ALAN S. BLINDER
Brookings Institution and Princeton University
ANGUS DEATON
Princeton University
The
Time
Function
Series
Consumption
Revisited
THERELATIONSHIP
between consumer spending and income is one of the
oldest statisticalregularitiesof macroeconomics-and one of the sturdiest. Like the aging movie star, it needs a little touchingup now and
again,but always seems to come bouncingback.
A dozenyearsago, boththe theoreticalderivationandtheeconometric
form of the aggregateconsumptionfunction were considered settled.
Most economists adheredto one of two ways of puttingFisher's theory
of intertemporaloptimizationinto operation: Milton Friedman's permanent income hypothesis (henceforth, PIH) or Franco Modigliani's
life-cycle hypothesis (henceforth,LCH).' Since each variantseemed to
have sound theoretical underpinnings,and since the two had similar
econometricforms that explainedthe datawell and had similarimplications for policy, there was not a greatdeal to quarrelabout. Perhapsthe
most contentious empirical issue was the apparentlylarge marginal
Thispaperhas benefitedfromthe commentsandsuggestionsof AlbertAndo, Whitney
Newey, and membersof the BrookingsPanel and from seminarpresentationsat the
Universityof Warwick,PrincetonUniversity,and JohnsHopkinsUniversity.We thank
Peter Rathjensand Lori Gruninfor researchassistanceand the NationalScience Foundationfor financialsupport.
1. Milton Friedman, A Theory of the Consumption Function (Princeton University
Press, 1957); Franco Modiglianiand Richard Brumberg, "Utility Analysis and the
ConsumptionFunction:An Interpretationof Cross-SectionData," in KennethK. Kurihara, ed., Post-KeynesianEconomics (Rutgers University Press, 1954), pp. 388-436;
AlbertAndoandFrancoModigliani,"The 'Life-Cycle'Hypothesisof Saving:Aggregate
Implications and Tests," American Economic Review, vol. 53 (May 1963), pp. 55-84.
465
466
BrookingsPapers on EconomicActivity,2:1985
propensityto consume out of transitoryincome, which was variously
explained by a "short horizon" (that is, a high discount rate) or by
liquidityconstraints.
Thingsare quite differentnow. Developmentsin economic research,
as well as actual events, have raised fundamentalquestions about the
consumptionfunction. At the same time, the rangeof experience of the
last dozen years has been great enough to hold out the hope of getting
some answersfromaggregatedata.This seems, therefore,an auspicious
time to take a fresh, and unabashedlyempirical,look at the time series
consumptionfunction.
Questions Raised by Modern Research
TheLucas Critique. Reasons aboundfor questioningthe traditional
consumptionfunction and its implicationsfor how tax policy affects
consumer spending.Robert Lucas has pointed out that, underrational
expectations, the PIH does not lead to a "structural" relationship
betweenconsumptionandincome, butratherto a statisticalrelationship
that should change whenever the stochastic process generatingincome
changes. The Lucas critique calls for estimation methods that treat
consumptionand incomejointly.2
The "Random Walk" Hypothesis. Robert Hall sharpened the impli-
cations of the PIH by showingthat the rationalexpectationshypothesis
impliesthatonly "surprises"in permanentincome shouldaffect current
consumption,once laggedconsumptionis controlledfor.3
The work of Hall and Lucas added a new dichotomy-that between
anticipated and unanticipatedchanges in income-to the traditional
permanent-transitory
dichotomy. It is this new dichotomy, ratherthan
the old one, thathasabsorbedthe attentionof contemporaryresearchers.
Hall's work in particularhas spawned an infant industry estimating
Euler equationslinkingcurrentand lagged consumptionin the manner
2. RobertE. Lucas,Jr., "EconometricPolicyEvaluation:A Critique,"inKarlBrunner
and Allan H. Meltzer, eds., The Phillips Curve and Labor Markets, Carnegie-Rochester
Conference Series on Public Policy, vol. 1 (Amsterdam:North-Holland, 1976), pp.
19-46.
3. Robert E. Hall, "Stochastic Implicationsof the Life Cycle-PermanentIncome
Hypothesis:Theory and Evidence," Journalof Political Economy, vol. 86 (December
1978), pp. 971-87.
Alan S. Blinder and Angus Deaton
467
impliedby the first-orderconditions of a Fisherianintertemporaloptimizationproblem.We have our doubts about the wisdom of modeling
aggregateconsumptionas the interiorsolution to a single individual's
optimizationproblemin adjacentperiods,4but in any case thinkit fairto
say that the researchdone to date has not supportedthe econometric
restrictionsimplied by the Euler equation approach. Nor has further
investigationvalidatedthe hypothesisthatthe responseof consumption
to income (henceforth, Y) reflects only the usefulness of current Y in
predictingfutureY.Instead,researchtypicallyfinds"excess sensitivity"
to currentincome.' But the case is by no means closed. So a central
questionof this study is whetherinformationknown at time t - 1, such
as anticipatedincome, has any predictive power for changes in consumption between times t - 1 and t.
The Barro Equivalence Hypothesis. A rather different objection to
standardconsumption functions, based on the idea that private and
government accounts should be consolidated, was raised by Robert
Barro.6The income (that is, disposable income) and wealth (that is,
household net worth, includinggovernmentdebt) variables normally
used in consumptionfunctions imply that intertemporalshifts in the
4. Cornersolutionsstemmingfromliquidityconstraintspose oneproblem.Aggregation
poses others. For example, with mortalconsumers, a constant age distributionof the
population,and neithersurprisesnor changesin interestrates, the ratioC,+ IC,(whereC
denotes consumption)would be 1 plus the growthrate of per capitaincome. The Euler
equationapproachmodelsthe growthrateof consumptionas a functionof the interestrate
andthe timediscountrate,anduses the observedgrowthrateto estimatethe intertemporal
elasticityof substitution.Backgroundgrowthof percapitaincomeseems to be ignored.
5. MarjorieFlavin,"TheAdjustmentof Consumptionto ChangingExpectationsabout
FutureIncome," Journal of Political Economy, vol. 89 (October 1981),pp. 974-1009;
RobertE. Hall and FredericS. Mishkin,"The Sensitivityof Consumptionto Transitory
Income:EstimatesfromPanelDataon Households,"Econometrica,vol.50 (March1982),
pp. 461-81; Ben S. Bernanke,"AdjustmentCosts, Durables,and AggregateConsumption,"JournalofMonetaryEconomics,vol. 15(January1985),pp.41-68;Ben S. Bernanke,
"PermanentIncome, Liquidity,and Expenditureon Automobiles:EvidencefromPanel
Data," QuarterlyJournalof Economics,vol. 99 (August1984),pp. 587-614;N. Gregory
Mankiw,JulioJ. Rotemberg,andLawrenceH. Summers,"IntertemporalSubstitutionin
Macroeconomics,"QuarterlyJournalof Economics,vol. 100(February1985),pp. 22551;MartinBrowning,AngusDeaton,andMargaretIrish,"A ProfitableApproachto Labor
Supply and CommodityDemands over the Life Cycle," Econometrica,vol. 53 (May
1985), pp. 503-43; Charles R. Bean, "The Estimationof 'Surprise'Models and the
'Surprise'ConsumptionFunction," Discussion Paper 54 (Centerfor Economic Policy
Research,February1985).
6. Robert J. Barro, "Are GovernmentBonds Net Wealth?"Journal of Political
Economy,vol. 82 (November-December1974),pp. 1095-1117.
468
BrookingsPapers on EconomicActivity,2:1985
patternof taxes, with no changein theirpresentvalue, produceshifts in
the time patternof consumption.This shouldnot be so, Barroargued,if
people can freely transferincome across generations.
The Barro equivalence hypothesis is not theoreticallyunobjectionable. In addition to the usual perfect capital markets assumption, it
requiresthat bequests be motivatedby intergenerationalaltruismand
thatpeople haveextremelylongtimehorizons.It also hastroubledealing
with childless people or with the possibility of "corner solutions" in
which the unconstrainedoptimalbequest cannotbe enforcedbecause it
is negative. Because of these andotherproblems,manyeconomists find
the equivalencehypothesisimplausibleon a priorigrounds.But a priori
reasoningis not the way to settle the issue, and empiricalstudies have
foundit surprisinglydifficultto rejectthe equivalencehypothesis.7More
evidence would be welcome, and we try to obtainsome below.
Intertemporal Substitution. Modern macroeconomic analysis has
reemphasized intertemporalsubstitution. Yet standardconsumption
functionsoften omit the rateof interestas an argument-not on theoretical grounds,but on empiricalgrounds.The consensus conclusion that
consumption,and hence saving, is insensitive to the rate of returnhas
been questionedby Michael Boskin and, more recently, by Lawrence
Summers.8Whatdo recent data say aboutthis issue?
7. For some theoreticalarguments,see MartinS. Feldstein, "PerceivedWealthin
Bonds and Social Security:A Comment,"Journalof Political Economy, vol. 84 (April
1976), pp. 331-36; Robert J. Barro, "Reply to Feldstein and Buchanan,"Journal of
Political Economy, vol. 84 (April 1976), pp. 343-49; and WillemH. Buiter and James
Tobin, "Debt Neutrality:A BriefReview of Doctrineand Evidence," in GeorgeM. von
Furstenberg, ed., Social Security vs. Private Saving (Ballinger, 1979), pp. 39-64. The
conclusionthattheempiricalevidenceis mixedis reachedby KarlBrunneraftera thorough
review of the literature.See Brunner,"Fiscal Policy in MacroTheory:A Survey and
Evaluation"(Federal Reserve Bank of St. Louis, January 1985). But both Roger C.
Kormendi,in "GovernmentDebt, GovernmentSpending,andPrivateSectorBehavior,"
AmericanEconomicReview, vol. 73 (December1983),pp. 994-1010,andJohnJ. Seater
andRobertoS. Mariano,in "New Tests of the Life Cycle andTax DiscountingHypotheses," Journal of Monetary Economics, vol. 15(March 1985), pp. 195-215, claim that the
data stronglysupportthe Barrohypothesis. For a recent contraryview, see MichaelJ.
Boskin and LaurenceJ. Kotlikoff, "Public Debt and U.S. Saving:A New Test of the
NeutralityHypothesis," WorkingPaper 1646(NationalBureauof EconomicResearch,
June 1985).
8. MichaelJ. Boskin,"Taxation,Saving,andtheRateof Interest,"JournalofPolitical
Economy, vol. 86 (April 1978),part 2, pp. S3-S27, and Lawrence H. Summers,"Tax
Policy, the Rate of Return,and Savings" WorkingPaper995 (NationalBureauof EconomicResearch,September1982).For a critiqueof Boskin'sworkandopposingresults,
Alan S. BlinderandAngusDeaton
469
Questions Raised by Recent Events
TemporaryTaxChanges. The purePIH with no liquidityconstraints
predictsthatpeople will reactmuchless to temporarythanto permanent
changes in taxes. And in 1968, when a temporarytax surchargewas
imposed, consumptiondid indeed decline less than simple Keynesian
consumptionfunctions predicted. Similarly,a temporarytax decrease
in 1975led to a strong surge in saving, but only a modest increase in
consumption. In the aftermath of these two episodes, both casual
observation of the facts and formal econometric research seemed to
supporta modifiedversion of the PIH.9But this inference rested on a
slenderdatabase.
Recent events have given us anotherepisode. Since the Reagantax
cuts of 1981-84came in threepreannouncedstages, they can be thought
of as a permanenttax reductionin August 1981coupledwitha temporary
tax increase of graduallydiminishingsize. Thus the PIH predicts that
savingshouldhave declined sharplyafterAugust 1981as the scheduled
permanenttax cut inducedhigherconsumption.Did it?
To answer this question we must, at a minimum,adjustthe data for
the sharpbusinesscycle thattook place duringthis period,for even very
weak versionsof the PIHimplythatsavingratesshouldfall in downturns
and rise in booms. Table 1 shows cyclically adjusted saving rates for
see E. PhilipHowrey and Saul H. Hymans, "The Measurementand Determinationof
Loanable-FundsSaving," BPEA, 3:1978, pp. 655-85. Robert Hall, in "Intertemporal
Substitutionin Consumption"(StanfordUniversity,July 1985),arguesthat the interest
elasticityof consumptionis quite small, and that previoushighestimatesare biased. For
two recent empiricalsurveys, see ThorvaldurGylfason, "InterestRates, Inflation,and
the Aggregate Consumption Function," Review ofEconomics and Statistics, vol. 63 (May
1981),pp. 233-45; and GeraldCarlino, "InterestRate Effects and IntertemporalConsumption,"Journal of Monetary Economics, vol. 9 (March 1982), pp. 223-34.
9. RobertEisner, "Fiscal and MonetaryPolicy Reconsidered,"AmericanEconomic
Review, vol. 59 (December 1969), pp. 897-905; ArthurM. Okun, "The PersonalTax
Surchargeand ConsumerDemand, 1968-1970,"BPEA, 1:1971,pp. 167-204;WilliamL.
Springer,"Did the 1968SurchargeReally Work?"AmericanEconomicReview, vol. 65
(September1975),pp. 644-59; FrancoModiglianiandCharlesSteindel,"Is a Tax Rebate
an EffectiveTool for StabilizationPolicy?"BPEA, 1:1977,pp. 175-203;Alan S. Blinder,
"TemporaryIncomeTaxes andConsumerSpending,"Journalof PoliticalEconomy,vol.
89 (February1981),pp. 26-53.
BrookingsPapers on EconomicActivity,2.:1985
470
Table 1. Cyclically Adjusted Net Saving as a Percentage of Net National Product,
1971
84a
Sector
1.
2.
3.
4.
5.
Government
Personal
Business (net)
Total private (2+3)
National (1+2+3)
1971-80
average
1980
1981
1982
1983
1984
-1.7
5.5
2.5
8.0
6.3
-1.4
4.5
2.1
6.6
5.2
-1.2
5.5
1.9
7.4
6.2
- 2.3
5.2
2.3
7.5
5.2
- 3.4
5.0
2.9
7.8
4.4
-4.6
5.9
2.7
8.6
4.0
a. Cyclical adjustment is based on regressions of particular saving rates on time, the unemployment rate, and the
change in the unemployment rate. The coefficients of the unemployment rate that are used to do cyclical adjustment
are as follows (with t-statistics in parentheses):
Item
Level of unemployment rate
Change in unemployment rate
Government
-0.76
(-5.2)
-0.17
(-1.1)
Personal
-0.25
(-2.4)
0.37
(3.4)
Business
-0.08
(-0.8)
Private
-0.33
(-2.4)
National
(-8.5)
-0.46
(-4.1)
-0.09
(-0.6)
-0.26
(-1.9)
-1.09
persons (households), businesses, and governmenton average for the
period 1971-80 and then annually for 1980-84.10 These data do not
suggest that the saving rate droppedafter the 1981tax act was passed.
As cyclically adjusted government dissaving rose steadily from 1.2
percentof net nationalproduct(NNP) in 1981to 4.6 percent of NNP in
1984,the cyclically adjustedpersonalsavingrate didfall slightlyin 1982
and 1983.But the adjustedrate of net business savingrose by more, so
that, if households "see through the corporate veil" by treating the
retained earnings of corporations as their own, the relevant saving
concept (total private saving) actually increased slightly in 1982 and
1983.But a simple look at the data is not the properway to answer the
question.We need to studywhetherconsumers'reactionsto the Reagan
"temporarytax increases" were consistent with their behaviorin 1968
and 1975.
Interest Rates and Inflation. Recent policy initiatives have focused
attentionanew on the sensitivityof savingto the after-taxrateof return.
Until a decade ago, the relatively small variance in the after-taxreal
interestratemadeinferencesaboutthe interestelasticityof consumption
10. Cyclicaladjustmentwasperformedby firstestimatingannualordinaryleastsquares
(OLS) regressions of the form: s, = a + bt + cU, + d[U, - U, I] + e, over the period
1954-84. Here s is any of the net saving rates listed in the table, U is the civilian
unemploymentrate, and t is time. The estimatesof the coefficientsc and d were used to
computecyclicallyadjustedsavingrates,thatis, the savingratesthatwouldhaveoccurred
if actualU hadbeenequalto RobertGordon'sestimateof the naturalrateof unemployment
each year. See RobertJ. Gordon,Macroeconomics(Little, Brown, 1983).
Alan S. Blinder and Angus Deaton
471
tenuous at best.1"Recent years, with their unprecedentedlyhigh and
volatile real after-tax interest rates, have remedied that problem and
should permit sharper inferences both about interest sensitivity and
about any direct effects that inflationmight have on consumption. In
particular,several years ago, Deaton suggested that the apparentdepressingeffect of inflationon consumptionmightbe due to the fact that
shoppersmistakenominalprice increases for real price increases when
inflationis unanticipated.12
Budget Deficits. The Barro hypothesis, which says (roughly) that
governmentspendingshouldreplacetaxes in the definitionof disposable
income,was also difficultto test untilrecentlybecausecyclicallyadjusted
budgetdeficits were small and variedlittle, except duringwars. Recent
events have changedthat. If Barrois right,privatesaving-and probably
personalsaving-should haverisendramaticallyto counteractthe effects
of government dissaving. Instead, table 1 shows that the cyclically
adjustedpersonal saving rate rose by only 0.4 of a percentage point
between 1981and 1984, while the cyclically adjustedgovernmentdissaving rate rose by 3.4 percentagepoints. If we look more broadly at
totalprivatesaving,13 the rise in the cyclically adjustedsavingrateis still
only 1.2 percentage points-about one-third of the decline in
governmentsaving. Thus, on quick inspection, the data appearhostile
to Barro'shypothesis:as the governmentdeficitincreased, net national
savingfell from6.2 percentof NNP (close to the averageof the previous
decade) in 1981to only 4 percent of NNP (the lowest level in the 195484 period)in 1984. But, once again, a more serious econometricinvestigationis in order.
Plan of the Paper
It has become traditionalto divide aggregateconsumerspendinginto
two components:purchasesof nondurablegoods and services, Ct, and
11. EugeneFama,."Short-TermInterestRates as Predictorsof Inflation,"American
Economic Review, vol. 65 (June1975), pp. 269-82.
12. See Angus S. Deaton, "InvoluntarySaving throughUnanticipatedInflation,"
Ametican Economic Review, vol. 67 (December1977),pp. 899-910.
13. Theequivalencehypothesis,it seems to us, suggeststhata decreasein government
savingshouldbe offset by an increasein personalsaving. But if householdsnot only see
throughthe corporateveil but actuallyreach throughit and get corporationsto do their
bidding,thenbusinesssavingcould offset the government'sactions.Thatis why we look
also at totalprivatesaving.
Brookings Papers on Economic Activity, 2:1985
472
purchases of durables. Modernresearch on the consumptionfunction
has focused on the formerandtypicallyhas modeledCtin isolationfrom
purchasesof durables.An alternativeprocedureis to workwith the sum
of the flow of services from durablegoods and purchasesof nondurable
goods and services. We adopt instead a middleposition in which each
component is treated separately, but relative price effects potentially
matter, as they would in a system of demand equations. To facilitate
comparisonwith the recent literature,and because modelingexpenditures on durablespresentsa host of specialproblems,the presentpaper
deals only with nondurablegoods and services.
We begin by developinga baseline consumptionfunctionfor Ctthat
includes as arguments such standard variables as income, wealth,
interest rates, relative prices, and inflation. We use this function to
addressa variety of basic questions. Is there any point in decomposing
income changes and other variablesinto anticipatedand unanticipated
components,oris the traditionalspecificationthatignoresthisdistinction
adequate?Is it only surprisechangesin variableslike incomeandwealth
that matter? Is consumption sensitive either to interest rates or to
inflation?Are there detectable relative price effects? Has consumer
behaviorchangedduringthe Reaganyears?
Next, we considervariousways to augmentthe baselinespecification
in order to test some more controversialhypotheses. Are permanent
and temporarytax changes treated differently by consumers? Is the
Barroequivalencehypothesis supported?The last section summarizes
what we thinkwe have learned.
A Basic ConsumptionFunction
THE
DATA
For purposesof this study, which covers the period 1954:1to 1984:4,
we make four changes in the official national income and product
accounts (NIPA) data.
First, we remove the 1975tax rebate from the income data until we
are ready to deal explicitly with the temporarytax issue. We do this so
as not to allow the 1975:2observationto exert undue influenceon our
baseline specification,for the raw data show a stunning23.8 percent
Alan S. Blinder and Angus Deaton
473
annualgrowth rate of real disposable income in 1975:2,followed by a
5.7 percent annual rate of decline in 1975:3.
Second, the NIPA include gross interest paymentsfrom businesses
to individuals(for example, on corporatebonds) in personal income,
and hence in disposableincome, without nettingout interestpayments
from individualsto businesses (for example, for consumercredit). But
the tax system essentially nets one against the other and levies taxes
only on net interest received. So we subtractedinterest paid by consumers to businesses from the NIPA definitionof disposable income.
This changelowers YslightlywithoutaffectingC. 14
Third, in the NIPA, personal "nontax payments" are groupedwith
personaltaxes. A closer examinationof this category reveals that such
things as tuition payments to state colleges and fees collected by
governmenthospitals are part of state and local "nontax payments."
Because these items seem more accuratelyclassified as personal consumption,not as taxes, we adjustedthe nationalaccountsby subtracting
state and local nontaxpayments,which totaled $46 billionin 1984,from
taxes andfromgovernmentpurchasesandaddingthemto consumption.
In doing so, we deflatedstate and local nontax payments by the NIPA
deflatorfor consumptionof services. This adjustmentraises C and Y
equallywithoutchangingthe governmentdeficit.
Fourth, since expenditureson durablegoods and expenditures on
nondurablesand services almostcertainlyrequiredifferenteconometric
explanations,they also requiresome classificationscheme. Wefollowed
the officialU.S. Bureauof EconomicAnalysis (BEA) classificationwith
one exception-we reclassifiedclothingandshoes, whicharenondurable
goods accordingto NIPA, as durables.
The resulting series on real purchases of nondurablegoods and
services, when put on a per capitabasis, became the basic variableto be
studiedand is henceforthdenoted by C. 15
14. The NIPA count interest paid as partof "personaloutlays," but not as partof
"personalconsumptionexpenditures."The adjustmentwe make is not a big one. For
example,in 1984interestpaidamountedto $78billion,whiledisposableincomewas $2,577
billion.
15. All series are seasonallyadjusted.We used total populationin makingper capita
conversions.Ourimplieddeflatorfor total consumerspending(henceforthP), which we
use to deflateour versionof disposableincome, differs triviallyfrom the NIPA deflator
becausethe two definitionsof totalconsumerspendingdifferslightly.
474
Brookings Papers on Economic Activity, 2:1985
FUNCTIONAL
FORM
AND
ESTIMATION
ISSUES
While the permanentincome model perhapsleads more naturallyto
a linearrelationshipbetween consumptionand some concept of income,
we adopt a logarithmicspecificationhere. We do not believe that any
importantresultsare sensitive to the choice between linearandlogarithmic form. But since we are interested in studying several price-like
variables, such as interest rates, the logarithmicform is convenient in
that it avoids the need for numerousinteractionterms to allow all the
coefficientsto depend,for example, on the interestrate. In addition,the
modern Euler-equationor "surprise" consumptionfunctions that we
wish.to studypredictthat, in the absence of new information,consumption grows from period to period at a rate that depends on the real rate
of interest. Once again, this is most easily modeled using a logarithmic
specification.
We do not, however, work with the currently fashionable "first
differences only" specification, because differencing obliterates the
single most strikingcharacteristicof the time series-the remarkable
constancyof the C/Yratio-and thereforeis silentaboutthe steady-state
propertiesof the system. Insteadwe attemptto captureboth the shortrun dynamics and the long-runpropertiesof the consumption-income
relationshipby adoptinga flexible distributedlag model that accommodates, or "nests," many of the specificationsthat have been discussed
in the literature-including both "Euler-type" specifications and the
error-correctionmodel that has been much recommendedby David
Hendryand several collaboratorsin the United Kingdom.16
Specifically,our basic functionalform is:
(1)
ACt= PO+ PIct-I + 12Yt+ 33Yt-l + qt8 + Zt-1y + ut,
wherec andy denote the naturallogarithmsof consumptionandincome.
In addition to income, there are two types of right-handvariables in
equation 1. The q variablesare contemporaneouslydated variableslike
16. JamesE.H. Davidson,David F. Hendry,FrankSrba, and StephenYeo, "Econometric Modelling of the Aggregate Time-Series Relationship between Consumers'
Expenditureand Incomein the U.K.," EconomicJournal,vol. 88 (December1978),pp.
661-92.
Alan S. Blinder and Angus Deaton
475
wealth, inflation,andrelativeprices. The zt_ variablesare eitherlagged
values of q variablesor othervariablesthatare knownat time t - 1 (such
as a time trend). The list of z and q variables changes as we examine
varioushypotheses duringthe course of the paper.
Wheneverwe estimate a version of equation 1, we also estimate an
augmentedspecificationthat decomposes y and q into anticipatedand
unanticipatedcomponents,namely:
(2)
z\c, = PO+ I3ct1 + 32EYt+ 3P*(yt-Eyt)
+ Eqt8 + (qt - Eqt)8* + Zt-ly + Ut.
We do this for two reasons. The first is to test the "surprises only"
predictionof the pure PIH without liquidity constraints, that is, the
implication that changes in consumption from t -1 to t should be
independentof informationabout income and wealth thatwas available
at time t - 1. We performthis test by droppinglaggedconsumptionfrom
the right-handside of equation2, re-estimatingthe equation, and then
testing whether the coefficients on anticipatedand lagged income and
wealth are zero.17 This test is precisely the "excess sensitivity" test
carriedout by MarjorieFlavin and others, adaptedto a more elaborate
specification.18
The second reason for decomposingall contemporaneousvariables
into anticipated and unanticipatedcomponents is econometric. The
currentvalue of a q variable(for example, the relativeprice of durable
goods) might have a well-definedtheoretical role in the consumption
function. But in addition, the "news" contained in any contemporaneouslydatedvariablemightinduceconsumersto revise theirestimates
of permanentincome, andthus to changetheirconsumption.If so, these
variableswill be correlatedwith the change in consumptionfor reasons
17. Thistest is sensitiveto assumptionsaboutthe presenceof transitoryconsumption.
If transitoryconsumptionis important,the differencedequationwill have movingaverage
residuals,in which case inclusionof laggedc will yield inconsistentestimates, and even
withoutlaggedc, the serialcorrelationwillinvalidatethe standardtest statistics.However,
we foundno evidenceof serialcorrelationwhetheror not laggedc was included(whichis
itself evidence againstthe purePIH), so we have no reason to doubtthe validityof our
test statistics.Sincec, appearsto be importantin severalof the regressions,we suppress
it only to makeour tests comparablewith those in the literature;test statisticswhen c,
is includedare muchless favorableto the surprises-onlyhypothesis.
18. See Flavin, "The Adjustmentof Consumption."
476
Brookings Papers on Economic Activity, 2:1985
having nothing to do with their inherent roles in the consumption
function, such as the substitutioneffects of relativeprices. The problem
is solved by constructinginstrumentalvariablesfor the q variablesfrom
first-stageregressionsusingdatadated t - 1 andearlier.This amountsto
replacingeach q by its "anticipated"value. We then allow separately
for the effects of the unanticipatedq variablesby includingthem in the
regressionsas well.
Estimation of equation 2 requires a way to deal with unobserved
expectations. As just suggested, we adopt the now standardmethodof
generating expectations as the one-period-aheadforecasts from an
estimatedvector autoregression(VAR)for the variablesy and q:
(3)
(Y)= AV,
+ et.
Here the vector Vt-I includes two lags each of c, y, and every variable
in q, plus all the zt_I variablesand a quadratictime trend.The expected
componentis the predictedvalue. The unanticipatedcomponentis the
series et.
The VAR equations themselves are of limited interest and change
every time we alter the specification of z and q; hence they are not
reported. However, since the unanticipated,or "surprise," variables
generatedby these equationsplay a majorrole in our analysis, a little
descriptionis in order.The VAR equationsfit the dataquitewell, so that
mostof the varianceofy andq is classifiedas "anticipated"(forexample,
R2forincomeexceeds 0.99). All the variablesare stronglyautoregressive
(generallyof second order),and, in addition,the stock of durablegoods
helpspredictbothincomeandinflation,the nominalrateof interesthelps
predict wealth, and time helps predict inflation.The appendixreports
the "data"on anticipatedandunanticipatedchangesof income, wealth,
and inflationgenerated by one importantversion of the model. The
simple correlationsbetween the actual changes and the surprises are
0.76, 0.66, and0.74 for income, wealth, and inflation,respectively.
After the VARs are estimated by ordinaryleast squares (OLS) and
used to create anticipatedand unanticipatedseries, equation2 is estimatedby OLS. This simpletwo-stepprocedurehas muchto recommend
it over more complicatedone-step proceduresthat treatequation2 and
equation3 as a system. First, it is simplercomputationally.Second, the
estimated coefficients in equation 2, the equation of interest, are less
Alan S. Blinder and Angus Deaton
477
contaminatedby specificationerrorsin auxiliaryequation3.19However,
thetwo-stepproceduredoes notyieldcorrectestimatesof allthe standard
errors, because it treats the anticipatedand unanticipatedvariablesas
known data, ratherthan as the statisticalestimates that they are. The
standarderrorsfor the coefficients of surprisevariablesare correct as
calculatedby OLS. But standarderrorsfor the other coefficients must
be obtainedfroma two-stageleast squares(2SLS) regressionthat omits
the surprisetermsand uses the VAR as the firststage.20
In applied econometrics, as in life, you rarely get something for
nothing.Some assumptionsmust be made in orderto identifya system
like equation2 andequation3. Ourmainidentifying,andthusuntestable,
assumptionis that transitoryconsumption, that is, the disturbancein
equation2, is orthogonalto the surprisesin incomeandinothervariables,
that is, to the disturbances,et, in equation3. Whilethis assumptioncan
certainly be questioned, it is far weaker than Friedman's original
assumption that transitory income and transitory consumption are
uncorrelated-because an income surprisedoes lead directly to a consumptionsurpriseaccordingto equation2.21
Note also that the estimatedcoefficientsof the no-surprisevariables
in equation2 are precisely those that would be producedby estimating
equation 1 by 2SLS. Put differently,the surprisemodel (equation2) is
observationallyequivalent to the traditionalmodel (equation 1) with
simultaneityaffecting y and q. A model with both simultaneityand
surprisesis thereforenot identifiable;the reader,like the authors,must
choose one interpretationor the other.22Throughoutthe paper,we adopt
the surprise interpretation.But readers preferringthe simultaneity
interpretationcan disregardthe coefficients of the surprise variables
19. This is just the standardargumentfor favoring limited-informationover fullinformationmethods.
20. See Adrian Pagan, "EconometricIssues in the Analysis of Regressions with
Generated Regressors," International Economic Review, vol. 25 (February 1984), pp.
221-48.
21. Flavin, in "The Adjustmentof Consumption,"identifiesthe model instead by
assuminga valuefor the coefficientof unanticipatedincomein equation2, a procedurewe
do not findappealing.
22. RobertHall, in "The Role of Consumptionin EconomicFluctuations,"in Robert
J. Gordon,ed., TheAmericanBusinessCycle(Universityof ChicagoPress,forthcoming),
explores the possibilityof estimating(a very parsimonious)consumptionfunctionwith
minimaluse of exogeneity assumptions.See also the commentsby Deaton that follow
Hall'spaper.
478
Brookings Papers on Economic Activity, 2:1985
in equation2 and treatthe othercoefficientsas 2SLS estimatesof equation 1.
Foreach pairof regressions,we routinelycarriedouta set of diagnostic
tests. First, in additionto the Durbin-Watsonstatistic, which is biased
against finding serial correlationowing to the presence of the lagged
dependent variable, and the Box-Pierce Q statistic, we constructed a
version of the Lagrangemultipliertest for serial correlationup to order
four. In none of the regressionswas thereever the slightesthintof serial
correlation,so we refrainfrom reportingall these test statistics. This
findingis of some interest,however, since the purePIH impliesthatthe
errorterm should be serially correlated(see footnote 17). Second, we
report the results of Chow stability tests over the two halves of the
sample and across the Reagan subperiod, 1981:3to 1984:4;marginal
significancelevels for rejectingparameterstability are labeled "halfsample stability"and "Reaganstability"in the tables. Except in a few
cases to be noted below, parameterstabilitycould not be rejected.
In additionto these diagnostics,we reporttests of two more economically interestinghypotheses. First, we always test the hypothesis that
the long-runelasticity of consumptionwith respect to income is unitysomethingthat is not obviously contradictedby the raw data. Second,
in all regressionsthat includethe rateof interest, we test the hypothesis
that only the real rate of interestmatters,that is, that the nominalaftertax interest rate and expected inflation(or actual inflationin the nosurpriseregressions)have equaland opposite coefficients.
A finalset of tests pertainsto the validityof the pure PIH and to the
value of decomposing variables into anticipatedand surprisecomponents. The hypothesis called "no decomposition" in the tables is that
the coefficientsof all anticipatedand unanticipatedvariablesare equal,
so thatonly actualvariablesmatter,thatis, the hypothesisthatequation
2 can be reducedto equation 1. Since the PIH suggests so stronglythat
anticipatedand unanticipatedincome and wealth should get different
coefficients, we next test the weakerhypothesisthat the decomposition
is irrelevantonly for these two variables,leavingothervariablesunconstrained.23Finally, we test the surprises-only hypothesis that was
23. Robert Hall pointed out at the September1985meetingof the BrookingsPanel
that, under some circumstances,the pure PIH is fully consistent with our no-surprise
specification.Specifically,if incomefollows a first-orderautoregressiveprocess, then y,
is the only variablerelevantto predictingfuturey variables.The appearanceof lagged
Alan S. Blinder and Angus Deaton
479
describedearlier, the hypothesis that lagged and anticipatedvalues of
income and wealth can be excluded from the regression. The readeris
remindedthat this test is not based on the surpriseregressionsreported
in the tables, but ratheron a constrainedversion thatomits c,_ fromthe
right-handside (see footnote 17above).
SIMPLEST
SPECIFICATION:
THE
CONSUMPTION-INCOME
RELATIONSHIP
We warmup by estimatingthe simplestpossible consumptionfunction: equations 2 and 1 with no z or q variables. Results are shown as
regressions2.1 and 2.2 in table 2. In this simple specification,which we
estimateonly becauseconsumption-incomerelationshipslikethisappear
so often in the literature,the surpriseand no-surpriseversions are very
similar,and we cannot come close to rejectingthe hypothesis that the
two income coefficients in regression 2.1 are equal. By contrast, the
hypothesis that only surprisesmatteris rejected at about the 3 percent
level. The implied steady-state elasticity of C to Y is 0.88, which is
differentfrom 1.0 at the 10percent(but not the 5 percent)level. Neither
longer lags of c nor longer lags of y were significantwhen added to
regression2.1 or 2.2.
This specificationis deficientin many respects. As a partialremedy,
we make three changes in moving to regressions2.3 and 2.4 in table 2.
First,we addwealthto the specificationby includingthreenew variables:
the logarithmof real wealth, w, divided into anticipatedand unanticipatedcomponentsin regression2.3, andthe laggedvalueof this variable.
Ourmeasureof wealth is the household net worth variableused in the
MPS model, except that we adjust the value of the governmentdebt
fromparto market.24
values of c and y in equation 1 can then be rationalizedby assumingthat transitory
consumptionis a first-orderautoregressiveprocess. Hall is correctthatit is very difficult
to discriminatebetweenpartialadjustment,whichwe tacitlyassume,andserialcorrelation
in the disturbance.However, the data stronglyrejectthe idea thatincomeis a first-order
autoregressiveprocess. Andneithertheorynorthe empiricalresultssuggestthatthe error
in the consumptionfunctionis seriallycorrelated.So we preferour interpretationover
Hall's.
24. The data necessary to do this come from W. Michael Cox, "The Behavior of
TreasurySecurities:Monthly, 1942-1984"(FederalReserve Bank of Dallas, February
1985),andwere kindlyprovidedby him.
'roooo
0
't
't
-
v
I
I
C1
rI
I
0
rI
*-
0
ON
I
e
I
00
ON
ri
't
'r
I
000t-
ON
%ir-
I
I
rI
.0
00
0ON
e\C00
0
00
00
\C
00
I
I
I
e\C
00
0
00t-ON
I
**'r
e
r1er1
0
-
I
I
0
o
0
-
0
sx
.
-
0
0
-
*rlr~~~~
0~
*o
,O
C * o a,
M
>~
M
.mPSo
O
O
O
M
~0
0
O
O
O
O
O
~~~o
O
O
1
O
D
'r
~~~~~~~~~~~~~~~~~~2.
3
C
s^?0.0
0-
^ ^ < O t t?2E
Ut
tQ
:
Ur E
O i)
o
6,
o -0
in.
03
~0I.
00
u
-0
(-
&3
. 0
_eE
E
o
c0
0
)
r
0
?~00o
M
0
o
to
N
~00
0 0-
>~~~~
~~~
x
00
03u3t
>.
-~~~-~~
~
-r
V
O.~~~~~.
0
w
zb
i
Y
=
Q;
r
2
~~)0 S
g
toC,,
~~~~~)Im.
g
-
0~
6,
- >
0
E-
b
C
>~
482
Brookings Papers on Economic Activity, 2:1985
Second, inrecognitionof the factthatCis nottotalconsumerspending,
we add a time trendto the specification.In doing this, we recognize the
dangerof pickingup a spurioustime trend.25Nonetheless, a time trend
seems calledforon economicgrounds;it couldrepresentslowly evolving
changes in tastes between durablesand nondurables,technologicalor
other changesin the availablemenuof goods, or other things.
Third,the basic PIH-LCHtheorysuggeststhatconsumptiondepends
on current wealth and on current and expected future labor income.
Income from capital is omitted because the current market value of
wealth is the best estimate of the discounted streamof income that will
be derivedfrom the assets currentlyowned. The problem,of course, is
that disposablelaborincome is difficultto measurebecause the income
tax is based on total income, not on laborandcapitalincome separately.
We constructeda real per capitadisposablelaborincome series, YL, as
follows. Startingwith the NIPA breakdownof personalincome into its
components,we apportionedproprietors'income into laborand capital
componentsin the ratioLII, where:
L is the sum of wages and salariesplus other laborincome, and
I is the sum of interest, dividends,and rentalincome.
Then we attributedpersonal income taxes to labor and capital in the
same ratio, treated all contributionsfor social insuranceas deductions
fromlaborincome, andattributedall transferpaymentsto laborincome.
Once these three changes are made, the fits of the equationsimprove
tremendously;R2 rises by about 0.13, and the standarderror falls by
about8 percent.Thereis no indicationof parameterinstability.Believers
in life-cycle theorywill findcomfortin the fact that, when regression2.4
is runusingtotal incomeratherthanjustlaborincome, therebyincluding
both wealthandthe incomefromwealthin the formulation,a non-nested
hypothesis test unambiguouslyselects labor income over total income
as the appropriateincome variable.26
25. The reasonsare discussedin N. GregoryMankiwandMatthewShapiro,"Trends,
RandomWalks,andTests of the PermanentIncomeHypothesis"(YaleUniversity,1984),
and Angus S. Deaton, "Life-CycleModels of Consumption:Is the Evidence Consistent
with the Theory?"(PrincetonUniversity, 1985).
26. The test is the Cox test describedin M.H. Pesaran,"On the GeneralProblemof
Model Selection," Review of Economic Studies, vol. 41 (April 1974), pp. 153-71. The test
rejects the hypothesis that total income is the correct variable,but cannot reject labor
income. Using laborincome, not total income, as the appropriateincomevariablemakes
irrelevantthe common argumentthat the part of interest income (and expense) that
Alan S. Blinder and Angus Deaton
483
The wealth coefficients in regression 2.3 show that the significant
wealth effect in regression 2.4 derives from unanticipatedchanges in
wealth, as suggested by life-cycle theory. This finding will persist
throughoutthe study. Because each specificationhas a sizable upward
time trend, the estimatedlong-runelasticity of consumptionto income
falls to below 0.6, and a unitaryelasticity can be rejected decisively.27
Since the impactelasticitiesareabout0.23, the dynamicsworkout faster
thanthey did in the previous regressions.
Whenincomeis redefined,the incomecoefficientschangemoderately
(compare regression 2.1 with regression 2.3, or 2.2 with 2.4), the
coefficientsof expected andunexpectedincomein regression2.3 remain
veryclose to one another,andexpectedincomeremainsquitesignificant.
On the basis of a comparisonof regressions2.4 and 2.3, the datacannot
reject the "no decomposition"hypothesis that only actual values of y
andw matter(marginalsignificancelevel = 46 percent).Butthe opposite
extremehypothesisthat only surprisesmatteris still rejected(marginal
significancelevel = 3.8 percent).
RELATIVE
PRICE
TERMS
IntertemporalPrices. All the equationsin table2 omit severalpotentiallyimportantvariables,such as inflationand interestrates. In theory,
the after-tax real interest rate should influence intertemporalchoice;
that, after all, is the basic insight of Euler equations. But what is the
effective tax rateon interestincome, and how shouldexpected inflation
be measured?
We measure one minus the effective marginaltax rate on interest
(1 - T) as the ratioof the yield on (tax-exempt)five-yearAAA municipal
representscompensationfor inflationshould be deducted from standardmeasures of
incomein arrivingat a trueconceptof Hicksianincome.However,thebehaviorof liquidity
constrainedconsumersis governedby cash flow, not by Hicksianincome.And, for them,
the needto makeinterestpayments,even if the paymentsmerelycompensateforinflation,
mightdeterconsumption.Wetestedforthisby addinginterestpaymentsto ourregressions,
butfounderraticandinsignificanteffects.
27. Throughoutthe paper, we calculate the long-runincome elasticity keeping the
wealth-incomeratioconstant, so that both directincome and indirectwealth effects are
included.If the timetrendis omitted,the estimatedlong-runelasticityis very close to 1.0.
It is hardlysurprisingthatthe presenceor absenceof a timetrendhas a dramaticeffect on
the estimatedlong-runelasticity.
484
Brookings Papers on Economic Activity, 2.1985
bonds to the yield on (taxable)five-year AAA corporatebonds. In the
sample, T varies between 0.18 and 0.39. We then constructan after-tax
nominal rate of interest by multiplyingthe three-monthTreasurybill
rateby (1 - T).28 The regressionsuse the interestrateknownat time t - 1
for carryingwealthforwardto time t because this is the naturaltimingto
use if we think of our specificationas representingthe intertemporal
choice between consumptionin periods t - 1 and t.29
Inflationaryexpectations are generatedfrom our VAR, based on the
deflatorfor total consumerspending,calledP. Both anticipatedinflation
(EAp) and unanticipated inflation (Ap - EAp) are entered into regressions
3.1 and 3.2 in table 3, the formeras a naturalcompanionto the nominal
interestrate, the latterto representthe price confusioneffect proposed
by Deaton. As noted earlier, however, we tested ratherthan imposed
the constraint that the coefficients of the nominal interest rate and
expected inflationbe equal and opposite-as would be true if only the
real interestrate mattered.
The regressionresults suggestthat nominalinterestrates are of little,
if any, importancefor consumption.Comparingregressions3.1 and 3.2
shows that the coefficientof rt- I is much largerin absolute value in the
surpriseversion, that is, when contemporaneousvariablesare instrumented.This was true in every specification.30
Inflation,on the otherhand,is quite significantin both specifications,
thoughonly in unanticipatedformin regression3.1. Note, however, that
expected inflationgets the wrong sign in regression 3.1. If only real
interest rates matter,EApand r should have equal and opposite coefficients. Thoughthis hypothesiscannotbe rejectedat the 10percentlevel
owing to large standarderrors, the sum of the coefficients is far from
zero. Why it shouldbe nominalinterestrates, not real rates, that matter
is a puzzle that has cropped up in other contexts.3' We have no good
28. As an alternative,we also tried using the five-yearmunicipalbond rate. Results
were similar.Since the shortrate makesbettertheoreticalsense, and usuallyprovideda
slightlybetterfit, we reportonly those results.
29. In responseto suggestionsfromsome readers,we also triedaddingcurrentr to the
regression.See footnote30.
30. Whenr,wasaddedto theregressions,theresultsdidnotchangemuch.Specificially,
r, and r,_ generally got large coefficients with opposite signs, reflectingcollinearity
betweenthe two variables.The sumof the two coefficientswas smallandnegative.
31. See, for example, ChristopherSims, "Comparisonof Interwarand Postwar
BusinessCycles:MonetarismReconsidered,"AmericanEconomicReview,vol. 70 (May
1980, Papers and Proceedings, 1979), pp. 250-57; Robert Litterman and Laurence Weiss,
Alan S. Blinder and Anguis Deaton
485
explanationto offer. But, in any case, the best guess at this stage is that
the significantcoefficientof inflationin regression3.2 reflects the price
confusioneffect.
There are no notablechanges in coefficients when interestrates and
inflationare addedto the regressions,althoughthe hypothesis that only
surprisesmattercan no longerbe rejected.
Relative Prices of Different Goods. In principle, consumers should
respondnot only to the relativeprice of the same good in differenttime
periods,butalso to the relativeprices of differentgoods in the same time
period.This is easy to test.
From among the prices of the three main components of consumer
spending-PS, the price of services; PND, the price of nondurable
goods; andPD, the price of durables-we selected the lattertwo for the
constructionof relativeprice variables.Regression3.4, the no-surprise
specification,adds the two relative prices in both currentand lagged
form; in logs, the new variables are (pn - p)t and (pn - p)t,I for
nondurables and (pd - p)t and (pd - p),- for durables. However, when
we tried to enter anticipated,unanticipated,and lagged relative prices
in the surprisespecification,regression3.3, the anticipatedand lagged
variableswere almostperfectlycorrelated.Consequently,3.3 omits the
laggedrelativeprices.
In addition,our regressionsought to include any other variablethat
has a stronginfluenceon the allocationof incomebetweendurablegoods
andC. The stock of durablesat the startof the quarter,Kt,is an obvious
candidate,since, given the desired stock, a higheropeningstock ought
to lead to lower spendingon durables,and hence to higherspendingon
C, otherthingsbeingequal.
Regressions 3.3 and 3.4 show a substantialimprovementin fit over
theirpredecessors-R2 rises by about0.08, andthe standarderrorof the
equationfalls by about5 percent.Five of the eightrelativepricevariables
in the two regressions are significant.In particular,the large negative
coefficients of (pn - p)t in regression 3.4 and of the surprise in (pn - p)t
in regression 3.3 both suggest a strong transienteffect of the fl'-ave
r
price of nondurablegoods. In either specification, a 1 percent rise in
"Money, Real InterestRates, and Output:A Reinterpretationof PostwarU.S. Data"
(Federal Reserve Bank of Minneapolis,January 1984);and Alan S. Blinder, "Retail
InventoryBehaviorandBusinessFluctuations,"BPEA,2:1981,pp. 443-505.
-s-
-s
s
-s
I--, I--, I--,
-s
0
I--, I-I--, ,
-
I
-s
-s
I
I
N-
I
t ?t
N
L
.2~
oo
O
-
PI,
ON
t
N'o0
~ ~
s
~
It
~
aN
-
(sN
~
~
O N
N
N
(N- --
0s
~
I
N
I
I
I
I
~~~~~~~~~~~~~~~~~~I
I
I
-s
~
00
O ONN
O
0
o
-
s
eCx
'IC
-
N
CN
0
(N
-
O0
(N
0
Cq
a:~~~~~~~~~~~~~~~~~~~~~C
~~~~~~~~
0N'I
--
00
X~~~~~~~~~1
CZ
l
EN
CU
0
0
CZ
CZ
0
4- 4
CZ
t
<
<
-
o00
?
0~~~~~~~~~~~~0
o
.4
0
U
0
CZ<
0
N
o
-Na
o
t
00
-e'
~00 7-~j
It
Itos
0
C1o
"C
-
"C o
oot
0
rO
.
.
.
ro::0
0
o
o-
0-
0
*
t-
.
CZ~~~~~~~c
<~0
.
^
-t oC
C
CZ
o~~~~~~~~~~~~~~~
o~~~~~~~0
~~ ~
CZ
CZ~~~~~~~~~~~~~~~~~
.
) 0
CZ~~~~~~~~~~~~~~~~~~
CZ
0~~~~~~~~~~~~~~~~~~~~~~~~c
cd
~~~~~~~~~~~C-
,0
2
Z
0
O.nu
0-
C
-~~~~~~~~
0~~~~~C
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Z
-
U
cc
Dz
-~~~~~~~~~~~~~~~~~~~~~~~~E
Gw
CZ
488
Brookings Papers on Economic Activity, 2:1985
PND relative to P (only if unanticipatedin regression3.3) reduces the
annualgrowth rate of C by about 0.9 percent in the first quarter.The
estimated steady-state (or anticipated)effects of pn - p are positive,
however.32
Resultsare weakerfor the relativepriceof durables,pd - p, although
the steady-stateeffect appearsto have the correct(positive)sign in both
specifications. While the coefficient of the opening stock of durables
gets insignificantcoefficients in both specifications, we leave it in the
regressionfor theoreticalreasons lest it interactin importantways with
any other variable.
Perhapsmore interestingthanthe estimatedrelativeprice effects per
se is the way the additionof relative prices changes some of the other
coefficients.Two changesareparticularlynotable.First, regression3.3,
unlikeregression3.1, stronglysuggeststhat only unanticipatedchanges
in income cause consumption to change. Second, the conclusion in
regression3.1 thatunanticipatedinflationmattersmorethananticipated
inflationis dramaticallyreversedin regression3.3. In 3.3, and in the rest
of the regressions estimated for this study, the effect of inflationon
consumption seems to derive from anticipatedinflation. The interest
elasticity is a bit stronger, though still not significant,when relative
prices are included.And the constraintthatthe coefficientsof r andElVp
are equal and opposite can now be rejected stronglyin regression 3.3
(marginal significance level = 0.5 percent).
Not surprisingly,given these changes in coefficients, we can now
reject, at the 0.8 percent level, the hypothesis that the split between
anticipatedand unanticipatedcomponentsdoes not matter.This result
nominatesregression 3.3 as our best consumptionfunction. In consequence, we displayin the appendixthe series on unanticipatedincome,
wealth,andinflationthatunderlieregression3.3. Nevertheless, we retain
both 3.3 and 3.4 as baseline specificationsto be used in the next section
because regression 3.3, but not 3.4, shows some slight evidence of
parameterinstabilityacross half samples,andbecause some economists
are,justifiably,skepticalof our methodfor decomposingvariablesinto
32. Thetheoreticallyexpectedsignforthe relativepriceof nondurablegoods is unclear
because the consumptionmeasureincludes services as well. Note too that the greater
importanceof the "surprise"as comparedwith the anticipatedrelativeprices could be
interpretedas evidence of simultaneitybetweenconsumptionandprices. See ourdiscussion above.
Alan S. Blinder and Angus Deaton
489
anticipatedand unanticipatedcomponents. Note also that our surprise
equationis not the pure surpriseversion of the PIH, because it includes
both laggedand anticipatedterms.
Further Investigation of Interest Sensitivity. The implied steady-
state semi-elasticityof consumptionto the nominalrate of interest in
regression3.3 is - 2.3. That means that, with the path of income held
constant, a 1 percentagepoint rise in r will eventuallydecrease C by 2.3
percent-a strongeffect. The result, however, seems to be fragile.The
strong elasticity is to the nominal interest rate and does not appearif
only the real rate is allowed in the regression. Furthermore,the strong
negativeeffect of the nominalinterestrate appearsonly in the surprise
version;the correspondingsemi-elasticityin the no-surpriseregression
is only -0.8. Since there has been so much concern lately with the
sensitivity of saving to the rate of return, we decided to look at the
interestelasticity issue more deeply.
Regressions4.1 through4.4 in table4 disaggregateC into its two main
components, nondurablegoods and services, and show, much to our
astonishment,thatit is actuallyspendingon services, not on goods, that
is sensitive to interest rates. We have a hard time believing that this
sensitivity represents intertemporalsubstitution, but cannot offer a
better explanation.The "equal and opposite" constrainton the coefficients of nominalinterestrates and inflationcontinues to be rejectedat
very exacting significancelevels for services, but not for nondurable
goods.
Dividing services into five component parts (housing, household
operation, transportationservices, other services, and state and local
nontax receipts) shows that the significantnegative interest elasticity
can be tracedmainlyto housingandtransportationservices. It may seem
surprisingat first that spending on these services should be interest
elastic, but each of these service flows relatesdirectlyto a durablestock
(houses and automobiles), the demand for which is probably quite
sensitiveto interestrates.33Thatobservationdoes not, of course, explain
why it is the nominalinterestrate that matters.
33. Severalreaderssuggestedthat the negativeinterestelasticityof housingservices
is an artifactof the BEA's imputationprocedures.That does not appearto be the case,
however.Services of owner-occupiedhouses are imputed,of course. But the imputation
is basedon actualobservedrent-to-valueratioson rentedhouses, deflatedby the consumer
price index for rent. There is no mechanicallinkagebetween interestrates and imputed
housing services.
.z
W')
CIA
2-
Cq
0
V
0
CiN
)
Ci
CI t-
=
-0
00
I*
^
~~~~~Ci
~II ~
~ ~
c7N
I
=
0
'
'IC
r
0
?
OOCiCi
Cf-
It
ON
O
ON0
0
N
I
0
N 'jIC
iN
0
AO
I
kn
-0
'q
#N
Ci~~~~~~~~~~~~~~~-0
CiA
'~~
CiO
~~I
OtN
I
II
I
0oNNNN
L<
f4Xo
N
Cu0
o~~~~~~~~~0
00~~~~~~~A
04
C
t
~^
O.
q
~
0o
0
O0)
t
0~~~~~~0E
O
"0
N??
"
o?
X
CU
0
)
0
O
o
CNO
OOO
OOONC
v
COO
0)CO
CZ-0
O
0)
z
E.
CZ^
0 0Jt-*1
-0--
CZ- Cq
CZ
t~~~~~~~~~"00
CZ b ? V-n N b o o o?*Vm
bt
^
vm
,
NO
0)
o
aN
C;
aN
rq
L4
CD
CN
0
0
aN
'IC
L4
0
00
aN
CN 'IC
It C aN
aN It "C
"C
CZ
7
CZ
CZ
7
7
7
CD
CZ
CZ
(Ueq 20
E
010 7
CZ>,
ts
VI
4-4
4.4
-cs 0
0
to
m
0 0
0,
Es
-5
z z
e
tw
tt
492
Brookings Papers on Economic Activity, 2:1985
As we had expected, the negative interest elasticities that we find
cannot be detected over a shorterdataperiodthat excludes the volatile
1980s, even thoughthe equationspass formal stabilitytests. When we
ranthe regressionsin tables 3 and 4 over a sampleendingin 1979:4,the
estimated interest elasticities were either insignificantlynegative or
positive.
THE BASELINE
CONSUMPTION
FUNCTION
Hereafter, we shall take equations 3.3 and 3.4 to be our baseline
consumptionfunctions, so it is worthdwellinga momenton the steadystate properties of the two. The implied steady-state consumption
functionsfromregressions3.3 and 3.4, respectively, are:
C = 7.85 yO.78 W0.42 K0005 (PND)
1.28(PD)
0.79e0.0019t 2.3r-8.8Ap
and
C = 6.76 Y053
WO25 K0015 (PND).31
PD
0.22 0019t-0.8r-1.5p
The two steady-statespecificationsdiffermainlybecause of different
coefficientson ct Iinregressions3.3 and3.4. Thisillustratesthe difficulty
of inferringlong-runpropertiesfromshorttime series.
Not surprisingly,most of the biggestresidualsoccur when c changes
the most;butno residualin the entiresampleis as largeas threestandard
errors.The biggestcomes in 1983:2,when ourregressionfails to capture
the huge (and transitory)accelerationof consumer spending. We also
miss the upward"blip" in spendingthat occurredin 1965:4.A plot of
the residuals (not shown) has no evident runs of positive or negative
residuals. In particular,there is no tendency for the regressions to
overpredictconsumptiongrowth (that is, produce negative residuals)
after the introductionof several special incentives for saving in the
August 1981 tax bill; in regression 3.4, the mean residual for the 13
quarters1981:4-1984:4is actually + 0.07 percent.
GovernmentPolicy and Consumption:Tests of Hypotheses
This section tests a variety of more controversialhypotheses about
how consumersreact to changes in governmentpolicy, such as tempo-
Alan S. Blinder and Angus Deaton
493
rarytaxes and budgetdeficits. In each case, the baseline consumption
functionof the previoussection serves as a startingpoint.
REACTIONS
TO TEMPORARY
INCOME
TAX
CHANGES
Friedman'spermanentincome hypothesis suggested long ago that
consumersshouldreactto temporaryincome tax changesless thanthey
do to permanentones, and Lucas made that point part of his famous
critiqueof econometricpolicy evalution.Furthermore,severalempirical
studies have detected such a difference. In the most comprehensive
study of the issue, Blinderestimatedthat consumerstreata temporary
tax changeas roughlyhalfordinaryincome change, halfpurewindfall.34
To study this issue, we begin, as Blinderdid in his previouspaper,by
dividing income along lines that differ from the usual permanenttransitorydistinction. Specifically,we define "special" income, S, as
the disposableincome, positive or negative,derivedfromtemporarytax
changesand "regular"income, R, as the rest; thatis, R, = Y, - S,. The
motivationfor this dichotomyis thatspecialincomeis observably"more
transitory"than regularincome, which is, itself, a blend of permanent
andtransitorycomponents.Because of this difference,the PIH suggests
that S shouldhave a smallereffect on consumptionthandoes R; that is,
the appropriateincome variableshouldbe:
R + RS = Y-
(1 -
)S = Y[1 -
(1 - R)S
where ,uis an empiricallyestimatedcoefficientbetween zero and one.
Since the log of a sum is not the sum of the logs, we approximatethe
naturallog of this income concept as follows:
lnY + ln[1 - (1 -
r)]
lnY- (1
-
)
which is an excellent approximation,since the income involved in
temporarytax changes is small, always less than 4 percent of other
income. Thus, to estimate p.,we simplyneed to add a new variable,the
ratio SIY,to our regressions. Since y = lnY enters in both currentand
laggedform, S/Yenters in currentandlaggedformas well. This gives us
34. See Blinder,"TemporaryIncomeTaxes."
494
Brookings Papers on Economic Activity, 2:1985
two differentestimates of (, - 1): an impact effect derived from the
ratioof the coefficientof (S!Y),to thatof y, anda long-runeffect derived
from the sums of the (Sl!j and y coefficients.35The pure PIH suggests
that ,u should be near zero, while ,u = 1 means that actual measured
income is the relevantincome concept.
Before lookingat the results, we need to explainhow we constructed
the S, series. We distinguishedfour different temporarytax change
episodes.
The 1968-70 tax surcharge. A one-year temporary tax surcharge
beganin 1968:3and was subsequentlyextended, at reducedrates, for a
second year. The 1968-70 episode was studied by Arthur Okun and
William Springer with conflicting results.36For this study, revised
estimatesof the revenuesfromthe surtaxduringthe eight quartersof its
existence were takenfrom the May 1978issue of the Surveyof Current
Business.
The1975rebate. A rebateof 1974taxes was paid mostly in May and
June of 1975, with traces tricklinginto the thirdand fourthquartersof
1975.This episode was studiedby Modiglianiand CharlesSteindel.37
The other 1975 cuts. The 1975 tax act also reduced bracket rates
"temporarily"for one year. Since these rate cuts were subsequently
extended in the Revenue AdjustmentAct of 1975and again in the Tax
ReformAct of 1976,and eventuallybecame a permanentfeatureof the
tax code, we hadto makeanarbitraryassumptionaboutwhenconsumers
ceased viewingthemas temporary.We assumedthatconsumersshifted
linearly from viewing the cuts as 100 percent temporaryin 1976:1to
viewingthemas 100percentpermanentby 1977:1. All threeof the abovementionedepisodes were studiedby Blinder;38
but dataon both 1975tax
reductionshave been revised since then, andwere kindlyprovidedto us
by the BEA.
TheReagan tax cuts. As mentionedin the introduction,the Reagan
tax cuts of 1981-84 have given us another "temporarytax" episode,
thoughone of a differentcharacterthatmayhave been treateddifferently
35. In the surpriseversion,we sumthe coefficientsof laggedandanticipatedvariables
to obtainthe long-runeffect. Thisprocedureis followedthroughout.
36. Okun,"The PersonalTax Surcharge";Springer,"Did the 1968SurchargeReally
Work?"
37. ModiglianiandSteindel,"Is a Tax Rebatean EffectiveTool?"
38. Blinder,"TemporaryIncomeTaxes."
Alan S. Blinder and Angus Deaton
495
by consumers.Froma strictPIH viewpoint,one shouldtreatthe phasedin tax cuts as a large permanenttax reductionin the middle of 1981:3
coupledwith a temporarytax increasethatdiminishedgraduallyto zero
by 1984:1. The series is offered for inspection in table 5 and explained
with the aid of the followingnotation:
A, is incomefromthe actualReagantax cuts in quartert (mostly BEA
data from the April 1985 Survey of Current Business);
T,is income from the theoretical "permanent"tax cut applicableto
quartert (constructedby us);
S, is income from the theoretical temporarytax hike applicableto
quarter t (A, - T,).
We dealt separatelywith each of the three reductionsin withholding
rates and the nonwithheldpartof the liabilities.
October 1981 withholding reductions. Though the Reagan tax cuts
were enacted in August 1981, the first stage was not effective until
October.We thereforeset T,for 1981:3equal to half the BEA value for
A, in 1981:4. Since A, = 0 in 1981:3, St = - Ttthat quarter. Thereafter,
the October1981cut was fully effective, so St fromthis source is zero.
July 1982 withholding reductions. BEA data assume that the July 1,
1982,rate reductionsbecame effective on that date. But tax rates apply
to calendaryears, so half the July 1 cuts actually appliedto the entire
calendaryear. So we began by changingthe BEA series on At by using
quarterlywithholdingpaymentsto allocatethe calendar1982tax reduction ($13.3 billionat annualrates) to the four quartersof 1982.We then
estimatedT,for each quarterof 1982by assumingthatthe full July 1 rate
reductions were effective throughoutthe year. Having done that, we
constructedStas At - Tt.For quartersbeyond 1982:4,the full July 1982
ratereductionswere in effect, so Stfromthis source is zero.
July 1983 withholding reductions. The procedure used here is exactly
the same as thatfor the July 1982reductions.
Nonwithheldtaxes. The BEA provideddata on At. We constructed
Ttby applyingthe 1984ratioof nonwithheldtax cuts to withheldtax cuts
to the quartersfrom 1981:4through1983:4(halfthe ratein 1981:3).Then
St was At - Ttas usual.
There is every reason, however, to treat the St series for Reagan's
"temporarytax hike" differentlyfromthe othersbecausethe otherthree
episodes were genuinechangesin incomeflows receivedby households,
while the mythical Reagan "hike" is a theoreticalconstruct based on
Brookings Papers on Economic Activity, 2:1985
496
Table 5. Estimated Consumer Income from Reagan Tax Reductions, 1981:3-1983:4a
Billions of dollars
Quarter
Actual
Permanent
Temporary
1981:3
1981:4
0.6
18.6
46.0
88.0
-45.4
- 69.4
1982:1
1982:2
1982:3
1982:4
39.7
40.4
40.9
41.3
91.1
93.4
95.4
96.8
-
51.4
53.0
54.5
55.5
1983:1
1983:2
1983:3
1983:4
81.7
83.5
85.1
87.6
97.0
99.4
101.7
105.0
-
15.3
15.9
16.6
17.4
Source: Authors' computations.
a. Starting in 1984:1, "actual" and "permanent" are equal, so "temporary" is zero.
the PIH. The following proceduresuggests itself. Suppose the income
concept relevantto consumptionin 1981:3-1983:4was:
Yt + a(YtP - Y,) = Yt + a( -S,*),
where YtP- Y
(-S*)
is the income from promised future tax cuts
that is currentlywithheld;this series is given, with sign reversed, in the
lastcolumnof table5. Thuswe see thata is analogous,butnot necessarily
equal, to the 1- ,uparameterfor the otherthreetemporarytax episodes;
thatis, a = 1 connotes the purePIH, while a = 0 meansthatconsumers
ignoredthe promisedfuturetax cuts.
Ratherthan repeat long tables of regressioncoefficients that hardly
change, we report in table 6 and in the tables that follow only the
estimates of the new coefficients and parametersof interest, plus the
relevanttest statistics.We willnote anyimportantchangesin coefficients
as we proceed. When temporarytax change variableswere added, the
most notable changes were an increase of about 50 percent in the
coefficientof lagged consumptionand a doublingof the estimatedtime
trend.In consequence, the estimatedlong-runelasticityof consumption
to income fell.
In table 6, the variable SlY pertains to the 1968-70 and 1975-76
temporarytax changes, while S*!Y pertains to the 1981-83 temporary
tax change. For both episodes, the point estimates suggest that the
effects of temporarytax changes are near zero, that is, that they are
ignoredby consumers. The long-runeffects are also near zero for the
Alan S. Blinder and Angus Deaton
497
Table6. Tests for TemporaryTaxes, 1954:1-1984:4a
Regression
Variable,parameter,and test
Ratio of income from temporarytax
to total laborincomeb
1968-70 and 1975-76 episode
(SlY1,
(S/Y1, l
1981-83 episode
WM*lY),
WM*lY),-,l
Parameterc
Long-runincome elasticity
Estimatedp.
Impact
Impactand lagged
Estimateda
Impact
Impactand lagged
Surprise
No-surprise
-0.152(- 1.2)
- 0.077(-0.7)
-0.200(- 2.0)
- 0.120(- 1.2)
- 0.014(-0.1)
- 0.010(- 0.1)
-0.014(-0.1)
0.041 (0.3)
0.83
0.62
0.12
- 1.37
- 0.06
- 1.51
0.08
0.24
0.07
-0.22
Test of hypotheses
(marginal significance levels)d
Strict PIH (p.= 0, a = 1)
Measuredincome (p.= 1, a = 0)
Unit elasticity
Real interestrate
No decomposition
No decompositionfor income and wealth
Surprisesonly
0.49
0.50
0.64
0.011
0.074
0.24
0.98
0.059
0.058
0.017
0.041
...
...
...
Sources: Same as table 2.
a. The dependent variable is the change in log of consumption of nondurable goods and services. New variables
are added to baseline equations 3.3 and 3.4.
b. These are coefficients of the indicated variables in the regressions.
c. As described in the text, the parameters are estimated as the ratio of coefficients from the regression.
d. This is the probability, expressed in a decimal, of getting the indicated coefficients if the hypothesis is correct.
Reaganepisode, but are actually strongly negative for the pre-Reagan
episodes-which meansthatconsumptionmovedin the wrongdirection.
These estimates are quite different from Blinder's 0.50 estimate of a
parametersimilarto ,upublishedfouryearsago, basedon unreviseddata
and a very differentspecification.The source of this discrepancyis an
intriguingquestionfor futureresearch.
The nullhypothesisa = 1and ,u = 0 correspondsroughlyto the strict
PIH in that it says that the relevant income concept excluded the 1968
and 1975-76temporarytax changesbut treatedthe Reagantax cuts as if
498
Brookings Papers on Economic Activity, 2:1985
they were fully effective in August 1981.This hypothesiscan be rejected
in the no-surpriseversion of the model, though only at the 6 percent
level; it cannotbe rejectedat any reasonablelevel in the surpriseversion,
owing to largerstandarderrors.The opposite extreme hypothesis, that
a = 0 and ,u = 1, says thatconsumersspendon the basis of conventional
measuredincome, with no specialallowancefor temporarytax changes.
This hypothesis againcan be rejectedin the no-surpriseversion only at
the 6 percentlevel, and not at all in the surpriseversion.
While these results are probably not precise enough to persuade
anyone to abandonstronglyheld a prioriviews, they do point towarda
mixed conclusion (a = 0 and ,u = 0) in which consumersdo not spend
on the basis of income received fromtemporarytax cuts (in accord with
the PIH) but nonetheless ignoredthe promisedtax reductionsof 198184 untilthey actuallyoccurred(an anti-PIHresult).
Because the 1975rebatewas received so late in the second quarterof
1975, we tried estimating separate coefficients for the rebate on the
suppositionthat (SIY),1 mightbe relativelymore importantthan (S!Y),
for the rebate than for other temporarytax episodes. This did not turn
out to be true, however. We also triedto distinguishbetween anticipated
and unanticipatedtemporarytax changes. But the estimateswere based
on so little datathat the standarderrorswere enormous.
IS THE
GOVERNMENT
VEIL
PIERCED?
Ourpreliminarylook at the data in the introductorysection made it
appearas thoughhouseholds (or households and businesses combined)
did not raise their savingto offset the rise in governmentdissavingafter
1981,as would be expected if debt and taxes were equivalent.Here we
investigatethe Barroequivalencehypothesismore rigorously.
If the Barrohypothesis is right, our consumptionfunctionis misspecifiedin two ways. First, the marketvalueof the governmentdebt should
be omitted from household net worth. This is easily done, thanks to
extremely accurateestimates of the marketvalue of the debt compiled
and kindly provided to us by Michael Cox.39 To test the Barro hypothesis
against the traditionalspecification,it is naturalto replace lnW in our
regressions by ln(W - D + RI1D),where D is the market value of
39. Cox, "The Behaviorof TreasurySecurities."
Alan S. Blinder and Angus Deaton
499
governmentdebt. If RI'= 0, the Barrohypothesis is correct;if RI = 1,
the full governmentdebt is included in wealth (that is, there is no tax
discounting).Using the same approximationas before, this amountsto
replacing lnW by two variables: ln(W - D) and D!(W - D). The ratio of
the two coefficientsis an estimateof RI.
Second, if Barrois right,disposablelaborincomeis the wrongincome
concept. But what income concept is right?We answerthis questionin
stages, first supposingthat we were workingwith total income rather
thanjust laborincome.
Consumerswho pierce the governmentveil presumablyunderstand
thatthe incomeactuallyavailableto themis net nationalproduct(NNP)40
minus governmentpurchases (G) minus transfersand interest paid by
the U.S. governmentto foreignersplus state and local personalnontax
payments.41Callthis concept of income YB.Manipulationof some NIPA
identitiesreveals that:
(4)
YB = Y + RET + SURP,
where Yistotaldisposableincomeas definedby us, S URPis the ordinary
NIPA governmentsurplus,andthe inclusionof RET,retainedearnings,
reflectsthe hypothesisthat the corporateveil is also pierced.
Now we must come to gripswith the fact that our baseline specification, regressions3.3 and 3.4, uses labor income, not total income, and
thatidentity4 does not hold in laborincome. We beginby defining
DIFF=
YB-
Y
as the difference between the Barro concept of total income and our
own. Since retentions are irrelevantin a specificationbased on labor
income, we eliminate them, which makes DIFF = SURP.42 Assume
thata portionX, = L!(L + 1)of the surplusgets into laborincome, where
40. Seaterand Mariano,in "New Tests," use GNP ratherthanNNP. We do not see
the argumentfor includingdepreciation.
41. Becausetransfersandinterestpaidby the U.S. governmentare sent abroad,they
are not availableto domestichouseholds.State and local personalnontaxpaymentsare
consideredto be personalconsumption,as explainedabove.
42. To check thatthis decisiondid not bias the case againstthe Barrohypothesis,we
ranregressionsthatincludeda retentionsvariableandtested the restrictionsthatexclude
this variable.The restrictionscould not be rejected.We also tested the Barrohypothesis
on the assumptionthatretentionsshouldbe includedas partof laborincome. The results
were not sensitiveto the way retentionswere handled.
500
Brookings Papers on Economic Activity, 2:1985
X,is the ratiowe used earlierto definelaborincome. Then a reasonable
concept of laborincome if the Barrohypothesisis correctis:
YBL
=
YL + XSURP.
Hence, using the usual approximation,we can replace ln(YL)in our
regressionsby:
ln(YL + OXSURP) = lnYL + OX
SURP
If Barrois right,0 = 1, which means that we shouldredefineincome by
adding back labor's share of taxes and subtractinglabor's share of
governmentpurchases.
But thatprocedurewould saddleBarro'shypothesiswith an auxiliary
constraint that is no part of the tax discounting idea, namely that
governmentpurchases have no direct effect on the marginalutility of
privateconsumption.If, instead,governmentnondefensepurchases,G,
are either substitutesor complementsfor privateconsumption,there is
no reasonto enter G and taxes, net of transfers,with equalandopposite
signs, as wouldbe done if we used the variableSURP. To allow for this,
we add to the regressiong = logG, both contemporaneously,decomposed into anticipatedand unanticipatedcomponents in the surprise
version, and lagged.43
The results are reportedin table 7. The coefficientson the current,or
surprise,versions of the new income variableX, (SURP/YL), are very
small, but those on the lagged, or anticipated,termsare largeenoughto
suggest that a substantialfraction(55 percentin the no-surpriseversion
and 92 percent in the surprise version) of future taxes is eventually
discounted back to the present. The coefficients of the new wealth
variablesimply estimates of RIof 1.0 or more. Hence, while the point
estimates of the income coefficients give some support to the Barro
hypothesis, the wealthcoefficientsare the opposite of what it requires.
This ratherodd combinationof coefficientsprobablyreflectsthe fact
thatthe bigmovementsin bothof the specialBarrovariables(forincome
and for wealth) came at the same time. Unfortunately,because some
43. The coefficientsof governmentspendingwere generallypositive and of marginal
significancein the regressions,indicatingthat governmentpurchasesand private consumptionarecomplementsratherthansubstitutes.
Alan S. Blinder and Angus Deaton
501
Table7. Testsof BarroHypothesis,1954:1-1984:4a
Regression
Variable,parameter,and test
Barroincome variableb,c
Lagged
Anticipated
Unanticipated
Ratio of governmentdebt to other wealthc
Lagged
Anticipated
Unanticipated
Surprise
No-surprise
- 0.009(- 0.1)
0.094 (0.7)
0.008 (0.2)1
0.039(1.0)
0.002(0.05)
.0(.5
-0.203(-0.6)
0.257 (0.6)1
0.157 (0.9)0
0.012(0.1)
0176(1 3)
Parameterd
Long-runincome elasticity
Estimated0
Impact
Impactand lagged
Estimated[t,
Impact
Impactand lagged
1.77
1.23
0.05
0.92
0.01
0.55
1.18
3.54
2.75
3.30
0.51
0.72
0.66
0.48
0.061
0.15
0.54
0.033
0.68
0.016
0.060
0.63
0.53
...
...
Test of hypotheses
(marginal significance levels)e
Barrohypothesis(0= 1, p,=0)
Traditionalhypothesis(0=0, t=, 1)
Unit elasticity
Real interestrate
No decomposition
No decompositionfor income and wealth
Surprisesonly
Half-samplestability
Reaganstability
0.0027
0.26
Sources: Same as table 2.
a. The dependent variable is the change in log of consumption of nondurable goods and services. For each variable,
the anticipated variable is defined as Ext, the unanticipated variable as xt-Ext.
New variables are added to
baseline equations 3.3 and 3.4.
X
b. The variable is defined as X(SURP/YL), where is the labor share, SURP is the government surplus, and YL
is labor income.
c. These are coefficients of the indicated variables in the regressions.
d. As described in the text, the parameters are estimated as the ratio of coefficients from the regression.
e. This is the probability, expressed as a decimal, of getting the indicated coefficients if the hypothesis is correct.
xt,
coefficientssupportBarroneutralityand others contradictit, the equation's implicationsfor the neutralityhypothesis are not transparent.To
understandbetter what the coefficients mean, we ran a simulationin
whichpersonalincome taxes fell by $25per capita($100at annualrates)
in 1981:4only, leaving the governmentdebt higher by $25 per capita
Brookings Papers on Economic Activity, 2:1985
502
thereafter,44
with all other variablesin the regressionheld constant.45If
the neutralityhypothesis were correct, such an event would have no
effect on consumption.The actual simulatedeffects on the level of real
per capita consumptionof nondurablegoods and services are given in
table 8, where they are juxtaposed against the results implied by the
traditionalmodel (0 = 0, RI = 1). The results are far closer to the
implicationsof the traditionalview than they are to the zeros called for
by the Barrohypothesis.
In the no-surprise specification, a formal test rejects the Barro
hypothesis at the 1.6 percentlevel, though,with a marginalsignificance
level of 6 percent, the traditionalview does littlebetter.The muchlarger
standarderrorsof the surprisespecificationprecluderejectionof either
extreme hypothesis. These equationsare the first ones in this paper to
show severe parameterinstability.Surprisingly,given the specification
and the recent behaviorof the governmentbudget deficit, it is the test
for stabilityover halfperiods, not the test comparingthe Reaganperiod
with the rest of the sample, that fails miserably. Tests of the Barro
hypothesisover a sampleperiodendingin 1981:2yield results similarto
those in table7. Finally, we note thatthe additionof the Barrovariables
changes a numberof the other coefficients. But, to conserve space, we
do not detailall these changes.
TAX
DISCOUNTING
WITHIN
A YEAR?
The standardPIH-LCH without bequests can be thought of as the
hypothesis that households smooth consumptionover their lifetimes.
Similarly,the Barrohypothesiscan be thoughtof as the hypothesisthat
dynastiessmoothconsumptionover periodsmuchlongerthana lifetime.
Ourbaseline specification,regressions3.3 and 3.4, supportsone importantimplicationof the standardPIH:thatonly unanticipatedmovements
in income andwealthlead to changesin consumption.But our results in
44. The governmentdebtvariablein the modelis realdebt percapita,whichnaturally
falls as populationandpricesrise.
45. Amongthe othervariablesheld constantis wealthexclusive of governmentdebt,
which falls if spendingrises. We cannot allow for this reaction, however, because our
measureof consumptionexcludesdurables.Hence, thenumbersintable8 slightlyoverstate
the spendingstream.
Alan S. Blinder and Angus Deaton
503
Table 8. Effects on Consumption of a Debt-Financed Temporary Tax Cut, 1981:41984:4a
Dollars
Surpriseequation
No-surpriseequation
Surprisetax cut
Estimated Traditional Estimated Traditional
Quarter effects hypothesis
effects hypothesis
1981:4
1982:1
1982:4
1983:4
1984:4
11.39
- 0.11
1.67
2.34
2.46
11.52
2.04
1.27
0.92
0.81
9.88
6.52
2.72
0.53
-0.31
10.37
6.24
4.74
3.82
3.37
Expectedtax cut
Estimated Traditional
effects hypothesis
2.65
1.24
1.11
1.00
0.97
8.79
4.35
2.29
1.10
0.60
Source: Authors' computations based on the equations in table 7.
a. Real per capita consumption of nondurable goods and services.
table 7 offer at least some evidence against the Barro equivalence
hypothesis.
A test of a much weaker smoothing hypothesis-that consumers
smooth their spendingover the annualApril 15 tax settlement dateprovides a useful cross check on our previous results. That consumers
do at least this much smoothingmay seem self-evident, but newspaper
reportslast winter claimed that consumer spendingwas low in 1985:1
because of InternalRevenue Service delays in processing tax refunds.
We can test the "smoothingover April 15" hypothesis because, while
official NIPA data use income tax payments rather than income tax
liabilities in defining disposable income, the BEA also publishes a
quarterlytime series on tax liabilities.46
A weak requirementof the PIH oughtto be thatconsumersbase their
spendingplans on income net of tax liabilities, not net of payments.
When TP is defined as payments and TL as liabilities, the concept of
laborincome relevantto consumersis:
YL + X(TP
-
TL),
ratherthanjust YL. The assumptionis thatlabor's sharein tax liabilities
46. The sample period is one year shorterfor these tests because 1984data on tax
liabilitiesare not yet available.Data sources are as follows: for 1953-75,U.S. Bureauof
EconomicAnalysis, Survey of Current Business, vol. 58 (May 1978);for 1976-79,Survey
of Current Business, vol. 63 (January1983);for 1980,Survey of Current Business, vol. 64
(April1984);for 1981-83,Survey of Current Business, vol. 65 (May 1985).
504
Brookings Papers on Economic Activity, 2:1985
is the same as labor's share in tax payments. To test whether it is
liabilitiesor payments that really matter, we replace lnYLwherever it
appearsby:
ln[YL +
2 A(TP -
TL)]
lnYL +
i2X(TP
-
TL)
YL
The test, however, is not likely to have much power because the
varianceof X(TP - TL) is smallcomparedto that of YL.
Table 9 summarizesthe results. Of the four impliedestimates of P2,
all but one, which is very poorly determined,are fairly close to unity.
Althoughthe formaltests cannotrejecteitherextremehypothesis, P2 =
Oor 2 = 1, owing to largestandarderrors,they preferthe latter. Given
that low discriminatorypower was to be expected, there appearsto be
no evidence against short-runsmoothing here; so we are inclined to
accept the theoreticalpredictionthatliabilities,not payments,influence
consumption.
Conclusion: What Have We Learned?
The hypotheses we have tested suggest one main change in our
baselineconsumptionfunctions:thatthe incomederivedfromtemporary
tax changes should be eliminatedfrom disposablelaborincome. In our
originalbaselinesin table3, we madethis adjustmentfor the 1975rebate,
but not for the other temporarytax episodes. The regressionsreported
in table 10makethis change.
For several reasons, we tend to think of the surpriseregression in
table 10as ourbest consumptionfunction.First,the "no decomposition"
hypothesisthatleads to the no-surprisespecificationis soundlyrejected.
Second, the surpriseregression has smaller unexplainedtime trends.
Third, and related to this, the long-runelasticity of consumptionwith
respect to income, when wealtheffects are includedby keepingthe ratio
of wealth to income constant, does not differsignificantlyfrom unity in
the surpriseregressionbut does in the no-surpriseregression.However,
the correspondingno-surpriseregression is also provided for readers
who prefer a more traditionalspecification that does not rely on an
admittedlyquestionableprocedurefordividingvariablesintoanticipated
and unanticipatedcomponents.
Alan S. Blinder and Angus Deaton
505
Table 9. Tests of Tax Discounting Within the Year, 1954:1-1983:4a
Regression
Variable,parameter,and test
Surprise
No-surprise
0.043 (0.2)
-0.207(-0.6)1
0.164 (1.4)0
-0.004(-0.03)
0155 (1.4)
Tax paymentsminus liabilitiesincome
variableb,c
Lagged
Anticipated
Unanticipated
Parameterd
EstimatedR2
Impact
Impactand lagged
0.84
-2.29
0.83
1.37
Test of hypotheses
(marginal significance levels)e
R2
=
0
1
Unit elasticity
Real interestrate
No decomposition
No decompositionfor income and wealth
Surprisesonly
Half-samplestability
Reaganstability
R2
=
0.47
0.62
0.44
0.004
0.018
0.041
0.94
0.23
0.57
0.31
0.78
0.14
0.076
...
...
...
0.30
0.52
Sources: Same as table 2.
a. Dependent variable is the change in log of consumption of nondurable goods and services. For each variable,
the anticipated variable is defined as Ext, the unanticipated variable as xt - Ext. New variables are added to
baseline equations 3.3 and 3.4.
b. The variable is defined as X[(TP-TL)/YL], where X is labor share, TP is tax payments, TL is tax liabilities, and
YL is labor income.
c. These are coefficients of the indicated variables in the regressions.
d. As described in the text, the parameters are estimated as the ratio of coefficients from the regressions.
e. This is the probability, expressed as a decimal, of getting the indicated coefficients if the hypothesis is correct.
xt,
The mainconclusions of this study are the following.
There do seem to be empiricalgains from decomposingincome and
wealth changes into anticipatedand unanticipatedcomponents, as the
"rationalexpectations"approachto the consumptionfunctionsuggests.
It seems to be mainlyunexpected, not expected, changesin income and
wealththatcause consumptionto change,just as modernversions of the
permanentincome hypothesis suggest. In fact, we cannot reject the
hypothesisthatlaggedand anticipatedincome andwealth are irrelevant
to changes in consumption.This may reflect the fact that in our formulations of expectations, much of the time series varianceof changes in
income andwealthis unexpected.
The rate of interest has an insignificantlynegative influence on
Table 10. AugmentedBaselineRegressionsa
Regression
Independent variable, summary statistic,
and test of hypothesis
Surprise
10.1
No-surprise
10.2
Constant
Time trend(x 103)
-0.19 (- 0.4)
0.26 (1.0)
Lagged consumption
- 0.126( - 1.5)
- 0.221( - 3.6)
0.141 (0.8)
- 0.048(-0.3)
0.178 (3.3)
- 0.053(- 1.0)
0.171 (3.4)
0.034
0.012
0.116
-0.274(-
-0.020(-0.6)
0.060 (2.0)
Income
Lagged
Anticipated
Unanticipated
Wealth
Lagged
Anticipated
Unanticipated
After-taxnominalinterestrate, lagged
Inflation
Anticipated
Unanticipated
Relative price of nondurablegoods
Lagged
Anticipated
Unanticipated
Relativeprice of durablegoods
Lagged
Anticipated
Unanticipated
Stock of durablegoods
Anticipated
Unanticipated
(0.7)
(0.2)
(3.0)
1.3)
- 0.928(- 2.7)1
-0.127( - 0.8)
...
-0.135(-2.7)
- 0.23.1(2).1)J
...
0.078 (2.1)]
0.022 (0.2)- 0.003(- 0.1)]
0.108 (0.6)
0.38
0.49
(1.4)
(2.6)
-0.141(- 1.0)
- 0.259(- 18)
-
0.281
(2.8)
-0.220(-2.1)
0.042
(0.5)
- 0.009(- 01)
-0001-0.1)
-.0(01
Summary statistic
R2
Standarderror(x 100)
Sum of squareresiduals(x 100)
Degrees of freedom
Long-runincome elasticity
0.541
0.340
0.122
106
1.08
0.515
0.343
0.129
110
0.71
0.86
0.008
0.015
0.081
0.81
0.063
0.31
0.081
0.038
...
Test of hypotheses
(marginal significance levels)b
Unit elasticity
Real interest rate
No decomposition
No decompositionfor income and wealth
Surprisesonly
Half-samplestability
Reaganstability
...
0.14
0.34
Sources: Same as table 2.
a. For each variable, x,, the anticipated variable is defined as Ext, the unanticipated variable as xt - Ext.
b. This is the probability, expressed as a decimal, of getting the indicated coefficients if the hypothesis is correct.
Alan S. Blinder and Angus Deaton
507
consumption.Whateffect it has is on services, ratherthanon nondurable
goods. Surprisingly,it is the nominal, not the real, interest rate that
seems to matter. Inflationhas a substantialindependentinfluence on
consumption.Otherthingsbeingequal,higheranticipatedinflationleads
to lower spending.
Relative prices matter. In particular,a rise in the relative price of
nondurablegoods leads to slower growth in spending on nondurable
goods andservicesin the shortrun.Relativepricesalso matterin another
importantsense: includingthem in the consumptionfunction changes
our assessment of the effects of several critical variables, such as
inflation,interestrates, and anticipatedincome.
Temporarytax changes appear to have had little, if any, effect on
consumption, as the PIH suggests. But consumers seem not to have
treatedthe Reagan"temporarytax hike" as a temporarytax.
There is some inconsistency in our equations regardingthe Barro
neutralityhypothesis. Thoughit appears that consumers discount the
future tax liabilities implied by governmentdebt, it also appears that
governmentbonds are viewed as net wealth. When the income and
wealthcoefficientsare combined,however, the equationsclearly imply
thata debt-financedtax cut raises consumptionsubstantially.
Simplespecificationsthatrelateconsumptionto incomeand,perhaps,
to one or two other variables, often give misleadingresults compared
with the fullermodels discussed here. Several of the precedingconclusions would not be apparentin the simple consumptionfunctions that
frequentlyappearin the literature.
APPENDIX
Time Series on Unanticipated Variables
unanticipatedincome, wealth, and inflationused in
our originalbaseline model (regression3.3) are given in columns 2, 4,
and6 in tableA- 1. Each series is juxtaposedagainstthe actualchangein
the correspondingtime series, in columns 1, 3, and 5. Changes are in
percentages,at quarterlyrates.
THE TIME SERIESon
Brookings Papers on Economic Activity, 2:1985
508
Table A-1. Income, Wealth, and Inflation, 1954-84
Percent
(1)
(2)
(3)
(4)
(5)
(6)
Change
in log
income
Unanticipated
income
Change
in log
wealth
Unanticipated
wealth
Change
in
inflation
Unanticipated
inflation
1954:1
1954:2
1954:3
1954:4
-0.866
-0.810
0.799
1.460
-0.345
- 0.866
0.603
0.452
0.650
1.356
2.156
1.853
-0.726
-0.688
0.127
-0.001
0.509
-0.287
- 0.907
0.654
0.315
0.199
- 0.776
0.003
1955:1
1955:2
1955:3
1955:4
0.428
1.941
1.033
1.078
-0.507
1.400
0.259
0.471
0.924
1.572
1.867
1.858
- 1.009
- 0.003
0.585
0.648
0.696
- 0.484
0.303
- 0.546
0.520
-0.175
0.119
- 0.378
1956:1
1956:2
1956:3
1956:4
0.608
0.183
0.049
0.641
-0.143
- 0.230
0.110
0.872
- 0.238
0.430
0.644
0.096
- 0.855
0.705
0.523
-0.043
0.384
0.391
0.018
0.066
-0.098
0.200
0.023
0.013
1957:1
1957:2
1957:3
1957:4
-0.329
0.029
-0.216
- 0.862
0.049
0.593
-0.122
- 0.682
- 1.046
0.793
-0.073
- 0.833
-0.719
1.500
-0.302
- 0.805
0.085
- 0.323
0.305
- 0.553
0.134
- 0.195
0.341
- 0.316
1958:1
1958:2
1958:3
1958:4
- 1.547
0.360
1.702
1.078
- 1.267
0.375
0.911
0.306
- 1.097
1.750
1.322
1.822
- 0.570
1.241
-0.236
0.805
0.692
-0.982
0.130
- 0.016
0.521
-0.412
- 0.204
-0.249
1959:1
1959:2
1959:3
1959:4
-0.298
1.006
- 1.198
- 0.031
-0.369
1.024
- 1.422
0.063
1.923
1.720
0.621
0.676
1.114
0.341
-0.137
0.544
0.486
-0.136
0.273
-0.476
0.153
-0.100
0.192
- 0.291
1960:1
1960:2
1960:3
1960:4
-0.009
0.504
-0.610
-0.785
- 0.512
0.387
- 1.136
- 1.446
-0.904
- 0.032
- 0.350
-0.326
- 0.756
0.646
- 0.899
-0.551
0.005
0.207
- 0.217
0.160
-0.144
0.158
- 0.177
0.249
1961:1
1961:2
1961:3
1961:4
0.518
1.131
0.741
1.252
-0.347
-0.105
-0.315
0.838
1.379
2.238
0.808
1.657
0.906
1.114
-0.811
1.054
- 0.524
0.124
0.350
-0.298
- 0.228
-0.042
0.207
- 0.206
1962:1
1962:2
1962:3
1962:4
0.729
0.510
- 0.246
0.009
0.152
0.171
- 0.594
-0.429
0.290
- 1.526
- 2.116
-0.121
-0.500
- 1.702
- 1.518
0.418
0.349
-0.043
- 0.140
0.163
0.118
-0.017
- 0.097
0.302
1963:1
1963:2
0.306
0.596
1.478
2.083
0.199
0.089
-0.580
-0.511
-0.071
-0.196
0.277
0.165
Alan S. Blinder and Angus Deaton
509
Table A-1. Income, Wealth, and Inflation, 1954-84 (Continued)
Percent
(1)
(2)
Change
in log
income
Unanticipated
income
(3)
(4)
(5)
(6)
Change
in log
wealth
Unanticipated
wealth
Change
in
inflation
Unanticipated
inflation
1963:3
1963:4
0.503
0.839
-0.763
-0.449
0.649
0.554
- 1.281
-0.511
0.248
-0.042
0.522
0.396
1964:1
1964:2
1964:3
1964:4
1.652
2.637
1.016
0.773
0.593
1.379
- 0.393
-0.141
1.195
1.835
1.163
1.491
0.330
0.579
-0.078
0.676
-0.000
- 0.297
0.224
-0.116
0.362
-0.024
0.170
- 0.127
1965:1
1965:2
1965:3
1965:4
0.670
0.714
2.192
1.506
0.002
0.357
1.238
0.523
0.790
0.860
0.592
1.142
0.223
0.590
0.243
1.183
0.282
0.147
-0.185
-0.022
0.057
0.042
- 0.116
-0.266
1966:1
1966:2
1966:3
1966:4
0.610
0.544
1.079
1.112
- 0.324
-0.048
0.112
0.390
0.003
- 1.398
- 2.565
- 1.397
- 0.324
- 1.174
- 1.780
-0.198
0.472
-0.157
0.047
-0.124
0.024
-0.287
- 0.111
-0.213
1967:1
1967:2
1967:3
1967:4
1.058
0.541
0.554
0.667
0.197
- 0.573
-0.347
-0.144
1.856
3.321
1.392
0.147
2.006
1.547
- 1.022
- 0.485
-0.472
0.333
0.304
0.015
- 0.528
- 0.122
0.025
-0.044
1968:1
1968:2
1968:3
1968:4
1.271
1.593
0.265
0.518
0.859
1.167
-0.684
-0.181
-0.571
2.018
1.905
1.539
-0.294
2.200
0.970
0.731
0.393
-0.334
- 0.001
0.205
0.263
-0.222
- 0.144
-0.003
1969:1
1969:2
1969:3
1969:4
-0.123
0.606
1.700
0.209
-0.880
0.058
0.826
-0.589
-0.385
- 1.078
- 1.202
- 1.048
-0.548
0.065
- 0.125
0.276
-0.136
0.286
-0.119
0.058
-0.163
0.127
0.044
0.113
1970:1
1970:2
1970:3
1970:4
0.368
1.846
0.489
-0.757
-0.018
1.192
-0.172
-1.301
- 0.893
- 1.825
- 1.637
1.024
- 0.216
- 1.758
- 1.211
0.635
- 0.092
-0.107
-0.114
0.429
- 0.007
-0.109
-0.157
0.363
1971:1
1971:2
1971:3
1971:4
1.676
1.170
-0.211
0.520
1.125
0.355
-0.856
0.472
3.147
2.855
-0.107
- 0.073
0.558
-0.521
-2.213
- 0.080
- 0.357
0.108
-0.077
- 0.284
- 0.017
- 0.003
-0.067
- 0.383
1972:1
1972:2
1972:3
1972:4
0.699
1.021
0.650
2.937
-0.016
-0.140
-0.681
1.814
1.584
2.815
1.157
1.882
0.969
0.960
-0.673
1.269
0.295
- 0.283
0.172
-0.005
0.049
- 0.258
- 0.154
- 0.262
510
Brookings Papers on Economic Activity, 2:1985
Table A-1. Income, Wealth, and Inflation, 1954-84 (Continued)
Percent
(1)
Change
in log
income
(2)
Unanticipated
income
(3)
(4)
Change
in log
wealth
Unanticipated
wealth
(5)
Change
in
inflation
(6)
Unanticipated
inflation
1973:1
1973:2
1973:3
1973:4
1.564
0.521
0.122
0.258
0.161
-0.037
0.096
0.619
1.494
-0.257
-0.501
- 0.338
0.701
-0.505
0.431
1.113
0.508
0.519
- 0.205
0.307
0.023
0.242
-0.143
0.010
1974:1
1974:2
1974:3
1974:4
- 2.068
- 0.736
-0.438
- 1.210
- 1.621
0.497
0.365
-0.781
- 2.003
- 1.432
- 2.694
-2.618
- 1.555
-0.440
- 1.975
- 1.874
0.887
- 0.354
-0.118
0.044
0.740
0.056
0.109
0.536
1975:1
1975:2
1975:3
1975:4
-0.762
1.647
1.748
0.709
-0.661
1.292
0.420
0.133
1.134
2.969
0.875
0.399
0.827
0.768
- 2.070
-0.884
- 1.116
-0.183
0.665
- 0.390
-0.295
- 0.222
0.616
- 0.092
1976:1
1976:2
1976:3
1976:4
1.400
0.222
0.173
0.229
1.279
-0.492
-0.358
- 0.262
3.179
3.661
1.761
0.440
2.190
1.268
-0.191
-0.419
-0.544
0.026
0.403
0.194
- 0.633
-0.411
0.022
0.042
1977:1
1977:2
1977:3
1977:4
0.026
1.238
1.485
1.133
- 0.585
0.406
0.183
0.097
- 0.759
-0.079
0.027
0.252
- 1.233
-0.016
-0.181
- 0.058
- 0.022
- 0.211
0.081
- 0.033
-0.121
- 0.287
-0.021
-0.083
1978:1
1978:2
1978:3
1978:4
0.271
0.549
0.281
0.369
- 0.532
-0.180
-0.270
- 0.184
0.321
1.393
2.453
1.869
- 0.192
0.519
0.759
0.245
0.194
0.665
- 0.325
0.096
0.079
0.663
0.079
0.129
1979:1
1979:2
1979:3
1979:4
0.059
-0.270
0.316
- 0.732
- 0.221
-0.584
0.230
- 0.716
0.506
0.754
0.313
- 0.064
-0.448
0.573
0.020
-0.007
0.227
-0.016
0.041
0.128
0.235
0.139
0.079
0.065
1980:1
1980:2
1980:3
1980:4
-0.449
- 1.910
0.573
0.370
-0.290
- 1.598
0.945
0.364
-0.019
0.242
1.367
1.584
0.229
0.170
0.440
0.278
0.310
-0.312
0.015
- 0.108
0.326
0.020
0.145
-0.027
1981:1
1981:2
1981:3
1981:4
-0.162
-0.949
0.492
-0.666
0.191
-0.258
1.132
-0.095
0.695
0.655
-0.081
- 0.356
0.828
1.031
0.207
0.332
-0.194
-0.234
0.056
- 0.194
-0.153
-0.295
0.034
-0.056
1982:1
1982:2
- 0.804
-0.133
-0.207
0.373
- 0.284
-0.727
- 0.358
-0.078
- 0.366
-0.360
- 0.268
- 0.368
Alan S. Blinder and Angus Deaton
511
Table A-1. Income, Wealth, and Inflation, 1954-84 (Continued)
Percent
(1)
Change
in log
income
(2)
Unanticipated
income
(3)
Change
in log
wealth
(4)
Unanticipated
wealth
-1.093
(5)
Change
in
inflation
(6)
Unanticipated
inflation
1982:3
0.398
0.302
- 1.003
0.506
0.516
1982:4
0.777
0.113
1.893
0.635
- 0.515
-0.004
1983:1
1983:2
1983:3
1983:4
0.302
0.818
1.352
1.708
-0.712
-0.256
-0.000
0.470
2.764
1.708
1.614
1.077
0.148
-0.569
- 0.132
- 0.165
-0.422
0.445
- 0.088
- 0.335
-0.400
0.290
0.052
- 0.265
1984:1
1984:2
1.474
0.773
0.421
0.069
-0.150
- 0.160
-0.747
0.035
0.353
- 0.386
0.158
- 0.316
1984:3
1984:4
0.055
0.529
- 0.483
0.156
- 0.724
0.734
- 0.452
0.558
0.527
-0.440
0.296
-0.118
Comments
and Discussion
Robert E. Hall: Alan Blinderand Angus Deaton make a brave assault
on some of the major outstanding issues in aggregate consumption.
Unhappily, the data do not always yield up definitive answers to the
questions the authors pose. But their efforts convince me that no
additionaleconometric wizardrywill get the answers out of U.S. time
series dataon majorcategoriesof consumption.
One of the questionsinvestigatedby BlinderandDeaton has received
a great deal of attention already, most notably in work by Marjorie
Flavin.' Can the first difference of consumptionbe predicted, or is it
purely a measureof the consumer'sreactionto new, inherentlyunpredictableinformation?Blinderand Deaton replicateFlavin's findingthat
the change in consumption is predictablefrom past income. In their
equation 2.1, from table 2, lagged income and the forecast of current
income based on laggedinformationare both significant,contraryto the
implicationsof a simple rational expectations model. However, in a
more elaboraterationalexpectations model, expressed in equation3.3,
fromtable3, with interestrates andrelativeprices, the predictivepower
of lagged income and wealth disappears. Flavin's rejection of rational
expectations may be an artifact of her neglect of predictors that are
consistent with rationalexpectations. However, Blinderand Deaton do
not pursuethis findingany further.
Blinderand Deaton are concernedwith anotherhypothesis they call
"no decomposition."Under it, the influenceof the surprisecomponent
of, say, incomeis the sameas the influenceof the predictablecomponent.
1. Marjorie A. Flavin, "The Adjustment of Consumption to Changing Expectations
about Future Income,"Journal of Poiitical Economy, vol. 89 (October 1981), pp. 9741009.
512
Alan S. Blinder and Angus Deaton
513
In that case, income itself is the propervariable,and the decomposition
intosurpriseandpredictedcomponentsis irrelevant.BlinderandDeaton
areextremelyunclearas to why we shouldbe interestedinthe hypothesis.
Theyjustifytestingthe hypothesison the groundsthat "the PIH suggests
so stronglythatanticipatedandunanticipatedincomeandwealth should
get differentcoefficients." However, they acknowledge in a footnote
that "undersome circumstances,the pure PIH is fully consistent with
our no-surprisespecification." In other words, sometimes a rational
expectations model satisfies the no decomposition hypothesis, and
sometimesit does not, dependingon issues that have nothingto do with
rationalexpectations. The paper is devoid of any analysis supporting
the proposition that rational expectations, or any other interesting
proposition, is related to the no decomposition hypothesis. The work
that I have done suggests that no interest attaches to the no decomposition hypothesis. I think the paper should stick to testing hypotheses
thathave been carefullyderivedfromtheory, such as the surprises-only
hypothesis.
The role of interestrates in the consumptionfunctionhas received a
good deal of attention recently, both within and outside the rational
expectations framework. Blinder and Deaton find slightly negative
coefficients,but cannotreject the hypothesisof no effect. The findingis
confused somewhat by the strong negative effect of expected inflation
on consumption. Plainly, expected inflationis not playing the role of
makingonly the real rate of interest affect consumption. Blinder and
Deaton attributethe negative role of inflationto consumer confusion
regardingreal and nominalchanges. I think they should also consider
the propositionthatinflationis a specificsymptomof unfavorableevents
in the U.S. economy. Overtheir sampleperiodthe two biggestburstsof
inflation occurred following oil price shocks. Their results say that
consumers react more strongly to an oil shock than they do to other
events with the same impacton real income and real wealth.
WhenBlinderand Deaton turntheir attentionto the importantissue
of the effect of temporarytax measureson consumption,they are again
unableto coax a definiteanswerout of the data. A temporarytax cut has
virtuallyno immediateeffect on consumption,as predictedby permanent
income-rationalexpectationstheory, but the effect in the long runis for
consumptionto fall, which is inconsistentwith any theory. Blinderand
Deatonalso makean ingeniousanalysisof the phased-intax cuts adopted
514
Brookings Papers on Economic Activity, 2:1985
in 1981.Theirresults suggest that consumerswaited untilthe cuts went
into effect to start spendingthem. This findingis consistent either with
a simpleview that consumersrespond,irrationally,only to theircurrent
disposable income, or with the view that consumers are rationally
skepticalabout the likelihoodof promisedfuturecuts.
The last majortopic is the Barroequivalencehypothesis,which states
thatconsumershold backduringperiodsof budgetdeficitsbecause they
know they will have to pay extra taxes later to support the resulting
governmentdebt. Blinderand Deaton set up the test in a way that looks
separately at an adjustmentto income (subtractingthe deficit from
income)andan adjustmentto wealth(subtractinggovernmentdebtfrom
total wealth). They find that consumers do seem to use the income
adjustment;deficitsdo reduceconsumptionvia thischannel.Thisfinding
is puzzling in light of table 1, which shows that private saving has not
risen substantiallyduringthe period of huge deficits startingin 1982.
Partof the answercomes fromtheirotherfindingthat consumersdo not
ignore government debt in calculating their wealth, as they should
accordingto Barro. Taken together, the net effect of an increase in the
deficit working on consumption through both the labor income and
wealth (nonlaborincome) channelsis to increase consumption.
R. GlennHubbard: Reverencefor"psychologicallaws" notwithstanding, the regularityof the relationshipbetween consumptionand income
must surelystrikeus as a conundrum.Explorationof this relationshipthat is, of the time series behaviorof aggregateconsumption-has been
a focus of macroeconomic research for decades. In recent years,
theoreticaland empiricalefforts have pursued two related avenues of
inquiry. One is the extent to which "consumption functions" are
consistent with the predictionsof the life-cycle model or the permanent
income hypothesis. The other is the "effectiveness" of fiscal policy
changes in raisingor loweringconsumption.
Blinderand Deaton have given us a carefuland innovativereexamination of both issues. The paper presents a systematic evaluation of
recent empiricalpropositionsregardingconsumptionmodels or, more
correctly, models of consumer spendingon nondurablesand services.
By addressinga largeset of variablespotentiallyaffectingconsumption,
their study offers a rich comparisonwith previous studies. Two issues
are particularlyimportant:the notion from some previous studies that
Alan S. Blinder and Angus Deaton
515
only unanticipatedincome changes matter,and the hypothesis associated with Barrothat individualspierce the veil of governmentfinance.
My focus is on those two issues and, more generally, on the relevance
of variousmodelingstrategiesfor policy analysis.
BlinderandDeatonbeginwitha flexiblefunctionalformfor estimation
andanalysisin equation1. The choice is a convenientone, as the model
nests many alternativehypotheses, and as is pointed out, the steadystate value of the average propensityto consume can be solved for to
test whetherthe results are sensible. The paper'sreferenceto an errorcorrectionmodelraises the questionsof whatfactors are motivatingthe
adjustmentprocess andwhethersuch a process of adjustmentmightalso
be influencedby transitoryunemploymentor liquidityconstraints.For
example, equation 1 is consistent with an error-correctionmodel in
which agents adjust toward a desired consumption-incomeratio, as
opposed to a desired consumptionlevel. As an empiricalmatter,based
on the estimatesin table 2, the speed of adjustmentof the consumptionincomeratiotowardits desiredlevel is quite slow.
The results in table 2 are an informativeexpansion of "traditional"
consumptionfunctions. The additionof household net worth variables
improvesthe fit of the regression, as one would expect. Whatconcept
of "net worth"to use here is a trickyquestion.For example, shouldone
use only financialnet worth?Moreimportant,whatshouldbe done about
household claims to social security and private pension benefits?The
former is unfunded, and the latter is not, but both clearly dominate
householdwealth. The issue is importantin tryingto disentangleeffects
of "news about future wealth" from those of changes in liquidity. A
(log) wealth-incomeratio variablecould reflecteithera scale parameter
for ln (CIY) or "precautionary"saving againstfutureevents leadingto
liquidityconstraintson consumption.Its expected effect is unclear.
A timetrendis also includedwhen the householdwealthvariablesare
introduced;I am not sure of its interpretationin equations2.3 and 2.4.
It is true that C here is not total consumer expenditures and that
componentparts mightexhibit "trends" if income and wealth were the
only explanatoryvariables.However, the inclusionof relativeprices of,
say, durableand nondurablegoods shouldmitigatethatproblem.When
this adjustmentis made, in table 3, the time trend is still generally
statistically significant.One can imagine other "trend" influences on
consumptionrelativeto income (for example, the increasinggenerosity
of social insuranceprograms)for which the time trendmay be a proxy.
516
Brookings Papers on Economic Activity, 2:1985
Perhapsmost interestingin table2 arethe resultsfor modelsincluding
labor income, the concept suggested by the life-cycle model. This first
set of resultsillustratesthatonly actualvariables,not theirdivisioninto
anticipatedand unanticipatedcomponents, matter.Even in this simple
case, however, one caveatis thatconsumers'expectationsaboutincome
or net worthpositions may be substantiallymoreinformedthanthose of
the econometrician,so thatpartof the differencebetween "actual"and
"expected" variablesis not reallyunexpected. The largeris that source
of "error"relativeto the "true" forecast error,the more likely are the
estimatesto give the resultthat "only actualvariablesmatter."
The additionof relativeprice effects, both intertemporaland across
categories of consumption, in table 3 brings still more novel, and
sometimespuzzling,findings.Nominalafter-taxinterestrates exert the
predicted negative effect on consumption,though the coefficient estimate is statistically insignificant.That the expected inflationvariable
does not have the positive sign required by a "real interest rate"
explanationcould reflecta relativeshiftto spendingon durablesin times
of inflation.Based on the results in equations 3.3 and 3.4, the "price
confusioneffect" noted by the authorsis suspect, given the importance
of anticipated inflationrelative to unanticipatedinflation. It is more
likely that effects of recurrentepisodes of inflation on consumption
reflect the importanceof the structuraldemand and supply variables
underlyinginflationaryperiods.
Inclusionof relativeprices of durableandnondurablegoods gives the
expected effects and improvesthe fit of the model. In addition,both the
hypotheses that "only actual variables matter" and "only surprises
matter"can be rejected.The findingthat nominalinterestsensitivityof
spendingon services exceeds that of spendingon nondurablegoods is
indeed unusual.That services are relativelymore complementarywith
durablegoods seems unlikely. The results on the interest sensitivity of
consumerspendingdo not shed muchadditionallighton the debateover
the sign and magnitudeof the interestelasticity of saving. In additionto
the problemsof carryingout such an analysiswith aggregatetime series
data, theory tells us that there can be no one "interest elasticity of
saving";resultsare specificto the individualthoughtexperiment.
Perhapsmost importantfor the applicationof the aggregateconsumptionfunctionto policy analysisis the issue of its abilityto assess whether
households internalize the saving decisions made for them by the
Alan S. Blinder and Angus Deaton
517
corporateandgovernmentsectors. Anobviousstartingpointforresearch
is to test for differences in consumer responses to temporary and
permanenttax changes. The Blinder-Deatonpaperfollows up Blinder's
importantearlierstudy and presents, amongother things, an early look
at the effects of the 1981EconomicRecoveryTax Act.' Whilethe results
are not precise, it appears that scheduled tax reductions during the
Reagan period were ignored by individuals. That is, individualsconsumedaccordingto actualafter-taxincomes ratherthan the permanent
after-taxincomes they would have once the tax reductionswere fully
effective. The estimates given in this paperfor temporarytax changes
in the pre-Reaganperiod differfrom those in Blinder'sprevious study;
here, the short-runeffects of temporarytaxes correspondmore closely
to those predictedby the permanentincome hypothesis.
Theauthorsdo notconsiderwhetherindividuals"piercethe corporate
veil." Nordo they questionwhetherindividualspiercetheirown veil,that
is, the extent to which householdnet worthheld in privatepensions, or,
for that matter,social securityclaims, acts like nonpensionwealthin its
effect on consumption.In other words, is household net worth held in
privatepensions substitutablefor nonpensionnet worth?
Most significantfor the policy debateover Ricardianequivalenceand
the governmentdebt, Blinderand Deaton examinewhetherhouseholds
distinguishbetweendebtandtaxes in governmentfinance.Theliterature
gives manyreasonswhy the Barrohypothesismightnot be borneout by
the data:unlikethe stylized model of Ricardianequivalence,tax policy
is not carried out on a lump-sumbasis; bequests may be at a corner
solution for many if not most consumers; in the presence of capitalmarketimperfections,bindingliquidityconstraintson manyindividuals
may restrict consumption movements. The last explanation seems
particularlyrelevanthere, given some of the otherevidence in the paper.
The authors do find evidence for discountingof future tax liabilities,
though it is not supportedby the wealth coefficients in table 7. The
estimatesare not precise enough to draw specific conclusions. Despite
theirrejectionof the strictrequirementsof the Barromodel, Blinderand
Deaton find evidence for short-termsmoothing, evidence requiredfor
even a pragmaticbelief in life-cycle or permanentincome models.
1. AlanS. Blinder,"TemporaryIncomeTaxes andConsumerSpending,"Jouirnal of
Political Economy, vol. 89 (February1981),pp. 26-53.
518
Brookings Papers on Economic Activity, 2:1985
A more direct way to contrast the Barro model and the life-cycle
model with time series data is to test whetherconsumptionvaries with
the age distributionof resources. Unlikethe life-cycle model, the Barro
model predicts that consumption is a function only of collective resources, not of the age distributionof resources. Data on the age
distributionof propertyandlaborincome are available,andone can test
whether"income share" variablesmatterin the consumptionequation.
As tax policies frequentlyredistributeincome and wealth along these
lines, this is likely to be a fertile area for appliedresearch. In addition,
such tests are likely to be importantin the continuingdebate over the
influenceof intergenerationaldebt policy on consumption.
Wheredo we go fromhere? By now, two puzzles figureprominently
in recentappliedresearchon consumption.First, given the professional
consensus on the modelingparadigmfor the life-cycle and permanent
income hypotheses, why do versions of "traditional" consumption
functions seem to fit the data well? Second, why is it that estimates of
aggregatesavinggeneratedfromtheoreticallybelievablelife-cycle consumptionsimulationmodels fall shortof "realistic"values?2
I suspect that these puzzles, though revealed in differentresearch
agendas,are related.Two modificationsof existingwork are suggested.
First, we need to consider more explicitly how restrictionson private
trades-specifically, borrowingrestrictions, or liquidity constraintsaffectconsumerspending,particularlyin extensions to modelsof spending on durables.For example, life-cycle microsimulationmodels can be
used to examinethe sensitivityof aggregateconsumptionandthe capital
stock to liquidityconstraints,helpingus to see whetherlikely effects are
sizable enough to explain the "anomalies" in aggregateconsumption
functions. Second, households most certainlyengage in precautionary
saving-against uncertaintyover earnings,health, lengthof life, and so
forth. Dynamic life-cycle simulationmodels may be able to shed light
here as well. To the extent that precautionarymotives in response to
individual-specificuncertainty are important, it may be difficult to
rationalizeresults of aggregateconsumptionfunctions with aggregated
2. See, for example, Laurence J. Kotlikoff and Lawrence H. Summers, "The Role of
Intergenerational Transfers in Aggregate Capital Accumulation," Journal of Political
Economy, vol. 89 (August 1981), pp. 706-32; and R. Glenn Hubbard and Kenneth L. Judd,
"Social Security and Individual Welfare: Precautionary Saving, Liquidity Constraints,
and the Payroll Tax" (Northwestern University, 1985).
Alan S. Blinder and Angus Deaton
519
results of simulationmodels of individualbehavior. These two qualificationsof the basic model-restrictions on borrowingandprecautionary
saving-are certainlyrelated.
The Blinder-Deatonpaperis an importantintegrationof variouslines
of inquiryin aggregatetime series studiesof consumption.New research
on issues involvedin linkingsolutionsto individualoptimizingproblems
to aggregatedata (in particular,the modeling of liquidity constraints)
will be an importantstep in extendingtheirwork.
General Discussion
Franco Modigliani questioned Alan Blinder and Angus Deaton's
exclusion of durablegoods services from their consumptionmeasure;
he arguedthat they ought to have imputedrentalvalues to the services
of durablegoods andincludedthose imputedvaluesalongwithpurchases
of nondurablesand services in total consumption.Blinderdefendedthe
consumptionmeasurechosen, noting that durablegoods services may
not be perfectsubstitutesfor otherformsof consumption,as would have
to be assumedto simplyadd them to the total, and that the dynamicsof
durable goods purchases are likely to differ considerably from the
dynamicsof otherconsumptionpurchases.
There was general discussion of whether significant"anticipated"
right-handside variables are inconsistent with a model in which consumptiondepends only on permanentincome, itself a random walk.
ChristopherSims observed that one explanation sometimes given for
the significanceof anticipatedvariablesis the presence of "transitory
consumption."But either transitoryconsumptionreflects information
availableto consumersbut not availableto economists modelingtheir
behavior-in which case changes in it should not be seriallycorrelated
so thatits presencecannotexplainsignificantcoefficientson anticipated
variables-or it reflects an inadequacyof the theory-in which case it
could be serially correlated, but no longer satisfies the identifying
assumptionsof the randomwalk model. In this last case, simultaneous
equationmethodsare necessary to test the theory.
Sims also urgedthat more carefulattentionbe given to time aggregation issues, which are critical if one is testing dynamic propositions
relatedto the importanceof innovations.The predictionsemergingfrom
520
Brookings Papers on Economic Activity, 2:1985
rationalexpectationstheorycan look quitedifferentwhenthey arebased
on dataaggregatedover two or threetime periods,as when monthlydata
are aggregatedinto quarters,than they do when they are based on data
for single periods. Stephen Goldfeld noted that shifts in the age and
income distributionof the population could introduce errors into the
aggregateconsumptionfunctionsBlinderand Deaton estimate.
StanleyFischercommentedthatthe categorizationof incomechanges
as expected or unexpecteddependscriticallyuponthe underlyingmodel
of expectations. Insofaras these distinctionsare crucialto the analysis,
he urgedthat more attentionbe paid to the vector autoregressionsthat
producethe income changepredictions.He also conjecturedthat in the
future,economistswill place less relianceon purelyeconometricmodels
of expectationformationand rely increasinglyon data from surveys in
which individualsare asked directlyabout theirexpectations. Goldfeld
noted a conceptual problem that arises whenever aggregateanalyses
resting upon an assessment of expectations are performed:there will
always be some dispersion in expectations within the population,and
thereis no fully satisfactoryrulefor decidingwhich expectationsto use.
The questionof whetherBlinderandDeatonhadadequatelycaptured
expected andunexpectedchangesalso came up in connectionwith their
test of the intertemporalsubstitution argument. They reached the
puzzling-if by now familiar-conclusion that it is nominalratherthan
real interestrates that seem to affect consumption.Sims put forwardhis
view thatthe resultreflectserrorsin the inflationpredictionthatunderlies
the distinctionbetween real and nominalinterest rates. Standardtechniques may lead to overly variableinflationforecasts; better forecasts
can be obtainedusingtechniquesthat shrinkthe variationin the inflation
forecasts towards zero. If the inflationforecasts do not vary so much,
then surprises in the nominal rate of interest turn out to be largely
surprisesin the real rate.
Sims also questioned Blinder and Deaton's treatmentof the Barro
hypothesis that increases in governmentdebt are seen by consumersas
equivalentto increasesin theirown indebtedness.BlinderandDeaton's
test is based on a single equationthat includes governmentdebt as an
explanatory variable; however, if there is a systematic connection
betweenlargedeficitsandotherevents thatchangepeople's expectations
about the likely course of future income, results based on this sort of
single equationframeworkcould be quite misleading.MauriceObstfeld
Alan S. Blinder and Angus Deaton
521
arguedthatresponses to increasesin the governmentdeficitwill depend
uponwhetherconsumersexpect any accompanyingincreasein governmentspendingto be temporaryor permanent.An increasein government
spendingthat is expected to be temporarywill have a much smaller
effect on consumptionthanwill an equal increase that is expected to be
permanent.
Georgevon FurstenbergquestionedBlinderand Deaton's characterizationof the Reagantax packageas a permanenttax cut combinedwith
a temporarytax increase. In makingthis characterization,Blinderand
Deaton assume a particularsort of forward-lookingbehavior by consumers estimating their future tax liability; a more sophisticated assumptionwould be that consumers estimatingtheir future tax liability
consider not only provisions alreadywritteninto the tax code but also
likely futurechanges in the code. In von Furstenberg'sview, the large
currentfederal deficit should lead consumers to expect future federal
tax increases, even thoughthey have not yet been writteninto law. This
could explainwhy savingrates did not fall followingpassage of the 1981
tax act.