The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
II
Milton Friedman and U.K. Economic Policy:
1938-1979
Edward Nelson
Milton Friedman’s publications and commentaries became the subject of enormous publicity and
scrutiny in the United Kingdom. This paper analyzes the interaction of Milton Friedman and U.K.
economic policy from 1938 to 1979. The period under study is separated into four subperiods:
1938-46, 1946-59, 1959-70, and 1970-79. For each of these subperiods, the author considers
Friedman’s observations on, and role in, key developments in U.K. monetary policy and in general
U.K. economic policy. (JEL E31, E32, E51, E52)
Federal Reserve Bank of St. Louis Review, September/October 2009, 91(5, Part 2), pp. 465-506.
INTRODUCTION
W
hen invited to comment on economic
developments in the United Kingdom,
Milton Friedman frequently prefaced
his remarks with a caveat. Thus in 1964 he testified, “I have not followed in detail the current
circumstances of the British economy.”1 And
much later in 2005, Friedman likewise stated,
“I have no expertise on recent British experience.”
But it was rare for him to confine his remarks to
this caveat. U.K. economic conditions were an
unrelenting source of interest to Friedman, a selfdescribed “life-long student of the monetary and
economic experience” of the United Kingdom,2
who, as we will see, was as early as 1943 citing
speeches by contemporary U.K. policymakers
and drawing on U.K. economic data.
In time, Friedman’s influence on U.K. economic discussion would become so pervasive
1
2
From the question-and-answer portion of Friedman’s March 3, 1964,
testimony, in Committee on Banking and Currency (1964, p. 1144).
Friedman (1980a, p. 55).
that he was part of the U.K. economic policy
debate whether he liked it or not. The fact is that
Friedman’s celebrity was proportionately much
greater in the United Kingdom than in the United
States. The January 1977 issue of the U.K. business
magazine Management Today referred to the
“present controversy, more acute in Britain than
anywhere else, over the teachings of Professor
Milton Friedman”; and even in 2001, long after
the peak of his fame, the London Independent
newspaper described Friedman as “one of the few
economists to have become a household name”
(Independent, August 28, 2001).3
Friedman’s emphasis on the effects of monetary policy, and his opposition to state intervention in the economy, guaranteed that he would
be classified as a marginal figure—if not ignored
outright—by U.K. academic and policy circles
in the early postwar period. Friedman discovered
this for himself during spells in the United
3
A bibliographical appendix gives details for newspaper and periodical articles cited in this paper. Sources in the appendix are arranged
chronologically.
The author is indebted to Charles Goodhart, David Laidler, Alvin Marty, Anna Schwartz, and Gloria Valentine for answers to many inquiries
on the subject matter of this paper. For help on specific issues, he thanks Terry Arthur, Anna Burke, Elizabeth Ennion, Robert Leeson, Mervyn
Lewis, and Louise North. The author also made considerable use of the services of the Federal Reserve Bank of St. Louis Research Library in
the course of obtaining material for this paper; the library staff who have helped include Adrienne Brennecke, Kathy Cosgrove, Barbara Dean,
Julia Seiter, Katrina Stierholz, and Anna Xiao. Luke Shimek and Faith Weller provided research assistance.
© 2009, 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.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
465
Nelson
Kingdom in the early 1950s. The marginal status
of Friedman and his positions persisted, with
short-lived exceptional periods, well into the
1960s. But from the late 1960s and afterward,
Friedman’s positions, while still encountering
resistance at the policymaking level, became the
subject of enormous publicity and scrutiny in the
United Kingdom. The control of inflation was at
the center of U.K. political debate from 1968 to
1979, dominating other policy issues during that
period in a way that it did not in the United States,
where Vietnam, Watergate, and superpower relations competed with, and frequently superseded,
inflation in prominence.
Particularly over this most intense period,
Friedman made interventions himself on the U.K.
scene. He provided commentary on British policy
developments during U.K. visits as well as by
long distance from the United States. Friedman’s
U.K. contributions also included some fundamental statements of his positions—most notably his
lecture, “The Counter-Revolution in Monetary
Theory” (Friedman, 1970a). This lecture, delivered
at the University of London in September 1970,
was treated by Bernanke (2004) as the most representative statement of Friedman’s views on
monetary matters and was what Friedman cited
as the place for a list of “some fundamental propositions of monetarism.”4
Friedman’s contributions to the U.K. scene
included several rebuttals to criticisms of his
research findings on monetary relations. In 1970
he had stated, “I am so happily blessed with critics
that I have been forced to adopt the general rule
of not replying to them.”5 In light of this policy,
the extent to which critics based in the United
Kingdom were able to smoke him out, and provoke a direct, published rejoinder from Friedman,
is impressive: Nicholas Kaldor in the 1970s,
Frank Hahn and Robert Neild in the 1980s, David
Hendry and Neil Ericsson in the 1990s.
The emergence of the United Kingdom as a
major battleground for the debate on Friedman’s
views, and particularly on Friedman’s version of
monetarism, was amplified by the positions of
the leading Keynesians in the United Kingdom.
As Cobham (1984, p. 160) observes, “British
Keynesianism has traditionally been more
‘extreme,’ more ‘hardline,’ than that prevalent
for example in North America.” In particular, in
the first several postwar decades, U.K. Keynesians
were more inclined than their U.S. counterparts
to dismiss altogether the importance of monetary
policy. The United Kingdom featured a greater
and much longer-lasting “nonmonetary,” or
“money does not matter,” brand of Keynesianism.
That this viewpoint was the establishment position in U.K. economics until the 1980s is reflected
in the names of those U.K. economists leading the
opposition to Friedman and monetarism. Among
them were an array of knights and barons: Sir Roy
Harrod, Sir John Hicks, Sir Alec Cairncross, Lord
Kahn, Lord Kaldor, and Lord Balogh.6
Because Keynesianism took a more militant
form in the United Kingdom than in the United
States, the U.K. debates on monetary policy were
more fundamental, and their outcome produced a
greater break in the direction of U.K. policymaking.
This brings me to the subject matter of this
paper, which is the interaction of Friedman and
U.K. economic policy over the period from 1938
to 1979. An obstacle to carrying out a study of this
kind is that Friedman never published a single,
definitive account encapsulating his views on U.K.
developments. True, Friedman and Anna Schwartz
wrote a detailed study of U.K. monetary relations,
their Monetary Trends (Friedman and Schwartz,
1982). But while Friedman once made a shorthand
reference to this book as a study of “U.K. monetary
history” (Wall Street Journal, February 12, 1987),
the volume was not, in fact, a U.K. counterpart
to Friedman and Schwartz’s (1963) A Monetary
History of the United States. Rather, its focus
was on the quantitative analysis of longer-term
economic relations, with Friedman and Schwartz
(1982, p. 605) acknowledging, “We have not made
a similarly exhaustive study of United Kingdom
monetary history.” Monetary Trends does contain
along the way many observations on U.K. develop6
4
Friedman, quoted in Snowdon, Vane, and Wynarczyk (1994, p. 174).
5
Friedman (1970b, p. 326).
466
S E P T E M B E R / O C TO B E R , PA R T 2
2009
After 1979, as governments more sympathetic to monetarism took
charge of the honors system, the tables were turned, and some of
the U.K. monetarist writers received titles: Sir Alan Walters, Sir
Samuel Brittan, Sir James Ball, Lord Griffiths, and so on.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
ments that are relevant to the present paper and
that are incorporated into my discussion. But the
book is not a sufficient statistic when it comes
to studying Friedman’s views on U.K. economic
developments; it does not contain most of
Friedman’s observations on the year-to-year course
of U.K. economic policy. For that, one must turn
to other places.
And, for a comprehensive account, this means
looking in a lot of places. Friedman’s remarks are
widely dispersed across time and location. Not
only his writings but also many interviews are
relevant, as they frequently contain, in the words
of Friedman and Schwartz (1982, p. 623), “illuminating side comments” on U.K. economic matters.
And of those interviews Friedman gave in which
the United Kingdom was the major topic, many
were in U.K. newspapers that have been neither
indexed nor electronically archived.
At first sight, the multiplicity of sources might
not seem too troublesome: Perhaps, it could be
argued, there are only a few basic Friedman references, the remainder being repetition and propagation of his key work. It is true that in the course
of countless lectures, writings, and interviews,
Friedman repeated himself on every dimension:
on the points he made, the historical examples
he cited, the analogies he drew, the anecdotes he
related. Putting aside actual reprints, the repetition is most evident in the considerable number
of his writings that include extended quotations
from previous works. Even the largely new
Friedman-Schwartz Monetary Trends opened its
concluding chapter with a lengthy excerpt from
a 1972 Friedman paper. And in his 1992 book,
Money Mischief, Friedman only makes it through
six lines of text before deploying a quotation from
an earlier book of his. On one occasion, Friedman,
using a stop-me-before-I-kill form of words,
acknowledged this practice: “I’m sorry to quote
myself all the time, but I can’t help it” (Fortune,
March 19, 1984).
Notwithstanding the heavy repetition, there
is usually some marginal contribution—perhaps
an added observation or an update or qualification to previous analysis—even in those works
of Friedman that drew most heavily on his previous writings. In other words, while Friedman
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
repeated himself often, he rarely repeated himself
completely. It would, furthermore, be misguided
to think that Friedman’s most relevant observations on a particular subject appeared in his most
prominent journal publications or in his most
widely cited articles. If anything, the opposite is
the case. This reflects the pattern summarized by
Johnson (1974, p. 346) as Friedman’s “life-long
habit of scattering his new empirical results and
ideas in unlikely places.” Friedman’s tendency
to “fractionate” his written output by spreading
it across an enormous variety of outlets means
that, to obtain the full picture, one has to reconstitute the record from this very wide base.
I have carried out such a reconstitution for
this study. The deployment of extensive source
material is a principal contribution of this paper.
The research here is based on an analysis of
Friedman’s publications, including many articles
neither appearing in his book collections nor
available electronically; his op-ed contributions;
his published interviews in newspapers, magazines, and journals, as well as my own meetings
with him; and much unpublished material. I have
built a database of Friedman’s public statements,
based on my extensive microfilm searches, on
physical inspection of hard copies of newspapers,
on information from search services offered by
companies and by newspapers, and on searches
of newspaper and other databases that are publicly electronically archived. My search through
Friedman correspondence included examination
of samples from the Hoover Institution Archives’
catalogued correspondence and of correspondence
yet to be catalogued by the Archives,7 and my own
correspondence with Friedman from 1991 to 2006.
Also, crucially, I draw extensively on material
(both correspondence and memoranda) provided
to me by Anna Schwartz from her own files.8 As
7
I am indebted to Friedman’s secretary, Gloria Valentine, for fulfilling my requests for information about as-yet-uncatalogued Friedman
files before her retirement in 2007, as well as answering many
inquiries from me on the subject matter of this paper.
8
As well as for the generous access she granted me to this material,
I am indebted to Anna Schwartz for answering numerous inquiries
on the subject matter of this paper. This includes responses to specific inquiries I made during the course of writing this paper and,
more generally, information conveyed in assorted correspondence,
conversations, and meetings with me from 1991 onward.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
467
Nelson
well as (obviously) covering much of her work
with Friedman on monetary policy generally,
these files cover such U.K.-relevant material as
Friedman’s lecture to the London School of
Economics in May 1952.9 I use Congressional
testimony and submissions by Friedman, including several items not included in his comprehensive published bibliographies. I also draw on
transcripts of television interviews Friedman gave
in the United States and the United Kingdom in
the 1960s and 1970s that have been infrequently,
if ever previously, cited.
The remaining discussion in this paper is
divided into chronological segments. For each
segment, I consider the main U.K. economic
events and Friedman’s interaction with them
and then, particular issues in each period. Brief
concluding remarks and a bibliographical appendix complete the paper.
1938-1946
Events
In 1938, Milton Friedman, then age 25, was
based in New York City at the National Bureau
of Economic Research (NBER), where he was
working primarily on completion of his dissertation. His dissertation work came under the
umbrella of what would subsequently be called
“microeconomics,” but Friedman also kept up
with the literature on monetary policy and business cycles. It was in this connection that, as he
told Brian Snowdon and Howard Vane, “I bought
[The General Theory of Employment, Interest and
Money (Keynes, 1936)] in 1938 and paid a dollar
and eighty cents for it.”10
Friedman’s recollection was that he was “if
anything[,] somewhat hostile” to the General
Theory (Friedman 1972a, p. 936), and that he was
influenced by the fact that among older economists
there had been “a good deal of skepticism and
dissatisfaction” in response to the book (Friedman,
9
Where known, I also identify, for the Friedman-Schwartz correspondence, the location of the corresponding copy of the material
in the Friedman papers at the Hoover Archives.
10
In Snowdon and Vane (1997, p. 195).
468
S E P T E M B E R / O C TO B E R , PA R T 2
2009
1982a, p. 9). Moreover, of the younger economists
closest to Friedman, Arthur Burns expressed
reservations about the novelty of the General
Theory, later contending that he had favored
expansionary measures “as early as 1930, before
Keynes’ theories were known.”11
Unlike the initial skeptics, Friedman did not
deny the novelty of Keynes’s theoretical contribution. The General Theory’s explanation for the
Depression and its rationalization for fiscal expansion, Friedman would conclude, were not merely
restatements of preexisting ideas; he would credit
Keynes with a “rigorous and sophisticated analysis” (Friedman, 1968b, p. 1) that provided “a new,
bold, and imaginative hypothesis” (Friedman
1972a, p. 908).
Friedman did share the concern that Keynes’s
book would be seen as giving the green light for
a permanent increase in the size of government.
To the critics, Keynes was providing a respectable theoretical rationalization for extensive
government intervention, through his depiction
of the income-expanding effects of government
purchases and his characterization of private
investment demand as destabilizing. In addition,
Friedman later argued that the underemploymentequilibrium argument in the General Theory was
“highly congenial to the opponents of the market
system” (Friedman and Schwartz, 1982, p. 43).
Friedman’s verdict at the end of the 1970s was
that the idea that “deficits…were a way of expanding the economy” led to a “tremendous growth in
government spending” (May 17, 1979, testimony,
in Committee on the Judiciary, 1980, p. 149).
These misgivings about the perceived policy
implications of the General Theory reflected
Friedman’s free-market, small-government attitudes, already entrenched by 1938. Friedman
assessed in retrospect that “I was mildly socialistic” before graduate study (Newsweek, June 15,
1998). But he had been converted to free-market
attitudes during the portion of his graduate studies that he took at the University of Chicago12—
11
Burns, October 2, 1975, testimony in Committee on the Budget
(1975, p. 170). Friedman described Burns as “really my mentor”
during Friedman’s early career (C-SPAN, November 20, 1994).
12
For example, Friedman (1976a, p. xxi) acknowledged, “I was influenced in this direction by my teachers at the University of Chicago.”
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
“an excellent Department of Economics, I think
the greatest in the country, even before I was
there.”13
What Friedman in 1938 called “this damn
European situation”14 led to the United Kingdom
going to war in September 1939. Following the
United States’ entry into World War II in 1941,
Friedman joined the U.S. Treasury. He later said
that his Treasury colleagues and superiors saw
him as a “starry-eyed theorist.”15 This being the
case, it was as a Keynesian theorist, for Friedman
had largely accepted the theoretical contribution
of the General Theory. In particular, he embraced
its skeptical perspective on monetary policy.
Friedman acknowledged in a television interview
in 1994 that “when I was at the Treasury, I was
essentially a Keynesian, as I believed that the way
to control inflation was by controlling government
spending. I paid very little attention to money”
(C-SPAN, November 20, 1994).
The Keynesian perspective is so clear in
Friedman’s early 1940s writings that monetarists
such as Laidler (2003) have marveled at the contrast with Friedman’s later work. Friedman
expressed a similar sense of surprise when looking at the 1940s work from the vantage point of
three decades later. “In a note on the inflationary
gap that I published in 1942,” Friedman said in a
November 1971 talk (Friedman, 1972b, p. 183),
“I never mentioned the quantity of money or
monetary factors at all!”
The Keynesian position that there was a
region where money and income had a very loose
relationship with one another was, to Friedman,
seemingly confirmed by his look at data. His (1943)
paper on inflation, written while at the Treasury,
13
Milton Friedman Speaks, Episode 8, p. 30 of transcript. Milton
Friedman Speaks was the name given to a series of Friedman talks
in the United States videotaped over 1977-78 and used to promote
interest in a projected television series for Friedman; see Friedman
and Friedman (1998, pp. 477-78, 604). The series was released on
a limited basis on videotape, accompanied by official transcripts,
in 1980, and was more recently repackaged as a commercially
available DVD set. References made in this paper are generally to
the transcripts, but all the quotations from the transcripts also
appear on the DVD releases.
14
March 17, 1938, letter from Milton Friedman to Rose Friedman,
quoted in Friedman and Friedman (1998, p. 77).
15
For example, in Newsweek, July 24, 1978.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
plotted growth rates of nominal money and nominal income for the United States for 1899-1929;
the plot led to Friedman’s judgment that the
relationship was “extremely unstable.”16 This
judgment seems untenable. Simple inspection of
the scatterplot in Friedman’s paper (Friedman,
1943, p. 121) indicates that the money growth–
income growth relationship is clearly positive
and reasonably tight by the standards of rate-ofchange data.
Friedman also embraced some of Keynes’s
post–General Theory ideas, notably those in
Keynes’s How to Pay for the War (1940).
Friedman’s contribution to “inflationary gap”
analysis was in this tradition. This work (Friedman,
1942, 1943) revealed a close following of U.K.
developments. Specifically, Friedman (1943)
discussed “recent English discussion of fiscal
policy [that] has centered on the so-called ‘inflationary gap,’” discussed U.K. gap estimates made
by British economist Frank Paish, and cited a 1943
House of Commons speech by the Chancellor of
the Exchequer, Kingsley Wood, a speech known
to have been drafted by Keynes (see Samuelson,
1946).
“Inflationary gap” analysis had in common
with Friedman’s later work the portrayal of inflation as demand inflation. The details of how
inflation emerged, however, were different in his
1940s’ analysis. Inflation in this analysis was seen
as serving to equalize the nominal value of potential output and the nominal volume of aggregate
spending. Potential output was assumed to have
a physical ceiling, so that price change took up all
the excess spending above this maximum. There
was, in contrast to later Keynesian and monetarist
work, no allowance for “overfull employment.”
Reflecting his later use of the overfull employment
concept, Friedman would say in 1972, “I think
people are wrong in supposing that there is a rigid
ceiling on output such that further increases in
real output are impossible…It is possible to have
overemployment as well as underemployment.”17
16
Friedman (1943, p. 119). Friedman contrasted this with what he
called the “considerably more regular” empirical relation between
consumption and income, which supported Keynesian theory
(1943, p. 120).
17
Friedman (1973a, p. 35).
S E P T E M B E R / O C TO B E R , PA R T 2
2009
469
Nelson
In postwar work, Friedman and others would
accordingly distinguish carefully between potential and maximum output.18 In particular, the
notion that output could temporarily exceed potential, and unemployment fall below its natural rate,
was a contribution of Friedman’s natural rate
hypothesis (e.g., 1968b, 1977a). Nevertheless, the
concepts of positive output gaps, and associated
overfull employment, were innovations neither
of Friedman nor of the Phillips curve literature;
they were in place earlier than the 1950s and
1960s. The possibility that overfull employment
could occur was specifically embodied in the U.K.
policymaking framework by the late 1940s.
“During World War II,” Friedman later recalled,
“governments everywhere had largely assumed
control of the economy. And it was simply almost
taken for granted that they would have to continue
to do so in the postwar period.”19 The Attlee government was elected in the U.K. general election
of July 1945. Friedman noted, “In Britain, the
Labour Party’s postwar victory over Winston
Churchill spelled a commitment to central planning” (Newsweek, July 14, 1975).
Issues
Nationalization and Central Planning.
Friedman observed that the postwar shift to
greater government economic control had been
justified on efficiency grounds: It was believed
“that centralized and comprehensive economic
planning and control by government is an essential requisite for economic development”
(Friedman, 1958a, p. 505). He noted that, in
particular, nationalization of industries was motivated by this consideration (The Listener, April 27,
1978). The Attlee government used the efficiency
argument when implementing a broad nationalization program after it came to power.
This nationalization program was believed
to be appealing to U.K. electors, to judge by the
notice of their plans that leading Labour politi18
19
Friedman would still believe that there was a physical ceiling on
output (see the expositions of his plucking model in Friedman
1964b, 1993), but he no longer treated this ceiling as synonymous
with the natural level of output.
Quoted in Levy (1992).
470
S E P T E M B E R / O C TO B E R , PA R T 2
2009
cians gave in the months approaching the election.
Stafford Cripps, later Chancellor of the Exchequer,
said, “We must replace the libertinism of private
enterprise by a planned system of economy which
calls for a considerable measure of state control
and ownership.” As did many in the West, Cripps
cited the Soviet Union as a successful economic
model: “In Russia you have a State-planned and
controlled industry, and I cite this as an example
to show that some form of centralized planning
and control helps and does not retard efficient
production and full employment” (NewsChronicle, December 18, 1944).
The nationalizations undertaken by the Attlee
government (1945-51) encompassed mining,
communications, the railway system, and steel.20
Friedman had anticipated that a still more comprehensive nationalization scheme would be carried
out. He observed in 1972 of the late 1940s, “If you
had asked us then about the health of capitalism
and free enterprise 25 years later, I think we would
have said it would be closer to its deathbed than
it actually is now.”21
In fact, Friedman’s Capitalism and Freedom
(1962a) contains remarks to the effect that he
thought that in the United Kingdom the move
toward greater government intervention had
peaked even before the Attlee government left
office. In particular, Friedman took comfort
from the fact that central planning, as opposed
to nationalization, had not endured in Western
economies beyond the 1940s. The detailed direction of resources, public and private, had been
foreshadowed by the Attlee government; as
Friedman noted, “immediately after World War II,
it was thought that the government was going to
get involved, especially in Britain…in central
economic planning on a large scale.”22 Efforts to
replace the price system with government direction of allocation decisions had, he argued, faltered and led to socialism peaking in the United
Kingdom in 1948 (Vision, April 1972). Friedman
20
See Childs (2006, p. 14) for a tabulation of the Attlee government’s
nationalizations.
21
Friedman speaking in Business and Society Review, Spring 1972;
reprinted in Friedman (1975a, p. 253).
22
Free to Choose, PBS debate 1980, Episode 3, p. 9 of transcript.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
(1962a, p. 11) singled out the fate of the Attlee
government’s Control of Engagements Order,
which, he said, would have meant a directedlabor economy if it had been enforced. The Order
was not, in fact, enforced heavily and was then
repealed, an event Friedman identified as a “turning point” (Friedman, 1962a, p. 11) when “central
planning came to a screeching halt” in the United
Kingdom.23
But Friedman further revised his opinion in
the 1970s: He observed in 1972, “I was much
more optimistic in 1962 than was justified by
what happened later.”24 Friedman continued to
acknowledge that the momentum for planning
and nationalization had stalled, noting that
“[t]here is less central planning in Britain now
than in 1946,”25 but he now judged that this had
“diverted…growth [in government] to a different
channel” (New York Times, August 13, 1994).
Greater government influence on resource allocation, he argued, had instead been achieved via
expansion of government spending (including
transfer programs) and of regulation (Friedman,
1976c; Friedman and Friedman, 1998, p. 582). This
changed perception was reflected in Friedman’s
descriptions of the U.K. system: Whereas he
characterized what was launched in the United
Kingdom in the 1940s as “a policy of welfare
statism and central planning” (Saturday Evening
Post, May/June 1977), Friedman argued that the
system evolved into “a socialist and welfare state”
(National Review, December 31, 1997).
Cheap Money. Many countries followed
“cheap money” policies in World War II and its
aftermath; the U.S. case is the subject of Friedman
and Schwartz (1963, Chaps. 10 and 11). In the
United Kingdom, the postwar “cheap money”
policy is associated with the attempt by
Chancellor of the Exchequer Hugh Dalton to
break with the practice his adviser, John Maynard
Keynes, had described as “the unwillingness of
most monetary authorities to deal boldly in debts
23
In Business and Society Review, Spring 1972; reprinted in Friedman
(1975a, p. 254).
24
Friedman, October 20, 1972, remarks, in Selden (1975, p. 51).
25
From Business and Society Review, Spring 1972, reprinted in
Friedman (1975a, p. 254).
of long term” (Keynes, 1936, p. 207). Although
announced by the Attlee government upon its
election, the long-term bond program began in
earnest in October 1946, several months after
Keynes’s death. Among the new government
bonds created in 1946 was a series of 2.5 percent
“irredeemable” securities, that is, securities that
might be held indefinitely as a source of interest
income; these new long-term securities were
unofficially known as “Daltons” or “Dalton consols.” By issuing very-low-yield medium- and
long-term securities, Dalton attempted to extend
the government’s existing low interest rate policy
to the entire term structure. “We have been gradually conditioning the capital market to a longterm rate of 2½%,” Dalton observed. “...I am sure
that our cheap money policy should continue to
be resolutely pressed home” (Financial Times,
October 17, 1946).
The U.K. and U.S. authorities’ interest in
influencing long-term rates rested heavily on the
Keynesian position that long-term rates mattered
for aggregate demand much more than short rates;
this interest was qualified by the consideration
that, as Friedman and Schwartz (1963, p. 700)
observe, even the sensitivity of demand to longterm rates was not thought to be substantial.
Low long-term interest rates were also perceived
as contributing to the flexibility of fiscal policy
by easing the financing of the national debt. So
the extension of the cheap money policy to the
long-term market had both Keynesian and debtmanagement motivations.
The Dalton program of October 1946, while
involving the creation of new debt instruments,
was intended to drive existing longer-term securities’ rates down to 2.5 percent too (see, e.g.,
Hallowell, 1950, p. 41); this contrasted with the
rates between 3 percent and 3.5percent prevailing for most of the period since 1932 (Hallowell,
1950, p. 23; Robertson, 1949, p. 22).26 Since Bank
Rate was left unchanged, the experiment was not
making use of the expectations theory of the term
structure. On the contrary, the expectations theory
26
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
When serving as a discussant of Friedman’s in 1970, Sir Roy Harrod
paraphrased Dalton’s rationale for the reduction in the long-term
rate as that if “we could run a great war at 3%, we ought to be able
to run the peace at 2.5%” (Harrod, 1971, p. 59).
S E P T E M B E R / O C TO B E R , PA R T 2
2009
471
Nelson
would suggest that keeping the short rate
unchanged tended to work against the success of
a policy to reduce long-term rates. From the perspective of the General Theory, however, the
approach made some sense: The General Theory
saw securities as becoming equivalents of money
before their yield became zero; insofar as shortterm interest rates were perceived as already
having hit their floor, but long-term rates had not
reached their floor, the monetary authorities could
carry out operations directly in long-term securities markets to encourage reductions in longerterm rates.
Wilson (1984, p. 76) observes that there were
“few British economists in the 1950s and 1960s
who advocated control of the money supply—
Robertson, Robbins, Paish, Dacey, myself [Thomas
Wilson], and one or two others.” Among those on
this list who were active in the 1940s, Robertson
was perhaps the leader and is acknowledged in
Friedman and Friedman (1998, p. 247) as an early
distinguished skeptic regarding Keynesian economics. That skepticism is evident in Robertson’s
discussions of the Dalton monetary policy, as
Robertson (1949, p. 22) counts himself among
those “who dared to question the wisdom of this
[1946] further turn of the [cheap money] screw.”
A fellow critic, Dacey (1947, p. 59) wrote that “it
is surprising that Mr. Dalton should have thought
it good statesmanship to press rates down further
at a time when inflationary forces are only kept in
check with the assistance of a formidable administrative apparatus [i.e., price controls].”
Friedman’s own later work contained critical
observations on the U.K.’s monetary framework
during the 1940s. There are many further criticisms
implicit in Friedman’s descriptions of what monetary policy can and should do. Friedman’s framework centered, first and foremost, on the point
that “monetary policy is not about interest rates;
monetary policy is about the rate of growth of the
quantity of money.” 27 The Fisher relation provided the only enduring channel by which the
central bank could affect interest rates, be they
short- or long-term. Monetary policy could exert
other, more transitory influences on interest rates,
but Friedman was skeptical that these influences
justified central banks’ claims that they could
control long-term interest rates (see, e.g., Friedman
and Schwartz, 1963, p. 514). Certainly, he believed
that a base money injection could produce some
temporary downward pressure on long-term rates,
both via the standard expectations channel (i.e.,
via the liquidity effect on current and expected
future short rates) and via a portfolio effect on the
long-short spread or term premium (see, e.g.,
his June 1966 memorandum to the Board of
Governors, published in Friedman, 1968a, p. 156).
But for the central bank to exploit these effects
in a way that made the long-term interest rate a
policy instrument would require being able to
overwhelm the “nonmonetary forces affecting
interest rates”28 as well as the Fisher effect, which
showed up in “long-term interest rates much
sooner”29 than in short-term rates and worked in
the opposite direction of the liquidity and portfolio effects of the money injection. Moreover,
Friedman noted, the sustainability of this policy
was doubtful, since for a central bank to “peg a
particular interest rate…it must accept whatever
happens to other magnitudes affected by the
[monetary] base, including the level of inflation”
(Wall Street Journal, April 5, 1990).
The Dalton attempt to bring down long-term
rates had the temporary appearance of success.
Rates on existing long-term securities fell toward
the new 2.5 percent baseline. For example,
Friedman and Schwartz’s (1982, Table 4.9, p. 133)
data on “old” consols (i.e., the perpetual-horizon
security already being traded before the release of
the new, “Dalton” consols) show an average rate
of 2.92 percent in 1945 and 2.6 percent in 1946.
But the effort to hold down long-term interest rates
did prove unsustainable, and in the course of
1947 long-term rates rebounded; Friedman and
Schwartz’s series shows an average for 1947 of
2.76 percent, rising to 3.21 percent in 1948.
Friedman (1970a, p. 8) observed, “Chancellor
Dalton tried to follow the Keynesian policy of
keeping interest rates very low. As you all know,
28
From Friedman’s October 1965 memorandum to the Board of
Governors, published in Friedman (1968a, p. 136).
27
29
Friedman (1983, p. 11).
From Friedman’s appearance on Meet the Press, October 24, 1976.
472
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
he was unable to do so and had to give up.” A few
months after Dalton left office, the Financial Times
reported, “The attempt to hold the rate of interest
on government long-term borrowing at 2½% has
now been officially abandoned” (Financial Times,
January 3, 1948).
The pressure on the cheap money policy in
the United Kingdom was more acute than in the
United States because the associated pressure on
aggregate demand was one-sided. An interest rate
peg can in principle be contractionary in effect,
as when the central bank has to withdraw base
money to enforce the peg. Friedman and Schwartz
(1963, p. 596) find that the U.S. cheap money
policy indeed produced deflationary pressure over
1948, and Friedman and Schwartz (1982, p. 76)
classify 1948-49 as a business contraction in the
United States. No corresponding contractionary
episode is evident during the U.K. postwar cheap
money period, and Friedman and Schwartz (1982,
p. 76) classify 1946-59 as a continuous expansion
in the United Kingdom. The U.K. experience consequently corresponded more literally than did
the U.S. case to Friedman’s summary statement
that “the stock of money rose as a result of the
cheap-money policies and so did prices, either
openly or in whatever disguise.”30
The Dalton policy on long-term interest rates
was the first element of the U.K. monetary framework to break under this one-sided pressure; the
exchange rate policy was next, with sterling devaluation taking place in 1949. The outbreak of the
Korean War in 1950 magnified the pressure on the
remaining component of the cheap money policy,
namely, the holding of Bank Rate at 2 percent.
The Financial Times noted (February 10, 1951):
“Both the British and American governments seem
determined not to use higher interest rates to combat inflation.” This determination contrasted with
Friedman’s position, which he articulated in 1952
as follows: “The purpose of monetary policy is to
maintain price stability, and on some occasions
this will call for actions that tend to raise interest
30
31
rates.”31 Bank Rate was finally raised by the
newly elected Churchill government in November
1951. “No country succeeded in stemming inflation without adopting measures that made it
possible to restrain the growth in the stock of
money,” Friedman observed at the end of the
1950s. “And every country that did hold down
the growth in the stock of money succeeded in
checking the price rise.”32
1946-1959
Events
Friedman’s first visit to the United Kingdom
in 1948, consisting of “two or three days in
England, in London,”33 left him convinced that
it was being “economically strangled by the law
obedience of her citizens” (Friedman, 1962b).
Friedman was persuaded by the argument made
by George Stigler, with whom he made the trip,
that price controls were distorting the United
Kingdom to an extent that they were not in France,
because of the more-extensive French underground economy.34 Price controls had been introduced by the United Kingdom in wartime,
Friedman later observed, in an “attempt to suppress the inflation arising from wartime spending,
financed largely by increasing the money supply”
(Newsweek, November 27, 1972). The Attlee government continued the controls into peacetime.
Friedman opposed price controls both as a wartime
and a peacetime measure (see his October 6, 1969,
testimony in Joint Economic Committee, 1970,
pp. 815-16) The peacetime controls in the United
Kingdom and on the Continent were, he argued in
an early intervention, impeding Europe’s economic
recovery (New York Times, January 11, 1948).
Other damaging restrictions, Friedman contended, came in the “foreign exchange controls
that strangled Western Europe after the war.”35
32
May 25, 1959, testimony, in Joint Economic Committee (1959a, p.
607; p. 138 of 1964 reprint).
33
Milton Friedman Speaks, Episode 1, p. 9 of transcript.
May 25, 1959, testimony, in Joint Economic Committee (1959a,
p. 607; p. 138 of 1964 reprint).
34
Friedman (1962b); and Milton Friedman Speaks, Episode 1, p. 9
of transcript.
March 25, 1952, testimony, in Joint Committee on the Economic
Report (1952, p. 736).
35
Friedman (1964a); p. 78 of 1969 reprint.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
473
Nelson
Foreign exchange controls were not initially
emphasized by Friedman in the 1940s, when his
focus was on price control. “I wrote my first
article [on exchange controls] in 1950,” Friedman
later recalled, “when I was in France in connection with the Common Market arrangements”
(Jerusalem Post, November 6, 1987). For the United
Kingdom, Friedman’s position, maintained from
the early 1950s, was that exchange controls had
such a depressing effect on the level of the pound
sterling that if the U.K. authorities floated the
pound and maintained exchange controls, the
pound would tend to depreciate, whereas if they
floated simultaneously with the removal of
exchange controls, the pound would appreciate
(Friedman, 1953b; Friedman and Schwartz, 1982a,
pp. 290-94). This conjecture was borne out when
the pound appreciated after the abolition of
exchange controls in 1979. Friedman criticized
the exchange controls on economic grounds, but
“entirely aside” from their economic aspects, he
opposed them on grounds of “human freedom.”36
It violated the “free market in ideas,” he said in
1977, “if a country, as Great Britain did immediately after the war, has exchange controls under
which no citizen of Britain may buy a foreign
book unless he got authorization from the Bank
of England.”37
Even by the early 1950s, the United Kingdom
had acquired a basket-case image in the United
States for its postwar performance, with extensive
economic aid by the United States to the United
Kingdom highlighting the problem. “When people
say, ‘Well, American aid bailed Germany out,’
I add that American aid also bailed Britain out,”
Friedman later observed. “The amount we gave
to Britain in the British-American loan was far
greater than anything Germany got” (Saturday
Evening Post, May/June 1977).
The United Kingdom was also a recipient of
aid from the United States via the Marshall Plan.
Friedman contended that “Europe would have
recovered with or without the Marshall Plan,”
and opposed the Plan at the time and in retrospect
(Friedman, 1982a, pp. 32-33). He argued that the
36
37
Free to Choose BBC2 debate, March 22, 1980, p. 21 of transcript.
Milton Friedman Speaks, Episode 3, p. 16 of transcript.
474
S E P T E M B E R / O C TO B E R , PA R T 2
2009
“Marshall Plan and similar programs” of the U.S.
government had “been harmful to the rest of the
world”38 because government-to-government
economic aid strengthened the government sector
at the expense of the private sector.39 It was,
nevertheless, as an adviser to the Marshall Plan
that Friedman made a second visit to Europe, in
late 1950, basing himself in Paris (Friedman, 1992,
p. 248; Friedman and Friedman, 1998, Chap. 12).
“Plans to spend a quarter at the London School
[of Economics] in the spring of 1952 fell through,”
Friedman recalled in 1994, “but did lead to my
making a three-week trip to Britain and France,
giving two lectures at the London School of
Economics.”40 Friedman’s talk, “Classical CounterRevolution and Monetary Theory and Policy,” at
the London School of Economics in May 1952,
opened with a major gaffe—or a clanger, to use
the U.K. nomenclature. His speaking notes state,
“With some hesitancy the American speaks on
this topic to an English audience. Basic contributions all English. Classical—Hume, Ricardo,
Thornton, Marshall…”41 The problem, of course,
was that David Hume was Scottish, not English.
Indeed, the two British economists with whom
Friedman identified most and whom he would
most often quote in his writings, Hume and Adam
Smith, were Scottish. To judge by Friedman’s later
statements, he learned his lesson and became
more careful about distinguishing the Scottish
from the English. In arguing against the Bretton
Woods system during Congressional testimony
in 1963, Friedman noted that what mattered to
the U.K. consumer was a good’s U.K. price, not
the same price expressed in U.S. dollars; different dollar values of the price were “all the same
to an Englishman—or even a Scotsman.”42
Around the time of Friedman’s 1952 visit, the
U.K. unemployment rate averaged 2.1 percent;
38
September 23, 1971, testimony, in Joint Economic Committee
(1971, p. 722).
39
For Friedman’s elaborations of this argument, see Friedman (1958a)
and Newsweek, December 21, 1970.
40
From Friedman’s notes “1952 MF trip,” in a 1994 letter to Anna
Schwartz; courtesy Anna Schwartz.
41
May 1952 lecture notes by Friedman; courtesy Anna Schwartz.
42
November 14, 1963, testimony, in Joint Economic Committee (1963,
p. 454).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
in 1951 it was only 1.2 percent, and in 1953 the
average was 1.8 percent.43 These were lower rates
than prevailing in the United States, but Friedman
later cautioned against interpreting such low rates
as a badge of honor. In his Nobel lecture, Friedman
cited the United Kingdom in the 1950s as an
example of a country with an inefficiently low
unemployment rate, reflecting the fact that a
“highly static rigid economy may have a fixed
place for everyone” (Friedman, 1977a, p. 459).
Friedman elaborated in 2004: “Progress depends—
it sounds funny, but it’s true—on unemployment…
Because that’s the only way you can provide the
necessary labor force for the new development,
the new industries that are coming along”
(Investor’s Business Daily, April 15, 2004).
The U.K. economy in the 1950s not only featured an inefficiently low unemployment rate,
but also a tendency for aggregate demand to be
expanded too rapidly, forcing unemployment
temporarily below its low natural value and creating inflationary pressure. Since U.K. policymakers
were aware by the late 1940s of the distinction
between full and overfull employment, the question arises why they kept overdoing expansion.
Some of the overheating in the late 1940s and the
1950s might be attributed to preemptive stimulus
in anticipation of a collapse in private demand.
Friedman (1973a, p. 5) noted that while “a great
post-war depression…was widely predicted,” it
“kept being expected but it never occurred.” This
observation was true of the United Kingdom, with
Chancellor Dalton stating in 1945 that the “government must arm itself with anti-slump powers, so
that never again, as in past years, shall prices, productivity, and employment all fall away through
the failure of private enterprise” (Financial Times,
November 23, 1945). In particular, “secular stagnation,” due to drying up of private investment
opportunities and to excessive consumer saving,
was feared in the 1940s and cited as a reason for
the government stepping in with its own demand
for output. Friedman was an early critic of the
secular stagnation thesis (Friedman, 1948, p. 262),
and the criticism had become widespread by the
1950s as the prospect of a consumption collapse
43
London and Cambridge Economic Service (1963, Table F, p. 11).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
dwindled. In fact, the secular stagnation theory
was one aspect of Keynes’s thinking that was
widely rejected in the United Kingdom even in
the Keynesian 1950s; the Financial Times, for
example, referred to “Keynes’ one-sided fear of
over-saving” (May 23, 1955) and to “Keynes’
incredibly shortsighted forecast of declining
investment opportunities” (October 15, 1956). But
precisely because the relevance of the secular
stagnation hypothesis was in so much doubt by
the 1950s, it is hard to cite belief in it as the reason
for continued U.K. overexpansion in that decade.
The repeated failures over the 1950s to deliver
the proper dosage of demand seem most attributable to the U.K. government’s misguided view of
how to affect demand. Here, fiscal policy received
pride of place, reflected in Chancellor of the
Exchequer Peter Thorneycroft’s observation in
1957, “The big instrument of government policy
in all these matters is the budget” (Daily Express,
July 13, 1957). This contrasts with Friedman’s
position on fiscal policy which, of course, was
this: “In my opinion, a budget deficit is ‘expansionary’ only if it is financed by printing money”
(Newsweek, February 15, 1971). According to this
interpretation, any apparent connection between
fiscal actions and aggregate demand was not an
indication of the working of the Keynesian multiplier process, but was instead a by-product (a
“disguised reflection” in the terminology of
Friedman and Meiselman, 1963) of the fact that
deficits in practice were monetized.
Figure 1 plots the ratio of the U.K. budget
balance to nominal gross domestic product (GDP),
and the growth rate of the U.K. monetary base.
The budgetary series is one available for 1948-99,
now discontinued but formerly reported as line
80 in hard copies of the International Monetary
Fund (IMF)’s International Financial Statistics
(IFS). Nominal GDP is the annual average series
for the United Kingdom from Haver-IFS (downloaded March 2009). The monetary base series is
the annual average of a series obtained by splicing annual averages of the Capie-Webber series
(1985) into the Bank of England break-adjusted
base money series. The plotted growth rate of the
base closely resembles that depicted in Benati
(2005, Chart 1), which was based on similar
S E P T E M B E R / O C TO B E R , PA R T 2
2009
475
Nelson
Figure 1
U.K. Budget-Balance Share of GDP versus Monetary Base Growth (1948-99, annual data)
Percent
20
Base Money Growth
15
Budget Share of GDP
10
5
0
–5
–10
1948
1953
1958
1963
1968
sources; in addition, for 1948-71, the growth rates
resemble those of the high-powered money series
in Friedman and Schwartz (1982, pp. 125-26).
Base growth is highly correlated with the budgetbalance share for 1948-79 (correlation = –0.80),
while the correlation for 1980-99 is negligible
(correlation = 0.08). The association between base
growth and deficits supports Friedman’s contention that in the early postwar decades U.K. government deficits were monetized,44 so that the period
simply does not provide clean evidence of the
effects of “pure” fiscal policy. Attempts to quantify a multiplier impact of deficits and surpluses
without attempting to hold constant the reaction of
the monetary authorities merely beg the question.
It may seem perplexing that my discussion
of demand policies in the United Kingdom in the
1950s has been able to proceed so far without a
discussion of the fact that the pound sterling was
44
Friedman (1975c, pp. 72-73) argues that fiscal deficits were stimulative in Western countries in the postwar period (up to that
point) because they were, in practice, monetized.
476
S E P T E M B E R / O C TO B E R , PA R T 2
2009
1973
1978
1983
1988
1993
1998
on a fixed exchange rate. I have occasionally seen
it argued that the U.K. authorities did not actually
pursue Keynesian policies before the 1970s
because they were constrained by their Bretton
Woods obligations.45 This argument overlooks
the extent to which foreign exchange controls
reduced the impact of the exchange rate constraint
on the formation of demand management policy.
As Friedman and Schwartz (1963, p. 105) observe
in discussing the postwar United Kingdom, “the
development of direct exchange and trade controls
gave it means of affecting its balance of payments
other than through movements in prices and
incomes”; relatedly, exchange controls gave the
U.K. authorities some room to separate fixing the
exchange rate from setting interest rates. There
was no occasion in the 1950s when there was a
Bank Rate increase that could not be justified by
domestic conditions; reflecting this, the Bank of
45
This argument was used, for example, by R.J. Ball in the Financial
Times, February 4, 1981.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
England governor testified in 1958 that foreign
exchange market considerations had determined
the timing (in terms of the specific week) of Bank
Rate moves, but not the actual moves themselves,
which were invariably also shaped by internal
considerations.46 Bretton Woods did not override
what the Financial Times (December 21, 1953)
called the “modern principle of shaping policy
by reference to domestic monetary needs.” The
coexistence of substantial monetary policy independence and a fixed exchange rate explains why,
during the 1950s and 1960s, Lionel Robbins simultaneously criticized the idea of floating rates and
advocated that the U.K. monetary authorities
manipulate the monetary base to achieve price
stability. As Friedman (1978) later pointed out, a
bona fide conflict between the exchange rate and
domestic considerations in the Bretton Woods
system led typically to the exchange rate giving
way, as it did in the United Kingdom in 1949,
1967, and 1972.
By the time of Friedman’s 1953-54 spell in the
United Kingdom, the development of Friedman’s
monetarism was well advanced, to the point
where, in reprinting his (1942) essay on inflation,
he added material on money, attributing its previous absence to the “prevailing Keynesian temper” of the 1940s (Friedman, 1953a, p. 253).
Friedman also had had a letter published in The
Economist (January 3, 1953) advocating that sterling be floated.
A review of Friedman’s Essays in Positive
Economics (1953a) appeared in the Financial
Times of February 8, 1954, apparently the firstever mention of Friedman in that newspaper. The
review was devoted mostly to Friedman’s argument for floating exchange rates (Friedman,
1953b). The review said that Friedman “grossly
overstates his case…when claiming that flexible
exchanges would have obviated the sterling crises
of 1947, 1949, and 1951.” The review apparently
regarded Friedman as neglecting the possibility
that devaluation could worsen the current account
balance measured in pounds. This was one of
several critiques of Friedman’s argument for float46
See the answers by Bank of England Governor Cobbold, in Radcliffe
Committee (1960, pp. 137, 155).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
ing exchange rates that took Friedman as implying
that a float removes current account deficits in
the balance of payments. It is true that Friedman
generally regarded depreciations as good for the
sterling trade balance and that a low-enough sterling exchange rate would remove the trade deficit;
he said so in his 1953 letter to The Economist, for
example. But, to my knowledge, Friedman did
not claim that a floating exchange rate would
converge to the value consistent with a zero trade
or current account balance; his claim for floats
was the correct and general one that they eliminate the possibility of balance of payments deficits
or surpluses, so that “[b]alance of payments
problems in the technical sense are a reflection
of price fixing.”47
Issues
The Early Monetarist. Two beliefs are widespread about Friedman’s origins as a monetarist.
The first belief is that his earliest monetarist
work appeared in 1956.48 The second is that, in
the 1956 paper and elsewhere, Friedman merely
dotted the i’s and crossed the t’s of existing work
by Keynes and of pre-1956 Keynesian work, so
that the theoretical innovations of monetarism
were negligible (at least if contributions regarding
the expectational Phillips curve are put to one
side). Both beliefs are misconceptions. They
are naturally handled together since the nonKeynesian aspects of Friedman’s framework are
not all present in his 1956 paper, but are evident
if the totality of his work over 1948-58 is considered. The discussion below shows that the
literature’s characterization of the 1956 paper as
the launching pad for Friedman’s monetarism
has obscured some of the major theoretical differences with Keynesianism that were already
visible in other work by Friedman in the 1950s.
It complements the cataloguing by Friedman
(1972a) and Meltzer (1977) of distinctions between
Friedman’s framework and Keynesianism but
includes items not in their lists. It also serves to
47
September 23, 1971, testimony, in Joint Economic Committee
(1971, p. 701).
48
See, for example, Eshag (1983).
S E P T E M B E R / O C TO B E R , PA R T 2
2009
477
Nelson
confirm Friedman’s observation (in his Reason
interview of June 1995) that the key arguments
he made in his 1960s’ and 1970s’ publications
were already present in his 1950s’ work.
First, Friedman argued as early as 1948 for a
focus on monetary policy for the control of inflation. A letter he and other Chicago faculty members wrote to the New York Times in early 1948
was entitled “Control of Prices: Regulation of
Money Supply to Halt Inflation Advocated”
(January 11, 1948). In claiming that “a marked
increase of the general level of prices unaccompanied by a marked increase in the supply of money
is a rare if not nonexistent phenomenon,” this
letter reflected early dissent by Friedman from
Keynesianism49 and was followed by Friedman’s
(1950, p. 474) sympathetic remarks about the
quantity theory. In 1952, Friedman was firmly
associated with the quantity theory position, and
Friedman published his finding that income and
price changes in U.S. wartime episodes were “more
readily explicable by the quantity theory than by
the income-expenditure theory” (Friedman, 1952,
p. 721).50
Second, Friedman (1951) argued for treating
prices and wages as endogenous variables at all
levels of employment, in contrast to the Keynesian
treatment (and Friedman’s in 1942-43) of prices
as insensitive to aggregate demand until full
employment was reached.
Third, Friedman advocated floating exchange
rates from 1950, when his (1953b) essay on the
subject was drafted, using arguments that rested
on the ability of monetary policy to deliver price
stability.
Fourth, in the 1950s Friedman rejected costpush factors as a source of sustained inflationary
pressure. While Friedman (1948) had given credence to cost-push factors as one factor driving
up wages, in 1951 he said, “My views about this
have changed considerably in the last few years”
(Friedman, 1951, p. 228). In 1952 Friedman testified, “I think the so-called wage-price spiral has
49
Brunner and Meltzer (1993) date Friedman’s earliest dissent to
Friedman (1944).
50
Friedman had also confronted Roy Harrod with his views on
velocity in a meeting in Chicago in early 1951 (Harrod, 1971, p. 58).
478
S E P T E M B E R / O C TO B E R , PA R T 2
2009
been enormously exaggerated, that what we have
had has been inflationary pressure pulling both
wages and prices up.”51 His rejection of cost-push
is clear also in his repudiation in Friedman
(1953b) of the idea that exchange rate depreciation
could trigger a self-sustaining wage/price spiral.
Friedman’s position from the early 1950s was that
cost-push factors had a zero mean in themselves;
upward pressure on wages or prices in one sector
would be “balanced by declines elsewhere in
other prices and costs.”52 Any tendency for inflation to exhibit a sustained rise in the face of a
positive wage or price shock reflected monetary
accommodation, so cost-push factors could not
alter inflationary expectations in the absence of
a monetary expansion. This rejection of cost-push
distinguished Friedman not only from Keynesians
(among whom the popularity of cost-push explanations increased over the 1950s, in both the
United States and the United Kingdom), but also
from some advocates of monetary control such
as Robbins. In contrast to Friedman, Robbins
believed that wage-push factors put a positive
bias into U.K. wage inflation, in the face of which
monetary policy needed to be contractionary
(rather than simply nonaccommodating) to deliver
price stability.53
Fifth, Friedman rejected the notion of a longrun trade-off between unemployment and inflation. In a 1950 symposium (Wright, 1951, p. 243),
Friedman said “I don’t know what you mean by
saying unemployment will police inflation.” In
1952 Congressional testimony he said, “Rather
51
March 25, 1952, testimony, in Joint Committee on the Economic
Report (1952, p. 727); see also Friedman (1952, fn. 7).
52
March 25, 1952, testimony, in Joint Committee on the Economic
Report (1952, p. 736).
53
Note that the view that unions can be an autonomous source of
wage-push is distinct from the view that unions can raise the natural
rate of unemployment, since wage- or cost-push refers to inflationary pressure created for a given difference between unemployment
and the natural rate. As for whether union pressure could affect
the natural rate, Friedman regarded the conditions for this to occur
as restrictive—he argued that unions could raise unemployment
in certain sectors, but not necessarily in the aggregate (Friedman,
1951; The Times, August 29, 1973; The Economist, September 28,
1974)—but he sometimes implied that the conditions for an effect
on aggregate unemployment might have been satisfied in the
postwar United Kingdom (e.g., Friedman 1963a; Friedman and
Friedman, 1980).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
than regarding the objectives of high employment and of price stability as inconsistent, I think
that fundamentally price stability will promote a
high level of output by avoiding a good many of
the interruptions to output that we have had in
the past, by giving people stable expectations,
and so on.”54
Sixth, the preceding two points combined
with his doubts about fiscal policy meant that
Friedman believed that monetary policy was
sufficient to control inflation. The “sufficient”
language was used in Friedman (1958b), and contrasts Friedman’s position directly with Keynes’s
view that monetary policy actions could not be
sufficient for delivering price stability (see Nelson
and Schwartz, 2008).
Turning now to contributions present in
Friedman (1956), a seventh 1950s’ contribution
by Friedman was to specify money demand as
dependent on a vector of interest rates. This means
that the monetary policy transmission mechanism
cannot be summarized by a single interest rate.
Patinkin (1969) claims that Friedman’s specifying
money demand as dependent on interest rates
makes his specification Keynesian. This overlooks the fact that Friedman does not condense
the nonmoney assets into a single asset, as
Keynesian analysis typically did. Moreover, preKeynes writers had made money demand interest
elastic, and the specifics of Friedman’s money
demand approach differ from those of Keynes.
Keynes had broken money demand into transactions and speculative components, with only the
second component interest-elastic and otherwise
“idle.” Friedman (1956) rejects the concept of idle
money and instead models every unit of money
as interest-elastic (possibly relative to own-rates),
and held for all motives at the same time (an aspect
of Friedman’s analysis acknowledged by Patinkin,
1965, p. 75).
Eighth, Friedman (1956) indicates that his
conception of money demand rules out the liquid54
March 25, 1952, testimony, in Joint Committee on the Economic
Report (1952, p. 727). Formulations such as this were precursors
to the descriptions of the inflation problem given by many who
worked in policymaking from the late 1970s onward, both in the
United Kingdom and the United States (regarding the latter, see
the statements by Paul Volcker and Alan Greenspan quoted in
Lindsey, Orphanides, and Rasche, 2005).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
ity trap, which he argued in Friedman (1972a)
was central to the General Theory. In light of the
discussions of the liquidity trap in recent years
by Paul Krugman and Lars Svensson,55 and their
attribution of liquidity-trap views to Keynes, it
may come as a surprise that Patinkin (1972a) and
many other Keynesians objected to Friedman’s
association of the General Theory with the liquidity trap. But it was hardly an off-the-wall interpretation on Friedman’s part. James Schlesinger,
by no means a close ally of Friedman, argued
strongly that the liquidity-trap thesis was central
to the General Theory in his 20-year retrospective
on the book. Schlesinger (1956), Friedman (1972a),
and Beenstock (1980) all provide their own, apparently independently constructed, lists of quotations
from Keynes (1936) supporting this interpretation,
and even Patinkin (1976a) acknowledged that
passages of the General Theory treat the liquidity
trap as empirically relevant. As Friedman (1972a,
p. 942) put it, again and again, Keynes’s “final
line of defense is absolute liquidity preference.”
Over 1948-58, all the elements of Friedman’s
monetarism fell into place and are recognizable
as the positions he took in what he termed the
“dispute in the 1950s or early 1960s” in the United
States56 and in the subsequent debate around the
world from the late 1960s. This crystallization of
Friedman’s framework was occurring when the
dominant thinking on monetary policy in the
United Kingdom was converging toward an almost
completely different framework.
THE ROAD TO RADCLIFFE
Friedman (1968c, p. 439) noted, “Experience
with monetary policy after World War II very
quickly produced a renewed interest in money
and a renewed belief that money matters.” But
later, viewing the 1950s and 1960s as a whole,
Friedman (1987, p. 13) concluded that the experience of the period “strongly reinforced” the
Keynesian critique of monetary policy, and
Friedman and Schwartz (1982, p. 17) argue that
55
See, e.g., Krugman (1998) and Svensson (2003).
56
Friedman (1977b, p. 12).
S E P T E M B E R / O C TO B E R , PA R T 2
2009
479
Nelson
the revival of the quantity theory of money did
not really take off until the 1960s. Evidently, the
1951 abandonment of cheap money policies was
not quite as great a breakthrough as Friedman and
other advocates of monetary policy had imagined.
What went wrong?
Friedman was, on the whole, pleased with
the course that monetary policy followed in the
United States during the 1950s. But even in the
U.S. case he was uneasy about the continuing
emphasis on fiscal and other nonmonetary influences when it came to accounting for economic
fluctuations; thus Friedman (1955, p. 32) referred
to “the intellectual climate of today and the recent
past, with its derogation of the significance of
monetary factors.” Furthermore, diagnoses of
inflation were becoming less orthodox with the
growing appeal from the mid-1950s of explanations that downplayed demand factors and instead
stressed “cost-push.”
In the United Kingdom, the trend of opinion
against monetary policy was even stronger. The
ending of cheap money was more hesitant; certainly interest rates were raised in 1951, but they
were cut in 1954 while the economy was gathering
steam. The really concerted tightening of monetary policy in the United Kingdom in the 1950s
was concentrated in the years 1955-58, which may
be why Friedman (1963b, p. 7) once said that the
U.K. cheap money period ended “a few years”
after 1951. The tightening began with increases
in Bank Rate in January-February 1955 under
Chancellor of the Exchequer R.A. Butler, and was
followed by further increases under his successors
in 1956 and 1957. The 1955-58 subperiod distinguishes itself from the preceding and subsequent
epochs by the extent to which the authorities
were attributed interest in control of the stock of
money. For example, Financial Times columnist
Harold Wincott contemplated what would happen
“if Mr. Butler continues with his policy of contracting the supply of money and credit” (Financial
Times, October 4, 1955), while the Financial
Times’s “Lombard” commentator said that the
“ultimate aim of the Government’s credit restriction drive is, of course, to exert a downward
pressure on the supply of money strong enough
to keep spending power within the limits of the
480
S E P T E M B E R / O C TO B E R , PA R T 2
2009
country’s available resources” (November 30, 1955;
emphasis added). In the September 1957 round
of tightening, policymakers themselves became
very explicit about their intention to restrict the
money supply, with speeches to that effect by
the prime minister (Harold Macmillan) and
Treasury ministers.
The emphasis on monetary aggregates at this
early stage may seem anomalous, as the official
money series (M0, M1, M3, and so on) that would
become familiar in later years were not available.
Many have noted that U.K. money supply data
were not available to Keynes when he wrote on
monetary affairs (see, e.g., Patinkin, 1976b), and
Walters (1970) conjectures that a historical series
for U.K. money was not put together until the early
1960s. One should not exaggerate the absence of
monetary data, however; the weekly release of
the Bank of England’s balance sheet gave most of
the information needed to construct currency and
monetary base series; and the various releases of
the clearing banks and other institutions provided
information on deposits. These releases were the
subject of regular attention in the financial press.57
The main problem for potential investigators of
monetary relations was constructing long series
free of breaks and double-counting. It was also
well known that the basic data for constructing a
long historical money series were available far
back for the United Kingdom; Friedman (1961b,
p. 270) referred to the availability of U.K. deposit
data back to the 1870s.
The would-be revival of monetary policy in
the United Kingdom suffered severe criticism
once the 1955 interest rate increases failed to
deliver the desired results during the year. The
Guardian’s financial editor had already claimed,
“It is now generally agreed that the experiment
of checking inflation by monetary policy alone
has not been a success” (Manchester Guardian,
December 12, 1955).
To many critics, the apparent failure of monetary policy to deliver low inflation vindicated
57
For example, for much of the 1940s and 1950s there was regular
space devoted in the Financial Times to the Bank of England’s
balance sheet release. Deposit data were discussed regularly too;
for example, in the Financial Times of November 30, 1955.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
the notion that monetary policy was ineffective
because aggregate demand was interest inelastic.
This notion, already embodied in the General
Theory to some extent,58 had been reinforced in
the United Kingdom in the prewar and early postwar period by surveys of firms carried out by
Oxford University. The survey results seemed,
as discussed, for example, by Schlesinger (1956,
p. 603), to vindicate the view that firms’ investment decisions were interest inelastic (and with
them the whole of aggregate demand, as most
Keynesian work had already narrowed the interest
rate channel to investment).
Friedman was scathing about the value of
questionnaires of businessmen. “That is not evidence…I do not care what they have said,” he said
on a panel in 1950.59 In 1979, Friedman added,
“Economics is a serious subject, and one of the
things we’ve learned in that subject is that if you
want to know how people behave, you don’t ask
them. You look.”60 The joint behavior of real
returns and the stock of productive capital led
Friedman to believe that investment was instead
“highly elastic” with respect to real interest rates
(Friedman and Schwartz, 1982, p. 494). Friedman’s
reaction to the survey results paralleled that of
his hero Dennis Robertson, who had said in 1949
that he had “a hunch that the reaction among the
neo-Keynesians against the importance of the
causal influence of the rate of interest on capital
outlay has been carried too far…Does anyone here,
I wonder, share my doubts—my very respectful
doubts—about the significance of those replies
to questionnaires?” (Robertson, 1949, p. 20).
Keynesians not only doubted the effectiveness
of monetary policy as a demand-control measure;
they argued that efforts to control inflation via
demand measures might in any case be misguided.
The Financial Times editorialized during the
initial tightenings that there was “still something
of a mystery about the origins of the inflationary
forces threatening the British economy”
(February 15, 1955). Despite their use of monetary
58
See, for example, the passage cited in Patinkin (1976a, p. 103).
59
Friedman, speaking in Wright (1951, p. 251).
60
In Anderson (1982, pp. 201-02).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
policy tightening, policymakers shared the view
that much inflation was cost-push in character, a
view evident in their repeated attempts to secure
an agreement with the private sector on wage
growth limits.
The skeptical sentiments regarding monetary
policy made by Keynesians since the 1930s were
synthesized and consolidated in the report on
monetary policy delivered by the Radcliffe
Committee to the U.K. government in August 1959.
The Radcliffe Report argued that monetary policy
was ineffective. It did not rely on the liquidity
trap argument but used lines of reasoning that
delivered as complete an ineffectiveness result
as that associated with the liquidity trap. Whereas
Keynes’s liquidity trap argument said that money
and government securities could become equivalent at a low interest rate, the Radcliffe Committee
argued that important asset prices were unaffected
by open market operations that switched money
for short-term securities, even when these operations changed the short-term interest rate. Thus
while open market operations could alter policy
rates, they affected only the composition, not the
aggregate, of “liquidity,” which was the financial
quantity that really mattered; financial innovation, it was argued, had eliminated much of the
difference between money and Treasury bills.
The interest rates that policy could affect, the
Committee argued, mattered negligibly for
aggregate demand, while the asset prices that
did matter for aggregate demand depended on
the “liquidity” total, which was generally not
susceptible to central bank manipulation. The
Committee also endorsed cost-push views of
inflation.61
Friedman and Schwartz (1982, p. 52) observe
that the Radcliffe Committee was “faithful to
Keynes” in emphasizing the ineffectiveness of
monetary policy arising from the alleged equivalence of money and securities. But the Committee,
by claiming that monetary policy was ineffective
generally, not just in Depression conditions, was
taking a harder-line position than Keynes usually
did. The Committee’s basis for this conclusion
61
See Radcliffe Committee (1959); and for references to the relevant
sections, see, e.g., Laidler (1989) and Nelson (2009).
S E P T E M B E R / O C TO B E R , PA R T 2
2009
481
Nelson
was that financial innovation—for example, the
growth of nonbank intermediaries—put the determination of important asset prices outside the
reach of monetary policy. While the emphasis on
financial innovation did not have a counterpart
in Keynes’s General Theory, it paralleled the
approach that Gurley and Shaw (1960) were taking in analyzing U.S. financial behavior. Noting
the connection, Friedman and Schwartz (1970)
treated Gurley-Shaw and Radcliffe as advocates
of the same “liquidity” argument, while Friedman
and Schwartz (1982, p. 209) referenced Radcliffe
and Gurley-Shaw together when citing studies
that minimized the significance of money and
monetary policy actions. Monetarists were not
impressed by the Radcliffe/Gurley-Shaw arguments from the beginning, and the monetarist
side of the argument was what—eventually—
won the day in the economics profession. Brunner
(1985, p. 22) observed witheringly that there really
was “no logical link between negative conclusions
bearing on monetary policy, and the discussion
of financial innovations…Gurley and Shaw
argued more than 20 years ago that the explosive
growth of savings and loans associations erodes
the potency of monetary policy. The subsequent
evolution discredited such fears or hopes.”
Friedman’s initial public response to the
Radcliffe Report was muted. Alvin Marty thanked
Friedman for “exceedingly helpful substantive
comments” on a paper published in early 1961 in
which Marty said the “Radcliffe Report is a striking
example of failure to offer a shred of evidence.”62
In a book review published at the end of 1961,
Friedman (1961a, pp. 1052-53) noted the problems
in defining “liquidity”; in 1964, he said that the
Radcliffe Committee’s liquidity concept was
“an undefined term which covers the universe,”63
while Friedman and Schwartz (1970, p. 130)
added that the Radcliffe Committee itself could not
settle on a firm liquidity definition.64 Friedman
denounced the theories offered by the Radcliffe
62
Marty (1961, pp. 56, 59).
63
Friedman (1964a; p. 73 of 1969 reprint).
64
The critical discussion of the Radcliffe Committee in Friedman
and Schwartz (1970) was originally part of their unpublished first
draft of their Trends study (Friedman and Schwartz, 1966).
482
S E P T E M B E R / O C TO B E R , PA R T 2
2009
Committee as “a false trail that will not in fact be
productive” (Friedman, 1963b, p. 9) and went on
to applaud the negative reception the Radcliffe
Report had received among “academic economists
and others.”65 Friedman’s appraisal that the Report
had been received negatively rested largely on
the U.S. reaction; in the United Kingdom, many
policy and academic figures greeted the Report
favorably.
In U.K. policymaking, confidence in monetary
policy restriction as the cure for inflation reached
its peak in 1957 and was followed by a period of
substantial monetary policy easing. Friedman
(1980b) argued that the shift to expansionary
policies (both monetary and fiscal) in the United
Kingdom from the later 1950s onward was a vindication of his (1954) prediction. He had predicted
that overreaction to actual or prospective minor
recessions would produce a tendency for the
authorities to overexpand. Complementing this
explanation is the fact that the U.K. authorities
after 1957 were much more inclined to view
incomes policies as the appropriate means of
fighting inflation. Even when inflation fell in the
late 1950s, to the point of delivering price stability
in 1960, the success was attributed to favorable
cost-push shocks rather than to the 1955-58
restrictive monetary policy. For example, The
Economist (August 29, 1970) attributed the fall
in inflation from 1958 to 1960 to less-militant
union behavior after the defeat of the 1958 London
bus strike.
1959-1970
Events
In a new round of monetary policy tightening
in July 1961, the Macmillan government raised
Bank Rate to 7 percent. This was not an unambiguous affirmation of the role on monetary policy,
because it was muddied by a simultaneous
attempt by the government at a wage freeze (a
“pay pause”). But after a year of tight monetary
policy, the Chancellor of the Exchequer, Selwyn
Lloyd, showed signs of determination to maintain
65
Friedman (1964a; p. 73 of 1969 reprint).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
Figure 2
U.K. M2 and Monetary Base Growth (1948-75, annual data)
Percent
30
Base Money Growth
25
M2 Growth
20
15
10
5
0
–5
–10
1948
1951
1954
1957
1960
a restrictive stance: “18 months ago there was
excess demand…Now, I think there is a measure
of disinflation…I think the economy is in better
shape, but you can’t have disinflationary measures
without there being, in result, a measure of disinflation…That means that some order books will
be shorter” (Yorkshire Post, July 11, 1962). Shortly
afterward, Lloyd was fired. The restrictive monetary policy episode turned out to be only an interruption in the de-emphasis on monetary policy
signaled by the Radcliffe Report. The expansionary
policies prevailing before 1961 were revived in a
more-intense form. The more-expansionary policy
was associated with a shift up in money growth,
whether measured by the monetary base or by
Friedman and Schwartz’s (1982) M2 measure,
reversing the moderation in growth observed
during the 1961 squeeze (Figure 2). Consequently,
while the pickup in U.S. money growth in the
1960s initially exceeded that in the United
Kingdom, Friedman and Schwartz (1982, p. 157)
note that, from the mid-1960s, “the United
Kingdom took the lead—if that is the right word.”
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
1963
1966
1969
1972
1975
Friedman viewed the U.K. monetary expansion of the 1960s as contributing to the mood of
the country. “[T]he fact is that most people enjoy
the early stages of the inflationary process. Take,
for example, Britain in the Swinging Sixties”
(The Listener, April 24, 1980). One aspect of the
sustained expansionary policies of the 1960s that
did cause concern to the U.K. authorities was one
that Friedman would prefer they had been sanguine about: the threat to the exchange rate. The
sterling/dollar rate was becoming more difficult
to maintain, even with the exchange control apparatus, and speculation against it increased after
the Labour Party under Harold Wilson was elected
to office in 1964. Friedman (1965, p. 179) said
that while he happened to disagree with the policies Wilson had promised to carry out, he found
it objectionable that foreign central banks and
other holders of sterling were perceived as having
a “veto power” over their implementation. He
elaborated in September 1965 that this meant
“that British internal policy was shaped by officials who were not responsible to the British
S E P T E M B E R / O C TO B E R , PA R T 2
2009
483
Nelson
electorate.”66 Friedman concluded that Wilson,
instead of negotiating a sterling rescue, should
have floated the pound on coming to office, blaming the predecessor government for the likely
depreciation.67
The Wilson government finally did devalue
the pound sterling in November 1967, with the
14 percent adjustment providing another example of what Friedman (1969, p. 20) called “this
awful business of holding and holding and holding to the last gasp and then having to make a big
change.” Friedman had anticipated the devaluation and had wanted to speculate $30,000 against
the pound, only to find that his Chicago banks
did not have the wherewithal to carry out the
foreign exchange transaction (Sydney Morning
Herald, October 9, 1986). After the devaluation,
the Wilson government started expressing policy
commitments to the IMF in terms of quantitative
financial targets, and in 1969 it announced a target for domestic credit expansion (DCE). Although
interpreted as a concession to monetarists, the
DCE targets had the decidedly un-monetarist
implication of encouraging the authorities to
regard money base growth that came from balance
of payments surpluses as “good.” The policy
framework of a fixed exchange rate, attention to
DCE at the expense of the aggregate monetary base,
and incomes policies, contrasted with Friedman’s
recommendation of a sterling float, no incomes
policy, and direct control of the aggregate base or
aggregate bank reserves.
Issues
Monetarism and the Quantity Theory. It is
difficult to convey the dramatic shift in the
amount of coverage given to monetary policy in
the U.K. financial press and political debate in
the years 1968 to 1970 compared with the preceding three years 1965 to 1967. The increased
degree of coverage turned out to be permanent.
Some flavor is captured by the observation of the
magazine Management Today (August 1976):
66
From Friedman’s 1965 Mont Pelerin Society meeting remarks,
published in Friedman (1968a, p. 274).
67
See, e.g., his remarks in Friedman and Roosa (1967, pp. 114-15)
and his 1968 memorandum in Friedman (1988).
484
S E P T E M B E R / O C TO B E R , PA R T 2
2009
“A decade ago economists, pundits and politicians alike would have been amazed to learn that
in the mid-1970s debates over monetary policy
would have come to dominate the national and
international economic scene.”
The upsurge in the coverage of monetary
policy was accompanied by greatly increased
discussion of Friedman. While “Chicago School”
was probably the most widely used term to
describe the school of thought emphasizing money
(e.g., The Sun, October 7, 1968; The Observer,
April 20, 1969), and “Friedmanite” was used from
an early stage (e.g., Sunday Times, November 10,
1968), the U.K. debate also rapidly proliferated a
term that The Economist had used as early as 1963,
but which was starting to become prevalent in
the United States: “monetarist.” Robert Solow
used “monetarism” repeatedly in an article he
wrote for The Times (December 23, 1968), and
Paul Samuelson criticized “crude monetarism”
in a contribution to the Sunday Telegraph
(December 15, 1968).
Friedman often publicly criticized the terms
“monetarist” or “monetarism.” In an interview
with The Times in 1976, Friedman said, “It is not
a new position, and that is one of the reasons I
don’t like the word monetarism” (The Times,
September 13, 1976).68 As Friedman saw it, he
was not launching a new theory but bringing quantity theorists’ work “down to date,”69 so that it
could be applied to the problems of the “bad old
present.”70 But Friedman (1978) confessed that
there was utility to the term monetarism because
there were some elements of older quantity theory
work that he and other monetarists had discarded.
In particular, an aspect of earlier quantity theory
analysis that Friedman explicitly rejected was
regarding velocity behavior as the outcome of an
institutionally determined payments process,
68
Anna Schwartz has suggested that another reason for Friedman’s
reservations about the word “monetarism” is that “I think he attributed it to Karl Brunner, who was not really a master of English.”
(Conversation with author, trip to Vermont Great Inflation conference, September 25, 2008.) Brunner is typically credited with the
term monetarist or monetarism, but both terms predate Brunner’s
usage of them (see Laidler, 2001; and the Oxford English Dictionary,
1976 and online editions).
69
Friedman (1972d, p. 12).
70
Friedman speaking in The Guardian, September 16, 1974.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
instead of as the result of decision problems by
households (e.g., Friedman, 1956, point 11;
Friedman, 1963b, p. 10; Friedman and Schwartz,
1982, pp. 38, 40, 62). Friedman himself used
“monetarism” and “monetarist” in his address
at the University of London (1970a). He became
accustomed to using these terms readily and without prompting, including in correspondence and
conversation.
On some occasions Friedman also associated
the older quantity theory with a further retrograde
aspect, namely the assumption of price flexibility,
or, more fairly, of not having a firm description of
the adjustment process of prices to money (e.g.,
Friedman and Schwartz, 1982, p. 44). But he
generally qualified this case by describing it as
the “simple quantity theory” (e.g., Friedman and
Schwartz, 1982, pp. 59, 398). Further consideration by Friedman during the 1970s of the quantitytheory literature had the effect of leading him, if
anything, to attribute more to the older writers.
In particular, Friedman was struck by how explicit
David Hume’s writings had been on the role of
expectations: “David Hume has a statement somewhere about the fact that an increase in the quantity of money stimulates economic activity only
so long as it keeps on increasing and people don’t
expect it” (The Times, September 13, 1976).
Friedman decided that while he, and Lucas and
Sargent after him, had expressed the point more
formally, the expectational Phillips curve idea was
due to Hume (Friedman, 1975b). While Patinkin
(1972b) claimed that Hume did not believe in a
long-run vertical Phillips curve, the explicit quotations given in Friedman (1975b) support the
crediting of this idea to Hume, and Friedman’s
interpretation was more recently reaffirmed by
Mankiw (2001) in his study of the same Hume
passages.
Friedman further credited Hume with the
demand-for-money perspective on the quantity
theory that Friedman had used (Friedman and
Schwartz, 1982, p. 621). So whereas Friedman
(1968c, p. 433) attributed to Hume the “broad
outlines of the quantity theory,” by the early 1980s
he was crediting Hume with both the aggregate
demand and aggregate supply aspects of his own
framework and so, he said, Hume was the true
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
originator of monetarism. Appropriately enough,
it was on British television that Friedman said:
“I really would like to make clear that the doctrines I proclaim are not original with me by any
means, in fact if I have to find a source for them
they are [from] David Hume.”71
In their interventions in the U.K. debate,
Samuelson and Solow argued that monetarism
was not making valid points about monetary relations that U.S. Keynesianism had not long since
incorporated. Samuelson, for example, said:
“Money was, so to speak, ‘rediscovered’ in my
country around 1950…When Professor Friedman
formulates his system in generality…it coincides
with the post-Keynesianism of the TobinModigliani type.” But the record does not support
the denial of Friedman’s contributions, nor
Samuelson and Solow’s confidence that U.S.
Keynesianism circa 1968 had incorporated the
role for monetary policy adequately. As we have
seen, Friedman’s emphasis on money predated
1950, and his elaboration of it incorporated a
general transmission mechanism not covered in
Keynesian work; in particular, more than one
interest rate, and (in contrast to positions taken
by Samuelson and Modigliani) a sensitivity of
consumption (not merely investment) to interest
rates.72 Furthermore, Solow and Samuelson in
both the 1960s and 1970s disputed Friedman’s
expectational Phillips curve analysis, contesting
both its long-run vertical property and its exclusion of a systematic role for cost-push factors.
In well-known lectures given in Manchester, for
example, Solow (1969) argued that cost-push factors mattered greatly in both the United Kingdom
and the United States, that demand factors barely
mattered at all for U.K. inflation, and that the U.S.
Phillips curve was permanently nonvertical.
Some influence of the money supply debate
was felt in what policymakers said: Chancellor
71
Free to Choose, BBC2 debate, March 22, 1980, p. 15 of transcript.
Likewise, at a press conference in Wellington, New Zealand, in
1981, Friedman said, “What is called monetarism, the quantity
theory of money, was developed by David Hume in the eighteenth
century. It is not ‘my’ theory—I have no patent on it” (Evening Post,
April 27, 1981).
72
See Blinder (1986) and Modigliani (1986) for characterizations of
the pre-monetarist Keynesian view as one that denied the interest
sensitivity of consumption.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
485
Nelson
of the Exchequer Roy Jenkins stated, “I attach the
greatest importance to monetary policy” (House
of Commons Debates, April 15, 1969, p. 1007);
and The Economist (April 18, 1970) referred to
the “new importance attached to monetary policy
in Britain.” As noted above, however, the changes
actually made in the macroeconomic policy framework were not truly in the direction Friedman
wanted. Indeed, at the U.K. Treasury, skepticism
prevailed from top to bottom about the attention
being given to Friedman. At a junior level, Treasury
economist Stephanie Edge spoke in favor of the
power of fiscal policy and criticized Friedman’s
findings to the contrary: “The idea that simple
one-equation comparisons can reveal anything is
one that should be vigorously attacked” (Edge,
1967, p. 205). At a senior level, Treasury adviser
Alec Cairncross wrote in his diary of October 6,
1968, that “the English press (and especially The
Times) was making such a fool of itself over Milton
Friedman” (Cairncross, 1997, p. 327).
Cairncross’s reference to The Times concerned
the articles being written by its economics editor,
Peter Jay. Jay was initially regarded with suspicion
on each side of the money supply debate as a
sympathizer with the other side. But further articles by Jay brought him out as a supporter of the
monetarist arguments, and Jay later identified
himself among “those…who began to advocate
proper control of the money supply from the
late 1960s” (Independent, September 23, 1991).
Friedman and Jay became good friends, appearing
together on several episodes of both the U.S. and
U.K. versions of the Free to Choose television program in 1980. In one of these programs, Friedman
said that he and Jay “are in almost complete agreement on the desirable monetary policy.”73
Another journalist, Samuel Brittan, serves as
an illustration of Friedman’s observation that
“accidents play an enormous role in mankind’s
experience.”74 Friedman happened to be visiting
Cambridge University while Brittan was an undergraduate student there, and Brittan happened to
73
Free to Choose, PBS debate, Episode 3, pp. 16-17 of PDF transcript.
74
April 16, 1996, talk by Friedman at Claremont College (broadcast
C-SPAN, December 26, 1996).
486
S E P T E M B E R / O C TO B E R , PA R T 2
2009
have Friedman assigned to him as his tutor. Brittan
joined the Financial Times after graduation.
Brittan considered the arrival of monetarism the
“most interesting event for a very long time in
the realm of economic ideas” (Financial Times,
January 8, 1970). A long article by Brittan, headed
“MONEY SUPPLY: The Great Debate” in an enormous font, appeared in the October 25, 1968, edition of the Financial Times, discussing Friedman’s
views and covering his American Economic
Association presidential address (Friedman,
1968b). Brittan and David Laidler, who had been
a Ph.D. student of Friedman’s at Chicago and from
the late 1960s was at Manchester University,
became two leading voices of monetarism in the
United Kingdom during the 1970s. Cobham (1984)
notes that although Laidler departed for Canada
in 1975, he remained prominent in the U.K. debate
during the second half of the 1970s. Neither Brittan
nor Laidler was an echo chamber for Friedman,
and both disagreed with him in print, but they
both had firsthand knowledge of his positions.
As Friedman observed, “You have the interesting
phenomenon that whereas David Laidler came
to Chicago, Chicago came to Sam Brittan.”75
Alan Walters, an academic and financial consultant who had been undertaking U.K. analogues
of some of Friedman’s empirical work since the
early 1960s, wrote to Friedman at the end of the
decade to let him know that he was close to being
a household name in the United Kingdom.76
The Beginning of Monetary Trends. In
November 1966, Friedman and Anna Schwartz
completed a draft of their manuscript on monetary
trends, concerned solely with U.S. data, and submitted it to the NBER review process. Friedman
(1955, p. 30) had written about the desirability
of studying U.K. monetary data, and, somewhere
along the line, Friedman and Schwartz elected
to cover U.K. data in the revised version of their
Trends manuscript, although it was not a change
specifically requested by the NBER. A major
obstacle, the construction of historical data, was
75
Friedman, speaking in September 1974, in Institute of Economic
Affairs (1974, p. 102).
76
Letter from Alan Walters to Milton Friedman, December 4, 1969,
Friedman office correspondence (uncatalogued as of end-2007).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
The Conservative Party, led by Edward Heath,
won the U.K. general election of June 1970.
Friedman had not met Heath (The Listener,
February 11, 1971). There is nevertheless evidence
of an influence of Friedman’s writings on Heath’s
statements. Heath’s introduction to his 1970 party
platform said, “[O]nce a policy is established, the
Prime Minister and his colleagues should have
the courage to stick to it. Nothing has done Britain
more harm in the world than the endless backing
and filling which we have seen in recent years”
(The Guardian, May 27, 1970). It is possible that
the drafting of this passage was influenced by
Friedman and Schwartz’s (1963, p. 289) characterization of the Federal Reserve’s history as one
of “so much backing and filling, so much confusion about purpose…”
There was less indication that the new government would be influenced by the Friedman-
Schwartz work when it came to monetary policy
formulation, as Anna Schwartz discovered on a
visit to the United Kingdom very soon after the
election result. Writing to Friedman about her
meetings with U.K. academic economists and
Bank of England officials, Schwartz reported:
“Much talk generally of what difference the
Conservative Government would make for monetary policy…[The Bank officials] didn’t see that
it would make any difference. Apparently, Bank
policy is perfect.”79
On the matter of the government’s role in the
market, the prospect seemed more favorable that
U.K. economic policy would go in a direction
favored by Friedman. Friedman said in 1971,
“My own personal view is that…the most effective
road to development is through free enterprise
and private investment, and that the government
can serve best by limiting itself to essential government functions.”80 In the same year, the Heath
government objectives were laid out by the
Chancellor of the Exchequer, Tony Barber, in a
form that seemed in keeping with Friedman’s
views: “Our object is to lessen government interference and reduce government subsidies; to extend
the opportunities for profitable enterprise; to
widen the area within which industry rather than
government will make decisions” (Dallas Morning
News, February 9, 1971). Friedman expressed
cautious approval, observing that the United
Kingdom had “potential, but only if you could
by some miracle get rid of the enormous mass of
controls, interventions, welfare-state measures
and so on…Heath has been moving somewhat in
that direction” (Vision, April 1972).
The Heath government moved away from
free-market policies during 1971 and made the
break more explicit with the passage of the
Industry Act 1972. The act introduced extensive
subsidies to private investment, contrary to
Friedman’s dictum, “Capital investment that has
to be subsidized is not worth having” (Wall Street
Journal, February 12, 1997).
77
Friedman, letter to David K. Sheppard, July 10, 1967, Friedman
office files (uncatalogued as of end-2007).
79
78
July 2, 1970, letter from Anna Schwartz to Friedman. The Hoover
Archives’ copy of this letter is in Box 91, folder 7, of the Friedman
papers.
October 6, 1969, testimony, in Joint Economic Committee (1970,
p. 826).
80
Friedman (1971a, p. 847).
partly overcome when David Sheppard, a Harvard
Ph.D. graduate shortly to return to the United
Kingdom, contacted Friedman to let him know
of his work in the area. In his reply to Sheppard,
Friedman indicated that U.K. data were being
incorporated into his new volume with
Schwartz.77 He and Schwartz thereafter used the
Sheppard data on money. Anna Schwartz used
the data in a 1969 paper for U.K. audiences
(Schwartz, 1969), and Friedman referred fleetingly to the Trends project during Congressional
testimony in October 1969, where he said he had
been “working on some British data which go
back a century. They show the same relation
[as in the United States].”78 Considering this
energetic start in the late 1960s, Friedman and
Schwartz surely could not have imagined that
Monetary Trends would not be finished until
the early 1980s.
1970-1979
Events
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
487
Nelson
The change in direction was felt also in the
Heath government’s policies against inflation.
The government started cutting Bank Rate in
April 1971. It had consistently seen inflation as
a nonmonetary problem and in November 1972
imposed wage-price controls.
Also in 1970, Friedman had his first published
exchange with a U.K. critic of his work. The
Radcliffe Report had given Friedman a heads-up
about the skepticism regarding monetary policy
prevailing in the United Kingdom. Along with
Richard Kahn, Nicholas Kaldor was regarded as
a major academic influence on the Report. Shortly
before Friedman’s visit, Kaldor (1970) restated the
U.K. anti-monetary policy position and used a
reverse causation argument to dispute Friedman’s
findings. If, Kaldor argued, the authorities actually
undertook measures that delivered them control
of the money stock, their actions on money would
face permanent, completely offsetting movements
in velocity. Historical relations between money
and other variables, according to this argument,
simply reflected reverse causation—the passive
creation of money in response to price and output
movements. Price and output behavior would, it
was argued, have been no different if the monetary
authorities had somehow prevented this money
creation from taking place.
The way that Friedman (1970c) answered
Kaldor was by appealing to the fact that money
had been connected to income and inflation under
many different monetary arrangements, undermining an explanation of the correlation like
Kaldor’s that relied on the existence of a particular
policy regime or on the institutional conditions
prevailing in the United Kingdom.
The Bretton Woods system collapsed during
the early 1970s, despite what Friedman said was
officials’ belief that “they can put Humpty Dumpty
together again.”81 The London Evening Standard’s
financial columnist blamed the foreign exchange
market turmoil on the “incredible influence of
economist Milton Friedman,” charging that
Friedman’s theories had discouraged international
coordination of policies (May 5, 1971). Though
81
September 23, 1971, testimony, in Joint Economic Committee
(1971, p. 699).
488
S E P T E M B E R / O C TO B E R , PA R T 2
2009
he would surely have liked to accept the credit
for the advent of floating rates, Friedman concluded that his advocacy of flexible rates had had
“absolutely no effect,” and that it was instead the
“brute force of events” that had forced governments to realize that fixed exchange rates were
untenable (The Listener, April 27, 1978). The
pound sterling began floating in June 1972.
In early December 1972, Friedman learned
that a debate had been taking place in the London
Times about his 1967 American Economic
Association presidential address. His statement
there (Friedman, 1968b, p. 11), that “full adjustment” to a shift in the inflation rate takes “a couple of decades,” had been interpreted by one critic
as implying that removing inflation would take 20
years of above-normal unemployment. Friedman
wrote a letter to The Times (dated December 6 and
published December 12) to clarify that by “full
adjustment” he meant resettling at the steady state.
“The important point is that while ‘full’ adjustment may well last several decades, the period
of unusually high unemployment is far shorter,
more like two to five years.” Around the same
time, in a submission to the U.S. Congress Joint
Economic Committee, Friedman had occasion to
convey his opinion of the U.K./France Concorde
project. That project had reached fruition partly
from the injection of funds from the U.K. government. The result was an air service that took
hours off of intercontinental travel, but only for
the elite class of customers who could afford the
ultrahigh ticket prices. Friedman (1958a) had
criticized the involvement of the government
sector in the creation of “monuments” that did
not raise ordinary living standards, and his
Congressional submission (dated December 11,
1972) urged that the United States government
not follow the Concorde precedent by subsidizing
a U.S. supersonic transport (SST). “A governmental decision to produce an SST largely at its
own expense is a step toward socialism and away
from free enterprise.”82
Though the preceding two items do not appear
in his published bibliographies, they could easily
have become the last things Friedman ever wrote
82
In Joint Economic Committee (1973a, p. 81).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
for publication. On December 15, Friedman had
open-heart surgery. The surgery was successful
and Friedman was discharged on December 26
(Kansas City Times, December 27, 1972), but
Friedman lost considerable weight in the wake
of the surgery, and his family medical history was
inauspicious. As Anna Schwartz has observed,
“Who would have thought at that time that
Friedman would live on to ninety-four?”83
From the United States, Friedman criticized
U.K. monetary policy developments during 1973,
observing in Congressional testimony that “defective as our policy has been, it has been less erratic,
more moderate than the policy of most other leading countries.”84 He observed later in the year that
Heath had gone from what Friedman perhaps too
generously called a “tight money policy” to “a
policy of stimulating…Now you have prices rising
in Britain at a rate of something over 10% a year”
(Friedman, 1973b, p. 33). The British magazine
Management Today pondered Heath’s and
Friedman’s contrasting diagnoses of the inflation
problem in its August 1973 issue: “But is the entire
phenomenon of unusually rapid and apparently
ineradicable inflation new in itself?,” its editorial
asked. “Is it a different variety, considerably more
virulent, of the disease to which Western society
has been susceptible for many decades past? The
temptation, of course, is to say that it is: to blame
union militancy…But economic historians half a
century hence may well not be impressed by this
argument. To them, the inflation will probably
seem a classic case of monetary inflation, engendered by the usual process of overproduction of
liquid currencies.”
The editorial just quoted was something of
an outlier in the general U.K. discussion in 1973.
The tendency to consider inflation to be nonmonetary in character intensified in the wake
of the oil and other commodity price shocks of
that year. Wherever he went during the 1970s,
Friedman found himself having to explain the
fallacy inherent in special-factor explanations
83
Remarks of Anna Schwartz to author, New York office of NBER,
May 27, 2008.
84
Friedman, June 21, 1973, testimony, in Joint Economic Committee
(1973b, p. 120).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
for inflation. “Arithmetic is one thing and economics is a very different thing,” was how
Friedman put it in one appearance (Meet the Press,
November 12, 1978). “The great confusion in this
area is to confuse particular prices with prices in
general.” In 1979 Friedman, perhaps near his wit’s
end, reiterated: “OPEC does not cause inflation;
no, sir.”85
What Friedman (1975a, p. 137) called the
U.K.’s “major economic crisis in early 1974,” with
U.K. inflation passing 15 percent, culminated in
an election that returned Harold Wilson to power.
Friedman noted that the recent U.K. elections
had helped refute the claim that governments
do not lose elections because of high inflation.
“Inflation surely helped to make Mr. Edward
Heath Prime Minister in 1970,” Friedman (1974,
p. 44) observed, “and, even more surely, ex-Prime
Minister in 1974.”
The centerpiece of the Wilson government’s
incomes policies measures was intended to be
its “Social Contract” agreement with the unions.
Friedman said that “the so-called Social Contract…
gives people a false impression of both the causes
of inflation and the way to cure it” (BBC2,
November 9, 1976). The Social Contract would
do “no good whatsoever as long as they continue
to run the printing press” (Friedman, 1975d, p. 20);
and, if money growth was slowed, the Social
Contract would be seen as having been successful,
even though the reduction in inflation would be
the same without the Contract (Newsweek,
September 20, 1976).
Friedman paid a one-week visit to the United
Kingdom in September 1974. Reflecting on his
visit a few months later, Friedman was particularly struck by the continuing popularity of the
wage-push explanation. “In Britain, the explanation that everybody gives for inflation is that
inflation is caused by trade unions, the greedy
grasping laborers who force up the wages that
cause inflation” (Friedman, 1975d, pp. 5, 7).
During his visit Friedman had written to The
Economist saying he had “been dismayed, even
in my few days in London, at the widespread
85
May 17, 1979, testimony, in Committee on the Judiciary (1980,
p. 154.)
S E P T E M B E R / O C TO B E R , PA R T 2
2009
489
Nelson
support of ‘union bashing’ as a way to attack
inflation” (The Economist, September 28, 1974).
While Friedman argued in 1978 that there
was “almost no one who any longer has a good
word to say about nationalization,” he had to
admit that in the United Kingdom the trend had
been to extension of nationalization (The Listener,
April 27, 1978). Indeed, Harold Wilson’s account
of his 1974-76 term in office (Wilson, 1979, p. 35)
includes a six-line list of the nationalization plans
his government outlined in 1974, and the Labour
government did proceed to extend nationalization to the shipbuilding industry. With Thomas
Balogh, one of Friedman’s Keynesian opponents,
as one of the ministers responsible, the Department
of Energy also increased the government’s stake
in the oil industry—leading to Friedman’s observation, “You have been nationalizing North Sea
oil” (The Listener, April 27, 1978). Some of the
oil-industry nationalization was wound back in
the austerity measures of late 1976, when the
Callaghan government announced the sale of
part of its interest in British Petroleum.
The U.K. private corporate sector over this
period was suffering a pronounced squeeze.
“Great Britain had a much more severe financial
crisis than we did,” Friedman observed.86 The
stock market experienced a major collapse, its
index standing in 1974 at its late 1950s’ value
(Bordo and Wheelock, 2004), and the U.K. longterm corporate bond market virtually disappeared
in the second half of the 1970s. Friedman noted
that a “proper climate” for growth required
“investment, enterprise, the ability to borrow
capital” (Dallas Morning News, October 17, 1975),
but in the U.K. case he observed that “the domestic capital markets are so disorganized by erratic
inflation, excessive taxation, and government
intervention” (Newsweek, December 27, 1976).
It was, however, foreign exchange market
turmoil that led to the United Kingdom negotiating
a loan from the IMF in late 1976. The Callaghan
government, including Chancellor of the
Exchequer Healey, appealed to the stringent terms
of the loan as the reason it had to undertake cuts
in government expenditure. Friedman maintained
that the government’s recourse to the IMF was
just a charade: “Your government has gone to the
IMF so that they can lay down rules for the management of your economy…It’s like the way big
corporations use management consultants. The
corporations know perfectly well what must be
done, but they want to blame the unpleasant remedies on someone else” (Daily Mail, September 30,
1976). No doubt there was a considerable element
of validity to this conjecture, as senior members
of the Callaghan government had indeed accepted
the need to shift the division of resources between
the public and private sectors. But Friedman went
too far with his further claim, “The British government knows that the only way to stop inflation is
for government to spend less and to create less
money” (Newsweek, October 11, 1976). This claim
attributed, yet again, acceptance of a monetary
view of inflation to the government. Such an
acceptance is not supported by the record of U.K.
policymakers’ views or behavior; on the contrary,
the government continued to point to the Social
Contract as a central part of its fight against inflation and to claim that monetary policy alone could
not defeat inflation. A wage-push view of U.K.
inflation continued to dominate, and the government saw monetary targets—the first publicly
announced target was for the financial year starting in April 1976—as a means of helping to avoid
adding demand-pull to the wage-push pressures.
Friedman subsequently pulled back from his late
1976 claim that the authorities knew inflation was
a monetary phenomenon. In a November 7, 1977,
talk, Friedman said, “If you listen to anybody
telling you about Great Britain’s plight, they will
tell you that the real problem in Great Britain is
that you have such strong trade unions, that they
push up wages and that causes inflation.”87
When the government announced its budgetary program in the wake of the IMF loan negotiation, Friedman pronounced himself unimpressed,
pointing to the modest nature of the public expenditure cuts, the use of devices such as asset
sales, and the failure to cut tax rates (Daily Mail,
December 17, 1976). But the U.K. government’s
expenditure did fall substantially after 1976 as a
86
87
Milton Friedman Speaks, Episode 5, p. 23 of transcript.
490
S E P T E M B E R / O C TO B E R , PA R T 2
2009
Milton Friedman Speaks, Episode 6, p. 6 of transcript.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
share of output. Acknowledging this, Friedman
said in February 1978 that a “rather curious reason
for hope” was the fact that a “Labour Government
for two years in a row has been led by political
pressures to reduce government spending as a
fraction of income.”88
Another source of hope for Friedman was a
speech that Prime Minister Callaghan made on
September 28, 1976. The speech, written with the
assistance of Peter Jay (who, as well as being an
advocate of money supply control, was Callaghan’s
son-in-law), was widely interpreted as signaling
a repudiation of fine-tuning and Keynesian
demand management, and invoked elements of
the natural rate hypothesis. Friedman quoted the
speech in his Newsweek column (December 6,
1976) and in his Nobel lecture given in December
1976 (Friedman, 1977a). Truth to tell, Friedman
cited the speech excessively89 and exaggerated
its significance. The speech was not so dramatic
a break with the past. The fact is that there were
many occasions since the 1950s when prime
ministers had talked about inflation moving up
together with unemployment and on the danger
of overstimulating the economy. Callaghan’s
speech sidestepped the greater problem with U.K.
policymakers’ outlook on inflation, namely their
appeal to nonmonetary explanations.
Friedman appeared on BBC television in late
1976, in a studio debate taped in Chicago with
former Wilson government adviser and minister
Thomas Balogh. In the debate, Balogh said, “I
think that the Professor really is terribly naïve.”
Friedman responded, “Well, I may be naïve but
let me point out first that Mr. Balogh is simply
defending his own record. Britain is in the position
that it is because it listened to his advice and the
advice of people who believe the way he does.”
Friedman emphasized that he was not referring
only to the Labour governments with which Balogh
had been affiliated, but to postwar governments
of both parties, which he said had “generally followed very much the same policies...I am trying
88
Milton Friedman Speaks, Episode 7, pp. 20-21 of transcript.
89
For example, in Free to Choose (both the television and book versions), in Milton Friedman Speaks, in Friedman (1992, 1997), etc.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
to argue against the general drift that has affected
both parties” (BBC1, December 6, 1976).
“That great prophet of monetarism, Milt [sic]
Friedman, is coming to Strathclyde University
[in Glasgow, Scotland] in April to lecture on inflation,” observed The Scotsman’s business columnist at the start of 1978 (January 25, 1978). In April
1978, Friedman, now 65 and sometimes describing himself as retired, duly appeared in the home
city of Adam Smith to give a lecture and press
conference. At his Glasgow appearances, Friedman
qualified his praise for the Callaghan government’s
reduction in the ratio of government spending to
GDP with criticism of its extensions of government
intervention in the marketplace. He also pronounced himself unimpressed by the practical
changes made in U.K. monetary policy. “In Britain,
monetary targets have been adopted but have not
been kept. Mr. Callaghan has said there will be no
fine-tuning, but Mr. Healey has been fine-tuning”
(The Scotsman, April 22, 1978).
Another aspect of the U.K. policy framework
that was anathema to Friedman was the continued
proffering of incomes policies—or in Friedman’s
(1976b, p. 233) blunt characterization, “general
price or wage controls, euphemistically referred
to as ‘incomes policies’”—as a part of the government’s anti-inflation strategy. Friedman had
observed early in the U.S. wage/price control
experiment (Newsweek, January 31, 1972):
“Experience in other countries [beside the United
States] suggests that for about a year such controls
generally look good; after about two years, they
collapse.” The incomes policies put in place in the
United Kingdom from 1972 to 1979 fell roughly
into the pattern Friedman described. Heath’s wageprice controls imposed in 1972 suppressed inflationary pressure for about a year before a breakout
at the end of 1973 and in early 1974. The Social
Contract of the Wilson government was largely
violated until a more legally binding version was
introduced in July 1975. U.K. inflation then generally declined for three years (1975-78), not two.
But monetary policy had been tightened in late
1975 and over 1976; it was only from early 1977
that the government’s incomes policies were
attempting to push inflation away from the direction implied by monetary policy. The substantial
S E P T E M B E R / O C TO B E R , PA R T 2
2009
491
Nelson
monetary stimulus created in 1977 was followed
by a collapse of the Social Contract at the end of
1978 and at the beginning of 1979. Shortly afterward, the Callaghan government lost a confidence
vote in Parliament and had to hold a general
election.
Friedman’s travels, interviews, and commentary on current events meant that the FriedmanSchwartz Monetary Trends study of the United
States and the United Kingdom was being slowed
down. Friedman told an audience in Sheffield,
England, in September 1970, “Mrs. Anna Schwartz
and I are currently engaged on a comparison of
U.S. and U.K. monetary trends…I had initially
hoped to present a paper on this work at this
seminar, but unfortunately the research did not
go rapidly enough” (Friedman, 1971b, p. 151). In
both that presentation and in Friedman lectures
in the following years (Friedman, 1972c, 1973a),
the U.K. coverage was limited mainly to discussions of data plots. Friedman and Schwartz (1972,
p. 32) admitted, “Our estimate of the time it would
take us to complete the manuscript on monetary
trends has been unduly optimistic in the past…
[W]e shall refrain from projecting a date for completion.” The publication of Schwartz (1975)
indicated that progress was being made, and by
1979 the project was edging to the finishing line,
with publication projected for sometime in the
early 1980s.
Issues
Common Market Entry. Though he sometimes
referred to the “European countries and Britain”
(e.g., in Pringle, 2002, p. 22), Friedman usually
counted the United Kingdom as part of Europe.
In 1948 Friedman referred to “Europe, including
England” (New York Times, January 11, 1948)
and classed the United Kingdom within Europe
or Western Europe on later occasions too, including in Friedman (1958a, p. 510) and Friedman
and Schwartz (1982, p. 309). A major issue for
the United Kingdom in the 1960s and continuing
into the 1970s was whether it should join other
major European countries in the European
Economic Community (EEC) or Common Market.
Friedman reminded people that he played a part
492
S E P T E M B E R / O C TO B E R , PA R T 2
2009
in the preparations for “the so-called Coal and
Steel Community—a precursor to the Common
Market” when he served as an adviser to the
Marshall Plan in 1950 (Newsweek, May 24, 1971).
In 1967, Friedman warned London newspaper
readers not to expect too much of EEC membership. “Membership of the Common Market may
or may not be good for Britain, but it is not a
necessary part of the solution to Britain’s difficulties. Germany in 1948 achieved an economicmiracle policy by decontrol without any Common
Market” (Sunday Telegraph, June 25, 1967).
After unsuccessful attempts to negotiate
entry in the 1960s, the United Kingdom joined
the EEC at the start of 1973. By then, Friedman
had expressed concern about the direction of the
EEC, in particular the danger that it was “dominated by the notion that it’s to serve as the supercentral planning body” for member countries
(Vision, April 1972). One of the planning measures undertaken by the EEC also went against
Friedman’s belief in free trade. “So far as the
Common Market is concerned, they have engaged
in agricultural protectionism on a large scale, as
you say,” Friedman observed in April 1978.90
“They are making a mistake in doing that.” In late
1978, the EEC started to move against something
Friedman regarded as one of the few favorable
economic policy developments in the 1970s:
floating exchange rates. The European Monetary
System (EMS) was set up at the end of 1978, to
commence in 1979. But, for now, the United
Kingdom would not be participating. The
Callaghan government had decided to stick to a
floating pound.
Democracy. Like Keynes before him,
Friedman in his work talked about the damaging
effects that inflation could have on the stability
of a democracy. For example, Friedman opened
his testimony to Congress in May 1959 with the
following: “Unless we can achieve both a reasonably stable economy in the short run and a reasonably stable price level in the long run, our free
enterprise economy is unlikely to be permitted
to survive.”91 In further testimony later that year,
Friedman said, “Wars aside, the chief economic
90
Milton Friedman Speaks, Episode 8, p. 26 of transcript.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
threats to the preservation of a free society have
come from the sharp fluctuations…in economic
activity and in prices…that have threatened to
tear the social fabric asunder.”92 Another theme
in his work, central to Friedman (1962a), was
the presence of a sizable private sector as a necessary condition for political freedom.
Friedman produced a storm when, in 1976,
he made use of these two themes to discuss the
state of the United Kingdom. On Meet the Press
(October 24, 1976), Friedman said, “Great Britain
is another horrible example…Britain is on the verge
of collapse.” Around the same time, Encounter
magazine published an article by Friedman, arguing that the public sector had become so large a
fraction of the U.K. economy that democracy
was threatened: “I fear very much that the odds
are at least 50-50 that within the next five years
British freedom and democracy, as we have seen
it, will be destroyed.”93 The controversy intensified when Friedman made similar remarks in a
60 Minutes special on the U.K. economy broadcast in the United States on November 28.
Friedman’s observations produced a backlash in the U.K. press. The Daily Mirror called
Friedman the “smiling man of woe,” and an editorial criticized his “biased view” and “doomsday
solutions” (November 30, 1976a and 1976b). The
Daily Express (November 30, 1976) said that
Friedman’s “sensible followers in this country—
particularly Mrs. Thatcher and Sir Keith Joseph—
must be in near despair” about his “absurdities.”
John Kenneth Galbraith joined in the criticism,
observing, “If the economists were right every
time they predicted a country was going down
the drain, there would be nowhere left” (Daily
Mirror, January 10, 1977).
The criticism that prompted Friedman to
react came from Samuel Brittan. Brittan published
an open letter to Friedman in his Financial Times
column, arguing that his “recent warnings about
the United Kingdom…represent personal hunches,
91
Friedman, May 25, 1959, testimony in Joint Economic Committee
(1959a, p. 605; p. 136 of reprint).
92
October 30, 1959, testimony, in Joint Economic Committee (1959b,
p. 3020).
93
Friedman (1976c, p. 9); also published in Sunday Telegraph,
October 31, 1976.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
individual value judgments or exaggerations”
that could detract from Friedman’s insights on
monetary matters (Financial Times, December 2,
1976). Friedman in turn had an “Open Reply” to
Brittan defending his statements (Financial Times,
January 6, 1977).
This backlash against Friedman’s warnings
reflected a certain inconsistency on the part of
U.K. commentators. A substantial number of U.K.
leaders and U.K. economists had made comments
similar to Friedman’s about the threat to democracy coming from economic instability. For example, in a September 1976 television interview,
Prime Minister Callaghan had said, “If we were
to fail, I don’t think another government could
succeed. I do not think that would mean a National
[coalition] government. I fear it would lead to a
totalitarian government of the Left or the Right”
(The Sun, October 1, 1976). An economist at the
Bank of England, Charles Goodhart, had warned
(1975, p. 221) that continued stagflation of the sort
the Western economies had faced in the 1970s
“may well serve to destroy the atomistic, democratic, capitalist structure of their existing system.”
One element that contributed to the controversy was Friedman’s emphasis on the threat to
democracy from a large government sector rather
than from inflation alone. Friedman was not, however, conflating the issues; he explicitly maintained that excessive growth of government did
social harm even if it were not accompanied by
inflation. “Ending inflation, in my opinion,”
Friedman said in 1981, “is a very desirable thing
to do. In my opinion, it is likely to be a necessary
precondition for resolving the other problems that
countries have, but it is not a be-all and end-all
of economic policy” (Evening Post, April 27, 1981).
In particular, Friedman contended that an inexorable rise in the government spending share of
GDP, “even if were accomplished without any
inflation whatever…would ultimately destroy
our freedom and society” (Evening Capital,
November 18, 1978).94
It was Friedman’s discussion of the U.K. government-to-GDP share that became the matter for
which a number of commentators took him to task.
94
A prior occasion on which Friedman separated the issues of inflation and the “threat to the maintenance of a free society” from a large
public sector was in a letter to The Economist (September 28, 1974).
S E P T E M B E R / O C TO B E R , PA R T 2
2009
493
Nelson
The figure he used in Friedman (1976c) and elsewhere of 60 percent was indeed the figure reported
by U.K. government publications as of early 1976.
But revisions during the year exposed doublecounting, and the official estimate was revised to
about 45 percent to 47 percent. Chancellor Healey
said that the corrected number refuted “the picture
of a profligate public sector as ignorantly presented by Professor Milton Friedman” (House of
Commons Debates, November 30, 1976, p. 715).
Some discussions, such as Begg (1987) and
Tomlinson (1990), mention Friedman’s use of
the 60 percent figure and create the impression
that, had the corrected number been known from
the start, there would have been no basis for
Friedman’s warnings about public expenditure
in the United Kingdom. This is questionable; for
one thing, Friedman stressed that his argument
did not rest on the present number being as high
as 60 percent (BBC1, December 6, 1976). For
another, Friedman would likely not have agreed
with all the statistical decisions used to reach
the 45 percent to 47 percent share.
Friedman likely would have insisted that
subsidies to firms and transfer payments to individuals be counted in the government spending
aggregate, and not (as is often the practice) as
“negative taxes.” My suspicion is that an estimate
that classed these items as spending—and which
was sure to include all government enterprises
in the government-spending estimate—would
show the share peaking above 50 percent during
the mid-1970s.
The Thatcher Government. “It’s not my job
to persuade people about things,” Friedman
argued (Omaha World-Herald, October 20, 1976;
“I just develop ideas and leave them around for
people to pick up.” Among those seen as picking up Friedman’s ideas in the late 1970s was
Margaret Thatcher, who had replaced Edward
Heath as Conservative Party leader in February
1975. Some have claimed that Thatcher had
monetarist ideas even in the late 1960s (Wapshott
and Brock, 1983, pp. 88, 187). But the Thatcher
statements offered as evidence on this point are
similar to those common among politicians at
the time—that is, she urged giving monetary
policy greater weight among the tools used for
494
S E P T E M B E R / O C TO B E R , PA R T 2
2009
demand management. The monetarist view of
inflation was not present in Thatcher’s 1960s’
statements. But there is no doubt that the position
on inflation taken by Thatcher and other senior
Opposition personnel converged in the late 1970s
toward the familiar monetarist one, and that the
policy framework of the Conservative Party on
returning to office in 1979 was shaped by the
monetarist position on inflation. Friedman himself is said to have first had detailed conversations with Thatcher in 1978 (Campbell, 2000,
p. 372). These probably took place during his
April 1978 visit to the United Kingdom.
Friedman, as discussed previously, was critical of the U.K. Conservative Party’s historical
record on economic policymaking. The impact
of his ideas on Conservative Party policy formation after 1975 did not come in for universal welcome on the part of conservatives in the United
Kingdom. The most well-known critics were
Edward Heath and other Conservative Party
advocates of incomes policy to fight inflation.95
But the economic substance of Friedman’s
arguments was nonpartisan. As early as 1968,
Robert Solow noted, “the association of monetarism with right-wing politics is not at all necessary”
(The Times, December 23, 1968). Friedman’s own
observations were in emphatic agreement with
Solow’s assessment. For example, Friedman (1978)
argued: “No doubt there are strong ideological
elements in the susceptibility of individuals,
including politicians and their advisors, to persuasion by either the monetarist or Keynesian
views. Yet the basic issue is scientific, not ideological…Whatever a man’s objectives, whatever
his ideology, he can pursue them more intelligently the better he understands how the world
works.”
A specific scientific question underlay
much of the U.K. political debate from 1974
onward. The issue was whether incomes policy
was a valid weapon against inflation or whether
instead only monetary policy could accomplish
disinflation. The outcome of this debate rested
on the scientific question of whether inflation
95
The projected companion paper covering 1979-2006 will deal in
detail with Heath’s disagreements with Friedman, including those
covered in their radio debate in 1980.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
was a monetary phenomenon. One Laboursupporting writer recognized the scientific aspect
to the debate in 1974, observing: “There is a danger that socialists will dismiss the monetarists’
argument simply because that school of thought
has hitherto been associated with right-wing conservatism. This is because the leading monetarist,
Professor Milton Friedman, has some very eccentric right-wing views. But, in fact, the analysis of
the rate of inflation is in no way related to ideological conservatism…Socialists will have to come
to terms with this school of thought if we are to
effectively fight inflation” (New Statesman,
October 25, 1974).
Governments of both political parties in the
United Kingdom eventually assigned inflation
control to monetary policy. This reflected not
the triumph of ideology, but the fact that policymakers wanted inflation down, and had accepted
that, as a technical matter, the only way this could
be accomplished was through monetary policy.
As Friedman put it, “It’s not what I advocate that
matters; there is only one way to do it” (St. Louis
Globe-Democrat, December 16, 1977).
That convergence of the political parties’
positions had not yet occurred when the 1979
U.K. general election was held. The Callaghan
government went to the election with incomes
policy prominent in its platform, including a new
union/government agreement on wages to replace
the Social Contract, and plans to extend price
control, while the Conservative Party under
Thatcher rejected incomes policy in favor of a
focus on monetary control. Interviewed by BBC
television several months after Thatcher’s election
victory, Friedman underlined the change in direction, both with respect to monetary policy and to
the role of government, implied by Thatcher’s
coming to power. “If the Thatcher government
succeeds,” Friedman said, “it will be an example
that will not be lost on the United States or the
rest of the world.”96
the implications of the change of government for
the direction of economic policy. “The mantle of
Keynes, and particularly the embroideries of his
followers, appears to have worn thin; and the
mode has shifted towards the sterner lines of
thought popularized by Milton Friedman. The
new Government comes to office in a climate of
opinion very different from that which influenced
its Conservative predecessor.” Nevertheless,
Nicholas Kaldor was able to boast accurately that
Friedman had “made comparatively few converts
among academic economists” in the United
Kingdom.97 The support for Friedman’s ideas was
also thin among members of the new government,
once one looked below the most senior levels. In
these circumstances, and despite his drawing of
distinctions between his own positions and those
of the Thatcher government, Friedman would find
himself a central figure in the debate over the new
economic policies. In 1979 he was about to shift
to an even-higher profile in the United Kingdom
and, in defending his positions on monetary policy
and on the role of government, would encounter
in debate some of the most formidable figures in
U.K. economics and some of the biggest names
in both major political parties.
REFERENCES*
Anderson, Martin, ed. Registration and the Draft:
Proceedings of the Hoover-Rochester Conference
on the All-Volunteer Force. Stanford, CA: Hoover
Institution Press, 1982.
Beenstock, Michael. A Neoclassical Analysis of
Macroeconomic Policy. Cambridge, UK: Cambridge
University Press, 1980.
Begg, David. “U.K. Fiscal Policy Since 1970,” in
Richard Layard and Rudiger Dornbusch, eds., The
Performance of the British Economy. Oxford, UK:
Oxford University Press, 1987, pp. 29-63.
96
CONCLUSION
October 11, 1979, broadcast of interview with Friedman on BBC
television program Newsweek.
97
Kaldor (1982, p. xxii).
The U.K. banking periodical Midland Bank
Review commented in its Summer 1979 issue on
* Sources for periodical articles and television appearances/
interviews are listed in the Bibliographical Appendix.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
495
Nelson
Benati, Luca. “The Inflation-Targeting Framework
from an Historical Perspective.” Bank of England
Quarterly Bulletin, Summer 2005, 45(2), pp. 160-68.
Bernanke, Ben S. “Friedman’s Monetary Framework:
Some Lessons,” in Mark A. Wynne, Harvey
Rosenblum, and Robert L. Formaini, eds., The
Legacy of Milton and Rose Friedman’s Free to
Choose: Economic Liberalism at the Turn of the
21st Century. Dallas, TX: Federal Reserve Bank of
Dallas, 2004, pp. 207-17.
Blinder, Alan S. “Ruminations on Karl Brunner’s
Reflections,” in R.W. Hafer, ed., The Monetary
Versus Fiscal Policy Debate: Lessons from Two
Decades. Totowa, NJ: Rowman and Allanheld, 1986,
pp. 117-26.
Bordo, Michael D. and Wheelock, David C. “Monetary
Policy and Asset Prices: A Look Back at Past U.S.
Stock Market Booms.” Federal Reserve Bank of St.
Louis Review, November/December 2004, 86(6),
pp. 19-44; http://research.stlouisfed.org/publications/
review/04/11/BordoWheelock.pdf.
Brunner, Karl. “Confusion of Language and the Politics
of Uncertainty.” Shadow Open Market Committee
Position paper, University of Rochester, March 24-25,
1985.
Brunner, Karl and Meltzer, Allan H. Money and the
Economy: Issues in Monetary Analysis. Cambridge,
UK: Cambridge University Press, 1993.
Cairncross, Alec. The Wilson Years: A Treasury Diary,
1964-1969. London: Historian’s Press, 1997.
Campbell, John. Margaret Thatcher. Volume 1: The
Grocer’s Daughter. London: Jonathan Cape, 2000.
Capie, Forrest and Webber, Alan. A Monetary History
of the United Kingdom, 1870-1982. Volume I: Data,
Sources, Methods. London: Allen and Unwin, 1985.
Childs, Bruce. Britain Since 1945: A Political History.
Sixth Edition. London: Routledge, 2006.
Cobham, David. “Convergence, Divergence and
Realignment in British Macroeconomics.” Banca
Nazionale del Lavoro Quarterly Review, June 1984,
149, pp. 159-76.
496
S E P T E M B E R / O C TO B E R , PA R T 2
2009
Committee on Banking and Currency, House of
Representatives. The Federal Reserve System After
Fifty Years. Washington, DC: Government Printing
Office, 1964.
Committee on the Budget, House of Representatives.
Second Budget Resolution: Fiscal Year 1976.
Washington, DC: Government Printing Office, 1975.
Committee on the Judiciary, House of Representatives.
Constitutional Amendments to Balance the Federal
Budget: Hearings. Washington, DC: Government
Printing Office, 1980.
Dacey, W. Manning. “The Cheap Money Technique.”
Lloyds Bank Review, January 1947, 2(1), 49-63.
Edge, Stephanie K. “The Relative Stability of Monetary
Velocity and the Investment Multiplier.” Australian
Economic Papers, December 1967, 6(9), pp. 192-207.
Eshag, Eprime. Fiscal and Monetary Policies and
Problems in Developing Countries. Cambridge, UK:
Cambridge University Press, 1983.
Friedman, Milton. “The Inflationary Gap: II.
Discussion of the Inflationary Gap.” American
Economic Review, February 1942, 32(2), pp. 314-20.
Friedman, Milton. “Methods of Predicting the Onset
of ‘Inflation,’” in Carl S. Shoup, Milton Friedman,
and Ruth P. Mack, eds., Taxing to Prevent Inflation.
New York: Columbia University Press, 1943,
pp. 111-53.
Friedman, Milton. “Book Review: Saving, Investment
and National Income by Oscar L. Altman.” Review
of Economics and Statistics, May 1944, 26(2),
pp. 101-02.
Friedman, Milton. “A Monetary and Fiscal Framework
for Economic Stability.” American Economic Review,
June 1948, 38(3), pp. 245-64.
Friedman, Milton. “Wesley Mitchell as an Economic
Theorist.” Journal of Political Economy, December
1950, 58(6), pp. 465-93.
Friedman, Milton. “Some Comments on the
Significance of Labor Unions for Economic Policy,”
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
in David McCord Wright, ed., The Impact of the
Union. New York: Harcourt Brace, 1951, pp. 204-34.
Friedman, Milton. “Price, Income and Monetary
Changes in Three Wartime Periods.” American
Economic Review, May 1952 (Papers and
Proceedings), 42(2), pp. 612-25.
Friedman, Milton. Essays in Positive Economics.
Chicago: University of Chicago Press, 1953a.
Friedman, Milton. “The Case for Flexible Exchange
Rates,” in Essays in Positive Economics. Chicago:
University of Chicago Press, 1953b, pp. 157-203.
Friedman, Milton. “Why the American Economy Is
Depression Proof.” Nationalekonomiska föreningens
förhandlingar (Stockholm), 1954, 3, pp. 58-77.
Reprinted in Milton Friedman, Dollars and Deficits:
Inflation, Monetary Policy and the Balance of
Payments. Englewood Cliffs, NJ: Prentice Hall, 1954,
pp. 72-96.
Friedman, Milton. “Money and Banking,” in Solomon
Fabricant, ed., Government in Economic Life:
National Bureau of Economic Research Thirty-Fifth
Annual Report, May 1955. New York: NBER, 1955,
pp. 30-33.
Friedman, Milton. “The Quantity Theory of Money:
A Restatement,” in Milton Friedman, ed., Studies
in the Quantity Theory of Money. Chicago:
University of Chicago Press, 1956, pp. 3-21.
Friedman, Milton. “Foreign Economic Aid: Means
and Objectives.” Yale Review, June 1958a, 47(4),
pp. 500-16.
Friedman, Milton. “What Price Inflation?” Finance
and Accounting, 1958b, 38(7), pp. 18-27.
Friedman, Milton. “Book Review: Inflation by Thomas
Wilson.” American Economic Review, December
1961a, 51(5), pp. 1051-55.
Friedman, Milton. “Monetary Data and National
Income Estimates.” Economic Development and
Cultural Change, April 1961b, 9(3), pp. 267-86.
Friedman, Milton. Capitalism and Freedom. Chicago:
University of Chicago Press, 1962a.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Friedman, Milton. “Is a Free Society Stable?” New
Individualist Review, Summer 1962b, 2(2), pp. 3-10.
Friedman, Milton. Inflation: Causes and Consequences.
Bombay: Asia Publishing House, 1963a.
Friedman, Milton. “The Present State of Monetary
Theory,” Economic Studies Quarterly, September
1963b, 14(1), 1-15.
Friedman, Milton. “Postwar Trends in Monetary
Theory and Policy.” National Banking Review,
September 1965a, 2(1), pp. 1-9. Reprinted in
Milton Friedman, The Optimum Quantity of
Money and Other Essays. Chicago: Aldine, 1969,
pp. 69-79.
Friedman, Milton. “The Monetary Studies of the
National Bureau,” in The National Bureau Enters
Its 45th Year, 44th Annual Report of the National
Bureau of Economic Research. New York: NBER,
1964b, pp. 7-25. Reprinted in Milton Friedman,
The Optimum Quantity of Money and Other Essays.
Chicago: Aldine, 1969, pp. 261-83.
Friedman, Milton. “Discussion.” American Economic
Review, March 1965, 55(1/2), pp. 178-81.
Friedman, Milton. Dollars and Deficits: Inflation,
Monetary Policy and the Balance of Payments.
Englewood Cliffs, NJ: Prentice Hall, 1968a.
Friedman, Milton. “The Role of Monetary Policy.”
American Economic Review, March 1968b, 58(1),
pp. 1-17.
Friedman, Milton. “Money: Quantity Theory,” in
David L. Sills, ed., International Encyclopedia of
the Social Sciences. Volume 10. New York:
Macmillan, 1968c, pp. 432-47.
Friedman, Milton. “The International Adjustment
Mechanism: Panel Discussion,” in Federal Reserve
Bank of Boston, The International Adjustment
Mechanism. Boston: Federal Reserve Bank of Boston,
1969; pp. 15-20.
Friedman, Milton. “The Counter-Revolution in
Monetary Theory.” IEA Occasional Paper No. 33,
Institute of Economic Affairs, 1970a. Reprinted in
S E P T E M B E R / O C TO B E R , PA R T 2
2009
497
Nelson
Milton Friedman, Monetarist Economics. Oxford,
UK: Basil Blackwell, 1991; pp. 1-20.
Friedman, Milton. “Comment on Tobin.” Quarterly
Journal of Economics, May 1970b, 84(2), pp. 318-27.
Friedman, Milton. “The New Monetarism: Comment.”
Lloyds Bank Review, October 1970c, 25(98),
pp. 52-53.
Friedman, Milton. “Government Revenue from
Inflation.” Journal of Political Economy, July/August
1971a, 79(4), pp. 346-56.
Friedman, Milton. “A Note on the U.S. and U.K.
Velocity of Circulation,” in George Clayton, John C.
Gilbert, and Robert C. Sedgwick, eds., Monetary
Theory and Monetary Policy in the 1970s:
Proceedings of the 1970 Sheffield Money Seminar.
London: Oxford University Press, 1971b, pp. 151-52.
Friedman, Milton. “Comments on the Critics,”
Journal of Political Economy, September/October
1972a, 80(5), pp. 906-50.
Friedman, Milton. “Monetary Policy,” in Proceedings
of the American Philosophical Society, June
1972b, 116(3), pp. 183-96.
Friedman, Milton. “Monetary Trends in the United
States and United Kingdom.” American Economist,
Spring 1972c, 16(1), pp. 4-17.
Friedman, Milton. “Have Monetary Policies Failed?”
American Economic Review, May 1972d (Papers
and Proceedings), 62(2), pp. 11-18.
Friedman, Milton. Money and Economic Development:
The Horowitz Lectures of 1972. New York: Praeger,
1973a.
Friedman, Milton. There’s No Such Thing As a Free
Lunch: Essays on Public Policy. LaSalle, IL: Open
Court, 1975a.
Friedman, Milton. “25 Years after the Rediscovery
of Money: What Have We Learned? Discussion.”
American Economic Review, May 1975b (Papers
and Proceedings), 65(2), pp. 176-79.
Friedman, Milton. Milton Friedman in Australia 1975.
Darlinghurst, NSW: Dryden Press, 1975c.
Friedman, Milton. Is Inflation a Curable Disease?
Alex C. Walker Memorial Lecture, December 1974.
Pittsburgh: University of Pittsburgh Graduate School
of Business, 1975d.
Friedman, Milton. “Foreword,” in Fritz Machup, ed.,
Essays on Hayek. New York: New York University
Press, 1976a, pp. xxi-xxiv.
Friedman, Milton. Price Theory. Second Edition.
Chicago: Aldine, 1976b.
Friedman, Milton. “The Line We Dare Not Cross.”
Encounter, November 1976c, 47(5), pp. 8-14.
Friedman, Milton. “Nobel Lecture: Inflation and
Unemployment.” Journal of Political Economy,
June 1977a, 85(3), 451-72.
Friedman, Milton. “Discussion of ‘The Monetarist
Controversy.’” Federal Reserve Bank of San
Francisco Economic Review, Spring 1977b, (Suppl.),
pp. 12-26.
Friedman, Milton. “How Stands the Theory and
Practice of Monetary Policy?” Presented at the
Mont Pelerin Society meeting, Hong Kong, 1978.
Friedman, Milton. “Facing Inflation.” Challenge,
November 1973b, 16(5), 29-37.
Friedman, Milton. “Memorandum: Response to
Questionnaire on Monetary Policy, June 11, 1980,”
in Treasury and Civil Service Committee,
Memoranda on Monetary Policy. London: H.M.S.O,
1980a, pp. 55-61.
Friedman, Milton. “Monetary Correction.” IEA
Occasional Paper No. 41, Institute of Economic
Affairs, 1974. Reprinted in Milton Friedman,
Monetarist Economics. Oxford, UK: Basil Blackwell,
1991, pp. 21-47.
Friedman, Milton. “Comment on Benjamin M.
Friedman, ‘The Changing Character of Financial
Markets,’” in Martin Feldstein, ed., The American
Economy in Transition. Chicago: University of
Chicago Press, 1980b, pp. 78-86.
498
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
Friedman, Milton. “Monetary Policy: Theory and
Practice.” Journal of Money, Credit, and Banking,
February 1982b, 14(1), pp. 98-118.
Friedman, Milton and Meiselman, David. “The
Relative Stability of Monetary Velocity and the
Investment Multiplier in the United States, 18971958,” in Commission on Money and Credit, ed.,
Stabilization Policies. Eaglewood Cliffs, NJ: Prentice
Hall, 1963, pp. 165-268.
Friedman, Milton. “Monetarism in Rhetoric and
Practice.” Bank of Japan Monetary and Economic
Studies, October 1983, 1(2), pp. 1-14.
Friedman, Milton and Roosa, Robert V. The Balance
of Payments: Free Versus Fixed Exchange Rates.
Washington, DC: American Enterprise Institute, 1967.
Friedman, Milton. “Quantity Theory of Money,” in
John Eatwell, Murray Milgate, and Peter Newman,
eds., The New Palgrave: A Dictionary of Economics,
Volume 4, Q to Z. London: Macmillan, 1987, pp. 3-20.
Friedman, Milton and Schwartz, Anna J. A Monetary
History of the United States, 1867-1960. Princeton,
NJ: Princeton University Press, 1963.
Friedman, Milton. On Milton Friedman. Vancouver,
BC: Fraser Institute, 1982a.
Friedman, Milton. “A Proposal for Resolving the
U.S. Balance of Payments Problem: Confidential
Memorandum to President-Elect Nixon [1968],” in
L. Melamed, ed., The Merits of Flexible Exchange
Rates: An Anthology. Fairfax, VA: George Mason
University Press, 1988, pp. 429-38.
Friedman, Milton. Money Mischief: Episodes in
Monetary History. New York: Harcourt Brace
Jovanovich, 1992.
Friedman, Milton. “The ‘Plucking Model’ of Business
Fluctuations Revisited.” Economic Inquiry, April
1993, 31(2), pp. 171-77.
Friedman, Milton. “John Maynard Keynes.” Federal
Reserve Bank of Richmond Economic Quarterly,
Spring 1997, 83(2), pp. 1-23.
Friedman, Milton. “Comment: Inflation,
Unemployment and the Pound,” in Subroto Roy
and John Clarke, eds., Margaret Thatcher’s
Revolution: How It Happened and What It Meant.
London: Continuum, 2005, p. 66.
Friedman, Milton and Friedman, Rose D. Free to
Choose. New York: Harcourt Brace Jovanovich, and
Middlesex, UK: Penguin, 1980.
Friedman, Milton and Friedman, Rose D. Two Lucky
People: Memoirs. Chicago: University of Chicago
Press, 1998.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Friedman, Milton and Schwartz, Anna J. Trends in
Money, Income, and Prices, 1867-1966 [manuscript
version]. New York: National Bureau of Economic
Research, 1966.
Friedman, Milton and Anna J. Monetary Statistics of
the United States. New York: Columbia University
Press, 1970.
Friedman, Milton and Schwartz, Anna J. “Money,”
in Innovations in Economic Research, 52nd Annual
Report of the National Bureau of Economic Research,
September 1972. New York: NBER, 1972, pp. 29-32.
Friedman, Milton and Schwartz, Anna J. Monetary
Trends in the United States and the United Kingdom:
Their Relation to Income, Prices, and Interest Rates,
1867-1975. Chicago: University of Chicago Press,
1982.
Goodhart, Charles A. E. Money, Information and
Uncertainty. London: Macmillan, 1975.
Gurley, John G. and Shaw, Edward S. Money in a
Theory of Finance. Washington, DC: Brookings
Institution, 1960.
Hallowell, Burton C. A Study of British Interest
Rates, 1929-50. Philadelphia: Connecticut General
Life Insurance Company, 1950.
Harrod, Roy.“Discussion Papers: (a) Sir Roy Harrod,”
in George Clayton; John C. Gilbert; John C. and
Robert C. Sedgwick, eds., Monetary Theory and
Monetary Policy in the 1970s: Proceedings of the
S E P T E M B E R / O C TO B E R , PA R T 2
2009
499
Nelson
1970 Sheffield Money Seminar. London: Oxford
University Press, 1971, pp. 58-63.
Kaldor, Nicholas. The Scourge of Monetarism.
Oxford, UK: Oxford University Press, 1982.
Institute of Economic Affairs. Inflation: Causes,
Consequences, and Cures. London: Institute of
Economic Affairs, 1974.
Keynes, John Maynard. The General Theory of
Employment, Interest and Money. London:
Macmillan, 1936.
Johnson, Harry G. “Review: Money and Economic
Development.” Economica, August 1974, 41(163),
346-347.
Keynes, John Maynard. How to Pay for the War: A
Radical Plan for the Chancellor of the Exchequer.
London: Macmillan, 1940.
Joint Committee on the Economic Report. Monetary
Policy and the Management of the Public Debt:
Replies to Questions. Washington, DC: Government
Printing Office, 1952.
Krugman, Paul. “It’s Baaack! Japan’s Slump and the
Return of the Liquidity Trap.” Brookings Papers on
Economic Activity, 1998, 29(2), pp. 137-87.
Joint Economic Committee. Employment, Growth,
and Price Levels, Hearings, Part 4. Washington, DC:
Government Printing Office, 1959a. Friedman’s
opening testimony reprinted as Milton Friedman,
“Monetary Theory and Policy,” in R.J. Ball and
Peter Doyle, eds., Inflation. London: Penguin, 1969,
pp. 136-45.
Joint Economic Committee. Employment, Growth,
and Price Levels, Hearings. Part 9A. Washington,
DC: Government Printing Office, 1959b.
Laidler, David. “Radcliffe, the Quantity Theory and
Monetarism,” in David Cobham; Richard Harrington,
and George Zis, eds., Money, Trade, and Payments:
Essays in Honour of D.J. Coppock. Manchester, UK:
Manchester University Press, 1989, pp. 17-37.
Laidler, David. “From Bimetallism to Monetarism:
The Shifting Political Affiliation of the Quantity
Theory.” Research Report 2001-1, Department of
Economics, University of Western Ontario, 2001.
Joint Economic Committee. The United States Balance
of Payments: Hearings. Part 3. Washington, DC:
Government Printing Office, 1963.
Laidler, David. “The Role of the History of Economic
Thought in Modern Macroeconomics,” in Paul
Mizen, ed., Modern History, Exchange Rates and
Financial Markets: Essays in Honour of Charles
Goodhart. Volume Two. Cheltenham, UK: Edward
Elgar, 2003, pp. 12-29.
Joint Economic Committee. Economic Analysis and
the Efficiency of Government: Hearings. Part 3.
Washington, DC: Government Printing Office, 1970.
Levy, David. “Milton Friedman: Interview.” Federal
Reserve Bank of Minneapolis The Region, June
1992, 6(2), pp. 6-13.
Joint Economic Committee. The President’s New
Economic Program: Hearings. Part 3. Washington,
DC: Government Printing Office, 1971.
Lindsey, David E.; Orphanides, Athansios and Rasche,
Robert H. “The Reform of October 1979: How It
Happened and Why.” Federal Reserve Bank of
St. Louis Review, March/April 2005, 87(2 Part 2),
pp. 187-235; http://research.stlouisfed.org/
publications/review/05/03/part2/Lindsey.pdf.
Joint Economic Committee. The Supersonic Transport:
Hearings. Washington, DC: Government Printing
Office, 1973a.
Joint Economic Committee. How Well Are Fluctuating
Exchange Rates Working? Hearings. Washington,
DC: Government Printing Office, 1973b.
Kaldor, Nicholas. “The New Monetarism.” Lloyds
Bank Review, July 1970, 25(97), pp. 1-18.
500
S E P T E M B E R / O C TO B E R , PA R T 2
2009
London and Cambridge Economic Service. Key
Statistics of the British Economy, 1900-1962.
London: Leagrave Press, 1963.
Mankiw, N. Gregory. “The Inexorable and Mysterious
Tradeoff between Inflation and Unemployment.”
Economic Journal, May 2001, 111(471), pp. C45-C61.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
Marty, Alvin L. “Gurley and Shaw on Money in a
Theory of Finance,” Journal of Political Economy,
February 1961, 69(1), pp. 56-62.
Patinkin, Don. Keynes’ Monetary Thought: A Study
of Its Development. Durham, NC: Duke University
Press, 1976a.
Meltzer, Allan H. “Monetarist, Keynesian and Quantity
Theories.” Kredit und Kapital, June 1977, 10(2),
pp. 149-82. Reprinted in Thomas Mayer; Martin
Bronfenbrenner; Karl Brunner et al., The Structure
of Monetarism. New York: W.W. Norton, 1978,
pp. 145-75.
Patinkin, Don. “Keynes and Econometrics: On the
Interaction Between the Macroeconomic Revolutions
of the Interwar Period.” Econometrica, November
1976b, 44(6), pp. 1091-1123.
Modigliani, Franco. The Debate Over Stabilization
Policy. Cambridge, UK: Cambridge University Press,
1986.
Nelson, Edward. “An Overhaul of Doctrine: The
Underpinning of U.K. Inflation Targeting.” Economic
Journal, June 2009, 119(538), pp. F333-68.
Nelson, Edward and Schwartz, Anna M. “The Impact
of Milton Friedman on Modern Monetary Economics:
Setting the Record Straight on Paul Krugman’s ‘Who
Was Milton Friedman?’” Journal of Monetary
Economics, May 2008, 55(4), pp. 835-56.
Oxford University Press. Oxford English Dictionary.
Oxford, UK: Clarendon Press, 1976.
Patinkin, Don. “An Indirect-Utility Approach to the
Theory of Money, Assets and Savings,” in Frank H.
Hahn and Frank P.R. Brechling, eds., Theory of
Interest Rates: Proceedings of a Conference Held
by the International Economic Association. London:
Macmillan, 1965.
Patinkin, Don. “The Chicago Tradition, the Quantity
Theory, and Friedman.” Journal of Money, Credit,
and Banking, February 1969, 1(1), pp. 46-70.
Patinkin, Don. “Friedman on the Quantity Theory
and Keynesian Economics.” Journal of Political
Economy, September/October 1972a, 80(5),
pp. 883-905.
Patinkin, Don. “On the Short-Run Non-Neutrality of
Money in the Quantity Theory.” Banca Nazionale
del Lavoro Quarterly Review, March 1972b, 100,
pp. 3-22.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Pringle, Robert. “Interview: Milton Friedman.”
Central Banking, August 2002, 13(1), pp. 15-23.
Radcliffe Committee. Committee on the Working of
the Monetary System: Report. London, UK: HMSO,
1959.
Radcliffe Committee. Minutes of Evidence: Committee
on the Working of the Monetary System. London:
HMSO, 1960.
Robertson, D.H. “What Has Happened to the Rate of
Interest?” Three Banks Review, March 1949, 1(1),
pp. 15-31.
Samuelson, Paul A. “Lord Keynes and the General
Theory.” Econometrica, July 1946, 14(3), pp. 187-200.
Schlesinger, James R. “After Twenty Years: The
General Theory.” Quarterly Journal of Economics,
November 1956, 70(4), pp. 581-602.
Schwartz, Anna J. “Why Money Matters.” Lloyds
Bank Review, October 1969, 24(94), pp. 1-16.
Schwartz, Anna J. “Monetary Trends in the United
States and the United Kingdom, 1878-1970: Selected
Findings.” Journal of Economic History, March
1975, 35(1), pp. 138-59.
Selden, Richard T., ed. Capitalism and Freedom:
Problems and Prospects; Proceedings of a Conference
in Honor of Milton Friedman. Charlottesville, VA:
University Press of Virginia, 1975.
Snowdon, Brian; Vane, Howard R. and Wynarczyk,
Peter. “Interview: Milton Friedman,” in A Modern
Guide to Macroeconomics: An Introduction to
Competing Schools of Thought. Aldershot, UK:
Edward Elgar, 1994, pp. 171-78.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
501
Nelson
Snowdon, Brian and Vane, Howard R. “Modern
Macroeconomics and Its Evolution from a Monetarist
Perspective: An Interview with Professor Milton
Friedman.” Journal of Economic Studies, July/
August 1997, 24(4), pp. 192-222.
Solow, Robert M. Price Expectations and the Behavior
of the Price Level. Manchester, UK: Manchester
University Press, 1969.
Svensson, Lars E. O. “Escaping from a Liquidity Trap
and Deflation: The Foolproof Way and Others.”
Journal of Economic Perspectives, Fall 2003, 17(4),
145-66.
Tomlinson, Jim. Public Policy and the Economy
Since 1900. Oxford, UK: Oxford University Press,
1990.
Walters, Alan A. “The Radcliffe Report—Ten Years
After: A Survey of Empirical Evidence,” in David R.
Croome and Harry G. Johnson, eds., Money in
502
S E P T E M B E R / O C TO B E R , PA R T 2
2009
Britain, 1959-1969. Oxford, UK: Oxford University
Press, 1970, pp. 39-68. Reprinted in Kent Matthews,
ed., The Economics and Politics of Money: The
Selected Essays of Alan Walters. Glouchestershire,
UK: Edward Elgar, 1998, pp. 91-120.
Wapshott, Nicholas and Brock, George. Thatcher.
London: Macdonald, 1983.
Wilson, Harold. Final Term: The Labour Government,
1974-1976. London: Weidenfeld and Nicolson, 1979.
Wilson, Thomas. “Monetarism in Britain,” in Thomas
Wilson, Inflation, Unemployment, and the Market.
Oxford, UK: Clarendon Press, 1984, pp. 41-78.
Wright, David McCord, ed. “Selections from the
Discussion of Friedman’s Paper,” in The Impact of
the Union: Eight Economic Theorists Evaluate the
Labor Union Movement. New York: Harcourt Brace,
1951, pp. 235-59.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
BIBLIOGRAPHICAL APPENDIX*
I. Periodical Articles Cited
“Cripps Attacks the Tory Policy,” News-Chronicle, December 18, 1944, p. 3.
“Government’s ‘Anti-Slump’ Powers,” Financial Times (London), November 23, 1945, p. 1.
“Departmental Deals in Gilt-Edged: ‘Net Sellers’ in the Past Six Months—Says Chancellor; Danger of
Trade Depression Outside U.K.,” Financial Times, October 17, 1946, p. 3.
“2½% Rate Dropped: ‘Locals’ on 3% Basis; Other Changes Forecast,” Financial Times, January 3, 1948,
p. 1.
Aaron Director, Milton Friedman, Abram L. Harris, Frank H. Knight, H. Gregg Lewis, Lloyd W. Mints,
Russell T. Nichols, and W. Allen Wallis, “Control of Prices: Regulation of Money Supply to Halt
Inflation Advocated,” New York Times, January 11, 1948, p. E8.
Editorial, “Prices,” Financial Times, February 10, 1951, p. 4.
Milton Friedman, “Living with the Dollar,” The Economist (London), January 3, 1953, p. 16.
A.S., “Bookshelf: The Dollar Grows Up—Mr. Harrod’s Lectures,” Financial Times, December 21, 1953,
p. 10.
P.E., “Shorter Notices: Essays in Positive Economics,” Financial Times, February 8, 1954, p. 8.
Editorial, “Industrial Stocks,” Financial Times, February 15, 1955, p. 6.
P.E., “Some Shorter Notices: John Maynard Keynes—Economist and Policy-Maker by Seymour Harris,”
Financial Times, May 23, 1955, p. 10.
Harold Wincott, “The Root of the Evil—1. Too Much Money,” Financial Times, October 4, 1955, p. 6.
Lombard, “The Fall in Bank Deposits,” Financial Times, November 30, 1955, p. 3.
“Money Market Control,” Manchester Guardian, December 12, 1955, p. 10.
P.E., “Shorter Notices: Introduction to Keynesian Dynamics by Kenneth K. Kurihara,” Financial Times,
October 15, 1956, p. 10.
Express Staff Reporter, “Folly to Go On Like This: Thorneycroft Talks of the Facts of Life,” Daily
Express, July 13, 1957, p. 2.
Andrew Alexander, “A Chancellor Needs a Thick Skin: Mr. Selwyn Lloyd Talks Freely to Andrew
Alexander,” Yorkshire Post, July 11, 1962, p. 5.
Milton Friedman, “The View from America: Floating the Pound,” Sunday Telegraph (London), June 25,
1967, p. 15.
Paul Bareau, “Inflation Limiting Britain’s Scope,” The Sun (London), October 7, 1968, p. 8
Samuel Brittan, “Money Supply: The Great Debate,” Financial Times, October 25, 1968, p. 16.
Harlow Unger, “The Economic Consequences of Mr. Nixon,” Sunday Times (London), November 10,
1968, p. 36.
Paul A. Samuelson, “Don’t Make Too Much of the Quantity Theory,” Sunday Telegraph, December 15,
1968, pp. 19 and 21.
Robert M. Solow, “Putting the Money Supply Dispute into Its True Perspective,” The Times,
December 23, 1968, p. 21.
Alan Day, “Jenkins Goes Chicago-Style,” The Observer (London), April 20, 1969, p. 12.
* Sources in the appendix sections are arranged in chronological order.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
503
Nelson
Samuel Brittan, “What the ‘New Old’ Economics Mean[s],” Financial Times, January 8, 1970, p. 17.
Editorial, “The Budget,” The Economist, April 18, 1970, pp. 61-64.
Edward Heath, “A Better Tomorrow: Foreword by Mr. Heath,” The Guardian, May 27, 1970, p. 6.
“Two Decades of Inflation,” The Economist, August 29, 1970, pp. 12-13.
Milton Friedman, “Development Fashions,” Newsweek, December 21, 1970, p. 82.
Editorial, “Heath Means Business,” Dallas Morning News, February 9, 1971, p. 2D.
Robert McKenzie, “The American Economist Milton Friedman, in Conversation with Robert McKenzie,
Gives His View of How Our Economic Problems Could Be Solved,” The Listener (London),
February 11, 1971, pp. 169-171.
Milton Friedman, “Needed: More of the Same,” Newsweek, February 15, 1971, p. 58.
Jack Prosser, “The D-Mark Dollar Crisis…Why Germany Didn’t Wait for the Crunch Again,” Evening
Standard (London), May 5, 1971, p. 47.
Milton Friedman, “The Mark Crisis,” Newsweek, May 24, 1971, p. 72.
Milton Friedman, Paul A. Samuelson, and Henry Wallich, “Three Views of Nixonomics and Where It
Leads,” Newsweek, January 31, 1972, pp. 74-75.
Richard Howe, “Talking to Vision: M. Friedman—Whatever Happened to Free Enterprise?” Vision
(London), April 1972, pp. 41-44.
John McClaughry, “Milton Friedman Responds” (interview), Business and Society Review, No. 1,
Spring 1972, pp. 5-16. Reprinted as “A Business and Society Review Interview” in Friedman
(1975a, pp. 240-56).
Milton Friedman, “A Cold Day for Britain,” Newsweek, November 27, 1972, p. 87.
Milton Friedman, “Unemployment and Monetary Growth,” The Times, December 12, 1972, p. 17.
Associated Press, “Economist Out of Hospital,” Kansas City Times, December 27, 1972, p. 2A.
Editorial, “The Real Inflation Threat,” Management Today, August 1973, p. 18.
Milton Friedman, “Public Spending and Inflation,” The Times, August 29, 1973, p. 15
Frances Cairncross, “Inflation ‘Immoral Tax No M.P. Would Approve,’” The Guardian, September 16,
1974, p. 12.
Milton Friedman, “Economic Policy,” The Economist, September 28, 1974, p. 4.
Paul Whiteley, “The Monetarists and Labour,” New Statesman, October 25, 1974, p. 574.
Milton Friedman, “National Economic Planning,” Newsweek, July 14, 1975, p. 71.
Theodore Kurrus, “Laissez Faire: Friedman Against Government Control,” Dallas Morning News,
October 17, 1975, p. 13B.
Editorial, “Changing Money,” Management Today, August 1976, p. 3.
David Sinclair, “Inflation: ‘The Tax That Never Has to Be Passed by Parliament,’” The Times,
September 13, 1976, p. 7.
Milton Friedman, “Money and Inflation,” Newsweek, September 20, 1976, p. 77.
Roger Nuttall, “We’re Battling for More Than Simply the Pound, We’re Battling for Our Very Freedom:
Milton Friedman of the Chicago School of Economics Talking to Roger Nuttall,” Daily Mail,
September 30, 1976.
Roger Carroll, “Grim Jim Warns the Nation—Our Last Chance: ‘We Could Have a Dictatorship If This
Government Sinks,’” The Sun, October 1, 1976, pp. 1-2.
Milton Friedman, “The Pound Crisis,” Newsweek, October 11, 1976, p. 91.
504
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Nelson
Rob Warden, “Economist Friedman’s Logic Loved, Hated,” Omaha World-Herald, October 20, 1976,
p. 44. (Originally in Chicago Daily News.)
Milton Friedman, “What Friedman Really Thinks of Britain,” Sunday Telegraph, October 31, 1976.
Milton Friedman, “High Living as a Tax Shelter,” Newsweek, November 8, 1976, p. 80.
Nicholas Davies, “Smiling Man of Woe,” Daily Mirror (London), November 30, 1976a, p. 2.
Editorial, “Mirror Comment: Right On?” Daily Mirror, November 30, 1976b, p. 2.
Editorial, “The Kindest Course?” Daily Express, November 30, 1976, p. 10.
Samuel Brittan, “An Open Letter to Professor Friedman,” Financial Times, December 2, 1976, p. 21.
Milton Friedman, “To Jimmy from James,” Newsweek, December 6, 1976, p. 45.
William Lowther, “Healey’s Budget Won’t Work, Says Friedman,” Daily Mail, December 17, 1976.
Milton Friedman, “How to Denationalize,” Newsweek, December 27, 1976, p. 54.
Editorial, “The Friedman Solution,” Management Today, January 1977, p. 3.
Milton Friedman, “An Open Reply from Milton Friedman,” Financial Times, January 6, 1977, p. 17.
Paul Callan, “Why Dr. Galbraith Thinks Britain Is Great,” Daily Mirror, January 10, 1977, p. 18.
Fred Kutchins, “Leaning against Next Year’s Wind” (interview with Milton Friedman), Saturday
Evening Post (New York), May/June 1977, pp. 16 and 18-20.
David M. Grebler, “Friedman Is Optimistic About Public’s Economic Understanding,” St. Louis GlobeDemocrat, December 16, 1977, p. 12C.
George McCarthy, “Column One,” The Scotsman, January 25, 1978, p. 3.
Michael Fry, “Prof. Friedman Scorns Big Seven Summit,” The Scotsman, April 22, 1978, p. 5.
Milton Friedman, “Has the Tide Turned?” The Listener, April 27, 1978, pp. 526-28.
Milton Friedman, “Inertia and the Fed,” Newsweek, July 24, 1978, p. 70.
Associated Press, “Conservatives Draft Plan to Control Taxes, Inflation,” Evening Capital (Maryland),
November 18, 1978, p. 2.
“Economic Outlook,” Midland Bank Review (U.K.), Summer 1979, pp. 1-4.
Milton Friedman, “Why Inflation Is Like Alcoholism,” The Listener, April 24, 1980, pp. 521-22.
R.J. Ball, “Unemployment and Keynesian Policies,” Financial Times, February 4, 1981, p. 15.
Michael Field, “Selling Friedman’s Miracle Cure,” Evening Post, April 27, 1981, p. 2.
Walter Guzzardi, Jr., “The Dire Warnings of Milton Friedman,” Fortune, March 19, 1984, pp. 28-34 of
U.S. edition; pp. 20-26 of international edition.
Maximilian Walsh, “The Tail That Wags the Dog,” Sydney Morning Herald, October 9, 1986, p. 17.
Milton Friedman, “Monetary History, Not Dogma,” Wall Street Journal, February 12, 1987, p. 24.
Daniel Doron and Steven Plaut, “Liberalization: ‘Israel Didn’t Go Far Enough,’” Jerusalem Post,
November 6, 1987.
Milton Friedman, “Fed’s Arsenal Has Only One Big Gun,” Wall Street Journal, April 5, 1990, p. 19.
Peter Jay, “Problems with the Virtuous Circle,” The Independent (London), September 23, 1991, p. 21.
Milton Friedman, “Once Again: Why Socialism Won’t Work,” New York Times, August 13, 1994, p. 21.
Brian Doherty, “Best of Both Worlds: Interview with Milton Friedman,” Reason, June 1995.
“Milton Friedman on Hong Kong’s Future,” Wall Street Journal, February 12, 1997, p. A16.
Milton Friedman, “Asian Values: Right…,” National Review, December 31, 1997, pp. 36-37.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
505
Nelson
Robert J. Samuelson, “The Age of Friedman,” Newsweek, June 15, 1998, pp. 30-31.
Philip Thornton, “Friedman Condemns the Single Currency as a ‘Grave Error,’” The Independent,
August 28, 2001, p. 11.
David Isaac, “To Keep U.S. Economy Growing, Curb Government, Take Risks: Friedman,” Investor’s
Business Daily, April 15, 2004, pp. A1 and A8.
II. Television Programs Cited
Friedman appearance on Meet the Press, October 24, 1976, NBC transcript.
Friedman appearance on Newsday, BBC2, November 9, 1976; BBC transcript.
Friedman appearance on segment “Will There Always Be an England?” 60 Minutes, November 28, 1976,
CBS transcript.
Friedman appearance on Panorama, BBC1, December 6, 1976; BBC transcript.
Friedman appearance on Meet the Press, November 12, 1978; NBC transcript.
Friedman appearance on Newsweek (BBC television program) episode entitled “New Tory Economics,”
October 11, 1979; excerpted on Pandora’s Box (episode title “The League of Gentlemen”), BBC2,
June 25, 1992 (also screened by ABC television, Australia, on August 10, 1992, on program Four
Corners, with episode title “From Keynes to Chaos”); BBC transcript.
Free to Choose, U.S. series, PBS, 1980; shows and transcripts available online.
Free to Choose, U.K. series, broadcast BBC2 February and March 1980; BBC transcripts.
Friedman interview with Brian Lamb, C-SPAN, November 20, 1994; transcript available on C-SPAN
website.
Friedman talk, April 16, 1996, at Claremont College, broadcast on C-SPAN on December 26, 1996.
Excerpts transcribed by author from off-air audio recording.
House of Commons Debates, April 15, 1969, and November 30, 1976.
House of Commons (various years). Parliamentary Debates. London: HMSO.
Milton Friedman Speaks, 1980 video and transcript releases of lecture series delivered in 1977 and
1978; also released as Idea Channel DVD series.
506
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
II
Do Macroeconomic Announcements
Move Inflation Forecasts?
Marlene Amstad and Andreas M. Fischer
This paper presents an empirical strategy that bridges the gap between event studies and macroeconomic forecasts based on common-factor models. Event studies examine the response of financial variables to a market-sensitive “surprise” component using a narrow event window. The authors
argue that these features—narrow event window and surprise component—can be easily embedded
in common-factor models that study the real-time impact of macroeconomic announcements on
key policy variables such as inflation or gross domestic product growth. Demonstrative applications
are provided for Swiss inflation that show that (i) the communication of monetary policy announcements generates an asymmetric response for inflation forecasts, (ii) the pass-through effect of import
price releases on inflation forecasts is weak, and (iii) macroeconomic releases of real and nominal
variables generate nonsynchronized effects for inflation forecasts. (JEL E37, E52, E58)
Federal Reserve Bank of St. Louis Review, September/October 2009, 91(5, Part 2), pp. 507-18.
A
n attractive feature of diffusion indices
is their ability to embed timely information from macroeconomic releases.
Studies using common-factor procedures by Forni et al. (2000) and Stock and Watson
(2002) show that updated forecasts have lower
forecast errors because additional observations
from macroeconomic releases are included in a
growing panel. Evans (2005) and Giannone,
Reichlin, and Small (2008) develop a procedure
that updates quarterly U.S. gross domestic product (GDP) nowcasts (i.e., forecasts for the current
quarter) as information from staggered macroeconomic releases arrives. Similarly, Altissimo
et al. (2007) argue that integration of early information at a monthly frequency improves quarterly
GDP nowcasts for the euro area. At a higher frequency, Amstad and Fischer (2009a) show that
weekly updates enhance the forecast accuracy
for monthly Swiss inflation. These studies argue
that sequentially updating the forecast on incoming macroeconomic information is informative for
analysts monitoring nominal and real activity.
A drawback of diffusion indices is that they
are statistical models without economic structure.
A naive method of uncovering the driving forces
behind forecasts from common-factor models
compares the forecasting performance between
included and excluded variable blocks in the
panel. Forni et al. (2001) use this method to show
that financial variables are important for inflation
forecasts. Analogous to the naive method, the
impact of macroeconomic announcements on
indices can be interpreted using an event study
framework. The “impact effect” is defined as the
difference between the forecast conditional on
Marlene Amstad was a visiting senior economist at the Federal Reserve Bank of New York when this paper was written; she is the head of
financial market analysis at the Swiss National Bank. Andreas M. Fischer is an economic adviser at the Swiss National Bank, research fellow
at the Centre for Economic Policy Research (London), and research associate at the Globalization and Monetary Policy Institute (Federal
Reserve Bank of Dallas). The authors thank Michael Dueker, Domenico Giannone, Simon Potter, Lucrezia Reichlin, and Robert Rich for
valuable discussions. Tobias Grassli provided valuable assistance in data support.
© 2009, 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, the regional Federal Reserve Banks, or the Swiss National Bank. Articles may be
reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation
are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank
of St. Louis.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
507
Amstad and Fischer
information after the macroeconomic release and
the forecast conditional on information before the
macroeconomic release. Event studies, which
measure the impact of an economic event on a
variable of interest, have a rich tradition in macroeconomics and finance.1 These studies often work
with a narrow event window to capture the financial market response to an announcement surprise
component. In a similar manner, forecast innovations from a common-factor model centered
on a macroeconomic release with a narrow and
fixed event window lend themselves readily to
an event study interpretation.
Our objective here is to bridge the gap between
event studies examining the impact of macro
announcements for financial variables and conventional macro models embodying a broad range
of macroeconomic information. More specifically, we want to know whether macroeconomic
announcement effects for a narrow event window
have a strong impact on the inflation forecast. It
is no surprise that wide event windows—say,
more than one month—generate large forecast
revisions, but it is unclear whether the same is
true for narrow event windows of one day. The
proposed identification procedure relies on generating forecast innovations for the macroeconomic
series based on panels updated on a daily basis
using the dynamic common-factor procedure
developed by Forni et al. (2000). The one-day
event window defined by the postrelease and prerelease dates of macroeconomic releases allows
interpretation of the announcement’s impact on
inflation forecasts.
The advantages of our procedure over previous
event studies that analyze announcement effects
are twofold. The first concerns the information
breadth captured in the anticipated component
of the event. The pre-event forecast from the
common-factor model is projected on a data-rich
environment, whereas previous event studies rely
on information from simple ordinary least squares
regressions and survey data or have no prior information. The second advantage is that announce1
See MacKinlay (1997) for a survey of the literature and empirical
tests.
508
S E P T E M B E R / O C TO B E R , PA R T 2
2009
ment effects can be analyzed for a wide range of
variables. They include all the variables in the
panel. Previous event studies focused exclusively
on financial variables to capture the announcement effect.
The empirical analysis considers three applications of the event study procedure to Swiss
inflation. The case studies are demonstrative,
reflecting the view that the proposed framework
has broad applications. The first exercise examines
the size of forecast innovations on days when the
Swiss National Bank (SNB) announces its target
range for its policy interest rate. Numerous studies
surveyed by Blinder et al. (2008) have examined
the response of financial markets to central bank
communications but not whether central bank
communications can have an impact on the inflation forecast through the market’s response and
subsequent effect on financial variables. We want
to know whether the financial variables in the
panel respond to SNB announcements and, in
turn, influence the inflation forecast. The second
exercise investigates whether forecast errors generated by the release of real and nominal macroeconomic variables influence inflation forecasts
in a synchronous manner. With this information
we want to understand how forecasts behave over
the cycle. The third exercise analyzes whether
inflation forecasts respond to import price releases.
We argue that the forecast innovation centered
on import price releases can be interpreted as an
alternative measure of the pass through from
import prices to consumer prices.
The paper is organized as follows. The next
section outlines the event study procedure for
common-factor models used for real-time forecasting. Then we discuss the structure of the panels
and the forecasting windows and the event-study
applications of common-factor models to Swiss
inflation.
THE IDENTIFICATION SCHEME
The identification scheme to analyze
announcement effects in macroeconomic models
with data-rich environments involves the following steps. The first step generates the projection
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Amstad and Fischer
for the variable of interest one day before the
release of macroeconomic information. The projections are based on panels that encompass realtime information from financial variables and
data releases that are updated on a daily basis.
Estimation follows the dynamic common-factor
procedure by Forni et al. (2000). The second step
reestimates the projections for the variable of
interest one day later that include cross-sectional
information from the macroeconomic release.
The third step constructs the forecast innovation
linked to the announcement surprise—that is, the
one-day difference in the two projections. The
main steps of the estimation procedure are defined
using the terminology of MacKinlay (1997).
Defining the Event
The monthly release of macroeconomic variables is defined as the event with the kth event
date τk = 共j,t 兲 corresponding to day j and month t
in calendar time and k = {1,…, K}. We assume that
new information attributed to the event stems
from the monthly macroeconomic release. This
assumption means that updated panels at the
time of the event are not subject to data revisions
on days when macroeconomic information is
released. Ideally, we focus on macroeconomic
releases that are large in the cross section (i.e., the
consumer price index [CPI] and its subcomponents) to reduce the influence of measurement
error in estimation.
Estimation
The empirical model relies on data-reduction
techniques that can handle real-time panels that
are updated daily. We follow the estimation procedures of Forni et al. (2000), Cristadoro et al.
(2005), and Altissimo et al. (2001). We provide
an informal outline of the estimation procedure,
but readers may refer to the individual papers
for specific details.
As in Forni et al. (2000) and following their
notation, we assume that the factor structure
has N variables in the generic panel, xt =
共x1,t , x2,t ,…, xN,t 兲′, where x1,t is the variable of
interest. In most cases, x1,t is either inflation or
output. The variables in the panel are first differF E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
enced when necessary for stationarity purposes.
Next, x1,t is assumed to be the sum of two unobservable components: a signal, x*1,t , and a component capturing short-run dynamics, seasonality,
measurement error, and idiosyncratic shocks, ei,t :
(1)
x 1,t = x 1∗,t + e1,t .
The objective of the generalized dynamic factor
model of Forni et al. (2000) is to estimate the signal, x*1,t , in equation (1) using information from
the present and past of the x’s (i.e., a contemporaneous linear combination of the x’s).
More formally, it is assumed that the variables
in equation (1), x1,t , can be represented as the
sum of two stationary, mutually orthogonal,
unobserved components. The first component is
the common component, χi,t , which is assumed
to capture a high degree of comovement between
the variables in the panel, xt . The second component is the idiosyncratic component, ξi,t . The common component is defined by q common factors,
uh,t , that are possibly loaded with different coefficients and (finite) lag structures, say, of order s.
Formally, Forni et al. (2000) specify x1,t as a generalized dynamic factor model:
(2) x i ,t = χi ,t + ξi ,t = Σ qh =1Σsk = 0bi , h, kuh,t − k + ξi,t ,
where ξi,t is the idiosyncratic component and
χi,t = xi,t – ξi,t is the common component.
The estimation procedure as in Cristadoro et al.
(2005) involves three steps. The first step estimates the common factors. In particular, the cross
spectra for the common component of χ1,t are estimated following Forni et al. (2000). The second
step computes the implied covariance of χ1,t and
the factors by integrating the cross spectra over a
specified frequency band. The last step involves
performing a linear projection of the common
component on the present and the lags of the
common factors:
(3) χˆ 1,t = Pr oj χ1,t uh,t − k , h = {1,, q; k = 0,, s } .
To generate the projections at time t, we apply
the shifting procedure for the covariance matrix
by Altissimo et al. (2001; see their Appendix B.4
on filling in incomplete observations for unbalS E P T E M B E R / O C TO B E R , PA R T 2
2009
509
Amstad and Fischer
anced panels).2 Altissimo et al. (2001) compute
values of χ̂i,t g months ahead by individually shifting each series in xi,t so that the most recent observation aligns g months ahead to form a balanced
panel. Afterward the generalized principal component is evaluated for the realigned xi,t .
Announcement Surprise Component
The event study literature frequently uses
the term “abnormal returns” for the response of
financial variables to an examined event. This is
defined as the actual ex post return of the (financial) variable over the event window minus the
normal return—the return that would be expected
if the event did not take place. Instead of returns,
we work with innovations of the projections. Thus,
to identify the influence of new information from
monthly releases in import prices, a measure of
innovations for event date τk = 共j,t 兲 is needed.
This is defined as the one-day difference in the
projections of χ̂1,t around the event (i.e., the release
dates). More specifically, ε 1,t is the innovation from
the projections for χ̂1,t|Pj,t conditional on the daily
panel, Pj,t , before and after the release of the
macroeconomic variable on day j in month t:
(4)
εˆ 1,t = χˆ 1,t P − χˆ 1,t P
j ,t
.
j – 1 ,t
Similarly, the h-ahead forecast innovation for
h > 0 is εˆ1,t+h = χ̂1,t+h|Pj,t – χ̂1,t+h|Pj –1,t . Equation (4)
represents the full-day impact from the macroeconomic announcement.3
The anticipated component for inflation in
equation (4), χ̂1,t|Pj –1,t , is conditional on a broad
range of information. In a similar manner, the
anticipated component can be derived for any
variable in the panel, Pj –1,t . This represents an
improvement over earlier studies reviewed in
MacKinlay (1997) that used survey data or simple
regression techniques projected on a handful of
variables to generate the anticipated component.
2
Giannone, Reichlin, and Small (2008) offer an alternative procedure
for forecasts of the common component based on the Kalman filter,
which are qualitatively the same.
3
Event studies frequently analyze the immediate impact, which is
generally defined as the market response 30 minutes before and
30 minutes after the macroeconomic announcement, rather than
the full-day impact.
510
S E P T E M B E R / O C TO B E R , PA R T 2
2009
As discussed in Rigobon and Sack (2008), anticipated components using survey data are problematic because of irregular timing and a limited
number of surveyed panelists. Rigobon and Sack
argue that these problems contaminate the surprise
terms and generate a biased impact effect in event
studies. They propose an error-in-variables procedure to overcome these problems stemming from
survey data. Equation 4 has no such problems.
THE (DAILY) PANELS AND DATA
RELEASES
All economic series used to construct the data
panels are from the SNB’s data bank. The dataset’s
construction is intended to replicate the contours
of a data-rich environment in which the SNB
operates.
The Panels
Because we are concerned with the problem
of how to weigh the most recent information
against what we already know at daily intervals,
we are interested in economic data that are frequently released, which means working with data
with a daily or monthly frequency. Table 1 shows
the breakdown of the 434 series into nominal
and real variables and their frequency. There are
27 financial variables at the daily frequency and
407 nominal and real variables at the monthly
frequency. Quarterly variables such as industrial
production or GDP were intentionally excluded
to avoid contaminating our estimates with revision
errors.4
Two types of panels are constructed. The first
uses end-of-month data from 1993:05 to 2003:11;
we generate our initial forecasts with this panel.
After 2003:11:01, we update the panels daily. The
starting date 1993:05 is chosen because a large
number of series do not go farther back than 1990
and 1993:05 coincides with a major revision in
the CPI.
An explicit intention in constructing the dataset was to transform the series as little as possible.
4
See Amstad and Fischer (2009a) for a discussion of data revision
at the monthly frequency. Also, preliminary estimates revealed
that the introduction of the quarterly information from GDP or
industrial production did not alter our estimates.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Amstad and Fischer
Table 1
Data and Release Frequencies
Release frequency
Data category
Monthly
Daily
Total
Nominal
254
Prices (CPI total, subcomponents, cores)
178
Money
9
Financial
6
9
Interest rates
12
11
Exchange rates
4
3
Foreign prices
10
Foreign interest rates
8
4
Real
180
Survey
40
External trade
83
Labor
14
Demand
16
Foreign industrial production
8
Foreign labor market
19
Total
First, we undertake no seasonal filtering because
of its reliance on future information. Amstad
and Fischer (2009a) demonstrate that seasonal
adjustment can be treated through band-pass filtering. The absence of seasonal revisions allows
better interpretation of the forecast innovations.
Several data transformations, however, were
necessary at the initial stages of estimation. The
series were filtered in the following manner. First,
logarithms were taken for nonnegative series that
were not in rates or in percentage units to account
for possible heteroskedasticity. Second, the series
were first-differenced, if necessary, to account for
stochastic trends. Third, the series were taken in
deviation from the mean and divided by their
standard deviation to remove scalar effects.
Clustered Data Releases
Figure 1 provides an example of the clustering
of macroeconomic releases for December 2003.
The number of data releases for a particular day
is listed on the vertical axis with the calendar
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
407
27
434
dates denoted on the horizontal axis. The releases
are divided into nominal (shaded bars) and real
variables (open bars). Of interest are the clusterings on December 2 and 19. The first spike stems
from CPI releases and their subcomponents,
whereas the second is the result of the release of
trade volumes across sectors.
APPLICATIONS TO SWISS
INFLATION FORECASTS
This section presents three empirical applications of analyzing the impact of macroeconomic
announcement effects on Swiss CPI inflation.5
The case studies were chosen to reflect the view
that the event study framework for common-factor
5
The empirical model is defined in Amstad and Fischer (2009a).
The same paper provides forecasting properties for a model with
12 static factors and 2 dynamic factors. Inflation is annualized
and uses a band-pass filter at 2π/12 to remove seasonality. This is
also the same dataset and estimation procedure used to estimate
the SNB’s monthly measure of core inflation, called dynamic factor
inflation. See page O15 of the SNB’s Monthly Statistical Bulletin.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
511
Amstad and Fischer
Figure 1
Nominal and Real Data Releases for December 2003
Number of Data Releases
250
Real Variables
Nominal Variables
200
150
100
50
0
12/02/03
12/05/03
12/08/03 12/11/03
12/14/03
models has broad applications. The first exercise
considers forecast innovations generated on days
when the SNB announced its target range for the
3-month London Interbank Offering Rate (LIBOR)
in 2004. In particular, we are interested in how
financial variables respond to SNB communication and its impact on the inflation forecast. The
second application asks whether forecast innovations generated by data releases of real and nominal variables to CPI inflation are synchronized.
In other words, do the data releases from real and
nominal variables influence the inflation forecast
in a similar manner? The last application examines whether forecast innovations generated by
import price releases influence CPI inflation. In
particular, we want to know whether the impact
is similar in magnitude to pass-through ratios
estimated in other studies that use traditional
time-series methods.
512
S E P T E M B E R / O C TO B E R , PA R T 2
2009
12/17/03
12/20/03
12/23/03
12/26/03
12/29/03
SNB Announcement Surprises in 2004
The SNB defines a target range of 100 basis
points for the 3-month LIBOR as its operating
target. To steer the LIBOR within the target range,
the SNB sets the 1-week repurchase (repo) rate.
Four times per year on scheduled dates, the SNB
releases a policy statement in which it announces
a change or no change in the target range.6 In 2004,
the announcement dates were March 18, June 17,
September 16, and December 16. We use these
four policy dates to generate the SNB announcement surprises. The SNB “announcement surprise”
is defined as the one-day difference in the inflation forecast conditional on postrelease information minus the inflation forecast conditional on
prerelease information. The difference in this
information set captures only information from
6
Outside these prearranged dates, the SNB reserves the right to
change the target range.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Amstad and Fischer
(daily) financial variables and their reaction to
the policy statement (i.e., no releases of macroeconomic data were made public on the four SNB
policy dates). These differences in the panels
pertain to daily updates in the 27 financial variables in our panel.
Figure 2 plots innovations of 24-month-ahead
inflation forecasts at the time of the four SNB
announcement dates. In June and September the
SNB’s board of directors raised the target range by
25 basis points, whereas in March and in December
the target range was left unchanged. The responses
to the SNB announcement surprises differ considerably. For the March release, there is no change
in the forecast. However, for the dates when the
SNB raised its target range, we observe a strong
response in the inflation forecast but in opposite
directions. Contractionary behavior is observed
for the June rate hike and expansionary behavior
for the September rate hike. For the last announcement surprise in December, we observe a weak
but expansionary response to the “no-change”
decision. Although the forecast innovations on
days when changes to the target range are larger
than on days with no changes to the target range,
we do not find them to be statistically significant
compared with forecast innovations on SNB days
in the years between 2000 and 2003. Next, we
focus on the direction of the forecast innovation.
How do we explain the differing reactions to
the change and no-change decisions in the target
range? The release dates that signal a change in
the target range account for larger reactions in the
inflation forecast. The stronger forecast response
on SNB days with a change in the target range
rests on the fact that many financial contracts in
Switzerland (i.e., automobile leases, home and
commercial property loans) are tied to the 3-month
LIBOR. To determine the innovation’s direction,
it is necessary to control for what the markets had
anticipated. As in Hamilton and Jorda (2002), one
possible method (aside from the projection one
day before the SNB announcement day) is to use
a spread of the SNB’s policy rates: the 3-month
LIBOR rate minus the repo rate. This interest rate
spread is plotted in Figure 3 along with the midpoint in the SNB’s target range for the 3-month
LIBOR.7 The interest rate spread shows that the
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
market anticipated the rate hikes in June and
September; the spreads widen. For the no-change
decisions, the spreads do not change in March and
widen slightly before the December policy release.
To understand the postrelease estimate, we
need to examine what happens to the spread the
day after the SNB policy statements are released.
For the March release, the spread does not change
between the preforecast and postforecast. This is
consistent with the March response of no reaction
to the SNB announcement surprises. For the June
release, the change in the spread is 0.01, whereas
for the September release it is –0.14. In the latter
case, the SNB did not raise the repo rates high
enough to move the 3-month LIBOR to the midpoint of the target range. In other words, the short
end of the yield curve was steeper than was anticipated by the market. This led to a rise in the postrelease estimate of inflation. The response to the
December release of no change in the target range
is similar to the response for the September
release. Although the reaction for September is
small, the change in the spread for the postrelease
and prerelease dates of –0.04 is consistent with
the innovation’s direction.
Are Real and Nominal Forecast
Innovations Synchronized?
Next, we test whether forecast innovations
from data releases of real and nominal variables
are synchronized. We generate the forecast innovations from the monthly trade releases (i.e., “real
innovations”) and the forecast innovation from
the monthly CPI releases (i.e., “nominal innovations”). A priori, we do not expect the two types
of forecast innovations to be similar. First, the size
and dynamics of the individual forecast innovations can differ from month to month. Second, the
comovement of real and nominal innovations
should not be restricted to be the same for each
month. In related empirical studies on the procyclicality of prices in the long run, Backus and
Kehoe (1992), Ravn and Sola (1995), and Smith
(1992) find that the cyclical properties of prices
and output are not stable.
7
The repo rate is either the 1-week or the 2-week repo rate; in most
cases, it is the former. See Dueker and Fischer (2005).
S E P T E M B E R / O C TO B E R , PA R T 2
2009
513
Amstad and Fischer
Figure 2
Forecast Innovations of SNB Announcements to the Target Range
No Change in SNB Target Range: March 18, 2004
Size of Forecast Innovation
6.00E-16
5.00E-16
4.00E-16
3.00E-16
2.00E-16
1.00E-16
0.00E+00
–1.00E-16
–2.00E-16
–3.00E-16
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
20
21
22
23
24
20
21
22
23
24
20
21
22
23
24
25-Basis-Point Raise in SNB Target Range: June 16, 2004
Size of Forecast Innovation
0
–0.005
–0.01
–0.015
–0.02
–0.025
–0.03
–0.035
–0.04
1
2
3
4
Size of Forecast Innovation
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
25-Basis-Point Raise in SNB Target Range: September 16, 2004
0.018
0.016
0.014
0.012
0.01
0.008
0.006
0.004
0.002
0
–0.002
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
No Change in SNB Target Range: December 16, 2004
Size of Forecast Innovation
0.012
0.01
0.008
0.006
0.004
0.002
0
–0.002
–0.004
514
1
2
3
4
5
S E P T E M B E R / O C TO B E R , PA R T 2
6
7
2009
8
9
10
11
12
13
14
15
16
17
18
19
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Amstad and Fischer
Figure 3
SNB Policy Rates in 2004
Interest Rate
0.8
Mid-Point of SNB Target Range
0.7
3-Month Libor Rate — 1-Week Repo Rate
0.6
0.5
0.4
0.3
0.2
0.1
0
01/05/04
02/16/04
03/29/04
05/12/04
06/25/04
08/06/04
09/20/04
11/01/04
12/13/04
SOURCE: Dueker and Fischer (2005).
Table 2
Synchronization of Forecast Innovations from Nominal and Real Variable Releases
November
2004
October
2004
September
2004
August
2004
July
2004
June
2004
επn,t+h| j,t , επr,t+h| k,t
0.174
0.348
0.130
0.348
0.826
0.565
επn,t+h| j,t , επn,t+h| j–1,t
0.522
0.522
0.870
0.822
0.391
0.261
επr,t+h| k,t ,
0.610
0.740
0.565
0.478
0.652
0.434
Forecast innovations
επr,t+h| k–1,t
NOTE: The forecast innovations generated by real and nominal variable releases are denoted by επr,t+h|k,t and επn,t+h|j,t . The index for
concordance by Harding and Pagan (2002) lies between 0 (countercyclical) and 1 (procyclical). The index is calculated for the months
June through November 2004.
To test whether the two types of forecast innovations are synchronous, we calculate the concordance index of Harding and Pagan (2002). The
application of the index examines whether the
comovement of real and nominal innovations
can be quantified by the fraction that both series
are simultaneously in the same state of expansion
(St = 1) or contraction (St = 0) with the index,
measuring the degree of concordance between
series 1 and 2, which are επr,t+h|k,t and ε πn,t+h|j,t in
our case.8
The concordance index can be used to determine whether nominal and real innovations to
inflation are procyclical or countercyclical. If
they are exactly procyclical, then the index is
unity, whereas a zero value denotes evidence of
countercyclical behavior. Table 2 presents the
24
I 1,2
∑ S1,t S2,t + (1 – S1,t )(1 – S1,t ) ,
= t =1
24
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
8
The concordance index has similar properties as the Cowles-Jones
test used for testing an i.i.d. random walk process.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
515
Amstad and Fischer
Figure 4
Exchange Rate Pass Through
Import Price Shock on CPI Inflation: November 2004
Size of Forecast Innovation
0.025
0.02
0.015
0.01
0.005
0
–0.005
–0.01
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
19
20
21
22
23
24
19
20
21
22
23
24
Import Price Shock on CPI Inflation: October 2004
Size of Forecast Innovation
0.005
0
–0.005
–0.01
–0.015
–0.02
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Import Price Shock on CPI Inflation: September 2004
Size of Forecast Innovation
0.01
0.005
0
–0.005
–0.01
–0.015
–0.02
–0.025
516
1
2
3
4
5
6
S E P T E M B E R / O C TO B E R , PA R T 2
7
8
2009
9
10
11
12
13
14
15
16
17
18
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Amstad and Fischer
degree of concordance between επr,t+h|k,t and ε πn,t+h|j,t
for June to November 2004. In the first row of
the table, the index values for επr,t+h|k,t and ε πn,t+h|j,t
show that the innovations behaved in a procyclical
manner in June and July, but the real and nominal
innovations to inflation behaved in a countercyclical manner from August through November.
In the second and the third rows of the table, information on the persistence of the innovations is
given by constructing the index for ε πn,t+h|j,t and
ε πn,t+h|j –1,t and επr,t+h|k,t and επr,t+h|k–1,t . Here, the evidence shows that the likelihood of the two types
of forecast innovations behaving in the same
manner (as in the previous month) is stronger for
real innovations than for nominal innovations to
inflation. In other words, the forecast innovations
from real data releases demonstrate a higher level
of persistence than do forecast innovations from
nominal data releases.
Do Inflation Forecasts Respond to
Releases in Import Prices?
The response of CPI inflation forecasts to
import price releases should be informative about
the pass through from import prices to consumer
prices.9 In our setup, the forecast innovation
around the import price release is defined as the
difference in the 24-month-ahead forecasts in CPI
inflation based on the daily panel that includes
the postrelease information from import prices
and the previous day’s panel that entails information from the prerelease.
Figure 4 shows the response of CPI inflation
to new information from import price releases for
November, October, and September 2004. A onestandard-deviation band, based on past innovations since December 2003, is depicted around
the forecast’s response. The evidence indicates
that the pass through under this measure is small.
In November and October, the innovations of the
import prices were slightly negative for the first
15 months and zero thereafter. The response for
September was stronger; again the effect of import
prices is absorbed within 18 months.
The finding that the Swiss pass through is
9
This section relies heavily on Amstad and Fischer (2009b).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
weak in 2004:Q4 does not contradict the crosscountry evidence by Campa and Goldberg (2005),
Gagnon and Ihrig (2004), and McCarthy (2000).
These studies find that the pass through for Swiss
prices is surprisingly small compared with the
empirical evidence for other small open economies.
CONCLUSION
Understanding the influence of real-time
information on inflation forecasts is vital for
policymakers. The proposed forecasting framework based on the common-factor procedure with
daily updated panels is a step in this direction.
As in event studies that focus on the response of
high-frequency financial data to new information
around a narrow event window, the identification
scheme herein relies on the recognition that
macroeconomic announcement effects can also
be interpreted as a forecast innovation with a oneday event window. The case studies for Swiss
inflation demonstrate that the event study framework for common-factor models is flexible to
handle numerous applications in real time.
REFERENCES
Altissimo, Filippo; Bassanetti, Antonio; Cristadoro,
Riccardo; Forni, Mario; Hallin, Marc; Lippi, Marco;
Reichlin, Lucrezia and Veronese, Giovanni.
“EuroCOIN: A Real Time Coincident Indicator of the
Euro Area Business Cycle.” CEPR Discussion Paper
No. 3108, Centre for Economic Policy Research, 2001;
www.cepr.org/pubs/new-dps/dplist.asp?dpno=3108.
Altissimo, Filippo; Cristadoro, Riccardo; Forni, Mario;
Lippi, Marco and Veronese, Giovanni. “New
Eurocoin: Tracking Economic Growth in Real Time.”
Banca d’ Italia, Temi di discussionedel Servizio
Studi No. 631, Bank of Italy, June 2007;
www.bancaditalia.it/pubblicazioni/econo/temidi/
td07/td631_07/td631/en_tema_631.pdf.
Amstad, Marlene and Fischer, Andreas M. “Are
Weekly Inflation Forecasts Informative?” Oxford
Bulletin of Economics and Statistics, April 2009a,
71(2), pp. 236-52.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
517
Amstad and Fischer
Amstad, Marlene and Fischer, Andreas M. “Monthly
Pass-Through Ratios.” Globalization and Monetary
Policy Institute Working Paper No. 26, Federal
Reserve Bank of Dallas, January 2009b;
www.dallasfed.org/institute/wpapers/2009/0026.pdf.
Backus, David K. and Kehoe, Patrick J. “International
Evidence on the Historical Properties of Business
Cycles.” American Economic Review, September
2002, 82(4), pp. 864-88.
Blinder, Alan S.; Ehrmann, Michael; Fratzcher, Marcel;
De Haan, Jakob and Jansen, David-Jan. “Central
Bank Communication and Monetary Policy:
A Survey of Theory and Evidence.” Journal of
Economic Literature, December 2008, 46(4),
pp. 910-45.
Campa, Jose M. and Goldberg, Linda S. “Exchange
Rate Pass-Through into Import Prices.” Review of
Economics and Statistics, November 2005, 87(4),
pp. 679-90.
Cristadora, Riccardo; Forni, Mario; Reichlin, Lucrezia
and Veronese, Giovanni. “A Core Inflation Indicator
for the Euro Area.” Journal of Money, Credit, and
Banking, June 2005, 37(3), pp. 539-60.
Dueker, Michael J. and Fischer, Andreas M. “Open
Mouth Operations: A Swiss Case Study.” Federal
Reserve Bank of St. Louis Monetary Trends, January
2005; http://research.stlouisfed.org/publications/
mt/20050101/cover.pdf.
Evans, Martin D.D. “Where Are We Now? Real-Time
Estimates of the Macro Economy.” International
Journal of Central Banking, September 2005, 1,
pp. 127-75.
Forni, Mario; Hallin, Marc; Lippi, Marco and Reichlin,
Lucrezia. “The Generalized Dynamic-Factor Model:
Identification and Estimation.” Review of Economics
and Statistics, 2000, 82(4), pp. 540-54.
Forni, Mario; Hallin, Marc; Lippi, Marco and Reichlin,
Lucrezia. “Do Financial Variables Help Inflation
and Real Activity in the Euro Area?” Journal of
Monetary Economics, 2001, 50(6), pp. 1243-55.
518
S E P T E M B E R / O C TO B E R , PA R T 2
2009
Gagnon, Joseph E. and Ihrig, Jane. “Monetary Policy
and Exchange Rate Pass-Through.” International
Journal of Finance and Economics, 2004, 9(4),
pp. 315-38.
Giannone, Domenico; Reichlin, Lucrezia and Small,
David. “Nowcasting: The Real-Time Informational
Content of Macroeconomic Data.” Journal of
Monetary Economics, May 2008, 55(4), pp. 665-76.
Hamilton, James D. and Jorda, Oscar. “A Model for
the Federal Funds Rate Target.” Journal of Political
Economy, October 2002, 110(5), pp. 1135-67.
Harding, Don and Pagan, Adrian. “Dissecting the
Cycle: A Methodological Investigation.” Journal of
Monetary Economics, March 2002, 49(2), pp. 365-81.
MacKinlay, A. Craig. “Event Studies in Economics
and Finance.” Journal of Economic Literature, 1997,
35(1), pp. 13-39.
McCarthy, Jonathan. “Pass-Through of Exchange Rates
and Import Prices to Domestic Inflation in Some
Industrialized Economies.” Staff Report No. 111,
Federal Reserve Bank of New York, September 2000;
www.ny.frb.org/research/staff_reports/sr111.pdf.
Ravn, Morten O. and Sola, Martin. “Stylized Facts
and Regime Changes: Are Prices Procyclical?”
Journal of Monetary Economics, December 1995,
36(1), pp. 497-526.
Rigobon, Robert and Sack, Brian. “Noisy Macroeconomic
Announcements, Monetary Policy, and Asset Prices,”
in John Campbell, ed., Asset Prices and Monetary
Policy (NBER). Chap. 8. Chicago: University of
Chicago Press, 2008, pp. 335-70.
Smith, R. Todd. “The Cyclical Behavior of Prices.”
Journal of Money, Credit, and Banking, November
1992, 24(4), pp. 413-30.
Stock, James H. and Watson, Mark W. “Macroeconomic
Forecasting Using Diffusion Indexes.” Journal of
Business and Economic Statistics, April 2002,
20(2), pp. 147-61.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
II
Challenges in Macro-Finance Modeling
Don H. Kim
This article discusses various challenges in the specification and implementation of “macrofinance” models in which macroeconomic variables and term structure variables are modeled
together in a no-arbitrage framework. The author classifies macro-finance models into pure latentfactor models (“internal basis models”) and models that have observed macroeconomic variables
as state variables (“external basis models”) and examines the underlying assumptions behind these
models. Particular attention is paid to the issue of unspanned short-run fluctuations in macroeconomic variables and their potentially adverse effect on the specification of external basis models.
The author also discusses the challenge of addressing features such as structural breaks and timevarying inflation uncertainty. Empirical difficulties in the estimation and evaluation of macrofinance models are also discussed in detail. (JEL E43, E44, G12)
Federal Reserve Bank of St. Louis Review, September/October 2009, 91(5, Part 2), pp. 519-44.
I
n recent years there has been much interest
in developing “macro-finance models,” in
which yields on nominal bonds are jointly
modeled with one or more macroeconomic
variables within a no-arbitrage framework.
Academic researchers and policymakers alike
have long recognized the need to go beyond
“nominal yields only” no-arbitrage models (i.e.,
to include a description of the macroeconomy or
other asset prices). Campbell, Lo, and MacKinlay
(1996), for example, have emphasized that “as
the term structure literature moves forward, it
will be important to integrate it with the rest of
the asset pricing literature.” Policymakers have
often used traditional theories such as the expectations hypothesis and the Fisher hypothesis to
extract an approximate measure of market expectations of interest rates and macroeconomic variables such as inflation, but they are also aware
that risk premia and other factors might compli-
cate the interpretation of the information in the
yield curve; policymakers therefore would welcome any progress in term structure modeling
that would facilitate greater understanding of
the messages in the yield curve.1
Despite much exciting work in macro-finance
modeling of late,2 as a central bank economist
who monitors markets regularly, I have found it
difficult to bring the current generation of models
to bear on the practical analysis of bond market
developments or to implement the models in real
time to obtain a reliable measure of the market’s
expectation of key variables such as inflation.3
1
See, for example, Bernanke (2004a).
2
Examples include Ang and Piazzesi (2003), Hördahl, Tristani, and
Vestin (2006), Rudebusch and Wu (2003), and Ang, Bekaert, and
Wei (2007 and 2008).
3
I emphasize that I speak as one of many central bank economists
and that my views as stated in this paper do not necessarily represent the general consensus among economists at central banks.
Don H. Kim is an assistant professor of finance at the Yonsei University School of Business. This article is a revised and abridged version of
a paper that appeared in the Bank for International Settlements Working Paper Series and the Board of Governors of the Federal Reserve System
Finance and Economics Discussion Series. The author thanks Claudio Borio, Stefania D’Amico, Peter Hördahl, Deb Lindner, Athanasios
Orphanides, Frank Packer, Min Wei, Jonathan Wright, and the seminar participants at the Bank of England and the BIS for helpful discussions.
© 2009, 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.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
519
Kim
The academic literature provides little evidence
in this regard (either for or against macro-finance
models). One exception is the recent paper of
Ang, Bekaert, and Wei (2007, ABW), who performed an extensive investigation of the out-ofsample inflation forecasting performance of
various models and survey forecasts. The authors
found that the no-arbitrage models they used
perform worse than not only survey forecasts but
also other types of models.4
It thus seems useful to review and discuss
various challenges in the specification and implementation of macro-finance models that might
help shed light on the lack of documented practicality of macro-finance models in general and
on the findings of ABW in particular. To this end,
I take a closer look at the role of the no-arbitrage
principle in macro-finance models and reconsider
the assumptions often made in this literature. The
no-arbitrage principle itself is clearly a reasonable
assumption, but the models also make additional
assumptions whose validity may not have been
discussed thoroughly in the existing literature. I
also discuss “more advanced” issues (such as
structural breaks and time-varying volatility) that
require going beyond the standard affine-Gaussian
framework of most macro-finance models and
the challenges encountered in this regard. Much
of the challenge in macro-finance modeling is
empirical; hence, I also discuss at length the difficulties in the implementation stage (estimation
and evaluation of models). Although the main
focus of this article is the extraction of information from the yield curve (particularly inflation
expectations), much of the discussion may be
relevant for macro-finance models developed to
address other issues, as they share some of the
key assumptions discussed in this article.
The state variables in the reduced-form noarbitrage model framework (on which most macro4
For example, the root mean square errors (RMSEs) for 1-year consumer price index (CPI) inflation forecasts based on the two noarbitrage models in ABW (what they refer to as MDL1 and MDL2)
are larger than those of autoregressive (AR)(1) and autoregressive
moving average (ARMA)(1,1) models by more than 30 percent for
the post-1995 window. Furthermore, all but 1 of ABW’s 11 regression models that involve term structure variables (what they refer
to as TS1-TS11) produce smaller RMSEs in forecasting 1-year CPI
(PUNEW) inflation than the no-arbitrage models in the post-1995
sample.
520
S E P T E M B E R / O C TO B E R , PA R T 2
2009
finance models are based) can be heuristically
viewed as forming a basis onto which to project
information in yields and other data. Herein I
make a distinction between models that use (what
I shall call) an “internal basis” versus those that
use an “external basis.” By an internal basis, I refer
to a basis that is determined inside the estimation;
hence, it is unknown before the estimation. Latentfactor models that describe inflation expectations
and term structure jointly (e.g., Sangvinatsos and
Wachter, 2005, and D’Amico, Kim, and Wei, 2008)
are examples of internal basis models. By an
external basis, I mean a basis that is a priori fixed
completely or partially, as when a specific macroeconomic variable (such as inflation) is taken as
a state variable. Note that no-arbitrage guarantees
the existence of some pricing kernel, but it does
not mean that the pricing kernel can be represented well by a priori selected variables. I shall
argue that external basis models involve strong
assumptions, and I discuss potential problems
that may occur with their use. All is not well with
internal basis models either: The weaker assumptions of these models may come at the cost of the
ability to give specific, intuitive interpretation of
the yield curve movements. Most important, internal basis models face many empirical difficulties
similar to those in the estimation of external basis
models, in particular, overfitting and small-sample
problems.
The remainder of this article is organized as
follows. The next section reviews the standard
affine-Gaussian setup of macro-finance models,
derives the affine bond pricing formula in a way
that emphasizes the replicating portfolio intuition,
and introduces a distinction between internal
basis models and external basis models. A critical
examination of the assumptions in both “lowdimensional” and “high-dimensional” external
basis models follows this review. Next, I then
discuss the challenge of accommodating nonlinear/non-Gaussian effects, such as structural
breaks and time-varying uncertainties, and potential problems with empirical techniques commonly used in the estimation and evaluation of
macro-finance models. I then return to why surveys perform better than models in inflation forecasting (as documented by ABW).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
THE BASIC MODEL
Affine-Gaussian Framework
(4)
Most macro-finance models in the literature
are based on the “affine-Gaussian” model,
denoted as
(1)
1
mt +1 ≡ logM t = − r ( xt ) − λ ( xt )′ εt +1 − λ ( xt )′ λ ( xt )
2
xt +1 = Φxt + ( I − Φ ) µ + Σεt +1,
0n ×1 = ρ + Bτ − ( Φ − ΣΛb )′ Bτ −1 ,
with boundary condition Aτ = 0 = 0, Bτ = 0 = 0n × 1
(see, for example, Ang and Piazzesi, 2003 [AP]).
The bond yield yτ,t 共= –log共Pτ,t 兲/τ 兲 is given by
(5)
r ( xt ) = ρo + ρ ′xt
λ ( xt ) = λa + Λb xt ,
where Mt is the pricing kernel, xt is an n-dimensional vector of state variables, rt is the nominal
short rate (i.e., one-period yield), and λt is the
market price of risk of the n-dimensional shocks
ε t+1 (Φ, Σ, and Λb are n × n constant matrices, ρ
and λa are constant n-dimensional vectors, and
ρo is a constant). A well-known result in finance
theory states that no-arbitrage implies the existence
of a pricing kernel (stochastic discount factor) of
the form (1).5
There is freedom in choosing the specific functional form of rt and λt and the dynamics of xt .
Use of the affine forms for rt and λt and the
Gaussian specification (VAR共1兲 specification) of
xt constitutes the affine-Gaussian model. This
form has certain limitations (discussed later), but
it is still quite general and capable of encompassing many of the known models in finance and
macroeconomics.
Using the recursion relation for the price of a
τ-period zero-coupon bond at time t,
(2)
Pτ ,t = E t ( Pτ −1,t +1M t +1 ),
it is straightforward to show that bond prices in
this model are given by
1
0 = ρo + Aτ − Aτ −1 − Bτ′ −1 ΣΣ′Bτ −1
2
− Bτ′ −1 (( I − Φ ) µ − Σλa ) an
nd
1
1
y τ ,t = − Aτ − Bτ′ xt ,
τ
τ
that is, it takes an affine form.
The original “finance term structure models”
such as those by Dai and Singleton (2000) and
Duffee (2002) were written for nominal bond
yields only. For example, the model defined by
equation (1) could be estimated with just nominal
yields data, with suitable (normalization) restrictions on the parameters Φ, µ, ρ,... to ensure that
the model be econometrically identified. The state
variables in this case are “latent factors” without
an explicit economic meaning.
In a seminal paper, AP proposed combining
this setup with a description of the macroeconomy. Their basic insight is that the well-known
Taylor-rule specification of the short rate also
has an affine form:
(6)
rt = ρπ π tY + ρg gapt + const,
where πtY is the annual inflation and gapt is the
gross domestic product (GDP) gap (log GDP minus
log potential GDP).6 Therefore, using variables
such as inflation and the GDP gap as part of the
state vector in equation (1), that is,
(7)
xt = πtY , gapt ,... ′ ,
where Aτ and Bτ are the solution of the difference
equations,
provides a system in which bond yields are linked
to key macroeconomic variables. Some macroeconomic variables might not be well described by
simple VAR共1兲 dynamics, but this is, in principle,
not a problem, as a higher-order VAR process
(VAR共q兲 model) can be written as a VAR共1兲 process
5
6
(3)
Pτ ,t = exp ( Aτ + Bτ′ xt ),
Duffie (2001) discusses this in the continuous-time formalism;
see also Cochrane (2001).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
To be precise, AP use GDP growth, instead of the GDP gap, in
their formulation.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
521
Kim
with an expanded state vector that includes
lags of these variables (e.g., [πtY, πtY–1,…, gapt ,
gapt –1 ,…]′).
Various macro-finance models differ by the
choice of the restrictions imposed on the matrices
(like Φ, ρ,..., etc.). For example, AP adopt an
atheoretical (statistical) approach, reminiscent
of Sims’s (1980) original VAR proposal; Hördahl,
Tristani, and Vestin (2006; HTV) impose more
structure, based on a New Keynesian model as
in Clarida, Galí, and Gertler (2000), though still
remain in the reduced-form framework.
These models are an innovation from the earlier approach of handling long-term bond yields
in macroeconomic models. In fact, most macroeconomic models have not dealt with long-term
bond yields at all, despite their importance for
savings and investment decisions in the economy.
Precursors to the macro-finance models, such as
the Federal Reserve’s FRB/US model, do contain
the 5-year and 10-year nominal yields, which are
specified as the expectations hypothesis–implied
yield plus a term premium (the 5-year term premium and the 10-year term premium are modeled
separately),7 but the framework (equation (1))
allows not just a few selected long-term yields but
information from the entire yield curve to be integrated with a description of the macroeconomy.
No-Arbitrage and Replicating Portfolios
Although the derivation of the affine bond
pricing equation (3) using the recursion relation
involving the pricing kernel is simple and elegant,
it is useful to re-derive it using the hedging (spanning) argument8 to get a better sense of the role
that the no-arbitrage principle plays in macrofinance models. Suppose there are n-dimensional
shocks underlying the term structure movements,
denoted by a standard normal random vector εt .
The change in the value of a bond with maturity
τ can be expressed generally as
δ Pτ ,t +1
= µτ ,t + γ τ′ ,t εt +1 ,
Pτ ,t
(8)
where I have used the notation δ Pτ,t +1 for Pτ –1,t +1
– Pτ,t (the change in the value of a bond which
was of time-to-maturity τ at time t) to avoid confusion with simple time-differencing; ∆Pτ,t +1 =
Pτ,t +1 – Pτ,t ; µτ,t is the one-period expected return
on a bond that has time-to-maturity τ at time t
(i.e., µτ,t = Et 共δ Pτ,t+1/Pτ,t 兲); and the n-dimensional
vector γτ,t is the loading on the shocks that determine the unexpected return.
Consider a portfolio formed by taking positions in n + 1 bonds with maturities τ1, τ2,…,τn+1,
with portfolio weights w1t ,…,wn+1,t . Denoting the
value of this portfolio, the return on the portfolio
V is given by
δ Pτ 1
δ Pτ n +1
δV
= w1
+ ... + w n +1
Pτ n +1
Pτ 1
V
(9)
n +1
=
∑ w i µτ
i
i =1
n +1
′
+ ∑ w i γ τ i ε,
i =1
where the time index t has been suppressed for
notational simplicity. If the portfolio is locally
risk-free
n +1
∑ w i γ τ i = 0 ,
i =1
then by no-arbitrage it should yield a risk-free
rate (one-period yield); that is,
n +1
∑ w it µτ ,t = rt ,
i
i =1
which is equivalent to
n +1
∑ w it ( µτ ,t − rt ) = 0
i
i =1
since
n +1
∑ w it = 1 .
i =1
Summarizing, we have
7
See, for example, Brayton et al. (1997).
8
The derivation here can be viewed as a discrete-time analogue of
Cox, Ingersoll, and Ross’s (1981) continuous-time derivation.
522
S E P T E M B E R / O C TO B E R , PA R T 2
n +1
2009
(10)
n +1
∑ w it ( µτ ,t − rt ) = 0, ∑ w it γ τ ,t = 0n ×1.
i
i =1
i
i =1
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
Recall that the latter of these equations is
n-dimensional, since γτi,t ’s are n-dimensional.
Equation (10) can be put in the matrix form,
(11)
µτ 1 ,t − rt µτ n +1 ,t − rt
γ τ 1 ,t
γ τ n +1 ,t
w
t
= 0( n +1)×1 ,
where wt = [w1t ,…,wn+1,t ]′. In order for this matrix
equation to have a nontrivial (i.e., nonzero) solution wt for an arbitrary choice of τi’s, the expected
excess return µτ,t – rt has to be a linear combination of γτ,t ; that is,
(12)
µτ ,t − rt = γ τ′ ,t λt ,
where the n-dimensional vector λ t (“market price
of risk”) expresses the linear dependence between
µτ,t – rt and γτ,t .
It is often more convenient to deal with log
prices and log returns on bonds, δ logPτ,t +1
共=logPτ –1,t +1 – logPτ,t 兲. From the discrete-time
version of Ito’s lemma,9 one has
(13)
δ logPτ ,t +1 = µ τ ,t + γ τ′ ,t εt +1,
where
1
δP 1
δP
(14) µ τ ,t = E t
− var
= µτ ,t − γ τ′ ,t γ τ ,t .
P 2 t P
2
Thus, equation (12) can be also written
(15)
1
µ τ ,t − rt + γ τ′ ,t γ τ ,t = γ τ′ ,t λt .
2
Note that the derivation thus far has been
quite general. If the short rate and market price
of risk are affine in the state variables and if the
state variables follow a VAR(1) process (i.e., equation (1)), one obtains a particularly simple result.
Positing that the bond prices take the form logPτ,t
= Aτ + B′τ xt , one has (from equation (13))
(16) µ τ ,t = Aτ −1 − Aτ + Bτ′ −1 ( I − Φ ) µ + ( Bτ′ −1Φ − Bτ′ ) xt
(17)
γ τ′ ,t = Bτ′ −1 Σ.
Substituting these (and the expressions for rt and
λ t ) into equation (15) gives the same difference
equation for bond prices as in equation (4)—hence,
the same bond prices, as promised earlier.
Internal Basis Models versus External
Basis Models
The key formula in the above derivation of
the bond pricing equation is equation (12), or
equivalently, equation (15). It states that the
expected return on a bond of arbitrary maturity
in excess of the short rate depends on the product
of the bond-independent market price of risk, λ t ,
and the bond’s sensitivity to risk, γτ,t . The basic
intuition underlying equation (12) is that the yield
curve is “smooth,” so the risks to a bond can be
hedged well by a portfolio of (a relatively small
number of) other bonds. This is well known from
the factor analysis of Litterman and Scheinkman
(1991) and other studies. One can also see this
from the regression of the quarterly change in the
5-year yield on the changes in 6-month, 2-year, and
10-year yields, which gives very high R-squareds
(e.g., 99 percent).
Note that equation (12) itself is silent about
the structure of the λ t vector, except for the condition that it does not depend on bond-specific
information (like maturity). In fact, the early generation of affine-Gaussian models assumed a constant market price of risk vector λ , which in effect
implied a version of the expectations hypothesis.
Later studies recognized that λ t can depend on
the state of economy; thus, a variable influencing
the market price of risk would also influence bond
prices.10 However, this creates, in a sense, too
large a set of possibilities: Any variable could, in
principle, enter the expression for the market price
of risk and, in turn, the expression for bond yields.
Latent-factor models of the term structure,
such as the affine-Gaussian model of Duffee (2002)
(the EA0共n兲 model in Duffee’s terminology), partly
get around this problem by implicitly defining
the model in statistical terms. A “maximally
flexible” n-dimensional affine-Gaussian model
(1) can be viewed as an answer to the question,
“what is the most general n-dimensional representation of the yield dynamics in which yields
10
9
See, for example, Campbell, Chan, and Viceira (2003).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Thus, a shock that changes λt, say ξt, should also be included in
the vector of shocks εt that moves bond prices.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
523
Kim
are Gaussian, linear in some basis, and consistent
with no arbitrage?” As the yield curve seems to be
well described by a small number of risk sources,
it stands to reason that there exists a suitable
representation for a relatively small n. Thus, the
no-arbitrage principle in this setting can help
describe the rich variation of the yield curve in a
tractable and relatively parsimonious way, while
allowing for a general pricing of risk (as opposed
to the expectations hypothesis).
Duffee’s (2002) affine-Gaussian model
describes only the nominal yield curve, but it is
straightforward to write down a “joint model” of
nominal yields and inflation in the same spirit
by combining equation (1) with the following
specification of the inflation process:
(18)
π t +1 = χ ( xt ) + σ ′εt +1
χ ( xt ) = ψ 0 + ψ ′ xt ,
where the one-period inflation πt +1共=log共Q t +1/Q t 兲,
Qt being the price level) consists of the one-period
expected inflation χ 共xt 兲 and unforecastable inflation σ̃ ′ε̃t+1. As in the case of the nominal short rate
rt , the one-period inflation expectation is specified
as an affine function of the state vector xt . The
disturbance vector ε̃t includes the vector of shocks
that move interest rates (εt in equation (1)) and a
shock (say ε Ⲛt ) that is orthogonal to the interest
rate shocks.11 As in the nominal-yields-only
model, the state vector xt is a vector of statistical
variables (latent variables), which is determined
only up to normalization restrictions (on parameter matrices Φ, ρo, ρ, ψo, ψ,...) that ensure the
(maximal) identification of the model. I shall refer
to such a model as an “internal basis model,” as
the state vector is unknown before the estimation
and is determined inside the estimation with
yields, inflation, and possibly other data.12
Such a joint model makes only fairly weak
assumptions: Writing the one-period inflation as
11
12
This shock (ε Ⲛt ) is introduced to allow for shocks to inflation that
are not spanned by interest rate shocks. (One can also define the
εt vector in equation (1) to include this shock.)
Perhaps the best-known example of internal basis models is factor
analysis (e.g., Litterman and Scheinkman, 1991). As in the noarbitrage internal basis models, the factors in factor analysis are
determined only up to an invariant transformation; thus, normalization restrictions are needed to define them.
524
S E P T E M B E R / O C TO B E R , PA R T 2
2009
the sum of expected inflation and unexpected
inflation in equation (18) is quite general, and it
makes intuitive sense to have the state vector xt
describe inflation expectations and bond yields
together, as a variable that moves inflation expectation would also be expected to move nominal
interest rates. At the same time, this formulation
relaxes the assumptions implicit in the two traditional theories of nominal yields: It goes beyond
the expectations hypothesis—as it now allows
for time-varying term premia—and the Fisher
hypothesis—as it now implicitly allows for a
general correlation between real rates and inflation. Note that the state vector xt in the joint model
has more economic meaning than the nominalyields-only model in the sense that it is now
(implicitly) related to objects such as inflation
expectations and inflation risk premia. However,
the fact that the xit ’s are still latent factors is potentially an unattractive feature and makes it difficult to discuss bond market developments in a
simple manner.
Thus, many papers in the macro-finance literature take all or part of the state vector to be specific macroeconomic variables (or variables with
clear macroeconomic interpretation) so as to make
the connection between the yield curve and
macroeconomy more explicit. These variables
form an external basis, in the sense that they are
a priori fixed, partially (“mixed” models) or completely (observables-only models). Simply speaking, internal basis models try to project information
in yields yτ,t and “observable” macroeconomic
variables f ito onto the state vector xt consisting of
unobservable variables f itu, while external basis
models try to project information in yields onto
“observable” macroeconomic variables f ito and
latent variables (if there are any). Schematically,
′
internal basis : { y τ ,t }, f ito ⇒ xt = f1ut , f2ut ,...
{ }
′
external basis : { y τ ,t } ⇒ xt = f1ot , f2ot ,..., f1ut , f2ut ,... .
As one moves on to external basis models, one
might be also moving away from the relative comfort of the original intuition behind no-arbitrage
(the smoothness of the yield curve); hence, close
scrutiny of the additional assumptions they
involve is warranted.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
EXAMINING THE ASSUMPTIONS
IN EXTERNAL BASIS MODELS
Unspanned Short-Run Inflation
One implication of having a macroeconomic
variable like inflation as a state variable in the setup of equation (1) is that short-run inflation risk
can be hedged by taking positions in nominal
bonds.13 Many practitioners, however, would be
skeptical about this claim. Policymakers are well
aware of large short-run variations in price indices
such as the producer price index (PPI) and consumer price index (CPI) that do not require a policy
response, and they are careful to “smooth through
the noise” in interpreting data on inflation. Blinder
(1997) puts this clearly and strongly: “[The noise
issue] was my principal concern as Vice-Chairman
of the Federal Reserve. I think it is a principal
concern of central bankers everywhere.”
Market participants are also (implicitly) cognizant of these issues. One evidence is the bond
market’s reaction to the announcement of total
CPI (also called “headline CPI” or simply “CPI”)
and core CPI (which is an inflation measure
obtained by stripping out the volatile food and
energy prices from total CPI): Bond yields are
known to react mainly to the surprise component
of core CPI, not total CPI.14 This raises the question
whether it is reasonable to treat the fluctuation
in total CPI as risks that are spanned by the yield
curve factors (an implicit assumption in most
external basis macro-finance models).
One can also consider the regression of the
change in quarterly inflation onto the changes
in 6-month, 2-year, and 10-year yields,15 which
13
Let the first element of the state vector xt in equation (1) be inflation.
The formalism (1) then implies that one can in general form a
portfolio of bonds that replicates the inflation shock ε1t.
14
One can regress the change in bond yield (around the announcement) on the surprise components in total CPI and core CPI (computed as the announced number minus the Bloomberg consensus
prediction). The coefficient on the total CPI surprise is found to
be insignificant.
15
Let yt denote a vector of n yields. In the affine model, one has
yt = a + Bxt, where xt is the n-dimensional vector of state variables,
a is an n-dimensional constant vector, and B is an n × n constant
matrix. Inverting it gives xt = B –1a + B –1yt . Thus, if πt is an element
of xt , this implies ∆πt = c ′∆yt , where c is a constant vector. This is
in the linear regression form without the residual error term (and
the intercept term).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
gives an R 2 of at most 10 percent in the 1965-2006
period, in stark contrast to the aforementioned
regression of the change in the 5-year yield (R 2 of
99 percent). Even when the lagged inflation terms
are included, as in
(19)
∆πt = a + b1 ∆πt −1 + b2 ∆π t −2 + b3 ∆π t −3
+b4 ∆y 6 M ,t + b5 ∆y 2Y ,t + b6 ∆y 10Y ,t + et ,
the R 2’s do not exceed 40 percent16; the use of
more than three yields does not make much difference. This exercise is similar in spirit to CollinDufresne and Goldstein (2002), who argue that
the relatively low R 2’s in the regressions of the
changes in interest rate derivative prices on the
changes in interest rates indicate the presence of
“unspanned stochastic volatility” in interest rates.
Note lastly that, although we have focused on
inflation (CPI) here, the concern about unspanned
shocks in macroeconomic variables is more general; for example, variables such as quarterly GDP
growth face similar problems.
Do Macro Variables Form a Suitable
Basis for Representing Expectations?
Let us now address a related question: whether
external basis models can properly describe inflation expectations, which, according to the Fisher
hypothesis intuition, is an important determinant
of the nominal term structure.
To those who engage in inflation forecasting
extensively, the poor inflation forecast performance of macro-finance models like those of ABW
might not be a surprise: A long line of research
has explored the inflation forecasting performance
of the yield curve information and generally
obtained disappointing results. Stock and Watson
(2003) summarize the situation thus: “With some
notable exceptions, the papers in this literature
generally find that there is little or no marginal
information content in the nominal interest rate
term structure for future inflation.”
Most of the regression-based inflation forecasting models in the literature include current and
16
In the regression (19), I have tried quarterly inflation based on
both the quarter-averaged CPI and the end-of-quarter (last month
of the quarter) CPI. In these cases, the quarter-averaged yields and
end-of-quarter yields were used, respectively.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
525
Kim
lagged inflation as regressors to take into account
the persistence of inflation. The expected inflation
over the next year in these models takes the form
(20)
(
)
Et π t +1Y ,t = a
+ b0π t∗
+ b1π t∗−1
+ + c ′zt ,
where π *t is either the one-period inflation or
annual inflation and the vector zt denotes other
regressors, which could include term structure
variables.
Consider a macro-finance model (1) that has
quarterly (one-period) inflation πt as a state variable. In other words, xt = [πt , z̃t ]′, where z̃t 共=[z̃1t ,
z̃2t ,…]′兲 denotes other state variables. The expected
inflation over the next year is
(
Et π t +1Y ,t
((
)
1 1t
)
2 2t
which is in the same form as equation (20).17
(The case is similar with models that use annual
inflation as a state variable.) As such, the difference between the macro-finance models formulated this way and the regression models is simply
in the coefficients, not in the basis. There is a possibility of an “efficiency gain” with no-arbitrage
models (through the imposition of useful constraints on the coefficients), but even this is not
ensured if the results in ABW are any indication.
More fundamentally, though, the frequently poor
inflation forecast performance of regression models
and macro-finance models like ABW raises questions about the efficacy of the basis itself.
Lessons from Simple Models
Some of the key conceptual issues in the representation of the yield curve and inflation expectations may be explained through a comparison
of two simple models of inflation—namely, AR共1兲
and ARMA共1,1兲 models:
17
If the z̃t vector includes latent factors, they can be “inverted” and
expressed in terms of yields (because of the linear relationship
between the yield and the factors), which again leads to the form
of equation (20). However, latent factors in these “mixed” models
are partly defined by their relation to the macro factors, and this
may entail complications, as discussed later in the “LowDimensional External Basis Models” subsection.
526
π t = (1 − φ ) µ + φπt −1 + εt ( AR ),
(23) π t = (1 − φ ) µ + φπ t −1 + εt − α εt −1 ( ARMA ).
The τ-period-ahead inflation expectations in
both models take the form
(24)
E t (π t + τ ) = φ τ −1 ( χt − µ ) + µ,
where the expected one-period inflation χt ⬅
Et 共πt +1兲 for the AR共1兲 model is given by
χt = φπ t + (1 − φ ) µ
(25)
and χt for the ARMA共1,1兲 model is given by
χt = φπ t − αεt + (1 − φ ) µ.
(26)
)
2
3
4
(21) = [1,0,…,0 ] Φ + Φ + Φ + Φ ( xt − µ ) + µ
= a + b π + b z + b z + ,
0 t
(22)
S E P T E M B E R / O C TO B E R , PA R T 2
2009
The estimate of φ in the AR共1兲 model, based on
U.S. quarterly CPI inflation data from 1960:Q1 to
2005:Q4, is 0.785(0.045), while the estimates of φ
and α in the ARMA共1,1兲 model are 0.935(0.031) and
0.341(0.081), respectively, with standard errors listed
in parentheses. These numbers imply fairly similar 1-quarter-ahead inflation expectations, as can
be seen in Figure 1A. (There is somewhat more
jaggedness in the AR共1兲 forecast.) The same parameter estimates, however, imply very different
longer-horizon inflation expectations (Figure 1B):
The 5-year-ahead (20-quarter-ahead) inflation
expectation from the AR共1兲 model is almost constant, while the 5-year-ahead inflation expectation
from the ARMA共1,1兲 model is more variable. (This
reflects the difference between 0.785 20–1 = 0.01
versus 0.93520–1 = 0.28 in equation (24).)
An almost constant 5-year-ahead inflation
expectation from the AR共1兲 model in the past 40
years is implausible. The main reason for the
qualitative difference between the AR and ARMA
models is that the ARMA共1,1兲 model tries to
separate the “unforecastable inflation” from the
expected inflation, while the AR共1兲 model does
not. This can be seen from the fact that the
ARMA共1,1兲 model is a univariate representation
of the following “two-component model”:
π t = χt −1 + ηt
(27) χt = (1 − φ ) µ + φχt −1 + et ,
(
)
(
)
ηt N 0,σ π2 , et N 0,σ 2x , corr (ηt ,et ) = ,
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
Figure 1
U.S. Inflation Expectations Based on AR(1) and ARMA(1,1) Models
A. 1-Quarter-Ahead Inflation Expectations
Percent
15
ARMA(1,1)
AR(1)
10
5
0
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
B. 5-Year-Ahead Inflation Expectations
Percent
8
ARMA(1,1)
AR(1)
6
4
2
1960
1965
1970
1975
1980
in which χt is an expected inflation process and
ηt is an unforecastable inflation.18 Though simple,
this two-component model (of which the internal
basis model [equation (18)], previously discussed
in the “Internal Versus External Basis Models”
subsection, can be viewed as an extension) is quite
useful for illustrating some of the key points in
this paper.19
18
The MA共1兲 coefficient in the ARMA共1,1兲 model is related to the
two-component model parameters as
1985
(
)
(
)
c ≡ 1 + φ 2 σ π2 + σ 2x − 2φσ π σ x 2 φσ π2 − σ π σ x .
See, for a derivation, Cochrane (2001, pp. 418-20).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
1995
2000
2005
The unforecastable inflation component ηt in
equation (27) can help explain several puzzling
empirical results in the literature. Among them
is the negative one-lag autocorrelation of the
changes in quarterly inflation ∆πt 共= πt – πt –1兲,
which, according to Rudd and Whelan (2006,
Sec III.C), is evidence against the New Keynesian
Phillips curve models (which generate positive
one-lag autocorrelation). In the case of the twocomponent model (27), one has
19
α = c − c 2 − 1, where
1990
This model of inflation has an interesting parallel with the
consumption-based asset pricing model of Bansal and Yaron (2004),
who argue that writing the consumption growth ∆ct as the sum of
expected component and unexpected component (∆ct = χt + ηt +1 )
can help resolve the equity premium puzzle.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
527
Kim
cov ( ∆π t , ∆πt −1 )
(28)
= cov ( ∆ηt , ∆ηt −1 ) + cov ( ∆χt −1 , ∆χt −2 )
∞
+cov ( ∆χt −1 , ∆ηt −1 ) .
(29)
The obviously negative first term dominates the
second and third terms at appropriate parameter
values, resulting in a negative cov共∆πt , ∆πt –1兲. The
unforecastable component ηt also plays the role
of putting an upper bound on the predictability
of inflation.
Economically, the ηt term represents veryshort-run effects in total CPI inflation, including
part of the food and energy prices that create the
wedge between total CPI and core CPI, as well as
the unforecastable components of the core CPI
inflation and potential errors in the measurement
of CPI.
The importance of the ηt term in the twocomponent model (27) has a parallel implication
for no-arbitrage macro-finance models: The failure
to separate out the “unspanned macro shocks”
in macro-finance models may produce problems
that mirror those of the AR共1兲 inflation model. It
is worth mentioning here that Stock and Watson
(2007) have also recently emphasized that separating inflation into a trend component and a serially uncorrelated shock (like ηt in equation (27))
is useful for explaining key features of U.S. inflation dynamics,20 though they do not discuss the
ramifications for macro-finance (no-arbitrage)
models.
It is instructive to ask about the basic variable
underlying the term structure of inflation expectations in the ARMA共1,1兲 model. As is clear from
equation (24), the basic variable is χt, not realized
inflation, πt. Note that in the case of the AR共1兲
model, χt is πt (up to a prefactor and an intercept),
as can be seen from equation (25). This is not the
case for the ARMA共1,1兲 model: It is straightforward to show (by solving for εt in equation (23)
and recursively substituting into equation (26))
20
Stock and Watson (2007) write the U.S. inflation process for the
past half century as πt = τt + ηt, where ηt is a serially uncorrelated
disturbance term. The τt term (what they refer to as the trend component) can be identified as χt–1 in equation (27). Stock and Watson’s
(2007) τt and ηt have time-varying volatilities, a feature which they
argue is important in the inflation persistence debate.
528
S E P T E M B E R / O C TO B E R , PA R T 2
that χt in the ARMA共1,1兲 model depends on an
infinite number of lags of πt:
2009
(
)
χt = (φ − α ) ∑ α j π t − j − µ + µ.
j =0
This is in the exponential smoothing form,
which has been familiar at least since the work
of Muth (1960).
The expression (29) suggests that (i) the connection between realized macroeconomic variables and state variables in no-arbitrage term
structure models could be complicated and (ii)
the poor inflation forecasting performance of
regression models and no-arbitrage models with
macroeconomic variables may be a more complex
issue than just a matter of having “efficient” coefficients (with conventional basis). To be sure, the
state variables in nominal term structure models
are not simply those that underlie the variation of
inflation expectations. Factors that affect the real
term structure and inflation risk premia should
also be included in the nominal term structure
model. However, it is not clear that these additional aspects would be any better described by
macroeconomic variables.
Low-Dimensional External Basis Models
Let us now consider some specific issues
that arise in external basis models with a “lowdimensional” state vector. Suppose that one has
a three-factor macro-finance model in the setup
of equation (1), with the state vector xt consisting
of all “observable” macroeconomic variables, say,
quarterly inflation, πt, quarterly GDP growth, gt ,
and the effective federal funds rate, fft . The inflation expectations in this model are then linear
functions of contemporaneous variables πt, gt, and
fft . (To see this, simply substitute z̃1t = gt , z̃2t = fft
in equation (21).) This type of forecast (VAR共1兲)
has more qualitative similarity to the AR共1兲 model
than the ARMA共1,1兲 model; in particular, despite
its multifactor nature, it still mixes “signal” with
“noise” and can therefore be expected to inherit
many of the problems with the AR共1兲 model.
Some of the macro-finance models in the literature, including Ang, Dong, and Piazzesi (2005;
ADP) and ABW, remain in a relatively lowF E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
dimensional framework but use a mix of latent
factors and macroeconomic variables, but these
“mixed models” may still have difficulties. Consider, for example, the ABW affine model (their
MDL1 model) with quarterly inflation and two
latent factors, that is, xt = [π t , f1t , f2t ]′. If the latent
factors f1t, f2t are interpreted as π t –1, π t –2, equation
(21) takes a form similar to the smoothing form
(29). However, besides the issue that two lags
might not be enough, one may not have the freedom to interpret ft’s this way, as that would deprive
the ability to describe other aspects of the nominal
term structure (e.g., real interest rates, the timevarying risk premium, or time-varying perceived
inflation target).
In the mixed models, having a macroeconomic
variable like πt as a part of the state vector may
cause a distortion in the inference, as the latent
factors can end up absorbing the “unspanned”
variation in πt. To illustrate this schematically,
suppose that the true model of the short rate is
(30)
rt = ρπ t + ft ,
where π̃ t is the “spanned” part of the one-period
inflation πt ; that is,
(31)
πt = π t + et ,
with et denoting the unspanned component. If
one uses realized inflation πt in place of π̃ t, then
(32)
rt = ρ (π t − et ) + ft = ρπt + (ft − ρ et ).
Thus, the latent factor ft would be distorted by
an amount ρet. Though one might be tempted to
regard this simply as a redefinition of the latent
factor, it would imply practical differences, such
as the reduced persistence of the factor dynamics.
High-Dimensional External Basis Models
Some of the external basis macro-finance
models in the literature use a fairly large number
of state variables that include lagged macroeconomic variables. Many such models (including
those of AP and HTV) use annual inflation
πtY 共= π t ,t –1Y 兲 as a state variable instead of oneperiod inflation. This may help alleviate concerns
about the problem with the use of one-period
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
inflation, since the year-on-year inflation partly
“smooths out” the noise in quarterly inflation:
πtY can be written
(33)
πtY = ∑wi π t − i ,
i
where the weights wi are 1/4 for i = 0,1,2,3, and 0
for i > 3.
Note, however, that the construction (33)
automatically implies a moving average structure
in πtY, which suggests that the simple VAR共1兲
description would not be a good description of
its dynamics. Thus, macro-finance models that
use annual inflation as a state variable typically
include additional lags, for example, AP use 12
monthly lags, in effect having a VAR共12兲 model.
Bond yields in this case depend on a “large” set
of state variables that include lagged macroeconomic variables.21
A problem with this type of “high-dimensional”
specification is that it inherits the well-known
problems of the unrestricted VAR models. In fact,
AP’s inflation dynamics is a conventional VAR.
They separate the vector of relevant variables into
an “observable” macro vector fto and an unobservable (latent) vector ftu, that is, x̃t = [fto ′, ftu ′]′,22 and
impose the restriction that the latent factors do
not affect the expectation of macroeconomic variables. Their macro vector dynamics are given by
the VAR共q兲:
(34) fto = Φo1 fto−1 + Φo2 fto−2 + + Φoq fto−q + c o + Σo εto ,
where q = 12. Although the parameters in the
matrices Φ1o,…,Φqo are, in principle, identified and
can be estimated by ordinary least squares (OLS),
this kind of unrestricted VAR is well known to
suffer from overparameterization problems (which
21
Since an invertible ARMA(1,1) model can be written as an AR model
with infinite lags, the use of lagged macroeconomic variables in
an external basis model may partly address the deficiency of the
AR(1) model (relative to the ARMA(1,1) model) discussed in the
“Low-Dimensional External Basis Models” subsection. However,
identifying inflation as a state variable may still be problematic conceptually (especially in the case of one-period inflation, πt ), in
view of our earlier discussion regarding the difference between χt
and πt .
22
Here I have attached a tilde to xt to clarify that this is not the full
state vector. The full state vector (on which bond yields depend)
o′
in AP is larger: xt = [fto′,ft–1
,…,ftu′]′.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
529
Kim
will be discussed further in the “Empirical Issues”
subsection).23
By having only the macroeconomic variables
describe inflation dynamics, AP suppressed the
possibility of the yield curve saying something
about future inflation. Unfortunately, it is difficult
to lift that restriction. The overparameterization
problem would worsen, as the full (maximally
identified) model would have an even larger number of parameters: In the specification of the state
vector dynamics
(35)
fto
ftu
+
=
Φ1ou
Φu1
Φ1o
Φ1uo
Φo2
Φou
2
Φuo
2
Φu2
fto−1
ftu−1
++ c
u
ft − 2
fto− 2
+ Σεt ,
the matrices Φ1ou,Φ2ou,... are now nonzero and have
to be estimated. Furthermore, the two-step estimation procedure that AP used is no longer
applicable; hence, the estimation now involves a
“one-step” optimization of a very-high-dimensional
likelihood function.
For specifying external basis models that
contain lags of macroeconomic variables in the
state vector, it is common practice to set the
coefficients of the market price of risk (Λb matrix
in equation (1)) that load on lagged macroeconomic variables to zero (e.g., AP and HTV). Even
with this restriction, the number of remaining
market price risk parameters is large, and modelers
often make additional ad hoc restrictions on the
Λb matrices to reduce the number of parameters
further.24 Unfortunately, setting the Λb coefficients
on lagged macroeconomic variables to zero may
be a problematic practice. It implies that the
expected excess return on a bond, µτ,t – rt , is completely spanned by contemporaneous macroeconomic variables (and latent factors, if there are
any). Recall, from equations (12) and (17), that
23
Models like HTV have more structure (in the form of the New
Keynesian Phillips curve and IS equations), which may help alleviate overparameterization concerns, but at a possibly greater
misspecification risk.
24
For example, AP and HTV assume that Λb is a block-diagonal
matrix (a block matrix for macro factors and a block matrix for
latent factors).
530
S E P T E M B E R / O C TO B E R , PA R T 2
2009
(36)
µτ ,t − rt = Bτ′ Σλt .
Therefore, if λt does not depend on lagged macroeconomic variables, neither does the bond return
premium. This means that while one has
(37)
y τ ,t = aτ + bτ ,1π t + bτ ,2π t −1 + bτ ,3π t −2 ,
one cannot have
(38) µτ ,t − rt = α τ + βτ ,1πt + βτ ,2π t −1 + βτ ,3π t − 2 + ,
that is, there is an asymmetry (in the way yields
and bond risk premia depend on lagged macroeconomic variables) that was not motivated by
theory. Thus, in order to cast the model in a “noarbitrage” framework, many external basis macrofinance models may be introducing arbitrary and
nontrivial assumptions about the market price of
risk.25
AFFINE-GAUSSIAN MODELS
VERSUS NON-AFFINE/
NON-GAUSSIAN MODELS
Structural Stability
One potential limitation of the general framework (1) is structural stability. To be sure, the
debate about the structural stability of macroeconomic relationships is not new (see, e.g.,
Rudebusch, 1998, and Sims, 1998). However, it
may have different ramifications for internal basis
models and external basis models, and hence
merits a discussion here.
Note that external basis macro-finance models
have often used a framework based on the Taylor
rule and VARs, but many have raised questions
about the instability of these specifications.26
One may hope that concerns about structural
25
Duffee (2006) has also recently questioned the modeling of the
term premium in the macro-finance literature, more specifically,
the finding in some macro-finance papers of a strong relationship
between the term premium and the macroeconomy. His point is
that these studies often do not provide alternatives other than the
“expectations hypothesis” (zero or constant return premium) and
a term premium that depends on macro variables, leading to an
exaggerated role of macro variables in term premium variation.
26
See, for example, Clarida, Galí, and Gertler (2000) about the instability of Taylor-rule coefficients and Stock and Watson (1996) about
the instability of VAR coefficients.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
instability would be alleviated if latent factors
are also included in external basis models. For
example, a macro-finance model with a Taylorrule–like mixed specification of the short rate
(similar to ADP, 2005)
(39)
rt = const + ρπ π tY + ρg gapt + ft ,
where ft is a latent factor, can be written as
(
)
(40) rt = const + πtY + (1 − ρπ ) πtY − π t∗ + ρg gapt ,
where πt* (=–ft /共1 – ρπ 兲兲 is the time-varying inflation target. However, the factor ft may have to play
a number of other roles in the model, for instance,
the interest rate smoothing term, time-varying risk
premium, and so on (analogously to an earlier
discussion in the “Low-Dimensional External
Basis Models” subsection regarding ABW’s affine
model). Thus, a model written with ft as a timevarying inflation target in mind might have some
difficulty capturing the intended effect.
Furthermore, there may be instabilities other
than the time-varying intercept: for instance,
changes in the conditional correlation of various
macroeconomic variables, changes in the persistence of the macroeconomic variables, and so
on. Imagine, heuristically, a situation in which
the “true” model is
(41)
rt = c + ρπ ,t π tY + ρ g ,t gapt ,
that is, a Taylor-rule–like short rate with timevarying loadings on the macroeconomic variables.
In this case, the two-factor affine model in which
the state variables are [πtY, gapt ]′ is obviously
misspecified. For another example, consider a
“time-varying inflation-persistence model”:
(42)
πtY = φt −1πtY−1 + c + εt .
Again, identifying πtY as a state variable in an
affine setting would be a misspecification.
One way to address this problem is to model
these effects explicitly in non-affine/non-Gaussian
models.27 However, these models, being richer
than affine-Gaussian models, may be even more
susceptible to overfitting concerns and may incur
27
For a work in this direction, see Ang, Boivin, and Dong (2007).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
a greater risk of misspecification. Alternatively,
the use of an internal basis (while still remaining
in the affine-Gaussian setup) may allay structural
instability concerns to some extent: Internal basis
models are agnostic as regards the definition of the
factors; thus, a model that is obviously unstable
from the point of view of an external basis may
not necessarily be so from the point of view of
an internal basis. For example, going back to
equation (41), choosing the state variable as
xt = [ρπ,t πtY, ρg,t gapt ]′ may be more effective than
having xt = [πtY, gapt ]′, although there may be an
even better internal basis for the problem (depending on how the rest of the model is defined).28
Of course, no-arbitrage models with an internal basis should not be expected to answer all
structural stability concerns. A strong structural
instability may be difficult to capture even with
an internal basis model, in which case it might be
better to use a shorter, structurally more homogeneous sample.
Time-Varying Uncertainty
Another limitation of the affine-Gaussian
models (both internal and external basis models)
is that they imply homoskedastic yields, while
there is copious evidence for time-varying volatility of yields.
Theoretically and intuitively, one should
expect a relation between term structure variables
and time-varying uncertainty about interest rates:
To the extent that bond market term premia arise
from risk, the changing amount of interest rate
risk should translate to a changing term premium.
It also stands to reason that at least a part of the
variation in interest rate volatility is linked to the
variation in the uncertainties about key macroeconomic variables. Various studies have noted
that macroeconomic uncertainties (inflation, GDP,
monetary policy) have declined since the Volcker
disinflation, a phenomenon often dubbed the
“Great Moderation.”29 One can expect this effect
28
If π Yt and ρπ,t are Gaussian processes, the process ρπ,tπ Yt would be
non-Gaussian (with time-varying volatility). However, one can
still think of the affine version as an approximation of the nonGaussian process.
29
Bernanke (2004b) discusses this phenomenon from a policymaker’s perspective.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
531
Kim
Figure 2
The Dispersion of Long-Horizon U.S. Inflation Forecasts in the BCFF and the 1-Year Rolling
Standard Deviation of Monthly CPI Inflation
Percent
6
5
BCFF Dispersion
CPI Rolling Standard Deviation
4
3
2
1
0
1990
1992
1994
1996
1998
to be accompanied by a corresponding reduction
in term premia in the bond market. Kim and
Orphanides (2007) indeed report positive relationships between the term premium in the 10-yearforward rate and proxies for uncertainties about
monetary policy and inflation based on the dispersion of long-horizon survey forecasts.30
However, much work remains to be done to
properly address the relationship between term
premia and macroeconomic uncertainties—in
particular, inflation uncertainty. The key difficulty
is measuring the relevant inflation uncertainty.
For instance, one can debate whether the survey
dispersion measure used in Kim and Orphanides
(2007) is a reliable proxy for uncertainty. Inflation
uncertainty measures based on a GARCH-type
model also would be problematic, as they posit
too tight a relationship between long-term and
near-term uncertainty.31 As can be seen in Figure 2,
1-year rolling standard deviation of monthly
(total) CPI inflation (a proxy for near-term inflation uncertainty) has been elevated from around
30
See also Backus and Wright (2007).
532
S E P T E M B E R / O C TO B E R , PA R T 2
2009
2000
2002
2004
2006
1999 on, but this does not seem to have translated
to an increase in the perception of longer-term
uncertainty, proxied by the dispersion of surveyed
forecasts of long-horizon inflation. Even granting
the imperfection of the long-horizon inflation
uncertainty measure, this contrast is noteworthy.32
The complexity of inflation dynamics can
thus create considerable challenge for attempts
to go beyond homoskedastic models: It may be
that a nonlinear model with time-varying inflation uncertainty can lead to poorer results if the
model’s inflation uncertainty is misspecified, as
when a model that does not make a qualitative
distinction between short- and long-run inflation
uncertainties tries to link the rise in the volatility
of short-run inflation of the recent several years
31
Consider a GARCH specification of one-period inflation,
πt +1 = f 共πt ,πt –1,…兲 + εt +1, εt +1 ~ N共0,σt2 兲, σt2 = α + βσ t2–1 + γε t2 . It is
straightforward to show that the uncertainty about multiperiod
inflation πt +τ,t = 共πt +1 + πt +2 + … πt +τ 兲/τ, has similar qualitative time
variation as σt (short-run inflation uncertainty).
32
Interestingly, earlier literature including Ball and Cecchetti
(1990) and Evans (1991) has also emphasized in another context
the need to distinguish between the short-run and long-run inflation uncertainties.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
(as seen in Figure 2) to bond market term
premia.33
EMPIRICAL ISSUES
Overfitting Problems
Flexibly specified no-arbitrage models tend
to entail much estimation difficulty. Part of the
problem is one that is familiar from unrestricted
VAR models explored in the 1980s macroeconomics literature. It is well known that unrestricted
VARs often lead to poor results,34 as these models
get easily overparameterized, and as the “atheoretical” (statistical) nature of these models means
that there is little structure in the model to prevent estimations from generating unreasonable
outcomes.
The no-arbitrage macro-finance models have
two features that exacerbate the difficulty. First,
unlike the unrestricted VARs, the estimation of
macro-finance models typically requires nonlinear optimization (instead of OLS), because of the
nonlinear relationship between bond yields and
parameters such as the market price of risk and
because of the latent factors (recall the discussion
regarding equation (35)).
Second, the overparameterization problem
can be exacerbated by a large number of additional
parameters unique to macro-finance models,
including those describing the dynamics of latent
factors and their interaction with observable facu
uo
tors (e.g., Φou
j , Φj , Φj in equation (35)) and those
describing the market price of risk (λa and Λb in
equation (1)). Note that the no-arbitrage principle
guarantees the existence of a pricing kernel, such
as equation (1), but the principle by itself does not
constrain the parameters of the market price of
risk (Λb matrix). Suppose, as in AP, that one has
in the state vector p observable variables, its q – 1
lags, and m unobservable (latent) variables, in
other words,
33
34
There has been much discussion about the low level of term premia in recent years (see, e.g., Backus and Wright, 2007). Trying to
explain the low term premium and high uncertainty would be a
daunting prospect.
See, e.g., Todd (1984) and Stock and Watson (2001).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
u
xt = f1ot ,.., f pto , f1ut ,.., f mt
, f1o,t −1 ,..,
(43)
′
f po,t −1 ,, f1o,t −q +1 ,.., f po,t −q +1 .
In that case, the Λb matrix in equation (1) can have
as many as 共p + m兲 . 共p . q + m兲 parameters.35 For
p = 2, q = 12, m = 3 (as in AP), there are 135 parameters for Λb to be determined; even if one chooses
a smaller q, the number of parameters is still quite
large.
Recall that the key innovation of the macrofinance models like AP, as compared to the traditional macro models, is that they link not only
the short rate rt 共=y1,t 兲 but also the rest of the term
structure ({yτ,t }τ >1) to the macroeconomy by casting
the problem in the no-arbitrage framework (1).
However, the fact that yields of various maturities
tend to be highly correlated (giving rise to the finding in factor analysis and principal components
analysis that there is a single dominant factor)
means that the pure additional information in
longer-term yields (beyond what is in the short
rate) may be modest in amount and perhaps too
delicate to capture with a specification of the
market price of risk that is liable to be overfitted;
the relation that one might see between yields
and macroeconomic variables in macro-finance
models may be more of a statement of the Taylor
rule (macro description of the short rate) than
no-arbitrage models.
The overparameterization problem may be
particularly severe with external basis models
that contain lags of macroeconomic variables.
However, internal basis models (which tend to
be implemented with a comparatively smaller
number of factors, e.g., three factors) may also
face serious overfitting concerns, because of the
especially flexible nature (the definitional freedom) of the latent-factor models. In particular,
latent factor models can do a good job of fitting
the data that they are asked to fit, even if the
model or the data are poor. For example, because
yield-fitting errors are minimized as a part of the
35
The lagged macro variables do not have market price of risk, but
the market price of risk of contemporaneous variables and latent
variables can still depend on them, as discussed in the “HighDimensional External Basis Models” subsection.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
533
Kim
estimation process, internal basis models with
three or four factors can fit the cross section of
the yield curve quite well (with much smaller
fitting errors than external basis models), but that
by itself might not be a sufficient reason to recommend internal basis models.
Small-Sample Problems
The implementation of macro-finance models
is also complicated by small-sample problems
that arise from the highly persistent nature of the
data. Both interest rates and inflation are known
to be persistent; unit root tests often fail to reject
a nonstationarity (unit root) null for them.
In light of this, many practitioners often use
nonstationary models to forecast inflation. For
example, many of the inflation forecasting models
used by the Federal Reserve staff impose the unit
root condition.36 By the Fisher-hypothesis intuition, unit root inflation dynamics implies unit
root interest rate dynamics.
By contrast, most of the estimated macrofinance models (or nominal term structure models)
in the literature assume stationarity. Stationarity
has an intuitive appeal: We do not expect interest rates and inflation to have infinite unconditional moments. Thus, we may posit that the “true
model” of yields is a stationary one, perhaps with
many factors to describe the complex dynamics
of yields and expectations; schematically,
(44)
y τ ,t = fτ ( x1t , x 2t , x 3t ,...., x Nt ).
In practice, however, one is forced to deal with
relatively low-dimensional models, because
either the limited amount of data makes it impossible to pin down the parameters of such a model
or one does not have enough knowledge to construct a very detailed model. In this case, it is
not clear whether the “best” low-dimensional
approximation
(45)
y τ ,t ≈ f τ ( x 1t ,..., x nt ) , ( n << N )
of the model (44) should be a stationary or nonstationary (unit root) model.
36
Federal Reserve staff make inflation forecast judgmentally, but they
do look at a variety of models to inform their judgments. The staff’s
forecasting procedure is discussed, for example, in Kohn (2005).
534
S E P T E M B E R / O C TO B E R , PA R T 2
2009
The distinction between stationary and nonstationary models could be semantic in the sense
that a stationary model that is close to the unit
root boundary is almost indistinguishable from
unit root models, but whether to assume stationarity or not can make a big difference operationally,
as conventional estimations have the tendency
to bias down the persistence of stationary time
series, the bias becoming stronger as the sample
gets smaller. This makes the expectations appear
to converge to a long-run level faster than they
actually do; thus, longer-horizon expectations of
inflation and interest rates in (estimated) stationary models are often artificially stable, varying
little from the sample mean of these variables.
Another manifestation of the small-sample
problem (besides bias) is imprecision: Highly persistent interest rates effectively make the size of
the sample “small”; no matter how frequently the
data are sampled, some of the key aspects of the
term structure model (those pertaining to expectations in the physical measure, as opposed to the
risk-neutral measure) are difficult to estimate.37
Problems with the Classical Approach
Most implementations of macro-finance
models have relied on classical methods such as
the maximum likelihood estimation (MLE) and
generalized method of moments (GMM), but these
methods may be less effective in this context than
is often presumed. At the heart of the matter is the
fact that reduced-form macro-finance models are
obviously an approximate representation of data,
and hence not very compatible with the classical
premise of having the “true model.” Though it
goes without saying that all models in finance are
approximate, this point is particularly relevant
here in view of the atheoretical (statistical) nature
of the model and the large number of parameters;
the MLE or GMM criterion function of these
models might thus contain multiple maxima,
which capture different aspects of data with differing degree of emphasis. The small-sample
problems discussed above add to the difficulty,
as they make asymptotic statistics a poor guide
to finite sample properties.
37
The bias and imprecision problems in term structure model estimations are discussed in detail in Kim and Orphanides (2005).
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
Note also that many of the classical estimation
approaches implicitly minimize fitting errors for
the one-period-ahead conditional moments. For
example, the MLE estimation can be viewed as
minimizing the one-period prediction errors or
the errors in the fit of the “likelihood score
moments,”
( θ logf ( yt | yt
∂
∂
−1 ,θ
)) ,
in a GMM framework. While in theory this could
yield an asymptotically correct estimate of the true
model (if the true model exists), the inherently
approximate nature of the model means that fitting
the one-period moments as closely as possible
might come at the expense of other aspects of the
model. Cochrane and Piazzesi (2006) in effect
make this point when they note that conventionally estimated affine models may have difficulty
producing the kind of term premia that they find
based on regressing multiperiod (1-year) excess
returns on a set of forward rates. Note also that a
GMM approach that matches unconditional
moments, such as
( )
1
T
T
∑t =1 y τ ,t ,
T
E ( y τ ,t y τ ′,t − j ) = T1 ∑t =1 y τ ,t y τ ′,t − j ,
E y τ ,t =
might not be effective, due to the closeness of
the interest rate process (and inflation process)
to the unit root behavior.
How Can We Evaluate Models?
The above discussion suggests that looking
at the fit of the moments that are often used in
the classical estimation might not necessarily be
a good criterion for model evaluation. Some
papers do look directly at practical implications
of the model, such as the multiperiod forecasts
of inflation and interest rates. Indeed, in view of
the fact that the second-moment aspects of affineGaussian models are trivial, much attention has
focused on these conditional first moments (the
forecasting performance) as a part of diagnostic
criteria, as in Ang and Piazzesi (2003), HTV, and
Moench (2008).
However, it is unclear to what extent summary
measures of forecasting performance examined
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
in these papers can help with model evaluation/
selection. To be sure, looking at the forecasting
performance can be useful for detecting problematic models. In Duffee (2002), for example, interest
rate forecast RMSEs that are substantially larger
than the random-walk benchmark were used to
highlight problems with certain stochasticvolatility no-arbitrage models (e.g., the EA2共3兲
specification). Similarly, the inflation forecast
RMSEs based on ABW’s no-arbitrage models that
are substantially larger than the univariate inflation model benchmark may signal problems with
the no-arbitrage models that they have used.
Nonetheless, the RMSE measures for in-sample
or out-of-sample forecasts are often ineffective in
discriminating between models. For instance,
ABW obtain very similar RMSEs for the 1-year
out-of-sample inflation forecasts from the AR共1兲
and the ARMA共1,1兲 models, although the AR共1兲
model implies qualitatively quite different inflation expectations than the ARMA共1,1兲 model, as
discussed in the “Lessons from Simple Models”
subsection.
Furthermore, because a large part of the inflation and interest rate variations are unforecastable,
the RMSEs themselves may have substantial
uncertainty (sampling variability).38 Thus, it may
happen that the “true model” generates an RMSE
that is no smaller than some other models. In this
sense, it may be actually misleading to focus on
the RMSE as a criterion for selecting the model
that best describes reality. With in-sample forecasts, this problem is exacerbated by the possibility that RMSEs are artificially pushed down
because of the use of “future information” in
generating the forecast, thus making interest rates
and inflation look more forecastable than they
actually are.
Often there are cases in which classical criteria
cannot easily tell if a model’s output is unreasonable, while practitioners can do so using “judgmental information.” For instance, many macrofinance models estimated with data going back to
1970s generate current (circa 2006) long-horizon
inflation expectations that exceed 4 percent.
38
Clark and McCracken (2006) emphasize that out-of-sample inflation forecast RMSEs may have weak power.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
535
Kim
(Recall also the AR and ARMA model outputs in
Figure 1B.) Though long-horizon expectations
are difficult to evaluate on purely econometric
grounds, as there are not many nonoverlapping
observations, most policymakers and market
participants would immediately say that a 4 percent long-horizon CPI inflation expectation is too
high; hence, models with such an output may fail
the test of relevance before any statistical tests.
Note also that even if two models generated similar forecast RMSEs, practitioners could have a
very different assessment of them, depending on
the details of the forecast errors from the models
(such as the direction of the errors).39
These discussions highlight the role of the
larger information set of practitioners (as compared with academic researchers). Unfortunately,
much of this extra information is difficult to cast
in the formal language of statistical tests, and the
proper evaluation of models remains a challenge
for macro-finance modeling.
Would a Bayesian Approach Help?
The use of Bayesian techniques to address
problems with conventional (classical) estimation
has a long history, but a particularly relevant early
example is the Bayesian approach to VAR forecasting. As discussed in the “High-Dimensional
External Basis Models” and “Overfitting Problems”
subsections, unrestricted VARs share some of the
key problems encountered in flexibly specified
macro-finance models, in particular, the statistical
(atheoretical) nature of the specification and the
tendency for overparameterization. Litterman
(1986) and others have documented that a
Bayesian implementation with an informative
39
For example, in the 1990s, inflation data often came in on the
“low” side, and it is widely believed that not all of this had been
predicted by market participants, that is, the “true” market forecast
of inflation in this period likely contained a mild upward bias.
(See Kohn, 1999, and Croushore, 1990, for Fed staff and private
sector forecasts in the 1990s, respectively.) Though an “unbiased”
multiperiod forecast is often viewed as a consequence of rational
expectations, to obtain it one needs tight assumptions that are difficult to justify in reality—in particular, the assumptions that there
is a relatively simple, structurally stable model of the economy and
that the agents fully know this structure. More realistic rational
expectations hypotheses that relax these restrictions, for example,
models that allow for learning and time-varying structure, are
consistent with biased expectations in “small” samples.
536
S E P T E M B E R / O C TO B E R , PA R T 2
2009
prior (“random-walk prior”) can generate better
results than the classical implementation. This
encourages us to take up a Bayesian strategy to
address the empirical difficulties with macrofinance models.
In the macro-finance context, ADP have in
fact already proposed a Bayesian approach, but
it is not clear that the particular priors that they
have used would help overcome the problems
with classical estimation discussed above. ADP
state that, except for the condition that the model
be stationary, their priors are uninformative. However, to the extent that the main problem with
the classical estimation of macro-finance models
is that the data by themselves are not fully informative about the model (especially as regards the
overfitting and small-sample problems), it is difficult to see how uninformative priors would solve
the problem. Recall that the superior performance
of Bayesian VARs (over conventionally estimated
unrestricted VARs) came from having an informative prior.
When ADP (2005) tried to estimate their
model using a classical method (maximum likelihood estimation), they found that the estimated
model explained most of the term structure movements in terms of the latent factor, and left little
role for macroeconomic variables to explain yield
curve movements,40 an outcome that is unappealing from the viewpoint of making a connection
between the macroeconomy and the yield curve.
However, even granting the problems with classical methods, there may be a reason for this—
namely, that the estimation marginalizes the
macroeconomic variables to avoid the counterfactual implication that shocks to inflation (and
other macroeconomic variables) have a tight relation to yield curve movements. This is a specification issue (i.e., one has to deal with “unspanned”
variation in macroeconomic variables in the
model.) Addressing the problem purely as an
estimation issue may lead to problems elsewhere
in the model.
In my view, the main challenge for a Bayesian
implementation is in coming up with suitable
40
Private conversation with Andrew Ang. ADP’s paper (2005) itself
does not describe the specifics of the outcome from the classical
estimation of their model.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
informative priors. This is particularly the case
when there are latent factors in the model (external
basis models with latent factors or internal basis
models): Because the economic meaning of many
of the individual parameters related to the latent
factors is unclear, it is difficult to provide sensible
priors for them. Recall that a flexibly specified
latent factor model can be normalized in many
different (but equivalent) ways. It would be problematic if a Bayesian prior that was stated for one
normalization of the model did not hold in another
normalization of the model.41
By stating priors about the variables that have
direct economic meaning, such as inflation expectations, interest rate expectations, and expected
bond returns, one can get around the problem of
normalization dependence: Surely these variables
must be normalization independent. Recall also
that the source of the small-sample problem is the
difficulty of estimating the parameters related to
expectations (in the physical measure); thus,
imposing priors on these variables would help
alleviate the problem. A prior about the 10-year
inflation expectation, for example, can be
expressed as
(
)
e
2
′
(46) i10
Y ,t = a10Y (θ ) + b (θ ) 10Y xt (θ ) N µt ,σ t ,
with θ denoting the model parameters collectively.
For µt, one can use a survey median forecast.
Setting σt = ⬁ corresponds to having no priors on
e
. Setting σt at a large value, but not large
i10Y
enough to be irrelevant, can be viewed as a quasiinformative prior. Other Bayesian priors that are
based on economic concepts and mechanisms
may be also useable.42
A statement like equation (46) can be conveniently incorporated within a Kalman-filter setting. Running a Kalman-filter–based MLE with a
survey median (or mean) forecast (of interest rates
and/or inflation) as a noisy proxy, as in D’Amico
et al. (2008), can be viewed as a “poor man’s
41
See the working paper versions of this paper (Kim, 2007) for
elaboration.
42
“Structural” priors can be also imposed in a Bayesian setting, as
in the dynamic stochastic general equilibrium (DSGE) modeling
literature.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Bayesian” implementation, with the point estimate serving as the mode of the Bayesian posterior.
UNDERSTANDING THE SUPERIOR
PERFORMANCE OF SURVEY
FORECASTS
The specification and implementation problems discussed so far may help explain why macrofinance models, which use more information than
past inflation data, could generate poorer results
than simple univariate inflation models. But is
the yield curve information useful at all for inflation forecasting? Why do survey forecasts perform
better than univariate models (and other models)?
One reason ABW offer for the superior performance of survey forecasts is that survey participants have more information about the economy
than econometricians. This is in line with the
point made in the “How Can We Evaluate Models?”
subsection that informational differences may
create a wedge between a practitioner’s and an
academic researcher’s evaluation of a model. But
it is worth exploring this issue further.
One could plausibly expect that survey
forecasts may have advantages at least at short
horizons, in that a potentially vast amount of
information that is relevant for forecasting nearterm inflation may not be easily summarized into
a small number of variables. Thus, it may be
instructive to examine the near-term expectations
in surveys and how they are linked to longer-term
expectations (i.e., the term structure of survey
inflation forecasts).
Fortunately, fairly detailed information about
the near-term term structure of survey inflation
expectations can be obtained, as survey forecasts
such as the Survey of Professional Forecasters
(SPF) and the Blue Chip Financial Forecasts
(BCFF) provide CPI inflation forecasts up to the
next four or more quarters. Figure 3 shows the
1-, 2-, and 4-quarter-ahead CPI inflation forecasts
from the BCFF survey, based on the surveys
published in January, April, July, and October
(taken at the end of December, March, June, and
September), from 1988 to 2006. The figure also
shows the BCFF long-horizon forecast (inflation
S E P T E M B E R / O C TO B E R , PA R T 2
2009
537
Kim
Figure 3
Forecasts of U.S. Quarterly CPI Inflation from the BCFF
Percent
1Q-Ahead
2Q-Ahead
6
4Q-Ahead
Long-Term
4
2
1988
1990
1992
1994
1996
expected between the next 5 and 10 years), which
is available twice per year, is also shown. It is
notable that this long-horizon forecast, which can
be viewed as a “quasi–long-run” mean of inflation,
has moved about (shifted down) significantly. It
is also notable how quickly the multiperiod forecasts approach the quasi–long-run value. The 4quarter-ahead forecast and the 2-quarter-ahead
forecast are already quite similar to the longhorizon forecast. Note that even in 1990:Q3, when
the 1-quarter-ahead inflation expectation peaked,
the expectations for longer horizons show that the
survey participants expected inflation to come
down quickly to the quasi–long-run level. Thus,
one comes to a conclusion that “the long term is
quite near.”
To get further insights into the survey forecasts, it is useful to compare them with ex post
realized inflation and the real-time forecasts from
the ARMA共1,1兲 model. Figure 4A shows the 1quarter-ahead inflation forecasts based on the
BCFF survey and the ARMA共1,1兲 model (20-year
rolling sample forecast), as well as the realized
quarterly inflation (πt plotted at t –1). The vertical
difference between realized inflation and the
survey forecast or the ARMA forecast is the forecast error. This error is indeed smaller for the
survey forecast. (The RMSEs of the 1-quarter-ahead
forecast are 1.19 percent and 1.40 percent in annual
percentage units for the survey forecast and the
ARMA共1,1兲 model, respectively.) Note that the
538
S E P T E M B E R / O C TO B E R , PA R T 2
2009
1998
2000
2002
2004
2006
1-quarter-ahead survey forecast is much less jagged
than realized inflation or the ARMA共1,1兲 forecast.
Granting the caveat that surveys might not necessarily be the best possible means of forecasting,
this suggests that a substantial part of short-run
inflation is unforecastable ex ante, lending support to a formulation like the two-component
model in equation (27) in which the inflation
process is separated into a trend inflation component and an unforecastable component.
Let us now examine the 1-year inflation forecast, shown in Figure 4B. The ARMA forecasts
(both the rolling and the expanding samples) performed worse than the survey forecast with RMSEs
of 1.04 percent for the 20-year rolling sample
ARMA, 1.15 percent for the expanding sample
ARMA, and 0.76 percent for the survey. The basic
reason for the superior forecast of the survey is
that the ARMA model–based forecasts substantially overpredicted inflation in the 1990s. It can
be seen that the ARMA forecasts lie notably above
the realized inflation (and survey forecast). This
overprediction is due in large measure to the fact
that the ARMA model in real time tended to generate “too high” values of the long-run mean level
(µ in equation (23)) to which the forecasts are converging. This is illustrated in Figure 5, where the
long-run mean parameter µ from the expanding
sample estimation lies significantly above the
long-horizon survey forecast. Because the expandF E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
Figure 4
Comparison of Realized CPI Inflation, the BCFF Survey Forecasts, and the
ARMA(1,1) Model Forecasts
A. 1-Quarter-Ahead Forecasts
Percent
7
6
5
4
3
2
Survey
1
Realized
0
−1
1988
ARMA (rolling)
1990
1992
1994
1996
1998
2000
2002
2004
2006
1998
2000
2002
2004
2006
B. 1-Year-Ahead Inflation Forecasts
Percent
7
6
5
4
3
2
Survey
Realized
1
ARMA (rolling)
0
−1
1988
ARMA (expanding)
1990
1992
1994
1996
ing sample includes periods of high inflation
(1970s and early 1980s), the estimated mean does
not fall quickly with declining inflation in the ’80s
and ’90s. The use of the 20-year rolling sample
produces lower µ (than the expanding sample)
as the estimation sample moves away from those
periods, but still the adjustment in the long-run
mean is not as fast as in the survey forecast.43
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
43
ABW also note that the ability of survey forecasts to quickly adapt
to major changes in the economic environment contributes to the
superior performance of the surveys. While the majority of ABW’s
estimations were done with expanding samples, they also examine
the forecast RMSEs based on rolling-sample estimation for a subset
of their models. Because their rolling sample (10 years) is shorter
than the 20-year rolling sample used here, ABW’s rolling-sample
results are even closer to those of the surveys. For example, the
ratio of the AR model RMSE and the survey RMSE in the post-1995
window is 0.879/0.861, very close to 1.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
539
Kim
Figure 5
Long-Run Means from the ARMA(1,1) Model Estimations of U.S. Quarterly Inflation Data
Percent
7.0
Rolling Sample
Expanding Sample
6.5
6.0
5.5
5.0
4.5
4.0
3.5
3.0
2.5
2.0
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
NOTE: The long-horizon BCFF survey inflation forecast is also shown (*).
The key point that emerges from this discussion is that surveys produce a more successful
forecast of inflation in large part because they capture the trend component of inflation better than
time-series models such as the ARMA共1,1兲 model.
In stationary time-series models (for example, the
models in Figure 4), forecasts tend to converge to
a value close to the sample mean, while nonstationary models put too much weight on the recent
past; thus, there is scope for judgmental information to play a role, especially if trend inflation
varies significantly over time. These considerations shed light on the attention that policymakers
pay to long-term inflation expectations (a better
indicator of the trend inflation than realized inflation) and on the use of judgmental forecasts at
central banks such as the Federal Reserve.
The importance of modeling the variation of
long-term expectations deepens the challenge
for macro-finance models: Besides the specification challenge, the nearly nonstationary nature of
the inflation process indicated by the substantial
variability of long-term survey forecasts poses
540
S E P T E M B E R / O C TO B E R , PA R T 2
2009
considerable empirical difficulties (discussed in
the previous section). These challenges notwithstanding, the discussions in this paper can be
viewed as encouraging for attempts to use term
structure models to extract inflation expectations:
It makes intuitive sense that the yield curve contains, at least, information about trend inflation,
and the indication that the near-term informational
advantage of surveys seems to wear out quickly
(beyond a few quarters) gives some hope that
models could capture much of the variation in
inflation expectations and compete with surveys.44
44
Although ABW find that survey forecasts cannot be improved by
combination with models that they consider, few policymakers
would regard survey forecasts as the ultimate measure of inflation
expectations. Consider, for example, the fact that between 1999 and
2006 the 10-year CPI inflation expectation from the SPF survey has
been almost stuck at 2.5 percent. While there is a broad consensus
that long-term inflation expectations were “better anchored” in
the 2000s than in the earlier decades, it may be a stretch to regard
that long-term inflation expectation has become so well anchored
as to be practically immovable. This may be one example in which
the yield curve contains useful information that is unavailable in
the SPF survey.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
CONCLUSION
These are some of my key points made in this
paper: (i) Not all of the variation in key macroeconomic variables is related to yield curve movements. (ii) The yield curve contains useful
information about the trend component of inflation. (iii) The no-arbitrage principle might not be
sufficient to guarantee sensible outputs from
macro-finance models in practice.
As I have stressed in the second section of
the paper (“The Basic Model”), the spanning
argument is the basis of the no-arbitrage framework; hence the presence of a short-run inflation
component that is not related to yield curve movements may undermine the validity of the models
that use inflation as a state variable. Such a component may also cause special difficulties when
one tries to go beyond the affine-Gaussian setup
to model time-varying uncertainties about macroeconomic variables explicitly. For example, as
discussed in the “Time-Varying Uncertainty” subsection, monthly CPI inflation in recent years has
been more volatile than in the 1990s, but there is
no strong evidence that this is reflected in the
yield curve (e.g., as an increased term premium);
an attempt to link them may thus lead to more
serious specification errors.
I have also argued in this paper that much of
the “spanned” component of inflation (the part
of inflation that is related to the yield curve) is
about the trend component (whose importance
was stressed in the discussion in the previous
section of why surveys perform better). This can
help resolve the puzzle that the “conventional
wisdom” that the change in nominal yields often
reflects changes in inflation expectations dies
hard, despite the poor performance of inflation
forecasting models involving term structure variables. In some sense, the latent factor models can
be viewed as a way to represent markets’ implicit
processing (filtering) of information.
No-arbitrage models of the term structure have
been viewed as a promising way to go beyond
the restrictive assumptions implicit in the expectations hypothesis (about how risk is incorporated
in the yield curve). However, reduced-form affineGaussian no-arbitrage models with flexible speciF E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
fication of the market price of risk can quickly
become “too unrestrictive,” with a profusion in
the number of parameters. In other words, the
no-arbitrage principle by itself may be too weak
to provide enough discipline in the model. Note
also that the two technical problems with estimation discussed in the “Empirical Issues” subsection (overfitting and small-sample problems) can
be viewed as an extension of the specification
discussion, as the main source of the problems
can be viewed as insufficient information in the
data or an incomplete structure in the model. For
further progress, it would be desirable to come
up with an effective and non-ad hoc structure on
the market price of risk and other parameters of
macro-finance models—or to come up with a
new, intuitively appealing way to represent the
term structure.
REFERENCES
Ang, Andrew; Bekaert, Geert and Wei, Min. “Do
Macro Variables, Asset Markets or Surveys Forecast
Inflation Better?” Journal of Monetary Economics,
2007, 54, pp. 1163-212.
Ang, Andrew; Bekaert, Geert and Wei, Min. “The Term
Structure of Real Rates and Inflation Expectation.”
Journal of Finance, April 2008, 63(2) pp. 797-849.
Ang, Andrew; Boivin, Jean and Dong, Sen. “Monetary
Policy Shifts and the Term Structure.” Working
paper, October 2007; http://neumann.hec.ca/pages/
jean.boivin/mypapers/abd.pdfworking paper.
Ang, Andrew; Dong, Sen and Piazzesi, Monika.
“No-Arbitrage Taylor Rules.” Working paper,
European Central Bank, August 2005; www.ecb.int/
events/pdf/conferences/ECB-BIS_2005/ADP2004.pdf.
Ang, Andrew and Piazzesi, Monika. “A No-Arbitrage
Vector Autoregression of Term Structure Dynamics
with Macroeconomic and Latent Variables.”
Journal of Monetary Economics, May 2003, 50(4),
pp. 745-87.
Backus, David and Wright, Jonathan H. “Cracking the
Conundrum.” Working Paper 2007-46, Federal
Reserve Board Finance and Economics Discussion
S E P T E M B E R / O C TO B E R , PA R T 2
2009
541
Kim
Series, 2007; www.federalreserve.gov/Pubs/feds/
2007/200746/200746pap.pdf.
Ball, Laurence; Cecchetti, Stephen G. and Gordon,
Robert J. “Inflation Uncertainty at Short and Long
Horizons.” Brookings Papers on Economic Activity,
1990, 1990(1), pp. 215-54.
Bansal, Ravi and Yaron, Amir. “Risks for the Long
Run: A Potential Resolution of the Equity Premium
Puzzle.” Journal of Finance, August 2004, 59(4),
pp. 1481-509.
Bernanke, Ben S. “What Policymakers Can Learn
from Asset Prices.” Speech before the Investment
Analyst Society of Chicago, April 15, 2004a;
www.federalreserve.gov/BoardDocs/Speeches/
2004/20040415/default.htm.
Bernanke, Ben, S. “The Great Moderation.” Speech
at the meetings of the Eastern Economic Association,
Washington, DC; February 20, 2004b;
www.federalreserve.gov/boarddocs/speeches/2004/
20040220/default.htm.
Blinder, Alan S. Commentary on “Measuring ShortRun Inflation for Central Bankers.” Federal Reserve
Bank of St. Louis Review, May/June 1997, 79(3),
pp. 157-60; http://research.stlouisfed.org/
publications/review/97/05/9705ab.pdf.
Brayton, Flint; Mauskopf, Eileen; Reifschneider,
David; Tinsley, Peter and Williams, John. “The Role
of Expectations in the FRB/US Macroeconomic
Model.” Federal Reserve Bulletin, April 1997;
www.federalreserve.gov/pubs/bulletin/1997/
199704lead.pdf.
Campbell, John Y.; Chan, Yeung, L. and Viceira,
Luis M. “A Multivariate Model of Strategic Asset
Allocation.” Journal of Financial Economics,
January 2003, 67(1), pp. 41-80.
Journal of Economics, January 2000, 115(1),
pp. 147-80.
Clark, Todd and McCracken, Michael W. “The
Predictive Content of the Output Gap for Inflation:
Resolving In-Sample and Out-of-Sample Evidence.”
Journal of Money, Credit, and Banking, 2006,
38(5), pp. 1127-48.
Cochrane, John H. Asset Pricing. Princeton, NJ:
Princeton University Press, 2001.
Cochrane, John H. and Piazzesi, Monika.
“Decomposing the Yield Curve.” Presented at the
September 2006 Brookings Papers on Economic
Activity Conference. Working paper (March 13, 2008,
revision); http://faculty.chicagobooth.edu/john.
cochrane/research/Papers/interest_rate_revised.pdf.
Collin-Dufresne, P. and Goldstein, Robert S. “Do Bonds
Span the Fixed Income Markets? Theory and
Evidence for Unspanned Stochastic Volatility.”
Journal of Finance, August 2002, 57(4), pp. 1685-730.
Cox, John C.; Ingersoll, Jonathan E. and Ross,
Stephen A. “A Reexamination of Traditional
Hypothesis about the Term Structure of Interest
Rates.” Journal of Finance, September 1981, 36(4),
pp. 769-99.
Croushore, Dean. “Low Inflation: The Surprise of the
1990s.” Federal Reserve Bank of Philadelphia
Business Review, July/August 1998, pp. 3-12;
www.philadelphiafed.org/research-and-data/
publications/business-review/1998/july-august/
brja98dc.pdf.
Dai, Qiang and Singleton, Kenneth J. “Specification
Analysis of the Affine Term Structure Models.”
Journal of Finance, October 2000, 55(5), pp. 1943-978.
Campbell, John Y.; Lo, Andrew W. and MacKinlay,
A. Craig. The Econometrics of Financial Markets.
Princeton, NJ: Princeton University Press, 1996.
D’Amico, Stefania; Kim, Don H. and Wei, Min. “Tips
from TIPS: The Informational Content of Treasury
Inflation-Protected Security Prices.” BIS Working
Paper No. 248, Bank for International Settlements,
2008.
Clarida, Richard H.; Galí, Jordi and Gertler, Mark.
“Monetary Policy Rules and Macroeconomic
Stability: Evidence and Some Theory.” Quarterly
Duffee, Gregory R. “Term Premia and Interest Rate
Forecasts in Affine Models.” Journal of Finance,
February 2002, 57(1), pp. 405-43.
542
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Kim
Duffee, Gregory R. “Are Variations in Term Premia
Related to the Macroeconomy?” Working paper,
Hass School of Business, University of California–
Berkeley, March 2006.
Duffie, Darrell. Dynamic Asset Pricing Theory. Third
Edition. Princeton, NJ: Princeton University Press,
2001.
Evans, Martin. “Discovering the Link Between Inflation
Rates and Inflation Uncertainty.” Journal of Money,
Credit, and Banking, May 1991, 23(2), pp. 169-84.
Hördahl, Peter; Tristani, Oreste and Vestin, David.
“A Joint Econometric Model of Macroeconomic and
Term-Structure Dynamics.” Journal of Econometrics,
March/April 2006, 131(1-2), pp. 405-44.
Litterman, Robert B. and Scheinkman, Jose A.
“Common Factors Affecting Bond Returns.”
Journal of Fixed Income, June 1991, 1(1), pp. 54-61.
Moench, E. “Forecasting the Yield Curve in a Data-Rich
Environment: A No-Arbitrage Factor-Augmented
VAR Approach.” Journal of Econometrics, September
2008, 146(1), pp. 26-43.
Muth, John, F. “Optimal Properties of the
Exponentially Weighted Forecasts,” Journal of the
American Statistical Association, 1960, 55, pp.
299-306.
Rudd, Jeremy and Whelan, Karl. “Can Rational
Expectations Sticky-Price Models Explain Inflation
Dynamics?” American Economic Review, March
2006, 96(1), pp. 303-20.
Kim, Don H. “Challenges in Macro-Finance
Modeling.” Working Paper 2008-06, Board of
Governors of the Federal Reserve System, Finance
and Economics Discussion Series, February 2008.
Rudebusch, Glenn, D. “Do Measures of Monetary
Policy in a VAR Make Sense?” International
Economic Review, November 1998, 39(4), pp. 907-31.
Kim, Don H. and Orphanides, Athanasios. “Term
Structure Estimation with Survey Data on Interest
Rate Forecasts.” Working Paper 2005-48, Finance
and Economics Discussion Series, Federal Reserve
Board, October 2005; www.federalreserve.gov/
pubs/feds/2005/200548/200548pap.pdf.
Rudebusch, Glenn D. and Wu, Tao. “A Macro-Finance
Model of the Term Structure, Monetary Policy, and
the Economy.” Working Paper 2003-17, Federal
Reserve Bank of San Francisco, 2003;
www.frbsf.org/publications/economics/papers/2003/
wp03-17bk.pdf.
Kim, Don H. and Orphanides, Athanasios. “Bond
Market Term Premium: What Is It, and How Can
We Measure It?” BIS Quarterly Review, June 2007,
pp. 27-40.
Sangvinatsos, Antonios and Wachter, Jessica A.
“Does the Failure of the Expectations Hypothesis
Matter for Long-Term Investors?” Journal of Finance,
February 2005, 60(1), pp. 179-230.
Kohn, Donald L. Comment on “Forward-Looking
Rules for Monetary Policy,” in John B. Taylor, ed.,
Monetary Policy Rules. Chicago: University of
Chicago Press, 1999, pp. 157-202.
Sims, Christopher A. “Macroeconomics and Reality.”
Econometrica, 1980, 48, pp. 1-48.
Kohn, Donald L. “Inflation Modeling: A Policymaker’s
Perspective.” Remarks to the International Research
Forum on Monetary Policy Conference, Frankfurt am
Main, Germany, May 20, 2005;
www.federalreserve.gov/Boarddocs/Speeches/2005/
20050520/default.htm.
Litterman, Robert B. “Forecasting with Bayesian
Vector Autoregressions—Five Years of Experience.”
Journal of Business and Economic Statistics,
January 1986, 4(1), pp. 25-38.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Sims, Christopher A. Commentary on “Do Measures
of Monetary Policy in a VAR Make Sense?”
International Economic Review, 1998, 39(4),
pp. 933-41.
Stock, James H. and Watson, Mark W. “Evidence on
Structural Instability of Macroeconomic Time
Series Relations.” Journal of Business Economics
and Statistics, January 1996, 14(1), pp. 11-30.
Stock, James H. and Watson, Mark W. “Vector
Autoregressions.” Journal of Economic Perspectives,
Fall 2001, 15(4), pp. 101-15.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
543
Kim
Stock, James H. and Watson, Mark W. “Forecasting
Output and Inflation: The Role of Asset Prices.”
Journal of Economic Literature, September 2003,
41(3), pp. 788-829.
Stock, James H. and Watson, Mark W. “Why Has U.S.
Inflation Become Harder to Forecast?” Journal of
Money, Credit, and Banking, February 2007, 39(s1),
pp. 3-33.
Todd, Richard M. “Improving Economic Forecasting
with Bayesian Vector Autoregression.” Federal
Reserve Bank of Minneapolis Quarterly Review,
Fall 1984, 8(4), pp. 18-29;
www.minneapolisfed.org/research/QR/QR843.pdf.
544
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
II
Investment Analysts’ Forecasts of Earnings
Rocco Ciciretti, Gerald P. Dwyer, and Iftekhar Hasan
The literature on investment analysts’ forecasts of firms’ earnings and their forecast errors is enormous. This paper summarizes the evidence on the distribution of analysts’ forecasts and forecast
errors using data for all U.S. firms from 1990 to 2004. The evidence indicates substantial asymmetry
of earnings, earning forecasts, and forecast errors. There is strong support for average and median
earning forecasts being higher than actual earnings a year before the earnings announcement. Such
differences between earnings and forecasts also exist across time periods and industries. A month
before the earnings announcement, the mean and median differences are small. (JEL G17, C53)
Federal Reserve Bank of St. Louis Review, September/October 2009, 91(5, Part 2), pp. 545-67.
D
o stock analysts provide information
on stocks, or are they merely salespeople issuing one-sided information
about stocks? In addition to forecasting earnings that are used by some investors
when they buy various firms’ stocks, analysts at
investment banks often have participated in other
activities such as convincing the same firms to
hire the investment bank to issue stock. These
activities were the basis of suits by the New York
attorney general against major investment banks.
Rather than proceeding to trial, the charges were
settled in April 2003. In the settlement, investment banks agreed to substantial changes in their
business practices designed to provide less incentive for analysts to be influenced by the investment banks’ other activities. The investment
banks also agreed to make payments totaling
$1.4 billion, which covered fines, payments to
investors, funding of investor education, and
funding of research by independent analysts.
This settlement brings into question the informativeness of analysts’ projections of earnings,
suggesting that analysts’ projections of earnings
largely or substantially reflect analysts’ interests
rather than an assessment of a firm’s prospects.
On the other hand, charges of an insidertrading scheme in 2007 suggest that analysts’ forecasts do contain information and affect prices.
This scheme involved an accomplice receiving
advance information about analysts’ forecasts
and taking positions before the announcements
(Smith, Scannell, and Davies, 2007). This scheme
makes no sense if analysts’ forecasts are uninformative and ignored. While indicating that at least
some analysts’ forecasts may be informative, such
activities do not imply that forecasts cannot be
improved. It is possible to take imperfect information and filter out predictable misinformation.
Rocco Ciciretti is an assistant professor of economic policy in the SEFeMEQ department at the University of Rome at Tor Vergata. Gerald P. Dwyer
is the director of the Center for Financial Innovation and Stability at the Federal Reserve Bank of Atlanta and an adjunct professor at the
University of Carlos III, Madrid. Iftekhar Hasan is the Cary L. Wellington Professor of Finance at Rensselaer Polytechnic Institute and a research
associate at the Berkley Center for Entrepreneurial Studies of the Stern School of Business at New York University. The Berkley Center helped
make these data available to the authors. Data from the Center for Research in Security Prices (CRSP), Booth School of Business, The University
of Chicago (2006), are used with permission. (All rights reserved. www.crsp.chicagobooth.edu). Much of this paper was completed while
Rocco Ciciretti was a visiting scholar at the Federal Reserve Bank of Atlanta, which provided research support. Gerald Dwyer thanks the
Spanish Ministry of Education and Culture for funding under project No. SEJ2007-67448/ECON. Budina Naydenova and Julie Stephens provided excellent research assistance. The authors also thank seminar participants in the XVI Tor Vergata Conference on Money, Banking, and
Finance, a DECE Seminar at the University of Cagliari, and a seminar at the Bank of Italy for helpful comments. Mark Fisher, Paula Tkac, and
two referees made helpful comments on earlier drafts.
© 2009, 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.
F E D E R A L R E S E R V E B A N K O F S T. LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
545
Ciciretti, Dwyer, Hasan
Are there predictable differences between
analysts’ earnings forecasts and actual earnings?
Many papers show that the analysts’ forecast errors
are predictably different from actual earnings.1
The evidence indicates that analysts’ forecasts of
earnings well before the announcement are higher
on average than actual earnings. Whatever earnings an analyst forecasts for a firm, a better prediction is a somewhat lower level of earnings. This
predictable difference is called a “bias” in the
forecasts.2 Some papers also suggest that analysts’
forecasts close to the earnings announcement
decline to less than the actual earnings. The rationale for this reverse bias is a suggestion that earnings greater than recent forecasts are interpreted
as a positive earnings surprise and the firm’s
stock price increases.
This paper provides an overview of analysts’
forecasts and the forecasts’ relationship to actual
earnings. Our data are for U.S. analysts’ forecasts
of U.S. firms’ earnings from 1990 through 2004.
These data show the usual result that analysts’
forecasts are greater than earnings on average. We
look at the distribution in more detail and find that
the distribution of earnings is asymmetric. The
distribution of earnings forecasts also is asymmetric but not sufficiently asymmetric that forecast
errors are symmetric; earnings forecast errors also
are asymmetric. We also find that median forecasts
are closer to actual forecasts than are mean forecasts. We examine differences between actual
earnings and earnings forecasts over time and by
industry. We find substantial differences in forecast accuracy across industries and larger forecast
errors during recessions. Forecast errors at the 1month horizon are small in magnitude.
ERRORS IN FORECASTING
EARNINGS PER SHARE
between actual earnings and these forecasts of
earnings. There is a scale problem with using the
level of forecasts across firms and over time. A
firm with the same total earnings as another but
half as many shares outstanding will have earnings per share that are twice as large. One way to
adjust for differences in the magnitude of earnings
per share and forecast errors across firms is to
divide the forecast error by the stock price. Dividing by the stock price assumes that errors in forecasting earnings per share relative to the stock
price are relatively homogeneous across firms.
Earnings per share relative to the stock price is
the inverse of the price-to-earnings ratio, often
used as part of the information used to evaluate
companies.3
The forecast error relative to the stock price
is indicated as follows:
(1)
Analysts forecast companies’ earnings per
share, and the forecast error is the difference
1
Sirri (2004) summarizes a few of these papers and provides references.
2
Not all research agrees that analysts’ forecasts are biased—for
example, Keane and Runkle (1990, 1998).
546
S E P T E M B E R / O C TO B E R , PA R T 2
2009
aTi − fTi,,tj
,
pTi – 1
i,j
where eT,t
is the computed relative forecast error
for company i by analyst j for year T made t months
before the release date, aTi is actual earnings per
i,j
is the forecasted
share of company i in year T, f T,t
earnings per share for company i by analyst j made
for year T with the forecast being made t months
i
is the stock price
before the release date, and pT–1
for company i at the end of the previous year, T–1.
The forecast horizon, t, is calculated as the
difference in months between the estimation date
(I/B/E/S [Institutional Brokers’ Estimate System]
variable ESTDATX) and the report date (I/B/E/S
variable REPDATX). We use the report date instead
of forecast period end date (FPEDATX) because
analysts can make forecasts between the fiscal
year’s end and the date earnings are reported.
The data on forecasts of earnings per share
and actual earnings per share are from the I/B/E/S
3
Data
eTi,,jt =
Another way to scale earnings per share is to divide by the level
of earnings to get the proportional error in forecasting earnings.
Earnings close to zero and negative earnings create serious problems for this normalization. Dividing by earnings can generate a
very large relative forecast error as earnings go to zero; dividing
by negative earnings would change the sign of the forecast error.
Stock prices cannot be negative and are strictly positive in our data.
Although prices can approach zero, earnings generally approach
zero at a related rate, which is another way of saying that earnings
per share relative to the stock price are relatively homogeneous
across firms.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
Detail History (with Actuals) database for 1990
through 2004. Any company with at least one
forecast between 1990 and 2004 is included in
the initial database.
The stock prices are from the Center for
Research in Security Prices (CRSP) database
from 1989 to 2003. The earnings in any year are
divided by the stock price at the end of trading
in the prior year. With this choice of stock price,
the stock price does not reflect the changes in
forecasts or the ensuing forecast errors made
during the year.
The initial number of observations on forecasts is 1,835,642. To avoid nonsynchronized timing of forecasts by year, we restrict the analysis to
companies with fiscal years ending in December.4
This reduces the number of observations to
1,207,445. We further restrict our analysis to forecasts by analysts located within the United States,
which reduces the number of observations to
678,427 forecasts for 6,731 companies. In this
paper, a company’s stock is defined by the six-digit
Committee on Uniform Securities Identification
Procedures (CUSIP) number followed by an “01”;
this indicates a common stock. We match U.S.
companies from I/B/E/S and CRSP databases by
CUSIP. We also associate an industry code according to the Global Industries Classification Standard
from Standard & Poor’s.
Finally, to eliminate possible transcription
errors, we cut off the distributions of both actual
and forecasted earnings per share relative to the
stock price at the 1st and 99th percentiles for each
year and forecast horizon.5 This results in a dataset
with 662,016 observations for 6,574 companies.
The number of firms included in the analysis
increases over time. The number of U.S. companies with a fiscal year ending in December and
an earnings’ forecast by at least one U.S. analyst
increased from 1,446 in 1990 to 2,569 in 2004.
The analyses by industry use the industry classification, which is not available for 104,840 obser-
Figure 1 shows the distributions of earnings
and forecasted earnings. The graphs show the
distribution of actual earnings and the distribution
of forecasts by analysts made 1 month, 6 months,
and 12 months before the earnings announcement.
For example, the first graph (Figure 1A) shows
actual earnings per share relative to the stock price
and forecasts made 1 month before the announcement of earnings. The second graph (Figure 1B)
shows the distribution of earnings and the distribution of the forecasts made 6 months before the
earnings announcement, and the third graph
(Figure 1C) shows the distribution of earnings and
the distribution of the forecasts made 12 months
before the earnings announcement.6 Deleting the
top and bottom 1 percent of the distribution still
leaves quite long tails to the distribution of earnings and, to a lesser but still easily discernible
extent, the forecasts. To avoid obscuring detail,
we also truncate these figures at –$0.50 and +$0.50
per dollar of share price. Table 1 shows the distribution of earnings, forecasts, and the forecast
errors without the truncation. Relative to the total
number of observations, the truncation excludes
a small number of observations, mostly in the
negative tail of the distributions.
The forecasts and actual earnings are strikingly
similar, which is consistent with the forecasts
being quite informative about actual earnings.
The histograms for forecasts and actual earnings
are distinguishable, but the overlap far outweighs
the differences. The dashed vertical lines are
drawn at the mean of actual earnings. The most
common—modal—values of forecasted and actual
earnings are similar. The solid curves in the figure
represent normal distributions with the same
4
6
When looking at data by year, having the same end date means
that the same events are occurring at the same horizon for all firms.
Firms with fiscal years ending in December represent about 74
percent of all firms in the I/B/E/S database.
5
The results in Tables 2 through 4 were computed with the tails of the
distribution of the data included. The results are broadly similar.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
vations. As a result, the analyses by industry use
557,176 observations instead of the whole sample
of 662,016 observations.
Distribution of Forecast Errors
The distribution of earnings is not the same at each of the horizons.
The figure shows the distribution of all forecasts and the distribution of the actual earnings that were predicted. Every firm with a
forecast appears in the figure; every firm with no forecast does not
appear in the figure. In addition, every firm with more than one
forecast appears in the figure the same number of times as the
number of its forecasts.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
547
Ciciretti, Dwyer, Hasan
Figure 1
Actual Earnings and Earnings Forecast
A. 1-Month Forecast Horizon
12
Actual Earnings
Earnings Forecast
10
8
6
Normal Distribution
of Actual Earnings
4
2
0
–0.5
–0.4
–0.3
–0.2
–0.1
0.0
0.1
0.2
0.3
0.4
0.5
B. 6-Month Forecast Horizon
12
Actual Earnings
Earnings Forecast
10
8
6
Normal Distribution
of Actual Earnings
4
2
0
–0.5
548
–0.4
–0.3
S E P T E M B E R / O C TO B E R , PA R T 2
–0.2
2009
–0.1
0.0
0.1
0.2
0.3
0.4
0.5
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
Figure 1, cont’d
Actual Earnings and Earnings Forecast
C. 12-Month Forecast Horizon
12
Actual Earnings
Earnings Forecast
10
8
6
Normal Distribution
of Actual Earnings
4
2
0
–0.5
–0.4
–0.3
–0.2
–0.1
0.0
0.1
0.2
0.3
0.4
0.5
Table 1
Summary of Minimum and Maximum Values and Observations Suppressed in Figures 1 and 2
12-Month horizon
Variable
Actual earnings
6-Month horizon
Number of
suppressed
Minimum Maximum observations
–1.6137
0.2844
150
1-Month horizon
Number of
suppressed
Minimum Maximum observations
–1.1820
0.3350
58
Number of
suppressed
Minimum Maximum observations
–0.9026
0.2844
11
Earnings forecasts –1.1532
0.2933
76
–0.7732
0.3267
21
–0.6487
0.2778
10
Forecast errors
0.7614
89
–1.1561
0.5533
15
–0.6085
0.3531
2
–1.2442
NOTE: For actual earnings and earnings forecasts there are no positive observations outside the –0.5 to +0.5 range. For forecast errors,
there are 6, 2, and 0 excluded positive observations at the 12-, 6-, and 1-month forecast horizons, respectively; the remainder are
negative observations.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
549
Ciciretti, Dwyer, Hasan
Figure 2
Forecast Errors
A. 1-Month Forecast Horizon
60
50
40
30
20
10
0
–0.5
Normal Distribution
of Actual Earnings
–0.4
–0.3
–0.2
–0.1
0.0
0.1
0.2
0.3
0.4
0.5
B. 6-Month Forecast Horizon
60
50
40
30
20
10
0
–0.5
550
Normal Distribution
of Actual Earnings
–0.4
–0.3
S E P T E M B E R / O C TO B E R , PA R T 2
–0.2
2009
–0.1
0.0
0.1
0.2
0.3
0.4
0.5
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
Figure 2, cont’d
Forecast Errors
C. 12-Month Forecast Horizon
60
50
40
30
20
10
0
–0.5
Normal Distribution
of Actual Earnings
–0.4
–0.3
–0.2
–0.1
means and standard deviations as actual earnings.
Actual and forecasted earnings have higher peaks
at the mean value than the normal distribution
and also have fatter tails. Because the total area
must add up to 100 percent, this implies that the
distributions of actual and forecasted earnings
have fewer observations between the tails and
the center of the distribution.
The graph of the 12-month-ahead forecasts
shows the bias in longer-term forecasts. Although
the distributions of actual and predicted earnings
are quite similar, the histogram shows the tendency of more forecasts of above-average earnings
and fewer forecasts of below-average earnings
than actual earnings. The distribution of the 6month-ahead forecasts shows less bias. The distribution of the 1-month-ahead forecasts is more
similar to the actual earnings.
The literature focuses on the deviations
between the earnings and the forecasts, which
makes it easy to lose sight of how informative the
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
0.0
0.1
0.2
0.3
0.4
0.5
forecasts are about actual earnings. Analysts’
earnings forecasts are quite informative about
actual earnings.
Figure 2 shows the distributions of the forecast errors. A “positive forecast error” means that
actual earnings exceed the forecasted earnings.
A “negative forecast error” means that actual earnings fall short of the forecasted earnings. If all
analysts forecasted earnings within a penny of
earnings per dollar of share price, all the forecast
errors would be within the two bars surrounding
zero. Recall that the share price is the price before
the start of the fiscal year, so this indicates that
the analysts are coming very close to forecasting
actual earnings. In fact, the forecast errors are
quite peaked near zero, whether 12 months, 6
months, or 1 month before the announcement of
actual earnings.
The earnings forecasts are closer to actual earnings 1 month before the earnings announcement
than 12 months before the earnings announcement.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
551
S E P T E M B E R / O C TO B E R , PA R T 2
Distribution of Forecast Errors by Year and Horizon
Minimum
1%
5%
10%
25%
Median
75%
90%
95%
99%
Maximum
Mean
Standard Skewness
deviation coefficient Kurtosis
12-Month horizon
2009
1990
–0.81
–0.4278
–0.1265
–0.0721
–0.0249
–0.0040
0.0003
0.0059
0.0121
0.0456
0.09
–0.0270
0.0754
–4.98
31.33
1991
–0.88
–0.3711
–0.1320
–0.0770
–0.0245
–0.0048
0.0002
0.0068
0.0177
0.0667
0.30
–0.0249
0.0711
–4.95
37.73
1992
–0.40
–0.2019
–0.0922
–0.0509
–0.0158
–0.0023
0.0012
0.0098
0.0193
0.0557
0.12
–0.0141
0.0418
–3.53
18.96
1993
–0.38
–0.1789
–0.0649
–0.0367
–0.0110
–0.0011
0.0022
0.0088
0.0185
0.0636
0.11
–0.0095
0.0368
–3.69
22.69
1994
–0.47
–0.1807
–0.0629
–0.0334
–0.0091
–0.0003
0.0024
0.0100
0.0194
0.0554
0.17
–0.0096
0.0431
–6.08
52.96
1995
–0.27
–0.1297
–0.0618
–0.0367
–0.0099
0.0000
0.0039
0.0118
0.0201
0.0633
0.18
–0.0071
0.0309
–2.50
16.08
1996
–0.29
–0.1455
–0.0697
–0.0379
–0.0100
–0.0001
0.0032
0.0134
0.0256
0.0593
0.20
–0.0078
0.0337
–2.20
13.34
1997
–0.45
–0.1566
–0.0608
–0.0329
–0.0093
–0.0008
0.0023
0.0085
0.0143
0.0400
0.11
–0.0094
0.0362
–5.56
49.00
1998
–0.49
–0.2378
–0.0704
–0.0495
–0.0198
–0.0035
0.0010
0.0060
0.0131
0.0419
0.27
–0.0154
0.0422
–4.19
29.79
1999
–0.76
–0.2484
–0.0743
–0.0391
–0.0119
0.0000
0.0050
0.0224
0.0430
0.1306
0.39
–0.0079
0.0576
–3.74
39.19
2000
–0.51
–0.2230
–0.0752
–0.0395
–0.0120
0.0003
0.0055
0.0276
0.0634
0.1277
0.31
–0.0054
0.0508
–2.41
17.01
2001
–1.24
–0.3840
–0.1364
–0.0785
–0.0335
–0.0086
0.0007
0.0091
0.0208
0.1803
0.76
–0.0265
0.0895
–4.00
50.19
2002
–0.74
–0.2228
–0.0656
–0.0370
–0.0114
–0.0002
0.0064
0.0234
0.0426
0.0976
0.32
–0.0067
0.0522
–5.09
53.33
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
2003
–0.71
–0.1839
–0.0617
–0.0339
–0.0104
0.0003
0.0092
0.0266
0.0443
0.0949
0.28
–0.0045
0.0464
–3.98
38.24
2004
–0.33
–0.1148
–0.0438
–0.0212
–0.0068
0.0010
0.0088
0.0264
0.0394
0.0812
0.14
–0.0003
0.0317
–3.10
26.77
6-Month horizon
1990
–1.16
–0.2730
–0.0955
–0.0427
–0.0122
–0.0016
0.0008
0.0060
0.0142
0.0575
0.20
–0.0162
0.0669
–7.95
92.95
1991
–0.54
–0.2171
–0.0642
–0.0353
–0.0097
–0.0015
0.0009
0.0074
0.0176
0.0600
0.18
–0.0108
0.0441
–5.33
44.17
1992
–0.32
–0.1301
–0.0444
–0.0219
–0.0071
–0.0006
0.0013
0.0062
0.0122
0.0357
0.11
–0.0066
0.0276
–5.01
39.50
1993
–0.16
–0.0814
–0.0247
–0.0137
–0.0037
–0.0001
0.0018
0.0066
0.0142
0.0409
0.18
–0.0024
0.0181
–2.34
24.80
1994
–0.17
–0.0705
–0.0284
–0.0159
–0.0041
0.0000
0.0024
0.0076
0.0129
0.0400
0.16
–0.0025
0.0170
–1.96
20.70
1995
–0.30
–0.0828
–0.0330
–0.0169
–0.0044
0.0000
0.022
0.065
0.111
0.2930
0.10
–0.0038
0.0198
–5.37
52.00
1996
–0.32
–0.0969
–0.0287
–0.0152
–0.0038
0.0001
0.0024
0.0090
0.0151
0.0389
0.19
–0.0029
0.0227
–4.78
54.34
1997
–0.27
–0.0907
–0.0275
–0.0132
–0.0030
0.0001
0.0023
0.0079
0.0146
0.0422
0.17
–0.0021
0.0206
–2.77
38.07
NOTE: *This test statistic has a chi-square distribution with two degrees of freedom under the null hypothesis. The value of this chi-square at the 0.001 level of significance
is 13.8. All of the values in the table have p-values less than 10–8.
Ciciretti, Dwyer, Hasan
552
Table 2
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Table 2, cont’d
Distribution of Forecast Errors by Year and Horizon
Minimum
1%
5%
10%
25%
Median
75%
90%
95%
99%
Maximum
Mean
Standard Skewness
deviation coefficient Kurtosis
6-Month horizon, cont’d
1998
–0.33
–0.0992
–0.0359
–0.0219
–0.0081
–0.0016
0.0008
0.0043
0.0094
0.0290
0.29
–0.0063
0.0226
–3.18
49.61
1999
–0.56
–0.1600
–0.0446
–0.0202
–0.0048
0.0001
0.0031
0.0109
0.0193
0.0533
0.55
–0.0052
0.0383
–3.74
78.39
26.48
2000
–0.36
–0.1101
–0.0447
–0.0221
–0.0059
0.0000
0.0022
0.0136
0.0261
0.0668
0.17
–0.0037
0.0273
–2.68
2001
–0.64
–0.1714
–0.0494
–0.0274
–0.0092
–0.0015
0.0012
0.0074
0.0141
0.0581
0.20
–0.0085
0.0391
–5.95
66.46
2002
–0.38
–0.0997
–0.0325
–0.0158
–0.0054
–0.0003
0.0027
0.0088
0.0159
0.0402
0.21
–0.0038
0.0269
–6.09
76.24
2003
–0.49
–0.0994
–0.0295
–0.0140
–0.0036
0.0004
0.0045
0.0125
0.0213
0.0667
0.38
–0.0011
0.0310
–2.52
68.31
2004
–0.29
–0.0617
–0.0284
–0.0184
–0.0045
0.0000
0.0032
0.0092
0.0164
0.0389
0.09
–0.0025
0.0195
–5.05
57.05
1-Month horizon
1990
–0.61
–0.0970
–0.0286
–0.0146
–0.0031
–0.0001
0.0014
0.0054
0.0131
0.0526
0.22
–0.0035
0.0342
–11.48
204.59
1991
–0.24
–0.0659
–0.0231
–0.0111
–0.0024
0.0000
0.0020
0.0074
0.0141
0.0395
0.13
–0.0015
0.0188
–2.99
48.29
1992
–0.14
–0.0698
–0.0118
–0.0053
–0.0010
0.0002
0.0025
0.0073
0.0144
0.0402
0.24
0.0006
0.0220
4.09
61.43
1993
–0.26
–0.0659
–0.0127
–0.0064
–0.0012
0.0001
0.0020
0.0062
0.0112
0.0400
0.10
–0.0005
0.0154
–4.97
71.88
1994
–0.11
–0.0274
–0.0079
–0.0039
–0.0007
0.0002
0.0020
0.0057
0.0104
0.0289
0.09
0.0006
0.0102
–1.20
41.64
1995
–0.22
–0.0455
–0.0093
–0.0048
–0.0009
0.0002
0.0019
0.0057
0.0114
0.0390
0.31
0.0004
0.0188
1.28
104.42
–0.20
–0.0277
–0.0078
–0.0036
–0.0005
0.0003
0.0017
0.0054
0.0097
0.0482
0.17
0.0008
0.0137
–0.90
89.84
–0.36
–0.0375
–0.0114
–0.0047
–0.0006
0.0003
0.0019
0.0054
0.0096
0.0325
0.19
0.0002
0.0145
–6.48
217.27
1998
–0.16
–0.0256
–0.0089
–0.0044
–0.0006
0.0003
0.0017
0.0050
0.0089
0.0285
0.20
0.0004
0.0102
1.12
110.97
1999
–0.23
–0.0410
–0.0069
–0.0031
–0.0004
0.0004
0.0023
0.0062
0.0116
0.0457
0.28
0.0011
0.0158
1.31
118.62
2000
–0.24
–0.0673
–0.0141
–0.0057
–0.0007
0.0002
0.0013
0.0044
0.0088
0.0291
0.11
–0.0011
0.0147
–6.61
83.84
2001
–0.18
–0.0371
–0.0101
–0.0038
–0.0005
0.0002
0.0014
0.0038
0.0066
0.0211
0.08
–0.0004
0.0104
–6.24
94.52
2002
–0.26
–0.0340
–0.0079
–0.0036
–0.0005
0.0003
0.0013
0.0038
0.0067
0.0211
0.35
–0.0002
0.0135
–0.63
268.15
2003
–0.36
–0.0645
–0.0100
–0.0047
–0.0007
0.0003
0.0018
0.0054
0.0097
0.0373
0.15
–0.0003
0.0157
–7.81
145.27
2004
–0.15
–0.0333
–0.0078
–0.0037
–0.0007
0.0004
0.0022
0.0052
0.0087
0.0255
0.15
0.0006
0.0092
–0.89
77.55
2009
NOTE: *This test statistic has a chi-square distribution with two degrees of freedom under the null hypothesis. The value of this chi-square at the 0.001 level of significance
is 13.8. All of the values in the table have p-values less than 10–8.
553
Ciciretti, Dwyer, Hasan
S E P T E M B E R / O C TO B E R , PA R T 2
1996
1997
Ciciretti, Dwyer, Hasan
This convergence is expected if the forecasts are
informed predictions. More information becomes
available as time goes on, and this information is
substantial: Eleven-twelfths of the year is past
when the 1-month-ahead forecast is made. Firms
announce earnings quarterly; when the 1-monthahead forecast is made, earnings for the first three
quarters of the year have been announced and
are known. Besides this relatively mechanical
effect as time passes, other information becomes
known about earnings as time passes and the
magnitudes of forecast errors can be expected to
decrease.
Over 90 percent of the forecasts made 1 month
before the earnings announcement are within
one penny of earnings per dollar of share price.
There is a clear asymmetry in the distribution of
these close forecast errors: 60 percent of the earnings are more than the forecasts and within a
penny; 30 percent of the earnings are less than
the forecasts and within a penny. The larger number of positive forecast errors can reflect analysts’
forecasts that the analyst knows are too low; it also
can occur for other reasons. For example, firms
with actual earnings less than forecasted earnings
may provide analysts with information before the
announcement and forecasts are revised accordingly. The forecast errors 12 months ahead and 6
months ahead also show asymmetry, with many
forecasts within a penny of actual earnings but
more above zero than below.
Table 2 shows detailed information about the
distributions of forecast errors by year at 12-month,
6-month, and 1-month horizons. The table shows
the maximum and minimum values, the mean,
standard deviation, measures of the skewness,
and kurtosis of the distribution of forecast errors
and selected percentiles of the distributions.
As Figure 2 suggests, the forecasts a month
before the earnings announcement are much
closer to actual earnings than are forecasts a year
in advance. The standard deviation of forecast
errors is a measure of the size of analysts’ errors,
independent of whether the forecast is above or
below actual earnings. The standard deviation is
substantially larger 12 months before earnings
are announced than 1 month before the earnings
announcement. For example, in 1990, the standard
deviation is 0.0754 at a horizon of 12 months,
554
S E P T E M B E R / O C TO B E R , PA R T 2
2009
0.0669 at a horizon of 6 months, and 0.0342 at a
horizon of 1 month. In 2004, the standard deviation is 0.0317 at a horizon of 12 months, 0.0195
at a horizon of 6 months, and 0.0092 at a horizon
of 1 month.
The mean forecast errors in the table also
decline as the announcement of earnings for the
year approaches. The largest magnitudes of mean
forecast errors in the table are for the 12-month
horizon, –2.7 cents per dollar of share price in
1990 and 2001 and –2.5 cents per dollar of share
price in 1991. The smallest magnitudes of mean
forecast errors are for the 1-month horizon; the
mean forecast error farthest from zero is –0.35
cents per dollar of share price in 1990. The mean
forecast error has been hundredths of a penny per
dollar of share price in most of the years since.
A large segment of the literature examines
these mean forecast errors. The negative mean
forecast errors are statistically significant and not
trivial in magnitude at the 12-month horizon.
Twelve months before earnings are announced,
analysts’ forecasts on average are overestimates
of actual earnings. This overestimation is predictable, in an interesting and specific sense. If only
the earnings forecasts are known a year in advance,
it is predictable that actual earnings will be less
on average. The difference is not large, but it is
not zero and it is predictable. If analysts are
attempting to forecast earnings well on average,
their performance is not as good as it could be.
In standard parlance, the forecasts are biased:
The average forecast error is not zero.
Besides the arithmetic average, the median
is another measure of the typical forecast. The
median is the middle forecast, the forecast that
divides the forecasts into two parts, with half the
observations above the median and half below
the median. The median forecast error is noticeably closer to zero than the average forecast error.
This indicates that the typical negative forecast
error is larger in magnitude than the typical positive forecast error. In other words, as Figure 2
shows, the distribution of forecast errors is not
symmetric. The percentiles of the distribution
clearly show this asymmetry of forecast errors.
The consistently negative values of skewness in
Table 2 also indicate what Figure 2 shows: NegaF E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
tive forecast errors are larger in magnitude than
the positive errors.7 Consistent with the figures,
the measure of skewness indicates that forecast
errors are skewed toward negative values.
Kurtosis measures how concentrated a distribution is around the mean compared with the
number of observations in the tails of the distribution.8 The positive values for kurtosis indicate
that the tails of the distribution have more observations than would be suggested by a normal distribution. Tests for normality of the distribution
of forecast errors uniformly are inconsistent with
a normal distribution.9
Figures 3 and 4 show aspects of the distributions of forecast errors for all horizons from 1990
to 2004. Figure 3 shows the mean and median
forecast errors as the horizon—the length of time
before the earnings announcement—approaches
zero. It also shows the median in combination with
the 25th and 75th percentiles of the distribution
of forecast errors. The mean forecast errors are
more strongly negative than the medians at long
horizons and consequently show more convergence to zero. The median forecast errors are negative, with the largest magnitudes in 1990, 1991,
1998, and 2001 (see Figure 4). With the exception
of 1998, these larger-magnitude median forecast
errors are associated with recessions.10 The mean
forecast errors are more strongly negative than
the median forecast errors but decrease to quite
close to zero by 1 month before the earnings
announcement.
Figure 4 shows the distribution of forecast
errors by year by graphing the median forecast
error and the 25th and 75th percentiles of the
7
The measure of skewness is the third moment about the mean
divided by the standard deviation cubed.
8
The measure of kurtosis is the fourth moment about the mean
minus 3, all relative to the fourth power of the standard deviation.
9
The test for normality is the Bera-Jarque test (1980). The inconsistency with a normal distribution matches up with the figures and
tables; a normal distribution is symmetric and does not have the
relatively fat tails indicated by the kurtosis statistics. The BeraJarque test statistics are not included in the table because the
p-values uniformly are inconsistent with a normal distribution
with p-values of 10–8 or below.
10
The National Bureau of Economic Research dates the recession in
1990 and 1991 from July 1990 to March 1991 and the recession in
2001 from March 2001 to November 2001.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
distribution for each horizon for each year from
1990 to 2004. The asymmetry of the distributions
is quite apparent. It also is clear that actual earnings fall short of the longer horizon forecasts during recessions; this is indicated by the much more
negative forecast errors during the recession years
1990, 1991, and 2001. Given the unpredictability
of recessions, this is not especially surprising. The
figure suggests that the distribution has become
more symmetric over time, although the occurrence of recessions clearly is associated with
greater asymmetry.
Table 3 presents the results of tests to determine whether the apparent skewness in the figures
is statistically significant and consistent across
horizons and years.11 The results of two tests are
presented. The first is the sign test, which determines whether the median equals the mean. If a
series’ median exceeds its mean, the value of the
statistic is positive. The p-value indicates the probability of that difference or a larger one if there
really were no difference between the median and
the mean. The second test determines whether
the skewness coefficient is zero. If the skewness
coefficient is zero and moments of the distribution
up to the sixth are finite, then the skewness coefficient has an asymptotic normal distribution that
can be used to construct a test.12
The sign tests indicate an asymmetry in forecast errors that persists from 1990 through 2004.
Tests for the equality of the median and mean at
all horizons are quite inconsistent with the equality of the two statistics. At the 12-month horizon,
the median forecast error is closer to zero than
the mean for all years from 1990 through 2004;
all of the differences are statistically significant
at any usual significance level. There is some
suggestion that the difference between the mean
and the median has been declining over time.
The difference is far smaller in 2004 than in earlier years but the difference still is statistically
11
The observations are repeated measures of forecasts by the same
analysts for the same industries. As Keane and Runkle (1998) argue,
this can introduce dependence in the data, which results in overstating the statistical significance of test statistics.
12
The mean of the asymptotic distribution of the skewness coefficient is zero under the null hypothesis and the variance is from
Gupta (1967, pp. 850-51.)
S E P T E M B E R / O C TO B E R , PA R T 2
2009
555
Ciciretti, Dwyer, Hasan
Figure 3
Forecast Errors by Horizon
0.002
–0.002
–0.006
–0.010
Median
Mean
–0.014
12
11
10
9
8
7
6
5
4
3
2
1
Forecast Horizon
0.010
0.000
–0.010
–0.020
–0.030
Median
25th and 75th Percentiles
–0.040
12
11
10
9
8
7
6
5
4
3
2
1
Forecast Horizon
556
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
Figure 4
Distribution of Forecast Errors by Year and Horizon
0.010
0.000
–0.010
–0.020
–0.030
Median
25th and 75th Percentiles
–0.040
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Year and Forecast Horizon
significant. The difference is one-tenth of a penny
per dollar share price in 2004. Given a typical
price-to-earnings ratio of 15 or 20, this implies a
forecast error in earnings on the order of 2 cents
per share per dollar of earnings 12 months ahead.
The tests using the skewness coefficient indicate that deviations from symmetry are persistent
from 1990 through 2004 only at the 12-month
horizon. The null hypothesis of symmetry for
the 12-month horizon cannot be rejected in 2002
at the 5 percent significance level, a result most
simply interpreted as due to chance rather than
anything special about 2002. There is less evidence of overall skewness in any year at the 6month horizon and scant evidence of asymmetry
at the 1-month horizon. This is an interesting contrast to the results using the median and mean.
While there are statistically significant differences
between the mean and median, the overall skewness of the distribution is less pronounced based
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
on the third moment, which summarizes the
asymmetry of the distribution.13
Forecasts Errors Across Industries
Forecast errors across firms and analysts are
likely to differ for a variety of reasons, one being
the likelihood that earnings are more predictable
for some industries than others.
Figure 5 shows forecast errors by two-digit
Global Industry Classification System categories.
Forecast errors vary substantially by industry.
All figures have the same scale to facilitate comparison of forecast errors across industries. Earnings in health care are predicted with relatively
13
Too many rejections of the null hypothesis are possible if data
have high kurtosis (Premaratne and Bera, 2005), as ours do. This
is an issue only at the 12-month horizon because only that horizon
shows rejections. Given the results for the median and mean and
the levels of significance, we are inclined to take the rejections as
being real rather than an artifact of kurtosis.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
557
S E P T E M B E R / O C TO B E R , PA R T 2
Ciciretti, Dwyer, Hasan
558
Table 3
Sign Test Statistics and Skewness Coefficients by Year and Horizon
Sign test
12-month horizon
2009
Year
Mean minus
median
p-Value
6-month horizon
Mean minus
median
p-Value
Skewness coefficient
1-month horizon
Mean minus
median
p-Value
12-month horizon
Coefficient
p-Value
6-month horizon
Coefficient
p-Value
1-month horizon
Coefficient
p-Value
1990
–0.0230
0.0000
–0.0146
0.0000
–0.0034
0.0000
–5.806
0.0000
–0.370
0.7116
–0.024
0.9807
1991
–0.0201
0.0000
–0.0094
0.0000
–0.0015
0.0000
–3.825
0.0001
–2.375
0.0176
–0.014
0.9885
1992
–0.0118
0.0000
–0.0060
0.0000
0.0004
0.0000
–16.085
0.0000
–2.999
0.0027
0.337
0.7360
1993
–0.0085
0.0000
–0.0023
0.0000
–0.0006
0.0000
–9.796
0.0000
–5.912
0.0000
–0.183
0.8546
1994
–0.0092
0.0000
–0.0025
0.0000
0.0004
0.0000
–2.374
0.0176
–1.901
0.0574
–0.009
0.9925
1995
–0.0071
0.0000
–0.0038
0.0000
0.0001
0.0007
–20.569
0.0000
–1.532
0.1256
0.046
0.9634
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
1996
–0.0077
0.0000
–0.0030
0.0000
0.0005
0.0000
–37.030
0.0000
–1.409
0.1588
–0.049
0.9611
1997
–0.0086
0.0000
–0.0022
0.0000
–0.0002
0.0000
–2.637
0.0084
–2.637
0.0084
–0.017
0.9867
1998
–0.0120
0.0000
–0.0047
0.0000
0.0002
0.0000
–14.384
0.0000
–2.011
0.0444
0.008
0.9933
1999
–0.0079
0.0000
–0.0053
0.0000
0.0007
0.0000
–3.588
0.0003
–0.552
0.5812
0.019
0.9849
2000
–0.0057
0.0000
–0.0037
0.0000
–0.0013
0.0000
–24.850
0.0000
–4.124
0.0000
–0.188
0.8507
2001
–0.0179
0.0000
–0.0070
0.0000
–0.0007
0.0000
–2.469
0.0136
–0.864
0.3877
0.000
0.9999
2002
–0.0065
0.0000
–0.0035
0.0000
–0.0004
0.0000
–1.841
0.0657
–0.721
0.4708
0.002
0.9987
2003
–0.0048
0.0000
–0.0015
0.0000
–0.0006
0.0000
–2.926
0.0034
–0.415
0.6780
–0.081
0.9358
2004
–0.0013
0.0000
–0.0025
0.0000
0.0002
0.0026
–7.336
0.0000
–1.362
0.1731
0.033
0.9739
Ciciretti, Dwyer, Hasan
Figure 5A
Distribution of Forecast Errors by Year, Horizon, and Industry
Materials
Industrials
0.030
0.030
0.004
0.004
–0.022
–0.022
–0.048
–0.048
–0.074
–0.074
Median
25th and 75th Percentiles
–0.100
Median
25th and 75th Percentiles
90
Year and Forecast Horizon
Consumer Discretionary
Consumer Staples
0.030
0.030
0.004
0.004
–0.022
–0.022
–0.048
–0.048
–0.074
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
19
Year and Forecast Horizon
19
90
19
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
–0.100
–0.074
Median
25th and 75th Percentiles
Median
25th and 75th Percentiles
–0.100
Year and Forecast Horizon
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
19
90
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
19
19
90
–0.100
Year and Forecast Horizon
S E P T E M B E R / O C TO B E R , PA R T 2
2009
559
Ciciretti, Dwyer, Hasan
Figure 5B
Distribution of Forecast Errors by Year, Horizon, and Industry
Health Care
Financials
0.030
0.030
0.004
0.004
–0.022
–0.022
–0.048
–0.048
–0.074
–0.074
Median
25th and 75th Percentiles
Median
25th and 75th Percentiles
–0.100
Year and Forecast Horizon
Year and Forecast Horizon
Information Technology
Utilities
0.030
0.030
0.004
0.004
–0.022
–0.022
–0.048
–0.048
–0.074
–0.074
Median
25th and 75th Percentiles
Median
25th and 75th Percentiles
–0.100
S E P T E M B E R / O C TO B E R , PA R T 2
2009
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
90
19
19
90
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
–0.100
Year and Forecast Horizon
560
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
19
19
90
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
19
19
90
–0.100
Year and Forecast Horizon
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
Figure 5C
Distribution of Forecast Errors by Year, Horizon, and Industry
Energy
Telecommunication Services
0.090
0.090
Median
25th and 75th Percentiles
0.030
0.030
0.000
0.000
–0.030
–0.030
–0.060
–0.060
90
19
Year and Forecast Horizon
small forecast errors, and earnings in energy firms
are predicted particularly poorly. It is plausible
that earnings forecasts in less-volatile industries
are smaller. Energy prices are subject to large
unpredictable price swings, which obviously
affect earnings. Although health care prices have
risen substantially in recent years, the increases
have been relatively persistent and therefore
predictable. Health care is virtually unaffected
by recessions, while the demand for energy falls
in recessions. Some other industries show low
earnings around recessions as well, such as materials and consumer discretionary goods. If recessions are not predicted, there is little reason to
think that these earnings decreases are predictable
either.
Sign tests not reported in the text are consistent with persistent differences between the
median and means of the forecast errors across
industries but suggest variation in the asymmetry
by industry. The evidence is noticeably weaker
for telecommunications and utilities.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
19
19
19
90
0.060
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
0.060
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
Median
25th and 75th Percentiles
Year and Forecast Horizon
UNBIASEDNESS OF EARNINGS
FORECASTS
Almost all of the existing literature on analysts’
forecasts examines whether their forecasts are
biased and, generally speaking, finds that analysts
overestimate earnings. This overestimation falls
as the earnings announcement approaches, as
indicated in Table 2, but future earnings typically
are noticeably less than the average forecast. Some
evidence and analysis suggests that analysts’ forecasts change from overestimates to underestimates
just before the earnings announcement. Such
near-term forecasts are intended to be helpful to
a firm’s management because the announcement
of higher-than-forecasted earnings generates
favorable publicity and a higher stock price after
the announcement.14
Asking for forecasts that are neither too high
nor low on average seems like a relatively simple
14
This is at least one reason to be dubious about this explanation if
the near-term underestimation of earnings is persistent and predictable. Investors are likely to notice and discount the overestimation
of earnings.
S E P T E M B E R / O C TO B E R , PA R T 2
2009
561
Ciciretti, Dwyer, Hasan
request, especially compared with asking that
forecasts be accurate. Even so, it is possible that
analysts process the information available to them
as best as possible, but some or all analysts do not
have an incentive to produce forecasts that are
correct on average.
Analysts’ Incentives and Forecasts
At first glance, it seems obvious that unbiased
forecasts are the best forecasts. A biased forecast
is high or low on average. Such a bias suggests
that the forecast can be improved by adjusting the
forecast by the bias. There are many conditions
in which an unbiased forecast is the best one.
A common criterion for forecast errors is mean
squared error. If a forecaster wants to minimize the
expected mean squared error of a forecast, then an
unbiased forecast is the best one.15 The expected
squared forecast error applies an increasing penalty
to forecasts farther from the average—a forecast
twice as far from zero is four times as bad.
The unbiased forecast—the mean—is not
necessarily the best forecast in all circumstances.
Suppose that someone is trying to forecast the
value shown when a fair die is thrown. The mean
forecast is the average of 1, 2, 3, 4, 5, and 6, which
is 3.5. If the forecaster’s earnings depend on how
close the forecast is to the actual value, the best
forecast in fact is 3.5. On the other hand, if the
forecaster gets paid only when the value shown
is the same as the value forecasted, this unbiased
forecast guarantees that the forecaster always loses.
The die will never have the value 3.5. If the forecaster is paid when the forecast is the same as the
value thrown and values from 1 to 6 are equally
likely, any integer forecast from 1 to 6 is equally
good and 3.5 never is predicted. While this is a
simple example, the point is more general. The
value forecasted depends on the forecaster’s incentives and the distribution of the data. An unbiased
forecast may not be the “best” forecast.
There also are objectives similar to minimizing
the expected squared error that lead to forecasts
being “biased.” If a forecaster wants to minimize
the expected absolute deviation of the forecast
15
A minimum expected squared error forecast minimizes the expected
value of the squared forecast errors.
562
S E P T E M B E R / O C TO B E R , PA R T 2
2009
error, then the median is the best forecast.16
The absolute forecast error applies an increasing
penalty to forecast errors farther from zero—a forecast error twice as far from zero is twice as bad.
The cost of forecast errors increases linearly with
the size of the error. The forecast that minimizes
the expected absolute forecast error is the median,
not the mean (or more precisely, the arithmetic
average). If the mean and the median are the same,
this is a distinction that does not matter. On the
other hand, if the distribution is not symmetric,
as the earnings distribution is not, the median is
a better forecast than the mean if a forecast error’s
cost increases linearly with the forecast error. 17
Analysts do not make forecasts in isolation.
Other analysts are making forecasts as well, and
the existence of other forecasts can affect an analyst’s forecasts in many ways. A simple, common
forecasting game illustrates that an unbiased forecast may not be an analyst’s best forecast. Consider
a forecasting game in which the smallest forecast
error wins and receives a prize; everyone else
receives nothing. Analysts’ situations may be
closer to this game than to isolated forecasts. In
this game, the incentive is to be the closest. If
you are not the closest, then it matters not at all
whether your forecast error is almost as good as the
best or is far away. More generally, any analyst’s
forecast will depend on what he or she thinks
other people will forecast or what others have
already forecasted. A simple example is one in
which two people guess someone else’s pick of a
number between 0 and 10. The unbiased forecast
is 5. Suppose that the first person picks 5. If the
second person picks 5, then he or she cannot win,
only tie. A pick of either 4 or 6 can increase the
expected winnings of the second person if there is
no payoff from tying. Neither 4 nor 6 is unbiased,
but that doesn’t matter. Either number maximizes
expected winnings, and it is winnings that matter.
This suggests that, even if analysts’ forecasts are
biased, it is important to consider analysts’ incentives before denouncing them as “irrational” or
“ignoring information readily available to them.”
16
A minimum expected absolute error forecast minimizes the
expected absolute value of the forecast errors.
17
Gu and Wu (2003) discuss this in more detail.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
Among others, Hong and Kubik (2003), Clarke
and Subramanian (2006), Ottaviani and Sørensen
(2006), and Ljungqvist et al. (2007) highlight factors that can explain a nonzero predictable forecast error. For example, Clarke and Subramanian
(2006) suggest that an analyst who performs
poorly and is at risk of being fired is more likely
to make a “bold” forecast that is unlikely to be
correct but will save the analyst’s job if it is correct.
Tests for Unbiasedness
The proposition that analysts’ forecasts are
biased is simple to determine with a test of
whether the average difference between actual
earnings and forecasted earnings is zero.18 Given
the evidence above that forecast errors are not
symmetric, it is worthwhile to test whether the
median forecast error is zero, in addition to testing
whether the mean forecast error is zero. A simple
t-test is used for the latter purpose. The test that
analysts’ median forecast errors are zero is the
sign test for deviations from zero.
Table 4 presents the mean and median forecast
errors by industry at the various horizons and
p-values for tests of whether the mean and median
forecast errors are zero. The mean forecast errors
are far smaller at the 1-month horizon than at
longer horizons. At the 12-month horizon, the
mean forecast error indicates that forecasted earnings are greater than actual earnings by about 1
cent per dollar of share price. At the 1-month
horizon, the mean forecast errors indicate that
forecasted earnings are greater than actual earnings by about one-hundredth of a cent per dollar
of stock price.
How big are these forecast errors? Mean earnings for all firms in our data are 2 cents per dollar
of share price; median earnings are 3.9 cents per
dollar of share price. A forecast error of 1 cent per
dollar of share price at the 12-month horizon is
large relative to average earnings of 2 cents. A
forecast error of one-hundredth of a cent at the
1-month horizon is relatively small and not obviously economically insignificant.
The median forecast error for all industries is
minus nine-hundredths of a cent per dollar of
18
The test is a standard t-test of whether the mean forecast error
equals zero using the asymptotic normal distribution.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
share price at the 12-month horizon. At the 6month and 1-month horizons, the median forecast
errors are minus two-hundredths of a cent per
dollar of share price and three-hundredths of a
dollar per dollar of share price. All these magnitudes based on the median are statistically significantly different from zero. Median forecast errors
of hundredths of a cent per dollar of share price
are not particularly large relative to median earnings of about 4 cents per dollar of share price.
The means and medians vary substantially
by industry. The mean forecast errors by industry
mirror the overall mean forecast errors, declining
in magnitude as the horizon shortens. The median
forecast errors show substantial variability across
industries in terms of magnitude. At the 1-month
horizon, all of the magnitudes are of the same
small order as the overall median, with the largest
being five-hundredths of a cent per dollar of
share price.
Table 5 shows the results of tests to determine
whether the average and median forecast errors
are zero by year. With the exception of the last
year in the table, 2004, all p-values for testing
whether mean forecast errors are zero at the 12month horizon are less than 10–4. All mean forecast errors are negative, indicating that forecasts
on average are greater than actual earnings. Mean
forecasts 6 months ahead look much like the forecasts at the 12-month horizon. The forecasts at the
1-month horizon look quite a bit different. At the
1-month horizon, there is little evidence in our
data of bias in the mean forecast: 8 of the 15 forecasts are positive and 7 are negative. Nine of the
forecasts are statistically significant at the 5 percent level, but they are not uniformly positive or
negative. There is little evidence to support a conclusion that mean forecasts at the 1-month horizon
are uniformly above or below zero.
The median forecasts in Table 5 are closer to
zero than the mean forecasts. The results of the
statistical tests that the median forecasts equal
zero indicate that they are not zero, but the magnitudes generally are hundredths of a cent per
dollar of share price.
At the 12-month horizon, the overall median
forecast error is negative, but this masks interesting variation by year. In five years—1995, 1999,
S E P T E M B E R / O C TO B E R , PA R T 2
2009
563
S E P T E M B E R / O C TO B E R , PA R T 2
Ciciretti, Dwyer, Hasan
564
Table 4
Forecast Errors by Industry and Horizon
12-month horizon
2009
Mean
p-Value
mean
equals zero
All industries
–0.0106
Consumer discretionary
Consumer staples
Industry
6-month horizon
Median
p-Value
median
equals zero
Mean
p-Value
mean
equals zero
0.0000
–0.0009
0.0000
–0.0048
–0.0124
0.0000
–0.0067
0.0000
–0.0017
0.0000
–0.0003
0.0000
1-month horizon
Median
p-Value
median
equals zero
Mean
p-Value
mean
equals zero
Median
p-Value
median
equals zero
0.0000
–0.0002
0.0000
–0.0001
0.3456
0.0003
0.0000
–0.0070
0.0000
–0.0039
0.0000
–0.0009
0.0000
–0.0002
0.3400
0.0003
0.0000
–0.0001
0.0078
–0.0002
0.4837
0.0002
0.0000
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Energy
–0.0002
0.8012
0.0002
0.1833
–0.0015
0.0001
–0.0003
0.0172
0.0003
0.4056
0.0005
0.0000
Financials
–0.0101
0.0000
0.0000
0.2480
–0.0050
0.0000
0.0001
0.0000
–0.0005
0.0410
0.0002
0.0000
Health care
–0.0043
0.0000
0.0000
0.6499
–0.0017
0.0001
0.0001
0.0000
0.0002
0.5447
0.0002
0.0000
Industrials
–0.0163
0.0000
–0.0025
0.0000
–0.0092
0.0000
–0.0012
0.0000
0.0007
0.0372
0.0003
0.0000
Information technology
–0.0159
0.0000
–0.0016
0.0000
–0.0043
0.0000
0.0000
0.5310
–0.0004
0.0890
0.0003
0.0000
Materials
–0.0208
0.0000
–0.0084
0.0000
–0.0078
0.0000
–0.0027
0.0000
0.0003
0.2840
0.0004
0.0000
Telecommunication
services
–0.0099
0.0000
–0.0018
0.0000
–0.0043
0.0001
–0.0002
0.0131
–0.0009
0.2061
0.0002
0.0001
Utilities
–0.0050
0.0000
–0.0009
0.0000
–0.0021
0.0062
–0.0003
0.0220
–0.0006
0.0732
0.0001
0.0004
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Table 5
Forecast Errors by Year and Horizon
12-month horizon
Year
1990-2004
6-month horizon
Mean
p-Value
mean
equals zero
Median
p-Value
median
equals zero
–0.0111
0.0000
–0.0010
0.0000
Mean
–0.0048
p-Value
mean
equals zero
0.0000
1-month horizon
Median
p-Value
median
equals zero
–0.0002
0.0000
Mean
p-Value
mean
equals zero
Median
p-Value
median
equals zero
–0.0000
0.7701
0.0003
0.0000
1990
–0.0270
0.0000
–0.0040
0.0000
–0.0162
0.0000
–0.0016
0.0000
–0.0035
0.0016
–0.0001
0.0253
1991
–0.0249
0.0000
–0.0048
0.0000
–0.0108
0.0000
–0.0015
0.0000
–0.0015
0.0286
0.0000
0.5331
1992
–0.0141
0.0000
–0.0023
0.0000
–0.0066
0.0000
–0.0006
0.0000
0.0006
0.4243
0.0002
0.0001
1993
–0.0095
0.0000
–0.0011
0.0000
–0.0024
0.0000
–0.0001
0.0012
–0.0005
0.1985
0.0001
0.0000
1994
–0.0096
0.0000
–0.0004
0.0000
–0.0025
0.0000
0.0000
0.6343
0.0006
0.0062
0.0002
0.0000
1995
–0.0071
0.0000
0.0000
0.6729
–0.0038
0.0000
0.0000
0.1360
0.0004
0.3581
0.0003
0.0000
1996
–0.0078
0.0000
–0.0001
0.2249
–0.0029
0.0000
0.0001
0.0117
0.0008
0.0069
0.0004
0.0000
1997
–0.0094
0.0000
–0.0008
0.0000
–0.0021
0.0000
0.0001
0.0019
0.0002
0.6129
0.0004
0.0000
–0.0155
0.0000
–0.0035
0.0000
–0.0063
0.0000
–0.0016
0.0000
0.0004
0.0338
0.0003
0.0000
–0.0079
0.0000
0.0000
0.8265
–0.0052
0.0000
0.0001
0.0015
0.0011
0.0008
0.0004
0.0000
2000
–0.0054
0.0000
0.0003
0.0024
–0.0037
0.0000
0.0000
0.0106
–0.0011
0.0002
0.0002
0.0000
2001
–0.0265
0.0000
–0.0086
0.0000
–0.0085
0.0000
–0.0015
0.0000
–0.0004
0.0297
0.0002
0.0000
2002
–0.0067
0.0000
–0.0002
0.1688
–0.0038
0.0000
–0.0003
0.0001
–0.0002
0.5212
0.0003
0.0000
2003
–0.0045
0.0000
0.0003
0.0289
–0.0011
0.0669
0.0004
0.0000
–0.0003
0.3086
0.0003
0.0000
2004
–0.0003
0.5223
0.0010
0.0000
–0.0025
0.0000
0.0000
0.2696
0.0006
0.0012
0.0004
0.0000
2009
565
Ciciretti, Dwyer, Hasan
S E P T E M B E R / O C TO B E R , PA R T 2
1998
1999
Ciciretti, Dwyer, Hasan
2000, 2003, and 2004—the median forecast error
at the 12-month horizon is positive, indicating
that the median forecast is an underestimate of
earnings. This is the opposite of the bias in the
mean forecast. It is interesting that these years
are toward the end of the period. For four years—
1995, 1996, 1999, and 2002—the median forecast
error is not statistically significantly different
from zero at the 5 percent significance level. Two
of these years have positive median forecast errors
and two have negative ones. At this 12-month
horizon, only 8 of the 15 years have median forecast errors that are negative and statistically significant. Moreover, of the medians at this 12-month
horizon from 1999 to 2004, only the recession
year 2001 has a negative median forecast error
that is statistically significantly different than
zero; 3 of the 5 years have positive median forecast errors that are statistically significant. These
results are consistent with the median forecast
errors not always being zero, but there is little
support for the median forecasts uniformly being
too high or too low.
At the 6-month horizon, median forecast
errors also provide little support for typical overestimation of earnings throughout the period. The
median forecast errors are negative in 8 of the 15
years, barely more than half the 15 years. The
median forecast errors are positive and statistically
significant at the 5 percent significance level in
years 1996, 1997, 1999, 2000, and 2003.
At the 1-month horizon, the median forecast
errors are positive in all years but 1990, a result
consistent with the stylized view in the literature
that forecast errors are underestimates close to
the announcement. It is interesting that our data
support such an inference using medians but provide much less support with means. All median
forecast errors at the 1-month horizon are quite
small, never larger in magnitude than fourhundredths of a cent per dollar of share price.
Economically, this is not that far from zero.
typical results: Analysts’ forecasts are greater
than earnings on average a year before earnings
are announced. Six months before the earnings
announcements, mean earnings forecasts also are
greater than actual earnings. On the other hand,
median earnings forecasts are about as likely to
be above actual earnings as below them at both
the 12-month and 6-month horizons. A month
before the announcement, mean forecast errors
provide little support for predictable differences
between average earnings and forecasts. Median
forecast errors at the 1-month horizon, though,
generally are positive and statistically significant,
indicating that the analysts’ median forecast is
less than earnings on average. These median
forecast errors are relatively small in magnitude,
though—on the order of hundredths of pennies
of earnings relative to the share price—when
average and median earnings are about 2 and 4
cents, respectively, relative to the share price.
Mean forecast errors and median forecast
errors differ substantially. The distribution of
forecast errors is asymmetric, with mean forecast
errors substantially larger in magnitude than
median forecast errors at the 6-month and 12month horizons. The distribution of earnings is
asymmetric. The distribution of earnings forecasts
also is asymmetric but not sufficiently asymmetric
that forecast errors are symmetric. There are substantial differences in mean and median forecast
errors across industries. We also find substantial
differences in mean and median forecast errors
by year, with the largest forecast errors in recession years.
CONCLUSION
Clarke, Jonathan and Subramanian, Ajay. “Dynamic
Forecasting Behavior by Analysts: Theory and
Evidence.” Journal of Financial Economics, April
2006, 80, pp. 81-113.
Our data for U.S. analysts’ forecasts of U.S.
firms’ earnings from 1990 through 2004 show
566
S E P T E M B E R / O C TO B E R , PA R T 2
2009
REFERENCES
Bera, Anil K. and Jarque, Carlos M. “Efficient Tests
for Normality, Homoscedasticity and Serial
Independence of Regression Residuals.” Economics
Letters, October 1980, 6(3), pp. 255-59.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
Ciciretti, Dwyer, Hasan
Gu, Zhaoang and Wu, Joanna S. “Earnings Skewness
and Analyst Forecast Bias.” Journal of Accounting
and Economics, April 2003, 35(1), 5-29.
Gupta, M.K. “An Asymptotically Nonparametric Test
of Symmetry.” Annals of Mathematical Statistics,
1967, 38(1), pp. 849-66.
Hong, Harrison and Kubik, Jeffrey D. “Analyzing the
Analysts: Career Concerns and Biased Earnings
Forecasts.” Journal of Finance, February 2003, 58,
pp. 313-51.
Keane, Michael P. and Runkle, David E. “Testing the
Rationality of Price Forecasts: New Evidence from
Panel Data.” American Economic Review,
September 1990, 80(4), pp. 714-35.
Keane, Michael P. and Runkle, David E. “Are
Financial Analysts’ Forecasts of Corporate Profits
Rational?” Journal of Political Economy, August
1998, 106(4), pp. 768-805.
Ljungqvist, Alexander; Marston, Felicia; Starks,
Laura T.; Weid, Kelsey D. and Yan, Hong. “Conflicts
of Interest in Sell-Side Research and the Moderating
Role of Institutional Investors.” Journal of Financial
Economics, August 2007, 85(2), pp. 420-56.
Ottaviani, Marco and Sørensen, Peter N. “The Strategy
of Professional Forecasting.” Journal of Financial
Economics, August 2006, 81(2), pp. 441-66.
Premaratne, Gamini and Bera, Anil. “A Test for
Symmetry with Leptokurtic Financial Data.”
Journal of Financial Econometrics, Spring 2005,
23(2), pp. 169-87.
Sirri, Erik. “Investment Banks, Scope, and Unavoidable
Conflicts of Interest.” Federal Reserve Bank of
Atlanta Economic Review, Fourth Quarter 2004,
89(4), pp. 23-35.
Smith, Randall; Scannell, Kara and Davies, Paul.
“A ‘Brazen’ Insider Scheme Revealed.” Wall Street
Journal, March 2, 2007, pp. C1, C2.
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
S E P T E M B E R / O C TO B E R , PA R T 2
2009
567
568
S E P T E M B E R / O C TO B E R , PA R T 2
2009
F E D E R A L R E S E R V E B A N K O F S T . LO U I S R E V I E W
@FedFRASER