniel L. Thornton

The Federal Open Market Committee has recently attempted to stimulate economic growth using
unconventional methods. Prominent among these is quantitative easing (QE)—the purchase of a
large quantity of longer-term debt on the assumption that it will reduce long-term yields through the
portfolio balance channel. Former Federal Reserve Chairman Ben Bernanke and others suggest that
QE works through the portfolio balance channel, which implies a strong, statistically significant positive relationship between the public’s holding of long-term Treasury debt and long-term Treasury
yields. The author uses the econometric approach of Gagnon et al. (2011) and others to investigate
the relationship between a variety of measures of the public’s debt holding and various yield measures in the literature. The empirical results provide virtually no support for the portfolio balance
channel. (JEL E52, E58, E43, E44)
Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 55-72.

All that quantitative easing (QE) does is to restructure the maturity of U.S. government debt in
private hands. Now, of all the stories you’ve heard why unemployment is stubbornly high, how
plausible is this: “The main problem is the maturity structure of debt. If only Treasury had issued
$600 billion more bills and not all these 5 year notes, unemployment wouldn’t be so high. It’s a
good thing the Fed can undo this mistake.”
Of course that’s preposterous.
—John H. Cochrane, December 7, 2010

he Federal Reserve aggressively increased the size of its balance sheet in the wake of
Lehman Brothers’ bankruptcy announcement on September 15, 2008. Coincident with
the massive increase in the supply of reserves, the federal funds rate fell to nearly zero.1
With the funds rate at effectively zero, the Federal Open Market Committee (FOMC) turned
to unconventional monetary policies. Prominent among these is the policy of large-scale asset
purchases (LSAPs), referred to as quantitative easing (QE). The goal of QE is to stimulate
investment and consumption by reducing longer-term yields (see, e.g., Woodford, 2001,
2012). Considerable research has been devoted to determining the impact of the Fed’s QE

T

Daniel L. Thornton is vice president and presidential adviser at the Federal Reserve Bank of St. Louis. The author thanks Clemens Kool for helpful
comments and Bryan Noeth, Sean Grover, and Li Li for valuable research assistance.
© 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views
of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. 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.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

55

Thornton

operations on long-term yields. This effort can be divided into two broad strands of the literature. The first is event studies that analyze high-frequency changes in bond yields following
QE announcements (e.g., Gagnon et al., 2011; Krishnamurthy and Vissing-Jorgensen, 2011;
Joyce et al., 2010; Neely, 2013; Swanson, 2011; Bauer and Rudebusch, 2011; Wright, 2012).2
The second strand of the literature uses lower-frequency (monthly) data to test the implication of the portfolio balance effect—namely, that there is a positive relationship between bond
term premiums (and, consequently, bond yields) and the maturity structure of the public’s
holding of Treasury debt and long-term Treasury yields (e.g., Gagnon et al., 2011; Hamilton
and Wu, 2012; Krishnamurthy and Vissing-Jorgensen, 2012; Greenwood and Vayanos, forthcoming). This strand of the literature investigates the relationship between bond yields or term
premiums and various measures of the public’s holdings of Treasury debt prior to the FOMC’s
QE activities. Evidence of a statistically significant and economically important positive relationship prior to the FOMC’s actions is used to infer how the FOMC’s actions to reduce these
supply measures should have affected term premiums or bond yields. This article contributes
to this literature by investigating the relationship between long-term yields and the public’s
holding of long-term debt using a wide array of public debt and bond yield measures found
in the literature. Following Gagnon et al. (2011) and Krishnamurthy and Vissing-Jorgensen
(2012), I investigate the relationship within the context of a simple reduced-form framework
that controls for a variety of macroeconomic and other variables. In implementing this methodology, I account for the trend in term premiums and bond yields over the sample period. To
preview the empirical results, when the trend is accounted for, there is little evidence of a
statistically significant effect of the Fed’s LSAPs on yields and no evidence of an economically
meaningful effect.
The article proceeds as follows. The next section briefly discusses the portfolio balance
channel. I then review the previous empirical work in the literature and examine a variety of
public debt, maturity/duration, and interest rate measures used in the literature. The next
section presents the empirical results using these measures.

THE PORTFOLIO BALANCE CHANNEL
Many researchers and policymakers hypothesize that the Fed’s LSAPs affect long-term
yields through a variety of channels (see, e.g., Krishnamurthy and Jorgensen, 2011). However,
many analysts (e.g., Bernanke, 2010, and Gagnon et al., 2011) have suggested that QE works
through the portfolio balance channel.3 For example, Gagnon et al. (2011, p. 7) say that QE
reduces long-term yields because the Fed’s LSAPs remove “a considerable amount of assets
with high duration from the markets. With less duration risk to hold in the aggregate, the
market should require a lower premium to hold that risk.”
For the portfolio balance channel to be operative, the market for long-term debt must be
effectively segmented from the rest of the financial market. Until recently, the idea that markets are segmented had gained relatively little traction among financial economists and policymakers. Skepticism that financial markets are segmented stems in part from the fact that yield
differentials create arbitrage opportunities that the market will exploit.
56

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Thornton

Consequently, it is not surprising that several analysts have expressed skepticism about
the empirical significance of the portfolio balance channel. For example, Cochrane (2011, p. 16)
suggests that the portfolio balance channel should be relatively weak because (i) “much of the
Treasury bond supply is locked away in central bank and pension fund vaults” and “arbitrageurs
take duration risk in mortgage-backed, corporate, and other markets” and (ii) the Fed’s QE
operations are “just a drop in the bucket.” Bauer and Rudebusch (2011, p. 6) make a similar
argument, noting that “the scale of the Fed’s purchases of $1.725 trillion of debt securities is
small relative to the size of [domestic] bond portfolios” and “the global bond market—arguably,
the relevant one—is several times larger.” They conclude that the portfolio balance channel
should be weak and suggest that the FOMC’s LSAPs affect long-term yield through the signaling channel. Kocherlakota (2010) suggests that QE merely shifts interest rate risk from bondholders to taxpayers, and as a result “QE ends up having no effects, except for those associated
with any new forward guidance that it signals.”

PREVIOUS LOW-FREQUENCY RESEARCH
Hancock and Passmore (2011) and Stroebel and Taylor (2009) use lower-frequency monthly
data to investigate the effect of the FOMC’s purchases of mortgage-backed securities and find
mixed results. Hancock and Passmore (2011) find a relatively large impact, while Stroebel and
Taylor (2009) find a relatively small or statistically insignificant effect.
D’Amico and King (2010) and D’Amico et al. (2012) investigate the effects of QE on the
Treasury yield curve using micro-transactions data. D’Amico and King (2010) estimate both
flow and stock effects; the former are the response of prices to ongoing purchases and the latter
are changes due to expectations about future withdrawals of supply. They find small and temporary flow effects. The stock effect based on a counterfactual yield curve from their model
suggests that the nearly $300 billion purchase of Treasury securities would flatten the yield
curve in the range of 10 to 15 years by 45 basis points. However, when the observations on
key QE announcements days are omitted, only one of the own response or cross-response
coefficients is statistically significant. This fact would seem to suggest that their results are
critically dependent on an announcement effect.
D’Amico et al. (2012) suggest that QE can affect long-term yields and term premiums
through three channels. The first is called the scarcity channel, which they define as “a mechanism under which the purchase by the Federal Reserve of assets with a specific maturity leads
to higher prices (and lower yields) of securities with similar maturities” (p. 2). The second is
called the duration channel, defined as “a mechanism under which the removal…of aggregate
duration from the outstanding stock of Treasury debt reduces term premiums on securities
across maturities” (p. 2). The duration channel seems to be very similar to the portfolio balance
channel. The third is the signaling channel. D’Amico et al. identify scarcity by creating maturity “buckets” consisting of the public’s holdings of Treasury securities of given maturities relative to total Treasury debt outstanding.4 They find that both the scarcity and duration channels
are statistically significant; however, the duration channel accounts for only a third or a fourth
of their estimate of the total effect. They find no evidence of an important signaling channel.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

57

Thornton

Greenwood and Vayanos (forthcoming) focus directly on the portfolio balance channel
by organizing their empirical analysis based on Vayanos and Vila’s (2009) segmented-markets
model. Specifically, they estimate regressions of bond yields and excess returns on the ratio of
maturity-weighted debt to gross domestic product (GDP). They find a positive relationship
between their debt measure and both yields and returns, but the estimates are statistically significant only for returns on bonds with maturities of five years or longer.
Just as Greenwood and Vayanos (forthcoming) do, Gagnon et al. (2011) investigate the
effect of QE on long-term yields using a measure of the public’s holding of Treasury debt.
Specifically, Gagnon et al. (2011) estimate the equation as follows:
(1)

it = α + Xt β + δ pdt + εt .

Gagnon et al. (2011) use two measures of i, the 10-year Treasury yield and an estimate of the
10-year Treasury term premium. X is a [1 × K] vector of macroeconomic variables and pd is
a measure of the public’s holding of Treasury debt, where a, b, and d are constant coefficients,
and e is a random error with a zero mean and a constant variance. They estimate the equation
over the period January 1985 through June 2008. Their estimate of the supply effect suggests
that the FOMC’s $1.75 trillion asset purchase (QE1) should have reduced the term premium
by about 52 basis points and the 10-year Treasury yield by about 82 basis points.
Hamilton and Wu (2012) investigate the effect of QE by estimating a three-factor affine
term structure model, using assumptions motivated by Vayanos and Vila’s (2009) model.
Specifically, they calculate arbitrageurs’ portfolio weights under the assumptions that arbitrageurs comprise the entire private sector and U.S. Treasury debt is held only by arbitrageurs.
They use the estimates from their model to infer how changes in the maturity structure of
Treasury debt can affect yields. Their estimates of the effect of QE1 on the 10-year Treasury
yield and term spreads are smaller than those reported by Gagnon et al. (2011), Greenwood
and Vayanos (forthcoming), and D’Amico and King (2010). The effect of QE2 was perverse:
The Fed’s asset purchase program increased Treasury yields and term premiums. Hamilton
and Wu (2012, p. 38) attribute this to the fact that the “fraction of publicly held debt of more
than 10 years maturity continued to increase even as the Fed was implementing its QE2 bond
purchases.”

THE RELATIONSHIP BETWEEN LONG-TERM YIELDS AND THE
SUPPLY OF PUBLIC DEBT
This section investigates the relationship between long-term yields and the supply of public
debt using the methodology of Gagnon et al. (2011). The analysis differs from previous work
in the literature in that I consider several debt supply measures used in the literature and alternative yield measures. The analysis begins with a discussion of the public debt measures used
in the literature.
58

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Thornton

Figure 1
Public’s Holding of Treasury Debt, Net of SOMA
$ Billions
4,500
4,000
3,500
3,000
2,500
2,000
1,500
1,000
500

Gagnon et al. (2011)
Hamilton and Wu (2012)

0
Jan-90 Jul-91 Jan-93 Jul-94 Jan-96 Jul-97 Jan-99 Jul-00 Jan-02 Jul-03 Jan-05 Jul-06 Jan-08

Alternative Debt Supply Measures
This section analyzes alternative debt measures used in the literature to investigate the
portfolio balance channel. The debt measures are those used by Gagnon et al. (2011), Hamilton
and Wu (2012), and Greenwood and Vayanos (forthcoming).5 Gagnon et al. (2011) and
Hamilton and Wu (2012) use data on the public’s holding of Treasury debt, less that held by
the Fed in the System Open Market Account (SOMA). Figure 1 shows these series for the
period January 1990 through June 2008. The series are nearly identical until the late 1990s
when they begin to diverge. The difference is likely due to the inclusion of Treasury inflationprotected securities (TIPS) in Gagnon et al.’s data. Hamilton and Wu’s data do not include TIPS.6
Gagnon et al. (2011) consider only the public’s holdings of government debt with maturities of one year or longer, net of SOMA. This series is shown in Figure 2. However, they make
several adjustments to this series. First, they subtract foreign official holdings of Treasury
securities with maturities of one year or longer because foreign governments are unlikely to
have a term premium similar to that of the private sector. The resulting series (S2) is also shown
in Figure 2.
Rather than using the S2 series, Gagnon et al. (2011) also subtract foreign official holdings
of agency and private sector debt with maturities of at least one year. This adjustment is inappropriate because agency and private securities are not included in S2. The resulting series (S3),
also shown in Figure 2, is negative beginning in November 2007, when foreign official holdings
of agency and private sector debt become larger than the public’s holding of Treasury debt.7 As
a final adjustment, Gagnon et al. (2011) express S3 as a percent of nominal GDP (S3 gdp).
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

59

Thornton

Figure 2
Alternative Measure of the Public’s Holding of Government Debt
$ Billions
3,000
2,500
2,000
1,500
1,000
500
S1
S2
S3

0
–500

Jan-90 Jul-91 Jan-93 Jul-94 Jan-96 Jul-97 Jan-99 Jul-00 Jan-02 Jul-03 Jan-05 Jul-06 Jan-08

Figure 3
Gagnon et al. and Greenwood and Vayanos Supply Measures
Percent of GDP

Maturity-Weighted Debt-to-GDP Ratio

25

4.5
4.0

20
3.5
15

3.0
2.5

10
2.0
5

1.5
1.0

0
Gagnon et al. (left axis)

0.5

Greenwood and Vayanos (right axis)

60

First Quarter 2014

7

06

l-0
Ju

4

nJa

03

l-0
Ju

1

00

l-0

nJa

Ju

8

nJa

97

l-9
Ju

5

94

l-9

nJa

Ju

2

nJa

91

l-9
Ju

9
l-8

nJa

Ju

nJa

l-8

6

88

0
Ju

Ja

n-

85

–5

Federal Reserve Bank of St. Louis REVIEW

Thornton

Figure 4
Average Maturity and Percent of Debt Greater Than 10 Years to Maturity
Percent

Months to Maturity
80

25

70
20
60
50

15

40
10

30
20

5
P10
AVE
0

10
0

Jan-90 Jul-91 Jan-93 Jul-94 Jan-96 Jul-97 Jan-99 Jul-00 Jan-02 Jul-03 Jan-05 Jul-06 Jan-08

Greenwood and Vayanos (forthcoming) use data from the Center for Research in Securities
Prices (CRSP) for every government bond issued between 1940 and 2007 without netting out
Fed or agency holdings. Specifically, they construct the maturity structure of the debt by aggregating cash flows across individual bonds—that is, the sum of all principal and coupon payments due over the maturity of each bond issued. They then construct a maturity-weighted
debt-to-GDP ratio supply measure. Figure 3 shows this measure with Gagnon et al.’s (2011)
S3 gdp measure.8 While the two measures differ greatly in scale, they follow a very similar pattern: Both rise until the mid-1990s and then decline. Gagnon et al.’s measure declines more
dramatically because they subtract foreign official holdings of agency and private debt, while
Greenwood and Vayanos’s measure is based solely on the public’s holdings of Treasury debt.
Hamilton and Wu (2012) do not use the debt measure shown in Figure 1 but rather the
average maturity of public debt (AVE) and the percent of public debt with maturities longer
than 10 years (P10) (Figure 4). The series behave similarly over the sample period; the correlation is 84 percent.

The Data: Term Premiums and Treasury Yields
This section considers alternative measures of interest rates used in the literature.
Gagnon et al. (2011) evaluate the effectiveness of QE by estimating the effect of the Fed’s purchase of securities on an estimate of the 10-year Treasury term premium (TP) and the zerocoupon 10-year Treasury bond yield (T10).9 Their estimate of TP is obtained from the term
structure model of Kim and Wright (2005). Figure 5 shows strong negative trends and similar
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

61

Thornton

Figure 5
10-Year Treasury Yield and Term Premium
Percent
14

T10
TP

12
10
8
6
4
2

7

06

l-0
Ju

4

nJa

03

l-0
Ju

1

n-

00

l-0

Ja

Ju

n-

8

97

l-9

Ja

Ju

5

n-

94

l-9

Ja

Ju

2

nJa

91

l-9
Ju

9

nJa

88

l-8
Ju

6

n-

l-8

Ja

nJa

Ju

85

0

Figure 6
Public’s Holding of Treasury Debt, Net of SOMA
Percent
5
SYC
TP
4

3

2

1

0

62

First Quarter 2014

7
l-0

06
nJa

Ju

4

03

l-0
Ju

n-

l-0

1
Ja

Ju

00
n-

l-9

97

8
Ja

Ju

n-

l-9

94

5
Ja

Ju

Ja

n-

2

91

l-9
Ju

Ja

n-

9

88

l-8
Ju

n-

l-8

6
Ja

Ju

Ja

n-

85

–1

Federal Reserve Bank of St. Louis REVIEW

Thornton

Table 1
Estimates of Equation (1) Not Accounting for the Trend (January 1985–June 2008)
Variable

Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value

Const.

0.203

0.775

–3.277

0.000

–2.730

0.000

–2.182

0.000

–2.503

0.000

gap

0.241

0.000

0.220

0.000

0.211

0.000

0.180

0.002

0.192

0.001

cpi

0.320

0.000

0.497

0.000

0.433

0.000

0.307

0.000

0.284

0.000

lrid

0.250

0.015

0.374

0.001

0.381

0.000

0.377

0.001

0.378

0.001

rv

0.492

0.053

1.225

0.000

1.094

0.000

0.943

0.000

1.049

0.000

S1

–0.001

0.003
0.001

0.000
0.001

0.000
0.044

0.000
0.289

0.000

S2
S3
S3 gdp
S GV
–
R2

0.812

0.816

0.842

0.847

0.843

SE

0.402

0.398

0.365

0.363

0.362

NOTE: SE, standard error.

cycles for T10 and TP. The correlation between TP and T10 is very high, 94 percent. Given
the similarity of these series, it is not surprising that the results are qualitatively similar with
either measure.
Hamilton and Wu (2012) investigate the effectiveness of the portfolio balance channel
using the slope of the yield curve (SYC), measured by the difference between the constant
maturity 10-year Treasury zero coupon bond yield and the 6-month T-bill rate. Figure 6 shows
SYC and TP over the period January 1985 through June 2008.

The Effectiveness of the Portfolio Balance Channel
This section reports the estimates of equation (1) using these alternative measures of it
(TP and SYC) and a variety of measures of pdt . The macroeconomic variables are those used
by Gagnon et al. (2011): the unemployment gap (gap), core consumer price index inflation
(cpi), long-run inflation disagreement (lrid), and 6-month realized daily volatility of the onthe-run 10-year Treasury yield (rv).10
Table 1 presents the estimates using TP as the dependent variable and the alternative
measures of the public’s holding of Treasury debt discussed previously (see “Alternative Debt
Supply Measures”). Contrary to the implication of the portfolio balance channel, the coefficient on the public’s holding of debt net of SOMA, S1, is negative and statistically significant.
However, when foreign official holdings of Treasury debt are netted out, the estimate is positive
and statistically significant. A coefficient of the same magnitude and statistical significance is
obtained when foreign official holdings of agency and private debt are netted out. Hence,
despite the abnormal nature of this adjustment, it has no effect on the results: A $600 billion
LSAP would reduce the term premium by 40 basis points.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

63

Thornton

Table 2
Estimates of Equation (1) Accounting for the Trend (January 1985–June 2008)
Variable

Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value

Const.

1.035

0.058

0.164

0.850

–0.380

0.671

–0.071

0.945

–0.399

0.715

gap

0.205

0.000

0.200

0.000

0.201

0.000

0.192

0.001

0.196

0.001

cpi

0.109

0.097

0.156

0.054

0.207

0.024

0.158

0.056

0.167

0.032

lrid

0.244

0.029

0.276

0.020

0.301

0.009

0.292

0.016

0.297

0.013

rv

0.394

0.056

0.574

0.005

0.665

0.001

0.590

0.003

0.681

0.002

Trend

–0.006

0.000

–0.006

0.000

–0.004

0.010

–0.005

0.053

–0.004

0.069

S1

–0.0001

0.533

S2

0.0002

0.409

S3

0.0003

0.145

S3 gdp

0.0160

0.368

S GV

0.1150

0.363

–
R2

0.855

0.856

0.856

0.857

0.851

SE

0.353

0.352

0.348

0.352

0.353

NOTE: SE, standard error.

The estimates using S3 gdp are, of course, identical to those of Gagnon et al. (2011). While
the coefficient on S3 gdp is larger than that on S3, the estimated magnitude of the effect of LSAPs
is smaller. A $600 billion LSAP is about 4.0 percent of 2009 GDP, so the estimated effect of
the same $600 billion LSAP would be half as large, about 19 basis points. Given the similarity
between S3 gdp and SGV, it is not surprising that the estimate of the coefficient also is positive and
highly statistically significant when Greenwood and Vayanos’s (forthcoming) measure is used.
Unfortunately, the results in Table 1 are the consequence of trends in TP and public debt
measures. A simple linear trend accounts for nearly 80 percent of the variation in TP and a
small but still relatively large (as much as 60 percent) portion of the variation in the supply
measures. The existence of trends can lead to spurious regression in that two variables that
trend in the same direction will be positively correlated even if they are independent of each
other. The importance of the trend in these measures is shown in Table 2, which reports the
estimates of equation (1) when a simple linear trend is included in the equation. The coefficient
on S1 remains negative but is not statistically significant. The coefficients on the other debt
measures remain positive but much smaller than the estimates in Table 1; more importantly,
none is statistically significant at even the 10 percent significance level. When the trend is
accounted for, the statistical support for the portfolio balance channel vanishes. This conclusion is the same if the trend is obtained using the Hodrick-Prescott filter or if the equation is
estimated in first differences. Hence, there is no statistically significant positive relationship
between the term premium and any of the debt measures considered here (i.e., no statistical
support for the portfolio balance channel) when the trend is accounted for.
64

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Thornton

Table 3
Estimates of Equation (1) Using Alternative Supply Measures and Accounting for the Trend
(January 1985–June 2008)
Variable

Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value

Const.

0.0831

0.9198

0.6770

0.2145

2.4388

0.0092

9.0658

0.0000

–0.0500

0.9602

gap

0.2169

0.0002

0.2068

0.0003

0.2055

0.0002

0.3410

0.0000

0.2028

0.0002

cpi

0.1874

0.0006

0.1882

0.0006

0.0923

0.2075

0.0227

0.6266

0.1670

0.0399

lrid

–0.0577

0.4817

–0.0611

0.4622

0.1729

0.0904

0.0821

0.1560

0.2898

0.0143

rv
Trend
AM

0.6981

0.0007

0.6682

0.0012

0.4253

0.0507

0.5823

0.0000

0.6110

0.0027

–0.0060

0.0000

–0.0069

0.0000

–0.0061

0.0000

0.0002

0.7840

–0.0055

0.0001

0.0130

0.1439
0.0246

0.2020
–0.2569

0.0497
–1.1908

0.0000

P10
DUR
DUR10
S2duradj

0.0003

0.2970

–
R2

0.8366

0.8357

0.8588

0.9227

0.8540

SE

0.2995

0.3003

0.3452

0.2555

0.3511

NOTE: SE, standard error.

Table 3 presents the estimates using five alternative supply measures: the average maturity
of the debt (AM), the percent of the public’s holding of debt with maturity of 10 years or longer
(P10), the duration of the public’s holding of the debt (DUR), the duration of the on-the-run
10-year Treasury securities (DUR10), and the S2 debt measure adjusted for the duration of the
debt using Gagnon et al.’s (2011) adjustment procedure (S2duradj). AM and P10 are calculated
from Hamilton and Wu’s (2012) data, which cover the period January 1990 through June 2008.
Hence, this is the sample period when these variables are used. DUR and DUR10 data were
provided by Gagnon et al. (2011).
The coefficients on AM and P10 are positive, but neither is statistically significant. The
estimates for the two duration measures are negative and statistically significant, suggesting
that a shortening of the duration of the public’s holding of government debt as the result of
LSAPs would increase the term premium. The coefficient on S2duradj is positive but not statistically significant. Hence, these alternative measures also provide no support for the portfolio
balance channel.11 Again, the conclusion is robust to the measure of trend used, whether the
equation is estimated using first differences or whether T10 is the dependent variable (Tables 4
and 5).
The portfolio balance channel is thought to reduce longer-term rates relative to shorterterm rates, so equation (1) is estimated using SYC as the dependent variable and all 10 supply
measures. Unlike TP and T10, there is no significant trend in SYC. However, SYC is highly
persistent, so SYCt–1 is included in the regression.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

65

Thornton

Table 4
Estimates of Equation (1) Using Dependent Variable T10 and Not Accounting for the Trend
(January 1985–June 2008)
Variable
Const.

Coefficient

p-Value

Coefficient

p-Value

Coefficient

p-Value

Coefficient

p-Value

4.8405

0.0000

4.0288

0.0018

4.3217

0.0036

4.4339

0.0059

gap

–0.2472

0.0037

–0.2529

0.0035

–0.2477

0.0036

–0.2576

0.0048

cpi

0.3100

0.0044

0.3770

0.0029

0.3734

0.0100

0.3389

0.0092

lrid

0.6654

0.0003

0.7113

0.0001

0.7003

0.0001

0.7002

0.0001

rv
Trend
S1

0.2827

0.4145

0.4109

0.2282

0.3425

0.3308

0.3032

0.3738

–0.0146

0.0000

–0.0123

0.0000

–0.0119

0.0000

–0.0117

0.0016

0.0003

0.3116
0.0004

0.2610
0.0002

0.4527
0.0165

0.5699

S2
S3
S3 gdp
–
R2

0.8999

0.9007

0.8998

0.8995

SE

0.5536

0.5513

0.5539

0.5548

NOTE: SE, standard error.

Table 5
Estimates of Equation (1) Using Dependent Variable T10 and Accounting for the Trend
(January 1985–June 2008)
Variable
Const.

Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value
6.6212

0.0000

6.1643

0.0000

11.0086

0.0000

19.8398

0.0000

4.4411

0.0012

gap

–0.2560

0.0021

–0.2567

0.0021

–0.2396

0.0039

–0.0044

0.9117

–0.2462

0.0035

cpi

0.3168

0.0034

0.3003

0.0045

0.2136

0.0402

0.1312

0.0259

0.3490

0.0052

lrid

0.5869

0.0009

0.5998

0.0009

0.3925

0.0088

0.3632

0.0000

0.6992

0.0001

rv

0.0956

0.7541

0.1266

0.6949

0.0761

0.8035

0.3908

0.0098

0.3286

0.3340

Trend

–0.0143

0.0000

–0.0139

0.0000

–0.0125

0.0000

–0.0019

0.0741

–0.0126

0.0000

AM

–0.0146

0.4315
–0.0364

0.3105
–0.8946

0.0000
–2.0904

0.0000

P10
DUR
DUR10
S2duradj

0.0004

0.4904

–
R2

0.8999

0.9001

0.9238

0.9594

0.8996

SE

0.5537

0.5532

0.4831

0.3527

0.5544

NOTE: SE, standard error.

66

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Thornton

Table 6
Estimates of Equation (1) with SYC as Dependent Variable (January 1985–June 2008)
Variable

Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value

Const.

0.0568

0.8479

0.2655

0.1046

0.1996

0.0456

0.1632

0.0646

0.1640

0.1060

gap

0.1086

0.0071

0.1135

0.0050

0.1088

0.0053

0.1137

0.0034

0.1108

0.0038

cpi

0.0338

0.2599

0.0264

0.1451

0.0326

0.0637

0.0486

0.0148

0.0422

0.0528

lrid

–0.1099

0.0001

–0.1194

0.0001

–0.1189

0.0000

–0.1212

0.0000

–0.1160

0.0001

rv

0.2528

0.0013

0.2125

0.0037

0.2214

0.0027

0.2343

0.0015

0.2164

0.0026

SYC t–1

0.9085

0.0000

0.9066

0.0000

0.9112

0.0000

0.9112

0.0000

0.9119

0.0000

S1

0.0000

0.8088
–0.0001

0.3315
0.0000

0.1937
–0.0051

0.0795

S2
S3
S3 gdp
S GV

–0.0170

0.4623

–
R2

0.9686

0.9687

0.9688

0.9691

0.9703

SE

0.2024

0.2018

0.2015

0.2007

0.1984

NOTE: SE, standard error.

Table 7
Estimates of Equation (1) with SYC and Alternative Supply Measures (January 1985–June 2008)
Variable
Const.
gap

Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value Coefficient p-Value
–0.4507

0.0115

–0.5291

0.0013

–0.8071

0.0531

–0.5240

0.3026

0.1946

0.1944

0.1708

0.0021

0.2173

0.0002

0.1289

0.0013

0.0998

0.0119

0.1105

0.0049

cpi

0.0257

0.1834

0.0189

0.2795

0.0482

0.0124

0.0568

0.0414

0.0282

0.1149

lrid

–0.0387

0.4037

–0.0399

0.3751

–0.0709

0.0150

–0.0998

0.0005

–0.1161

0.0001

rv

0.5325

0.0001

0.5211

0.0001

0.2856

0.0006

0.2563

0.0006

0.2257

0.0022

SYC t–1

0.8471

0.0000

0.8177

0.0000

0.8905

0.0000

0.9112

0.0000

0.9075

0.0000

AM

0.0027

0.3846
0.0201

0.0959
0.1375

0.0268
0.0705

0.2073

P10
DUR
DUR10
S2duradj

0.0000

0.6032

–
R2

0.9740

0.9741

0.9699

0.9689

0.9686

SE

0.1948

0.1940

0.1980

0.2014

0.2023

NOTE: SE, standard error.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

67

Thornton

Table 6 presents the estimates using S1, S2, S3, S3 gdp, and SGV. None of these measures
provides support for the portfolio balance channel. The coefficients on S1 and S3 are positive
but not statistically significant. The coefficients on S2, S3 gdp, and SGV are negative; however,
none is statistically significant at the 5 percent significance level.
Table 7 presents the estimates using the five alternative supply measures. These results
are somewhat more supportive of the portfolio balance channel. The coefficient estimates for
AM and P10 are positive but not statistically significant at the 5 percent level. The coefficient
on P10 is marginally significant at the 10 percent level but is small: A 1-percentage-point
increase in the percent of debt with maturity of 10 years or longer increases SYC by only 2
basis points.
The coefficient on DUR is positive and statistically significant. A 1-year increase in DUR
would increase the SYC by nearly 14 basis points. However, the standard deviation of DUR over
the sample period is about 0.5 years. Consequently, a relatively large change in DUR would be
required to have a very large effect on SYC. D’Amico et al. (2012) indicate that the average
duration of the Treasury securities held by the public declined from 4.42 years to 4.30 years
during the first LSAP. Hence, this would account for only about a 2-basis-point flattening of
the yield curve during this period. They note that QE1 removed only 0.10 years of duration
from the market, so the duration effect of QE2 would be even smaller.
The duration-adjusted S2 supply measure also provides no support for the portfolio balance channel. The estimated coefficient is positive, but it is also very small and not statistically
significant.
Overall, the evidence is more supportive of the portfolio balance effect using SYC. Two of
the 10 supply measures yield coefficients that are correctly signed and statistically significant.
However, in either case the effect on SYC is modest. Consequently, the measures cannot account
for the well-documented decline in long-term interest rates and the term premium reported
in the event study literature.

CONCLUSION
With its principal policy tool—the federal funds rate—effectively at zero, the FOMC has
attempted to stimulate aggregate demand by reducing longer-term rates through the so-called
signaling and portfolio balance channels of policy. The portfolio balance channel assumes
that the market for long-term Treasury securities is segmented from the rest of the financial
market and hypothesizes a positive relationship between the term premium in long-term bonds
and the quantity of long-term debt held by the public. By implication, the portfolio balance
channel suggests that term premiums, and consequently long-term Treasury yields, can be
reduced through LSAPs or by purchasing longer-term securities while simultaneously selling
an equal quantity of shorter-term securities.
This article uses the reduced-form methodology of Gagnon et al. (2011) and Krishnamurthy
and Vissing-Jorgensen (2012) to investigate the portfolio balance channel with 3 interest rate
measures and 10 public debt supply measures from the literature. The results indicate there is
68

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Thornton

no statistically significant positive relationship between either the term premium or the 10-year
Treasury yield and any of the 10 supply measures. However, there is a statistically significant
relationship between the SYC and 2 of the 10 supply measures. The estimate of the effect is very
small and cannot account for the estimates of the effect of LSAPs on long-term yields reported
in the event study literature. Hence, there appears to be no empirical support for the idea that
these purchases reduced long-term yields or flattened the yield curve by reducing the public’s
holdings of long-term debt as the portfolio balance channel suggests it should have. ■

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

69

Thornton

NOTES
1

See Thornton (2010).

2

See Thornton (2013) for an analysis of the extent to which the announcement effects in the event study literature
are identified.

3

Specifically, Bernanke notes that “The channels through which the Fed’s purchases affect longer-term interest
rates and financial conditions more generally have been subject to debate. I see the evidence as most favorable
to the view that such purchases work primarily through the so-called portfolio balance channel, which holds that
once short-term interest rates have reached zero, the Federal Reserve’s purchases of longer-term securities affect
financial conditions by changing the quantity and mix of financial assets held by the public.”

4

This formulation seems at odds with the concept of scarcity. A more natural measure would seem to be the public’s holding of Treasury securities of given maturities relative to the total supply of those maturities.

5

I would like to thank the authors of these studies for providing the data. The data for the article by Hamilton and
Wu (2012) can be found at http://research.stlouisfed.org/econ/thornton/.

6

The results are quantitatively similar and the qualitative conclusions identical if the Hamilton and Wu base series
is used, suggesting that including or excluding TIPS has only a minor effect on the results.

7

They provide no reason for making this adjustment. They merely note that they made it.

8

The Greenwood and Vayanos measure is available only through December 2007. Hence, estimates of equation (1)
using their measure are based on monthly data over the period January 1985 through December 2007.

9

Bauer and Rudebusch (2011) have an alternative estimate of the risk premium. However, their measure behaves
similarly to that of Gagnon et al. (2011). Indeed, the qualitative conclusions presented in this section are the same
when the Bauer and Rudebusch measure is used.

10 See Gagnon et al. (2011) for the precise definitions of these variables.
11 As before, the qualitative conclusions are unchanged if the 10-year Treasury yield or Bauer and Rudebusch’s

(2011) measure of the term premium is used. For completeness, Hamilton and Wu’s (2012) supply factors were
also used. These factors are available for the period January 1990 through July 2007. None of these supply factors
was statistically significant when either TP or T10 was the dependent variable.

REFERENCES
Bauer, Michael and Rudebusch, Glenn. “The Signaling Channel for Federal Reserve Bond Purchases.” Working Paper
No. 2011-21, Federal Reserve Bank of San Francisco, September 2011;
http://www.frbsf.org/economic-research/files/wp11-21bk.pdf.
Bernanke, Ben S. “The Economic Outlook and Monetary Policy.” Speech at the Federal Reserve Bank of Kansas City
Economic Policy Symposium, Jackson Hole, Wyoming, August 27, 2010;
http://www.federalreserve.gov/newsevents/speech/bernanke20100827a.htm.
Cochrane, John H. “Sense and Nonsense in the Quantitative Easing Debate.” VOX, December 7, 2010;
http://www.voxeu.org/index.php?q=node/5900.
Cochrane, John H. “Inside the Black Box: Hamilton, Wu, and QE2.” Unpublished manuscript, University of Chicago
Booth School of Business, March 3, 2011;
http://faculty.chicagobooth.edu/john.cochrane/research/papers/hamiton_wu_term_structure.pdf.
D’Amico, Stefania; English, William B.; Lopez-Salido, J. David and Nelson, Edward. “The Federal Reserve’s LargeScale Asset Purchase Programs: Rationale and Effects.” Finance and Economics Discussion Series No. 2012-85,
Board of Governors of the Federal Reserve System, December 2012;
http://www.federalreserve.gov/pubs/feds/2012/201285/201285abs.html.
70

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Thornton
D’Amico, Stefania and King, Thomas B. “Flow and Stock Effects of Large-Scale Treasury Purchases.” Finance and
Economics Discussion Series No. 2010-52, Board of Governors of the Federal Reserve System, September 2010;
http://www.federalreserve.gov/pubs/feds/2010/201052/201052abs.html.
Gagnon, Joseph; Raskin, Matthew; Remache, Julie and Sack, Brian. “The Financial Market Effects of the Federal
Reserve’s Large-Scale Asset Purchases.” International Journal of Central Banking, March 2011, 7(1), pp. 3-43.
Greenwood, Robin and Vayanos, Dimitri. “Bond Supply and Excess Bond Returns.” Review of Financial Studies (forthcoming). Advance Access published online January 14, 2014;
http://rfs.oxfordjournals.org/content/early/2014/01/14/rfs.hht133.full.pdf+html.
Hamilton, James D. and Wu, Jing Cynthia. “The Effectiveness of Alternative Monetary Policy Tools in a Zero Lower
Bound Environment.” Journal of Money, Credit, and Banking, February 2012, 44(Suppl. s1), pp. 3-46.
Hancock, Diana and Passmore, Wayne. “Did the Federal Reserve’s MBS Purchase Program Lower Mortgage Rates?”
Journal of Monetary Economics, July 2011, 58(5), pp. 498-514.
Joyce, Michael; Lasaosa, Ana; Stevens, Ibrahim and Tong, Matthew. “The Financial Market Impact of Quantitative
Easing.” Working Paper 393, Bank of England, August 2010;
http://www.bankofengland.co.uk/research/Documents/workingpapers/2010/wp393.pdf.
Kim, Don H. and Wright, Jonathan H. “An Arbitrage-Free Three-Factor Term Structure Model and the Recent Behavior
of Long-Term Yields and Distant-Horizon Forward Rates.” Finance and Economics Discussion Series No. 2005-33,
Board of Governors of the Federal Reserve System, August 2005;
http://www.federalreserve.gov/pubs/feds/2005/200533/200533pap.pdf.
Kocherlakota, Narayana. “Economic Outlook and the Current Tools of Monetary Policy.” Speech at the European
Economics and Financial Centre, London, England, September 29, 2010;
http://minneapolisfed.org/news_events/pres/speech_display.cfm?id=4555.
Krishnamurthy, Arvind and Vissing-Jorgensen, Annette. “The Effects of Quantitative Easing on Long-Term Interest
Rates.” Brookings Papers on Economic Activity, Fall 2011, pp. 215-65.
Krishnamurthy, Arvind and Vissing-Jorgensen, Annette. “The Aggregate Demand for Treasury Debt.” Journal of Political
Economy, April 2012, 120(2), pp. 233-67.
Neely, Christopher J. “Unconventional Monetary Policy Had Large International Effects.” Working Paper No. 2010-018D,
Federal Reserve Bank of St. Louis, July 2010, revised August 2013;
http://research.stlouisfed.org/wp/2010/2010-018.pdf.
Stroebel, Johannes and Taylor, John B. “Estimated Impact of the Fed’s Mortgage-Backed Securities Purchase Program.”
NBER Working Paper No. 15626, National Bureau of Economic Research, December 2009;
http://www.nber.org/papers/w15626.pdf?new_window=1.
Swanson, Eric T. “Let’s Twist Again: A High-Frequency Event-Study Analysis of Operation Twist and Its Implications for
QE2.” Working Paper No. 2011-08, Federal Reserve Bank of San Francisco, February 2011;
http://www.frbsf.org/economic-research/files/wp11-08bk.pdf.
Thornton, Daniel L. “Can the FOMC Increase the Funds Rate Without Reducing Reserves?” Federal Reserve Bank of
St. Louis Economic Synopses, 2010, No. 28, October 6, 2010;
http://research.stlouisfed.org/publications/es/10/ES1028.pdf.
Thornton, Daniel L. “An Evaluation of Event-Study Evidence on the Effectiveness of the FOMC’s LSAP Program: The
Reasonable Person Standard.” Working Paper No. 2013-033A, Federal Reserve Bank of St. Louis, October 2013;
http://research.stlouisfed.org/wp/2013/2013-033.pdf.
Vayanos, Dimitri and Vila, Jean-Luc. “A Preferred-Habitat Model of the Term Structure of Interest Rates.” NBER Working
Paper No. 15487, National Bureau of Economic Research, November 2009;
http://www.nber.org/papers/w15487.pdf?new_window=1.
Woodford, Michael. “Monetary Policy in the Information Economy,” in Economic Policy for the Information Economy.
Proceedings of the Economic Policy Symposium sponsored by the Federal Reserve Bank of Kansas City, Jackson
Hole, Wyoming, August 20-September 1, 2001, pp. 297-370;
http://www.kansascityfed.org/Publicat/sympos/2001/papers/S02wood.pdf.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

71

Thornton
Woodford, Michael. “Methods of Policy Accommodation at the Interest-Rate Lower Bound,” in The Changing Policy
Landscape. Proceedings of the Economic Policy Symposium sponsored by the Federal Reserve Bank of Kansas
City, Jackson Hole, Wyoming, August 20-September 1, 2012, pp. 185-288;
http://www.kansascityfed.org/publicat/sympos/2012/Woodford_final.pdf.
Wright, Jonathan H. “What Does Monetary Policy Do to Long-Term Interest Rates at the Zero Lower Bound?”
Economic Journal, November 2012, 122(564), pp. F447-66.

72

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

The Evolution of Federal Reserve Policy and the
Impact of Monetary Policy Surprises on Asset Prices
Brett W. Fawley and Christopher J. Neely

This article describes the joint evolution of Federal Reserve policy and the study of the impact of
monetary policy surprises on high-frequency asset prices. Since the 1970s, the Federal Open Market
Committee has clarified its objectives and modified its procedures to become more transparent and
predictable. Researchers have had to account for these changes to procedures and perceived objectives in developing methods to study the effects of monetary surprises. Unexpected changes to the
Committee’s federal funds target and postmeeting statements strongly and consistently affect asset
prices, including interest rates, exchange rates, and (for target changes) stock prices. The study of
monetary surprises on asset prices provides important insight for policymakers, financial market
participants, and economic models. (JEL E52, E58, G14)
Federal Reserve Bank of St. Louis Review, First Quarter 2014, 96(1), pp. 73-109.

ince the 1970s, monetary policy has been the primary macroeconomic stabilization
instrument. In light of this fact, many researchers have studied how monetary policy
affects asset prices, consumer prices, output, and employment to improve such policy.
This large literature has used two main methods to study the effect of monetary shocks on
macroeconomic variables: vector autoregressions (VARs) and studies of high-frequency
monetary shocks on asset prices.1
VARs offer the advantage of directly studying the effects of monetary policy shocks on
key variables—prices, output, and employment—rather than indirectly studying them
through their effects on asset prices (see Litterman and Weiss,1985; Strongin,1995; Edelberg
and Marshall, 1996; Evans and Marshall, 1998; Bernanke and Mihov, 1998; and Christiano,
Eichenbaum, and Evans, 1999). It is difficult to isolate the effects of policy-induced changes
in interest rates on monthly or quarterly macroeconomic variables from changes induced by
other factors, however, and it is equally difficult to definitively differentiate the effects of monetary policy shocks from the effects of variables to which monetary policy reacts. That is, VAR

S

Brett Fawley is a quantitative analytics and reporting analyst at Rosen Consulting Group. Christopher J. Neely is an assistant vice president and
economist at the Federal Reserve Bank of St. Louis. This article was written while Fawley was a senior research associate at the Federal Reserve
Bank of St. Louis. The authors thank Rasmus Fatum, Ken Kuttner, Ed Nelson, Carlo Rosa, Juan Sánchez, Eric Swanson, and Dan Thornton for helpful comments and Sean Grover for research assistance.
© 2014, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views
of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. 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.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

73

Fawley and Neely

analysis requires controversial identification assumptions to identify simultaneous causality
because time aggregation of data to lower frequencies—such as the monthly or quarterly data
used in VAR analysis—generally produces simultaneous causality in economic data even if
there is unidirectional causality at very high frequencies. In other words, although it is unlikely
that macro variables or asset price changes within the meeting day influence Federal Open
Market Committee (FOMC) policy decisions, asset price changes in the weeks before such
decisions very likely have influenced such decisions. The combination of simultaneity and the
omission of many variables that affect asset prices inherently leave a great deal of uncertainty
about the effect of monetary policy on monthly prices, output, and employment.
It is far easier to identify the effect of high-frequency (daily, hourly) monetary shocks on
asset prices. If the monetary policy instrument and market expectations for its value are known,
then it is possible to characterize the impact of monetary policy shocks—deviations from
expectations—on asset prices, which react quickly to news and transmit monetary policy to
the economy. Because financial markets are forward looking, one would normally expect asset
prices to react only to the unexpected portion of monetary policy changes, as the expected
portion would already be priced into assets. Such high-frequency studies of the effect of monetary shocks on asset prices interest both market participants and economists and constitute a
useful first step to answering larger questions about the effects of monetary policy on macro
variables.
Why study the effect of monetary policy shocks when systematic monetary policy presumably has greater total effects? Both systematic and unsystematic policy actions might be
expected to affect asset prices. However, the effects of the systematic policy arise as new
information (e.g., data releases, policy statements) becomes available and reshapes market
expectations about the economy and the likely policy reaction. These expectations about
economic conditions and the central bank’s reaction function form over time and are influenced by both monetary and nonmonetary events. Therefore, it is very difficult to estimate
the effects of these unobservable changes in expectations of systematic monetary policy on
asset prices, though they will have an effect. In contrast, when a central bank makes a discrete
change to policy, the monetary surprise changes expectations immediately—by definition—
and it is easy to determine the effects of such surprises on asset prices, which inform us about
the transmission of all monetary policy.
Such investigations have proliferated since the seminal work of Kuttner (2001), but efforts
to draw those lessons together have been only limited and fragmented. This article remedies
that deficiency by reviewing the literature that has sought to both characterize the response
of asset prices to high-frequency monetary policy shocks and—more ambitiously—to explain
those reactions.2
A central theme in this literature is that financial markets’ expectations of monetary policy
have become more accurate as Federal Reserve policy has become more transparent in its
objectives and procedures. In the 1970s, the FOMC allowed inflation to rise to intolerable
levels, which was symptomatic of the lack of clarity in the Fed’s ultimate objectives. The fact
that Cook and Hahn (1989) were able to link federal funds target changes—not just surprises—
in the 1970s to asset price changes suggests that the Fed’s lack of clarity produced such poor
74

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

expectations of funds target changes that such changes were largely unexpected, nearly equivalent to surprises.
In 1979, however, the Federal Reserve’s new Chairman, Paul Volcker, clarified that one of
the Fed’s long-run objectives was price stability, as he sought to dramatically reduce the inflation rate. By the late 1980s, the Fed had reestablished its commitment to price stability and was
functionally again using the federal funds rate to achieve its objectives.3 As markets came to
understand Fed objectives and its likely reactions, only the unexpected portion of federal funds
target changes (the surprise) affected asset prices by the late 1980s (Kuttner, 2001). Researchers
exploited a liquid federal funds futures market to accurately estimate market expectations
and reduce the amount of measurement error in expectations (and, thus, surprises).
The Fed continued to become more transparent in its procedures in 1994-95 by greatly
reducing unscheduled changes in the funds target, explicitly announcing the funds target at
the conclusion of FOMC meetings, and describing the FOMC’s view of economic conditions
and monetary policy after each policy meeting. Each of these actions improved the market’s
ability to anticipate or react to Fed actions. Reducing intermeeting target changes made the
timing of those changes easier to estimate and reduced the problem of omitted variables.4
Announcing the new target allowed market prices to reach the new target quickly. FOMC
meeting statements allowed the FOMC to influence long rates by communicating its view of
the economy and policy and presumably improved market forecasts of future policy (Poole,
Rasche, and Thornton, 2002; Swanson, 2006). The influence of FOMC statements on the
yield curve prompted Gürkaynak, Sack, and Swanson (2005) to develop a two-factor model
of monetary policy shocks that has influenced further research.
In brief, researchers have found that federal funds rate surprises have consistent and sizable
effects on other asset prices, including long-term interest rates, foreign exchange, and equities.
For example, Kuttner (2001) estimates that a 100-basis-point surprise increase in the federal
funds rate would raise 10-year interest rates by about 30 basis points. Andersen et al. (2003)
calculate that the same increase would raise the value of the dollar by 66 to 107 basis points,
and Bernanke and Kuttner (2005) find that such a surprise would reduce equity prices by
about 4 percent.
The rest of this article reviews the challenges involved in defining monetary policy shocks
and accurately estimating their impact on asset prices.

RESEARCH ON MONETARY POLICY SHOCKS AND ASSET PRICES
The efficient markets hypothesis implies that, because financial markets are forward looking, only the unexpected portion of a monetary policy change should influence asset prices
and it should do so very quickly (Fama, 1970). Therefore, any study of monetary policy must
decompose actions into expected and unexpected portions; that decomposition depends on
market perceptions of Federal Reserve objectives, procedures, and communications. A central
theme of this article is that research on monetary policy shocks has evolved jointly with those
FOMC objectives, procedures, and communications. As the Federal Reserve has become
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

75

Fawley and Neely

more transparent in its objectives and communications, it has become better able to influence
asset markets without disrupting them (Bernanke and Kuttner, 2005; Gürkaynak, Sack, and
Swanson, 2005).

Early Research on Effects of Monetary Shocks
Any study of the effects of monetary policy must first define the monetary policy
instrument, the primary tool that a central bank uses to influence the economy. Monetary
instruments are closely related to, but distinct from, the stance of monetary policy—the contribution that monetary policy makes to economic and financial conditions—because any change
in the instrument is likely to imply a change in the stance of monetary policy in the short run,
though the stance can certainly change without a change in the instrument in the longer run.
That is, researchers are interested in the monetary policy instrument because unexpected
changes in the instrument equate to unexpected changes in the stance of monetary policy,
though the stance of policy can change without a change in the instrument.
It is natural to think of monetary policy as operating through changes in measures of
money and monetarists, such as Friedman, commonly characterized the stance of monetary
policy with reference to the growth rate of money.5 Indeed, the Fed described its own policies in the 1970s and 1980s in terms of targeting various measures of reserves or monetary
aggregates.
Thus, empirical studies of monetary policy during the 1970s and 1980s typically associated changes in monetary policy with changes in measures of the money supply. Sims (1980),
for example, describes the monetarist view that “the time path of the money stock is a good
single index of monetary policy.” He questions the efficacy of monetary policy on the grounds
that money adds no additional explanatory power to a system that includes output, prices,
and a short interest rate.
The immediate predecessor of the literature examining high-frequency monetary policy
shocks was the literature that searched for a liquidity effect: an injection of reserves that decreases
the nominal interest rate.6 A large literature searched for this liquidity effect in various ways:
estimating investment-savings, liquidity-preference-money-supply (IS-LM) models; regressing interest rates on functions of money growth; and studying asset price behavior around
monetary stock announcements. The results were mixed, at best, with many studies finding
negative or perverse results with aggregate data. Reichenstein (1987) surveys this literature
and concludes that “the Fed appears to have little control over month-to-month changes in
[short-term] interest rates.” Despite the lack of clear success, researchers used VARs with aggregate data to continue to look for a liquidity effect well into the 1990s (Thornton, 2001b). The
increasing adoption and recognition of overt interest rate targeting by central banks ultimately
brought an end to this line of research, however (Friedman and Kuttner, 2011).

Measuring Monetary Policy with the Federal Funds Rate
Although the Federal Reserve described its policy objectives in the 1970s and much of the
1980s in terms of targets for some measure of money or reserves, most central banks normally
76

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

conduct monetary policy by trading short-term securities or managing short-term loans to
banks to target a short-term interest rate. Observing that central banks usually conduct monetary policy through a short-term interest rate and that monetary aggregates do not predict
output or interest rates very well, McCallum (1983) reasoned that if the Fed’s behavior is determined by an interest rate rule, “policy innovations—i.e., the unsystematic portion of the Fed’s
behavior—are then represented as the stochastic disturbances appearing in this interest rate
rule, not in some rule determining the value of the money stock.”
Building on such observations, researchers began to equate the federal funds rate with the
U.S. monetary policy instrument by the late 1980s. Cook and Hahn (1989) argue that, because
the Fed uses the federal funds rate as its instrument and does not quickly reverse target changes,
the funds rate should influence other interest rates.7 These authors regress changes in bill or
bond rates on 75 federal funds target changes from September 1974 through September 1979,
a period in which the Fed tightly controlled the funds rate:
(1)

∆Rt = β1 + β 2 ∆RFFt + ut ,

where DRt is the change in a bond or bill rate and DRFFt is the federal funds target change.
Contrary to Reichenstein’s (1987) conclusions surveying studies of the impact of money on
interest rates, Cook and Hahn’s results imply that a 1-percentage-point increase in the federal
funds rate was associated with a 55-basis-point increase in 3-month Treasury bills and a 13basis-point increase in 10-year bonds.8 Table 1 summarizes the empirical results from important papers in this literature while Table 2 describes their contributions.
Bernanke and Blinder (1992) implicitly support this use of the funds rate, arguing that the
federal funds rate is “a good indicator of monetary policy” because it forecasts macro variables
better than any other variables—suggesting that it might affect these macro variables—and
responds systematically to unemployment and inflation, which reflect the Fed’s dual mandate.
Furthermore, they argue that because the funds rate is insensitive to changes in demand for
reserves, it “is mostly driven by policy decisions.”
Some would criticize the use of regressions such as equation (1) to determine the effect of
changes in the federal funds target on the grounds that the effect of the announcement change
on asset prices is measured only over one day—or a few days—and might be temporary. Such
criticisms are misplaced. Because uncertainty about asset prices usually rises with the forecast
horizon, no one can know the long-term effects of any event on asset prices. The efficient
markets hypothesis implies that the market’s best guess must have been that the effects of the
federal funds target change would persist. Otherwise, expectations of a temporary impact of
a policy announcement would create a risk-arbitrage opportunity for investors to bet on the
reversal of the policy’s effects.

The Importance of Expectations
Studies on later samples failed to confirm the Cook and Hahn (1989) results, however. In
particular, Radecki and Reinhart (1994) and Roley and Sellon (1995) failed to find any significant relationship between federal funds target changes and interest rates over later samples.9
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

77

First Quarter 2014

Estimated Impact of Monetary Policy Surprises on Interest Rates
Kuttner
(2001)
Table number in
original article
Sample

Poole,
Poole,
Poole and
Rasche, and
Rasche, and
Rasche (2000) Thornton (2002) Thornton (2002)

Rigobon and
Sack (2004)

Rigobon and
Sack (2004)

Gürkaynak,
Fleming and
Sack, and
Piazzesi (2005) Swanson (2005)

Hamilton
(2008)

Table 3

Table 3

Table 7

Table 3

Table 4

Table 4

Table 1

Table 1

Table 2

1989:06–
2000:02

1988:10–
2000:02

1987:08–
1993:12

1994:03–
2001:05

1994:01–
2001:11

1994:01–
2001:11

1994:01–
2004:12

1991:07–
2004:12

1998:10–
2006:12
4,552

Observations

42

53

38

62

73

73

93

120

Daily

Daily

Daily

Daily

Daily

Daily

1-hour

30-minute

Daily

Expectations
measure

Currentmonth FF

Nextmonth FF

Nextmonth FF

Nextmonth FF

Currentmonth FF

Currentmonth FF

Currentmonth FF

Expected and
surprise?

With
anticipated

With
anticipated

With
anticipated

With
anticipated

No correction

No correction

Data frequency

Correction?

Errors-in-variables Errors-in-variables

correction

correction

Current 3-month Current 3-month
eurodollar
eurodollar
future
future
Surprise only

Surprise only

Surprise only

Surprise only

Surprise only

ID through

ID through

Intraday data

Intraday data

No correction

66.6
(4.4)

53.7
(4.0)

65.8
(2.2)

60.9
(7.4)

52.2
(5.7)

70.6
(2.1)

heteroskedasticity heteroskedasticity

(IV)

(GMM)

Maturity of regressand

Federal Reserve Bank of St. Louis REVIEW

3-month

79.1
(9.4)

6-month

71.6
(8.4)

12-month

71.6
(9.2)

2-year

61.4
(10.2)

73
(9.0)

78
(9.0)

82.3
(10.0)

80.8
(28.0)

89.9
(11.0)

63.5
(20.0)

87.6
(11.5)

47.1
(13.0)

91.8
(10.0)

54.6
(19.0)

75.6
(9.3)

27.6
(12.7)

76.1
(14.0)

36.4
(20.0)

79
(11.2)

15.5
(11.6)

74.8
(2.3)
48.3
(13.0)

45.5
(8.7)

3-year

68.5
(2.9)
64.1
(3.0)

5-year

48.1
(11.2)

10-year

31.5
(10.2)

30-year

19.4
(8.4)

48
(10.0)

56.9
(14.0)

18.2
(23.0)

93
(12.6)

12.5
(13.9)

31.3
(14.8)

26.4
(8.1)

42.6
(11.0)

2.7
(22.0)

61.1
(13.7)

0.8
(10.2)

8.3
(15.6)

12.5
(5.8)

27.7
(10.0)

–7.5
(13.0)

35.2
(13.6)

–13.3
(8.3)

42.6
(2.8)

NOTE: The lower half of the table displays regression coefficients (yield changes in basis points) and standard errors (in parentheses) from a number of articles on the impact of
monetary policy (federal funds rate target) surprises on changes in interest rates of varying maturities. FF, federal funds target rate; GMM, generalized method of moments; ID,
identification; IV, instrumental variables.

Fawley and Neely

78

Table 1

Federal Reserve Bank of St. Louis REVIEW

Table 2
Summary of the Most Important Papers in the Literature on Monetary Policy Surprises
Article

Method/Innovation

Results

Regression on federal funds target changes

“We find that changes in the target caused large movements in short-term rates and smaller
but significant movements in intermediate- and long-term rates.”

Kuttner (2001)

Identification of federal funds target surprises
using futures contracts

“Interest rates’ response to anticipated target rate changes is small, while their response to
unanticipated changes is large and highly significant.”

Poole, Rasche, and
Thornton (2002)

Correction for errors-in-variables bias

“The estimates…suggest that Treasury rates respond significantly to unexpected changes in
the Fed’s funds rate target…[T]he response of the Treasury rate to unexpected target
changes declines as the term lengthens…We find that the response of the 3-month T-bill rate
is nearly identical before and after [the 1994] procedural change. The magnitude and significance of the response of longer-term rates, however, declines after this procedural change.”

Rigobon and Sack (2004)

Correction for simultaneity/omitted variables

“The results indicate that an increase in short-term interest rates results in a decline in stock
prices and in an upward shift in the yield curve that becomes smaller at longer maturities. The
findings also suggest that the event-study estimates contain biases that make the estimated
effects on stock prices appear too small and those on Treasury yields too large.”

Bernanke and Kuttner
(2005)

Full sensitivity analysis for equities/Decomposition
of equity return

“We find that, on average, a hypothetical unanticipated 25-basis-point cut in the Federal
funds rate target is associated with about a 1% increase in broad stock indexes. Adapting a
methodology due to Campbell and Ammer, we find that the effects of unanticipated monetary policy actions on expected excess returns account for the largest part of the response of
stock prices.”

Gürkaynak, Sack, and
Swanson (2005)

Two-factor model

“We test whether [monetary policy] effects are adequately captured by a single factor—
changes in the federal funds rate target—and find that they are not. Instead, we find that two
factors are required. These factors have a structural interpretation as a ‘current federal funds
rate target’ factor and a ‘future path of policy’ factor, with the latter closely associated with
Federal Open Market Committee statements...According to our estimates, both monetary
policy actions and statements have important but differing effects on asset prices, with statements having a much greater impact on longer-term Treasury yields.”

Fleming and Piazzesi (2005)

Tick-by-tick analysis

“Analysis of high-frequency data shows that Treasury note yields are highly volatile around
FOMC announcements, even though the average effects of fed funds target rate surprises on
such yields are fairly modest.”

Hamilton (2008)

Daily model with uncertainty over event days

“This paper develops a generalization of the formulas proposed by Kuttner (2001) and others
for purposes of measuring the effects of a change in the federal funds target on Treasury
yields of different maturities…Although the methods are new, the conclusion is quite similar
to that reported by earlier researchers—changes in the fed funds target seem to be associated
with quite large changes in Treasury yields, even for maturities of up to 10 years.”

Hausman and Wongswan
(2011)

Two-factor model applied to full set of
international assets

“This paper analyzes the impact of U.S. monetary policy announcement surprises on foreign
equity indexes, short- and long-term interest rates, and exchange rates in 49 countries…
Global equity indexes respond mainly to the target surprise; exchange rates and long-term
interest rates respond mainly to the path surprise; and short-term interest rates respond to
both surprises.”

79

Fawley and Neely

First Quarter 2014

Cook and Hahn (1989)

Fawley and Neely

Thornton (1998) criticizes Cook and Hahn’s target change series as being endogenous—
that is, partially expected and therefore not true policy surprises. Likewise, Kuttner (2001)
argues that the failure to confirm Cook and Hahn’s results on later samples is due to the failure
to decompose monetary policy shocks into their expected and surprise components.10 Expected
funds rate changes should not influence other asset prices because financial markets are forward looking; only the unexpected component of changes in the federal funds target should
change other asset prices.11 In the 1970s, funds rate expectations were apparently so poor that
changes in the funds target were good proxies for their unexpected component. By the late
1980s, however, clearer Fed objectives and procedures allowed markets to anticipate a large
portion of federal funds target changes, so changes in the funds target became poor proxies
for their unexpected component. Therefore, researchers began to consider how to decompose
federal funds rate changes into expected and unexpected components to test the effect of the
latter with high-frequency data.
The most common method to compute expectations of the federal funds target is due to
Kuttner (2001), who used prices from the federal funds futures market. The basic idea is that
the futures market implies an average federal funds rate for a particular contract month and—
because the New York Open Market Desk can keep the average federal funds rate near the
target—this implies an average federal funds target for the contract month. Thus, the market’s
expectation of the change in the target on the FOMC meeting date can be calculated if the
target at the start of the month and the date on which the target might be changed are known.
Appendix A on the federal funds futures market details this procedure.
Using these procedures to decompose the surprise and expected components of federal
funds target changes from June 1989 to February 2000, Kuttner (2001) estimates
(2)

∆R tn = α + β1∆ rt e + β 2 ∆ rtu + ε t ,

where DRtn is the change in the yield of an n-year Treasury bond on date t, and Dr̃ te and Dr̃ tu
are the expected and surprise components of the federal funds target change on day t. Using
42 days of changes in the FOMC target, Kuttner finds that an unexpected 1-percentage-point
increase in the federal funds target raises 3-month Treasury yields by 79 basis points and 10year yields by 32 basis points. The reaction to the surprise component is significant at all maturities analyzed, while the impact of expected changes is always small and insignificant. Table 3
illustrates Kuttner’s results on the set of all FOMC events (all FOMC meetings plus intermeeting rate changes) from October 1988 to June 2007. Some anticipated target changes are statistically significant at the very short end of the yield curve, presumably because the central bank
can control the very short end of the yield curve through the open market operations that pin
down the federal funds rate. The unanticipated component, however, clearly has the much
stronger and more statistically significant impact on interest rates at all horizons.
Kuttner (2001) claims that federal funds futures offer three advantages over other procedures to identify expectations of monetary policy: (i) Futures require no model; (ii) futures
data are not revised and so there is no “data vintage” problem; and (iii) futures do not entail
an errors-in-variables problem as do VARs.12 In addition, Gürkaynak, Sack, and Swanson
80

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

Table 3
Response of Interest Rates to Expected and Unexpected Components of Federal Funds
Target Surprises (1988-2007)
Maturity

Intercept

Anticipated

Unanticipated

R2

3-month

–0.01
(–3.7)

0.05
(2.2)

0.66
(15.2)

0.60

6-month

–0.02
(–3.8)

0.05
(2.3)

0.64
(14.5)

0.58

12-month

–0.01
(–1.9)

0.03
(1.5)

0.62
(13.3)

0.53

2-year

–0.01
(–1.1)

0.05
(1.7)

0.46
(8.1)

0.30

5-year

0.00
(–0.8)

0.02
(0.7)

0.32
(5.2)

0.15

10-year

0.00
(–0.5)

–0.01
(–0.2)

0.17
(3.2)

0.06

30-year

–0.01
(–1.4)

–0.01
(–0.6)

0.08
(1.9)

0.02

NOTE: The table shows the results of a regression similar to that of Kuttner (2001, equation (2)) that uses all FOMC events
from October 1988 to June 2007. The event set includes 177 events, including all regularly scheduled FOMC meetings,
plus intermeeting rate changes. Shaded numbers denote significantly positive (blue) or negative (red) t-statistics.

(2007) show that federal funds futures are the best-available forecasts of federal funds targets
at horizons of up to 6 months.13 These advantages have made federal funds futures the standard
metric for computing expectations of the federal funds target.
The most attractive alternative to using federal funds futures data to calculate expectations
for the federal funds target is to use some sort of survey data. Andersen et al. (2003), for example, use Money Market Services (MMS) survey data in their broad study of the effects of macro
announcements on foreign exchange returns and volatility. Using data from January 1992 to
December 1998, they find that positive funds target surprises significantly appreciate the
dollar for four of the five exchange rates.
The use of expectations from the federal funds futures market, rather than from MMS
survey data, implies similar effects of funds target surprises on the foreign exchange market.
Fatum and Scholnick (2006) determine that changes in 2-month-ahead federal funds futures
on days without monetary policy news, which the authors interpret as policy expectations,
are significant predictors of three exchange rates at the daily frequency and the response is
rapid, within the day. Faust et al. (2007) find somewhat larger results with FOMC meetingday data from March 1995 through December 2002 and expectations from the federal funds
futures market. Their regressions using 20-minute windows imply that a 100-basis-point surprise increase in the funds target would depreciate the dollar 123 basis points against the
deutsche mark/euro (DEM/EUR) and 66 basis points against the British pound (GBP). The
impacts are highly statistically significant.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

81

Fawley and Neely

Noisy Measures of Monetary Policy
Computing shocks to the federal funds target entails confronting the fact that expectations of the federal funds target—and therefore the surprises in the federal funds target—are
measured with some error because of bid-ask spreads, stale expectations, and risk premia.14
When regressors are measured with error, the coefficients on those regressors are biased and
the expected values are attenuated toward zero. Poole, Rasche, and Thornton (2002) correct
for such errors-in-variables bias by estimating the size of the measurement error from policy
actions that were correctly anticipated. The authors gauge the degree of correct anticipation
from surveys and commentary in the “Credit Markets” section of the Wall Street Journal published on the days before FOMC meetings. Using this correction and with data from March
1994 through May 2001, the authors report that a 1-percentage-point reduction in the federal
funds target reduces 3-month Treasury yields by 77 to 82 basis points and 10-year Treasury
yields from 40 to 43 basis points. The corrected estimates are somewhat larger than (but
fairly close to) the uncorrected estimates, indicating that measurement error is not a major
problem.15
Hamilton (2008) develops a method to measure monetary policy surprises to account for
both noise induced by deviations in the effective federal funds rate from its target and uncertainty about the date of policy actions (as may have been important prior to 1994).16 The
method extracts the monetary policy signal from daily federal funds futures changes under
the assumption that the econometrician does not know the dates of policy actions and must
take them to be equally likely on all dates in the sample. Using this procedure, Hamilton (2008)
estimates his regression over all business days, not just those with FOMC meetings or known
target changes, to find that a 1-percentage-point surprise to the federal funds target increases
3-month Treasury yields by 66 basis points and 10-year yields by 43 basis points.

Simultaneity and Omitted Variables in Asset Price Reactions to Monetary Policy
If nonmonetary news influences monetary policy and asset prices within the event window, or if monetary policy responds contemporaneously to asset price changes within the event
window, then that complicates the study of monetary policy’s effect on asset prices. The first
problem is omitted variables bias, while the second problem is simultaneity bias.17 Appendix B
describes these problems in some detail. In either case, a naive event study estimate of the
impact of target rate changes on asset prices will inconsistently estimate the true impact of
those target rate changes. To understand these biases, consider a linear system in which monetary policy (Dmt) and asset prices (Dpt) are determined simultaneously as follows:
(3)

∆pt = a0 + a1∆mt + a2newst + e p ,t

(4)

∆mt = b0 + b1∆pt + b2newst + em ,t ,

where newst denotes macro releases that potentially affect monetary policy and asset prices.
82

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

Omitted variables bias occurs when nonmonetary news within the event window moves
financial prices but the nonmonetary regressor is incorrectly excluded. That is, a2 ≠ 0 but the
estimated regression imposes a2 = 0. In that case, the event study estimate of the impact of
target rate changes will be biased.
If monetary policy reacts to asset price changes within the observation interval for the
data—for example, within the day for daily data—then b1 will be nonzero in equation (4)
and an ordinary least squares (OLS) estimate of a1—the effect of the monetary policy change
on the asset price—will be inconsistent. As the observation interval around a policy change
becomes arbitrarily small, the parameter b1 will tend to go to zero—equation (4) will contain
only lagged values of Dpt , which present no simultaneity problem—and the OLS estimator of
a1 becomes consistent.
Although the FOMC rarely directly reacts to asset price changes within a day—that is,
it seems likely that b1 = 0 but there are lags of Dpt in equation (4)—it is more plausible that
omitted variables bias presents a problem. Specifically, prior to February 1994, Chairman
Alan Greenspan changed the federal funds target on several days of weak employment reports,
presumably in response to those releases (i.e., a2 ≠ 0 in equation (3)).18 A negative employment
report will tend to directly reduce equity prices and interest rates, but it will also tend to make
the FOMC reduce interest rates, which will tend to increase equity prices but reduce longerterm interest rates. If the whole effect were naively ascribed to the policy shock—ignoring possible joint-response bias—then equity responses would be underestimated and yield responses
to policy shocks would be overestimated. Bernanke and Kuttner (2005) report that failing to
account for the omitted variables bias related to target changes that occur on the same day as
employment reports attenuates the average stock market response in their sample by 1 percentage point.
To address these occasions when policy surprises react to asset price changes or (much
more frequently) react to other news, researchers have sought to identify the effects of monetary
policy shocks with (i) explicit identification schemes with daily data or (ii) higher-frequency
data and narrow event windows.19
Rigobon and Sack (2004) pursue the first strategy: These authors identify monetary shocks
from FOMC meetings and the semiannual monetary policy testimony from the daily conditional heteroskedasticity of nearby futures on 3-month eurodollar interest rates over a sample
from 1994 through November 2001. The nearby futures on 3-month interest rate contracts
are used to reduce problems with timing issues (as discussed more extensively in the following
subsections). These authors exploit the fact that policy shocks display greater variance on
announcement days than nonannouncement days to identify the effect of monetary policy
shocks, solving the usual problems of simultaneity/omitted variables in daily interest rate data
in a way that does not require the strong assumptions of daily event studies.20
The two heteroskedasticity estimators employed by Rigobon and Sack (2004) diverge
widely in the degree to which they indicate that event studies overstate yield responses. Rigobon
and Sack (2004) find that a 1-percentage-point surprise federal funds target cut (i) increases
broad stock indexes by 4.85 to 10.06 percentage points and (ii) lowers 6-month Treasury yields
by 47 to 88 basis points and 10-year Treasury yields by 1 to 61 basis points. The authors attribFederal Reserve Bank of St. Louis REVIEW

First Quarter 2014

83

Fawley and Neely

ute the large effects on stock prices to the lack of downward bias in their estimates compared
with event study estimates, but they also report that using federal funds futures instead of
futures on 3-month eurodollars produces smaller estimates of the impact of monetary shocks.
Rosa (2011b) reexamines the biases that Rigobon and Sack’s (2004) identification-throughheteroskedasticity methods are designed to confront. Rosa (2011b) finds a small but statistically significant bias in event study estimates of asset market reactions. Nevertheless, Rosa
recommends the event study estimator because its bias is small and it outperforms the heteroskedasticity-based estimator.
The second strategy to avoid omitted variables and/or simultaneity is to use highfrequency (intraday) data to estimate the effect of monetary policy on asset prices (a1). At a
sufficiently high frequency, asset prices presumably have no effect on monetary policy (b1 = 0)
and even news variables are predetermined, so the relation can be simply estimated by OLS.
Gürkaynak, Sack, and Swanson (2005) and Fleming and Piazzesi (2005) use intraday data from
1994 through 2004 and 30-minute event windows to estimate the effects of policy surprises
on yield changes. The Fleming and Piazzesi (2005) results generally confirm those from daily
event studies: A 1-percentage-point cut in the federal funds target lowers 3-month Treasury
yields by 67 basis points and 10-year Treasury yields by 8.3 basis points (although the latter
response is insignificant).21

Federal Funds Surprises and Equity Prices
A large literature debates the extent to which monetary policy affects the economy through
interest rate channels and/or credit channels. Monetary policy can affect consumption, investment, and the international competitiveness of domestic goods by influencing the foreign
exchange value of the domestic currency and stock prices. Credit channels exploit the fact
that easier monetary policy can improve consumers’ and firms’ balance sheets and thus reduce
the effect of certain financial frictions—adverse selection and moral hazard—that hinder borrowing (Mishkin, 1995).22 Thus, monetary policy is often thought to generate a substantial
portion of its effects through equity markets. Therefore, economists study how much and
why monetary surprises affect stock prices and what these answers imply for monetary policy
channels.
To answer these questions, Bernanke and Kuttner (2005) studied monetary policy effects
on equity prices using data from May 1989 through December 2002. This study carefully considered factors that could affect the response, such as distinguishing scheduled from unscheduled target changes, simultaneity between news and target changes, and the timing of target
changes—an advancement or postponement of an expected action—or whether they signal
persistent changes in interest rates. Positive (negative) federal funds target shocks significantly
reduced (raised) equity prices; high-tech sectors reacted more strongly than did broad indexes.
The authors found no evidence for asymmetry in the magnitude of positive/negative target
shocks, but they did find that reversals—for example, a target increase after a series of decreases—
had particularly strong effects on equities, although they cautioned there are only five such
observations in the sample. The authors also showed that six “outliers”—four of which are intermeeting moves—strongly influenced the estimated impact coefficients. The authors speculated
84

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

that these very strong effects were due to the fact that intermeeting moves convey much more
urgency than equally sized surprises at regularly scheduled FOMC meetings.
Perhaps the main contribution of Bernanke and Kuttner (2005), however, is their use of
federal funds surprises, measured with futures data, to study the source of the equity returns
with a Campbell and Shiller (1988) decomposition of excess equity returns into expectations
of future dividends, interest rates, and excess returns.23 This methodology implies that positive federal funds surprises reduce expected excess returns or dividends—depending on the
sample—but not real interest rates. The authors interpret an effect on expected excess returns
as arising from an increase in the riskiness of stocks or the willingness of investors to bear
stock risk. Alternatively, the change in expected excess returns might stem from overreaction
or excess sensitivity to policy.
The aggregated equity returns used by Bernanke and Kuttner (2005) cannot shed much
light on differential responses by individual stocks and industry portfolios to monetary policy
surprises. To investigate such heterogeneity, Ehrmann and Fratzscher (2004) regress daily
returns for individual U.S. firms and industry portfolios on monetary policy surprises from
79 FOMC meetings from 1994 through January 2003. The expectations are computed using
Reuters’ surveys of market participants. The results have mixed implications for the credit
and interest rate channels, however. The fact that stock prices of financially constrained firms
react strongly to monetary policy tends to support the credit channel, while the strong reaction
of firms in cyclical and capital-intensive industries tends to support the interest rate channel.
Basistha and Kurov (2008) investigate the reactions of individual stocks to find that all stocks
tend to react particularly strongly to monetary policy in recessions and tight credit conditions.
Stock prices of financially constrained firms display particularly strong asymmetry, however,
which the authors interpret as further evidence for the credit channel.
Two papers extend this research to investigate the impact of U.S. funds target surprises
on international equity prices. Ehrmann and Fratzscher (2009) consider how national characteristics—openness, exchange rate regime, and so on—determine the strength of the transmission process. On average, a 100-basis-point surprise U.S. tightening reduces equity prices
by 2.7 percent, although there is a great deal of heterogeneity in responses to U.S. monetary
policy surprises, both across countries and across sectors. Financially open and more integrated
countries have stronger equity market reactions. The reaction of U.S. short-term interest rates
to the target surprise appears to govern the strength of the overall transmission to foreign
equities, and countries with strong equity market reactions to U.S. target surprises also tend
to exhibit strong exchange rate and interest rate reactions.
Ammer, Vega, and Wongswan (2010) examine the impact of U.S. target surprises on firmlevel equities from the United States and 21 foreign countries.24 Consistent with Bernanke
and Kuttner’s (2005) results for U.S. equities, they find that an unexpected tightening of 100
basis points reduces U.S. and foreign equity prices by 6.4 percent and 6.8 percent, respectively.
The authors interpret the sensitivity of cyclical industries as supporting the demand (interest
rate) channel of monetary policy. In contrast to the Ehrmann and Fratzscher (2009) results,
Ammer, Vega, and Wongswan (2010) find that countries with fixed exchange rates respond
more strongly to U.S. target surprises than do countries with flexible exchange rates.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

85

Fawley and Neely

Increasing Predictability in Fed Policy
In 1994, the FOMC greatly reduced the frequency of intermeeting funds target changes
and began to issue statements after policy meetings. Such statements could potentially contain
three distinct pieces of information: (i) the current policy action in terms of the federal funds
target or asset purchases; (ii) a statement on current or future economic conditions; and (iii)
a forecast for the path of policy.25 Table 4, excerpted from Middeldorp (2011), describes changes
in FOMC communication policy from 1993 to 2007.
The information in the announcements was initially very simple—a qualitative description
of the change in the funds target:
Chairman Alan Greenspan announced today that the Federal Open Market Committee decided
to increase slightly the degree of pressure on reserve positions. The action is expected to be
associated with a small increase in short-term money market interest rates. (FOMC, 1994a)

Such postmeeting statements later became more elaborate, with numerical changes to the
target and brief discussions of economic conditions, as in this excerpt from the August 1994
FOMC press release (FOMC, 1994b):
The Federal Reserve announced today the following monetary policy actions:
• The Board of Governors approved an increase in the discount rate from 3½ percent to
4 percent, effective immediately.
• The Federal Open Market Committee agreed that this increase would be allowed to show
through completely into interest rates in reserve markets.
These measures were taken against the background of evidence of continuing strength in the
economic expansion and high levels of resource utilization. The actions are intended to keep
inflationary pressures contained, and thereby foster sustainable economic growth.

The practices of reducing the frequency of intermeeting target changes and issuing statements allowed markets to better anticipate the timing and direction of policy target changes
(see the boxed insert). In the pre-February 1994 sample, Poole, Rasche, and Thornton (2002)
“find few instances where there was a widespread expectation that the Fed would take an action
on a particular day.” In contrast, Poole, Rasche, and Thornton (2002) identify only 18 “surprise”
events in the set of 62 policy events from March 1994 to May 2001, using articles from the
Wall Street Journal to assess expectations. Swanson (2006) demonstrates that the private sector
has become better at forecasting interest rates and less surprised by Fed actions since the late
1980s. He attributes this improvement to increased Fed transparency, showing as a control
that GDP and inflation forecasts did not improve over the same period. These factors reduced
the measured effects of policy surprises on longer-term interest rates (Poole and Rasche, 2000;
Poole, Rasche, and Thornton, 2002). For example, compare the decline in coefficients in the
“Target” columns from the first to the second panel in Table 5.26

Multidimensionality: Two Monetary Policy Factors
The increasing predictability of Fed policy, as documented by Swanson (2006), might well
be partly due to the FOMC’s strategy of releasing statements regarding the economic outlook
86

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

Table 4
Changes in FOMC Communication Policy (1993-2007)
Date

Label

Nature of change

March 1993

Minutes

Merging of FOMC “Minutes of Actions” and “Policy Record” into one new
document labeled the “Minutes of the FOMC”

February 1994

Statement

First postmeeting statement with qualitative description of change in policy

August 1994

Rationale

Some reasoning behind the decision is added to the statement

July 1995

Federal funds target

Inclusion of actual numerical federal funds target in statement

May 1999

Bias

Inclusion of FOMC’s asymmetric policy directive in statement

January 2000

Balance of risks

Revised statement language discussing balance of risks toward growth or
inflation rather than bias for federal funds target

March 2002

FOMC vote

Inclusion of vote with name(s) of dissenters in statement

August 2003

Guidance

Statement language explicitly indicating the likely direction of rates over
extended period

January 2005

Earlier minutes

Minutes released three weeks after meeting

November 2007

Enhanced projections

More detailed, frequent, and extended projections

NOTE: This table is excerpted from Middeldorp (2011).

and likely future policy after its meetings. This suggests that the FOMC might be able to influence the whole yield curve—not just the very short end—by communicating its intentions to
markets. The reason for this is that long-term interest rates depend, in part, on the expected
path of future short-term interest rates, which the FOMC can influence.
Fleming and Piazzesi (2005) and Gürkaynak, Sack, and Swanson (2005) identify specific
cases where the response of long rates appears to be unrelated to the measured surprise to the
funds target. Specifically, Gürkaynak, Sack, and Swanson (2005) identify large responses in
2- to 5-year Treasury yields after the January 28, 2004, policy meeting, despite the fact that
there was almost no surprise in the federal funds target. The authors attribute these unusual
responses to changes in expectations shaped by the FOMC’s meeting statements rather than
the policy action itself. The independence of long rate changes and federal funds surprises
motivates Gürkaynak, Sack, and Swanson (2005) to develop a two-factor model of monetary
policy shocks (see Appendix C).27 A single factor fails to adequately describe yield curve reactions (up to one year ahead) to monetary policy shocks, but statistical tests cannot reject the
same null of two factors for yields up to a year. The first factor (target or timing) is closely
related to current-month federal funds futures surprises, while the second factor (level or path)
correlates strongly with 1-year eurodollar futures (i.e., 1-year-ahead policy expectations).
The second factor significantly increases the power of monetary policy actions to explain
medium- to long-term interest rate changes, with the largest improvements at the longest
maturities: The 10-year yields respond almost three times more to the path factor than the
target factor, and the R-squared rises from 8 percent to 74 percent (Gürkaynak, Sack, and
Swanson, 2005). The path factor is less important for equity returns, however: The coefficient
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

87

Fawley and Neely

The Evolution of Fed Policy
One needs to define a monetary policy instrument to study the effects of monetary policy surprises on
asset prices. The nature of Federal Reserve monetary policy has changed over time, however, from a
hybrid of interest rate targeting to hit ranges for monetary aggregates, to targeting for various categories
of bank reserves, and back to explicit interest rate targeting. This evolution of Fed policy can be interpreted as a series of moves toward greater transparency in objectives and procedures that have strongly
influenced the methods and assumptions of monetary policy researchers, as well as the impact of monetary policy on asset prices.
From 1970 to October 1979, the Federal Reserve targeted growth rates for monetary aggregates, primarily M1 but also M2 (Meulendyke, 1998). The Full Employment and Balanced Growth Act of 1978 (the
Humphrey-Hawkins Act) enshrined monetary targeting into law by requiring the Federal Reserve to set
targets for calendar years and to explain any deviations. To realize these money supply growth targets,
the FOMC chose a target for the federal funds rate and instructed the New York Fed Trading Desk to trade
appropriately to correct deviations of the funds rate from this target. Gradually, the Desk began to control the federal funds rate fairly closely (Meulendyke, 1998).
Unfortunately, Federal Reserve monetary policy in the 1970s failed to control inflation.1 Annual consumer price index inflation reached 12.2 percent per year in September 1979, eventually peaking at 14.8
percent per year in March 1980. Many analysts blamed federal funds rate targeting for producing too
much inertia in policy (Meulendyke, 1998). As a result, in October 1979, Chairman Volcker announced
that the FOMC would target nonborrowed reserves, rather than the funds rate, to achieve desired M1
growth.
In 1983, the lack of a stable relation between M1 growth and economic activity led the FOMC to change
procedures again, this time to targeting borrowed reserves (total reserves less nonborrowed reserves).2
In practice, analysts interpreted this procedure as a move back toward “soft” interest rate targeting
(Thornton, 1988). This procedure, however, depended on a stable function for bank borrowing from the
Fed. Banking problems—particularly those of Continental Illinois in 1984—soon led banks to become
very wary of borrowing from the Federal Reserve, lest investors, depositors, and/or regulators conclude
they were in financial trouble (Meulendyke, 1998). As with M1 targeting, the policy of targeting borrowed
reserves failed because the borrowing function was unstable; banks became less willing to borrow from
the Fed and borrowed reserves were not closely related to prices or economic activity (Thornton, 1988).
1 Researchers broadly agree that changing ideas on the objectives and scope of monetary policy were critical to the
development of the Great Inflation and the subsequent Great Disinflation, but they disagree about why the Fed failed
to act. DeLong (1997) believes that the Great Depression left the Federal Reserve with no mandate to control inflation
at the expense of unemployment. In contrast, Romer and Romer (2002) implicitly argue that the Fed used a fairly
sophisticated but deeply flawed model that claimed to offer an exploitable inflation-unemployment trade-off. Nelson
(2005a,b) and Nelson and Nikolov (2004) argue that “monetary neglect”—emphasis on nonmonetary factors in inflation—largely explains the Great Inflation not only in the United States but also in Canada, Australia, New Zealand, and
the United Kingdom.
2 Banks can borrow reserves directly from the Fed through the discount window to meet reserve requirements, avoid

overnight overdrafts, or meet seasonal funding needs. The demand for borrowed reserves theoretically reflects the
tightness of credit conditions in the nonborrowed reserves market.

88

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

The Evolution of Fed Policy (cont’d)
By the late 1980s, the FOMC was effectively targeting the federal funds rate, although it did not announce
changes in the federal funds target immediately after FOMC meetings. Instead, markets had to infer new
targets from the Desk’s trading patterns, which could take a day or two. In addition, during the late 1980s
and early 1990s, the FOMC’s standard practice was to change the funds target between scheduled meetings, often in response to incoming economic news.
In 1994, the FOMC changed its procedures in three ways. First, the FOMC began publicly announcing
changes to the target immediately after the meeting or decision. Second, the FOMC almost eliminated
intermeeting target changes, which had been common in prior years.3 From October 1988 to December
1993, the FOMC changed the federal funds target 32 times, but only 9 times at a scheduled meeting.
From 1994 to 2012, the FOMC has changed the target 60 times, 53 of which were at a scheduled meeting.4 Third, the FOMC began to issue statements on the state of the economy and monetary policy
immediately after meetings. The statements initially were short and qualitatively described the federal
funds target policy but became more detailed in August 1994 and included a numerical description of
the funds target by July 1995.
FOMC communication policy continued to evolve. In May 1999, the FOMC began to issue a statement of
its “bias,” or the direction in which its next move was most likely. But this approach was replaced by a “balance of risks” statement in January 2000 that switched the statement’s emphasis to the likely risks to
growth or inflation. From August 2003 to December 2005, the FOMC added implicit forward guidance
about the likely future course of the funds target. This practice began again in December 2008 in the
wake of the financial crisis.
Of course, the financial crisis in the summer and fall of 2008 also motivated the FOMC to institute many
temporary special programs to support credit markets and to announce the first of several asset purchase programs in November 2008.
3 The FOMC had been reducing the frequency of intermeeting target changes prior to February 1994. The seven inter-

meeting target changes since 1994 occurred on 4/18/1994, 10/15/1998, 1/3/2001, 4/18/2001, 9/17/2001, 1/22/2008,
and 10/8/2008. The 10/15/1998 target change came on the heels of the Long-Term Capital Management collapse,
which had been affected by the Russian default (8/17/1998); and the 9/17/2001 change followed the 9/11 terrorist
attacks.
4 FOMC transcripts reveal that Chairman Greenspan came to consider the practice of making decisions only at scheduled meetings to be a “very useful procedure.” The Chairman would sometimes hint at policy decisions between
meetings, however (FOMC, 1998). The FOMC has also changed the federal funds target much less frequently since
1994. The FOMC did not begin making target changes in multiples of 25 basis points until late 1989; previously,
changes in multiples smaller than 25 basis points were common (Poole, Rasche, and Thornton, 2002).

on the target factor is four times that of the path factor in a regression of the Standard and
Poor’s (S&P) 500 index returns on the two factors, and the R-squared improves little.
Table 5 shows results of an exercise similar to that in Gürkaynak, Sack, and Swanson
(2005), in which one regresses daily interest rate changes on target and path factors, plus a
constant (coefficient not shown). The top panel shows the results for all FOMC meetings; the
bottom panel shows the results with employment reports removed. The target coefficients in
the 1988-93 subsample are uniformly larger than those in the 1994-2007 subsample, particularly at longer horizons, and tend to be more statistically significant despite the shorter sample. At the same time, the coefficients on the path factor become larger and more statistically
significant in the second subsample. This pattern presumably occurs because the introducFederal Reserve Bank of St. Louis REVIEW

First Quarter 2014

89

Fawley and Neely

Table 5
Response of Interest Rates to Target and Path Shocks
1988-93
Target

1994-2007

Path

R

2

Target

1988-2007

Path

R

2

Target

Path

R2

All meeting observations
3-month

0.94
(14.9)

0.18
(3.3)

0.79

0.63
(12.3)

0.35
(7.7)

0.66

0.74
(20.1)

0.18
(8.5)

0.73

6-month

0.98
(14.5)

0.17
(2.8)

0.78

0.58
(16.1)

0.48
(15.1)

0.82

0.73
(23.4)

0.24
(13.4)

0.81

12-month

1.07
(22.8)

0.14
(3.4)

0.90

0.47
(14.7)

0.56
(19.9)

0.85

0.69
(27.2)

0.31
(20.8)

0.87

2-year

1.00
(22.4)

0.10
(2.4)

0.89

0.27
(8.3)

0.73
(25.5)

0.87

0.52
(20.8)

0.41
(28.2)

0.88

5-year

0.86
(13.5)

0.04
(0.7)

0.75

0.13
(3.6)

0.73
(22.7)

0.83

0.37
(12.5)

0.43
(25.2)

0.82

10-year

0.64
(9.6)

0.03
(0.5)

0.60

–0.01
–(0.4)

0.58
(17.4)

0.73

0.20
(6.4)

0.35
(19.8)

0.71

30-year

0.50
(7.6)

–0.01
–(0.1)

0.48

–0.08
–(2.1)

0.36
(10.3)

0.50

0.10
(3.1)

0.25
(13.6)

0.53

With meetings on days of employment reports omitted
3-month

0.78
(10.2)

0.19
(4.0)

0.69

0.67
(12.0)

0.27
(7.7)

0.65

0.73
(17.0)

0.15
(8.3)

0.68

6-month

0.83
(10.0)

0.17
(3.4)

0.67

0.61
(15.5)

0.36
(14.9)

0.81

0.69
(19.0)

0.20
(13.1)

0.76

12-month

0.91
(16.6)

0.17
(4.9)

0.85

0.49
(14.0)

0.43
(19.7)

0.84

0.62
(21.2)

0.26
(20.3)

0.84

2-year

0.88
(17.4)

0.18
(5.8)

0.86

0.26
(7.5)

0.55
(25.1)

0.86

0.43
(15.0)

0.34
(27.3)

0.85

5-year

0.71
(9.9)

0.17
(3.8)

0.67

0.11
(2.9)

0.55
(22.3)

0.82

0.26
(8.0)

0.35
(24.6)

0.80

10-year

0.51
(7.0)

0.14
(3.1)

0.52

–0.04
–(1.0)

0.43
(17.0)

0.73

0.10
(2.9)

0.28
(19.4)

0.70

30-year

0.38
(5.3)

0.10
(2.2)

0.38

–0.11
–(2.4)

0.27
(10.0)

0.50

0.00
(0.0)

0.20
(13.5)

0.52

NOTE: The table shows results of an exercise similar to that in Gürkaynak, Sack, and Swanson (2005), in which one
regresses daily interest rate changes on target and path factors, plus a constant (coefficient not shown). Columns 2
through 4 show results using all FOMC events for the October 1988–December 1993 period; columns 5 through 7 show
similar results for the January 1994–June 2007 period; and columns 8 through 10 show results for the whole sample.
The top panel shows the results for all FOMC meetings; the bottom panel shows the results with employment reports
removed. For the top (bottom) panel, the first subsample includes 65 (57) events and the second subsample 112 (111)
events, for a total of 177 (168) events. Shaded numbers denote significantly positive (blue) or negative (red) t-statistics.

90

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

Figure 1
Relationship Between Interest Rate Futures and Target and Path Surprises (1994-2007)
MP1

ED12

50

50

40

40

30

30

20

20

10

10

0

0

–10

–10

–20

–20

–30

–30

–40

–40
–50

–50
–50 –40 –30 –20 –10

0

10

Target Factor

20

30

40

50

–100 –80 –60 –40 –20

0

20

40

60

80 100

Path Factor

NOTE: MP1 denotes the Kuttner (2001) federal funds shock measured from near-month federal funds futures contracts and ED12 denotes the
change in 12-month-ahead eurodollar futures rates, using all FOMC events from 1994 through 2007. The 45-degree blue lines denote a one-toone relationship; the dashed red lines denote the least squares fit from regressing the y variable on the x variable and a coefficient. The slope is
determined by construction (see Appendix C).

tion of FOMC meeting statements in 1994 not only improved market confidence in its expectations of the path but also made target surprises less informative.
The bottom panel of Table 5 shows the same regression results but with days of employment reports removed. The target coefficients in the top panel are bigger (artificially inflated)
because omitting employment reports from the regression tends to increase the estimated
impact of monetary policy. In the bottom panel, the days of employment reports are completely
omitted from the sample so there is no omitted variables bias.28
Gürkaynak, Sack, and Swanson (2005) emphasize that the path factor is not an independent monetary policy tool, but rather exists because the FOMC can influence medium-term
rates by shaping expectations of the target’s path. The ability to influence medium-term rates
is important because much of monetary policy’s effect on the economy occurs through mediumterm rates and the incentives they provide for cyclical spending, such as business and residential
investment. In addition, the ability to influence longer-term rates would prove particularly
valuable near the zero lower nominal bound.
Although the FOMC does not directly choose the path surprise in the same way that it
chooses the federal funds target, the FOMC can strongly influence the path with its meeting
statement (Gürkaynak, Sack, and Swanson, 2005). This raises the question of whether such a
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

91

Fawley and Neely

Figure 2
Relationship Between Federal Funds Surprises and Eurodollar Futures
1988-1993

3-Month Eurodollar
20

3-Month Eurodollar
20

10

10

0

0

–10

–10

–20

–20

–30

–30

1994-2007

–40

–40
Slope = 0.83
–50
–50

–40

–30

–20

–10

0

10

20

Slope = 0.65
–50
–50

–40

–30

–20

1988-1993

12-Month Eurodollar

–10

0

10

20

Federal Funds Surprise

Federal Funds Surprise

1994-2007

12-Month Eurodollar
30

20
20
10
10
0

0

–10

–10

–20

–20

–30

–30

–40
Slope = 0.76
–50
–50

–40

–30

–20

–10

0

10

20

–40

Slope = 0.35

–50
–50

–40

–30

–20

Federal Funds Surprise
One-for-One

–10

0

10

20

30

Federal Funds Surprise
Less than One-for-One

More than One-for-One

Perverse

NOTE: The figure plots changes in 3-month-ahead eurodollar futures (top panel) and 12-month-ahead eurodollar futures (bottom panel) against
federal funds rate surprises measured using the methodology of Kuttner (2001). The left (right) column plots all FOMC events for the October
1988–December 1993 period (January 1994–June 2007 period). The solid blue line denotes a one-for-one change in the x and y variables; the
dashed red line denotes the OLS fitted line from regression of the y variable on the x variable plus a constant. The slope of this line is identified in
the lower-right corner of the figure.

statement reflects a commitment by the FOMC to pursue a certain policy that is inconsistent
with its expected reaction function or whether it simply reflects the FOMC’s view of the normal
policy response to likely economic conditions. Campbell et al. (2012) argue that the FOMC
has sufficient experience with communication to influence the economy by committing itself
to an unusual policy path. These authors term forward policy guidance that links policy to
the forecast of economic activity in the normal way as “Delphic” forward guidance and policy
that commits the FOMC to a particular policy as “Odyssean” forward guidance.
92

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

How should the path factor be interpreted? Hausman and Wongswan (2011) show that
the somewhat complex Gürkaynak, Sack, and Swanson (2005) transformation of the yield
curve to target and path shocks is nearly equivalent to simply using federal funds surprises
and changes in the 12-month-ahead eurodollar futures rate. Figure 1 shows this close relation
as scatterplots of the federal funds surprises and changes in the 12-month-ahead eurodollar
futures rate versus the target and path factors. The data points in the first panel have a 45degree slope by construction (see Appendix C).
The intimate link between changes to year-ahead policy expectations and path surprises
raises the question of how the relationship between current target surprises and near-term
expectations shifted in 1994. Figure 2 shows a scatterplot of federal funds surprises and corresponding changes in 3- and 12-month-ahead eurodollar rates, along with a 45-degree line
denoting one-for-one changes, for two different subsamples. Figures from 1988 to 1993 are
shown in the left panel, while figures from 1994 to 2007 are shown in the right panel. Squares
(circles) denote surprises that changed eurodollar rates more (less) than one for one. If a surprise to the federal funds target simply moves an expected policy action forward in time, one
would expect it to have a less than one-for-one effect on 3- to 12-month interest rates, such as
the 3-month eurodollar. In contrast, if a surprise increases expectations of further moves in
the same direction, one might expect a more than one-for-one effect on 3- to 12-month interest rates. The panels show that both 3-month and 12-month eurodollar rates responded less
strongly—the fitted lines have flatter slopes—to a given target surprise after the FOMC began
issuing statements in 1994. This suggests that FOMC target changes produced expectations
of less-persistent effects in the latter period.
How do the target and path factors affect asset prices? Tables 6 and 7 show the results of
regressing equity and foreign exchange returns and interest rate changes, respectively, on the
target and path factors for different samples.29 Consistent with Gürkaynak, Sack, and Swanson
(2005), the second panels in each table show that the path factor emerges following the FOMC’s
1994 decision to issue postmeeting statements. As in Table 5, Table 7 also documents a significant decline in the importance of target surprises to international interest rate changes in
the post-1994 sample as FOMC meeting statements made policy surprises relatively less informative. Note that, in both tables, the coefficients on the path factor in the 1988-93 sample are
very large—but often statistically insignificant—because of the lack of path factor variation
and lack of information in the path factor during that sample.
Statistical methods are not necessarily the only way to identify target and path factors,
however. Rosa (2011a,c) attempts to identify the second factor using a “narrative” approach
in which he first summarizes the tone of each FOMC meeting statement about the future direction of monetary policy and then approximates the unexpected components of the statement
by estimating forecasting regressions. Both the surprise component of policy decisions and
the statement’s tone significantly influence stock prices (Rosa, 2011a) and dollar exchange
rates (Rosa, 2011c). The surprise component of the statements accounts for most of monetary
policy’s effect on asset returns.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

93

Fawley and Neely

Table 6
Response of the S&P 500 and Exchange Rates to Target and Path Shocks
1988-93
Target

1994-2007

Path

R

2

Target

Path

1988-2007
R

2

Target

Path

R2

S&P500

–3.38
–(2.8)

1.53
(1.4)

0.13

–6.27
–(5.6)

–1.68
–(1.7)

0.24

–3.73
–(4.4)

–1.06
–(2.2)

0.12

CAD/USD

–0.03
–(0.1)

0.10
(0.3)

0.00

–0.57
–(1.3)

0.75
(1.9)

0.05

–0.30
–(1.0)

0.35
(2.1)

0.03

DEM-EUR/USD

1.48
(1.1)

1.60
(1.4)

0.05

–1.42
–(2.2)

1.94
(3.3)

0.13

0.01
(0.0)

1.12
(3.2)

0.06

GBP/USD

0.95
(0.8)

0.72
(0.7)

0.02

–0.58
–(1.1)

1.33
(2.8)

0.08

0.15
(0.3)

0.74
(2.4)

0.03

CHF/USD

1.91
(1.6)

1.48
(1.4)

0.07

–0.68
– (1.0)

2.15
(3.6)

0.12

0.56
(1.0)

1.16
(3.4)

0.07

JPY/USD

1.28
(1.4)

0.81
(0.8)

0.04

0.85
(1.2)

2.17
(3.4)

0.11

0.99
(1.8)

0.97
(3.1)

0.07

NOTE: The table shows results of an exercise similar to that in Gürkaynak, Sack, and Swanson (2005), in which one
regresses daily S&P 500 and exchange rate returns on target, path factors, and a constant (coefficient not shown).
Columns 2 through 4 show results using all FOMC events for the October 1988–December 1993 period; columns 5
through 7 show similar results for the January 1994–June 2007 period; and columns 8 through 10 show results for the
whole sample. The first subsample includes 65 events and the second subsample 112 events, for a total of 177 events.
Shaded numbers denote significantly positive (blue) or negative (red) t-statistics. CAD, Canadian dollar; CHF, Swiss
franc; DEM, German deutsche mark; EUR, euro; GBP, British pound; JPY, Japanese yen; USD, U.S. dollar.

International Effects of Monetary Shocks
The literature on the effects of monetary shocks spurred similar research into the effects
of monetary policy shocks on international asset prices. Craine and Martin (2008) extend
Rigobon and Sack’s (2004) identification-through-heteroskedasticity method to study spillovers
of monetary shocks between the United States and Australia, with a special emphasis on
accounting for nonmonetary shocks. The authors find that U.S. monetary policy shocks spill
over to Australian interest rate and equity markets, but Australian shocks do not seem to affect
U.S. financial markets.30 Nonmonetary surprises are more important than monetary surprises
for long maturity yields and in equities. Valente (2009) follows Craine and Martin’s work with
a study of the interaction between U.S. monetary policy and the yield curve in Hong Kong
and Singapore.
Two-factor methods have also been applied to international data. Hausman and Wongswan
(2011) regress interest rate changes, equity index returns, and exchange rate returns in 49 different countries on target and path factors for U.S. monetary policy surprises and obtain results
similar to those of Gürkaynak, Sack, and Swanson (2005): Short-term interest rates respond
to both path and target surprises, exchange rates and long-term rates respond primarily to
path surprises, and equity indexes respond primarily to target surprises. The insensitivity of
equity prices to path shocks is somewhat surprising as equity prices should reflect the present
value of profits into the infinite future and thus equities should be sensitive to the whole yield
94

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

Table 7
Response of International Long Yields to Target and Path Shocks
1988-93

1994-2007

Path

R

U.S. 10-year

0.64
(9.6)

0.03
(0.5)

0.60

–0.01
–(0.4)

0.58
(17.4)

0.73

0.20
(6.4)

0.35
(19.8)

0.71

Canadian10-year

0.37
(4.1)

0.04
(0.4)

0.23

0.04
(0.7)

0.50
(10.8)

0.54

0.15
(3.5)

0.26
(10.5)

0.43

0.05
(1.0)

0.21
(5.1)

0.21

0.11
(2.2)

0.12
(5.1)

0.20

German 10-year

Target

Path

1988-2007

Target

2

R

2

Target

Path

R2

French 10-year

0.15
(1.9)

–0.02
–(0.3)

0.06

–0.09
–(2.0)

0.03
(0.8)

0.04

0.00
(0.1)

0.04
(1.8)

0.02

U.K. 10-year

0.13
(1.5)

–0.03
–(0.3)

0.03

0.00
–(0.1)

0.14
(2.6)

0.06

0.07
(1.5)

0.06
(2.1)

0.04

Japanese 10-year

0.29
(3.9)

0.02
(0.4)

0.22

0.03
(0.6)

0.12
(2.8)

0.08

0.08
(2.1)

0.11
(4.8)

0.16

NOTE: The table shows results of an exercise similar to that in Hausman and Wongswan (2011), but using the Gürkaynak,
Sack, and Swanson (2005) two-factor methodology, in which one regresses daily international long yields on target,
path factors, and a constant (coefficient not shown). Columns 2 through 4 show results using all FOMC events for the
October 1988–December 1993 period; columns 5 through 7 show similar results for the January 1994–June 2007 period;
and columns 8 through 10 show results for the whole sample. The first subsample includes 65 events and the second
subsample 112 events, for a total of 177 events. Shaded numbers denote significantly positive (blue) or negative (red)
t-statistics.

curve. As in Tables 5 and 6, Table 7 shows a strong path factor effect on foreign long rates
after 1993.

CONCLUSION
The speed and flexibility with which monetary policy can be employed has made it the
primary policy for macroeconomic stabilization. The importance of this policy for inflation,
output, and employment has motivated researchers to explore its impact on the economy.
Because of the difficulty in directly discerning the impact of monetary policy on monthly and
quarterly macroeconomic variables, researchers have studied the effect of such policies on
quick-reacting asset prices as a first step in understanding the broader implications of monetary policy for macro variables. A key lesson from this literature is that researchers must
properly account for Federal Reserve procedures to draw the correct inference about the
impact of monetary surprises. Specifically, the increasing transparency of Fed procedures
and objectives, coupled with credibility won over several decades, has enabled the FOMC to
influence asset prices with statements rather than large and disruptive surprises in overnight
interest rates.
Early research on the effects of low-frequency changes in monetary or reserve aggregates
found no consistent effect of these variables on asset prices, despite the Fed’s description of
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

95

Fawley and Neely

its own objectives in terms of the former variables. But Cook and Hahn (1989) found strong
effects of federal funds target changes—the FOMC’s monetary policy instrument—in the
1970s. Kuttner (2001) established the importance of decomposing policy actions into expected
and unexpected components with data from federal funds futures markets, as asset prices
should respond only to the unexpected component. Other researchers (e.g., Poole, Rasche,
and Thornton, 2002, and Hamilton, 2008) advocated methods that are robust to the measurement error inherent in constructing monetary surprises.
FOMC procedural changes, particularly those in 1994-95, have influenced researchers’
methods. The FOMC greatly reduced the frequency of intermeeting target changes, announced
target changes as they were decided, and began to issue postmeeting statements to guide public
understanding of the economy and likely future monetary policy. These changes reduced the
necessity of accounting for other news that might affect asset prices and the simultaneity of
monetary policy and asset price changes (Rigobon and Sack, 2004). In addition, the resultant
increased Fed transparency has greatly improved the market’s ability to forecast and price in
monetary policy actions well before they happen (e.g., Poole, Rasche, and Thornton, 2002,
and Swanson, 2006).
An apparent disconnect of movements in longer-term interest rates from federal funds
surprises led researchers to realize they should account for the effect of FOMC policy statements
on year-ahead interest rates (Bernanke and Kuttner, 2005; Gürkaynak, Sack, and Swanson,
2005). In the post-1994 sample, these path surprises significantly affect exchange rates and
international interest rates across the yield curve; however, they do not affect U.S. or international equity prices. Gürkaynak, Sack, and Swanson (2005) argue that the influence of these
path surprises indicates that the Fed can credibly commit to a policy path.
Researchers have used such institutional knowledge to establish that monetary policy
surprises influence the prices of many asset classes, including fixed income, equity, and foreign
exchange. They have also investigated the impact of monetary policy surprises on equity
prices from different types of firms and industries and what that reveals about the relative
importance of the asset price and credit channels of monetary policy (Ehrmann and Fratzcher,
2004, 2009; Basistha and Kurov, 2008). Most recently, researchers have found considerable
international impact of monetary policy surprises (Craine and Martin, 2008; Hausman and
Wongswan, 2011). ■

96

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

APPENDIXES
Appendix A: Expectations of the Federal Funds Target from Futures Prices
The Chicago Board of Trade has sponsored trading in the 30-day federal fund futures
contract—an interest rate derivative—since October 3, 1988. Emmons, Lakdawala, and Neely
(2006) describe federal funds futures and options on those futures in some detail.
The federal funds futures market is a derivatives market whose final settlement price is
determined by the average federal funds rate over the contract month. Thus, the final settlement
price on the March 2013 contract is determined by the average federal funds rate during that
month. If the contract month has M trading days, the final settlement price (price settle,t ) will be
price settle ,t = 1 − ff settle ,t = 1 −

(A.1)

1
∑ Mj=1 ff jt ,
M

where ff settle,t is the average federal funds rate implied by the final settlement price.
On day i, the federal funds futures rate implied by the day i future price (ffit = 1 – Priceit )
for a given contract month, t, is equal to the expected final settlement price plus a risk premium:
ffit = Ei ff settle ,t + rpit = Ei

(A.2)

1
M
t
t
∑
j =1 ff j + rpi .
M

Piazzesi and Swanson (2008) show that implied federal funds rates from futures are modestly
biased predictors, likely to slightly overpredict the rate implied by final futures settlement by
3 to 6 basis points per month of forecast horizon, on average. The authors estimate the following
t
regression with heteroskedastic-consistent errors: ftn – rt+n = a (n) + et+n
, where ftn is the n-periodahead rate implied by the federal funds futures price in month t and rt+n is the actual (realized,
ex post) average funds rate in month t + n. For values of n = {1,2, 3, 4, and 5}, a (n) = {3.4, 7.4,
12.5, 19.2, 27.6, and 36.7}, respectively. Some analysts argue that such risk premia are too large
to be plausible (Carlson, Craig, and Melick, 2005). But because the risk premium changes very
slowly—at business cycle frequencies—daily changes in federal funds futures prices closely
approximate daily changes in the market’s expectation of the funds target (Piazzesi and
Swanson, 2008).
The New York Fed is able to keep the actual federal funds rate “close” to the federal funds
target desired by the FOMC. Therefore, the average federal funds rate over a month will be
very close to the average federal funds targets for that month:
M
M
1
1
ff jt ≈ ∑ ff jtarget ,t .
∑
j
=
1
M
M j =1

(A.3)

If there is an FOMC meeting within a month, on day d of the month, the average federal funds
target will be a weighted average of the (known) target prior to the FOMC meeting and the
new target chosen at the FOMC meeting:
(A.4)

d
1
1
1
M
t
ff jtarget ,t + ∑ Mj =d+1 ff jtarget ,t .
∑
∑
j =1 ff j ≈
M j =1
M
M

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

97

Fawley and Neely
target ,t
If we denote the known target going into the meeting as ff start
and the target chosen at the
target ,t
FOMC meeting, on day d of the month, as ffmeeting , then the average funds rate over the month
is equal to

1
d target ,t M − d target ,t
M
t
ff start +
ffmeeting
∑
j =1 ff j ≈
M
M
M

(A.5)

or, solving for the target to be chosen at the FOMC meeting:
target ,t
ffmeeting
≈

(A.6)

M 1
d target ,t 
M
t
ff start  .
 ∑ j =1 ff j −

M −d  M
M

Using the fact that the average expected funds rate for a target month is equal to the funds rate


1
implied by the current futures prices  ffit = Ei ∑ Mj =1 ff jt + rpit  , the expected target, as of


M
day i, at the FOMC meeting in month t can be written as follows:
target ,t
Ei ( ffmeeting
)≈

(A.7)

M  t
d target ,t 
t
ffstart  .
 ffi − rpi −

M −d 
M

If the FOMC meeting day is near the end of the contract month, then M–d will be small and
the calculation will be very sensitive to small errors in the data, perhaps caused by bid-ask
spreads. In this case, it is better to use the next month’s contract—when there will be no FOMC
meeting—to estimate the market’s expectation of the FOMC decision. For relatively short
forecast horizons, researchers commonly ignore the risk premium in equation (A.7), assuming that it will be very small.
Equation (A.7) can be used to decompose FOMC target changes into expected and surprise components. The expected change in the federal funds target is the expected target at
the meeting less the current (start of month) target, expressed as
target ,t
target ,t
target ,t
Ei ( ∆ffmeeting
) = Ei ( ffmeeting
) − ffstart

M  t
d target ,t 
target ,t
ffstart  − ff start
 ffi − rpit −


−
M d
M

M
d target ,t M − d target ,t 
t
t
≈
ffstart −
ff start 
 ffi − rpi −

M −d 
M
M
M  t
M target ,t 
t
≈
ffstart 
 ffi − rpi −

M
M −d 
M
target ,t
≈
( ffit − rpit − ff start
).
M −d
≈

(A.8)

(A.9)

target ,t
target ,t
target ,t
≈
Ei ( ∆ffmeeting
) = Ei ( ffmeeting
) − ffstart

(

M
target ,t
ffit − rpit − ff start
(
).
M −d

)

target , t
The surprise component ffmeeting is the actual change less the change that was expected just
before the meeting:

98

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

(A.10)

target ,t
target ,t
target ,t
∆ ffmeeting
= ∆ffmeeting
− Ei ( ∆ffmeeting
).

Kuttner (2001) recognizes that a time-varying risk premium, rpit, potentially contaminates
the federal funds shocks calculated from equations (A.9) and (A.10), and that the multiplier
M
amplifies this measurement error, particularly near the end of the month. Therefore,
M −d
he proposes a more robust estimate of federal funds shocks using changes in futures prices
around FOMC events. Specifically, he observes that the futures price in equation (A.7) can be
solved for to obtain the following:
(A.11)

ffit =

M −d
d target ,t
target ,t
+
Ei ( ffmeeting
ffstart + rpit ,
)
M
M

which implies that a daily difference in the futures price should estimate the sum of the daily
change in the target for month t plus the daily difference in the risk premium:
(A.12)

ffi t − ffit−1 =

M −d
target ,t
∆Ei ( ffmeeting
) + rpit − rpit−1 .
M

If the day-to-day change in the risk premium is small, then he can estimate the policy surprise—that is, the change in the expected target on the day of the meeting—as follows:
(A.13)

target ,t
∆ ffmeeting
=

M
ffit − ffi−t 1 ) .
(
M −d

Changes in futures prices around FOMC events reflect unanticipated changes to the funds rate.
While surprises calculated from equation (A.13) may still suffer from measurement error,
they remain robust under much weaker assumptions about the sources of such error.

Appendix B: Simultaneity and Omitted Bias
The monetary policy shock literature typically seeks to determine the effect of monetary
policy on some function of asset prices. Researchers have been primarily concerned with the
effect of some function of monetary policy (Dmt) on some function of asset prices (Dpt), which
could include changes in bond yields, stock prices, or exchange rates, but the relation could
also include other news, such as macroeconomic releases (newst ), and/or other variables:
(B.1)

∆pt = a1∆mt + a2newst + e p ,t .

Although we would like to estimate the parameter, a1, if monetary policy reacts to asset
price changes within the observation interval for the data (e.g., within the day for daily data),
then b1 will be nonzero in equation (B.2) and this will affect the estimation of equation (B.1):
(B.2)

∆mt = b1∆pt + b2newst + em ,t .

For simplicity of notation, we assume that all variables are known to be mean zero and
that the structural errors, ep,t and em,t , are uncorrelated over time and contemporaneously.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

99

Fawley and Neely

They have diagonal covariance matrix V and are also uncorrelated with exogenous news
shocks, whose variance is normalized to equal 1. Equations (B.1) and (B.2) can be written in
terms of the matrixes, M, P, N, Em, and Ep, which represent the T × 1 vectors whose tth elements
are Dmt , Dpt , newst , em,t , and ep,t , respectively:
(B.3)

P = Ma1 + Na2 + E p

(B.4)

M = Pb1 + Nb2 + E m .

Or, this can be written in matrix form:
 1 −b 
1
 P M  
 = N  a2 b2 +  E p Em  ,

 

1
a
−

 1

(B.5)

and equation (B.5) can be rewritten as follows:
 P M  = NDC −1 +  E E C −1 ,


 p m

(B.6)

 1 −b 
1  1 b1 
1
 , and so C −1 =
where D =  a2 b2 , C = 
. The following matrixes


1 − a1b1  a1 1 
 −a1 1 
are used to define the unconditional covariance matrix of the endogenous variables:
(B.7)


2
2
 1  
(a2 + a1b2 )
(b1a2 + b2 ) (a2 + a1b2 ) 
−1
′
C D′N ′NDC = N ′N 


2
 1 − a1b1   (b1a2 + b2 )(a2b1 + b2 )
(b1a2 + b2 )


−1

C ′VC −1 =
−1

(B.8)

 V + a 2V
1
 11 1 22
2
(1 − a1b2 )  b1V11 + a1V22


b1V11 + a1V22 
,
b12V11 + V22 

where V11 and V22 are the variances of the structural errors of the price and money equations.
The OLS estimates of the parameters in equation (B.3) are as follows:
(B.9)

−1
−1
 â  
  M ′P   M ′M M ′N   M ′Ma1 + M ′Na2 + M ′E p
M
M
M
N
′
′
1

=
 
 
=
 â2   M ′N N ′N   N ′P   M ′N N ′N   N ′Ma1 + N ′Na2 + N ′E p



.



Taking the inverse, multiplying through, and eliminating terms that are identically zero
yields the following expression for the OLS estimates:
 ( N ′NM ′M − M ′NN ′M ) a + ( N ′NM ′− M ′NN ′) E
 â 
1
1
P
1

=
(B.10) 
 â2  M ′MN ′N − ( M ′N ) 2  ( M ′MN ′N − M ′NM ′N ) a2 + ( M ′MN ′− M ′NM ′) EP
100

First Quarter 2014


.



Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

The behavior of the coefficients in large samples is given by the probability limits (plims)
of the expressions in (B.10), if the limits exist:
(B.11)

 â   a
1
= 1
plim 
T →∞  â2   a2
 



 N ′NM ′Ep

1

+ plim 
2
 T →∞  M ′MN ′N − ( M ′N )  − M ′NM ′E p


 .



Fortunately, the rules of probability limits allow us to pass the plim function through
products and quotients. That is, the plim of a product is the product of its plims, and the plim
of a quotient is the quotient of the plims. This allows us to determine the plims of the coefficient
estimators in equation (B.11). We use the following limits:
(B.12)

plim
T →∞

(B.13)

plim
T →∞

(B.14)

1
N ′N = 1
T

1
b 2V + V
2
M ′MN ′N − ( M ′N ) = 1 11 222
2
T
(1 − a1b1 )

(

plimT →∞

)

1
V b
M ′E p = 11 1
T
1 − a1b1

(B.15)

 (b a + b ) 2 + b 2V + V 
1 11
22
plim [ M ′M ] T =  1 2 2

2
T →∞


(1 − a1b1 )

(B.16)

1
 a b +b
plim  M ′N  = 2 1 2 .
 1 − a1b1
T →∞  T

Equation (B.14) uses the assumption that the structural errors are uncorrelated with the news
1

shocks—that is, plim  N ′E p  = 0. Using these probability limits in (B.11), we obtain the

T →∞  T
following:


V11b1



2
 â   a 
1 − a1b1
  a1 

1
a
b
−
b1
(
)
1
1
1
1
 (1 − a1b1 )


+
=
(B.17) plim   =  + 2

 
2

 

T →∞  â2   a2  (b1 V11 + V22 )

 −  a2b1 + b2   V11b1    a2  (b1 + V22 V11 )  − (a2b1 + b2 )

 
  1 − a1b1  1 − a1b1  




.



Note that the OLS estimation of an equation with an endogenous regressor (Dmt)—that
is, b1 ≠ 0—will generally produce inconsistent estimates of both regressors.31 In this case, the
OLS estimate, â1, is consistent if the variance of the structural asset price shocks equals zero
(V11 = 0) or if Dpt has no contemporaneous effect on Dmt; that is, b1 = 0, in which case Dpt
would be considered predetermined. Otherwise, if the ratio of variance of monetary policy
shocks to the variance of price shocks (V22 /V11) gets arbitrarily large, then â1 will converge in
probability to a1.
Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

101

Fawley and Neely

One way to drive the variance ratio (V22 /V11) of the structural shocks arbitrarily large is
to take an arbitrarily short interval around the monetary policy event—an FOMC announcement, for example. In such an interval, the variance of monetary shocks will be very large and
the variance of prices can be made arbitrarily small. Also, in such a short interval, the monetary authorities are very unlikely to be reacting to any changes in asset prices, which means
that Dpt has no contemporaneous effect on Dmt ; that is, b1 = 0.
Note too that if the news variables in equation (B.4) are omitted from the estimated regression, there is a different problem: omitted variables bias. In this case, the OLS estimate of the
coefficient on the monetary shock, a1, is as follows:
(B.18)

−1

−1

â1 = [ M ′M ] M ′P = [ M ′M ]  M ′Ma1 + M ′Na2 + M ′E p  .

Again, we can use the plims that exist for

(B.19)

plim â1 = a1 +
T →∞

1
1
1
M ′M ,
M ′N , and M ′E p to show that
T
T
T

(1 − a1b1 ) ((a2b1 + b2 ) a2 + V11b1 )
.
2
(b1a2 + b2 ) + (b12V11 + V22 )

If a2 = 0, there is no omitted variable, but simultaneity (b1 ≠ 0) still creates inconsistency:
(B.20)

plim â1 = a1 +
T →∞

(1 − a1b1 ) b1
b V11 + b12 +V22 V11
2
2

.

If b1 = 0, then the monetary policy shock is predetermined; there is no simultaneity. In
this case, if either b2 or a2 were also zero—that is, news does not contemporaneously affect
both asset prices and monetary—then equation (B.19) shows that â1 would be consistent.
Otherwise—still assuming that b1 = 0—then as the effect of news on monetary policy (b2) or
the variance of monetary policy shocks (V22) gets arbitrarily large, the estimate approaches
consistency (see below):
(B.21)

plim â1 = a1 +
T →∞

b2 a2
a2
.
= a1 +
b + V22
b2 + V22 b2
2
2

Appendix C: Target and Path Surprises
Gürkaynak, Sack, and Swanson (2005) identify monetary policy shocks as the unobserved monetary factors F from
(C.1)

X = FΛ + η ,

where h are white-noise disturbances and L contains the loadings of F on X. Gürkaynak, Sack,
and Swanson (2005) include five variables in X: (i) the surprise to the federal funds target
measured from current-month federal funds futures; (ii) the surprise change in expectations
of the federal funds target two FOMC meetings ahead, measured from the appropriate federal
102

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely

Table C1
Response of Federal Funds Surprises and 12-Month Eurodollar Futures to the
Normalized Target and Path Factors
1988-93

1994-2007

Target

Path

R

MP1

1.12
(15.3)

0.00
(0.0)

0.79

ED12

1.22
(25.0)

(0.02)
(0.5)

0.91

2

Target

1988-2007

Path

R

0.93
(40.5)

0.00
(0.0)

0.94

0.31
(16.6)

1.01
(59.9)

0.97

2

Target

Path

R2

1.00
(41.6)

0.00
(0.0)

0.91

0.58
(41.0)

0.58
(70.9)

0.97

NOTE: The table shows the result of regressing the federal funds surprise (MP1) and the daily change in 12-month
eurodollar futures (ED12) on the target and path factors used as regressors in Tables 5 through 7. The full sample results
(1988-2007) show the effect of the normalization procedure, while the subsample results (1988-93 and 1994-2007)
show the effect of sampling variation.

funds future contract; and the price change in (iii) 6-month, (iv) 9-month, and (v) 12-month
eurodollar futures contracts.
Gürkaynak, Sack, and Swanson (2005) structurally identify the two monetary policy factors, F, as a linear transformation of the first two principal components (Z) of X. Specifically,
(C.2)

F = ZU ,

where Z is the first two principal components of X, and U is a 2 × 2 matrix whose elements
are identified by imposing the following restrictions: (i) the columns of U have unit length,
(ii) the columns of F are orthogonal, and (iii) F2 , the second column of F, does not influence
the current federal funds shock. The last restriction, which implies two equations, provides
the structural interpretation of F1 and F2 as the target and path surprise, respectively. In other
words, F1 contains all information from the first two principal components that explains the
current federal funds surprise, and F2 contains all residual information.
Finally, Gürkaynak, Sack, and Swanson (2005) rescale F1 and F2 to provide comparability
between coefficients and against earlier studies. Specifically, they scale F1 to move one for one
with the current federal funds surprise (MP1) and F2 to have the same magnitude effect on
12-month eurodollar futures (ED12) as F1. Table C1 illustrates the relationship between MP1
and ED12 and the normalized target and path factors used as regressors in Tables 5 through 7.
Please note the effect of sampling variation when comparing the size of subsample coefficients
in Tables 5 through 7 with previous studies.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

103

Fawley and Neely

NOTES
1

The Romer and Romer (1989, 1994) narrative approach is another method to identify monetary policy shocks. The
terms “shocks” and “surprises” are closely related but not quite synonymous. A “shock” denotes the unexpected
component of a variable in a statistical model. Some researchers reserve the term for the unexpected component
of a variable in a structural statistical model. In contrast, a “surprise” denotes any unexpected occurrence, particularly an event that markets did not expect. Most of the research discussed in this article uses statistical methods in
which expectations are derived from futures market prices, in which case there is no distinction between shocks
and surprises.

2

This paper focuses on the literature studying the reactions to conventional monetary policy shocks, not on reactions to the unconventional policies, including long-term security purchases, which debuted in 2008.

3

The federal funds market is an overnight market in which depository institutions lend reserve balances to other
depository institutions.

4

In a regression, the omission of relevant explanatory variables will generally bias the estimated coefficients on
included variables, unless the omitted explanatory variables happen to be uncorrelated with the included variables.

5

Friedman (1968) argued that “interest rates are such a misleading indicator of whether monetary policy is ‘tight’ or
‘easy.’” He viewed nominal interest as reflecting the stance of past monetary policy via inflation expectations.

6

Early studies measured the unexpected component of money growth as the residuals from a univariate or multivariate autoregressive model of money growth (Barro, 1978) or as the difference between announced money
growth and Money Market Services (MMS) median survey forecasts (Cornell, 1982, 1983; Hardouvelis, 1984).

7

The Federal Reserve did not begin announcing a numerical federal funds target in its statement until July 1995
(Middeldorp, 2011), but it has published an official federal funds target dating back to 1982
(http://research.stlouisfed.org/fred2/series/DFEDTAR?cid=118).
Although a monetary policy instrument is not the same as the monetary policy stance—for example, “high” nominal interest rates do not necessarily indicate tight policy if inflation expectations are sufficiently high—a discrete
change in the instrument typically implies a discrete change in the stance as price levels and inflation expectations typically do not “jump” to offset a change in the instrument.

8

The end of Cook and Hahn’s (1989) sample coincides with Chairman Volcker’s decision to deemphasize the role of
the federal funds rate in setting monetary policy. Cook and Hahn (1989) document the market’s adeptness at
interpreting pre-1994, unannounced policy changes from open market operations. Cook and Hahn (1988) find
that when discount rate changes signaled federal funds changes, they also moved interest rates. This result held
in their 1973-79 and 1979-85 subsamples.

9

Roley and Sellon (1995) do find some evidence that 30-year Treasury yields anticipate future policy.

10 Fatum and Scholnick (2008) confirm this finding and argue that failing to correctly model the unexpected com-

ponent of monetary policy news leads to underestimating the effects of such news. They also confirm that systematic reactions occur rapidly, within the day of the announcement.
11 Poole, Rasche, and Thornton (2002) present evidence that markets were commonly able to predict policy actions

at least two weeks ahead after 1994.
12 Surprises measured with futures data also contain some measurement error from bid-ask spreads and risk pre-

mia, but probably much less than other methods.
13 Some researchers have used longer-term interest rates to identify shocks on the grounds that they better meas-

ure surprising actions by the Fed while minimizing measurement error when only the timing of the action is a surprise. For example, if the Fed surprises markets by lowering interest rates one meeting earlier than expected, then
near-month federal funds futures will measure a large surprise, while 3-month eurodollar futures might remain
unchanged. Cochrane and Piazzesi (2002) measure policy surprises from 1-month eurodollar deposit rates;
Ellingsen and Soderstrom (2004) use the 3-month Treasury bill rate; Rigobon and Sack (2004) use the 3-month
eurodollar futures rate; and Bomfim (2003), Poole and Rasche (2000), and Poole, Rasche, and Thornton (2002) use
month-ahead federal funds futures. Bernanke and Kuttner (2005) check the robustness of their results to those
obtained with 3-month eurodollar futures.
104

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely
14 Rigobon and Sack (2008) argue that correcting for the noise in macroeconomic announcements significantly

increases the implied information content of those announcements.
15 Piazzesi and Swanson (2008) argue that risk premia on federal funds futures are small and fairly stable at a daily

frequency (also see Hamilton, 2009).
16 Hamilton’s (2008) methodology nests Kuttner’s (2001) specification when the effective federal funds rate always

equals the target and the dates of policy actions are known.
17 Gürkaynak, Sack, and Swanson (2005) note a number of instances in which major news was released on days of

FOMC meetings or target changes.
18 Thornton (2001a) describes the shift away from unilateral discretion for the chairman to adjust the funds target in

favor of consultation with the FOMC before making any changes.
19 In this context, “identification” means that the assumed model permits one to sort out the two-way causality in

asset price changes and FOMC actions. That is, the assumed data-generating process in equations (3) and (4) must
permit consistent estimation of a1 and b1 if the contemporaneous effects are to be considered identified.
Gürkaynak, Sack, and Swanson (2005) find that policy surprises constructed from federal funds futures in 30minute and 1-day windows around policy events are nearly identical; the only notable deviations occur on pre1994 event days that coincide with the release of the employment report.
20 The results are somewhat sensitive to the inclusion of the semiannual report to Congress.
21 Narrow event windows might not capture the full impact on asset prices. Fleming and Piazzesi (2005) report that,

with the exception of intermeeting moves, asset prices respond very quickly to FOMC announcements.
22 In the context of borrowing and lending, adverse selection is the tendency of individuals and firms with bad credit

to be more likely to seek loans from banks. Moral hazard is the tendency of borrowers to engage in risky activities
that will make it less likely they will repay their loans. Both adverse selection and moral hazard are problems
because of the existence of asymmetric information, which means that borrowers know things about their ability
to repay that lenders do not.
23 Bernanke and Kuttner (2005) follow Patelis (1997) in using the Campbell and Ammer (1993) methodology to

decompose stock returns into changes in the discount factor (interest rates), expected dividends, or expected
excess returns.
24 Ammer, Vega, and Wongswan

(2010) also consider the impact of path shocks described by Gürkaynak, Sack, and
Swanson (2005). But, consistent with Gürkaynak, Sack, and Swanson’s (2005) results, path shocks have little effect
on equity, so Ammer, Vega, and Wongswan (2010) report only the results for the target shocks.

25 If the FOMC follows a “policy rule” that links policy to economic conditions (e.g., a Taylor rule), then a forecast of

economic conditions would imply a policy forecast and vice versa. Faust, Swanson, and Wright (2004) find little
evidence that Federal Reserve policy surprises can be used to improve forecasts of statistical releases, which suggests that the Federal Reserve does not necessarily have superior information about the state of the economy.
26 Kool and Thornton (2012), however, argue for a more skeptical view. Their study of forward guidance in New

Zealand, Norway, Sweden, and the United States finds limited evidence that forward guidance improves the private sector’s ability to forecast monetary policy.
27 Gürkaynak, Sack, and Swanson’s (2005) two-factor model is arguably very similar to Bernanke and Kuttner’s (2005)

use of level and timing shocks in their study of equity reactions. In fact, the two sets of explanatory variables span
the same space.
28 The authors thank Brian Swanson for suggesting this exercise and interpretation.
29 The negative “Target” coefficients in Table 6, which would imply a perverse exchange rate response to interest

rates, are the product of the use of noisy daily data. Use of intraday exchange rate data produces positive coefficients, as expected.
30 Neely (2013) characterizes the impact of the Federal Reserve’s unconventional policies on international bond

yields and exchange rates. Bauer and Neely (2013) investigate the channels through which such effects occur.
31 An estimator is consistent if it converges in probability to the parameter as the sample size increases.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

105

Fawley and Neely

REFERENCES
Ammer, John; Vega, Clara and Wongswan, Jon. “International Transmission of U.S. Monetary Policy Shocks:
Evidence from Stock Prices.” Journal of Money, Credit, and Banking, September 2010, 42(Suppl. s1), pp. 179-98.
Andersen, Torben G.; Bollerslev, Tim; Diebold, Francis X. and Vega, Clara. “Micro Effects of Macro Announcements:
Real-Time Price Discovery in Foreign Exchange.” American Economic Review, March 2003, 93(1), pp. 38-62.
Barro, Robert J. “Unanticipated Money, Output, and the Price Level.” Journal of Political Economy, August 1978, 86,
pp. 549-80.
Basistha, Arabinda and Kurov, Alexander. “Macroeconomic Cycles and the Stock Market’s Reaction to Monetary
Policy.” Journal of Banking and Finance, December 2008, 32(12), pp. 2606-16.
Bauer, Michael D. and Neely, Christopher J. “International Channels of the Fed’s Unconventional Monetary Policy.”
Federal Reserve Bank of St. Louis Working Paper No. 2012-028D, December 12, 2013;
http://research.stlouisfed.org/wp/2012/2012-028.pdf.
Bernanke, Ben S. and Blinder, Alan S. “The Federal Funds Rate and the Channels of Monetary Transmission.”
American Economic Review, September 1992, 82(4), pp. 901-21.
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.
Bernanke, Ben S. and Mihov, Ilian. “Measuring Monetary Policy.” Quarterly Journal of Econometrics, August 1998,
113(3), pp. 869-902.
Bomfim, Antulio N. “Pre-Announcement Effects, News Effects, and Volatility: Monetary Policy and the Stock
Market.” Journal of Banking and Finance, January 2003, 27(1), pp. 133-51.
Campbell, Jeffrey R.; Evans, Charles L.; Fisher, Jonas D.M. and Justiniano, Alejandro. “Macroeconomic Effects of
Federal Reserve Forward Guidance.” Brookings Papers on Economic Activity, Spring 2012, pp. 1-80.
Campbell, John Y. and Ammer, John. “What Moves the Stock and Bond Markets? A Variance Decomposition for
Long‐Term Asset Returns.” Journal of Finance, March 1993, 48(1), pp. 3-37.
Campbell, John Y. and Shiller, Robert J. “The Dividend-Price Ratio and Expectations of Future Dividends and
Discount Factors.” Review of Financial Studies, 1988, 1(3), pp. 195-228.
Carlson, John B.; Craig, Ben R. and Melick, William R. “Recovering Market Expectations of FOMC Rate Changes with
Options on Federal Funds Futures.” Journal of Futures Markets, December 2005, 25(12), pp. 1203-42.
Christiano, Lawrence J.; Eichenbaum, Martin and Evans, Charles L. “Monetary Policy Shocks: What Have We Learned
and to What End?” in John B. Taylor and Michael J. Woodford, eds., Handbook of Macroeconomics. Volume 1A.
Amsterdam: Elsevier B.V., 1999, pp. 65-148.
Cochrane, John H. and Piazzesi, Monika. “The Fed and Interest Rates: A High-Frequency Identification.” American
Economic Review, May 2002, 92(2), pp. 90-95.
Cook, Timothy and Hahn, Thomas. “The Information Content of Discount Rate Announcements and Their Effect on
Market Interest Rates.” Journal of Money, Credit, and Banking, May 1988, 20(2), pp. 167-80.
Cook, Timothy and Hahn, Thomas. “The Effect of Changes in the Federal Funds Rate Target on Market Interest
Rates in the 1970s.” Journal of Monetary Economics, November 1989, 24(3), pp. 331-51.
Cornell, Bradford. “Money Supply Announcements, Interest Rates, and Foreign Exchange.” Journal of International
Money and Finance, August 1982, 1, pp. 201-08.
Cornell, Bradford. “The Money Supply Announcements Puzzle: Review and Interpretation.” American Economic
Review, September 1983, 73(4), pp. 644-57.
Craine, Roger and Martin, Vance L. “International Monetary Policy Surprise Spillovers.” Journal of International
Economics, May 2008, 75(1), pp. 180-96.
DeLong, J.B. “America’s Peacetime Inflation: The 1970s,” in Christina D. Romer and David H. Romer, (eds.), Motivation
and Strategy. NBER Studies in Business Cycles. Volume 30. Chicago: University of Chicago Press, 1997.
106

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely
Edelberg, Wendy and Marshall, David. “Monetary Policy Shocks and Long-Term Interest Rates.” Federal Reserve
Bank of Chicago Economic Perspectives, March 1996, 20(2), pp. 2-17;
http://www.chicagofed.org/digital_assets/publications/economic_perspectives/1996/epmar96a.pdf.
Ehrmann, Michael and Fratscher, Marcel. “Taking Stock: Monetary Policy Transmission to Equity Markets.” Journal of
Money, Credit, and Banking, August 2004, 36(4), pp. 719-37.
Ehrmann, Michael and Fratzscher, Marcel. “Global Financial Transmission of Monetary Policy Shocks.” Oxford Bulletin
of Economics and Statistics, December 2009, 71(6), pp. 739-59.
Ellingsen, Tore and Soderstrom, Ulf. “Monetary Policy and the Bond Market.” IGIER Working paper, Innocenzo
Gasparini Institute for Economic Research, 2004.
Emmons, William R.; Lakdawala, Aeimit K. and Neely, Christopher J. “What Are the Odds? Option-Based Forecasts of
FOMC Target Changes.” Federal Reserve Bank of St. Louis Review, November/December 2006, 88(6), pp. 543-61;
http://research.stlouisfed.org/publications/review/06/11/Emmons.pdf.
Evans, Charles L. and Marshall, David A. “Monetary Policy and the Term Structure of Nominal Interest Rates:
Evidence and Theory.” Carnegie-Rochester Conference Series on Public Policy, December 1998, 49(1), pp. 53-111.
Fama, Eugene F. “Efficient Capital Markets: A Review of Theory and Empirical Work.” Journal of Finance, May 1970,
25(2), pp. 383-417.
Fatum, Rasmus, and Scholnick, Barry. “Do Exchange Rates Respond to Day-to-Day Changes in Monetary Policy
Expectations When No Monetary Policy Changes Occur?” Journal of Money, Credit, and Banking, September 2006,
38(6), pp. 1641-57.
Fatum, Rasmus and Scholnick, Barry. “Monetary Policy News and Exchange Rate Responses: Do Only Surprises
Matter?” Journal of Banking and Finance, June 2008, 32(6), pp. 1076-86.
Faust, Jon; Rogers, John H.; Wang, Shing-Yi B. and Wright, Jonathan H. “The High-Frequency Response of Exchange
Rates and Interest Rates to Macroeconomic Announcements.” Journal of Monetary Economics, 2007, 54,
pp. 1051-68.
Faust, Jon; Swanson, Eric T. and Wright, Jonathan H. “Do Federal Reserve Policy Surprises Reveal Superior
Information about the Economy?” Contributions in Macroeconomics, October 2004, 4(1), pp. 1-31.
Federal Open Market Committee. “Press Release.” February 4, 1994a;
http://www.federalreserve.gov/fomc/19940204default.htm.
Federal Open Market Committee. “Press Release.” August 16, 1994b;
http://www.federalreserve.gov/fomc/19940816default.htm.
Federal Open Market Committee. “Conference Call.” September 21, 1998;
http://www.federalreserve.gov/monetarypolicy/files/FOMC19980921confcall.pdf.
Fleming, Michael J. and Piazzesi, Monika. “Monetary Policy Tick-by-Tick.” Unpublished manuscript, Federal Reserve
Bank of New York, August 2005.
Friedman, Benjamin M. and Kuttner, Kenneth N. “Implementation of Monetary Policy: How Do Central Banks Set
Interest Rates?” in Friedman, Benjamin M. and Woodford, Michael (eds.)., Handbook of Monetary Economics.
Volume 3B. Amsterdam: North-Holland, 2011, pp. 1345-438.
Friedman, Milton. “The Role of Monetary Policy.” American Economic Review, March 1968, 58(1), pp. 1-17.
Gürkaynak, Refet S.; Sack, Brian P. and Swanson, Eric T. “Do Actions Speak Louder Than Words? The Response of
Asset Prices to Monetary Policy Actions and Statements.” International Journal of Central Banking, May 2005, 1(1),
pp. 55-93.
Gürkaynak, Refet S.; Sack, Brian P. and Swanson, Eric T. “Market-Based Measures of Monetary Policy Expectations.”
Journal of Business and Economic Statistics, April 2007, 25(2), pp. 201-12.
Hamilton, James D. “Assessing Monetary Policy Effects Using Daily Federal Funds Futures Contracts.” Federal
Reserve Bank of St. Louis Review, July/August 2008, 90(4), pp. 377-93;
http://research.stlouisfed.org/publications/review/08/07/Hamilton.pdf.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

107

Fawley and Neely
Hamilton, James D. “Daily Changes in Fed Funds Futures Prices.” Journal of Money, Credit, and Banking, June 2009,
41(4), pp. 567-82.
Hardouvelis, Gikas A. “Market Perceptions of Federal Reserve Policy and the Weekly Monetary Announcements.”
Journal of Monetary Economics, September 1984, 14(2), pp. 225-40.
Hausman, Joshua and Wongswan, Jon. “Global Asset Prices and FOMC Announcements.” Journal of International
Money and Finance, April 2011, 30, pp. 547-71.
Kool, Clemens J.M. and Thornton, Daniel L. “How Effective Is Central Bank Forward Guidance?” Federal Reserve
Bank of St. Louis Working Paper No. 2012-063A, December 2012;
http://research.stlouisfed.org/wp/2012/2012-063.pdf.
Kuttner, Kenneth N. “Monetary Policy Surprises and Interest Rates: Evidence from the Fed Funds Futures Market.”
Journal of Monetary Economics, June 2001, 47(3), pp. 523-44.
Litterman, Robert B. and Weiss, Laurence. “Money, Real Interest Rates, and Output: A Reinterpretation of Postwar
U.S. Data.” Econometrica, January 1985, 53(1), pp. 129-56.
McCallum, Bennett T. “A Reconsideration of Sims’ Evidence Concerning Monetarism.” Economic Letters, 1983,
13(2-3), pp. 167-71.
Meulendyke, Ann-Marie. U.S. Monetary Policy and Financial Markets. Federal Reserve Bank of New York, 1998;
http://research.stlouisfed.org/aggreg/meulendyke.pdf.
Middeldorp, Menno. “FOMC Communication Policy and the Accuracy of Fed Funds Futures.” Federal Reserve Bank
of New York Staff Report No. 491, April 2011; http://www.newyorkfed.org/research/staff_reports/sr491.pdf.
Mishkin, Frederic S. “Symposium on the Monetary Transmission Mechanism.” Journal of Economic Perspectives, Fall
1995, 9(4), pp. 3-10.
Neely, Christopher J. “Unconventional Monetary Policy Had Large International Effects.” Federal Reserve Bank of
St. Louis Working Paper No. 2010-018E, July 2010, updated August 2013;
http://research.stlouisfed.org/wp/2010/2010-018.pdf.
Nelson, Edward. “The Great Inflation of the Seventies: What Really Happened?” B.E. Journal of Macroeconomics:
Advances in Macroeconomics, July 2005a, 5(1), pp. 1-50.
Nelson, Edward. “Monetary Policy Neglect and the Great Inflation in Canada, Australia, and New Zealand.”
International Journal of Central Banking, May 2005b, 1(1), pp. 133-79.
Nelson, Edward and Nikolov, Kalin. “Monetary Policy and Stagflation in the U.K.” Journal of Money, Credit, and
Banking, June 2004, 36(3), pp. 293-318.
Patelis, Alex D. “Stock Return Predictability and the Role of Monetary Policy.” Journal of Finance, December 1997,
52(5), pp. 1951-72.
Piazzesi, Monika and Swanson, Eric T. “Futures Prices as Risk-Adjusted Forecasts of Monetary Policy.” Journal of
Monetary Economics, May 2008, 55(4), pp. 677-91.
Poole, William and Rasche, Robert H. “Perfecting the Market’s Knowledge of Monetary Policy.” Journal of Financial
Services Research, December 2000, 18(2-3), pp. 255-98.
Poole, William; Rasche, Robert H. and Thornton, Daniel L. “Market Anticipations of Monetary Policy.” Federal
Reserve Bank of St. Louis Review, July/August 2002, 84(4), pp. 65-93;
http://research.stlouisfed.org/publications/review/02/07/65-94PooleRasche.pdf.
Radecki, Laurence and Reinhart, Vincent R. “The Financial Linkages in the Transmission of Monetary Policy in the
United States,” in National Differences in Interest Rate Transmission. Basel: Bank for International Settlements,
1994, pp. 291-337.
Reichenstein, William. “The Impact of Money on Short-term Interest Rates.” Economic Inquiry, January 1987, 25(1),
pp. 67-82.
Rigobon, Roberto and Sack, Brian P. “The Impact of Monetary Policy on Asset Prices.” Journal of Monetary Economics,
November 2004, 51(8), pp. 1553-75.
108

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW

Fawley and Neely
Rigobon, Roberto and Sack, Brian P. “Noisy Macroeconomic Announcements, Monetary Policy, and Asset Prices,” in
John Y. Campbell, ed., Asset Prices and Monetary Policy. Chicago: University of Chicago Press, 2008, pp. 335-70.
Roley, V. Vance and Sellon, Gordon H. Jr. “Monetary Policy Actions and Long Term Interest Rates.” Federal Reserve
Bank of Kansas City Economic Quarterly, Fourth Quarter 1995, 80, pp. 73-89;
http://www.kc.frb.org/publicat/econrev/pdf/4q95role.pdf.
Romer, Christina D. and Romer, David H. “Does Monetary Policy Matter? A New Test in the Spirit of Friedman and
Schwartz,” in Olivier Jean Blanchard and Stanley Fischer, eds., NBER Macroeconomics Annual 1989. Volume 4.
Cambridge, MA: MIT Press, 1989, pp. 121-84.
Romer, Christina D. and Romer, David H. “Monetary Policy Matters.” Journal of Monetary Economics, August 1994,
34(1), pp. 75-88.
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.
Rosa, Carlo. “Words That Shake Traders: The Stock Market’s Reaction to Central Bank Communication in Real Time.”
Journal of Empirical Finance, December 2011a, 18(5), pp. 915-34.
Rosa, Carlo. “The Validity of the Event-Study Approach: Evidence from the Impact of the Fed’s Monetary Policy on
U.S. and Foreign Asset Prices.” Economica, July 2011b, 78(311), pp. 429–39.
Rosa, Carlo. “The High-Frequency Response of Exchange Rates to Monetary Policy Actions and Statements.” Journal
of Banking and Finance, February 2011c, 35(2), pp. 478-89.
Sims, Christopher A. “Comparison of Interwar and Postwar Business Cycles: Monetarism Reconsidered.” American
Economic Review, May 1980, 70, pp. 250-57.
Strongin, Steven H. “The Identification of Monetary Policy Disturbances Explaining the Liquidity Puzzle.” Journal of
Monetary Economics, June 1995, 35(3), pp. 463-97.
Swanson, Eric T. “Have Increases in Federal Reserve Transparency Improved Private Sector Interest Rate Forecasts?”
Journal of Money, Credit, and Banking, April 2006, 38(3), pp. 791-819.
Thornton, Daniel L. “The Borrowed-Reserves Operating Procedure: Theory and Evidence.” Federal Reserve Bank of
St. Louis Review, January/February 1988, 70(1), pp. 30-54;
http://research.stlouisfed.org/publications/review/88/01/Borrowed_Jan_Feb1988.pdf.
Thornton, Daniel L. “The Information Content of Discount Rate Announcements: What Is Behind the Announcement
Effect?” Journal of Banking and Finance, January 1998, 22(1), pp. 83-108.
Thornton, Daniel L. “The Codification of an FOMC Procedure.” Federal Reserve Bank of St. Louis Monetary Trends,
March 2001a; http://research.stlouisfed.org/publications/mt/20010301/cover.pdf.
Thornton, Daniel L. “The Federal Reserve’s Operating Procedure, Non-Borrowed Reserves, Borrowed Reserves and
the Liquidity Effect.” Journal of Banking and Finance, September 2001b, 25(9), pp. 1717-39.
Valente, Giorgio. “International Interest Rates and U.S. Monetary Policy Announcements: Evidence from Hong Kong
and Singapore.” Journal of International Money and Finance, October 2009, 28(6), pp. 920-40.

Federal Reserve Bank of St. Louis REVIEW

First Quarter 2014

109

110

First Quarter 2014

Federal Reserve Bank of St. Louis REVIEW