Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/longlargedeclineOOblan
DEWa
31
415
o:
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
THE
LONG AND LARGE DECLINE
U.S.
IN
OUTPUT VOLATILITY
MIT
John Simon, Reserve Bank of Australia
Olivier Blanchard,
Working Paper
01 -29
April 2001
RoomE52-251
50 Memorial Drive
Cambridge, MA 02142
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http://papers.ssm.com/abstract=277356
MASSACHUSEHS
INSTITUTE
OF TECHNOLOGY
AUG 1
5 2001
LIBRARIES
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
THE
LONG AND LARGE DECLINE
U.S.
IN
OUTPUT VOLATILITY
MIT
John Simon, Reserve Bank of Australia
Olivier Blanchard,
Working Paper
01 -29
April 2001
RoomE52-251
50 Memorial Drive
Cambridge, MA 02142
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http://papers.ssm.com/abstract""277356
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
THE
LONG AND LARGE DECLINE
U.S.
IN
OUTPUT VOLATILITY
MIT
John Simon, Reserve Bank of Australia
Olivier Blanchard,
Working Paper
01 -29
April 2001
RoomE52-251
50 Memorial Drive
Cambridge, MA 021 42
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http://papers.ssm.com/paper.taF7abstract
id=XXXXX
The
long and large decline in U.S. output
volatility.
Olivier Blanchard
and John Simon
*
April 20, 2000
*MIT and NBER, and
Reserve Bank of Australia respectively.
and our discussant, Benjamin Friedman,
Ph.D. thesis
[2000].
We
comments. The views
to the
NBER
for
or Reserve
thank the editors
comments. The paper builds on Simon's
thank Robert Solow, and participants
in this
We
in the
MIT
MIT
macro lunch
for
paper are those of the authors and should not be attributed
Bank
of Australia.
Output
volatility
Abstract
The
is
two U.S. expansions have been unusually long. One view
last
that this
is
the result of luck, of an absence of major adverse shocks
over the last twenty years.
We
argue that more
a large underlying decline in output
volatility.
is
at
work, namely
This decline
is
not a
recent development, but rather a steady one, starting in the 1950s,
interrupted in the 1970s and eeirly 1980s, with a return to trend in the
late 1980s
and the 1990s. The standard deviation
of quarterly
growth has declined by a factor of 3 over the period. This
enough to account
We
is
more than
length of expansions.
reach two other conclusions. First, the trend decrease can be
traced to a
ity in
for the increased
output
number
of proximate causes,
from a decrease
government spending early on, to a decrease
in
in the volatil-
consumption and
investment volatility throughout the period, to a change
in
the sign
of the correlation between inventory investment and sales in the last
decade. Second, there
is
a strong relation between movements in out-
put volatility and inflation
volatility.
This association accounts
for the
interruption of the trend decline in output volatility in the 1970s and
early 1980s.
Output
volatility
Since the early 1980s, the U.S.
The
expansions.
and
from 1982 to 1990, lasted 33 quarters. The second,
first,
But
in 1991, is giving signs of faltering.
which started
quarter,
economy has gone through two long
it
is
is
already the longest U.S. expansion on record.
is
that these two long expansions are simply the result of luck,
One view
of an absence of major adverse shocks over the last twenty years.
in this
in
Furthermore, we argue, this decline
volatility.
development
is
We
argue
paper that more has been at work, namely a large underlying decline
output
talent
38th
in its
— but
— the by-product of a
"New Economy"
or of
is
not a recent
Alan Greenspan's
rather a steady one, starting in the 1950s (or earlier, but this
difficult to establish,
because of a lack of consistent data), interrupted in
the 1970s and early 1980s, with a return to trend in the late 1980s and the
1990s. ^
The magnitude
of quarterly output
is
of the decline
is
substantial:
The standard
deviation
growth has declined by a factor of 3 over the period. This
more than enough
to account for the increased length of expansions.
Having established
this fact,
we reach two other
conclusions. First, the
decrease can be traced to a number of proximate causes, from a decrease in
the volatility in government spending early on, to a decrease in consumption
and investment
volatility
throughout the period, to a change
in the sign of
the correlation between inventory investment and sales in the last decade.
Second, there
is
and movements
in
output
a strong relation between movements in output volatility
in inflation volatility.
volatility
is
The
interruption of the trend decline
associated with a large increase in inflation volatility
in the 1970s; the return to trend
is
associated with the decrease in inflation
volatility since then.
'Wliat has happened to output volatiHty over a much longer time span
subject of a well
known debate,
and Balke and Gordon
[1989].
to which
we do not
return. See
Romer
hcis
[19S6],
been the
Weir
[19S6],
Output
volatility
Our paper
namely the decrease
in
output
GDP, and show
process for
We start
volatility. We
organized as follows.
is
by presenting our basic
fact,
then look at the stochastic
that the decrease in volatility can be traced
primarily to a decrease in the standard deviation of output shocks, rather
than to a change
in the
dynamics of output.
we show how
Finally,
this
decrease in the standard deviation of innovations accounts for the increase
in the length of expansions.
We
then take up the question of whether recessions are special, in a
the formalization used earlier does not do justice
take up the ciuestion of whether what
years
this
is
is
Put another way, we
we have seen over the
simply the absence of large shocks and nothing more.
not the case.
The measured decrease
do with the absence of large shocks
We
to.
in
last
We
twenty
show that
output volatility has
little
is
level of inflation,
volatilities
We
a strong relation both between output volatility and the
and between output
went up
in the 1970s,
factors,
volatility
and
inflation volatility.
and have come down
does not, however, imply causality.
due to third
to
in the recent past.
then turn to the relation between output volatility and inflation.
show that there
way
Correlation
since.
Both evolutions may
for
Both
example be
such as supply shocks in the 1970s. This leads us to
consider the evidence from the
G7
countries.
Our motivation
is
that the
different timings of disinflation across the countries can help separate out
the effects of inflation from those of supply shocks.
We
first
show that
these other countries have also experienced a decline in output volatility,
although with some differences
exception
is
output
timing and
in
magnitude.
An
interesting
Japan, where a decline in output volatility has been reversed
since the late 1980s.
time fixed
in
We
then show that, even after controlling for
effects, inflation volatility still
common
appears to be strongly related to
volatility.
As a matter
of accounting, the decline in output volatility can be traced
either to changes in its composition, or to changes in the variance
and
co-
Output
volatility
variances of
its
underlying components. With this motivation, we look at
the components of
tion at which
GDP. We
find that, at least at the level of disaggrega-
we operate, changes
in
composition explain essentially none of
the trend decline. Composition has changed, but the effects of the various
We
changes have mostly cancelled each other.
also find that, apart
from
a decrease in the volatility of government spending early on in the period,
most
of the decrease in output volatility can
consumption and investment
volatility,
be traced to a decrease
in
both
and more recently to a change
in
inventory behavior, with inventory investment becoming more countercyclical.
We
it is
conclude by discussing the agenda for further research. In particular,
clear that
we have only gotten
The deeper
decline.
to the proximate causes of the volatility
causes, from changes in financial markets to better
counter-cyclical policy, remain to be identified.
1
The
Figure
real
decline in output volatility
shows the evolution of the rolling standard deviation of quarterly
1
output growth (measured at a quarterly
sure of output
chain-weighted
is
GDP. We
rate), since 1952:1.
use a
so the standard deviation reported for quarter
deviation over quarters
weighted
GDP
is
figure
—
19 to
1947:1, so the
of the growth rate
The
t
is
shows a
The
The
first
first
is
of 20 quarters,
the estimated standard
available observation for chain-
observation for the standard deviation
1952:1.
clear decline in the standard deviation over time,
from about 1.5% per quarter
late 1990s.
t.
t
window
The mea-
decline
is
in the early 1950s to less
than 0.5%
in the
not continuous: Volatility increases from the late
1960s to the mid-1980s, followed by a sharp decline in the second half of the
1980s.
One may
think of other ways of measuring volatility.
An
alternative
is
to
o
C
J
1
o
Z3
O
o
D
>
O
O
D
O
^ C
o
if)
a^
c
O^
1
o
fY
R-
9'
L
V
6'
L
0'
L
9'0
^'0
^0
Output
volatility
look at the standard deviation of the deviation of the level of (the logarithm
of)
is
output from trend,
example, from a Hodrick-Prescott
for
to look at annual rather than quarterly changes in
tives yield very similar conclusions.
The
filter.^
Another
GDP. These
basic reason
is
alterna-
that the standard
deviation of quarterly output gTowth reflects primarily the high frequency
properties of the series, which are largely invariant to the detrending method.
Changes
The
output process
in the
natural next step
movements over
time,
is
to think
and ask how
it
about the process generating output
has changed. Does the lower volatility
of output reflect a lower standard deviation of output shocks, or a change
in the
dynamic process through which these shocks
More
concretely,
affect output, or
both?
assume that output growth follows an autoregressive
process given by:
(Ayt
where
yt
difference,
g
-
=
g)
-
a{L){Ayt-i
denotes the logarithm of output
is
The standard
(Tf,
and a{L)
deviation of output
cxy
is
put growth follows a
first
be AR(1); while
^This
is
is
t,
this
A
is
denotes a
first
a white noise
a lag polynomial.
If
a{L)
—
o, for
example, out-
=
a^j\J\
—
a^,
the standard deviation of output.
points in mind,
1947:1 on, again with a
for
quarter
(1.1)
order autoregressive process and ay
so the higher a, the higher
makes
CYt
then depends both on the standard
deviation a^ and on the lag polynomial a{L).
it
in
+
the underlying growth rate of output, eyt
shock with standard deviation
With these
g)
we estimate
window
(1.1) over a rolling
of 20 quarters.
We
sample from
assume the process
to
does not fully capture the dynamics of output growth,
an easier interpretation of the changes
the approach taken by Taylor [2000].
in the process,
and
all
Output
volatility
of our conclusions extend to a higher order
Figure
The top panel
2.
gives the estimated
gives the estimated
AR(1)
coefficient.
AR. The
growth
results are plotted in
The bottom panel
standard deviation of the shock. The other two
The middle panel
rate.
lines in
gives the estimated
each panel give two-
standard-deviation bands on each side of the estimate.
The
conclusions from Figure 2 are straightforward. Neither the growth
show
rate nor the
AR(1)
coefficient
slightly lower at the
is
but the difference
coefficient
is
movements over
clear
end
not significant.'^
time.
The AR(1)
of 1990s than in the rest of the sample,
The standard
deviation of the output
shock shows the same evolution as the standard deviation of output growth.
when
Indeed,
plotted in the
same graph,
their evolutions
appear nearly
identical.
In short, the decrease in output volatility appears to
come from smaller
shocks, rather than from a decrease in persistence of the effects of these
shocks on output.
Back to the length of expansions
Having estimated the process
for
To do
issue of the length of expansions.
We
output growth, we can return to the
so,
we proceed
in
two
steps.
estimate two processes, one for the period 1947:1 to 1981:4, the other
We
for the period 1982:1 to 2000:4.
choose the
split
date to coincide with
the peak of the cycle preceding the last two expansions.
The
intent of the
capture the major changes in the process between the start and
split is to
the end of the sample.
^A perhaps obvious
point:
Changes
in
the estimated AR(1) coefficient for the univariate
representation of output growth do not imply a change in the dynamic structure of the
economy.
dynamic
used
Output movements come from many underlying shocks, each with
effects.
At
different times, different shocks
for estimation, leading to different
may dominate
its
own
the (short) subsample
estimated univariate dynamics of output.
Roiling
Figure 2
mean growth
rote
Rolling ar(l) coefficient
Rolling
standard deviation
of
residual
t
,
J
\
4_/-^H
/
/
J
^
N
\
—
*v
^^
_^^
Output
volatility
The two estimated equations
are given by;
47:1 to 81:4
{Ayt
-
0.S7)
=
0.31{Ayt^i
-
0.87)
+
eyt
ct^
=
1.12
82:4 to 00:4
(Ayt
-
0.85)
=
0.48(Ayf_i
-
0.85)
+
eyt
ct^
=
0.56
Using the
estimated equation, we generate a sequence of 100,000
first
observations for output growth, b;Tsed on draws of the shocks from a nor-
mal
distribution. Following a long tradition of approximating
by a simple
rule,
we
NBER dating
define the beginning of a recession as two consecutive
quarters of negative growth, and the beginning of an expansion as two consecutive quarters of positive growth following a recession.
mean and median
tions.
We
The
for the
We
compute the
lengths of expansions in the sample of 100,000 observa-
then do the same using the second estimated equation.'*
shown
results are
in
Table
1.
mean and median expansion
35 quarters
length for the
are 17 and 13 quarters
first
subsample; 51 and
second subsample. These means compare to an actual
for the
mean expansion
The estimates
length of 19 cjuarters for 1950:1 to 1981:4 (with recessions
being defined with the same rule as
and of 36 quarters
in
the simulation, not by
for 1982:1 to 2000:4 (but
NBER dating),
with the second expansion not
having ended yet). In other words, the differences between the two estimated
''Note that the average length of expansions
lying parameters of the
starts;
AR
process.
By
is
a very non-linear function of the under-
construction, an expansion ends
when a
recession
under our definition of a recession, this requires two consecutive quarters of nega-
tive growth.
rate, the
The
probability of such an event depends non linearly on the average growth
standard deviation of the residual, and the
below the average growth
standard deviation
is
deviation will have
little effect
far
AR
coefficient.
rate, small
on the probability of a
recession.
If,
changes
for
example, the
in the
standard
If it is closer,
the same
small changes will have a substantial impact on the probability of a recession, and in turn
on the length of expansions.
Output
volatility
autoregressive processes account well for the increase in expansion length
from the
to the second sample.
first
To show which parameter changes
are responsible for this increase,
we
then show the results of switching either the mean, or the AR(1) coefficient,
or the standard deviation of the shocks, across the two samples.
Not
the two samples, switching
AR
Switching the
persistent,
them has no
effect
making
on the length of expansions.
effect of
it
more
a negative shock on output growth
likely that
output
will decrease for
a row), longer in the second subsample. But nearly
from switching standard deviations.
the same as in the
now be only
first
If
all
subsample, the
we have
expansions are likely to be
much
now more
two quarters
the action comes
mean
length of expansions would
14 quarters, the median 11 quarters. In short,
the two long expansions
is
the standard deviation had remained
decrease in the standard deviation of output shocks which
just experienced.
And
is
it is
the large
at the root of
unless this changes,
longer in the future than they were in the
past.
Table
1.
Mean and median
Simulated
(Actual)
length of expansions (quarters)
1947:1 to 1981:4
1982:1 to 2000:4
mean/median
mean/median
17/13
51/35
(19/15)
(36/-)
17/12
55/38
15/11
83/55
99/67
14/11
Switching:
Growth
AR
a',s
in
coefficients leads to shorter expansions in the first sub-
sample (because the
in
growth rates are nearly the same
surprisingly, given that the
rates
coefficients
Output
10
volatility
Are recessions
2
There
is
special?
a widespread view of recessions and of output volatility under which
the estimation and the exercise
we
carried out in the previous section
According to that view,
largely tautological at best, misleading at worst.
cessions are largely the result of infrequent large shocks
large
and
identifiable that they often
have names: the
shocks, the Volcker disinflation and so on.
dominate output
and there
volatility,
decline in output volatility:
We just
is
is
re-
— indeed sufficiently
and second
first
oil
Under that view, these shocks
no great mystery
in
the measured
have not had large shocks over the
last
two decades.^
To
see whether this
In the
approaches.^
first,
our measure of volatility
In the second,
and
indeed what has been going on, we explore two
is
we look
if
we look
what happens
we simply exclude
The measured
to the evolution of
recessions from the sample.
for signs of large shocks,
kurtosis, in the relevant distributions.
conclusions.
and associated skewness
Both approaches
decline in output volatility
absence of large shocks over the
is
at
last
twenty years.
What
is
it
yield similar
not due to the
captures instead
the decline in the volatility of "routine" quarter-to-quarter changes in
GDP
If
growth.
the decline in volatility simply reflected the absence of large negative
shocks and associated recessions, excluding recessions would then eliminate
our findings.
Indeed, excluding recessions from the sample
strong a correction under our null hypothesis.
It
is
clearly too
corresponds to eliminating
^This view was forcefully communicated to us by the editors of the Panel at the start
of this project.
^recessions often
come with unusually
large negative realizations of the shocks also
follows from the definition of the recession,
and the
fact that the distribution of shocks,
conditional on being in a recession, implies a higher probability of large negative shocks.
Output
11
volatility
large negative realizations just because they
But,
this overly strong correction still
if
makes
And
for convincing evidence.
we
this approach,
but allowing
same
presence of a
dummy
each of the quarters of an
for the
NBER
shows a decrease
rolling regression as
line),
The
dated recession.
is
in volatility,
it
To implement
we did
earlier,
resulting time series
plotted in Figure 3 (as
together, for ease of compari.son, with the standard deviation
obtained without recession dummies (the series shown
and plotted here as the dotted
The
and negative.
variable taking the value of one in
estimated standard deviation of the residual
the solid
large
this is indeed the case.
re-estimate the
for the
happen to be
results are quite clear.
(by construction).
in
Figure
1 earlier,
line).'
Output
volatility
But the general evolution
is
is
indeed lower in recessions
very similar, with a clear
trend downwards, from roughly 1.2 at the start of the sample, to 0.4 at the
end.
The
other approach
is
to actually look for signs of infrequent, large,
shocks. For example, under the hypothesis that the
economy
subject to
is
two types of shocks, frequent and small on the one hand, infrequent and
large
on the other, we would expect the distribution of output growth to
exhibit either skewness
or kurtosis
(if
large infrequent shocks are equally likely to be positive or
(if
negative), or both. Other,
implications.
more complex, models
of recessions have similar
While, by the Wold representation theorem, we
even these models are
(1.1),
large infrequent shocks are typically negative),
still
know
that
consistent with the linear representation given by
the residuals are likely also to exhibit either skewness or kurtosis.*
^Introducing a
mean growth
dummy
for recessions
can be thought, of a way of allowing
rate in recessions. In this sense, this estimation
switching process estimated by Hamilton [1989] for U.S.
^Take, for example, the idea that
is
in the spirit of
for a
lower
the Markov-
GDP.
serial correlation of
output
is
higher in expansions
if)
0)
Z5
C
o
CO
CO
(D
O
CD
Z5
O
C
O
O
>
(D
X)
X5
^^
D
X3
C
O
CO
c
o
(D
D
9'
L
V
L
2'
L
0"
I
tiuinmiii^iiumiuc
Output
12
volatility
This suggests looking at the evokition of skewness and kurtosis of
residual obtained from estimation of (1.1).
Each point represents the estimate
4.
The
results are
shown
in
e,
the
Figure
of skewness (top panel) or excess
kurtosis (bottom panel) of the residual from estimation of an
AR(1) process
over the current and previous 19 quarters. Each of the two panels also gives
the standard
The two
95%
confidence band for the hypothesis that each equals zero.
panels yield similar conclusions. Except for a brief period during
the 1980 recession, there
kurtosis.^
is little
evidence of either significant skewness or
^^
than in recessions, clearly a non linear feature. Capture
this
by the assumption that output
a two-state Markov process, with a high probability that output growth remains
growth
is
high
if
initially high,
low.
It is
and a low probability that output growth remains low
easy to show that this
Markov process
will
if
initially
have an AR(1) representation given
by (11), with the distribution of the residual skewed so that most residuals are small and
positive but with a long negative tail associated with recessions.
^
Under the assumption that large shocks are indeed infrequent, the use of a short
window
(20 quarters) implies that there are
occurs.
Those subsamples
expect
many
or
will
many subsamples during which no
large shock
not show evidence of skewness or kurtosis. But we would
most recessions to be associated with me;isured skewness or
kurtosis. This
does not appear to be the case. Nor do we see more evidence of skewness or kurtosis
we use a
longer window.
Over the sample treated
as a whole, there
of significant kurtosis, but this appears simply to be
is
due to the decrease
if
indeed evidence
in the
standard
deviation over time (The distribution of draws from a set of normal distributions with
difTerent variances will exhibit kurtosis.)
'"Other evidence that skewness and kurtosis are not important here
is
that the expansion
length simulation results presented in the previous section are roughly mialTcctcd
draw shocks by sampling with replacement from the estimated
if
we
residuals rather than from
Figure 4
Roiling regression
1950
1950
1970
Rolling
I960
1970
1980
skewness
1990
2000
2010
2000
2010
regression excess kurtosis
1980
1990
Output
We
[1986],
13
volatility
have explored other approaches. Following Blanchard and Watson
we estimated a
specification in
which the output shock
is
assumed to
be the sum of two underlying shocks, one drawn every period from a normal
with probability
distribution, the other equal to
(1
— p), and drawn from a
normal distribution with larger variance, with probability
p was equal to
reject the hypothesis that
of a decrease in
p.
and could not
zero,
p over time. In other words, we could not
We
could not
find evidence
find evidence that
the decrease in output volatility has been due to a decrease in the likelihood
of large shocks over time.^^
3
Output
volatility
and
Having estabhshed the basic
are at least two
1
inflation
fact,
ways to look
we now turn
to
at the evolution of
its
interpretation. There
output
volatility in Figure
— or equivalently, at the evolution of the standard deviation of the residual
in Figure 2, as the
The
first,
two are nearly
identical.
which we have implicitly
relied
on
until now,
is
to see the
evolution as a trend decline, temporarily interrupted in the 1970s and early
This interpretation
1980s.
5,
is
shown graphically
in the
top panel of Figure
which reproduces the evolution of the standard deviation of output from
Figure
1,
and draws
in addition
an estimated exponential trend over the
period.
The
second, which has been suggested in a
particular
McConnell and Perez-Quiros
a normal distribution
"One
hypothesis
quarters,
exercise
— as we did
is
are nearly identical
is,
of recent papers (in
instead, of a step decrease
our stochastic simulations
earlier.
that there are large shocks, but their effects appear over a
making them more
would be very
in
[2000])
number
difficult to detect.
different
if
If
this
we were to use lower
when using annual
few-
were the case, the results of our
freqxiency data. In fact, the results
rather than quarterly data.
Figure
5:
Trend decline or break
in
volatility?
Log linear time trend
?950
1960
2000
1970
Break
in
2010
1986:1
?nin
Output
volatility
some time
in the
14
in the early to
mid-1980s. This interpretation
bottom panel of Figure
can also
fit
5,
the assumption of a step decline in 1986:1.
et al,
shown graphically
which shows how an estimated step function
The
the general evolution of volatility.
work of McConnell
is
step function
The more
is
drawn on
careful econometric
estimating rather than a^isuming the break date,
finds a slightly earlier date, 1984:1, as the
most hkely break
This second interpretation suggests looking for factors
point. ^^
in
the economic
environment which changed around the mid-1980s. Plausible candidates are
an improvement
by Taylor
by Kahn
[2000])
et al.
Under the
likely to
in
,
the conduct of monetary policy (as argued for example
or changes in inventory behavior (as argued for
example
[2001].)
interpretation however, which
first
be the right one, one needs to look
for
we
two
shall argue
is
sets of factors.
more
First,
^'^
the factors behind the underlying trend decline over the last 50 years.
Second, the factors behind the interruption of the trend in the 1970s. Put
another way, the focus shifts from what happened in the 1980s (to explain
the step decline in volatility) to what happened in the 1970s (to explain
the interruption of the trend for a bit
interpretation
Inflation
we
prefer,
and
and the route we follow
different
by major increases
aflFected
is
the
in the rest of the paper.
inflation volatility
That the 1970s were
was
more than a decade). This
is
in
not very controversial.
The
U.S.
economy
the prices of raw materials, including
oil.
Inflation increased, to return to a lower level only after the disinflation of
'^The difference conies from our use of a rolling window to capture
in 1984:1 will
of our
A
decline
not necessarily show up until enough earlier observations have dropped out
window
— something that happened around
'^Or indeed over the past century,
from
volatility.
earlier research
on
if
1986:1.
one takes a longer view, informed by the evidence
volatility since the late 1800s.
Output
15
volatility
the early 1980s. That these shocks, and perhaps inflation
led to
more output
top panel plots output volatility (dotted
of the inflation rate (solid hne)
GDP deflator.
have
volatility.
The
does not seem implausible.
volatility
Figure 6 shows the relation between inflation and output
mean
may
itself,
The bottom panel
inflation volatility (solid line),
line) against the
— with
inflation
20-quarter rolling
measured using the
plots output volatility (dotted line) against
both constructed as 20-quarter rolling stan-
dard deviations. (All variables, including mean
inflation, are
measured
at
quarterly rates).
The temporary
is
increase in output volatility in the 1970s and early 1980s
temporary increase with the
clearly correlated with the
Output
volatility
level of inflation.
seems however more strongly related to the
volatility
than
to the level of inflation. Simple regressions of output volatility on the level of
and an exponential time trend show
inflation, inflation variability
factors to be significant, with the level
roughly quantitatively similar
important and
roles,
and the
all
three
variability of inflation playing
and the negative time trend remaining
'^
significant.'^
Correlation between inflation and output volatility does not imply how-
ever causality from inflation to output volatility.
that the correlation reflects a
alternative
is
and output
volatility
'"One worry
is
At
least
one plausible
common dependence
of inflation
on third factors, such as the supply shocks of the 1970s.
that measurement noise in the decomposition of nominal
GDP will create
a spurious positive correlation between output and inflation volatility. But the results are
very similar
if
we use the CPl, where the
'^Onc problem with such regressions
tions
and means both on the
appropriate
GARCH
model
be a function of inflation
results.
left
for
is
issue
is
likely to
be
less
important.
the use of moving averages for standard devia-
and right hand
sides.
Estimation of a potentially more
output growth, allowing the variance of output shocks to
volatility,
the inflation
level,
and a time trend yields very similar
Figure
6:
Volatility of
standard deviation
of
output and inflation
output and
mean
inflation
Output
16
volatility
Here, international evidence can help:
First,
and obviously,
resentative of
can
it
what happened
tween output and inflation
to
tell
us whether the
to output volatility,
volatility elsewhere.
US
evolutions are rep-
and to the
But
relation be-
we
also, if
are willing
assume that the supply shocks of the 1970s were largely common across
countries, then
them
gives us a
it
way
of controlling for their presence by treating
as fixed eflFects in a cross country panel regression.
In other words,
such a regression can help us establish the relation between output and
flation volatility, controlling for
G7
turn to the evidence from the
A
look at the other
supply shocks. With this
G7
mind, we now
in
countries.
countries
G7
Figure 7 shows the evolution of output volatility for the
For
clarity,
we have grouped the
countries in three panels.
includes the United States, the United
Italy.
in
1960 (1982
The top panel
The bottom
Japan. In each case, the measure of output volatility
deviation of output growth, using a
countries.
Kingdom, and Canada. The middle
panel includes West Germany, France, and
data we have start only
in-
window
is
panel shows
the rolling standard
of 20 quarters.
Because the
for Italy), the different
measures are
available only from 1965:1 on (1987:1 for Italy), a shorter sample than the
one used
The
for the
United States above.
top and the middle panels show that six of the
had roughly similar evolutions over the period. In
all
G7
countries, the standard
deviation of output has declined, from a range of 1.5% (for
little
below 1.0%
(for
is
indeed
how
One
closely,
to a
of the striking characteristics of these
common, long
The
all
two
similar the standard deviation of output growth
across these countries today.
the presence of
Germany)
the U.S.) in the early 1960s, to around 0.5% for
countries in the late 1990s.
panels
countries have
is
general decline and convergence suggest
lasting, forces across countries.
Looking more
however, there are also clear differences across countries, especially
'
"
Figure 7
%
1.5
1.0
0.5
%
/
West Germany
\
1
1.5
1.5
t
*^ 1
i"*
1.0
_^—
1.0
Italy
'i,
W.-*"^
'.I
0.5
l/'^-w.,*^^.
0.5
France
%
o/
/o
1.5
1.5
1.0
1.0
0.5
0.5
Q Q
'
I
I
I
1965 1970
I
I
I
I
I
1975
I
I
I
I
I
1980
I
I
I
I
I
1985
I
I
I
I
I
1990
I
I
I
I
I
I
I
I
I
Q Q
1995 2000
Output
17
volatility
After the general increase of the early 1970s, the decrease in
in timing.
taken place earlier in Germany, later in Canada.
volatility has
The only G7 country
bottom
in the
to have a clearly different evolution
panel. After a decrease
Japan, shown
is
from the early 1960s to
late 1980s, the
standard deviation of output has increased in the 1990s, and
than
it
To the extent that
at the start of the sample.
was
slump of the
largely coincides with the long Japanese
to search in that direction for an explanation.
by both firms and consumers
liquid assets
of cash flows on
effects of
is
1990s,
now higher
this increase
it is
tempting
For example, a decrease in
may have
led to stronger effects
consumption and investment, leading to stronger multiplier
shocks on output. The floor on interest rates
We
monetary policy responses.
may have
constrained
have not explored these hypotheses further
paper, but we find the coincidence of the long slump and the increase
in this
in volatility intriguing,
and potentially useful
in learning
what has happened
in other countries.
Leaving Japan aside and focusing on the other six countries, we return
to the relation
between output
volatility
and
inflation.
To do
so,
we rim the
following panel regression;
^yit
where
effects,
the
of output
If
+ /?t +
refers to country,
fit
and
and
t
ai7r,j
and
tt is
(3.1)
e,i
to time, so the j3iS are country fixed
are time fixed effects, Oy and
inflation,
+ C2cr^., +
a rolling
a-„
are rolling standard deviations
mean
of the inflation rate.
the effects of the supply shocks of the 1970s on output volatility were
indeed
lation
i
/3,
common
across countries, then this specification will give us the re-
between output and
The assumption
is
inflation volatility controlling for
probably too strong however: The
supply shocks.
effects of
were different across countries, and these differences
may
supply shocks
well have been
associated with different evolutions of both the level and the volatility in-
Output
18
volatility
flation. In this case,
the coefficient on inflation will
up some
pick
still
of the
supply shocks. Nevertheless, even in this case, this cross country
effects of
specification
an improvement over the U.S. regression
is
which we could
(in
^^
not introduce fixed time effects) presented earlier.
Estimation yields coefficients of ao
a\
=
best
way
The top
for the
is
is
inflation volatility, rather
summarize the implications of the regression
to
which are given
is
through a
in the three panels of Figure 8:
panel gives the actual and the fitted values of output volatility
United States (dropping
changed by
this).
specification does a
The
it
which appears to matter, and to matter strongly.
level of inflation
set of graphs,
•
0.67 with a t-statistic of 13.7, and
—0.02, with a t-statistic of -0.7. Thus,
than the
The
~
The
good job
in inflation,
from the panel
regTe.ssion
conclusion to be drawn
other two panels show
movements
Tiit
is
— nothing
that the panel
of fitting the U.S. evolution.
how much
of the evolution
and how much
is
due to the
comes from
common
time
components:
•
The second panel
volatility (repeated
gives the evolution of the fitted value of output
from the
of the inflation component,
first
panel), together with the evolution
a20"7r,(.
The panel makes
clear that the
increase in inflation volatility accounts for the reversal in trend from
the early 1970s to the early 1980s.
'^Note that, even under the assumption of common supply shocks, this specification does
not resolve other potential identification problems, such as the possibility that the relation
between output and
inflation volatility reflects causality
from output to inflation
volatility
(through the response of monetary policy), or a dependence on other third causes, such as
an improvement
volatility.
in
the conduct of monetary policy leading to lower output and inflation
Figure 8
o/
/o
Actual and
fitted
values of
/o
a.
1.0
1.0
^-vA.
0.5
0.5
a.
o/
/o
%
component
Inflation
1.0
1.0
0.5
0.5
%
Common
o/
/o
time (G6) component
1.0
1.0
0.5
0.5
Q Q
I
I
I
'l965
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
Q Q
"
1970
1975
1980
1985
1990
1995 2000
Output
•
19
volatility
The
third panel gives the evolution of the fitted value of output volatil-
ity (again
of the
repeated from the
first
common time component,
panel), together with the evolution
(3f
This suggests a steady underlying
trend decrease from 1960 on.
In short, this decomposition suggests a trend decrease in volatility, tem-
porarily interrupted by an increase in inflation volatility.
Under that
inter-
pretation, the sharp decline in output volatility in the 1980s appears to be
mostly the result of a sharp decline in inflation
To
volatility.
get a better understanding of both the trend decrease and the tem-
porary reversal
in
output
volatility,
the last section goes one level down, and
looks at the evolution of the individual components of
4
A
GDP.
look at the components
In his 1960 Presidential address to the American Economic Association,
Arthur Burns (Burns [I960]) argued that a trend decline in output volatihty
was indeed under way. Composition
improvements
in capital
smooth consumption
effects
and the steady
shift to services,
markets and the increasing ability of consumers to
in the face of variations in income, the increase in the
income tax and stronger automatic
stabilizers, all led,
and, he argued, would
continue to lead, to more economic stability.
He was
clearly right
This section makes a
about the trend.
first
Was
pass at the answer.
he right about the channels?
From a
statistical
accounting
point of view, one can think of the volatility of output as depending on three
sets of factors, the volatility of its
relative weight.
We
Volatility of
look at
all
components, their co-variation, and their
three in tmn.
output components
Take the standard decomposition of
GDP
by type of purchase and type
of purchaser, consumption, investment, government spending, net exports
Output
20
volatility
and inventory investment. Let each of these components,
denoted by Xi
in real terms,
be
so:
For each of the components, we consider two measures of volatihty:
The
GDP
the same as for
first is
little
namely the
rolling
standard
Xn, which we denote by Gxn- This measure
deviation of the rate of growth of
makes
earher,
sense however for inventories and net exports v/hich change signs
and are frequently close to
Thus, we construct the rolling standard
zero.
deviation of growth for consumption, investment, and government spending
only.^'
The second measures the
volatility of
a variable commonly called the
"growth contribution" of each component, which adjusts
the component in
GDP. A very
overall output volatility
able
is
if it
volatile
component may have
rolhng standard deviation. Note that this measure
GDP
little effect
is
on
vari-
once again the
well defined for
all
com-
regardless of whether they change sign or arc close to zero.
Note
also that the variable
so,
the share
if
is
the share of
GDP. The
accounts for a small share of
defined as /\Xit/Yt^\, and our measure of volatility
ponents of
for
can be rewritten as {^X^t/Xit-\){X^t-\/yt-\),
stable at high frequency, the standard deviation will be
is
roughly equal to the share of the component of
GDP,
times the standard
deviation of the component's growth rate. For both volatility measures, the
window we use
to
compute standard deviations
change are quarterly, not annual
'
'
In parallel with
component.
The
20 quarters.
is
GDP, we have estimated
the
same
as for
GDP.
decrease in volatility comes from the decrease in the volatility of
change
in
dynamics.
We
Rates of
rates.
our exploration of
general conclusion
is
do not present the
results here.
AR
processes for each
For the most part, the
tlie
shocks than from a
Output
The
21
volatility
given by the solid
by the dotted
move
line,
line,
and the
scale
is
on the
with the scale on the right
left axis.
axis.
first
measure
The second
is
given
is
Note that the two
lines
largely together at high frequency, reflecting the stability of the shares.
From Figure
9,
• Volatility of
we draw the
following conclusions:
government spending (and of
very high during the Korean war.
and has remained low ever
•
The
evolutions are plotted in Figures 9 and 10.
There
is
no
It
fiscal policy in general)
rapidly went
down
was
in the 1950s,
since.
clear trend in the volatility of net exports.
There
is
also
no
clear trend in the volatility of inventory investment, although volatihty
has been low in the 1990s (this together with the change in correlation
shown below, suggests a recent change
in the behavior of inventory
investment.)
•
Most
of the decrease in overall volatility can
be traced to a decrease
in
the volatility of consumption and investment. After a large decrease in
the 1950s, consumption has continued to decrease, from about 0.75%
in the
1960s to 0.30% in the late 1990s.
volatility has
measure
is
The
decrease in investment
been more limited. The standard deviation of our second
nearly the
same
in the late
1990s as
it
was
in the 1960s.
Given that much of the action comes from consumption and investment.
Figure 10 goes one further step
of
down and shows the evolution
of the volatility
consumption and investment components.
•
Relative declines in the volatility of
tion,
all
three components of consump-
spending on durables, on non-durables, and on services, are
roughly similar.
Their timing
is
slightly different, with
much
of the
trend reversal in consumption in the 1970s and early 1980s coming
from consumption of
services.
1
1
1
1
1
r>
-r-
1
1
'\
^
1
r
r
)
^
-
.
:
S
s
X
Q
X
>^
Q
'-
V
^
-
r
^
Ol
J
c
o
c
a
'-
s
1
3
p
J
0)
(A
>
!
1
\
V,
s
5
O
>
'
-
:
<
>
:
I
1
r
cz
^->^
-
^
1
(0
.
1
1
1
1
1
eg
1
>-
1
^
>
T3
C
3
U)
OJ
T-
1
1
1
1
~/
a>
o
X
o
.^
- - '
.--
c
r^
^ — —
2
"^~~—-,
1
1
ii)
—
-^
<u
w
C
o
c
*-»
c
y
^
.j>
.2
»-
•-
CM
^
'-
J
~>""^::=^
T
J
_
'^- "C-.
•-^
j^(
0)
JE
^^;5^
LJ. ^
en
0)
.
>
.^^^^
__II7
C
J>
^
(/)
/
J
^~^^~^
"~
^
/»
o
_
^
^»
1
1
1
1
1
1
1
•-
<-
1
1
>
<D
T3
(N
05
1
1
1
1
1
1
1
1
1
^
i.
-.
^-"^
L
1
t
K
L
>
\
2__^
y
m
C
Q)
U)
0)
T3
r
p*
J"
c
-^
o
J
_j
D.
L
o
Xy
Q.
t,
E
3
W
o
/
1
!=.
1
—
*
\,
^^
:
>
/
^
^e^
—
^^
^
^
^
C
3
c
o
o
r^
?
"
r-
-^
r--
E
~Ilii7
c
o
C;^
r
o
c
"
'
(
13
3
a
O)
3
<
^
i
^
re
^m
^
^ ~/~^
->
X
n
5=^=.
^
^
1
1
1
1
^
^
^__^
___j
J
^
C
1
\
\
05
O
»-
1
1
1
1
1
1
1
f
CNJ
1—
Output
22
volatility
We
think these are slightly surprising iindings.
One might have
ex-
pected improvements in financial markets to lead consumers to choose
a smoother consumption path, thus leading to smoother consumption
and non durables. But one would
of services
also have expected
an
improved abihty to borrow and lend to lead to a stronger stock-flow
adjustment
for
purchases of durables, and thus potentially to more
volatility of durable purchases.
This does not seem to be the case in
the data.
•
The two
series for
Non
tions.
investment volatihty exhibit quite different evolu-
residential investment
shows a steady decline, and a limited
increase in the 1970s. Residential im'estment shows a steady increase
from the 1950s to the mid 1980s, and a sharp decrease after that. The
sharp decrease corresponds to the elimination of interest rate ceilings
on savings and loan institutions (the end of Regulation Q), making
it
a plausible candidate explanation.
Correlations of output components
The standard
viations of
its
deviation of output depends not only on the standard de-
components, but also on their correlations. To show what has
happened, we construct the correlation of each component with
(i.e
GDP
minus inventory investment)
AXit/Yt-i with ASt/Yt-\, where St
final sales
— more specifically the correlation of
is final sales.
Again, we use a window
The
correlations change over
of 20 quarters.
These evolutions are shown
time (as we would expect
if
in
Figure
11.
the subsamples are dominated by different
shocks, with different implications for the correlation between each com-
ponent and
trends.
final sales).
The exception
sales. Until the
is
But, except for one
series,
they do not show clear
the correlation between inventory investment and
mid-1980s, inventory investment tended to
move with
sales,
if)
Q
o
o
0)
3
Li-
Output
23
'
volatility
leading to a higher variance of production than of sales
— a fact
studied at
length by the research on inventory behavior. Since the mid 1980s, inventory
investment has become countercyclical, leading to a decline in the variance
of output relative to sales.
et al.
[2001],
methods
This
must have come from a change
of firms.
It is
in
the inventory management
clearly one of the factors
volatility in the 1980s,
output
which has been examined by Kahn
fact,
although only the
behind the decrease
last
one
in
in a long series of
structural changes.^*
Composition
efFects
The composition
GDP
has changed substantially over the
The main three changes
years.
we
of
last fifty
(at the level of disaggregation at
which
are looking) are the increase in the share of (high volatility) fixed non
residential investment,
from 9.6%
in
1950 to 13.8% in 2000, the decrease
non durables consumption, from 36.9% to 20.1%, and the
in the share of
mirror increase in the share of (low volatility) consumption of services, from
20.8%
To
to 39.3%.
characterize the effects of composition on output volatility, a simple
approach
is
to
compute the
volatility of a counterfactual series for
output
growth, using 1947 shares rather than current shares to weigh components.
More
specifically,
follows.
"^There
we construct the counterfactual output growth
Write the rate of growth of output
is
a puzzle here as well:
with the introduction of "just
have
and the
roughly coincides roughly
management methods. These methods
not clear however
why
they should have
change from positive to negative: Better tracking and forecasting of
ability to
less procyclical
is
as:
in correlation
time" inventory
led to lower inventory-to-sales ratios. It
led the correlation to
sales,
in
The change
series as
maintain a stable inventory to sales ratio should lead to more not
inventory investment...
Output
volatility
24
(AYt/Yt^r)
=
^ AX,,/y,_i + J2AX,t/Yt-,
i
where the terms
tive,
and the terms
in the first
in the
J
sum
second
are the terms which are always posi-
sum
are the terms which change sign in
We
the sample (net exports, and inventory investment).
expression
as:
[AYt/Yt-i)
where Oit-i
= ^a,t-iAXu/Xu^i + Y,AX,t/Yt-i
the share of component
is
constructing AXjt/Xjt-i does not
We
at time
i
make much
exports as they are frequently around zero.
separately.
can rewrite this
i
—
1.
Once
again,
sense for inventories and net
Consequently, we treat them
then construct the counterfactual series for output growth
as:
{AYt/Yt-iT
= ^a,i947AX,,/X„_i +
I
where 0^947
is
j
the 1947 share of component
standard deviations
for actual
figure, as the
The
in
changes
i.
We
then construct rolling
and counterfactual output
no need to present a
different
^ AA>/yt_i
two
series.
composition nearly offset each other, and the
with the general evolution of output volatility over the
5
Conclusions
We
have documented the long and large decline
We
have shown that
is
series are nearly indistinguishable.
values are within .05% of each other. Composition effects have
last half-century.
There
in
last
little
final
to do
50 years.
output volatility over the
this evolution
does not have one, but
—
Output
many proximate
up
in
25
volatility
Among
causes.
them, the evohition of inflation volatihty,
the 1970s and early 1980s,
down
since then; a steady decrease in
investment volatihty, and even more so of consumption
in the volatility of
volatility;
A change
government spending early on.
to countercyclical inventory investment in the 1990s.
a decrease
from procyclical
Many questions remain
however.
First,
pohcy
of
about the deeper causes of the decrease
— especially
monetary policy
in volatility,
from the role
— to the role of structural changes
especially changes in financial markets.
•
Our
findings suggest a complex role for
hand, the trend decrease
give
much support
result of a
policy, of a
r-
output
in
for the idea that
effect.
volatility in the
first is
cyclical
what we are seeing
dramatic recent improvement
Greenspan
policy.
On
the one
output volatihty from 1950 on does not
On
in
is
primarily the
our conduct of monetary
the other, the dramatic decrease in
mid- 1980s can be interpreted
both of them related to monetary
The
monetary
in
two ways,
policy.
that this decrease was indeed the result of smarter counter-
monetary
the 1980s on.
policy, leading to better
output stabilization from
This explanation runs into a puzzle however.
Given
the lags in the effects of monetary pohcy on output, one would have
expected better monetary policy to show up primarily as shorter lived
effects of
shocks on output, and thus as a decrease
in
the
efficient in the univariate autoregressive representation.
AR(1)
There
is
co-
no
evidence that this has been the case.^^
'^A more sophi.sticated argument
is
that, despite the lags in
policy might have reduced the variance of
measured output shocks
expect shorter-lived effects of the underlying shocks on
these sliocks in the
first place.
monetary
GDP, and
is
policy, better
leading agents to
thus to react less to
But, even in this case, better policy should be reflected in
Output
The other
the
same
that
is
it
was associated with
decrease in inflation volatihty which occured around
time.
But even
The
second interpretation impHes a role
this
monetary
policy.
been due,
in large part, to better
Our
— and may have been largely
— the
caused by
•
26
volatility
findings also suggest a role for
the argument
is
be have
in financial
markets
not straightforward.
is
policy.
improvements
consumption and investment
in reducing
likely to
increased inflation stability
monetary
volatility.
On
for
But, here again,
theoretical grounds,
it
is
not obvious that more efficient financial markets should lead to lower
consumption
volatility.
While, for given interest rates, they plausibly
lead to a decrease in the volatility of consumption services
and non
durables, they also allow consumers to adjust faster to their desired
stock of durables, leading, other things equal, to more volatility of
spending on consumer durables.
The evidence
vestment.
is
The same argument
however of a decrease
applies to in-
in volatility in all
components of consumption and investment.
•
The
issue of the relative role of
monetary policy and
ket improvements in reducing output volatility
is
that respect, the evolution of Japan in the 1990s
inconclusive.
Japan
in
the 1990s. But was
markets
in financial
obvious.
As we saw, output
Monetary
(or to
policy,
it
financial
a fascinating one. In
is
both intriguing and
volatility increased substantially in
due to monetary
something
else)?
policy, or to
The answer
is
was
clearly disrupted.
— thus
in
far
from
— the
liq-
And, because of the problems of banks, intermediation
Only a more disaggregated examination
both a smaller variance of measured output shocks, and
on output
changes
both current and anticipated, was clearly
limited by the constraint that interest rates be non negative
uidity trap.
mar-
a decrease in the
AR(1)
coefficient.
in
will
shorter lasting effects of shocks
Output
27
volatility
help attribute blame.
Second, about the implications of our findings.
•
We
reasonably confident in predicting that the increase in the
feel
length of expansions
United States
is it
will
here to stay (This
is
not a prediction that the
not go through a recession in the near future, nor
a statement that the
cycle...):
is
The decrease
New Economy
has eliminated the business
output volatility appears sufficiently steady
in
and broad based that a major reversal appears
a
•
much
unlikely.
This implies
smaller likehhood of recessions.
Lower output
volatility suggests lower risk,
and thus changes
in risk
premia, in precautionary saving, and so on. Interestingly, the decrease
in
output volatihty has not been reflected
The
asset price volatility.
in a parallel decrease in
last figure of this paper,
Figure 12, plots the
evolution of output and of stock market price volatility, the second
being measured as the rolling standard deviation of the rate of change
of the
Dow
Jones index.
evidence of a trend in
•
Dow
As documented by
Jones
And, obviously, ultimately what
but the
risk facing individuals.
others, there
is
little
volatility.-*^
important
is
We
is
not aggregate
risk,
do not know what has happened
to the volatility of idiosyncratic shocks during the period.
We
"°This
intend to explore
is
all
these avenues in the near future.
not neces,sarily a puzzle.
If
we think
of the better use of
monetary policy
as one of the factors behind the decrease in output volatihty, stronger stabilization
require larger
cisset prices.
let
movements
There
is
in interest rates,
however
little
alone nominal interest rates.
may
and thus potentially stronger movements
in
evidence of increased volatility in real interest rates,
o
o
05
O
CO
O
O
in
CM
O
O
p
CD
CD
O
o
m
c
o
(0
>
Q>
T3
T5
OX
CO
CD
T3
o
C ^
OS
^
o
CO Q
"O
O) c
c
^^
H
O
03
a.
Ci
(0
Q^
cvi
T"
Q>
3
O)
LJ.
o
o
in
o
o
c>
CD
CD
o
o
o
o
o
LO
CNJ
o
o
Output
volatility
References
Balke, N. and Gordon, R., 1989,
The estimation
methodology and new evidence, Journal of
of prewar gross national product,
Political
Economy 97(1),
Blanchard, O. and Watson, M., 1986, Are business cycles
in
38-92.
25, 123-156,
all alike?,
"The American Business Cycle: Continuity and Change, Robert Gordon
ed,
University of Chicago Press, Chicago.
Burns, A., 1960, Progress towards economic stability,
-
Amencan Economic Review
50(1).
Hamilton,
series
Kahn,
J.,
1989,
A new
and the business
J.,
approach to the economic analysis of non statinary time
cycle,
Econometnca 57, 357-384.
McConnell, M., and Perz-Quiros, G., 2001, The reduced
U.S. economy; Policy or progi'ess?, Mimeo, Federal Reserve
McConnell, M. and Perez-Quiros, G., 2000,
States;
What
volatility of the
Bank
Output fluctuations
of New York.
in
the United
has changed since the early 1980s?, American Economic Review
90(5), 1464-1476.
Romer, C, 1986,
data?,
Simon,
.J.,
Is
the stabilization of the post-war economy a figment of the
American Economic Review 76, 314-334.
2000,
The long boom. Essays
in empirical
macroeconomics, Ph.D.
thesis,
MIT, Cambridge, MA.
Taylor,
J.,
2000, Recent changes in trend and cycle, mimeo, Stanford.
Weir, D., 1986,
The
cyclic stability,
reliability of historical
macroeconomic data
Journal of Economic History 46.
for
comparing
'SQU5
Date Due
MIT LIBRARIES
3 9080 02246 2771
HUlKHllUiHimiiim!