F. THOMAS JUSTER* National Bureauof Economic Research PAUL WACHTEL* City Universityof New Yorkand National Bureauof EconomicResearch Inflation and the ConsumerTo JUDGE FROM THE RECENT PRONOUNCEMENTS of the forecastingfra-ternity,uncertaintyabout the behaviorof consumersis at the heart ofdifferencesin view aboutmacroeconomicpolicyat the presenttime. Theinabilityof the economicrecoveryto gathersufficientforce to bite intounemploymenrtateshasbeenwidelyattributedto the continuedhesitancyand cautionof consumers,as reflectedin exceptionallyhighratiosof per-sonal savingto disposableincome.And differencesof opinionabouttheprobablevigoror sluggishnessof therecoveryareduein considerablepartto differencesin judgmentsaboutthe probableconsumerresponseto theunfoldingeconomicsituation.One of the majorsourcesof uncertaintyabout consumerreactionsconcernsthe way that priceinflation,both ex-pected and realized,influencesconsumerdecisionsabout spendingorsaving. This paperposes a numberof questionsand providessome tentativeanswers: L. Haveconsumerreactionsto the standardeconomicdeterminantsofbehavior(income,relativeprices,the size of durablestocks relativetoincome,and the like) been differentor less predictableduringthe recentperiodof historicallyhighpriceinflation? * The researchreportedheregrows out of work supportedby the National ScienceFoundation,the National Bureauof Economic Research,and the U.S. Bureauof theCensus.Supportforthecomputerworkin the paperwasprovidedbyRapidata,Inc.,andfor researchassistanceby the CommercialCreditCorporation.TeresitaRodriguezandTedi McDermotthelpedwith the computations. 71

72 Brookings Papers on Economic Activity, 1:1972 2. Havepredictionsaboutconsumerbehaviorbasedon explicitsurveymeasuresof expectationsand attitudesbeenless reliable? 3. Have the economicvariablesthat seem to be associatedwith suchsurveymeasurestendedto be different? 4. What do consumerspendingmodels, appliedto recentexperience,tell us aboutthe effectof actualinflationandinflationaryexpectationsonspending? 5. Giventhe optimalpredictionmodel,whatis the probablecourseofconsumerspendingand saving behaviorunder alternativeassumptionsaboutthe behaviorof key variables? Thispaperattemptsto minesurveymeasuresof consumerexpectations,attitudes,andplansin an effortto determinetheirusefulnessin predictingand explainingconsumerbehavior.We ask two questions:First,what isthe role of these anticipatoryvariablesin models designedto forecastconsumerbehavior?Second,whatarethe economicvariablesunderlyingthe movementof consumeranticipations? Theplanof thepaperis, first,to examinedurablegoodsdemandmodels-for both automobilesandnonautodurables-basedon nonanticipatoryor objectivevariables,andthento examinesimilardemandmodelsbasedon anticipationsvariables,analyzingthe featuresof an optimaldemandmodelthatusesboth consumeranticipationsandothertypesof variables.Next, we look specificallyat the role of priceinflationin these models.Thefollowingsectionfocuseson modelsdesignedto predictthe two prin-cipalsurveyvariablesusedin theanalysis,theindexof consumersentimentand the indexof expectedpurchases,in both of whichthe role of priceinflationis of specialinterest.The next sectionlooks explicitlyat savingratherthandurablegoodsexpenditurefunctions,usinga verysimplesys-tem of equationsin whichall the principalallocationsof disposablein-come (saving and durableand nondurableexpenditures)are regressedagainsta standardset of independentvariablesincludingincome,pricechange,and the anticipationsvariables.The last sectionsummarizestheprincipalresultsof the paper. ObjectiveModelsof DurableGoodsDemand Ourmodelof demandforconsumerdurablegoodsis somewhatdifferentfromothersthat arealso basedentirelyon the standard\"objective\"eco-

F. ThomasJusterandPaul Wachtel 73nomic variables. Certain features, we think, make it a better specificationthan others,but it is generallyrepresentativeof the standarddistributedlagmodels of durablegoods demand found in the recentliterature.The generalspirit of the model is that target stocks of durable goods are determinedbyexpected values of income and relative prices, and that observed expendi-tures reflect both the attempt to adjust beginning-period stock to (chang-ing) target values and the immediate impact on expenditures of transitoryincome changes that are independent of the expected value of income.These transitory expendituresdo alter durable goods stocks and hence thegap between actual and target stocks. Gross investment or expenditures, G, is thus divided conceptually intoplanned (GP) and transitory (GT)components:(1.0) G = GP+ GT.The formeris determinedby long-run expectations and average adjustmentlags, while the latter represents the influence of unexpected economicphenomena on durables expenditures. The familiar partial adjustment model is applied to the planned compo-nent. As shown by (1.1), the expendituresa household plans for the quarterwill close some proportion of the gap between its desired stock, K*, anddepreciatedexisting stock, (1 -6)K,_:(1.1) GJD=P 3[K*- (I -6)Kt-1KThe adjustment coefficient is f and the quarterly depreciation ratio is 6.It is assumed that a constant fraction of stock existing at the end of agiven quarter will depreciate during the next quarter. The desired level ofstock is determinedby the household's expectations about economic condi-tions, as in(1.2) K*-f(Z= )where Ze is the expected value of a set of objective economic variablessuch as income and prices. It is assumed that the expected value, Ze, isformed by the adaptive expectations model of(1.3) Zf - Ze p(Z1-zt)where p is the coefficient of expectations.' In this model expectationschange in response to the difference between current experience and the 1. For simplicity, in (1.3) we assume an identical expectationalstructurefor allvariablesin the function,Z, set, andomit the functionalnotationthatappearedin (1.2).

74 BrookingsPaperson EconomicActivity,1:1972previouslyformedexpectation.Finally,transitoryinvestmentis specifiedto be a functionof variablesin the set, T, as in(1.4) G = g(Tt). The reducedform of the model that includesits expectations,adjust-ment,andtransitoryfeaturesis givenby(1.5) Gt = 3pZt- 3(p- 6)Kt- + (1 - )(1 -p)G + g(T) - (1 - p) g(Tt_1).2Themodelincludeslaggedstock,laggedexpenditurest,he determinantosfdesiredstock, and both the currentand lagged values of the transitoryfunction.Thelaggedstockandlaggedexpenditureasrebothin thereducedformbecausetherearetwo lag mechanismsin the model(forstockadjust-mentandforexpectationsformation)T. heappearanceof thelaggedtransi-toryfunctionis explainedin AppendixA. Least squaresestimatesof (1.5) overidentifythe parameters,Band p.Theexpectedsignsof the coefficientson laggedstockandlaggedexpendi-turesareknownbecauseboth 3 andpmustbe positiveandlessthanunity.Thecoefficienton Gt-l mustbe positiveandlessthanunity;it willbe largeif adjustmentor expectationslags (or both) are long. The coefficientonKt-, will be negativeunlessexpectationsareformedveryslowly(so thatp < 8, the quarterlydepreciationrate).3 The determinantsof desiredstock, the elementsof the set Z, are in-comeandrelativeprices;theZ functionis assumedto be linear.Two alter-nativeversionsof the transitoryfunctionare tested, the unemploymentrateandan independenetstimateof transitoryincome.Transitoryincomeis definedasthedifferencebetweenactualincomeandestimatedpermanentincomein eachperiod;permanentincomeis estimatedas a smoothgrowthpathof actualincomeusingthetrendequationasdescribedinAppendixB. All monetaryvariablesaremeasuredin price-deflatedollarsperhouse-hold. Dummies are included in automobileregressionsfor strikes in 2. A detailedderivationis providedin AppendixA. 3. The model includesthe same featuresas the objectivemodel in F. ThomasJusterand Paul Wachtel,\"Anticipatoryand ObjectiveModels of DurableGoods Demand,\"AmericanEconomicReview(September1972,forthcoming),which includesa fullerdis-cussionof long-runequilibriumproperties.That model, however,appliesthe systemtonet investmentonly; that is, all replacementdemandis plannedand met withoutdelay.The modeljust describedyields the same reducedform but a somewhatdifferentinter-pretationof the lag coefficients.

F. ThomasJusterandPaul Wachtel 751959:4, 1964:4, 1967:4, and 1970:4, and for compensatingexpendituresin two post-strikequartersT. hecoefficientson the strikedummiesarenotreported.The followingvariablesareusedin the empiricalanalysis: Y = disposableincomeperhousehold,in constantprices YP = expectedor permanentincomeper household,in constant prices YP' = permanentincomebasedon disposableincomeper house- hold less transferpaymentsPA/PCE = relativepriceof automobiles YT= transitoryincomeperhousehold,in constantprices YT'= transitoryincomeperhouseholdless transferpayments U = total unemployment rate.AUTOMOBILES The resultsof applyingthe model to quarterlyexpenditureson auto-mobilesareshownin equation(2):4(2) CAR, = 54 + 0.059 YP' - 33 Ut + 26 Ut- (3.2) (-5.5) (4.6) -1O1 (PA/PCE), - 0.181 K-11+ 0.654 CAR,-(-1.1) (-2.6) (7.2) Periodof fit: 1953:4 to 1971:2.R2= 0.931; standarderrorof estimate= 19.1; Durbin-Watsonstatistic= 2.11.The dependentvariableis CAR,real expenditureson new cars and netpurchasesof usedcars.5Expenditureson automobileparts,whichincludethe largeand growingmobilehomecomponent,areclassifiedas nonautodurablesand arethusexcluded. The statisticalspecificationseemsgenerallysatisfactoryexceptfor thecoefficienton relativeprices,whichis small and insignificantT. he coeffi-cientof relativepriceis verysensitiveto the timespanusedfor estimation 4. Hereand in subsequentequations,the numbersin parenthesesare t ratios. 5. Becausethe equationincludesdummyvariablesfor strikeand post-strikeeffects,the laggeddependentvariable(CAR,-1)is correctedfor the strikeinfluencein all equa-tions so as to removestrikeeffectsfrom the estimationof the lag structure.This proce-dure is discussedin the forthcomingstudy by Justerand Wachtel,\"AnticipatoryandObjectiveModels,\"to bereprintedwithadditionalappendixesbytheNationalBureauofEconomicResearch.

76 Brookings Papers on Economic Activity, 1:1972and to the specificationof the model generally.The specificationusedaboverequiresthatthecombinedcoefficientson the stockvariableandthelaggedexpenditurevariablebe of suchmagnitudeand sign as to producestabilityin thesystem,andtheseconditionsaresatisfied.Finally,thespeci-ficationrequiresthat the coefficienton the currentunemploymenrtatethe transitoryvariablein equation(2)-be negativeand the coefficienton the laggedunemploymenrtate positiveand smallerin absolutesize.6Theseconditionsare all satisfied,and the lag structureand incomeelas-ticityimpliedby the modelseemreasonable. An alternativespecificationof the transitoryfunction,usingtransitoryincomeproperin currentandlaggedform,is also a reasonablespecifica-tion, but it explainsless variancethanequation(2). In otherexperiments,involvingalternativeunemploymentrates, the total unemploymentrateturnedoutto be marginallybetterthanthe unemploymenrtateof marriedmalesandmuchbetterthanthe insuredunemploymenrtate.In addition,experimentws itha rateformarriedmalesinconjunctionwiththedifferencebetweenit and the ratefor all men revealedno gain fromdistinguishingemploymentratesfor headsof householdsfromthose for others.(Thesetests were performedon the anticipatorymodel discussedbelow ratherthanthe objectivemodel.) AnticipatoryModelsandData The availabilityof surveydata on householdplans,expectations,andattitudesmakesit possibleto constructandtestconsumerdemandmodelswitha structurentirelydifferentfromthatjustdescribedP. lannedchangesin durablegoodsstockscanbe directlyrepresentedby a surveymeasureofhouseholdplansor intentionsto purchase,whiletransitoryor unplannedinvestmenits againthoughtof as influencedby eventsthatwereunforeseenorimperfectlyforeseenatthetimetheplanswereformulatedT. heexpecta-tions of the householdabout externalevents(for example,income andprices)enterintothe formulationof itsplans,whileanydifferencebetween 6. Our model interpretsunemploymentas a transitoryincome phenomenonthataffectstransitoryinvestmentin durablesstock.An alternativeinterpretationof equation(2) is that the correctunemploymentvariableis the changein unemploymentrates,thatAUis partof the plannedinvestmentfunction,andthatthereis no transitoryinvestment function.

F. ThomasJuster and Paul Wachtel 77eventsand expectationsaffectsthe relationbetweenhouseholdplans topurchaseand actual purchases.7Finally,the generalstate of consumersentimentor attitudesmay influencethe relationbetweenplans and be-havior via an effect on the firmnesswith which plans are formulated.Alternativelys,entimentmightaffectthe wayin whichdifferencebs etweeneventsandexpectationsmodifybehaviorrelativeto plans. DATA In this paperwe use threevariablesobtainedfromconsumersurveys:the indexof consumersentiment(S); an indexof expectedpurchasesforautomobiles(A*); and an indexof pricechangesexpectedby consumers(CPI*). The indexof consumersentimenthas beenpublishedsince 1953by theSurveyResearchCenterat the Universityof Michigan(SRC).The indexof expectedpurchasesof automobilesis constructedfromboth SRC andCensusBureaudata.Plansorintentionsto purchaseautomobileshavebeenobtainedregularlyby the SRCsincethe early1950s.Thebureaubegantocollectsimilarinformationin 1959,andin recentyearshasbeengatheringdata on the mean(subjective)probabilitythat householdswill purchaseautomobilesand houses. Comparabledata on purchaseplans for otherdurablesare less plentifuland consistent.8The index of consumerpriceexpectationsis compiledfromthreedifferentsegments,all basedon SRC 7. Expectationsand plansof businessesand householdsarenot alike. Businessfirms,especiallylargeones, could hardlysurvivewithoutexplicitsalesforecastsor investmentplans, and they are apt to devote significantresourcesto such corporateplanning.Buthouseholdsobviouslycan surviveand even prosperwithouteitherexplicitforecastsorplans,and typicallyare unlikelyto spendmuchtime or energyon planning.Hencedataon investmentplans, and on expectationsabout income or prices,obtainedfrom cor-porateenterprisesare probablydifferentfrom those obtainedfrom households.Opera-tionally, these differencessuggest interpretinghouseholdanticipationsdata in a morerelaxedframeworkthan might be appropriatefor businessanticipationsdata, althoughthe conceptualframeworkshould not be so relaxedas to disappear. Interestinglyenough,dataon investmentplansof smallbusinessenterprisesc, ollectedby the CommerceDepartmentand the Securitiesand ExchangeCommissionand byMcGraw-Hillmust, like household expenditureplans, be substantiallyadjusted forstrongand systematicbiases. 8. To combinethe availabledataon expectedpurchasesinto a singlemeasureis itselfa small researchproject,which the authors undertooksome time ago. Otherways ofcompilingandaggregatingtheavailabledataon expectedcarpurchaseswillnot yieldthesame resultsas those in this paper,and the readershould be awareof that fact.

78 Brookings Papers on Economic Activity, 1:1972data. The segments differ in what they measure and in the precision withwhich the measurementsare provided. Some of these problems in compilation and use of the survey data aresufficientlysharp that we have used the availability of survey materials asone criterion in choosing time spans for the empirical analysis. For exam-ple, in examining automobile outlays we chose to start with 1960:1, pri-marily because the index of expected purchases (which plays a key role inour analysis) presents problems of sampling variability before that time.9 AUTOMOBILE DEMAND MODELS We have experimented extensively with alternative specifications of anautomobile demand model based on the survey measures of purchaseexpectations and attitudes. The expected purchase variable for automo-biles, A *, for part of the period is a direct estimate of the mean (subjective)probability of car purchase of U.S. households and, for the remainder ofthe period, is a constructed variable based on purchase intentions. Al-though one could argue that such a variable encompasses all of the adjust-ment lags and the expectational structure underlying desired stock in theobjective models, experience indicates that it must be supplemented, sinceapparently it does not capture some aspects of consumer optimism andconsumer uncertainty. For example, whether consumer optimism is stable or unstable-eitherrising or falling-makes a difference to the way in which unanticipatedevents influence actual spending, for favorable intervening events mighthave more impact on actual purchasesif optimism is rising (or unfavorableevents more if it is falling). In addition, the expected purchase measurerepresentsonly an estimate of the mean value of a probability distribution,and tells nothing either about dispersion around the mean or about thereliability of any household's estimate of the likelihood it will make apurchase. The Survey Research Center's index of consumer sentiment is a logicalcandidate to capturesome of these influences.We have examined a numberof the alternativeways to use it in econometric modeling: the index proper 9. Before 1960, the indexcan be obtainedonly from SurveyResearchCentermate-rials,whosetypicalsamplesizeis about 1,300.After 1960,the indexcan be obtainedalsofromthe U.S. CensusBureau,whosesamplesizeis approximately15,000households.

F. ThomasJuster and Paul Wachtel 79(denoted S), a two-quarter moving average of \"filtered\" changes in S(denoted SZ), and Almon distributed lag versions of these variables.10 Of these, the filteredvariable, SZ, gave the best results and is the versionreported on here. It is constructed on the hypothesis that the index ofexpected purchases, A*, provides an unbiased estimate of future purchaserates only if S is not changing in a systematic way. Therefore, SZ equalsthe change in S provided S has been changing significantly and system-atically, and otherwise equals zero. The nature of the index of expected purchases suggests that the appro-priate dependent variable is consumer purchases of automobiles in units,since the expected purchase variable measures the probability that thefamily will purchase either a new or a used car without regardto its value.Thus the dependent variable, A, is the proportion of U.S. households pur-chasing new cars, seasonally adjusted at annual rates. Purchase rates areestimated from data on gross expendituresfor new cars and average pricespaid for new cars, both derived from Department of Commerce data.Thus estimates of real expendituresfor new cars based on the anticipationsmodel must be derived from a combination of equations that explain pur-chase rates and real car prices. The anticipations model for automobile purchaserates follows the sameframework as the objective model for expenditures.The following are thecorresponding equations:(1.0*) A =AP + AT;(1.1*) AP ==f(A*, S);(1.4*)(1.5*) AT = g(U); A = f(A*, S) + g(U).The unit purchase equation (1.0*) has planned and transitory components.The survey measures (A* and an S variant) constitute the anticipatorycounterpartto planned gross investment(1.1*). Equation (1.1*) substitutesfor appendix equation (A2), which is the reduced form of the adjustmentand expectations mechanismsthat determineplanned gross investment; the 10. All variablesaredefinedin AppendixB. The filteringprocedureis discussedin F.ThomasJusterand Paul Wachtel,\"Uncertainty,Expectationsand DurableGoods De-mand Models,\" in HiimacnBehacviorin Econiomic Affairs: Esscaysin Honior of GeorgeKatoicn(Amsterdam:North-Holland,forthcoming),andin SaulH. Hymans,\"ConsumerDurableSpending:Explanationand Prediction,\"BrookinigsPcapersonlEcotionoicActivity(2:1970), pp. 173-99.

80 BrookingsPaperson EconomicActivity,1:1972anticipatory model is, of course, a much simpler equation. The transitoryfunction (1.4*) is the same in the anticipatory model as in the objectivemodel. The reduced form for purchase rates is given by (1.5*); both func-tions are assumed to be linear.11However, (1.5*) explains only unit pur-chases and must be supplemented by an equation to explain average realcar price. Deflated car price (V) is a distributedlag function of permanentincome and the price of cars relative to other goods and services:(1.6*) V = h(YP, PA/PCE).Automobile expenditures are then the product of (1.5*) and (1.6*). In general, the Almon lag versions of S in the equations explainingunit purchases prove superior to the simple variables, and the SZ Almonlag version contains the optimum specification. Equation (2 *)below showsthe optimal anticipatory model for unit purchases, where the dependentvariable is the proportion of households purchasing a new automobile, A: 3(2*) At = 0.45 + 0.115 A* + 0.033 SZti - 0.043 Ut. (8.2) (6.3) i=O (4.6) Periodof fit: 1960:1 to 1971:2.= 0.883; standarderrorof estimate= 0.047; Durbin-Watsonstatistic= 1.90.The addition of standard economic variables does not appear to improvethis equation. For example, the income level takes on the wrong sign; andincome change adds nothing to the explanation. As noted above, equation (2*) must be supplementedby an equation onaverage real car prices to permit estimates of total real expenditures onautomobiles. The best explanation of real car price is a distributed lag onpermanent income and relative car price. Other variables that were im-portant in the unit purchase equation, especially those with a strongcyclical influence, have only a random effect on average car price. Sometypical results are shown in Table 1. Although relative prices show upweakly in the post-1960 equations, their importancein the full-periodequa-tion indicates that they belong in the model. The income variable suggeststhat rising real income per family will spur upgrading of car purchases, acommonly noted phenomenon. The relative price variables suggest thatrising relative prices of cars will affect the extent of upgrading, and that 11. Note that the reducedform of the anticipatorymodel has only a currentperiodtransitoryvariable(U) and does not containthe laggedtransitory(U,-1). Thisis becausethe adjustmentand expectationslags of the objectivemodel are replacedby (1.1*);theanticipatoryreducedformdoes not involvethe solutionof a lag function.

F. ThomasJuster and Paul Wachtel 81Table1. EstimatingEquationforAverageRealCarPrice,1953-71andSubperioda RegressionstatisticsPeriod Regressioncoefficientsb.Q Durbin- Constant YP' (PA/PCE)t Vt1 Standard Watson R2 error statistic1953:4-1971:2 1,354 0.057 -798 0.683 0.939 64.1 1.931960: 1-1971:2 143 (2.4) (2.6) (9.0) 0.934 66.8 2.12 0.076 0.797 0.970 49.5 2.02 1,006 (3.1) ... (12.2) 0.971 49.3 2.03 60 0.154 0.535 (1.8) -635 (4.7) 0.209 (0.8) 0.532 (4.1) ... (4.7) Sources: See Appendix B for derivation of permanentincome. Car prices are based on unpublished datafrom the U.S. Department of Commerce, Office of Business Economics, and are weighted averages of theaverage priceof foreign and domestic cars deflated by the implicit pricedeflator. Also see discussion in text. a. The dependentvariableis average new car pricein constant (1958) dollars, Vt. b. The numbersin parentheseshere and in subsequent tables are t ratios. c. Symbols used in this table are defined as follows: YPt = Real permanentincome per household less transfer payments.(PA/PCE)g = Price of cars relative to other goods and services. Vg-l = Average real car price.people tend to trade down when car prices are rising relative to otherprices. Thus the anticipations model for real automobile expenditures has twoequations with rather different independent variables. The unit purchaseequation is dominated by the index of expected purchases and the filteredsentiment variable, and, among the standard economic variables usuallyfound in such equations, includes only unemployment rates. The real carprice equation includes only real income and relative price as determiningforces. TIME-SPAN DIFFERENCES We noted earlier that the available survey measures differ rather mark-edly in their reliabilityand consistency, both within and between measures.For example, over the 1953-59 period the expected purchase variable hasa very much larger sampling error than in later years. Moreover, the ex-pected purchase variable is measured from 1967 on as a mean subjectiveprobability of purchase, but before that as the weighted sum of plans topurchase.12Finally, the impact of differentinflation rates can be exploredby defining periods characterizedby varying rates of price inflation. 12. The purchaseprobabilitydesign,in cross-sectionalanalysis,is clearlysuperiortothe purchaseplanor intentionsdesign.See F. ThomasJuster,\"ConsumerBuyingInten-tionsand PurchaseProbability:An Experimentin SurveyDesign,\"Journlaolf theAmeri--cacnStatisticalAssociationiV, ol. 61 (September1966),pp. 658-96.

82 Brookings Papers on Economic Activity, 1;1972 To test forthe existenceof differencesamongtimespanswe reestimatedboth the anticipatoryunit purchaseequationand the objectivemodelofrealautomobileexpenditurefsor two subperiodswithineachof two over-lappingmajorperiods.First,we estimatedthe 1954-59and 1960-66sub-periodsof the 1954-66span,and the 1960-66and 1967-71subperiodsofthe 1960-71span.The firsttwo subperiodscorrespondto differencesbe-tweenmeasurementosf the expectedpurchasevariablewithhighand lowsamplingerror,whilethe secondtwo correspondto the purchaseplan orintentionand the purchaseprobabilityversionsof the CensusBureau'sexpectedpurchasevariable.Furthermoret,he rate of priceinflationwassystematicallyhigherduringparts of 1954-59and 1967-71 than during1960-66. The anticipatorymodel unit equationsare summarizedin Table 2.Accordingto the Chow test for differencesin structurebetween sub-periods,the anticipatoryunitpurchasemodeldoesnot havea significantlydifferentstructureduringanyof thesubperiodsa, lthoughindividualcoeffi-cientsdisplayverylargedifferencesF. or example,expectedpurchasesarea less helpfulvariableduringthe periodpriorto 1960,whenthe variableis derivedfrom SRC data (negativesign, and t ratio of 0.5) than afterTable 2. Changesin Structureof AnticipatoryVersion of Unit PurchaseRate Equations,Various Subperiods1954-71a Regressioncoefficientsb RegressionstatisticsSubperiod Constant At 3 Ut Durbin- Standard Watson Z SZt_t A2 error statistic i-o1954:2-1966:4 0.45 0.117 0.032 -0.045 0.803 0.079 0.961954:2-1959:4 0.651 0.104 0.681960: 1-1966:4 (8.5) (4.3) (3.1) 0.934 0.039 1.651960:1-1971:2 1.79 -0.040 0.035 -0.085 0.883 0.047 1.911967:1-1971:2 0.712 0.056 2.19 (0.5) (3.0) (2.3) 0.44 0.122 0.033 -0.049 (7.2) (3.7) (3.4) 0.45 0.115 0.033 -0.043 (6.3) (8.2) 0.024 (4.6) 1.04 0.051 (2.2) -0.044 (0.6) (2.4) Sources: See Appendix B for sources and explanation of indexes of expected purchasesand of consumersentiment. The unemploymentrate is based on total unemployment of all experiencedworkers over 16 yearsof age, seasonally adjusted, from the U.S. Bureau of Labor Statistics. Also see discussion in text. a. The dependent variable, Ai, is the proportion of households purchasinga new automobile, seasonallyadjusted at annual rates. When applicable, each equation includes a dummy variable for major automobilestrikes. b. Symbols used in this table are defined as follows: At = Index of expected car purchases. 3E SZt_i = Four-quarterfirstdegree Almon lag on the filteredindex of consumer sentiment.i-o U= Unemployment rate.

F. ThomasJuster and Paul Wachtel 831960when CensusBureaudata become available(t ratio of 7.2). Inter-estinglyenough,althoughthe coefficientof expectedpurchasesis muchweakerin 1967-71thanin 1960-66,thereis littledifferencein the residualvarianceof the anticipatorymodelin the two spans.'3However,whentheobjectivemodelof automobileexpenditureiss estimatedfor the sametimeintervals,the Chow test indicatesa significantlydifferentstructurefor1960-66(a periodof low inflation)than for 1967-71(a period of highinflation),althoughthereis no structuradl ifferencebetween1954-59and1960-66. A plausibleexplanationfortheseresultsis theabsenceof a priceinflationvariablein the objectivemodel,sincethe structuradl ifferenceshowsup inthe high-inflationand low-inflationsubperiodsof the 1960-71span. Theanticipatorymodel may not show a comparablestructuraldifferencebe-causethe effectof priceinflationon automobilepurchaseratesis reflectedin the behaviorof the two anticipatoryvariablesthemselves-the indexofexpectedpurchasesandthe indexof consumersentiment.Thisspeculationis consistentwith resultsdiscussedbelow, which demonstratethe sub-stantialinfluenceof priceinflationvariableson both surveymeasures. JOINT OBJECTIVE-ANTICIPATORY AUTOMOBILE MODELS A questionthatarisesnaturallyis whetherthe optimalpredictionmodelfor real expendituresper householdon automobilesshouldincludebothobjective and anticipatoryvariables.If the anticipatoryvariablesareviewedas partof the desiredstock functionin the objectivemodel,theycan be introducedsimplyby addingboth expectedpurchasesand the fil-teredsentimentvariableto the objectivemodel; on this view, the antici-patoryvariableswould be part of the Z set in equation(1.2), or couldconstitutethe entireZ set. Alternativelyt,hesevariablescould replaceallthose reflectingstock adjustmentand formationof expectationsin the 13. The reason for this differencemay be that purchaserates for automobilesshowless total variabilityoverthe 1967-71span, hencethereis relativelyless systematic(andexplainable)variabilityand relativelymore rcanidovnamriabilityin purchasesduringthisperiod. Under these circumstances,the regressioncoefficientsof the independentvari-ables in the unit purchaseequationwill tend to havelargestandarderrors.And in fact,although the coefficientof A* in the 1967-71 periodis not significantlydifferentfromzero,neitheris it significantlydifferentfromtheA*coefficientduring1960-66.The othervariablesin the unit purchaseequationhavecoefficientsof approximatelythe same sizein eitherpartof the 1960-71 span,althoughthe standarderrorsare greaterin 1967-71.

84 Brookings Papers on Economic Activity, 1:1972objectivemodelin additionto servingas proxiesfor the level of desiredstock.In thateventtheywoulddisplaceall the economicvariablesin theobjectiveplannedexpendituremodel(stock, laggedexpenditurei,ncome,andrelativeprices),leavingonlythe transitoryinvestmentpart.Sincethisfunctionis thesamein bothmodels,thislastsituationsimplywouldreducetheobjectivemodelto theanticipatorymodel.Theissuecomesdownto anempiricalone, sincetheoryofferslittleguidance. Thesimplestwayto examinethequestionis to addbothsurveyvariablesto the objectivemodelandexaminethe resultingstructureT. able3 showsa selectionof such results,with equationsestimatedfor time spansthatcorrespondroughlyto differencesin the precisionwith whichthe surveyvariablesaremeasured. For the longestavailablespan, 1953-71,both expectedpurchases,A*,and the filteredsentimentindex,SZ*, significantlyimprovethe objectivemodel.The mechanismsfor stock adjustmentand formationof expecta-tionsareseriouslydiluted,andthepermanenitncomecoefficientfallsbelowstandardsignificancelevels.In additionthetransitorypartof theobjectivemodellosespoweralmostentirely.For theshorterspan,1960-71,in whichthe measureof expectedpurchaseshas substantiallyless samplingvari-ability,the resultsare even stronger.Here, additionof the two surveyvariablesdestroysthe structureof the originalobjectivemodel,with thesurveyvariablesandthe transitoryunemploymenrtatevariablebeingtheonly significantones in the equation.Estimatesfor the still shorterspan1960-67strengthenthis conclusion;herethe two surveyvariablesaretheonly ones close to statisticalsignificanceT. heseresultssuggestthat, forautomobiledemandmodels,the surveyvariablewill tend to dominateajoint modelprovidedestimatesarerestrictedto the periodbeginningwith1960,whenthe indexof expectedpurchaseshasreasonablysmallsamplingerrors. Thesimplestcombinedmodelcanbe interpretedas a reducedformthatincorporatesthe anticipatorymodel for unit purchasesand a supple-mentaryequationfor realpriceperunit.We showedabovethatthe prin-cipalvariablesin the firstarethe two surveyvariablesandthe unemploy-mentrate,andin the second,permanenitncome.To combinethetwo intoa singleexpenditureequationin principlerequiresmultiplicationof oneby the other.Theresultingspecificationcontainsall independenvt ariablesin eitherequationand all the cross-producterms.Eliminatingthe latteralongwiththe lag structurein the priceequationand the erraticrelative

! 3.Q;o_o- (Y~~~~~o~~~~~~C o\ . \ Fts C 11 11i- i D s _o ^ Is'D ^ v N; a = t l 6 ri (, 6 1 ,N ,R 3 XR < ~~~~;PC Lf____,SS > 0 90 6 0 t 0 > ^ _ ? >RE e . s= ~~~~~~~~n~-,,toCv ,Dt_~~_~3~~~~C CCt a Cm lo 0es X U) vo vo U) U=*;~n~~~~~~A~z~~~~~~~~~~~~~~~~~~~~~~~~~ FS o o cz

86 BrookingsPaperson EconomicActivity,1:1972pricevariableyieldsequation(3). The standarderroris smallerthan thatfor the objectivemodelby itselfshownin Table3.(3) CARt = 1.14 + 0.037 YP' + 29.9 A* + 4.69 SZt - 25.1 Ut. (4.5) (4.5) (2.7) (7.2) Periodof fit: 1960:1-1971:2. = 0.943; standarderrorof estimate= 16.2; Durbin-Watsonstatistic= 1.70. InflationEffects Whetherthe rate of price inflation-actual, anticipated,or unantici-pated-has an independentinfluenceon expendituresfor automobilesisan interestingquestionthat invitesexaminationat some length.Eventhedirectionof any inflationeffectis ambiguouson a priorigrounds. Economistshave usuallytakenthe viewthat an expectedrise in pricespromotesan increasein expenditureP: eoplewill tendto substitutegoodsfor moneyand thus to spendmoreand save less. The analysisis usuallycouchedin termsof previouslyunanticipatedchangesin the rateof priceinflation,althoughthat distinctionis not alwaysexplicit.Thusthe \"stan-dard\"viewis that a fully anticipatedrisein priceswill have no effectonrealeconomicdecisionsbecauseall the impactshavebeenfullydiscountedandembodiedin currentprices,interestrates,andso forth.Butthe expec-tation of a rise in prices,that was previouslyunanticipatedw, ill makegoods, particularlydurables,more attractiveand money less attractive. A persuasivealternativeview, whichprobablyowes its originmoretopsychologistsand sociologiststhan economists,is that a rise in prices(anticipatedornot)willtendto increasesavingandreducespending.1T4 heargumentis oftenput in termsof the impactof risingpriceson consumerconfidenceor consumeroptimism:Risingprices,accordingto surveydata,tend to be associatedwith unfavorableconsumerreactionsand weakerconfidence. One interpretationof the survey-basedfindingthat rising prices stirpessimismfocuseson the effecttheyhaveon consumerexpectationsaboutrealincome.Historicallyh, ighinflationratestendto be associatedwith a 14. One of the earliest and most consistent proponentsof this view was GeorgeKatona of the Survey Research Center. See, for example, The Powerful Consumer:PsychologicalStudiesof theAmericanEconomy(McGraw-Hill,1960).

F. ThomasJusterandPaul Wachtel 87relativelyhigh variancein the rate of infla'tionI.f consumerscommonlybelievethat the rate of increasein nominalincomewill be less variablethan the rate of increasein prices,the expectationof risingpriceswillgenerategreaterdispersionof expectationsabout real income. A widerdispersionmaynot havesymmetricaelffectson behavior,in thatthe pros-pectof decliningrealincomemaycarrymoreweighton consumerdecisionsthanthe prospectof risingrealincome,eventhoughthe two areregardedas equallyprobable.In short,consumersmay be muchmore concernedthat priceinflationwill erodetheir real incomethan pleasedthat risingnominalincomeswilloutweighrisingprices.If so, theappropriatereactionto inflationaryexpectationswouldbe to curtailspendingin an attempttoguardagainstdecliningrealincome,thus,as a corollary,raisingthe savingrate. Thesameasymmetryshowsup in theeffectof priceinflationon the realvalue of assets.Consumerswith variablepriceassetshave no reasontosupposethatrisingpriceswillerodetherealvalueof theirassets.Butmanyhave only fixed-priceassets whose purchasingpowerwill be erodedbyrising prices.If the relationbetweenasset position and expendituresisstrongerfor those whose assets are mainlyfixed-price,the net effect ofinflationon expenditurestemmingfromthe assetssidewillbe restrictive. Thereseemsto be no way to settlethis argumenton a priorigrounds.Probably,steadyand moderateratesof priceinflationwould have quitedifferenteffectson consumerexpendituresthan high or variablerates;moreover,the anticipatorybuyingeffectsmay not be strongat all unlessratesof priceinflationarequitehigh.EvidencefromtheSRCsuggeststhat,amongU.S. consumersr, isingpricesinhibitrealexpendituresa,lthoughthesurveyis ambiguousabouttheexactreasonswhyrisingprices(andwages!)make people feel less optimistic.15Ultimately,one is forced back onempiricajludgments. We have experimentedextensivelywith the effects of inflationrates,both anticipatedand unanticipatedo, n spendingand savingbehavior.Asnoted earlier,we have compileda serieson expectedpricechangesfromSRCdata;it is quiteunsatisfactoryin manyrespects,however,becauseofdifferencesin measuringexpectations.Priorto 1966,the surveyreportedonlywhetherconsumersexpectedpricesto go up or go down or remain 15. To be precise,the surveyfinds that consumerscommonly associaterisingpriceswith \"bad\"economicconditions. But it is difficultto identifythe precisesets of eventsthat constitutethese unsatisfactoryconditions.

88 Brookings Papers on Economic Activity, 1:1972unchangeda; ndit wasonlystartingin 1966thatthosewhoexpectedpricesto rise wereaskedto specifya percentageW. hilewe have constructedauniformindexof priceexpectationsfrom1953to thepresent,it maynot bea whollysatisfactorymeasureof expectedpricechange.The problemisdiscussedin AppendixB. In ourexperimentswithalternativemeasuresofactualpricechange,the consumerpriceindexappearsto be morecloselyassociatedwith actual behaviorthan the implicitdeflatorfor personalconsumptionexpendituresa,ndis the variantreportedhere. In general,we use two variablesto reflectthe influenceof inflationonconsumerbehavior:the expectedrate of inflationconstructedfrom theSRCdata,labeledCPI*,and the actualrateof inflation,CPL.16 Theregressioncoefficientson actualandexpectedinflationcanbe inter-pretedto show the effectsof anticipatedand unanticipatedinflation.Atotal inflationeffectis givenby a, CPI + a2 CPI*,whereal and a2 are regressioncoefficients.Inflationis fully anticipatedwhenexpectedinflationis equalto actualinflationandthuscanbe substi-tutedfor it, as givenby a, CPI* + a2 CPI*.The effectof unanticipatedinflationis the total effectless the anticipatedeffect,or a, (CPI - CPI*).Thus,the sum of the regressioncoefficientsis interpretedas the effectofanticipatedinflation,and the coefficienton actualinflationis interpretedas the effectof unanticipatedinflation.Thatis, the unanticipatedinflationeffectis the partialderivativewith respectto the actualrate of inflation,holdingexpectedinflationconstant.17 AUTOMOBILE DEMAND MODELS AND INFLATION The resultsfromintroducingthe priceinflationvariableinto the antici-patory model of automobileunit sales provideevidencethat both the 16. CPI* is the averageof the expectedrate of inflationfrom surveystakenin t - 1and t - 2. CPIt is the averageof the actualrateof inflationin the samequarters. 17. George Perrysuggestedthis interpretation.Although a differentalgebraicap-proachwill yield a differentinterpretationt,his one has the most logical appeal.

F. ThomasJusterandPaul Wachtel 89expectedandthe actualrateof priceinflationhavea negativeinfluenceonautomobilepurchaserates.Table4 containsa collectionof suchresultsforthe 1960-71period.Theinflationeffectsaresignificantandnegativewhenthe filteredsentimentvariableis used withoutAlmon distributedlags.Whileeitherinflationvariablehasa negativesignin the optimumequationform,whichuses the Almon lag, the effectsarenot significantlydifferentfromzero. Theseresultssuggestthat muchof the inflationeffectmayal-readybe containedin the surveyvariables,a subjectexaminedin greaterdetailbelow. We alsoexaminedtheeffectsof inflationon the objectivemodelof auto-mobiledemand,andon thesupplementarayutomobilepriceequationusedin the anticipatorymodel.In the objectivemodel,no strongor systematiceffectof anyof the inflationvariablesappearedI. n the realpriceequation,however,the expectedrateof pricechangehasa significantnegativeeffecton averagerealcarprice,whileactualpricechangehad a weakerpositiveeffect.Thusa fully anticipatedinflationhas a negative impacton averageprice,whileanyunanticipatedpricechangehas a positive impact. The possibleinfluenceof specialinflationeffectsin the real car priceequationduringtheperiodof relativelyhighinflationrates,1967:3through1971:2,was also explored.Thetestconsistsof addingvariables,operativeTable 4. AutomobileUnit PurchaseEquationswith InflationVariables,First Quarter 1960 throughSecond Quarter 1971a Regression coefficientsb Regressioni statisticsConstant A* 3 U, Inflation variables Durbini- CPIt CPI* _ E SZ'_ _ Standard WVatson SZ, iO R2 error statistic _0.47 0.123 0.014 ... -0.054 -0.016 ... 0.865 0.050 1.72 (6.8) (2.3) (5.1) (2.5)0.57 0.129 0.016 ... -0.059 ... -0.064 0.867 0.050 1.77 (6.8) (3.2) (5.5) (2.6)0.52 0.130 0.014 ... -0.057 -0.008 -0.041 0.866 0.050 1.79 (6.9) (2.4) (5.3) (0.8) (1.2)0.42 0.123 ... 0.027 -0.047 -0.010 ... 0.887 0.046 2.01 (4.2) (4.9) (1.6)(8.4)0.47 0.124 ... 0.029 -0.047 ... -0.028 0.833 0.046 1.98 (7.8) (4.7) (4.6) (1.1) Source: See Appendix B and sources for Table 2. a. The dependent variable is the automobile purchase rate, A. Regressions also include strike dummyvariables: A single dummy is used in the Almon lag equations, four separate dummies in the others. b. Symbols used in this table are defined as follows: At = Index of expected car purchases. U= Unemployment rate. SZt = Filtered index of consumer sentiment. CPI, = Actual rate. 3 CPJt = Expected rate.Z SZ?-= Almon lag on filtered index of consumer sentiment.i-l,

90 BrookingsPaperson EconomicActivity,1:1972only during the 1967-71 period, to an equation that explains average realcar price in terms of income, lagged real price, and both actual and ex-pected price change. These \"interaction\"terms consist of variables definedas Z1X, where X is an inflation variable and Z, has a value of unity duringthe 1967-71 (high-inflation) period, and of zero at other times. Estimatingan equation with both X and Z1X permitsidentificationof both the generaleffects of price inflation during the period as a whole, through the coeffi-cient of X by itself, and any special effects associated with the 1967-71period, through summing the coefficients of X and Z1X. We have appliedthe interaction of the high-inflation period to the expected and actualprice change variables as well as to the intercept. The interaction terms aredesignated CPID and CPI*D for actual and expected inflation, respec-tively, and INFD for the intercept. With these interaction terms used to isolate any special characteristic ofthe 1967-71 period, inflation effects appear to have been quite differentinthose years from the rest of the period; but the standard errors are so highas to limit confidence in this result. The real price equations, with expectedand actual price change and with the interaction terms added, are shownbelow.(4) Vt = 95 + 0.225 YP' + 0.531 Vt-i - 56.7 CPI* + 8.8 CPIt. (4.3) (5.3) (3.3) (1.3) Periodof fit: 1953:4-1971:2. = 0.942; standarderrorof estimate= 62.8; Durbin-Watsonstatistic= 1.93.(4.1) Vt = 133 + 0.221 YP' + 0.523 Vtl - 58.1 CPI* + 52.5 CPI*Dt (4.1) (4.6) (3.3) (0.5) + 12.1 CPIt - 23.8 CPIDt - 57 INFDt. (1.6) (1.3) (0.2) Periodof fit: 1953:4-1971:2. R2 = 0.941; standarderrorof estimate= 63.3; Durbin-Watsonstatistic= 1.99.The net effect of inflation on averagerealcar price in dollars was as follows: Fully Unanticipated anticipated inflation inflation (n = percentage rise in prices)Withoutinteraction -47.9n 8.8nWith interaction -46.On 12.1n 1953-66 -17.3n - 57 -11.7n - 57 1967-71

F. ThomasJusterandPaul Wachtel 91 Takenat facevalue,the resultsindicatethat,fullyanticipatedi,nflationusuallytendsto reducereal expendituresper car substantiallyw, hile un-anticipatedi,t tendsto increasethemslightly.However,duringthe 1967-71period,accordingto ourbest estimate,inflationhadthe specialcharacter-istic of reducingreal expenditureper car, whetherit was anticipatedorunanticipatedt,houghtheeffectwassomewhatmorenegativein theformercase. Demandfor NonautoDurables An objectivemodelfor nonautodurablesis almostidenticalto that forlautomobiles;the model includesan income variable,both currentandtaggedunemploymenrtates,a stockadjustmenmt echanismas reflectedbythe beginning-of-periosdtocksof otherdurables,and an adaptiveforma-tion-of-expectationms echanismE. quation(5)belowcontainsthe optimumversionof this model,in whichall parametershavethe appropriatesignsand all but beginning-periodstocks have satisfactorysignificancelevels.Otherdurables(OD) aredefinedas total expenditureos n consumerdura-bles(conventionaDl epartmentof Commercedefinitionl)essgrossexpendi-tureson newcarsandnet purchasesof usedcars,deflated,perhousehold.(5) OD = -232 + 0.072 Yt - 9.0 Ut + 5.8 UtI (5.5) (2.8) (2.0) + 0.056 Kt-I + 0.468 ODt-1. (1.5) (4.2) Periodof fit: 1953:4-1971:2. J2 = 0.996; standarderrorof estimate= 9.6; Durbin-Watsonstatistic = 1.60. We did not constructa pure anticipatorymodel for other durables,sincethe expectedpurchasevariableusedin the automobileequationhasno good counterpartO. rdinarilyt,he availablemeasuresof expectedpur-chasesfortheotherdurablescategoryhavelittleif anynetassociationwithactualexpenditurea,ndthe difficultiesof comparabilityovertimeareevenmore seriousherethan for automobiles.Nonetheless,it is of interesttotest,in the otherdurablesmodel,thesametwo anticipatoryvariablesusedfor automobiles.The filteredsentimentvariableis presumablyas appli-cableto expenditureson otherdurablesas it is to those on automobiles,whilethe variablefor expectedautomobilepurchasesis not irrelevantifthey are representativoef expectedpurchasesof durablesgenerally.We

6 'IO C~ co 42 aS8 oM ? oFc ab~C~~~C ON sro C08 1N ~o~ o o~ON o~ No~CDI R cc II .e- Cl)~~~~~~~o' ~~o~_~~~~~o~.00~~o~'*o ~~~~~~11~11~3~~co~~~~~~~~~~~~~~i~.~o ~ l? R 6 o6 6. -N . . mim po I I I I I me >kqI-d0'~~~~~~~~~0 - ar~- aog c.c o E = e~cl~~~~~~~~~~~_ I ~ ~ ~ ~ ~ ~ ~~~~c0~~~~~~~~~~~~~c V.~0 O .00Po ~~~~~~~~ Mo MO 0 0~~~~ .00 ~~~~~~~~~~w 'aE z oCIS I Iq

F. ThomasJuster and Paul Wachtel 93also tested the expected and actual price inflation variables in the otherdurables regressions. Table 5 shows the results of these experiments:The first line contains thestandardmodel for other durables; the second and third lines add the twosurvey variables to the standard model for other durables, and test boththe S and SZ versions of the consumer sentiment index; the last two equa-tions add the two price inflation variables first to the standard model forother durables, and then to the model with the survey variables. Both survey variables appear to be marginal candidates for improvingthe specification of a model for other durables: They are significantlyweaker than in the comparable automobile expenditure equation. In thethird equation, using S, the sentiment index itself, in place of the filteredvariable SZ, S had a very weak influence on expenditures. INFLATION The two price inflation variables improve the equation significantly:Expected price change has a significant negative impact on expendituresfor other durables, while the impact of actual price change is also negative,but smaller and not quite significant at conventional levels. Thus a fullyanticipated inflation has a strong negative impact on real expenditures forother durables, while unanticipated inflation has a small negative impact.Thereis no evidence that the impact of inflation is differentin the 1967-71part of the period from the period as a whole. We tested in the other dura-bles equation the interaction terms used in the equation for real car prices,but the coefficients were generally small and all terms had standarderrorswell in excess of the coefficients. Finally, the survey variablesare weakenedwhen the inflation variables are added; the filtered sentiment index, inparticular,apparently is redundantto the price inflation variables. Predictionsof Survey Measures The evidence seems to bear out that the expected purchaseand consumersentiment variablescan improve the explanation of spending on both auto-mobiles and other durable goods. The optimum form of prediction equa-tions for a durable goods demand model seems to involve both surveyvariables,probablyin conjunction with some standardeconomic and finan-

94 BrookingsPaperson EconomicActivity,1:1972cial variables.The degreeto whichsuch modelsare usefulfor short-termforecastingdependsupontheanswersto twoquestions.First,do thesurveyvariablesimprovepredictionfor morethan one quarterinto the future?Second,how well can the surveyvariablesthemselvesbe predicted?Asubsidiaryquestionof considerableinterestis, How does priceinflationaffectthe surveyvariables? EXPECTED PURCHASES We havehadonly moderatesuccessin devisingequationsto predicttheindexof expectedpurchasesfor automobiles.The A* variableusedin theeconometricworkaboveis actuallya weightedaverageof currentandpastvalues;hencein this sectionwe usethe originalvariable,A**,ratherthanthe weightedaverage.The presumptionis that it is betterto predictA**,andfromit calculatethe weightedA*, thanto tryto predictA* directly.18 We estimateequationsfor both expectedpurchasesand expectedexpen-dituresperhousehold(expectedpurchasestimesaveragecarprice,A**V).Thelattervariableis suggestedby theargumenthatexpectedexpendituresarea reasonableproxyfortheplannedinvestmentcomponentof theobjec-tivemodel.We candirectlysubstituteA**Vfor GPin the objectivemodel,as the unitsof measuremenatrecommensurate. The reducedform equationfor GPis shown in AppendixA as (A5).It specifiesa positivecoefficientfor the laggeddependentvariableand anegativecoefficientforlaggedexpendituresw, hilethe stockcoefficientcanbe of eithersign. The results,shown in the top line of Table6, are consistentwith thismodelexceptforthe signof CAR,-,, thelaggedexpenditurevariable.Thatresultcouldeasilybe due to the strongcollinearitythat mustexist in theequation,sinceA**V,CAR,K,and YPall havestrongupwardtrends.Thedeterminantosf plannedexpenditureosn automobilesimpliedbytheequa-tion arepermanentandtransitoryincomealongwiththe expectedrateofpriceinflation.The secondequationshowsthe best predictionmodel forthe expectedpurchaserate variable,A**. It includesthe same variablesas thefirstequationexceptthatthelaggedstockandexpenditurevariablesaredroppedandthe inflationinteractiontermsforthe 1967-71periodareadded.The thirdequationreestimatesthe firstmodel with the inflation 18. At is defined as 0.6A*t*+ 0.3A ** + O.A**.

.SQ2 a2\" t^=C %) t? '' D X . ~~~~~~~~a0,0 a CA ? , o_ ' ?.0 MrrxMx~~~~~~~~~~~~~~~~~~~~~~~Cl I l l 11 11 1111~~~~~~~~~~~~~~r e E4 I O'-W,l0N0 !2 }| O o?~~~~~~~~~~~~~~~C F U) Q0 Q0 Q0 Q0 E c -

96 Brookings Papers on Economic Activity, 1:1972interaction terms added and the CAR,-1 variable, which had the wrongsign, omitted. Expected price change, which has a negative coefficient throughout, isthe only inflation rate variable with any measurable strength in the ex-pected purchase equations. There is evidence, primarily from the A**prediction model, that the effect of price inflation on expected purchaseswas not the same during 1967-71 as during 1960-66, as shown by theinteraction tests in the second equation. Here, the results imply that antici-pated inflation generally has a negative influence on expected purchases,which during the 1967-71 period was independent of variation in inflationrates but more sharply negative than before. This interpretation followsfrom comparison of the slope and intercept interaction variables in thesecond row. While the net effect of different anticipated inflation ratesduring the 1967-71 period is nil (the sum of the CPI* and CPI*D coeffi-cients is virtually zero), there is a large negative constant term applicableto the period 1960 as a whole. Thus the level of expected purchasesappearsto have been depressed during 1967-71 relative to the earlier part of theperiod, other things equal. A similarbut weakereffect shows up in the thirdequation, where expected expenditures is the dependent variable. INDEX OF CONSUMER SENTIMENT Predictions for the consumer sentiment variable19are shown in Table 7;basically, we follow the analysis suggestedby Hymans, with some modifica-tion.20 Our best results suggest that the stock price level has an influence,although Hymans did not find it useful, and we also find some interestingand different results on the impact of price inflation on the consumersentiment index. However, our results are extremely sensitive to the defini-tion of a price inflation variable, as regards not only the effect of inflationon S, but also the influence of stock prices. Because we were interestedin comparability, the S predictionsare basedon the period since 1960. The explanatory variables that proved to be im-portant are stock prices, a variable reflecting the change in stock prices,a price inflation variable, and the lagged dependent variable. The best fitsare obtained with the price inflation variable suggested by Hymans, labeled 19. Thedependentvariablehereis S, thelevelof theindexof consumersentiment.Wedo not attemptto predictSZ, the filteredindex usedin the econometricwork. 20. Hymans,\"ConsumerDurableSpending.\"

N CN 00 r- II zt x= 0 ll% O 13 09 O OCld 0 Eas 0 CZ C,3 0 Ca f0l r- C's CN 0 0-4 > cl=43 0r-) Ecrscli Cfj tn Ca UQ;t O . 0 C1o3 'c o o 2 Ca r Ca V) uu tn M o V) 00 09 eq QL4 cri F-T 'o. L14 09 o 00 aQ,O cfsQ, 00 cq tr; c,3 C,3 kE4n 4E) >' -1 o 0'a0 -O aL.s Z - f-01 to 00 to C,3 O,.,4 4) u .. C :5 C's ooo Ci tn C,:3o6nvZ -0 00 c- q E (: E 0> CZ - u X0 O.tDON m t- C:-> a V v) 'O C C; X .04) r. rc.q I. ro-0u E 10, E00r- 0 Ca u o 00 t- v) 0 Li 0 C,3Po cl

98 BrookingsPaperson EconomicActivity,1.1972PCH; the CPI and CPI*variablesusedthroughoutthe analysisnot onlyareless effectivein theseequations,but also reducethe influenceof stockpricesvirtuallyto zero. Both measuresof actualinflationrates have anegativeimpacton sentimentwhileexpectedinflationhas a smallpositiveeffect.Thefirstandthirdequationsin Table7 suggestthatfullyanticipatedinflationhas no effecton sentiment,while unanticipatedinflationhas anoticeablenegativeeffect. Table7 also showsequationswith 1967-71interactionterms,designedto testfordifferenceisn theeffectsof priceinflationon consumersentimentbeforeandafter1967.Butthe equationswitha timeperiodinteractiononthemosteffectivepriceinflationvariable(PCH)provedto be unsuccessful;in general,no significandt ifferenceappearedbetweentheeffectsof PCHonsentimentduring1967-71and the effectsover the entire 1960-71 span.For the otherinflationvariables,however,somedifferentiaelffectsshowedup. For expectedinflation,CPI*,as well as actualinflation,CPI, the co-efficientof the price inflationvariableis sizable and negativefor the1967-71interactionterm.On the otherhand,the interceptinteractionsinthese equationsare positive.The combinationof a negativeslope inter-actionand a positiveinterceptinteractionsuggeststhatmoderateratesofpriceinflationhad, on balance,a positiveinfluenceon sentimentduringthe 1967-71period,whilehighrateshad a net negativeinfluence. TheAllocationof Incometo SpendingandSaving Earliersectionsexaminedtheroleof consumersurveydatain specifyingdurablesdemandmodelsand the influenceof inflationin these models.Theseimportantand relatedissueshelp in formingan understandingofrecentconsumerbehavior.We have arguedthat the surveyscontaineco-nomicinformationnot availablein an objectivemodelandthatthe infla-tionaryexperiencemakesrecentyearsunique.If theseargumentsarevalid,the purviewof this papershouldnot be limitedto automobilesand otherdurables.If, at a givenlevelof income,consumersentimentandtherateofpricechangehave an effecton one sectorof consumerexpendituret,heymust necessarilyhave an oppositeeffect on some other aspectsof con-sumerbehavior.For example,if, at a givenlevelof income,an increaseinconsumersentimentpromotesexpenditureson durables,some othertypeof expenditureor savingmustfallby a like amount.It is necessarythere-

F. ThomasJusterandPaul Wachtel 99fore to examinea full model of incomeallocationand determinationinorderto understandfullyboththeeffectsof inflationandthe usefulnessofsurveydata. A full modelis beyondthe scopeof thispaper,but the issuesareexam-inedwithinthecontextof a simplesystemof incomeallocation.A seriesofequationsof the followingformareestimatedby least squares: Ei =ai + bli YP + b2i YT + ,giiXi i = 1, . ,n, iwheretheEi arethe allocationsof income; YPand YTarepermanenat ndtransitoryincome,respectively;and the Xj are the surveyand inflationvariables.Systemsof expenditureequationsof this form are useful forexaminingthe issuesat hand becauseleast squaresestimateswill satisfyallthenecessaryconstraintsconcerningsummation;thatis, when n ,Ei = YP + YT, i=1the least squaresestimateswill satisfythe constraintsthatand that nn =bli b2i=1, i=1 i=1 gi= 0 for allj,21wherethej subscriptdenotesseparatevariablesin the X vector. Four allocationsof disposableincomeare examined:expendituresondurables(DUR); expenditureson nondurablesand services(CX); andpersonalsaving(PS) dividedinto the netincreasein consumerinstallmentcreditliabilities(CC) and othernet saving(ONS). Since personalsavingis equalto othernetsavingless the increasein consumerinstallmentcreditliabilities,we defineCC as havinga negativesign.Algebraicallyt,herefore, 21. A simple proof that the constraints hold is found in S. J. Prais and H. S.Houthakker,The Anialysisof FamilyBudgets(second impression,abridged;London:CambridgeUniversityPress, 1971). In general,for any system of equationswith com-mon independentvariablesthat includesthe sum of the dependentvariables,the cross-equationcoefficientsumswill be unityfor the dependentvariablesandzero for the othervariables.In the presentcontextit is expectedthatan increasein theconsumersentimentindex,forexample,willlead to increaseddurablesexpenditureso, therthingsbeingequal.The constrainedsystemwill reveala negativeeffecton some otherallocationof income.The offsettingeffectsof the surveyandinflationeffectsrevealedby the modelindicatethedirectionand degreeof substitutability.

100 BrookingsPaperson EconomicActivity,1:1972Y = CX + DUR + PS, and Y = CX + DUR + ONS + CC, where thelast term is negative when installment credit increases, and positive whenit decreases.The data are all from the national income accounts except forthe change in consumer credit, which is obtained from the Federal ReserveBoard. Two specifications of the system are shown in Table 8. In Panel A, eachregression includes permanent and transitory income, the survey measureof expected inflation (CPI*), and the actual inflation variable (CPI). PanelB results include the income variables, the index of expected purchasesTable 8. RegressionEstimates of Allocation of Incometo SpendingandSaving, Second Quarter1954 throughSecond Quarter1971a Regressioncoefficientsb Effects Saving SpendingIndependentvariableand ONS CC DUR CX regressionstatisticConstant -383.1 71.1 Panel A 312 -312 (4.8) (1. 1) 0.11 0.89Real permanentincome -931.9 1,243.9 0.17 0.83 per household, YP 0.1325 -0.0212 (12.8) (15.1) -37 +37 (9.5) (2.0) +22 -22Real'transitory income 0.2843 0.6044 per household, YT 0.3758 -0.2078 (22.5) (42.5) ... ... (5.6) (4.0)Expected inflation rate, CPI* 0.4019 0.4302 -27.81 -8.85 (6.6) (6.3)Actual inflation rate, CPI (3.0) (1.2) -28.24 64.89R2 12.57 8.94 (3.3) (6.8)Durbin-Watson statistic (2.3) (2.1) -1 .42 -20.08 0.828 0.368 (0.3) (3.6) 0.79 0.49 0.962 0.993 0.70 0.66 Panel BConstant -307.6 111.8 -809.0 1,004.9 -196 196 (6.3) (3.6) (17.9) (15.3) 0.20 0.80Real permanent income 0.38 0.62 per household, YP 0.1845 0.0107 0.2238 0.5816 -88 +88 (10.1) (0.9) (13.4) (23.9)Real transitory income -5 +5 per household, YT 0.4481 -0.0644 0.2415 0.3748 ... (5.5) (1.2) (3.2) (3.4)Index of expected car purchases, A* -56.24 -31.45 28.82 58.87 (3.7) (3.2) (2.0) (2.9)Filtered index of consumer sentiment, SZ 5.86 -10.38 7.59 -3.07 (2.0) (1.9) (2.8) (0.8)T2Durbin-Watson statistic 0.836 0.594 0.963 0.988 0.76 0.72 0.68 0.29 Sources: Basic data for the dependent variables are from U.S. Office of Business Economics, and, forconsumer credit, Board of Governors of the Federal Reserve System. For independent variables, see Ap-pendix B and discussion in text. a. All monetary variables are in constant dollars per household. b. Symbols used in this table are defined as follows:ONS = Other net saving (personal saving minus consumer credit change). CC = Decrease in consumer installment credit liabilities.DUR = Expenditures on durables. CX = Expenditures on nondurables and services.

F. ThomasJuster and Paul Wachtel 101(A*), and the filteredindex of consumer sentiment(SZ). The impacts of theinflation variables and the survey variables on income allocation are ex-amined separately since the survey variables presumably reflect the in-fluence of inflation. The expendituresystem is, of course, a highly simplifiedmodel, designedas a frameworkfor examining inflation and expectational effects on house-hold behavior, and not as a fully specifiedbehavioral model. It will provideestimates of marginal propensitiesto spend and save out of permanentandtransitoryincome, and of the effects on allocation of changes in the rate ofinflation and in buying intentions and sentiment. The effect of the last twois expected to be positive on expenditures, especially for durables. Sincethese variables measure a general willingness to buy, a tradeoff againstsaving is expected. The inflation effects are ambiguous. As noted earlier,inflation might prompt a shift into tangible assets in the form of durablesand out of financial assets as a form of protection against inflation, orinflation might aggravate consumer uncertaintyabout the outlook for realincome and induce curtailment of spending as a form of precaution. Thetwo hypotheses can be distinguished only empirically. The coefficients on the inflation variables are small, but the t ratiosexceed two in six of eight cases in Panel A. The effects of fully antici-pated and unanticipated inflation implied by the results in Table 8 aresummarized in Table 9. They suggest that unanticipated inflation doescause the household to reallocate income toward saving and away fromspending. The negative effect of unanticipated inflation on consumptionof nondurables and services and on total spending including durables,supports the uncertainty hypothesis. Fully anticipated inflation raises con-Table 9. Summaryof InflationEffects on HouseholdAllocationof IncometoSpendingand Saving, Second Quarter 1954 throughSecond Quarter 1971Constant1958dollarsof allocationper householdper percentof inflationperyear Inflation ONSa CC D UR Effectsof inflatioexpectation CX Saving SpendingFully anticipated -15.2 +0. 1 -29.6 44.8 -15.1 +15.1Unanticipated +12.6 +8.9 -1.4 -20.1 +21.5 -21.5 Source: Derived from Table 8, Panel A. a. Symbols used in this table are defined as follows:ONS = Other net saving = personialsaving minus consumer credit change. CC Decrease in consumer installment credit liabilities.DUR = Expenditureson durables. CX = Expenditureson nondurables and services.

102 BrookingsPaperson EconomicActivity,1:1972sumptionexpenditureosn nondurablesandservicesat the expenseof bothsavingandexpenditureson durables. Theseresultsarehighlytentative.The equationsexhibita greatdeal ofserialcorrelation,however,and the possibilitythat the omittedvariableshavebiasedthe resultsis thereforegreat.For example,inflationratesandinterestratesare positivelycorrelated;the effectsof the former,whichisincluded,couldrepresenthoseof thelatter,whichis omitted.Thetradeoffeffectsshouldbe consideredwith thesereservationsuntil a full model ofhouseholdbehavioris analyzed. The influenceof the sentimentand buyingplansvariableson expendi-turesis clearand striking.The coefficientson A* and SZ in PanelB arealmostall significanat ndreveala definitepatternof substitutionE. xpectedpurchasesmeasurea generalwillingnessto spend on both durablesandotherconsumption;the negativeeffecton savingis strong.Somewhatdif-ferentresultsarefoundfor the filteredsentimentindex.Thatvariablewasdesignedfor the durablesmodelsandindicatesthat greateroptimismhas anegligibleeffecton consumptionof nondurablesand services. A finalpointof interestis therelativemagnitudein bothsetsof expendi-tureequationsof thecoefficientson permanenat ndtransitoryincome.Thetransitoryincomecoefficientsarelargerthanthose on permanentincomefor all equationsexceptconsumptionof nondurablesand services.Theseresultsare not inconsistentwiththe permanentincomehypothesis.How-ever,thetotalpropensitietso spendout of permanenat ndtransitoryincomearenot verydifferenta, ndthe transitoryincomeeffecton consumptionofnondurablesand servicesis as largeas it is on durables.22 FORECASTING CONSUMPTION AND SAVING The resultsfromPanelA of Table8 wereusedto generatesome fore-casts beyondthe sampleperiod using actualdata for the period up to1972:1 and hypothesizedinflationrates and incomegrowthratesthere- 22. The permanentincomehypothesisholdsthatthe flow of consumption,definedtoinclude the services from durablesbut not their purchase, should be unaffectedbytransitoryincome.To the extentthattheyincludesomechangesin inventoryratherthanpure consumption,expenditureson nondurablesand services ought to be related totransitoryincome,but the coefficientseems muchtoo highto be the resultof inventorychangealone. See Milton Friedman,A Theoryof the ConsumptionFunction(PrincetonUniversityPressfor the National Bureauof EconomicResearch,1957).

F. ThomasJusterandPaul Wachtel 103after.Alternativeassumptionsforthegrowthrateof realdisposableincomewere(a)6 percentfortherestof 1972and5.5percentduring1972;and(b)7percentthroughoutheforecastperiod.Foralltheinflationratesexamined,the higherlevelof incomegrowthyieldedhigherpredictedratiosof savingto incomeanddurablesexpendituretso income,andcorrespondingllyowerratiosof otherconsumptionexpendituresto income.The differencesgen-eratedby the alternativeincomegrowthassumptionsare small,however,andthe discussionbelowusesassumption(a) above.As both inflationratevariablesin the model use lagged data, the actual numbersavailablethroughthe first quarterof 1972 provideestimatesthrough 1972. Forpredictionsfor 1973, alternativeinflationassumptionsweremade: highanticipatedinflation(CPI = CPI* = 6 percent);low anticipatedinflation(CPI= CPI* = 3 percent); high unanticipated(CPI = 6 percent andCPI* = 3 percent),andlow unanticipated(CPI = 3 percentand CPI*0). Althoughthe modelunderpredicttshe recenthighratesof personalsav-ing (8.2 percentin 1971),it accuratelypredictsthe 7.0 percentrate in1972:1. Under the assumptionof high anticipatedinflationthe rate ispredictedto be 7.7 percentin 1973:1 and 7.8 percentin 1973:4, whileunderthe assumptionof highunanticipatedinflationthe savingratiowillbe higher:8.3 percentin 1973:1 and 8.4 percentin 1973:4. The lowerinflationratesreduceall the predictedsavingrates,but by no morethan0.1 percent.In other words, a given amount of unanticipatedinflation(actualexceedsexpectedby 3 percentagepoints)willincreasethe personalsavingratesubstantially(by0.6 percent),whileanequivalentrisein antici-patedinflation(from3 percentto 6 percent)willleadto an increasein thesavingrateof only0.05 percent. Thesameassumptionsaboutincomegrowthandinflationratesproducea differentpatternfor expenditureson durables.With high anticipatedinflation,the ratio of durablesoutlaysto incomeis predictedto be 15.8percentin 1973:1; with low anticipatedinflation,16.8 percent.For un-anticipatedinflation,bothratiosincreasesubstantially1: 6.7percentforthehighassumptiona, nd17.7percentforthelowassumptionA. smallincreasein each of theseratiosis predictedduring1973,giventhe incomegrowthassumptions. Thus,in general,higherratesof inflationincreasesaving,reducedura-bles expenditures,and increaseexpenditureson nondurablesand ser-vices; on balance,spendingis reducedslightly.But if the inflationis un-

104 Brookings Papers on Economic Activity, 1:1972anticipatede, itherhigh or low ratesof inflationwill increaseboth savingand expenditureson durables,and will sharplyreduceexpendituresonnondurablesandservices;on balance,spendingis reducedmarkedly. SummaryandConclusions Two principalconclusionsemergefrom this paper.The firstconcernsthe usefulnessof data on consumeranticipationsin demandmodels,thesecondthe role of anticipatedand unanticipatedinflationon consumerspendingbehavior. ANTICIPATIONS DATA Ourmost firmlydocumentedconclusionis that explicitlyexpectationalmeasuresof consumerintentionsand attitudesare of considerableimpor-tance in models forecastingexpenditureson automobiles:The index ofexpectedpurchasesfor automobiles,as well as the filteredversionof theconsumersentimentindex,contributessignificantlyto the explanationofpurchasebehavior.We find that forecastmodelsthat involvean appro-priatespecificationof thesetwo surveyvariablesadequatelyexplainall ofthe aspectsof adjustmentlag and expectationsformationin the modelsordinarilyused in econometricwork, and the only economic variablebesidesthe two surveymeasuresthatis relevantis one thatcapturestran-sitoryandunexpectedincomechange,suchas the unemploymenrtate.Fornonautodurables,economicvariablescontinueto exerta stronginfluenceand the surveyvariablesappearto be marginallyuseful in predictionmodels.Thus,successfulpredictionsof durablegoodsdemanddependonabilityto extrapolatethe two surveyvariablessome time into the future. Interestinglyenough,it appearsthatthe two surveyvariablesexaminedhereareusefulbeyondthespecificpurposes-predictionof consumerdura-ble goods demand-for whichthey wereconstructed.Both are generallysignificantin functionsdesignedto explainpersonalsavingandconsumerexpenditureson nondurablesand services.Althoughthe equationsystemwithinwhichwe examinesavingandnondurablesconsumptionis a verysimpleone and is a relativelyweakbehavioralspecificationt,he strengthof the two surveyvariablesin it encouragesbeliefthat even a fully speci-fied model of savingor expendituremighthave a usefulrole for survey

F. ThomasJuster and Paul Wachtel 105measures of consumer plans and attitudes. Hence, we find persuasive evi-dence that these expectational variableswill be useful in any fully developedeconometric system of short-term demand forecasts, although perhapsmore so in automobile demand and saving equations than elsewhere. INFLATION We have examined the role of price inflation in the structureof demandfor durables, in functional relationships designed to explain the surveyvariables, and in the simple income allocation system. The results aresometimes hardto interpret,and are sensitive to the selection of an inflationmeasure. Some results that seem relatively unambiguous, and that areimportant for policy purposes, can, however, be identified. The price inflation variables have a moderate influence in fully devel-oped demand models for automobiles and other durable goods as well asin equations designedto predictthe surveyvariables that play an importantrole in the models. The strongest effects of inflation show up in equationsthat seek to explain average real expendituresper car and real expenditureson nonauto durables. There is some inflation effect in equations designedto explain unit purchases of automobiles, but they are relatively weak. Ingeneral, the most consistent effect in these demand models is a negativeinfluence of expected price change on real expenditures, for both auto-mobiles and nonauto durables. As a result, a fully anticipated inflationappears to have a negative impact on real expendituresfor durables; how-ever, unanticipated inflation has very little impact. Second, unanticipatedprice inflation appearsto have a negative influenceon both purchasing plans and attitudes of consumers, as reflected, respec-tively, in the index of expected purchases, A*, and the index of consumersentiment, S. Even when anticipated, inflation has an influence on expectedpurchases, but has no identifiable effect on consumer sentiment. There issome evidence that the effects of inflation on these variables were not thesame during 1967-71 as they were for the 1960-71 period as a whole, andthat the more rapid inflation during the later part of this period tended todepress the purchase and sentiment indexes relative to predictions made onthe basis of the earlier part of the period. Our third and most tentative conclusion relates to the way inflation,as measured in the income allocation equations of Table 8, affects totalexpenditures and saving. Unanticipated inflation appears to have a nega-

106 BrookingsPaperson EconomicActivity,1:1972tive influenceon consumerexpendituresfor nondurablesand services,while a fully anticipatedinflationhas a strongerpositiveinfluence.Fordurablegoods,unanticipatedinflationhas almostno impact,whilea fullyanticipatedinflationhas a negativeone. For saving,thefirsthas a positiveinfluence,whilethe secondhas a somewhatweakernegativeeffect. In generalterms,the resultssuggestthata fullyanticipatedinflationwillincreaseconsumerexpenditureson nondurablesand servicesand reduceboth spendingon durablesand personalsaving.On the otherhand,a to-tally unanticipatedinflationhas the oppositeeffect:Here,consumerex-pendituresm, ainlyfornondurablesandservices,arereducedsubstantially,whilesavingrisescommensuratelyT. heseresultssupportthe view that aprimaryeffectof unanticipatedinflationis to reducespendingandincreasesaving,possiblybecauseinflationdeepensuncertaintyaboutreal incomeexpectations. APPENDIX A ReducedFormof ObjectiveModel The reducedformof the modelis derivedby writing(1.3), withthe lagoperator,solvingfor Ze, and substitutingin (1.2)' to yield(Al) K*=- I-(1-PZ1p)L,'Substitutingfor K* in (1.1)yields(A2) GP (1 p)L (1 -)KtCombining(A2) with(1.0) and(1.4)yields(A3) Gt -(1-p)L- (1 - 5)Kt) +f(Tt).Multiplyingout by the lag operatoryields(A4) Gt = pZt- (1 - 6)Kt- + p(1 - )(1 -P)Kt +f(Tt) - (1 - p)f(T-1). 1. The functionalnotationis droppedfrom(1.2)for simplicity.Thelag operator,L, isdefinedas Lxt = xt-i. The othervariablesaredefinedon pp. 73-74.

F. ThomasJusterandPaul Wachtel 107Equation(A4) showsthat the modelincludestwo laggedstock variablesbecausethereare two lag processes(adjustmentand expectations)A. lso,a laggedtransitorytermappearsbecausethe transitoryvariablesare notpartof thelaggedformation-of-expectationcosmponentof the model.Thereducedformin the text, equation(1.5),is derivedfrom(A4)by substitu-tion fromthe identity Gt-l = Kt - (1 - )Kt-2A reducedform for the plannedcomponentof expenditurealone can bederivedin the sameway:(AS) GI = fpZ, + (1 - p)GPl + 3(-p + 5)St-I - ((1 - p)Gt-. APPENDIX B Data Estimationand Sources PermanenItncome Permanentincomewasestimatedfromthe followingbasicequation: YPt= Yt-(1 - n)( + y) YPt-1,where Y is currentreal income per household,' r is the coefficientofadjustment,and y the trendrate of growth.The adaptiveformationofexpectationsof realpermanentincome, YP,followsFriedman'sapproachandthe estimationprocedurewas suggestedby Darby.2 The growthrate,-y,is generatedfromthe quadratictrendfunction, In Y = a + bt + Ct2.The quadratictermis usedbecausethe t2termis highlysignificant,and, 1. Currentrealincomeperhouseholdwasderivedbydividingdisposablerealincome,publishedin Surveyof CuirrenBituisiniessb,y an estimateof the numberof households,based on annual data publishedin Bureauof the Census, Statistical Abstractof theUniitedStates. 2. See MiltonFriedman,A Tlheoryof theContsuimptiFonuznzctio(nPrincetonUniversityPressfor the National Bureauof EconomicResearch,1957),and MichaelDarby,\"TheAllocation of TransitoryIncomeAmong Consumers'Assets,\" America,zEcon2omiRc e-view(1972, forthcoming).

108 BrookingsPaperson EconomicActivity,1:1972although it implies an increasing trend, it yields a better fit during thesample period. The growth rate is given by y = .dddint Y = b + 2ct.The initial value for permanent income (when t = 0) is also derived fromthe trend function, YPo = ea. The value of -qis assumed to be 0.19. This figure is derived from Fried-man's estimate of -q= 0.4 from an annual model. Mundlak3 shows thatwhen adjustmentsare truly made quarterlyand the model is estimated fromannual data, the annual adjustmentcoefficient ('qA) and the true quarterlyadjustmentcoefficient (-qQ)are related as follows:(l -nA)= 44-qQ[I - (I - qQ)4]. Two permanent income series, YP and YP', are used in the paper. Theseries for YP is based on disposable income. The trend equation isln Y = 8.6805 + 0.00234 t + 0.0000426 t2. (4.6) (6.5)= 0.965; standarderrorof estimate= 0.023; Durbin-Watsonstatistic = 0.12.The series for YP' is based on disposable income less transfer payments.This variable is used in the objective models of automobile demand, wherenontransfer income is more relevant. The trend equation for calculatingpermanent nontransfer income isln Y = 8.6799 + 0.00234 t + 0.0000425 t2. (4.6) (6.5)2= 0.964; standarderrorof estimate = 0.023; Durbin-Watsonstatistic 0.12. AutomobileandOtherDurableStocks Thestockexistingattheendof thequarteris thesumof depreciatedpastexpenditures in real terms. That is, the stock is given by n Kt= 2(1 -)iXt_, i=Owhere K is the stock; X, quarterlyreal expenditures;n, the life of the good; 3. See Yair Mundlak,\"AggregationOverTime in DistributedLag Models,\"Inter-national Economic Review, Vol. 2 (May 1961), pp. 154-63.

F. ThomasJusterandPaul Wachtel 109and 6, the depreciation rate. It is assumed that the goods depreciate at aconstant rate, 6, for n quarters, at which point they are scrapped. For automobile stocks, the assumptions about depreciation and length oflife used by Hymans are adopted: a = 0.078 and n = 40 quarters.4Theexpenditureseriesis the U.S. Department of Commerce personal consump-tion expenditurecomponent of gross auto product, published in SurveyofCurrentBusiness. For other durables (personal consumption expenditureson durables lessauto expenditures, as above) the assumptions are n = 40 and a = 0.098.The depreciationrate is the average rate over 1954-68 implied by Hymans'procedurefor other durables.5The stock series are deflatedby the numberof households at mid-quarter. Index of Expected Purchases The index of expected purchases of automobiles, A*, is a weighted vari-able defined by A*= 0.6 A** + 0.3 A** + 0.1 A**2,where A** is constituted from data from the Survey Research Center(SRC), the Quarterly Surveyof Intentions(QSI), and ConsumerBuyingEx-pectations (CBE), described below. From 1953 through 1959 the only source of information on buyingintentions is data from the Survey Research Center of the University ofMichigan. Since the data are taken from several sources and are not avail-able in a consistent form nor for every quarter,some processingis necessaryto put them into useful form. This was done for earlieryears by Arthur M.Okun, in \"The Value of Anticipations Data in Forecasting National Prod-uct,\" in The Quality and Economic Significanceof Anticipations Data, AConference of the Universities-National Bureau Committee for EconomicResearch(Princeton University Press for the National Bureauof EconomicResearch, 1960), p. 446; later data are taken from various issues of theSurveyof ConsumerFinances, published annually by the Survey ResearchCenter of the University of Michigan. 4. Saul H. Hymans, \"ConsumerDurable Spending:Explanationand Prediction,\"Brookings Papers on Economic Activity (2:1970), p. 196. 5. Ibid., p. 199.

110 Brookings Papers on Economic Activity, 1:1972 From first quarter1953to first quarter1956Okunprovidesdata fornine periodsof the thirteenquartersin the form of intentions(measuredby the sum of \"willbuy,\" \"willprobablybuy,\" and one-halfthe \"maybuy\"responses)for new and usedcars.The new and usedcar intentionsareassignedweightsof 0.6 and0.3,respectivelyF. rom1956on, secondandfourthquartersurveysareavailablewiththe dataclassifiedby \"willbuy,\"\"willprobablybuy,\"and \"maybuy\"new autos.Weightsof 0.7, 0.5, and0.3, respectivelyw, ereassignedaswellas a 0.3weightforusedcarpurchaseplansanda 0.4 weightfor \"don'tknow\"responses.Thefirstquarterdataareavailablein a new-usedclassificationwith\"don'tknow\"responsesallo-cated.Consistentweightsfor theseclassificationsbasedon the meansizeof eachcategorywerecalculated(0.32forusedcars,0.54fornewcars).Thetwo sectionsof the SRCdatawerethenlinkedon the basis of an overlapperiod. Data for missingquarterswere interpolatedand the seriesseasonallyadjustedwith the X-11 movingseasonalprogram.After adjustment,themissingquarterswerecorrectedto be interpolationsof the seasonallyad-justeddata.The SRCportion(1953-60)of the basicintentionsserieswasthenlinkedto thelevelof theQSI-CBEportionbasedon anoverlapperiod.The derivationof the QSI-CBEportionfollows. For 1960through1966,the CensusBureau'sQuarterlySurveyof Inten-tionsis used; for 1967 on, purchaseprobabilitydata from the bureau'sConsumeBr uyingExpectationasreused.Firstweconstructa weightedmea-sure of the basic QSI intentionsdata: Six-monthdefinite,probable,orpossibleplansto buy new cars are assignedweightsof 0.7, 0.5, and 0.3,respectively;twelve-monthplans are assigneda weightof 0.3; used-carplans a weightof 0.2; and \"don'tknow\"responsesa weightof 0.3. ForCBE data, six- and twelve-monthcar purchaseprobabilitieswere givenequalweightsin constructingthe series.Theconstructedindexusedin theequationsis givenin TableB-1. Indexof ConsumerSentiment The SurveyResearchCenter'squarterlyindexof consumersentimentispublishedin BusinessConditionDs igest.ThevariableS is thesurveylaggedone quarterand with missingquarters(all prior to 1962) interpolatedlinearly. Thefilteredsentimentindex,SZ (TableB-2),reflectsthe notionthatthe

F. ThomasJuster and Paul Wachtel 111TableB-1. Indexof ExpectedPurchasesA, *, 1953-71 QuarterYear First Second Third Fourth1953 ... ... 6.26 5.661954 6.17 6.89 7.56 8.131955 7.79 7.60 7.50 7.471956 7.32 7.42 7.64 7.911957 7.78 7.74 7.45 7.031958 6.79 6.56 6.52 6.531959 7.07 7.37 7.37 7.201960 7.75 7.74 7.60 7.531961 7.63 7.60 7.86 7.971962 8.04 8.30 8.21 8.351963 8.40 8.64 8.84 8.821964 8.97 9.16 9.05 9.411965 9.55 9.58 9.62 9.611966 9.58 9.46 9.47 9.581967 9.28 8.93 9.24 9.121968 9.23 9.37 9.42 9.5719691970 9.55 9.57 9.54 9.441971 9.41 9.36 9.41 9.44 9.74 9.54 ... ... Sources: Derived from data in the following publications: 1953-56-Arthur M. Okun, \"The Value ofAnticipations Data in ForecastingNational Product,\" in The QualityandEconomicSignificanceof Anticipa-tions Data, A Conference of the Universities-NationalBureauCommitteefor Economic Research(PrincetonUniversity Press for the National Bureau of Economic Research, 1960), p. 446; 1957-59-University ofMichigan, Survey Research Center, Surveyof ConsumerFinances, relevantissues; 1960-66-U.S. Bureauofthe Census, Quarterly Survey of Intentionts,relevant issues; 1967-71-Bureau of the Census, ConsumerBuying Indicators, relevantissues.changein S is relevantwhenit is eitherlarge or persistent,hence whenuncertaintyis unmistakablyincreasingor decreasingT. he rule for deter-miningwhethera changein S is systematicis arbitraryb, ut reasonable:It holds that a move is persistentwhenit proceedsin the samedirectionfor at leastthreeconsecutivequarters.Interpolatedquartersarecounted;andthreemorequarterlymovementsarenot necessaryto reintroducetheseriesaftera breakin a seriesof upwardor downwardmovements.Thecriterionis whetherthenextquarterafterthebreakreinforcesthe previouspatternby registeringa newlocal high(or low) value;if so, the serieswillbe interruptedonlyby the quarterbreak,andif not thebasicdecisionruleapplies,and the seriesmustrestart.The ruleis relaxedin the case of twoconsecutivechangesthat are quantitativelylarge (definedto be at leastsevenpercentagepointsontheSRCindex,whichhasa baseof 1963= 100).

112 Brookings Papers on Economic Activity, 1:1972TableB-2. FilteredIndexof ConsumeSr entimentS, Z, 1953-72 QuarterYear First Second Third Fourth1953 ... ... 0.00 0.001954 -1.65 -1.65 0.00 1.001955 2.05 4.10 6.05 3.151956 0.30 0.15 0.00 0.001957 0.00 0.00 -1.85 -4.001958 -4.60 -5.05 -2.60 2.501959 4.95 3.60 2.25 1.101960 0.00 0.00 0.00 -0.701961 0.00 0.501962 -1.40 -0.70 1.40 0.00 1.05 1.951963 0.00 0.001964 0.00 0.00 1.05 1.051965 0.00 1.05 1.40 0.851966 1.05 1.05 -2.15 -4.401967 0.50 0.00 0.00 0.80 -3.70 -1.451968 0.00 0.001969 0.80 0.00 0.00 -2.601970 0.00 0.00 -2.15 -1.351971 -5.95 -4.15 1.701972 0.00 0.00 2.10 0.40 ... ... ...Sources: Derived from index of consumer sentimeiit in Business Conditions Digest, relevant issues.Formally,the filteredvariableiswhere SZt = 0.5 Zt(,AS,) + 0.5 Zt- (ASt-1),and whereZt = ASt= St -St1 1if ASt-ifor i = 0, 1, 2 areof the samesign,or if lASt+ ASti >79or if Zt-2 = 1 and Zt-i = 0 and IASt > IASt-|;otherwise, Zt = 0. Indexof ExpectedPriceChange The expectedrateof inflation,CPI*, is constructedfromSRCdata onexpectedpricechangesw, hichhavebeengatheredsince1952.Howeverp, rior

F. ThomasJuster and Paul Wachtel 113to 1966:2,thesurveyascertainedonlywhetherpriceswereexpectedto rise,fall, or remainthe same.Since1966,thosewho expecteda priceincreasehavebeenasked,\"Howlargea priceincreasedoyouexpect?\"Exceptinthelast fewyears,the surveyswerenot takenregularlyand missingquartersareinterpolated(see TableB-3). The basic surveyseriesfor post-1966yearsreportsthe meanexpectedpriceincreasein the next twelvemonths.The calculationuses the classmid-pointsof the fourclassesexpectingpricesto go up less than 10 per-cent; 10percentfortheclassesexpectinga greaterthanor equalto 10per-centrise;the meanof positiveresponsesforthe \"don'tknowhow much\"class;and0 for the \"willnot go up\" responses.For the earlierperiod,anindexof expectedinflationis constructedby assigningweightsof +1 to\"willgo up\"responses,0 to \"staythesame\"and\"don'tknow\"responses,and -1 to \"willgo down\"responsesT. hisindexis linkedto the post-1966databy anoverlapperiod.Priorto 1960,the surveyquestionrefersto ex-pectedpricechangesfor householdgoods,appliances,andclothingin theTable B-3. Basic Index of Expected Price Change, CPI**, 1953-72 QuarterYear First Second Third Fourth1953 ... ... -0.61 -0.711954 -0.80 -0.90 -0.87 -0.8519551956 0.13 0.42 0.73 1.041957 1.25 1.46 1.65 1.84 1.58 1.32 1.39 1.461958 0.94 0.42 1.01 1.601959 2.04 2.16 2.28 2.411960 2.53 2.39 2.25 2.111961 1.97 1.97 2.00 2.041962 2.27 2.04 1.85 2.011963 2.53 2.30 2.34 2.231964 2.41 2.60 2.62 2.631965 2.64 2.66 2.68 2.601966 2.67 2.74 3.33 3.121967 2.91 3.01 3.12 3.441968 3.38 3.32 3.47 2.991969 3.07 3.73 3.74 2.801970 3.29 3.73 3.06 3.471971 3.50 3.25 2.39 2.411972 3.04 ... ... ... Sources: Derived from University of Michigan, Survey Research Center, Surveyof ConisunmFerinances,various issues, table on price expectations for the next year.

114 BrookingsPaperson EconomicActivity,1:1972next year.The indexof expectedinflationwas constructedwith the sameweightsandlinkedto the priceinflationindexby anotheroverlapperiod. The main apparentdeficiencyin the data is the persistenttendencyofconsumersto expectalmostas muchpriceinflationin theearlyandmiddle1960sas in thelate 1960s.Similarlysurprisingaretheexpectationsof defla-tionin theperiodaftertheKoreanWar.Ontheotherhand,thesedatamayvery well reflectconsumerexpectationsand accuratelyindicatethat thecurrentinflationis largelyunanticipated. The variableused in the econometricwork is the index of expectedinflation,CPI*, an averageof the basic surveyseries, CPI**, over twoquarters,as shownby CPIt = 0.5 CPItt*l + 0.5 CPItt*2.The data are laggedbecausethe SRC surveysare takenat varioustimesduringthequarteranddo not becomeavailableforsometime.Thesurveysareaveragedin orderto smoothrandomvariation.

CommentsandDiscussionSaulHymans:TheJuster-Wachteplaperattemptsto interrelatethe entirerangeof stockadjustmenpt rocesses,sentiment-expectatiodnata,andanal-ysis of inflationeffectsin modelspredictingconsumerexpendituresS. omeof the data employedare notoriouslyimperfect.And althoughthe paperseeksto be as hardon these data as possible,theirqualityremainsonereasonwhy I feel the resultsshouldbe consideredlittle more than sug-gestiveof fruitfulavenuesto pursuein furtherinvestigationsI. also havesome technicalquestionsto raiseaboutthe analysisandthe specificationof some of the empiricalmodels. There are at least three troublesometechnicalpoints. First, Z is thefunctioncontainingthe determinantsof desiredstock. The adaptiveex-pectationsmodel allegesthat changesin expectationsabout Z are pro-portionalto the differencebetweenthecurrentvalueof Z andthepreviousexpectationof Z. However,if the currentvalueof Z is available,an expec-tationvaluewouldbe unnecessaryI.t is the laggedvalueof Z thatshouldappearin the adaptiveexpectationsequation,in the subsequentreducedform equationthat is fitted to the data, and in all the other equationsinvolvingthe objectivemodel. Second,thecombinationof adaptiveexpectationsandpartialadjustmentcan playrealtrickswiththe structurael rrortermin the model.This pos-sibilityis neglectedby the authorsin makingregressionestimatesand, Ibelieve,worksto the relativedisadvantageof the objectivemodelwhenitis comparedwiththe anticipatorymodel.The objectivemodelis subjectedto both of these mechanisms-partialadjustmentand adaptiveexpecta-tions.The anticipatorymodelis subjectedto neither. Third,the authorsattemptto predictthe index of expectedauto pur-chases,the A**variable,andthe valueof thosepurchasest,he A**V vari-able, whichresultedin the threeequationsin Table6. In this modelthe 115

116 Brookings Papers on Economic Activity, 1:1972laggedautostockandlaggedexpenditureos n autosshouldbe includedasexplanatoryvariables.Butwhentheyare,in thefirstequationof thetable,theyarenot statisticallysignificantandlaggedexpenditurehs asthe wrongsign.Thevariablesareapparentlytoo highlycorrelatedto appearasregres-sors in the sameequation.It mightbe possibleto constructa statisticaltechniquethatwoulduseana prioritheoreticaslpecificationaboutthevari-ables to get around.this problem.But when the problemis avoidedbyusingequationssuchas the secondandthirdequationsin Table6, whichomitthelaggedexpenditurevsariable,thestructuraml eaningof theresultsis unclear,includingthe resultsin the inflationvariables. Turningto empiricapl roblemsT, able2 comparestheanticipatorymodeloverdifferentimeperiods.Priorto 1960,the expectedpurchasevariable,A*, is simplyconstructedfrombad data.Thatis probablyreasonenoughto explainthe 1954-59equation.But whataboutthe 1967-71equations,whenthe coefficientof A* is againneithersignificantlynor evenprobablymeasurablydifferentfrom zero?Apparentlythat variableis useful onlyfor the 1960-66period.The authorsconcludedthatthe residualvariationin the anticipatorymodelis not muchhigherin 1967-71thanin 1960-66,butin factthe standarderrorof estimateis 44 percenthigherin the formerperiodthan in the latter.The samecomparisonwith the objectivemodelshowsonly a 21 percentincreasein the standarderrorof estimatedespitethe factthatthe Chowtest seemsto like one and not the other. It is apparentthat, whatevertheirtype, all of the modelsgo bad in thelate 1960s.Thatfact seemsto me of fargreatersignificancethanany nit-pickingoverthetwo bits'worthof explanatorypowerby whichthe objec-tive and the anticipatorymodelsdiffer. The materialon the incomeallocationsystemis potentiallyinterestingbutsubjectto too manyproblemsforme to be impressedwiththe results.Thespecificationfsortheexpendituresystemaretoo simplisticto be in thesame paperwith the equationsdiscussedearlier.As the Durbin-Watsonstatisticsdemonstratet,he timingpatternsare so inadequateas to vitiatethe results.I suspectthe authorswouldhavebeenbetteroff puttingall thelaggeddependentvariablesin each equation,or puttinglagged durablestocksin eachequation,or, as I thinkbest,fittingthe simpleexpendituresystemto annualdata, wherethe timingrelationshipsarefarless crucial.This methodwould still let them deal with the issue that intereststhemmost-the way inflationaffectsthe allocationof incomeamongvarioussavingandspendingcategories.In short,I wouldhavebeenhappierwith

F. ThomasJuster and Paul Wachtel 117a far morecarefullyconsideredjob on a propersubsetof the topicsthatthe authorstook up.JamesDuesenberryI: thought the Juster-Wachtepl aperwas very inter-estingand suggestiveof a numberof possibilitiesfor furtherresearch.Inplacesit movesus aheada significantdistance.ButI also had someprob-lemswithit. Firstof all, I wassurprisedat theinitialdiscussionof thebasicmodel becausethey used a standardstock adjustmentmodel that dealswith automobilesbut not withany substantivetheoryof the behaviorofthe public. I would have thoughtsomeoneas interestedas Justeris insurveysof consumerattitudeswouldlook morecloselyat thespecificdeci-sion makingthat is involved.We are dealingwith a situationin whichpeoplearefacedwitha set of long-runchoices:how oftentheytradetheircars; whetherthey buy secondcars;and whatkind of car they buy andwithwhatdegreeof luxury.Presumablym, uchof thebasicpatternof theirchoicesis accountedfor by the slow responseto permanentincome,torelativeprices,and,possibly,to therelativerepaircostsof usedcars,whichoughtto be somewherein thissytem.Then,in addition,therearetransientfactors,suchas changesin actualincomeor expectationst,hatleadpeopleto postponepurchasesor to decideto tradedown or up and thus causevariationsaroundthe basicpatternof purchases.More attentionshouldhave been paid to the bolts and nuts of the decision-makingprocessinspecifyingthe basicmodel.The authorswentin this directionwhentheyseparatedtheproblemof predictingautoexpenditureisntohowmanycarswouldbe boughtandat whataverageprice. An integratedargumentarisesfrom lookingat the paperas a whole.The evidenceseemsto show that thereis somethingdifferentabout thelastfiveyearsin comparisonwiththeearlieryears.The paperargues,first,that some of that differenceappearsto be pickedup by the surveyvari-ables,althoughthatconclusionis calledinto questionby the observationsthatHymansmade,andremainsproblematicA. nd it argues,second,thatinflationmayaccountfor the differenceaboutthe last fiveyears. Both of theseargumentsb, ut especiallythe firstone, are weakenedbythe way the authorscompromisedin usingthe manyvariablesfrom theobjectivemodeltogetherwiththe surveyvariables.Theyendedup withaweak specificationof the objectivemodel, particularlyin its transitoryelements.If an objectivemodel were built from scratchto includethesurveyvariables,it wouldbe morecomplicatedthanthis one. In dealing

118 BrookingsPaperson EconomicActivity,1:1972with transitoryfactorsthe unemploymenrt ate is not enoughto link thebasicmodelto the surveyvariables.Moredetailaboutthecarstockmighthavebeenusefulin copingwiththisproblem.Becausethecombinedmodelis inadequatei,t weakenstheconclusionthatsomestructuraclhangein thelateryearsis pickedup by the surveyvariablesand not by the objectiveones. I thoughtthe inflationanalysismadesomerealprogress;separatingtheinflationintoanticipatedandunanticipatedelementsmakesthewholeargu-mentmoresensible.On the wholethereis somepositiveevidencein favorof thepropositionthatinflationwasa significantvariablein the latterpartof the periodand not earlier.But I wasnot overwhelmedby the evidencethatthe surveyvariablespickedthatup. Table3, in whichthe anticipatoryvariablesare addedto the objectivevariables,poses anotherproblem.Unfortunatelya, ddingthe anticipatoryvariablescausesbig changes,in both the size of the othercoefficientsandtheirsignificancea, t least in the 1960-71and 1960-67periods.Giventheweaknessesin the specificationof the objectivemodel, the importanceofthe explanatorypowerof the anticipatoryvariablesis questionable.Par-ticularlyin this situation,addinga new variablethat knocksanotheroneout does not mean the new one is the rightone. It tells you only that,becauseof collinearityproblems,the resultis ambiguous. The paperdoes move us aheadon a complicatedsubject.But a gooddeal of ambiguityremains.To reduceit callsfor workingthe surveydatainto a model that makesuse of more detail about consumerbuying-whetherpeopleare buyingnew or old cars,tradingup or tradingdown,and that sort of thing-as a way of providingmorecertaintyaboutwhatmotivatespeopleand whatthe surveyscan tell us.Tom Justerand Paul Wachtel:RegardingHymans'first point, in a lifecycleapproachto a demandmodel,expectationsarerelevantwhetherornot currentinformationis available.It mightbe arguedthatcurrentinfor-mationthat may not be availableshouldnot be used to predict,if theobjectivemodelis viewedas a predictiveone. In thatcase,the onlychangenecessaryis to lag the objectivevariables,incomeandprices,one quarter.We have looked at equationsof that sort and nothingis fundamentallychangedfromthose reportedin our paper.In general,expectationsareaveryimportantpartof the model.Theyarerelevantevenwhencurrentin-formationis available,andthatishowwejustifyourformulation.

F. ThomasJusterandPaul Wachtel 119 We triedto give the objectivemodela fair shakein comparingit withthe anticipatorymodel, althoughboth Hymansand Duesenberryfelt weshouldhave done betterby it. In any sort of objectivemodel with a lagstructure,a completespecificationwould be complicatedand we wereforcedto somesimplificationsS. canningpossibleserialcorrelationparam-eters for first,second,or higherorderserialcorrelationswould improvethe fit and givethe objectivemodela betterchancein thesecomparisons.But withoutan a priorireasonto expecta particularform of the errorterm,the improvedfit may not genuinelyreflectan improvedmodel. WeentirelyagreewithDuesenberry'nsotionthatone wantsto get moremicroaboutthe determinantos f auto purchases.Manyof his suggestionswould apply to the design of an equationto explainA**, the index ofexpectedpurchasesT. hatvariableis supposedto representa netjudgmenton the partof consumersas to whattheywill do in the nextthree,six, ortwelvemonths,whichthey presumablymake on the basis of their pastincome,expectedincome,car stockrelativeto income,and developmentsin usedandnewcarprices.All of thoseingredientsgo intothe determina-tion of the indexof expectedpurchases.Thereforeit is not surprisingtofindthat,whenthatvariableis introducedintoanequationincludingother,objective,variables,the othersbecomeredundant.That is exactlywhathappenedin the 1960-71periodwhenan accuratemeasurewas availablefor the indexof expectedpurchases. GeneralDiscussion Severaldiscussantscommentedon how the surveyvariablesshouldbeinterpretedand whatthe most usefulway to studythem wouldbe. SaulHymansarguedfor sharplydifferentiatinbgetweenthe buyingintentionsandconsumersentimentvariables.He notedthatwhenthe sentimentandbuyingintentionsvariableswereused togetherwith objectivevariables,asthey havebeenin this and otherresearch,the sentimentvariabledid notdisplaceobjectivevariableswhile buying intentionsdid. He cautionedagainsta casualinferencethat both types of surveysreflect\"attitudes.\"JamesDuesenberrypointedout thatif buyingintentionsweretreatedlikethe data on plantand equipmentspendingplans,it wouldbe naturaltouse one set of objectivevariablesto explainbuyingintentionsandanotherset of variables,includingthe consumersentimentdata, to explainthe

120 Brookings Papers on Economic Activity, 1:1972deviationof theintentionsfromactualconsumerexpendituresP. artof thesentimentvariablepresumablywouldbe explainedby objectivevariables,andit wouldbe theremainingu, nexplainedp, artof the sentimentvariablethatwouldbe usefulnewinformationaboutconsumerbehavior. LawrenceKleinalso arguedthe importanceof goingas far as possiblewithobjectivevariablesratherthanseeinghow fartheycouldbe replacedby surveydata.He preferredusinga carefullystructuredmodelof objec-tive variablesto explainconsumerbehaviorand supplementingthis withsome functionof the sentimentvariablethat representedthe partof con-sumerbehaviorthat could not be explainedby objectivevariables.Hereportedbeingableto predictthesentimentvariablefairlywellusingpricechangesand unemploymentrates.In a similarvein, R. J. Gordon saidthat the realneedis for consumptionfunctionsin an econometricmodelthat wouldpredictthe resultof alternativemonetaryand fiscalpolicies.Thisrequiredmoreresearchon explainingthe determinantosf the surveyvariables.Startingwithobjectivefunctionslikethe onesin this paper,andwith improvementsto them fromsurveydata, good explanationsof thesurveydataarethe basicrequirementT. heultimateconsumptionfunctionwouldthenincludeall the determinantosf the surveyvariablesas well asthe objectivevariables. WhileArthurOkunagreedthat one importantuse of surveydata wasto improveourunderstandinogf consumerresponseto objectivevariables,he notedthat surveyinformationon inflationaryexpectationsitself pro-videdobjectivemeasuresof that importantdeterminantof consumerbe-havior.Thepresentpaperwasthe firststudyhe hadseenthatusedsurveyresponseson expectedinflationto helpexplainconsumptionH. e andAlanGreenspanboth notedthat consumerscould be expectedto cut back onspendingin the face of higherinflationrates,whetheranticipatedor not,becausemoreinflationwouldbe associatedwithgreateruncertaintyaboutindividualrealincomes,evenif realincomeswereunchangedon average.The resultsin the paperwereconsistentwiththisview. RobertSolowfoundit hardto takeseriouslyresultsbasedon the surveyof inflationaryexpectationsbecausethesurveydataseemedso implausiblea priori.He notedthatindividualresponsesto the surveywerewildlyoutof line with actual price experienceand suspectedrespondentsdid notunderstandthe questionsthey were being asked and would not in factbehavein a waythatwasconsistentwiththeiranswersW. illiamNordhaus,GeorgePerry,and otherswereinclinedto reservejudgmenton this, since

# Inflation and the Consumer - Brookings Institution

##
**Description: ** F. THOMAS JUSTER* National Bureau of Economic Research PAUL WACHTEL* City University of New York and National Bureau of Economic Research Inflation and the

### Read the Text Version

No Text Content!