View original document

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

r

FEDERALRESERVE
RESERVE BANK
BANK OF
OFDALLAS
DALLAS
FEDERAL
Quarter 1994
1994
FiFirst
rst Quarter

The Federal Funds Rate as an
all Indicator
ofMonetary
of
Monetary Policy: Evidencefrom
Evidence from the 1980s
Nathan S. Balke and
Kenneth M.
M. Emery

Demographics
Demographia and the Long-Term Outlook
for Housing
HOusing Investment
John K.
K. Hill and
John
O'Ann M. Petersen

f

I

APrimer
ofBusiness
A
Primer on the Nature of
Business Cycles
Gregory W. Huffman
Gregory

This publication was digitized and made available by the Federal Reserve Bank of Dallas' Historical Library (FedHistory@dal.frb.org)

Economic Review
Federal Reserve Bank of Dallas
Robert D. McTIe', Jr.
"....1ttIIa.J~0Ifiu;

Tony J .

S.",,,io

'dlra"--"(1wI~_ilIrl<or

HlntlY

Rosellblum

s.-~~..,o.-f/t_

W. Michie' Cox
va ~ lilt! ftooono: AI:M1a
Ger.ld P. O'Driscoll, Jr.
Vu~ MIl Ftmon. A<Wor

St.phen P. A. Brown
~va~_s...Et_

Robert W. Gilmer
~OtIar,

_/II""""

Economists

ZsoIt BeeSI
Robert T Clall
Jo/J1 V Ouca
Kennettl M Emery
Beverly J fox
David M Goold
William C Gruben
Joseph H Haslag
EYan F Koen'!l

O' Ann M Petersen
Keith R Phillips
Fiona 0 Sigalla
lofi l T3'/1of
l ucinda Vargas
Kelly ~ Whealan
Mark A. Wynne
Kevlfl J Yeats

Mille K Yoce'

ReMlreh Auociltn

Professor Nathan S Balke
Soo/hern Methodist /Jnll'efSJty
Professor Thomas B fumby
Southern Methodist UniverSIty
I'fofeuor Gregory W HuHman

Soothern Methodist Umversity
PrOlesSOf Roy J RuH;n
Unwersiry of HoosrOfl
Professor Ping Wang

Pennsylvania SI318 University
Editors
Ahooda Harris
Virginia M Rogers

Desill"
GeneAl,I\ry

Gllpflic'lnd Typog,.p"y
laufa J Bell
Tile fC{lfl()lfll€ ~ IS ~,shed by IN fld",1 Re5trYe Sant of Oiliu it. 'MWS
. , pressed Ir.tholl, oIlM lutbOl's end do not neceswuly f.llecllhe POSItIons of the Fedellil
ReslMISank 01 O.II.s Of the fede ••• Rese~ S,$lem
Sima'IKNlIIS ill! WI,1abIe flee 01 d'IIfOe Please send IeqUeSIS lor s.onglMqJr and mull'.
OO!I'I' lUbscI1ptOlS. bid'. _ . ItId Jddrm changes 10 lhe l'Iboc Alia-IS ~ feder,l
fIew;e Barlt. at 0.1111$. P 0 8oJ: 6S5EO>. 0.1115. tx J!I265.!o!Di. 11141 922{)2!il
Arbela IftIY be llI!J'onted on the tondollOll IlIaI the $(Ute 1$ uedned and !hi! Reseadl
Oepiortmln1 '$ p<oYIOeC! W1m a toP\' oj !lie po.obI.caloon eootl.n,i'9 the repoonted 111i118f1i1

On thl COlli: 1ft arclllt.cturllrlnd.llnt ollile new Federal Rese", hn' II OatlH

1M"",art'fI.

Contents
Page 1

The Federal Funds Rate
as an indicator of
Monetary Policy:
Evidence from the 19BOs
Nathan S. Balke and
Kenneth M. Emery

Recently, several {."Conomists have argued that movements

in the federal funds rate are a good proxy for c ha nges in
monetary policy. In this al1idc, Nathan Balke and Ke nne th

Emery critically examine this view a nd the evidence supporting it. Using simple veClOr auto regressions, they find that
before 1980 the correl:ttions between the feder-AI funds rate

and other important macroeconomic variables are consistent
with a traditional monetary policy interpretation o f the federal
funds rate . However, they show that after 1982 the relationships between the federa l funds ratc a nd o ther macroeconomic

variables change s ignificantly. Most important , the correlations
between the feder,1I fun ds rate and other macroeconomic
variables observt.'<i during the 1980s arc not as consistent with
a traditio nal mo netary policy view of the federal funds rate as
they were before 1980.
Balke and Emery's work highlights how relationships
between important macroeconomic variables can c hange
when institutions or policy regimes c hange. While the federal
funds rate may still be a good indicator of monetary policy,
its relationship with ot her important macroeconomic variables
is now clearly diffe rent from what it was before 1980.

Page 17

Demographics and the
Long-Term Outlook for
Housing investment
John K. Hill and
O'Ann M. Petersen

John Hill and O'Ann Petersen measure the importance
of projected shifts in the size and age d istributio n of the U.S.
population for domestic housing investment Their analysis
runs through the year 2010 and provides separate estimates
for Single-family and multifamily investment.
Hill and Petersen fin d that the contractionary effects of
the population slowdown are already being felt in the hOUSing
industry and probably have been since the latter part of the
19805. In Hill and Petersen's s imulatiOns, demograph ic shifts
lower net ho using investment by 17 percent from the late
1980s through the first half of the 199Os. Popu lation factors
then reduce net investment a n additional 22 percent through
the year 2005 before rurning favorable .

(Continued on the next page)

Contents
(Contillued from the Pfl?II;OUS page)

Demographics and the
Long-Term Outlook/or
HOusing Investment
John K. Hill and

O'Ann M. Petersen

Hill and Petersen discuss the impliC::llio ns of their findings for construction jobs and housing prices. They suggest
that the population slowdown need n OI produce an absolute
contraction in housing employment It will, however, reduce

housing's share of national emplo}'lllcl1\ by as much as onetbird. According to the authors, the changing demographics
do not provide a compelling reason for average home prices
to suffer a deep decline . They do s uggest. however. that significant rdative prkc adjustme nts may need to take place
between different types of homes.

Page 27

APrimer on the Nature
a/Business Cycles
Gregory W, Huffman

Discussions of the dfects of monct:l ry ;md rlSCal policy
sometimes center on the impact o f such policies in a meliorating nuctuatio ns associat(.'(\ with the business cycle. However,
though familiar wilh Ihe te ml "business cyde ,~ many people
are not aware of what it refers 10 exactly. In this article,
Gregory Huffman presents ;10 explanation o f the term and
provides a detailed illustration of post - World War II U.S .
business cycles. He also contrasts the behavior of various U.s.
economic time series over the b usiness cycle with simih r
Canadian statistics and points OUl some apparent anoma lies
in the data.

Nathan S. Balke

Kenneth M. Emery

Associate Professor of Economics
Southern Methodist University and
Visiting Consultant
Federal Reserve Bank of Dallas

Senior Economist
Federal Reserve Bank of Dallas

The Federal Funds Rate as an Indicator
of Monetary Policy: Evidence from the 1980s

H

ow monetary policy affects economic activity
is a perennial question in macroeconomics.
One of the main impediments to answering this
question is the absence of agreement on what is
an accurate gauge of monetary policy. Historically, many economists have used changes in the
quantity of money as an indicator of changes in
monetary policy. One problem with this approach,
however, is that changes in money can result
from factors other than changes in monetary
policy. For instance, economic conditions can
significantly influence money growth over the
course of the business cycle.
Several economists have argued that movements in short-term interest rates, particularly
movements in the federal funds rate, may be a
better indicator of changes in monetary policy than
are changes in the quantity of money (McCallum
1983, Laurent 1988, Bernanke and Blinder 1992,
and Goodfriend 1992). This view is based on the
observation that, with the exception of the 1979–
82 period, the Federal Reserve appears to have
implemented its monetary policy by targeting the
federal funds rate.
In support of the federal funds rate as a
gauge of monetary policy, Bernanke and Blinder
(1992) present evidence that the federal funds rate
is a better predictor of future economic activity
than are other interest rates or other monetary
aggregates. Additionally, using data before 1979,
they show, using a simple vector autoregression
(VAR), that changes in the federal funds rate are
systematically related to changes in inflation and
unemployment. (See the box entitled “Vector
Autoregressions.”) Specifically, the federal funds
rate rises in response to unexpected increases in
Economic Review — First Quarter 1994

inflation and falls in response to unexpected increases in unemployment. Thus, their results are
consistent with a monetary policy that “leans
against the wind,” or reacts countercyclically to
the business cycle.
In this article, we use simple vector autoregressions to examine whether the relationships
found by Bernanke and Blinder for the pre-1979
period persist after 1982, when the Federal Reserve
returned to a policy of explicitly targeting the
federal funds rate. The results provide more recent
evidence on whether sensible monetary policy
reaction functions can be derived using the federal
funds rate as an indicator of the stance of monetary policy, and whether the funds rate has information about future inflation and unemployment.
Also, because the vector autoregression methodology uncovers correlations among macroeconomic
variables, the results of this examination may
reveal changes in the correlations of macroeconomic variables after 1982 that shed light on
several monetary policy issues.1 For instance, if
monetary policy is now less countercyclical, does
the federal funds rate now respond less to innovations in unemployment or inflation? Do changes in

The authors thank John Duca, Joseph Haslag, and Evan
Koenig for helpful comments and suggestions. We also
thank Adrienne Slack for her capable research assistance.
Any remaining errors are the authors’ responsibility.
1

According to the Lucas Critique, changes in policy regimes
or economic institutions will likely change the relationships
among macroeconomic variables.

1

the federal funds rate now have less information
content for economic activity? If so, does this
imply that monetary policy became less effective
during the 1980s? While answering these types of
questions using the vector autoregression methodology is difficult, VARs do provide correlations
with which economic models must contend.
In this article, we first review the recent
literature on measuring monetary policy. We then
present the empirical results, and finally, we outline some interpretations of these findings.
Vector autoregressions and monetary policy
Traditional monetarists (for example, Friedman and Schwartz 1963) view the growth of the
money stock as a good indicator of monetary
policy. While traditional monetarists argue that the
money supply has important effects on the real
economy in the short run, they typically stress that
policymakers should avoid the temptation to temporarily stimulate real economic activity by rapidly
expanding the money supply. Traditional monetarists fear such actions would increase inflation in
the long run and exacerbate the business cycle.
With the advent of rational expectations in macroeconomics, however, most economists view only
unexpected policy actions as having real effects on
the economy. Thus, rational expectations monetarists (for example, Sargent and Wallace 1975)
take the view that only unexpected changes in the
money supply will have temporary real effects,
while expected changes in the money supply will
be immediately reflected in the price level.
Sims (1980) questions the importance of
unexpected changes in money for future changes
in economic activity. Using a four-variable VAR,
he shows that once the information content of
interest rates is taken into account, only a small
portion of the unexpected variation in output can
be attributed to unexpected changes in the money
supply. While some researchers have questioned
the robustness of Sims’ results, the conclusion that

2

2

When post-1979 data are added to their sample, the statistical significance of the federal funds rate in prediction
equations generally declines. Their results are consistent
with those of Bernanke and Blinder (1990).

interest rates have substantial information content
about future economic activity has held up. For a
time, these results were considered damaging to
the view that monetary policy is an important
factor in explaining business cycles.
However, later research (McCallum 1983 and
Laurent 1988) maintains that Sims’ results do not
imply that monetary policy is unimportant in determining economic activity. These economists argue
that because the Federal Reserve conducts policy
by targeting the federal funds rate and because
changes in the money supply can be caused by
factors other than changes in monetary policy,
unexpected changes in the federal funds rate may
be a better measure of monetary policy than unexpected movements in the money supply.
If the federal funds rate is a good indicator
of monetary policy and monetary policy has real
effects on the economy, then the federal funds
rate should be a good predictor of economic
variables. Bernanke and Blinder (1992) show that
the federal funds rate is a good predictor of major
macroeconomic variables before 1979 and that the
federal funds rate better predicts macroeconomic
variables than other interest rates or monetary
aggregates.2 Bernanke and Blinder also discover,
using a simple VAR, that the federal funds rate
responds to variables the Federal Reserve has been
traditionally concerned with—unemployment and
inflation. In other words, “reaction functions” can
be estimated in which monetary policy (changes
in the federal funds rate) reacts countercyclically
in response to unexpected movements in unemployment and inflation.
In the next section of this article, we examine
whether the relationships found by Bernanke and
Blinder for the pre-1980 period held up during
the 1980s.
Empirical results
Pre-1980 results. Bernanke and Blinder (1992)
specify a series of three-variable VARs consisting
of a measure of monetary policy, the prime-age
(25–54) male unemployment rate, and the inflation rate as measured by the consumer price index.
Each variable is regressed on six lags of itself
and six lags of the other two remaining variables.
The data are monthly, starting with July 1959 and
ending in September 1979, when interest rate
Federal Reserve Bank of Dallas

Vector Autoregressions
Vector autoregressions (VARs) are
time series models that use only past values
of the variables of interest to make forecasts.
For instance, a three-variable VAR system of
interest rates, unemployment, and inflation
can be expressed as

Rt = β 1 + Σ Rt –i + Σ Ut –i + Σ πt –i + ⑀ Rt
Ut = β 2 + Σ Rt –i + Σ Ut –i + Σ πt –i + ⑀ Ut
πt = β 3 + Σ Rt –i + Σ Ut –i + Σ πt –i + ⑀ π t ,
where R, U, and π are the interest rate,
unemployment rate, and inflation rate, respectively. β is an intercept term, t is a time
subscript, and ⑀ is an error term. Thus, each
of the three variables is expressed as a linear
function of past values of itself and past values of other variables in the system.
In practice, the estimated error terms
from each equation are correlated so that it is
not correct to assume that, for instance, ⑀Ut
represents an independent surprise movement in the unemployment rate. To better
interpret the dynamic relationships present in
the data, the residuals from the VAR are
broken up into linear combinations of independent (orthogonal) shocks. A common orthogonalization is to assume that the VAR
system is recursive so that there is a chain of
causality among surprises in the variables
during any given period. For example, a pos-

targeting was de-emphasized. Bernanke and
Blinder impose recursivity on the system using
the Choleski decomposition, with the ordering
from the policy variable to the unemployment
rate to the inflation rate.
Bernanke and Blinder use both the federal
funds rate and the spread between the federal funds
rate and the ten-year U.S. Treasury bond rate,
henceforth the spread, as indicators of monetary
policy. The spread is an alternative indicator of
monetary policy because it controls for the general
Economic Review — First Quarter 1994

sible recursive system of the VAR above is
one in which the interest rate responds to an
exogenous shock, and unemployment responds to the contemporaneous interest rate
and an exogenous unemployment shock,
while the inflation rate responds to the contemporaneous interest rate, contemporaneous unemployment rate, and an exogenous
inflation shock. In effect, new surprises, or a
shock term for each variable, are created that
are now uncorrelated with each other. The
transformation of the original shocks into recursive, orthogonal shocks is called the
Choleski decomposition.
The Choleski decomposition is controversial because if the VAR is used to draw
economic inferences, then the recursive restriction imposed on the system should be
supported by economic theory. If the identifying assumption of recursivity is not justified,
then the estimated parameters will be a mixture of both structural and reduced-form parameters. However, for forecasting purposes,
the use of Choleski decompositions in VARs
does not pose a problem because no economic inferences are being drawn from the
estimated parameters.1
1

For more on VARs, see Todd (1990), Runkle (1987), Sims
(1986), Cooley and LeRoy (1985), and Hakkio and Morris
(1984).

level of market interest rates and therefore provides
further information about whether a particular
level of the federal funds rate represents a restrictive or loose monetary policy. Figure 1 plots both
the federal funds rate and the spread.3 Note that
the spread is nearly a mirror image of the federal

3

A similar figure appears in Bernanke and Blinder (1992).

3

funds rate. In general, as Bernanke and Blinder
point out, run-ups in the federal funds rate have
preceded the onset of all recessions since 1959.
Because the estimation results for the individual equations within a VAR system are of little
interest, they are not reported here. Table 1, however, reports the marginal significance level of
exclusion tests for lags of the right-hand side variables. As in Bernanke and Blinder, the hypothesis
that lags of inflation or unemployment can be
excluded from the federal funds rate equation is
easily rejected. This result indicates that the recent
state of the economy, as measured by lagged inflation and unemployment, contains information about
future movements in the federal funds rate. In addition, lags of the federal funds rates are significant
in both the inflation and unemployment equations,
suggesting, at a minimum, that knowledge of the
federal funds rate helps predict these variables.
The results for the spread between the ten-year
Treasury bond rate and the federal funds rate are
qualitatively similar to those of the federal fund
rate; for this reason, we do not present them here.
When M2 growth is added to the threevariable system, the federal funds rate still retains
its importance (as measured by the significance
levels) in explaining the behavior of inflation; the
rate is somewhat less important for unemployment. Furthermore, we can reject the hypothesis
that the federal funds rate can be dropped from
the four-variable system consisting of federal funds
rate, M2 growth, unemployment, and inflation.

4

4

We employ a Choleski decomposition with the ordering of
federal funds rate, M2, unemployment, and inflation. The
qualitative results are unchanged if M2 and the federal
funds rate or unemployment and inflation are switched in
the ordering.

5

The 90-percent confidence bands for the variance decompositions and impulse response functions (reported below)
are generated using a Monte Carlo procedure and are
available from the authors on request.

6

Using the spread variable, the forecast error variance
decompositions are similar. However, shocks to the spread
contribute less to the total inflation forecast error variance,
and shocks to inflation contribute less to the total spread
forecast error variance.

Figure 1

Federal Funds Rate and Spread
Percent
20

PT

P T

P

T

P T P T

PT

15

10
Federal
funds rate

5
Spread

0

–5

–10
’59

’62

’65

’68

’71

’74

’77

’80

’83

’86

’89

’92

SOURCE: Board of Governors, Federal Reserve System.

Table 2 presents the forecast error variance
decompositions for the VAR, including the federal
funds rate and M2.4 A variance decomposition
divides the total forecast error variance, at different
forecast horizons, into portions attributable to
shocks in each of the variables in the system.5
From Table 2, we find a substantial proportion of
the forecast error variance in the federal funds rate
is caused by uncertainty about the future values
of unemployment and inflation. In other words,
knowledge about future states of the economy
tells us something about future movements in the
federal funds rate. The decompositions for inflation and unemployment indicate that uncertainty
about future values of the federal funds rate contributes only modestly to uncertainty about their
future values. The contribution of federal funds
shocks to the forecast variance of unemployment
tends to be greater at longer horizons than that of
M2. On the other hand, M2 tends to contribute
more to the forecast variance of inflation than does
the federal funds rate. Still, the majority of the
forecast error variance for inflation and unemployment arises from uncertainty about shocks to those
variables themselves.6
Overall, the federal funds rate had important
predictive content for unemployment and inflation
during the period July 1959–September 1979. This
Federal Reserve Bank of Dallas

Table 1

Marginal Significance Levels for Exclusion of Lags, 1959–79
Three-variable system (federal funds rate, unemployment, inflation)
Marginal Significance Levels
Federal funds

Lags of
Unemployment

Inflation

.0000
.0092
.0000

.0000
.0000
.0698

.0002
.2300
.1304

Equation
Federal funds
Unemployment
Inflation

Four-variable system (federal funds rate, M2, unemployment, inflation)
Marginal Significance Levels

Equation
Federal funds
M2
Unemployment
Inflation

Federal funds

M2

.0000
.0000
.1140
.0000

.0357
.0000
.1032
.0063

Lags of
Unemployment

Inflation

.0003
.0897
.0000
.2523

.0035
.4875
.1032
.2376

Tests for dropping variable from four-variable system
M2
Federal funds

χ2 (18) = 40.88
χ2 (18) = 85.50

(.0016)
(.0000)

is the case even after allowing for the effect of
money aggregates, as measured by M2. This finding
provides indirect evidence that the federal funds
rate may be a good indicator of monetary policy.
An additional source of evidence provided
by Bernanke and Blinder is that the response of
the federal funds rate to shocks in unemployment
and inflation is consistent with a “lean against the
wind” policy; that is, the federal funds rate rises in
response to a positive inflation shock and falls in
response to a positive unemployment shock.
Figures 2 and 3 plot impulse response functions for the three-variable VAR that includes the
federal funds rate, unemployment, and inflation
and is estimated over the sample period up to
September 1979.7 Figure 2 displays the response
of the federal funds rate over time to unexpected
movements in inflation and unemployment.8 As
underscored by Bernanke and Blinder, the plots
look very much like “lean against the wind” moneEconomic Review — First Quarter 1994

tary policy reaction functions. The federal funds
rate rises in response to an unexpected increase
in inflation and falls in response to an unexpected
increase in unemployment.
Figure 3 displays the responses of unemployment and inflation to an innovation in the
federal funds rate. The qualitative pattern of the
impulse response function for unemployment is

7

The qualitative behavior of the impulse responses using the
spread variable as the monetary policy variable is very
similar.

8

The ordering of the unemployment rate and the inflation rate
in the VAR does not affect the results that follow. Also, the
inclusion of money growth, as measured by M2, in the VAR
system does not qualitatively affect the impulse responses
of the other three variables.

5

interpreting federal funds rate innovations as
unexpected monetary policy changes.9 This price
level effect, does not, however, necessarily provide evidence against the effectiveness of a “lean
against the wind” policy. The reason is that such a
countercyclical policy, to the extent that it is predictable, would be embodied in the coefficients of
the VAR and not necessarily be reflected in unexpected movements of the federal funds rate.10

Figure 2

Responses of Federal Funds Rate to Inflation
and Unemployment Shocks
Percent
.5
.4
Inflation shock

.3

Post-1982 results

.2
.1
0
–.1
–.2
Unemployment shock

–.3
1

6

11

16

21

26

31

36

Months

broadly consistent with the view that unexpected
changes in the federal funds rate represent changes
in monetary policy. Figure 3 shows that after a
temporary and short-lived fall, the unemployment
rate rises in response to an unexpected increase
in the federal funds rate. On the other hand, the
inflation rate response moves the wrong way if
one interprets a surprise increase in the federal
funds rate as a tightening of monetary policy. This
“price level” effect, noted by Eichenbaum (1992) in
his comment on Sims (1992), raises doubts about

In this section, we examine whether the
relationships found by Bernanke and Blinder
persist during the 1980s. There are several reasons
they may not. First, financial innovation and
deregulation during the 1980s may have changed
the effectiveness and the transmission mechanism
of monetary policy (Bosworth 1989 and Kahn
1989). Second, the high-inflation decade of the
1970s may have changed the way the public reacts
to inflation. In particular, the Phillips curve may
have steepened, lessening even the short-term
trade-off between inflation and unemployment.
Furthermore, financial markets may have become
more sensitive to inflation fears, and these concerns are more readily reflected in increases in
long-term interest rates. Finally, the Federal Reserve
may have focused more of its policy on control-

Figure 3

Responses to Federal Funds Rate Shocks
Percent
.7
9

10

6

Similarly, Gordon and Leeper (1993) cite this effect in
arguing that innovations in the federal funds rate are an
inappropriate indicator of monetary policy. As an alternative, they construct a structural model of the money market
to identify monetary policy surprises.
Another explanation for this price level effect would be to
distinguish between nominal and real federal funds rate
innovations. Many analysts would argue that a monetary
policy tightening occurs only when the real federal funds
rate rises. The price level effect found here may only signal
that increases in the nominal federal funds rate are not as
large as the contemporaneous increases in inflation and,
therefore, do not constitute a tightening of monetary policy.
An examination of the viability of the real federal funds rate
as an indicator of monetary policy is left for future research.

.6
.5
.4
Inflation rate

.3
.2
.1

Unemployment rate

0
–.1
1

6

11

16

21

26

31

36

Months

Federal Reserve Bank of Dallas

Table 2

Forecast Error Variance Decompositions, 1959–79
Federal funds rate
Percentage of forecast error variance explained by

Forecast
horizon

Federal funds

M2

Unemployment

Inflation

6
12
24
36

77.7
51.7
30.5
23.9

2.9
9.3
31.9
41.6

8.9
18.7
22.2
21.8

10.5
20.3
15.5
12.7

M2
Percentage of forecast error variance explained by

Forecast
horizon

Federal funds

M2

Unemployment

6
12
24
36

22.5
21.1
23.1
24.3

70.3
64.0
60.0
58.0

3.3
8.1
9.8
9.4

Inflation
4.0
6.8
7.1
8.3

Unemployment rate
Percentage of forecast error variance explained by

Forecast
horizon

Federal funds

M2

Unemployment

6
12
24
36

1.1
1.7
13.2
23.8

2.8
6.0
10.2
8.7

96.2
90.4
68.1
57.5

Inflation
.3
.6
3.6
8.0

Inflation rate
Percentage of forecast error variance explained by

Forecast
horizon

Federal funds

M2

Unemployment

Inflation

6
12
24
36

12.9
17.2
15.5
14.2

7.7
11.8
24.5
33.6

2.9
4.7
6.7
8.0

76.5
66.2
52.3
44.2

ling inflation before inflation accelerates. This shift
may be the result of a heightened aversion to
inflation by the public or increased awareness by
the Federal Reserve of the limitations of monetary
policy in affecting the real economy.
Is the VAR evidence still consistent with the
view that the federal funds rate is a good indicator
of monetary policy? The first step in answering
this question is to test whether there is any evidence of structural instability in the VAR system
Economic Review — First Quarter 1994

after 1982.11 VARs are estimated over the November
1982–September 1992 period and are compared
with VARs estimated over the July 1959–September
1979 period. Using standard likelihood-ratio tests,

11

We exclude the October 1979–September 1982 period,
when the Federal Reserve de-emphasized the targeting of
interest rates in setting monetary policy.

7

Table 3

Test for Structural Change Between the 1959–79
and 1982–92 Samples
Three-variable system (federal funds rate, unemployment, inflation)
Entire system

χ2 (57) = 137.0

(.0000)

Significance levels for structural change equation by equation
Federal funds
Unemployment
Inflation

.0352
.0081
.0000

Four-variable system (federal funds, M2, unemployment, inflation)
Entire system

χ2 (100) = 219.5

(.0000)

Significance levels for structural change equation by equation
Federal funds
M2
Unemployment
Inflation

.0043
.0000
.0442
.0000

the hypothesis of stability is easily rejected for the
VAR systems using both the federal funds rate
and the spread and including and excluding M2
(Table 3).12 Taken equation by equation, there is
also evidence that the correlation structure from
the 1959–79 period differs from that of the 1982–92
period.
Does the federal funds rate retain its predictive ability in the post-1982 period? Table 4 displays
the marginal significance level of exclusion tests
for the post-1982 period. The federal funds rate
still has predictive content for inflation and unemployment, with significance levels close to or less

12

8

Evidence of a unit root in the level of the federal funds rate
and the unemployment rate made us cautious about using
the data in levels form to make inferences using likelihoodratio tests. However, instability was also found when the
VAR system was estimated in first differences. There was
also evidence of instability when the system was estimated
in error-correction form using a cointegrating vector found
by the methodology developed by Johansen and Juselius
(1990).

than 10 percent; therefore, we can strongly reject
the hypothesis that the federal funds rate can be
dropped from the system. M2, however, loses
much of its predictive power in the later sample;
therefore, we cannot reject the hypothesis that M2
can be dropped from the system. Furthermore, the
significance levels of lags of unemployment and
inflation in the federal funds equation indicate that
these variables do not explain much of the movement in the federal funds rate in the later period.
Examining the forecast error variance decompositions for the VAR from the pre-1980 and post1982 periods (Table 5 ), the federal funds rate still
explains a modest percentage of the forecast
variance of unemployment and inflation. The contribution of M2, on the other hand, falls substantially in the post-1982 period. Thus, it appears that
the federal funds rate has become more important
relative to M2 in explaining the behavior of inflation and unemployment in the post-1982 period.
Perhaps the most striking difference in the
variance decompositions is the small percentage
of the forecast error variance for the federal funds
rate that can be attributed to uncertainty about
future inflation. In other words, inflation shocks
Federal Reserve Bank of Dallas

Table 4

Marginal Significance Levels for Exclusion of Lags, 1982–92
Three-variable system (federal funds rate, unemployment, inflation)
Marginal Significance Levels
Federal funds

Lags of
Unemployment

Inflation

.0000
.0923
.0133

.1700
.0000
.5909

.8184
.6415
.0003

Equation
Federal funds
Unemployment
Inflation

Four-variable system (federal funds rate, M2, unemployment, inflation)
Marginal Significance Levels

Equation
Federal funds
M2
Unemployment
Inflation

Federal funds

M2

.0000
.0000
.1163
.0362

.1060
.0033
.6812
.8822

Lags of
Unemployment

Inflation

.3003
.7979
.0000
.6827

.3911
.0322
.6765
.0006

Tests for dropping variable from four-variable system
M2
Federal funds

χ2 (18) = 18.66
χ2 (18) = 65.04

(.4129)
(.0000)

account for little of the variability in the federal
funds rate in the later period. This finding raises
questions about the traditional monetary policy
interpretation of the federal funds rate, since that
interest rate does not appear to respond to inflation shocks in the post-1982 period.13 Additionally,
the variance decompositions indicate that a larger
percentage of the forecast variance in the federal
funds rate can be attributed to uncertainty about
innovations to the federal funds rate itself. It seems
the federal funds rate has become more independent of the other variables in the VAR during
the 1980s.
Figures 4 and 5 present impulse responses
for the post-1982 period for the three-variable
VAR.14 Figure 4 displays the reaction of the federal
funds rate to unexpected increases in inflation and
unemployment. The response of the federal funds
rate to an unexpected increase in the unemployment rate in the post-1982 period is less than that
Economic Review — First Quarter 1994

in the previous period, although it does indicate a
loosening of monetary policy in response to a surprise increase in unemployment. Additionally, the
90-percent confidence bands of the two responses
overlap and, hence, are not statistically different
from each other. On the other hand, the response
of the federal funds rate to an unexpected increase

13

Again, the variance decompositions using the spread are
similar except that uncertainty about future inflation contributes little to the forecast error variance of the spread (and
vice versa) in both the earlier and later periods.

14

The differences in the impulse responses between the two
periods result mainly from differences in the estimated
coefficients of the VAR equations and not from differences
in the sizes of the standard deviation shocks. The size of the
shocks for each variable are very close across the two
periods.

9

Figure 4

Responses of Federal Funds Rate to Inflation
and Unemployment Shocks
Percent
.5
.4
Inflation shock, 1959–79

.3
.2
.1
Inflation shock, 1982–92

0
Unemployment shock, 1982–92

–.1
–.2

Unemployment shock, 1959–79

–.3
1

6

11

16

21

26

31

36

federal funds rate is negative in the post-1982
period rather than positive as in the 1959–79
period (the 90-percent confidence bands of two
responses do not overlap after thirteen months).
The response of inflation in the post-1982 period
to a federal funds rate shock, as in the pre-1980
period, moves in the wrong direction but is much
less persistent in the post-1982 period than in the
earlier period.
Thus, while the federal funds rate has become
more important relative to M2 in explaining inflation and unemployment, it appears that the Federal
Reserve no longer “leans against the wind” with
respect to inflation shocks in the post-1982 period.
As a final note, the above results are robust to
using the core-CPI inflation rate rather than just the
CPI inflation rate and to trying to control for the
effect of the decline in 1986 oil prices by introducing a dummy variable for this period into the VAR.

Months

Possible interpretations of the results
in inflation does not correspond to a “lean against
the wind” reaction function. The federal funds
rate for the most part fails to respond at all to an
inflation shock, and the confidence bands of the
two responses do not overlap over a horizon of
three to twenty months.15 A “lean against the wind”
policy would suggest an increase in the federal
funds rate in response to a positive inflation
shock. Therefore, whereas in the earlier sample
a monetary policy reaction interpretation could
be applied, in the 1982–92 period the response
of the policy variables do not look like typical
reaction functions.16
Finally, the response of unemployment and
inflation to shocks in the federal funds rate casts
further doubt on the interpretation of federal
funds rate innovations as monetary policy changes
in the post-1982 period (Figure 5 ). The response
of unemployment to unexpected increases in the

As noted in the box on vector autoregressions, making economic inferences from estimated
VARs is controversial. The fundamental difficulty
is that the estimated relationships are derived
from reduced-form equations. Thus, VARs provide
evidence on correlations in the data, but these
correlations may be consistent with a number of

Figure 5

Responses to Federal Funds Rate Shocks
Percent
.8
.7
.6
.5
.4

Inflation rate, 1959–79
.3
.2

Inflation rate, 1982–92

Unemployment rate, 1959–79

.1
15

The spread between the ten-year Treasury bond yield and
the federal funds rate actually increases in response to a
positive inflation shock.

0
–.1

Unemployment rate, 1982–92

–.2
16

10

Supporting this point, lagged inflation is no longer statistically significant in the federal funds rate equation.

1

6

11

16

21

26

31

36

Months

Federal Reserve Bank of Dallas

Table 5

Forecast Error Variance Decompositions,
November 1982–September 1992
Federal funds rate
Percentage of forecast error variance explained by

Forecast
horizon

Federal funds

M2

Unemployment

6
12
24
36

89.4
75.7
72.5
73.3

5.7
10.6
11.6
12.7

4.4
12.4
14.3
12.7

Forecast
horizon

Federal funds

M2

Unemployment

Inflation

6
12
24
36

18.6
24.4
28.7
32.2

63.1
56.7
52.8
50.2

.3
1.4
2.2
2.5

18.0
17.5
16.3
15.1

Inflation
.6
1.2
1.5
1.3

M2
Percentage of forecast error variance explained by

Unemployment rate
Percentage of forecast error variance explained by

Forecast
horizon

Federal funds

M2

Unemployment

6
12
24
36

22.6
21.3
23.4
26.6

.6
1.3
1.8
2.6

76.5
75.3
70.3
66.2

Inflation
0.3
2.1
4.5
4.5

Inflation rate
Percentage of forecast error variance explained by

Forecast
horizon

Federal funds

M2

Unemployment

Inflation

6
12
24
36

11.0
11.5
12.3
12.6

3.6
4.2
4.3
4.3

1.7
3.2
3.7
3.7

83.7
81.0
79.8
79.3

economic theories.17 In this section, we discuss
several plausible interpretations of the changed
relationships found from the VARs above. These
interpretations rely on developments in monetary
policy issues that arose during the 1980s.
That the Federal Reserve no longer “leans
against the wind” with respect to inflation in the
post-1982 period is somewhat surprising given the
supposedly higher premium the Federal Reserve
put on price stability during this period. Economic
Economic Review — First Quarter 1994

models of monetary policy indicate that a monetary authority’s increased desire to fight inflation
results in policy’s being tightened more severely
in the case of an unexpected increase in the inflation rate (Ball 1990 and Alogoskoufis and Smith

17

In other words, the economic theories have “observationally
equivalent” reduced forms.

11

1991). If the federal funds rate is a good indicator
of monetary policy, these results indicate just the
opposite: the policy response to an increase in the
inflation rate became muted during the 1980s.
One plausible explanation of the results that
reconciles the muted response of the federal funds
rate with an increase in the Federal Reserve’s
resolve to fight inflation is evidence that the
behavior of inflation changed during the 1980s.
Inflation since the early 1980s exhibits substantially less persistence than in the previous period,
so that increases in inflation one month are temporary and, in fact, tend to be followed by a
decrease in the next month (Emery, forthcoming).
This change in the persistence of inflation is
reflected in the difference between the response
of inflation to federal funds shocks in the two
periods noted in Figure 5. The behavior of inflation during the 1980s implies that lagged values of
inflation provide little information about inflation
in the future. As a result, unexpected movements
in inflation, as measured by the VAR, no longer
require a monetary policy response.
An additional explanation is that the Federal
Reserve has abandoned a “lean against the wind”
policy and instead conducts monetary policy in a
more forward-looking manner. That is, the Federal
Reserve increases the federal funds rate in anticipation of inflation so as to not get behind the
curve with respect to fighting inflation. Since lowdimension VARs, such as the ones examined in
this article, probably do not reflect all the information available to the monetary authorities at the
time policy decisions are made, the estimated
impulse response function for the federal funds
rate may not adequately capture the forwardlooking nature of monetary policy.
Sims (1992) uses a similar argument in
explaining the perverse inflation response to a
federal funds rate innovation. Because the VAR
does not reflect all the information available to the
monetary authority when it conducts policy, the
federal funds innovations in these VARs still reflect
systematic policy responses. Therefore, the positive response of inflation to a positive federal
funds innovation is really capturing the increase in
the federal funds rate in anticipation (correctly) of
inflation. The resultant tightening is not sufficient
to completely eliminate the subsequent inflationary pressures. If, indeed, this price puzzle does
12

Figure 6

U.S. Inflation
(Six-month moving average)
Percent
16
14
12
10
8
6
4
2
0
–2
’59

’62

’65

’68

’71

’74

’77

’80

’83

’86

’89

’92

SOURCE OF PRIMARY DATA: Bureau of Labor Statistics.

reflect a forward-looking Federal Reserve, then
because this price puzzle is present in both periods
(even more so in the earlier period, as indicated
in Figure 5), it appears that the Federal Reserve is
not substantially more forward-looking in the
1982–92 period than in the 1959–79 period.
Of course, it is very possible that the two
explanations offered here are related. A more
forward-looking Federal Reserve could conceivably
be better able to diminish the persistence of inflation by effectively offsetting fluctuations in the
underlying trend inflation rate, so that the price
level effect is diminished. Many analysts have
maintained that the Federal Reserve stabilized the
inflation rate around 4 percent during most of the
1980s (Goodfriend 1992) and that as a result of
this policy, deviations of inflation away from 4 percent were temporary. By contrast, movements in
inflation before 1980 tended to be more indicative
of rising or ebbing inflationary pressures because
the Federal Reserve did not respond quickly
enough to changing price pressures (Figure 6 ).
Conclusions
The vector autoregression evidence on the
federal funds rate as an indicator of monetary
Federal Reserve Bank of Dallas

policy weakens when the period since 1982 is
examined. Specifically, in contrast to the pre-1980
period, the federal funds rate no long displays a
“lean against the wind” response to inflation—that
is, it does not increase in response to unexpected
increases in inflation. However, this change does
not necessarily imply that the federal funds rate is
no longer an indicator of monetary policy. The
vector autoregression results indicate that after
1982, the federal funds rate responds to unexpected changes in the unemployment rate in a
manner similar to that before 1980 and consistent

with a traditional monetary policy interpretation.
Furthermore, there exist several possible explanations that are consistent with a monetary policy
interpretation of the federal funds rate.18
Nonetheless, the main message of this article
is to highlight how correlations between important macroeconomic variables can change when
institutions or policy regimes change. While the
federal funds rate may still be a good indicator of
monetary policy, its relationship with unemployment and inflation is now clearly different from
what it was before 1980.

18

Economic Review — First Quarter 1994

See Goodfriend (1992) for a narrative approach that supports the use of the federal funds rate, in addition to other
long-term rates, as a good indicator of monetary policy
during the 1980s.

13

References
Alogoskoufis, George S., and Ron Smith (1991),
“The Phillips Curve, the Persistence of Inflation, and the Lucas Critique: Evidence from
Exchange Rate Regimes,” American Economic
Review 81 (December): 1254 –75.
Ball, Laurence (1990), “Time-Consistent Policy and
Persistent Changes in Inflation,” NBER Working Paper Series, no. 3529 (Cambridge, Mass.:
National Bureau of Economic Research,
December).
Bernanke, Ben S., and Alan S. Blinder (1992),
“The Federal Funds Rate and the Channels of
Monetary Transmission,” American Economic
Review 82 (September): 901–21.
——— (1990), “On the Predictive Power of Interest
Rates and Interest Rate Spreads,” Federal
Reserve Bank of Boston New England Economic Review, November/December, 51– 68.
Bosworth, Barry (1989), “Institutional Change and
the Efficacy of Monetary Policy,” Brookings
Papers on Economic Activity, no. 1, 77–110.
Cooley, Thomas F., and Stephen F. LeRoy (1985),
“Atheoretical Macroeconometrics: A Critique,”
Journal of Monetary Economics 16 (November): 283–308.
Eichenbaum, Martin (1992), “Comments on ‘Interpreting the Macroeconomic Time Series Facts:
The Effects of Monetary Policy’ by Christopher Sims,” European Economic Review 36
( June): 1001–11.
Emery, Kenneth M. (forthcoming), “Inflation Persistence and Fisher Effects: Evidence of a
Regime Change,” Journal of Economics and
Business.
Friedman, Milton, and Anna J. Schwartz (1963),
A Monetary History of the United States, 1867 –
1960 (Princeton, N.J.: Princeton University
Press).

14

Goodfriend, Marvin (1992), “Interest Rate Policy
and the Inflation Scare Problem: 1979–1992,”
Federal Reserve Bank of Richmond 1992
Annual Report, 7–19.
Gordon, David B., and Eric M. Leeper (1993),
“The Dynamic Impacts of Monetary Policy:
An Exercise in Tentative Identification,” Federal Reserve Bank of Atlanta Working Paper
93–5, April.
Hakkio, Craig S., and Charles S. Morris (1984),
“Vector Autoregressions: A User’s Guide,”
Federal Reserve Bank of Kansas City Research
Working Paper 84 –10, November.
Johansen, Soren, and Katarina Juselius (1990),
“Maximum Likelihood Estimation and Inference on Cointegration—With Applications to
the Demand for Money,” Oxford Bulletin of
Economics and Statistics 52 (May): 169–210.
Kahn, George A. (1989), “The Changing Interest
Sensitivity of the U.S. Economy,” Federal
Reserve Bank of Kansas City Economic Review
74 (November): 13–34.
Laurent, Robert D. (1988), “An Interest Rate-Based
Indicator of Monetary Policy,” Federal Reserve
Bank of Chicago Economic Perspectives,
January/February, 3–14.
McCallum, Bennett T. (1983), “A Reconsideration
of Sims’ Evidence Concerning Monetarism,”
Economics Letters 13 (2–3): 167–71.
Runkle, David E. (1987), “Vector Autoregressions
and Reality,” Journal of Business and Economic Statistics 5 (October): 437–42.
Sargent, Thomas J., and Neil Wallace, (1975),
“ ‘Rational’ Expectations, the Optimal Monetary Instrument, and the Money Supply Rule,”
Journal of Political Economy 83 (April): 241–54.

Federal Reserve Bank of Dallas

Sims, Christopher A. (1992), “Interpreting the
Macroeconomic Time Series Facts: The Effects
of Monetary Policy,” European Economic
Review 36 ( June): 975–1000.
——— (1986), “Are Forecasting Models Usable for
Policy Analysis?” Federal Reserve Bank of
Minneapolis Quarterly Review 10 (Winter):
2–15.

ered,” American Economic Review 70 (May):
250–57.
Todd, Richard M. (1990), “Vector Autoregression
Evidence on Monetarism: Another Look at
the Robustness Debate,” Federal Reserve
Bank of Minneapolis Quarterly Review 14
(Spring): 19–37.

——— (1980), “Comparison of Interwar and Postwar Business Cycles: Monetarism Reconsid-

Economic Review — First Quarter 1994

15

16

Federal Reserve Bank of Dallas

John K. Hill

D’Ann M. Petersen

Arizona Public Service Company

Assistant Economist
Federal Reserve Bank of Dallas

Demographics and the Long-Term Outlook
for Housing Investment

H

ousing construction has been a driving force
behind U.S. economic recoveries. On average,
residential construction has accounted for twothirds of the increase in final sales during cyclical
upturns (Perry and Schultze 1993). In the most
recent recovery, however, homebuilding activity
has been modest, accounting for less than a third of
the increase in final sales. With housing affordability at a twenty-year high, those in the industry
are puzzled and concerned by the lackluster
growth of home construction.
Why has the housing industry not boosted
the economy as much during this recovery?
Demographics would seem to be an important
part of the answer. The average annual increase
in the population aged 25 and over is projected to
fall from the 2.6 million experienced during the
1980s to 1.7 million in the 1990s. As growth in the
adult population slows, so will the demand for
new housing.
The purpose of this article is to measure the
importance of projected shifts in the size and age
distribution of the U.S. population for the rate of
growth in housing demand (that is, net housing
investment). We wish to give the reader a sense of
just how much and for how long the population
slowdown is likely to restrain housing demand.
The analysis runs through the first decade of the
next century and provides separate estimates for
single-family and multifamily housing.
Our results indicate that the contractionary
effects of the population slowdown are already
being felt in the housing industry and probably
have been at work since the latter part of the 1980s.
In our simulations, changes in the size and age
distribution of the population lower net housing
investment by 17 percent from the late 1980s
through the first half of the 1990s. Population
Economic Review — First Quarter 1994

factors then reduce net investment an additional
22 percent from the mid-1990s through the first
half of the first decade of the next century before
turning favorable.
On a percentage basis, the effects of the
population slowdown are greatest in multifamily
building. Population shifts reduce net investment
in multifamily units by 60 percent from the late
1980s through the end of this century. Singlefamily building is not spared, however. Population
factors decrease net investment in single-family
homes by one-third from the late 1980s through
the middle of the first decade of the next century.
Are the demographics inexorable? Is it
possible that changes in immigration policy could
offset the slowdown in the native population? The
numbers show that to stave off a decline in new
home construction, immigration quotas would
have to be doubled, from the current limit of
700,000 people per year to around 1.5 million
per year.
We also investigate whether the effects of
the population slowdown could be reversed by
changes in cohabitation patterns. In the scenario
most favorable to housing investment, we assume
that high economic growth encourages substantial
new household formation and that baby boomers,
who in their younger years had less of a taste for
marriage than did their parents, continue to live as
single adults in relatively high proportions. The
implied changes in household formation have a

Our thanks to John Duca, Evan Koenig, and Mine Yücel for
their valuable comments and suggestions.

17

strong effect on the mix of new housing demand,
greatly favoring multifamily investment. However,
these changes can only reduce the projected
decline in total net investment from 36 percent to
22 percent from the late 1980s through the early
part of the twenty-first century.
Population shifts may not be the only influence on housing demand over the next few
decades. Housing demand is also greatly affected
by interest rates and tax laws. But given what we
know about the size of the decline in births following the end of the baby boom, demographics are
certain to play a major role in the future of the
U.S. housing industry. Any long-term assessment
of housing demand must begin with basic population arithmetic.
Framework for analysis
To calculate the effect of demographic shifts
on housing demand, we use a method similar to
the one developed by Jaffee and Rosen (1979).
We begin with individual population projections
by age group and use historical headship rates to
estimate the household population by age of
household head and type of household (family or
nonfamily). Projections of housing demand by
type of home (single-family or multifamily) then
are developed by combining the estimates of
household population with historical propensities
to demand housing of a particular type by age
and type of household.
The computational framework is given
formally by
HSF = K s ∑ POPi [ fiθ fi + niθni ]

(1)

i

and
(2)

HMF = K m ∑ POPi [ fi (1 − θ fi ) + ni (1 − θni )],
i

where HSF and HMF are the stock demands for
single-family and multifamily housing in a given

1

18

The baby boom generation refers to those born between the
years 1946 and 1964.

year; Ks and Km are scale factors; POPi is the
number of people in age bracket i ; fi is the likelihood of a person of age i heading up a family
household (the family headship rate); ni is the
likelihood of a person of age i heading up a nonfamily household (the nonfamily headship rate);
θf i is the likelihood that a family household headed
by a person of age i would occupy a single-family
home; and θni is the likelihood that a nonfamily
household with head of age i would occupy a
single-family home.
In our basic population analysis, all of the
terms in equations 1 and 2 are treated as constants,
except for the population data. We simply run the
population numbers through the equations to see
how much of an effect on housing investment we
should expect from population shifts alone.
To specify the equations, the scale factors Ks
and Km were chosen so that the two simulated
series on net investment would replicate, respectively, the average value of new single-family and
new multifamily homes put in place during the
period 1970–89. The resulting values are Ks = 73.2
and Km = 45.0, both in thousands of 1982 dollars.
Values used for the remaining parameters are based
on 1980 data and are shown in Table 1. The headship rates are Bureau of the Census estimates.
Following Census convention, family households
are either married couples or single-parent households with at least one child. Nonfamily households include singles and two or more unrelated
individuals sharing a residence. The propensities
to occupy single-family housing are from the
Current Population Survey.
Especially important to the analysis are the
terms in brackets in equations 1 and 2. These
terms represent the number of single-family and
multifamily housing units demanded per capita.
Their values are shown in the last two rows of
Table 1. Within the adult age categories, the most
significant variation in the numbers occurs between
the age groups 25–34 and 35–44. The frequency
of single-family home demand rises by a third and
that of multifamily demand falls by a little more
than a third as households move from one age
group to the next. These differences in per capita
housing demand play a significant role in our
analysis, because every member of the baby boom
generation passes between these age groups
sometime during the 1980s and 1990s.1
Federal Reserve Bank of Dallas

Table 1

Anatomy of Per Capita Housing Demand in 1980
by Age Group and Type of Structure
Age of householder
<25

25–34

35–44

45–54

55–64

≥65

Family (fi )

.04

.36

.47

.47

.43

.34

Nonfamily (ni )

.03

.11

.07

.08

.13

.29

Family (θf i )

.55

.75

.86

.88

.89

.86

Nonfamily (θn i )

.31

.38

.46

.55

.63

.62

Single-family homes per capita
(fi θf i + ni θn i )

.03

.32

.43

.46

.46

.46

Multifamily housing units per capita
[fi (1– θf i ) + ni (1– θn i )]

.04

.16

.10

.09

.10

.17

Headship rates, by type of household

Percent of households occupying
single-family homes, by type of household

The effect of population shifts
on net housing investment
Equations 1 and 2 provide the analytical
framework for measuring the effect on housing
demand of shifts in the size and age distribution
of the population. To carry out the analysis, we
use population data for the period 1980–2010,
with data through 1990 being Census estimates of
actuals and data beyond 1990 being Census projections. The demographics have their most visible
impact on the rate of growth in housing demand,
or net housing investment. Thus, we show the
calculated path of net housing investment (Figure
1). To help smooth the series, the results are presented as annual averages over five-year periods.
The first point to notice is that the population slowdown does not bring about an absolute
decline in housing demand. Net investment remains
positive throughout the forecast period. What the
demographics do is reduce the rate of growth in
housing demand. From the latter half of the 1980s
through the period 2000–04, total net housing
investment falls by 36 percent. On a percentage
Economic Review — First Quarter 1994

basis, the population slowdown is most important
for multifamily building. Net investment in multifamily units declines by 60 percent from the late
1980s through the late 1990s. The demographics
reduce single-family investment throughout the
1990s and into the first half of the first decade of
the next century. Net investment in single-family
homes falls by one-third over this period.
Table 2 details the results by age group. The
most significant patterns in the numbers relate to
the baby boom generation. Baby boomers enter
the 35–44 age group in the early 1980s, producing
a bulge in housing investment in that age bracket
through the mid-1990s. We can then follow the
bulge as the cohort matures. The bulge appears
in the 45–54 age bracket beginning in the early
1990s, and it reappears in the 55–64 group at the
turn of the century. The baby boomers also leave
their mark as they vacate an age bracket. Their
maturation is the reason for the absolute drop in
housing demand in the 25–34 age group during
the 1990s and in the 35–44 age group during the
first decade of the next century.
Also noticeable, although much less signifi19

Figure 1

Net Housing Investment by Type of Structure:
Simulated Series Based on Shifts in the Size
and Age Distribution of the U.S. Population
(Five-Year Averages of Annual Rates)
Billions of 1982 dollars
100
90
80
70
Total
60
50

Single-family

40
30
20
Multifamily

10
0
1980–84

1985–89

1990–94

1995–99

2000–04

2005–09

cant in size, is the effect of the baby bust generation born in the Great Depression. The relatively
low number of births during the 1930s is the reason
for the low net investment numbers in the 55–64
age group during the period 1985–94 and in the
65-plus age group during the period 1995–2004.

population forecast is immigration. In this section
we give the reader a sense of how much more
open U.S. immigration policy would have to be if
a slowdown in housing demand is to be avoided.
To quantify the impact of alternative immigration policies, we use the same algebraic framework as in the previous section but modify the
population numbers to reflect an increased flow of
immigrants. In our simulations, an infusion of new
immigrants occurs each year beginning in 1991
and continuing throughout the forecast period.
Upon arrival, these new immigrants are distributed
across age groups in the same way as legal immigrants who were admitted into the United States
between 1980 and 1988. Except for age, immigrants are identical to natives.2
The increase in immigration allowed for in
the simulations is something over and above the
immigration assumed by the Census Bureau in its
projections. In the Census projections we used for
our base case, net immigration was assumed to
occur at a rate of about 500,000 people per year.
These projections were made before the Immigration Act of 1990, which raised legal immigrant
quotas to about 700,000 per year. We consider
three alternatives to our base case. The first provides for an increase in flows of 200,000 immigrants per year. This case gives us a rough idea
of how important the 1990 reforms will be to the

Housing and immigration policy
Figure 2

Our prediction that net housing investment
will fall sharply over the next two decades is made
essentially on the basis of a projected decline in
the growth of the U.S. adult population. We can
be confident in our assumptions about the future
growth of the native population, because in a
forecast that goes out no more than twenty years,
the size of adult age groups can be estimated
from known birth rates. The major risk in the

Housing and Immigration Policy
(Simulated Net Investment)
Billions of 1982 dollars
110

100

90

80

70
2

20

There are, of course, significant differences between immigrants and natives. Most crucial for housing demand is that
immigrants on average earn less income, even after a long
assimilation period (see Borjas 1990). Thus, our simulations probably overstate the impact of new immigrants on
housing demand.

60

50
1980–84

Base + 1 million
Base + 500 thousand
With 1990 reforms
Base
1985–89

1990–94

1995–99

2000–04

2005–09

Federal Reserve Bank of Dallas

Table 2

Effect of Population Shifts on Net Housing Investment by Type of Structure
and Age of Householder, 1980–2009
(Average Annual Rates, in Billions of 1982 Dollars)
Age of householder
<25

25–34

35–44

45–54

55–64

≥65

Total

–1.1
–.1
.1
.2
.0
–.1

21.4
7.8
–13.8
–17.7
–5.3
7.3

38.0
38.4
28.2
10.0
–18.8
–23.9

–1.1
19.4
39.1
39.9
29.6
10.7

3.9
–6.6
–.3
19.1
38.0
38.3

19.2
20.5
15.0
7.5
9.5
20.9

80.4
79.5
68.3
59.1
52.8
53.3

–.1
–.1
.0
.0
.0
.0

6.6
2.4
– 4.3
– 5.5
–1.6
2.2

5.6
5.7
4.1
1.5
–2.7
–3.5

–.1
2.4
4.9
5.0
3.6
1.3

.5
–.8
.0
2.5
5.0
5.0

4.3
4.5
3.2
1.6
2.1
4.6

16.7
14.1
8.0
5.2
6.4
9.5

Change in demand for
single-family homes
1980–84
1985–89
1990–94
1995–99
2000–04
2005–09
Change in demand for
multifamily housing
1980–84
1985–89
1990–94
1995–99
2000–04
2005–09

housing industry. In our second case, immigrant
flows are raised by 500,000 people per year. In a
final scenario we assume immigration quotas are
increased by 1 million per year.
The results are shown in Figure 2. As one
would expect, housing investment rises uniformly
with each successive increase in the quota limit.
The 1990 reforms are seen to have a modest effect
on housing investment. In the base case, net residential investment drops by 33 percent from the
late 1980s through the end of the first decade of
the twenty-first century. In the scenario with 1990
reforms, investment still falls by 26 percent. To
avoid a decline in net housing investment, immigration quotas have to be raised to 1.5 million per
year, more than double the amount under current
policy.
Housing and headship rates
In projecting housing demand from demographic data, it is necessary to know not only how
Economic Review — First Quarter 1994

many people there are but how they group themselves into households. Over the past two decades,
there has been a growing trend toward singleadult households. Rising divorce rates, delayed
marriages, and greater societal acceptance of
singleness have contributed to this trend. Because
single adults have a higher propensity to rent
apartments than do families, we would expect
that the trend toward singleness has tilted the
demand for housing away from single-family and
toward multifamily units. With a greater number
of households being formed from a given population, the overall level of housing demand may
also have been raised.
The purpose of this section is to determine
the importance of recent and possible future
changes in cohabitation patterns for housing investment. We once again use equations 1 and 2,
but now we allow headship rates to change over
time. For the years 1980 through 1990 we use
Census estimates of actual headship rates. For the
years 1991 through 2010, we consider a range of
21

possible values based on the work of Hendershott
(1988). Hendershott’s projections run from 1990
through the year 2000. We use these projections
to extend the actual headship rates to 2000. For
the first decade of the twenty-first century, we
assume that the trends over the previous ten years
continue but at only one-half the rate.
From the group of alternative scenarios
suggested by Hendershott, we chose two that
produce a wide range of possible outcomes for
housing investment.3 In the scenario we label
“high,” the projected headship rates are based
on the assumption that economic growth will be
high and that baby boomers will continue to have
a strong preference for living as single adults,
even as they grow older. In the “low” scenario,
income growth is assumed to be low and there is
a less rapid decline in the married-couple share
of households.4
The results of the simulations involving
changing headship rates are shown in Figures 3
through 5. For comparison, we also present the
base case. Beginning with actuals, the changes in
headship rates that occurred during the 1980s
appear to have altered the mix of housing investment but not the total amount. Comparing the
base case with the series simulated from actual
headship rates, net investment in multifamily
homes is raised 25 percent by the changes in
headship rates. This rise is offset by a comparable
absolute decline in single-family investment, however, so that total net residential investment is
essentially unchanged.

22

3

In addition to the Hendershott scenarios, we considered an
alternative forecast using projected headship rates from the
Bureau of the Census. The results using these projections
fell between the other two scenarios. Thus, we did not
include these results in the text.

4

Hendershott projects total headship rates (family plus
nonfamily) on the basis of assumptions about future economic conditions. The relationships are consistent with the
theory that privacy is a normal good. More households will
be formed from a given population the greater the ability of
the population to afford housing. The breakdown of total
headship rates by type of family depends upon assumptions about future tastes for marriage.

Figure 3

Housing and Headship Rates: Alternative
Scenarios Involving Changing Patterns
of Cohabitation (Simulated Net Investment)
All Homes
Billions of 1982 dollars
110

100

90

80
High

70
Low

60
Base-fixed

50
1980–84

1985–89

1990–94

1995–99

2000–04

2005–09

Turning to the projections, the results from
the “low” scenario are similar to those for the
1980s. The projected changes in headship rates
have a large percentage effect on multifamily
investment, but only a small effect on total investment. The total investment series with changing
headship rates is, over the entire forecast period,
only 5 percent higher than the base case series.
The changes in headship rates assumed in
the “high” scenario have a more sizeable impact
on housing investment. Once again the results are
most dramatic in multifamily investment. The
projected changes in headship rates almost double
the average annual rate of multifamily investment.
Because of substantial increases in total headship
rates in this scenario, single-family investment is
also raised, by an average of 11 percent over the
forecast period. In total, net housing investment is
19 percent higher because of the projected changes
in headship rates. These gains are not sufficient,
however, to offset the contractionary effects of the
population slowdown. Total net housing investment continues to fall in this scenario.
Long-run forecasts of multifamily building
clearly must factor in the possibility of a continued
trend toward single-adult households. Not to do
so would understate investment by one-quarter or
Federal Reserve Bank of Dallas

more. Future changes in cohabitation patterns are
probably less crucial in the overall outlook for
residential construction. We obtained numerically
significant results for total housing investment
only after making extremely aggressive assumptions about future rates of household formation.
Even in this case, total residential investment is
projected to fall 22 percent from the late 1980s
through the period 2000–04.
Implications
Our analysis has focused on the rate of
growth in housing demand, or net housing investment. We chose to present our results in this way
because shifts in the size and age-mix of the
population speak more directly to this variable
than to any other. Those interested in the future
of the housing industry, on the other hand, are
probably more concerned with what the population slowdown will mean for construction jobs
and home prices. We conclude with a discussion
of what our results suggest will happen to these
variables. In our discussion, we use results from
the base-case simulations, with fixed headship
rates. This represents something of a worst-case
scenario. But absent a major liberalization of
immigration policy or rapid economic growth, it
may not be far off the mark.

Figure 4

Housing and Headship Rates (Continued)
Single-family
Billions of 1982 dollars
85
80
75
70
65
High

60

Low

55

Base-fixed
50
45
1980–84

1985–89

1990–94

1995–99

Economic Review — First Quarter 1994

2000–04

2005–09

Figure 5

Housing and Headship Rates (Continued)
Multifamily
Billions of 1982 dollars
20
18
16
14

High

12

Low

10
8
6
4
1980–84

Base-fixed

1985–89

1990–94

1995–99

2000–04

2005–09

To assess the outlook for residential construction employment, we need to think in terms
of gross investment rather than net investment.
That is, we need to consider the construction that
is needed to maintain and replace worn out
buildings as well as that required to provide for a
growing household population. To obtain estimates
of gross housing investment, we assume that the
stock of single-family homes depreciates at an
annual rate of 2.25 percent and that apartments
depreciate at an annual rate of 4 percent. Gross
investment, then, is the sum of net investment
plus what is needed to offset depreciation. In
Figure 6 we show the resulting series on gross
residential investment, along with the baseline
series on net housing investment. To make comparisons easy, we index each series to equal 100
over the period 1980–84.
The population slowdown will bring about a
sharp reduction in net housing investment but no
significant change in gross investment. Thus, the
homebuilding industry need not contract absolutely. There will be little if any job growth, however, and the industry is certain to play a smaller
role in the economy. The top line in Figure 6
shows how much gross investment would have to
rise to keep pace with historical and projected
growth in the U.S. labor force. The width of the
gap between this line and the line on gross housing
23

Figure 6

Gross Investment Needed to Maintain
Housing’s Share of Employment
Index (1980–84) = 100
155
For constant employment share
145
135
125
115
Projected gross investment

105
95
85
75

Projected net investment

65
55
1980–84

1985–89

1990–94

1995–99

2000–04

2005–09

investment indicates the degree to which demographic changes will reduce the share of residential construction in national employment. With
gross investment being essentially flat and the
labor force growing about 50 percent from the
early 1980s through the year 2010, housing’s share
of employment is reduced by one-third.
Turning to home prices, it is useful to think of
the price of a home as reflecting two components:
the price of the land and the price of the structure. Given a certain fixity in the supply of land
suitable for residential development, land prices
will move with the stock demand for housing—
rising as housing demand rises and falling as
housing demand falls. Our analysis shows that
future demographic shifts will reduce the rate of
growth in housing demand but not its absolute
level. Housing demand will continue to rise over
the foreseeable future. There is, then, no apparent
reason for residential land prices to be weakened
by the population slowdown. It is always possible
that real estate markets have failed to appreciate
the extent of the slowdown in housing demand,
having capitalized excessively any future appreciation in land values and having set themselves
up for a price correction. But this would be a
matter of some speculation and certainly not a
necessary consequence of the demographics.
24

With a rising supply price for new home
construction, the price of residential structures will
vary directly with the rate of gross investment
demand. How much prices would fall in response
to a decline in investment demand depends on
the size of the drop in demand and the price
elasticity of supply of new structures. From the
work of Muth (1983), it is widely believed that the
supply of new homes is highly elastic in the long
run, ensuring a limited price adjustment whatever
the shift in demand. Our analysis further suggests
that the shift in demand is not likely to be large in
the first place. In our simulations, the demographics halt the growth of gross housing investment,
but they do not reduce it.
Considering both land and structures, it is
difficult to see in the population numbers a compelling reason for average home prices to fall.
Thus, we strongly disagree with the conclusion of
Mankiw and Weil (1989) that home prices may fall
by half over the next two decades because of the
demographic slowdown. There is more potential,
we believe, for relative price adjustments to take
place between different types of homes. The stock
demand for housing will fall sharply for households in the age group 25–34 during the 1990s
and for those aged 35–44 during the first decade
of the next century. Prices of homes specialized to
suit people in these age brackets (starter homes,
homes for families with young children) may well
weaken. On the other hand, the demographics
will serve to strengthen the prices of homes that
are popular with older adults who have graduated
their children, the so-called empty nesters.
The population slowdown is an important
economic and social event with the potential to
substantially reduce the importance of homebuilding in the economy and to alter the prices of some
single-family homes. However, these changes will
be consumer driven and so should not be resisted.
The changes also will take decades to play out
and are relatively easy to forecast. They would not
seem to pose a significant threat to macroeconomic
stability. Policymakers need to be well-informed
about the extent of the change in the housing
industry that can be expected from the demographics to avoid overstimulating the economy
and causing undue delay in the process of structural change.

Federal Reserve Bank of Dallas

References
Borjas, George J. (1990), Friends or Strangers: The
Impact of Immigrants on the U.S. Economy
(New York: Basic Books).
Bureau of the Census, Current Population Reports,
series P-20, Population Characteristics, and P25, Population Estimates and Projections,
various issues.
DRI/McGraw-Hill (1993), Review of the U.S. Economy: Long-Range Focus, Summer, 35–36.
Hendershott, Patric H. (1988), “Household Formation and Homeownership: Impacts of Demographic, Sociological and Economic Factors,”
Housing Finance Review 7: 201–24.
Jaffee, Dwight M., and Kenneth T. Rosen (1979),
“Mortgage Credit Availability and Residential
Construction,” Brookings Papers on Economic
Activity, no. 2, 333– 86.

Economic Review — First Quarter 1994

Mankiw, N. Gregory, and David N. Weil (1989),
“The Baby Boom, the Baby Bust, and the
Housing Market,” Regional Science and Urban
Economics 19: 235–58.
Muth, Richard F. (1983), “Effects of the U.S. Tax
System on Housing Prices and Consumption,”
in The Urban Economy and Housing, ed.
Ronald E. Grieson (Lexington, Mass.: Lexington Books).
Perry, George L., and Charles L. Schultze (1993),
“Was This Recession Different? Are They All
Different?,” Brookings Papers on Economic
Activity, no. 1, 145–211.
U.S. Department of Justice, Immigration and
Naturalization Service, 1989 Statistical Yearbook of the Immigration and Naturalization
Service, Table 11, 24.

25

26

Federal Reserve Bank of Dallas

Gregory W. Huffman
Associate Professor of Economics
Southern Methodist University and
Research Associate
Federal Reserve Bank of Dallas

A Primer on the Nature of Business Cycles

F

or a considerable time, economists have devoted
much effort to obtaining a greater understanding of the causes of the business cycle, or (as it
used to be called) the trade cycle. Business cycles,
in themselves, are thought by many to be undesirable. Therefore, a greater understanding of the
nature and causes of business cycles would be
useful in leading to the development of government fiscal or monetary policies that alleviate their
impact. Dynamic economic models developed in
the past decade have been especially useful in
enhancing our understanding of observed business cycles. Although the economics profession
apparently still has some way to go to understand
the full nature and causes of these fluctuations, it
is possible at this stage to describe which features
or sectors of economies contribute the most to
observed business cycles.
Economists and analysts sometimes disagree
about what might be the primary source of observed
business cycles. Some might say that these cycles
are caused by changes in technology, while others
would ascribe much of the culpability to the
behavior of government or central banks. The
point of this article is not to settle this issue, nor
even to describe the controversies. Instead of
focusing on issues on which economists may disagree, this article is intended to study the issues
about which little disagreement can take place.
Specifically, the intention here is to describe the
behavior of observed economic aggregates over
the course of the business cycle. This article, then,
is to be a “user’s guide” to obtain a better understanding of the business cycle in the United States
in particular and in market economies in general.
The article is organized as follows. The next
section will show exactly how the different components of aggregate output behave over the
course of the business cycle. Additionally, the
Economic Review — First Quarter 1994

different categories of consumption and investment
will be studied in more detail. The behavior of
labor productivity over the course of the business
cycle will then be analyzed. Lastly, the businesscycle properties of the U.S. and Canadian economies will be compared.
Defining the business cycle
There are several different ways to define
“the business cycle.” Loosely speaking, the term
usually refers to fluctuations of economic aggregates around their trend values. This is easily
understood by looking at Figure 1. The solid line
is the actual time path of the U.S. gross national
product from the first quarter of 1947 through the
third quarter of 1991.1 The dotted line is what one
might identify as the trend value of output over
the same period.2 The difference between these
two lines will be referred to as the fluctuation of
actual output around its trend value. Fluctuations
of output above and below its trend are what is
usually referred to as the business cycle. (See the

The comments of Stephen Brown, Kenneth Emery, and
Mark Wynne are gratefully acknowledged.
1

The data are in 1982 dollars. This data set is the most recent
for which government spending is available back to 1947.

2

It should be noted that there are several ways to describe
the behavior of the trend of an economic aggregate such as
GNP. The method here is that employed in Hodrick and
Prescott (1980). This method happens to give rise to a
variable trend growth rate. An alternative description of the
trend would change the definition of the cycles. However,
for the most part, the magnitude and correlations of these
fluctuations, described later, would be roughly the same
even with an alternative definition of the trend.

27

Figure 1

Actual and Trend Levels of GNP
Trillions of constant (1982) dollars
4.5
Actual GNP
Trend level of GNP

4

3.5

3

2.5

2

1.5

1
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

SOURCE OF PRIMARY DATA: U.S. Department of Commerce.

box titled “Is There a Trend in Economic Time
Series?” on page 40.)
Obviously, aggregate output is not the only
economic aggregate that exhibits growth and
fluctuations; almost all the aggregates do. Therefore, for any such aggregate, its business-cycle
fluctuations can be described as the fluctuations
around its trend value.
Aggregate spending in the United States can
be separated into its components of aggregate consumption spending, investment and government
spending, and exports and imports. Furthermore,
these aggregates can be broken down into narrower
categories, as will be shown later. It is then enlightening to inquire, Which components of aggregate output contribute to its observed fluctuations?
Consumption. In Figure 2, the fluctuations of
aggregate output around its trend are the solid
line. The vertical axis is a measure of the percentage deviation of the variable from its trend value.
The dotted line represents the fluctuations of
aggregate consumption. Analysis of this diagram
reveals that the fluctuations in aggregate consumption are somewhat smaller than those of

aggregate output. Perhaps this is not too surprising. Consumers apparently wish to “smooth” their
consumption patterns: they do not increase their
spending too much when times are good and do
not cut back too much when times are bad.
The reason for this behavior can perhaps be
best illustrated by the simple microeconomic
experiment depicted in Figure 3. This diagram
illustrates the choices open to an agent who must
make consumption choices in two time periods.
The horizontal axis measures the amount of
consumption in the first period, while the vertical
axis measures second-period consumption. The
agent can spend M units on consumption in the
two periods, and R is the net real interest rate.
M represents the maximum real discounted value
of consumption in the two periods. Given this
information, the agent will then have indifference
curve I 0 tangent to the budget constraint with
wealth M. The agent will choose to consume C 1*
in the first period and C 2* in the second period.
Should some event occur, such as an increase in
wealth from M to M ′, the agent can afford to
consume more. The diagram illustrates the case
in which the agent chooses to consume more in
each period and, so, consumes C 1′ in the first
period and C 2′ in the second period. This is the
sense in which consumers are said to want smooth
consumption patterns.3

Figure 2

Aggregate Output and Consumption
Deviation from trend value
(Percent)
4

2

0

–2

–4

–6

3

28

This is nothing more than a restatement of the permanent
income hypothesis, as described by Milton Friedman (1957).

Aggregate output
Aggregate consumption

–8
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

Federal Reserve Bank of Dallas

Figure 3

Consumption Choices over a
Two-Period Horizon
Period 2
consumption
M′(1 + R)

I1
M(1 + R)
I0
C2′

C*2

C1*

C1′

M

M′

Period 1
consumption

A practical illustration of such a phenomenon
is that a person who wins a lottery and has a large
increase in income, or a person who temporarily
loses his job and has a decrease in income, rarely
changes consumption purchases by an amount
equal to the change in income but, instead, spreads
the change in income out by changing both present
and future levels of consumption purchases.
Consequently, with this analysis in mind, it is
not surprising to observe that aggregate consumption does not exhibit pronounced fluctuations
relative to those of aggregate output. Although
aggregate consumption obviously fluctuates, its
fluctuations could hardly be said to be the driving
force behind aggregate output fluctuations. For
aggregate output to fluctuate as much as it does,
some component of output other than consumption must fluctuate more than does consumption.
Figures 4 through 6 present a further
breakdown of the behavior of aggregate consumption. Figure 4 shows that the consumption of
nondurable goods fluctuates very little relative to
the level of aggregate output. Figure 5 illustrates
similar behavior for the consumption of services.
On the other hand, Figure 6 shows that the
consumption of durable goods fluctuates much
more than does the level of aggregate output.
This diagram indicates that consumer purchases
of such items as appliances and automobiles
increase (decrease) substantially when output is
Economic Review — First Quarter 1994

growing (falling) relative to its trend value.
The behavior of the various components of
aggregate consumption is further illustrated in
Table 1. The first column indicates the relative
volatilities of the components of aggregate output,
as measured in percentage standard deviations,
for the sample period. The percentage standard
deviation of aggregate output over the period is
1.92 percent, and aggregate consumption fluctuates slightly less than does aggregate output.
Nondurables consumption and consumption of
services fluctuate less than does total consumption, but the consumption of durable goods
fluctuates more.
The second column of Table 1 shows how
the various aggregates are correlated with aggregate output. The closer are these numbers to 1,
the more likely the relevant variable will tend to
move in the same direction as aggregate output. It
is clear from the table that all categories of consumption are procyclical; that is, on average, they
tend to grow when aggregate output grows and to
decline when output declines. However, these
variables differ in the amount of fluctuation over
the course of the business cycle.
Because of the relatively small fluctuations in
aggregate consumption, some researchers have indicated that the presence of business cycles should

Figure 4

Aggregate Output and Nondurables
Consumption
Deviation from trend value
(Percent)
4

2

0

–2

–4

–6

Aggregate output
Nondurables consumption

–8
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

29

Table 1

Cyclical Behavior of Various U.S. Economic
Time Series, 1947:1–1991:3
Percentage
standard deviation

Correlation
with output

Aggregate output

1.92

1.000

Aggregate consumption
Durable goods
Nondurable goods
Services

1.24
5.36
1.23
.71

.681
.441
.635
.671

Aggregate investment
Producer durable equipment
Nonresidential structures
Residential structures

8.72
6.25
4.55
10.77

.777
.793
.476
.422

Aggregate government
Federal defense
Federal nondefense

4.38
9.31
12.11

.346
.411
–.164

Exports

6.92

.427

Imports

5.04

.647

Inventories

1.90

.645

Hours of work

1.85

.898

.86

.302

Labor productivity

SOURCE OF PRIMARY DATA: U.S. Department of Commerce.

30

4

For example, Lucas (1987) constructs a model that allows
him to ask how much average lifetime consumption the
typical consumer would be willing to forgo to fully insure
himself against future consumption fluctuations. For a wide
range of plausible parameter values, Lucas finds that such
a consumer would be willing to give up less than one-tenth
of 1 percent of the average consumption level to rid himself
of these fluctuations. This is not to say that consumers or
policymakers should be indifferent about business cycles.
Instead, the implication is that the societal costs of the
fluctuations are likely to be small relative to the costs
imposed by, say, distortional fiscal policy or compared with
the benefits to be gained by even moderate increases in the
consumption growth rate.

5

Table 2 measures the simple correlations of aggregate
output and the past or future levels of various components
of consumption. For example, the correlation of output with
the level of purchases of consumer durables two quarters
ago is 0.465. The correlation of output with the level of
consumption of services three quarters in the future is
0.181.

not be a matter of great concern.4 The reason is
that consumers, for the most part, care about the
quantity of goods they are able to consume.
However, as indicated in Table 1, the quantity of
aggregate consumption does not fluctuate very
much over the course of the business cycle.
Table 2 shows how the components of
consumption change as output changes at various
lags.5 All components of consumption are fairly
highly correlated with output several quarters in
the future. This means that purchases of consumption goods will begin to increase even before other
components of aggregate output begin to rise and
will fall before aggregate output begins to fall.
Investment. Figure 7 presents a comparison of
the fluctuations in aggregate output and those of
aggregate investment. It is apparent that the fluctuations in investment are much larger than those
in output. Table 1 also illustrates the relatively
large volatility of investment. Not all categories of
Federal Reserve Bank of Dallas

Figure 5

Figure 6

Aggregate Output and Services Consumption

Aggregate Output and Durables Consumption

Deviation from trend value
(Percent)

Deviation from trend value
(Percent)
25

4

Aggregate output
Durables consumption

20
2

15
0

10
5

–2

0
–4

–5
–6

Aggregate output
Services consumption

–10

–8

–15
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

investment, however, behave in the same manner
over the course of the business cycle. In particular, investment in producer durable equipment
and investment in residential structures are
especially volatile. Nevertheless, all components
of aggregate investment are procyclical.
As Table 1 indicates, of all the components
of aggregate output, aggregate investment and its
subcomponents apparently are likely responsible
for most of the observed fluctuations in output.
Investigating the behavior of each subcomponent
of aggregate investment reveals which fluctuates
most over the course of the business cycle. As
shown in Table 1, investment in residential structures (houses and apartments) exhibits extreme
fluctuations, with investment in producer durable
equipment and nonresidential structures being
somewhat less volatile but still more volatile than
aggregate output itself.
Table 3 shows how the different components of investment behave at different points in
the business cycle. In the first column, the correlation between aggregate output and residential
investment two quarters ago is 0.610, a relatively
high value. Apparently, just before the growth rate
of aggregate output begins to rise (fall), residential
construction begins to rise (fall). Hence, one
might think of residential construction as a leading
indicator of aggregate output. The behavior of
Economic Review — First Quarter 1994

’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

nonresidential investment is not at all similar to
that of residential investment. The correlation
between aggregate output and nonresidential
investment two quarters ago is 0.016. However,
the correlation of output and nonresidential
investment two quarters in the future is 0.557.
This might mean that producers or firms are
reluctant to undertake this type of fixed invest-

Figure 7

Aggregate Output and Investment
Deviation from trend value
(Percent)
30
Aggregate output
Aggregate investment

20

10

0

–10

–20

–30
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

31

Table 2

Correlation of Various Components of Consumption with Aggregate Output
for United States at Various Lags
Correlation of output
Lag length
(Quarters)

With consumption of
durable goods

With consumption of
nondurable goods

With consumption of
services

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

.065
.127
.187
.253
.348
.411
.465
.480
.441
.249
.033
–.201
–.346
–.424
–.387
–.343
–.268

–.211
–.099
–.039
–.191
–.335
.455
.575
.648
.635
.520
.339
.139
–.013
–.146
–.223
–.283
–.307

–.243
–.207
–.130
–.009
.187
.397
.580
.692
.671
.518
.336
.181
.043
–.058
–.115
–.168
–.211

ment until other components of output have
already begun to increase.
Similar behavior is seen for investment in
producer durable equipment. The correlation
between aggregate output and investment in this
equipment two quarters ago is 0.369, while the
correlation for two quarters in the future is 0.674.
Aggregate output is very highly correlated with
investment in equipment in the same period, with
a correlation coefficient of 0.793. Table 3 indicates
that producers are willing to begin investment in
durable equipment slightly earlier than in structures (nonresidential investment). This might be
attributed to the fact that investments in equipment are typically smaller than those in structures,

6

32

An increase or decrease in inventories of finished goods,
semifinished goods, or raw materials is classified as an
investment in inventories, and this is a component of aggregate GNP.

and producers are reluctant to make the larger
investments until they are confident that sales
have increased.
Table 3 also illustrates an interesting behavior for inventories. The correlation between aggregate output and inventories one to three quarters
in the future is high. There is one obvious possible reason. As aggregate output rises, consumers
increase their purchases of goods and thereby
help deplete producer inventories, which firms,
after several quarters, seek to replenish. Conversely, when aggregate output begins to fall,
consumers hold off on these purchases, which
helps to increase firms’ inventories. Firms then
move to lower the level of inventory holdings to
minimize costs.
Table 1 and Figure 8 also show the behavior
of aggregate inventories. This variable fluctuates
to just about the same degree as does aggregate
output. However, as will be shown, the change
in inventories is also a component of aggregate
investment, and it fluctuates tremendously.6
Federal Reserve Bank of Dallas

As noted above, it is of interest to investigate
which components of aggregate investment contribute most to the large fluctuations in the total.
The prime candidate for being the primary source
of these fluctuations would seem to be aggregate
investment. Within this category, residential construction causes much of the fluctuations in total
investment. However, another variable (excluded
from Table 1) also contributes a great deal to the
fluctuations in investment—namely, investment in
business inventories. We can break down the
variance of aggregate inventories according to this
equation: ∆I = ∆INVEN + ∆OI, where ∆I represents the change in total investment, ∆INVEN
represents the change in inventory investment,
and ∆OI is the change in all other forms of investment. This equation can then be used to show the
following relationship for the respective variances:7

Figure 9

Changes in Aggregate and Business
Inventory Investment
Billions of constant (1982) dollars
80
Change in aggregate investment
Change in business inventory investment

60
40
20
0
–20
–40
–60
–80
–100
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

var(∆I ) = var(∆INVEN ) + var(∆OI )
+ 2cov(∆ INVEN,∆OI ).
This equation provides us with a tool to
describe to what degree the changes in inventories
are responsible for the behavior of changes in
total investment. For the U.S. data, the values computed for the variables are as follows: var(∆I ) =
473.57, var(∆INVEN ) = 269.64, and var(∆OI ) =
125.27. The covariance in the equation is of negli-

Figure 8

Aggregate Output and Business Inventories
Deviation from trend value
(Percent)
6

gible size. In other words, changes in inventory
investment apparently are responsible for a very
large quantity of the change in total investment
over the course of the business cycle. This is
especially surprising because the average change
in business inventories represents only 3.4 percent
of average changes in total investment in the
post–World War II period.
Figure 9 further illustrates this behavior. Shown
here are the change in aggregate investment and
the change in business inventory investment. From
the diagram it is hard to see the difference between
these two variables, but this is the important point.
Despite the fact that business inventory investment
is a very small portion of total investment, most of
the changes in the latter variable from quarter to
quarter are due to changes in inventory investment.

4

2

0
7

–2

–4

Aggregate output
Business inventories

–6

–8
’47

’52

’57

’62

’67

’72

’77

Economic Review — First Quarter 1994

’82

’87

’92

The variance of a variable is a measure of the degree of
fluctuation exhibited by the variable in question. For example, if var(∆I) equals zero, then it must be the case
that the change in investment is always the same.
Cov(∆INVEN,∆OI) is the covariance of the change in inventory investment and the change in all other forms of investment. This is a measure of the degree to which these two
variables move together over the course of the business
cycle.

33

Table 3

Correlation of Various Components of Investment and Inventories
with Aggregate Output for United States at Various Lags
Correlation of output
Lag length
(Quarters)

With lagged
residential
investment

With lagged
nonresidential
investment

With lagged
producer durable
equipment

With lagged
inventories

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

.032
.115
.213
.330
.468
.565
.610
.572
.422
.172
–.078
–.276
–.398
–.423
–.396
–.350
–.304

–.335
–.383
–.413
–.399
–.331
–.186
.016
.261
.476
.561
.557
.466
.337
.201
.060
–.043
–.128

–.336
–.309
–.259
–.182
–.056
–.124
.369
.614
.793
.811
.674
.467
.227
.016
–.136
–.248
–.323

–.351
–.361
–.368
–.349
–.283
–.131
.094
.362
.645
.805
.844
.766
.596
.392
.176
–.022
–.193

It should also be noted that by far the largest portion of business inventory investment is attributable
to changes in inventories of nonfarm businesses.
Government spending, exports, and imports.
Table 1 and Figure 10 indicate that total government spending is more than twice as volatile as
aggregate output. Additionally, both defense
spending and nondefense spending of the federal
government are much more volatile than is
aggregate output. Given this behavior of federal
spending, government spending at the state and
local levels clearly is much less volatile over the
course of the business cycle.
Although total government spending is
moderately procyclical, federal nondefense spending is countercyclical. This behavior might be
attributed to what are sometimes referred to as
“automatic stabilizers.” That is, some types of
government spending programs actually increase
(decrease) more when aggregate output is declining (increasing), which contributes to the countercyclical spending pattern.
34

Table 1 and Figures 11 and 12 show that
exports and imports are much more volatile than
is aggregate output, but both are procyclical.
Exports to other countries typically increase when
income in the other countries increases and the
foreign consumers demand more American-made
goods. To the extent that income in other countries moves in tandem with that in the United
States, one would expect U.S. exports to be procyclical. For identical (but reversed) reasons, U.S.
imports also would be procyclical.
Productivity and employment hours. In
analyzing the behavior of the business cycle, it is
important to consider employment, or total hours
of work, and labor productivity. Productivity
refers to how much is produced, on average, by
each hour worked. In other words, one usually
gauges the productivity of the U.S. economy by
calculating the quantity of goods and services
produced and then dividing by the quantity of
hours worked in producing those goods and
services.
Federal Reserve Bank of Dallas

Figure 13 and Table 1 indicate that hours of
work are about as volatile as is aggregate output
over the course of the business cycle. But Figure
14 and Table 1 indicate that labor productivity is
less volatile than output or hours of work. This is
of interest because one might tend to believe that
changes in productivity are closely linked to changes
in hours of work. That is, when workers are most
productive, it will be in the interest of employers
to hire more workers or to have employees work
longer hours. Apparently, however, relatively
small changes in productivity help produce larger
swings in the quantity of hours worked.
A well-known tenet is that the growth rates
of some components of aggregate output increase
before those of other variables. A variable whose
growth rate increases just before a period of faster
economic growth is referred to as a leading
indicator of aggregate output.8 It is important to
investigate which variables exhibit this behavior
and which variables exhibit faster growth after
most other variables.
Analysis of Figure 14 shows why some
people think of labor productivity as a leading
indicator of aggregate output. Changes in productivity tend, on average, to be followed by changes
in aggregate output, in the same direction, from
three to five quarters later. Therefore, if labor
productivity were to begin to rise substantially this

Figure 10

Aggregate Output and Government Spending
Deviation from trend value
(Percent)
15

10

5

0

Figure 11

Aggregate Output and Exports
Deviation from trend value
(Percent)
20
15
10
5
0
–5
–10
–15

Aggregate output
Aggregate exports

–20
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

quarter, one might reasonably expect aggregate
output to increase in about a year. Note that
Figure 14 shows that this sequence does not
happen for every period but on average.
This relationship is illustrated in the middle
column of Table 4, which lists the correlations
between output and labor productivity at various
lags. Changes in productivity provide practically
no information about how output will behave two
years into the future but are a good leading indicator of output at shorter ranges, such as one year.
However, contemporaneous productivity and
output exhibit a much weaker correlation. Note
also that this relationship is not symmetric; output
is a “negative” leading indicator of productivity.
This analysis illustrates why some economists
suggest that the key to higher economic growth
in the United States is to raise labor productivity.
Table 1 indicates that a rise in labor productivity
is likely to be accompanied by a subsequent increase in aggregate output. Furthermore, the more
productivity rises, the more future output will

–5

–10
Aggregate output
Aggregate government spending

–15

8

–20
’47

’52

’57

’62

’67

’72

’77

Economic Review — First Quarter 1994

’82

’87

’92

Productivity is only one of several measures of future
economic activity. Koenig and Emery (1993) provide an
analysis of the performance of the U.S. Commerce
Department’s composite index of leading indicators.

35

residential investment and investment in producer
durable equipment are leading indicators. They
begin to increase just before aggregate output
rises. In addition, Table 2 shows that consumption
of nondurable goods and consumption of services
are also leading indicators.

Figure 12

Aggregate Output and Imports
Deviation from trend value
(Percent)
15
10

Some international comparisons

5
0
–5
–10
–15
Aggregate output
Aggregate imports

–20
–25
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

increase. Figure 14 shows how the recent increase
in productivity has apparently been accompanied
by a subsequent rise in output, which has helped
pull the U.S. economy out of the latest recession.
Table 1 and Figure 13 show that the level of
output and total hours of work are very highly
correlated. This is not surprising. If the quantity of
goods and services produced is to rise, more
work must be undertaken to produce the goods
and services because, over short periods, increasing the use of labor is easier than increasing the
quantity of capital. However, with output and
employment tending to move in tandem, it is not
surprising that labor productivity is a good leading
indicator of future employment hours as well. The
third column of Table 4 shows that a change in
labor productivity tends, on average, to be followed
by a change in employment hours, in the same
direction, from four to six quarters later.
Lastly, labor productivity is not the only
leading indicator or gauge of future economic
activity. Table 3 indicates that, to some extent,

An appropriate question at this juncture is,
How robust are the above-described features of
the U.S. business cycle? That is, do all market
economies exhibit the same sort of cyclical
fluctuations as does the U.S. economy, or is each
economy very special or different? If all economies
exhibit very different types of business cycles,
then the policy remedies used to deal with them
might need to be quite distinctive. On the other
hand, if economies exhibit similar business cycles,
then unique and economy-specific policies do not
have to be used.
Fortunately, many market economies apparently exhibit cyclical fluctuations that are very
similar to those observed in this country. The
Canadian economy is a good example. Table 5
presents statistics for the Canadian economy that
are the counterpart of the U.S. statistics in Table 1.9
The two economies are strikingly similar in many
respects. First of all, both have aggregate invest-

Figure 13

Aggregate Output and Employment Hours
Deviation from trend value
(Percent)
4

2

0

–2

–4

9

36

In 1991, Canada accounted for 20.4 percent of U.S. merchandise exports and 19 percent of U.S. merchandise
imports. Japan, the next biggest trading partner, accounted
for only 11.3 percent and 18.7 percent, respectively.

–6

Aggregate output
Hours of work

–8
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

Federal Reserve Bank of Dallas

Table 4

Correlation of Aggregate Output and Employment Hours
with Labor Productivity for United States at Various Lags
Lag length
(Quarters)

Correlation of output
with lagged productivity

Correlation of hours
with lagged productivity

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

.160
.315
.419
.480
.492
.486
.488
.390
.302
.001
–.183
–.314
–.343
–.325
–.294
–.264
–.220

.262
.427
.521
.561
.523
.441
.301
.095
–.149
–.310
–.400
–.389
–.343
–.273
–.219
–.175
–.134

ment that is much more volatile than aggregate output, which, in turn, is more volatile than aggregate
consumption. For both countries, durable goods
consumption is more volatile than nondurable
goods consumption, which is more volatile than
service consumption in both countries. In both
economies, the variability of investment in residential construction is greater than that of producer
durable equipment, which is greater than the
variability of investment in nonresidential structures. Government spending, imports, and exports
exhibit similar degrees of variability. The correlations of these economic time series with aggregate
output for the respective countries are also similar.
There are other business-cycle features that
are of much interest. Many models commonly
used to study business cycles generally imply that
movements in consumption in two different
countries should be highly correlated. This implication is especially strong when the countries have
a great deal of trade in goods and capital, as do
Canada and the United States. The reason is simple.
As illustrated earlier, there is a general presumption that consumers prefer smooth consumption
Economic Review — First Quarter 1994

patterns, rather than “feasting” today and “famine”
tomorrow. If some temporary event in the United
States causes consumers to cut back on consumption today, they should be able, at least to some
Figure 14

Aggregate Output and Labor Productivity
Deviation from trend value
(Percent)
4

2

0

–2

–4

–6

Aggregate output
Labor productivity

–8
’47

’52

’57

’62

’67

’72

’77

’82

’87

’92

37

Table 5

Cyclical Behavior of Various Canadian Economic
Time Series, 1947:1–1991:3
Percentage
standard deviation

Correlation
with output

1.84
1.77
6.96
1.29
1.09
5.01
7.79
6.29
8.14
4.38
4.04
5.69
4.19

1.000
.753
.665
.478
.471
.624
.568
.367
.370
.184
.536
.762
.300

Aggregate output
Aggregate consumption
Durable goods
Nondurable goods
Services
Aggregate investment
Producer durable equipment
Nonresidential structures
Residential structures
Aggregate government
Exports
Imports
Inventories
SOURCE OF PRIMARY DATA: Statistics Canada.

extent, to borrow from abroad (through financial
markets) to increase present consumption and pay
back the loan by forgoing future consumption.
Hence, even the vagaries of the business cycle
should still leave consumption in different countries highly correlated, despite any other similarities or differences in aggregate behavior.
This description applies, in particular, to the
presumed behavior of consumption of services
and nondurable goods, because they are easily
purchased. The general economic presumption,
however, is that purchases of durable goods
might be somewhat less correlated across countries. The reason is that consumers are somewhat
inhibited in purchasing and selling durable goods
to smooth their consumption paths of these
goods. Furthermore, consumers sometimes have
to acquire outside financing to purchase houses,
cars, and televisions. Arranging such financing can
be costly, and a consumer may be reluctant to

10

38

Backus, Kehoe, and Kydland (1992) document that for
many countries, aggregate consumption is less highly
correlated across countries than is aggregate production.

purchase and sell numerous durable assets.
Additionally, because of the relatively thin resale
market in used consumer durable goods, it is
costly for a consumer to trade in them to smooth
consumption patterns.
Some economists would suggest that there is
perhaps less reason to believe that the correlation
of various categories of investment in two countries will be higher over the course of the business cycle than the correlation of, say, the
components of consumption. Technological
innovations in the two countries might make their
investment change in a different manner. Lastly,
there would seem to be little economic presumption that government spending in the two countries should be strongly correlated.
It is, then, of interest to see if these predictions are in accord with the data for Canada and
the United States. Table 6 presents the correlations
of the various aggregates in this country and
Canada.10 The correlation between aggregate output in these two countries is 0.637. However, except
for government spending, none of the subcategories of GNP has a greater degree of correlation.
A very surprising result in Table 6 is that
durable goods consumption and investment in
Federal Reserve Bank of Dallas

Table 6

Correlation of the Components of Aggregate Output
for Canada and United States
Correlation coefficient
Aggregate output
Aggregate consumption
Durable goods
Nondurable goods
Services
Aggregate investment
Producer durable equipment
Nonresidential structures
Residential structures
Aggregate government
Exports
Imports
Inventories

.637
.540
.578
.270
.152
.194
.387
.260
.533
.714
.393
.435
.608

residential construction are two of the components of aggregate spending that are the most
highly correlated in the United States and Canada.
For the reasons described earlier, the components
of aggregate output that are most likely to be
highly correlated across countries—consumption
of services and consumption of nondurable
goods—have the lowest correlations. Durable
goods consumption is more highly correlated than
are the other components of consumption.
Inspection of Tables 1, 5, and 6 reveals that
investment in residential construction in the United
States and Canada is more highly correlated than
it is with the output of their respective countries.
Furthermore, both investment in residential structures and investment in producer durable equipment
in the two countries are more highly correlated
than is the consumption of services. These outcomes are especially surprising in light of the fact
that the amount of trade in goods, services, and
capital between the countries apparently is large
and they have relatively similar economic systems.
In addition, Table 6 shows that the degree
of correlation between government spending in
the two countries is rather high, although there is
no natural economic reason why it should be.
Lastly, the degree of correlation of inventories in
the countries is high as well.
Economic Review — First Quarter 1994

Final remarks
This article has documented exactly how the
various aggregates in the U.S. economy fluctuate
over the course of the business cycle. Some
aggregates increase just before aggregate output
begins to rise, while other variables lag aggregate
output. It has been shown that labor productivity
is a leading indicator of aggregate output and that
these two variables are highly correlated.
The article also shows that the business
cycles observed in Canada and the United States
since 1947 are very similar in many respects.
There is a very strong parallel pattern between the
aggregates in the two countries. This pattern of
behavior does not mean, however, that their
business cycles are coincident or identical. In fact,
although economic theory might predict that
various components of consumption in the two
countries should be highly correlated, the data
appear to indicate the opposite. For example, the
correlation between the consumption of services
and the consumption of nondurable goods is very
low. Future research needs to be done to provide
a greater understanding of this seemingly anomalous behavior.

39

Is There a Trend in Economic Time Series?
To some extent, analysts disagree about
what is meant by “business cycle.” Many would
support the idea that this term should refer to
the fluctuations of output around its trend value.
However, there is also disagreement about
what constitutes the trend value of output.
At one extreme is the view that the trend
level of output grows at some constant rate of,
say, 2.5 percent per year. People who support
this view are said to maintain that output is
trend stationary, with a constant trend growth
rate. Fluctuations of actual output around this
trend are referred to as the business cycle.
At the other extreme is the view that
there is no identifiable constant “trend level of
output” that would allow isolation of the business-cycle fluctuations. This view maintains
that the future trend level is the level of output
that the economy would produce in the future
if output grew at the current level forever. That
is, the economy is always on its trend path,
and there are no deviations from trend. If the
growth rate changed in the future, then the
trend level would also change. This view is
that the only fluctuations to be concerned with
are changes in the growth rate or the trend,
rather than deviations from trend. Researchers using this technique are said to view
economic aggregates as growth or difference
stationary. They might use the term “business-cycle fluctuations” to refer to the changes
in the growth rates of the various economic
time series.
The approach in the article here is somewhere between these two polar views. As

40

Figure 1 shows, the trend (dotted) line is not
a straight line. This means that the trend level
of output grows at a variable rate. This way of
decomposing the growth and fluctuation components of economic time series, suggested
by Hodrick and Prescott (1980), has become
popular in business-cycle research. King and
Rebelo (1993) analyze this filter in detail and
make some comparisons with other detrending
methods. Christiano and Eichenbaum (1990)
discuss whether it can even be determined if
there is a trend in economic time series.
Just as analysts might disagree about
what is meant by “business cycle,” they might
also disagree about what is meant by “recession” and “expansion.” Does “recession” refer
to periods when output is declining and, if so,
declining for how long? Or does the term refer
to when output is below trend and, if so, how
much below which trend? One popular method
for defining these terms is that used by the
National Bureau of Economic Research. The
NBER looks at a broad range of economic
time series that can be said to characterize
aggregate economic activity. It uses these
data to identify peaks in economic activity—
which help to identify the onset of recessions—and troughs—which indicate when
the recessions have ceased. This approach is
described in Zarnowitz (1992). The ancestor
of this research is the original work of Burns
and Mitchell (1946). More recently, Wynne
and Balke (1993) describe similar dating
schemes and use them to identify the asymmetries in the U.S. business cycle.

Federal Reserve Bank of Dallas

References
Backus, David K., Patrick J. Kehoe, and Finn E.
Kydland (1992), “International Real Business
Cycles,” Journal of Political Economy 100
(August): 745–75.

King, Robert G., and Sergio T. Rebelo (1993),
“Low Frequency Filtering and Real Business
Cycles,” Journal of Economic Dynamics and
Control 17 ( January/March): 207–31.

Burns, Arthur F., and Wesley C. Mitchell (1946),
Measuring Business Cycles (New York: National
Bureau of Economic Research).

Koenig, Evan F., and Kenneth M. Emery (1993),
“Why the Composite Index of Leading Indicators Doesn’t Lead,” Federal Reserve Bank of
Dallas Research Paper no. 9318 (Dallas, May).

Christiano, Lawrence J., and Martin Eichenbaum
(1990), “Unit Roots in Real GNP: Do We
Know, and Do We Care?” Carnegie –Rochester
Conference Series on Public Policy 32: 7–61.
Friedman, Milton (1957), A Theory of the Consumption Function (Princeton: Princeton University Press for National Bureau of Economic
Research).
Hodrick, Robert J., and Edward C. Prescott (1980),
“Post-War U.S. Business Cycles: An Empirical
Investigation,” Carnegie Mellon University Discussion Paper no. 451 (Pittsburgh, November).

Economic Review — First Quarter 1994

Lucas, Robert E., Jr. (1987), Models of Business
Cycles (Oxford and New York: Basil Blackwell).
Wynne, Mark A., and Nathan S. Balke (1993),
“Recessions and Recoveries,” Federal Reserve
Bank of Dallas Economic Review, First Quarter, 1–17.
Zarnowitz, Victor (1992), Business Cycles: Theory,
History, Indicators, and Forecasting (Chicago:
University of Chicago Press for National
Bureau of Economic Research).

41