Digitized
by
the
Internet
Archive
in
2011
with
funding
from
Boston
Library
Consortium
Member
Libraries
working
paper
department
of
economics
massachusetts
institute
of
technology
50 memorial
drive
Cambridge,
mass.
02139
Consumption and
the
recession of 1990-1991
Olivier Blanchard
MIT and
NBER
No.
93-5
Jan.
1993
J.T.
UBRAR
AUG
12
1993'Consumption
and
the recession of1990-1991.
Olivier
Blanchard,
MIT
and
NBER
*January
15,1993
Abstract
The
paper looksatthe recession of1990-91. Itreaches twoconclusions:Notall recessions have thesamedynamics. Decreases in output which
come
fromother sourcesthanconsumptionshocks tendtobeshortlived,andfollowed by sharp
recoveries. Decreases in output which
come
from consumption shocks, decreasesin consumption given income, tend to be longer lasting, and followed byweaker
recoveries.
In contrastto its predecessors, the recession of 1990-91 was causedprimarilybya
consumption shock. Thus, conditionalon the natureofthose shocks, the slow
re-coverycomesasnosurprise. Asoftheendof1992,theeconomywasnearly exactly
onthe trackgivenby adynamicforecastbasedoninformation uptothefirstquarter
Recession 2
Duringthelasttwoquarters of1990andthefirstquarter of 1991, the
US
economy
experienced negativegrowth.
These
three quartershavenow
been designatedasthe 1990-1991 recession by the
NBER.
But both in the year precedingand
the year following, growth wasanemic
as well.Growth
since the beginning of 1989hasaveraged
0.7%
atan annual rate, afarcry from its post-1973mean
of 2.2%.Incontrast toitspredecessors, this recessiondoes not have an obvious proximate cause.
And
precisely because of that, explanations abound, ranging, to cite only the leading candidates, from a longexpansiondying ofold age, toconsumer
de-pression, a
mix
ofdebt overhangand
therealizationoflowergrowthprospects, tothecreditcrunch, thecombinationofimprudentbehavior byfinancial institutions and the tightening ofregulation, to the
end
of theCold
War
and thedecrease indefense spending, tostructural adjustmentsrequiredby globalcompetition*.
The
purpose ofthis paper is to look at the data, putting justenough economic
structure
on
the econometricstopinpoint ifnot the deep, atleast the proximatecauses of the recession.
The
storywhich
emerges is relatively clear. By far, themain
proximate cause of the recessionwas a "consumption shock" a decreaseinconsumption
in relation to its normal determinants. Because the effectsofsuch shocks are longlasting, this also explains why, incontrast topreviousrecoveries,the recoverywas aslow and
weak
one.The
issueleftunanswered
iswhether this"consumption shock" was
due
to animal spirits/taste shocks, orsimply toantici-pations of the slowgrowth to come. Circumstancial evidencepoints to a role of
Recession 3
I.
Looking
at theComponents
ofGNP.
A. Estimating a
VAR.
A
simplefirstpassis toestimate the jointbehaviorof thecomponents
ofGNP, and
look at the residuals. I thus
decompose
realGNP
(Y), as thesum
ofconsump-tion
on
non
durablesand
services (C),consumption
expenditureson
durables(CD), residential investment (IR),
non
residential (INR) investment,govern-ment
spending (G), inventory investment (1NV)and
net exports (NX),lb
achievestationarity, I log-difference C,
CD,
IR,INR
and
G,and
divideINV
and
NX
by trend output, obtained byfitting abroken exponential trendtoGNP,
witha breakin 1973.
I then run a
VAR
for the seven transformed variables, for the period 1959:1 to1992:3, with 3 lags ofeach variable, a constant and a post-73
dummy
in eachequation. Issuessuchastreatment oftrend,or the incorporation of cointegrating
relations, while important if
we
were to look at impulse responses, are oflittleimport
when
the focusis, as here,on
the residuals toeachequation.Given
the 7 estimated equations,an auxiliary equationisneeded
tocharacterize thebehaviorofGNP
and
itsresidual.Given
thenon-linear transformations of the individualseriesusedtoinducestationarity, there aretwo waysofdoingthis.The
first is to constructGNP
residuals for each period as a weighted average of theresiduals ineachequation, multiplied by the time varying ratiosofeach
compo-nentto
GNP.
The
otheristo regress log-differencedGNP
on
thesetofrighthand
side variables of the
VAR,
and
use the residuals from this equation. This second approachyields resultsvery similartothe first,and
issimpler to implement. ThisRecession 4
B. Identifying the Shocks.
Denote
the residuals to each equation by lower case letters, forexample
c forthe
consumption
equation, yfor the-auxiliary-outputequation.These
residuals are simply forecast errors.They
are, not surprisingly, generally positively corre-latedacrossequations. Thisreflectstheir jointdependence on
common
underly-ing shocks, as well as their direct
dependence on each
other.To
theextent thatconsumption
respondstoincome
within the quarterforexample, partof the con-sumption residual, c,reflectsalsotheshocksto othercomponents.We
canmake
some
progressand
get closer to the underlying shocks bymaking
two identifyingassumptions.
The
firstassumption isthat, within thequarter, thecomponents
ofGNP
depend
on
each other only throughGNR
Thus, I assumethat, forexample, cdependsonly
on
y-theresidualto theoutputequation-, notseparately
on
ir,inrand
soon.The
second assumptionisthatg,thegovernment
spendingresidual, isexogenous.
Under
thosetwoassumptions, gcanbe usedasan instrument toestimate theeffectsof yon
eachof the othersixresiduals.Denote
the residuals from these instrumentalvariable regressions by e, thus using ec for
consumption
forexample,and
refertothem
asthe "shocks".These
new
residuals are stillcross-correlated, but lessso thantheoriginalones.The
two identifying assumptions are crude,and
forecast errors ingovernment
spending arenot a very powerful instrument. Nethertheless, the estimates of the
contemporaneous
effectsofyon
itscomponents
make
sense. Using samplemean
values togo fromestimatedelasticities to derivatives, the estimates imply that a
one
dollar increaseinGNP
increasesconsumption
ofnon
durablesand
servicesby 12 cents, durable
consumption
by 11 cents, residential investment by-1 cent,in-Table 1.
The
source ofshocksAccumulated
normalizedshocksQuarter
Uv/°v)
E(Cc/^c)
EO.rMr)
1989: 1 0.00 0.00 0.00 1990:2 2.32 -2.50 0.25 1990:3 1.38 -2.29 -1.15 1990:4 0.79 -4.07 -1.61 1991:1 -0.39 -5.42 -2.87 1991:2 0.18 -5.35 -2.17 1992:3 0.46 -6.12 -0.15
The
estimation of the shocks, the e's, isdescribed in thetext.The
accumulatedseriesfollow
random
walks with zeromean
and
unitstandard deviation.Thus
forexample, starting at 0.00 in 1989:1, the distribution of the
sum
in 1992:3 hasRecession 5
vestmentby -15 cents.
The
next step is then to look at the sequence ofshocks, the €x's, and look forunusually large individual shocks, or for sequences ofshocks of the
same
sign.A
convenientway
todo
so is to look at the accumulated shocks to the variouscomponents, each normalized byits estimatedstandard deviation, starting from
some
datewhich
Ichoose tobethefirstquarter ofslowgrowth, 1989:2.The
main
results are presentedin table 1.
The
firstcolumn
gives thesum
ofnormalizedGNP
residuals, y/cry.Itshows
that,despite the fact that output growth was below its
mean
in 1989and
early 1990,residualstotheoutputequationwere, duringthat period,smallbutpositive. Thus,
the
VAR
explains theslowdown
ofoutput pre-1990:3 through the internaldy-namics of the
economy
rather than through adverse shocks. Itwould
be worthfollowing thisfinding through; I have not
done
ityet.The
secondand
thirdcolumns
give thesum
ofaccumulated shocks toconsump-tion of
non
durablesand
services,and
to residential investment respectively.These
arethe onlytwocomponents
ofGNP
which
show
largenegativeshocks(al-though, interestingly,
even
forthose, no single residualislarger inabsolute valuethan 2 standard deviations).
A
negative shock to residential investmentdomi-nates thefirst quarter of the recession, 1990:3.
Two
negativeshocks toconsump-tiondominate the next twoquarters.
While
the largest shocks take placeduring the recession, thewhole period since 1989:2 isdominated
by negative shockstoconsumption.
The sum
is still large in 1992:3.The
rest of the paper focuseson
Recession 6
II.
Looking
atOutput
and
Consumption
A.
From
7 to 2 Variables.7-variable
VAR's
are unwieldy, their dynamics hard to understandand
harder todescribe. Fortunately, examination of the
dynamic
structure of theVAR
abovesuggests a short cut. Tests of the significance ofeach variable in each equation
(which,becauseofspaceconstraints,arenot reported)
show
thestrong predictivepower
ofnon
durables and servicesconsumption
for nearly allcomponents
ofGNP; and
theweak
predictivepower
of most other components. This suggeststhat
we may
not lose toomuch
by focusingon
a bivariate system in outputand
consumption. Thisis
what
I do in thissection.I thus estimatea bivariatesysteminthelogarithm of
consumption
ofnon
durablesand
services,and
thelogarithm ofGNR
There
isweak
evidenceof cointegration,that the ratio of
consumption
toincome
is stationary. I thus specify theVAR
interms offirstdifferences, allowing forthree lags ofeachfirst-differenced variable,
the lagged value ofthe log
consumption
toincome
ratio, a constantand
a post-1973dummy
2.B.
From
Residuals toShocks
Estimation of the system yields residuals for
consumption
and
income, cand
yrespectively. 1 follow the
same
approach asbefore to get toconsumption
shocks,using the unexpected
component
ofgovernment
spending as an instrument ina regressionofcon
y.Thisyieldsanelasticityof.12 ofconsumption
toincome, thusan increase of
consumption
of7cents within the quarter fora dollar increasein income.The
"consumptionshock", ec, isdefined astheresidualofthisregression.1 i
o
io
iO
io
o
03O
•vl IS3 en CDL
iO
b
o
p
CO8
S
m
3
3
I
1
o
1o
o
o
o
o
o
o
-fc.o
N3o
o
o
N3o
o
eno
CD U3O
<n 1 l I l -i i r Tlo
m
o
>
—1>
O
—1c
>
I -* t 1 i | i | / | I t :; t... . ':/-y~M
0-0*: /W+M
:::M
}\
1 \ 1 \ 1 \ \ \ • \ 1 V 1 \ f \ I \\
\\
\\
\\
\\
\ J \ \ v I \ 1 \ \ \ I\\
\8
II
I
I
>icz
70
N>
1
o
1o
o
o
o
o
o
o
o
o
o
o
o
-b.o
eno
COV
\ A ':V
• \ 4 / to
70m
O
>
in>
o
—ic
>
O
8
i <>>m
Recession 7
A
regressionof y in turnon
ec yields an elasticityof.51 (thus an increase of96cents in
income
for an increase inconsumption
ofa dollar, a multiplierslightlybelow one within thequarter), and an
"income
shock" as the residual, ey.
Sum-marizing, the
mapping
from residuals toshocks is-givenby:c
=
A2y
+
e cy
=
.51ec+
eyAn
examination of thesequenceofrealizedshocksgives a picture similarto thatin table 1.
The
first quarter of the recession, 1990:3 isdominated
by anincome
shock,
which
we
know
from earlier can be traced mostly tonon
residentialin-vestment.
The
next two quarters aredominated
byconsumption
shocks. Thislastfindingisthekeytounderstanding
why
therecoveryhasbeen
soweak.To
see why, I proceed in twosteps, first looking at thedynamic
effects ofbothincome
and
consumption
shocks,and
then looking atdynamic
forecasts ofoutput as of 1991:1.C.
Impulse
Responses,and
theWeak
Recovery.The
effects ofone-standard-deviation negativeconsumption and income
shocks aredrawn
inFigure 1.The
figure hastwomain
features.The
first is thatshocks toincome
haveonlytransitory effectson consumption and
income; their effectislargely
gone
within two years3.The
second is that shocks toconsumption
havelong lasting,
hump
shaped, effectson
output,and
to alesserextenton
consump-tion4.
Recession 8
for output based
on
the bivariateVAR
using information upand
including thelast quarter of the recession, 1991:1 (Thus, both lines are the
same
up toand
including 1991:1).
What
isstriking isnot onlyhow
weak
the recovery has beensince 1991, butalsothat a
weak
recoveryisalsowhat would
havebeen
predicted by the bivariatemodel
as of theend
of 1991:1. Indeed the actual recovery hasbeenslightly fasterthanthe
dynamic
forecast... (Note that thedynamic
forecastsare independentofouridentificationof theshocks.).
For comparison purposes, Figure 3 performs the
same
exercise for the previousrecession, giving output, actual and forecast as of the last quarter ofthat
reces-sion, 1982:4.
Note
that, in that case, thesame
bivariatemodel
largely predictsthe strong recovery
which
followed 5.Given
our decomposition of the shocks, and the characterization ofimpulsere-sponses, the key to Figure 2
and
Figure 3 is easily given.More
so than inpre-vious recessions, thedecrease in output in 1990-91 was due to adverse shifts in consumption.
Those
shifts have long,hump
shaped effectson
output,and
theirdynamic
effects explainwhy
the recovery has been slow.There
is an importantlesson: recessions are not necessarily followedby fastrecoveries; this
depends
on
thenature ofthe shocks
which
triggered the recession.HI. Foresight, or
Animal
Spirits ?I have so far established that the recession was associated with large negative
"consumptionshocks", thatsuch shocks have long lasting effects
on
output,and
that thisexplains
why
therecovery hasbeenso slow.Thisraiseshowever
thenextRecession 9
from ?
And,
inparticular,where
did those of 90-91come
from ?One
interpretation, call it "foresight", isthatconsumption
shocks aresimply thereflectionof anticipationsbyconsumersof othershocksand theireffect
on
future income.The
interpretationofimpulse responsesinFigure 1isthenthat consump-tion shocks are simply followed by thechanges inincome
which
triggeredthem
in the first place;
consumption
shocks are a mirror, not a cause.Another
inter-pretation
however
isthatconsumption
shocksreflect inpartmovements
in con-sumptionnotduetochangesinexpectations offutureincome. Reasonsmay
rangefrom increasing prudence, to changesin intertemporal preferences, to-stepping outside ofthe usual maximizing
model-
sudden realizations ofpastoverborrow-ing, panic, and so on.
Under
that interpretation, call it "animalspirits", impulse responses in Figure 1show
how
shifts inconsumption
lead through a combina-tion ofdynamic
multiplierand
accelerator effects to ahump
shapedresponse of output.Can
one tell these two interpretations apart ?To some
extent, one can.Under
theanimalspiritsinterpretation,
and
under the plausibleassumption thatanimalspiritshavelittle or
no
long runeffecton
output, the impulse response ofoutputto a
consumption
shock should eventually return to zero. This isclearlyviolatedinFigure 1, suggesting that
consumption
shocks must reflect in part foresightofshocks with
permanent
effects.The
foresight interpretationon
the otherhand
imposes constraints
on
the relation of theconsumption
response to thesubse-quent response ofincome, and ofthe shapeof the
consumption
impulse responseitself. Following the route of putting
more
theoreticalstructureon
thedata totryto disentangle foresight and animal spirits
would
takeme
too far. I shall insteadRecession 1
nature of the negative
consumption
shocksof 1990-91:I see theoverall evidence, from the timing of
monthly
declines in the index ofconsumer
confidence (from theMichigan
Surveyofconsumers), in the indexofleadingindicators,
and
incommercialforecastsof output,aspointingtomore
thanconsumer
foresight atwork. First, incontrast to earlierrecessions, the decline in confidence was largely prior to-and
much
stronger thanwould
havebeen
pre-dictedby-either thedecline of leadingindicators or commercialforecasts of the
recession.Second, perhapsnotcoincidentally,thefirstlargedeclineinconfidence
inAugust 1990,wasassociated with
an
importantbutlargelynon economic
event, theinvasion ofKuwaitbyIrak. Third, afterhavingdropped,consumer
confidence remainedverylow inthe following twoyears,much
lower thanwould
have beenpredicted
on
thebasisofhistorical relationswith aggregatevariables.Each
piece ofevidencecan, withsome
effort, be reconciled with theforesight interpretation;together however,I find
them
suggestive ofa role foradropinconfidencecoming
Recession I 1
References
Cochrane, John, "Permanent and Transitory
Components
ofGNP
and
Stock Prices",
mimeo,
UniversityofChicago,July 1992.Sinai,Allen,"What's
Wrong
with theeconomy
?", Challenge,forthcom-ing, 1992
Walsh, Carl,
"What
caused the 1990-1991 recession ?",mimeo,
FederalRecession 1
2
Footnotes
*
Department
ofEconomics, MIT, 50Memorial
Drive,Cambridge
MA
02139.I thank the
NSF
for financial support, Roger Brinner, Ricardo Caballero,Julio Rotemberg, AllenSinai
and Lawrence
Summers
for usefuldiscussionsand
comments.
1
Two
usefuldiscussionsofpotential causesaregiven byAllen Sinaiand CarlWalsh
.2
Such
systemshavebeen
estimated bymany
others.John Cochrane
in par-ticularfocusesalsoon
thedifferentdynamic
effectsofconsumption and
in-come
shocks. Hisapproach toidentificationishowever somewhat
differentfromthat used in this paper.
3
The
fact that the long run effect ofa negative shock is positive is mildlyembarassing, but thislongrun effectisnot significantlydifferent from zero.
4 Thispoint isalsoemphasized byCochrane.
5 Results for the post -recession episodes of 1971
and
1975 are qualitatively similar to those for 1982, with fast predicted and actual growth followingDate
Due
291994
MIT LIBRARIES DUPl