sign in sign up
<<
Home , Market (economics), May 1975

BARRY BOSWORTH Brookings

The and the

THE STOCKMARKET decline of 1973-74 marked the longest and steepest fall in corporate-stockprices since the depressionof the 1930s.The loss of stockholderwealth in marketprices amounted to $525 billion, or 43 per- cent.'The magnitudeof this declinein stockvalues, in conjunctionwith the subsequentcollapse of aggregatedemand in 1974-75, has sparkeda re- newed discussionof the role of the stock marketin businesscycles. The debate-as is so frequentlythe case-is not new to .Several sig- nificantcontributions recently made at both the conceptualand empirical levels seem, however,to justify a reexaminationof the issues. The disputeabout the import of changesin the stock marketrevolves around their causal role in economicfluctuation: Are they a source of variationin aggregatedemand? Does the causationrun solely in the op- posite direction?Or do the levels of economicactivity and of stock simplyrespond similarly to other,more basic, economic forces, with no di- rect causal link betweenthe two? This third interpretationis consistent with a view that the stock marketreflects investors' attempts to forecast economictrends. The fact that movementsin stock prices foretellmajor Note: I am gratefulto LeonardHerk for researchaid in writingthis article.Members of the Brookingspanel offeredvaluable comments and suggestionsin the preparationof the draft. David A. Wyss of the Federal Reserve Board staff provided the computer simulationsof the MPS model and answerednumerous questions. 1. Derived as the change between and , as shown in Board of Governorsof the FederalReserve , unpublisheddetail accounts, from the flow of funds (). 257 258 BrookingsPapers on EconomicActivity, 2:1975 cyclesin businessactivity is, thus, only evidencethat investors'forecasts are betterthan randomguesses. As figure 1 demonstrates,the stock marketand economicactivity do move in similarcyclical patterns.Because several previous studies have foundthat changesin the stockmarket tend to precedechanges in business conditionsby an averageof about four months,2the stock-priceindex is a majorcomponent of the indexof leadingindicators.3 The cyclicalpattern is particularlyevident in the latest .Stock pricespeaked at the be- ginningof 1973,then declinedsharply for two years.Industrial , however,continued to rise throughoutmost of 1973, and its collapsedid not begin beforelate 1974. On the otherhand, as othersare quickto point out, this behaviordoes not implya causalrelation between the stockmarket and real economic ac- tivity. Furthermore,focusing only on major cycles in economic activity means ignoringthe numerousperiods when the changed sharplyand nothinghappened to the economy.Saul Hymans, for example, reportsa negativeand insignificantcorrelation between lagged stock prices and an of currenteconomic activity for the period 1948-72 in an equationthat includesother cyclicalvariables.4 Thus, it remainsunclear whetherthe collapseof the stock markethad a directrole in the 1974-75 recession,or simply mirroreddepressive developments elsewhere. In the sectionsthat follow, an effortwill be made to assessthe role of changesin stockprices in two majorcomponents of aggregatedemand: personal con- sumptionand businessfixed .

ConsumerDemand

In the vast literatureon consumerdemand, only a few studieshave fo- cuseddirectly on the influenceof changesin stock pricesand of theirasso- ciatedcapital gains and losses. In addition,those studiesthat haveexplored

2. See, for example, Geoffrey H. Moore and Julius Slhiskin,Inidicators of Business Expansionsand Contractions,Occasional Paper 103 (ColumbiaUniversity Press for the National Bureauof Economic Research, 1967), pp. 44-45. 3. For a descriptionof the currentversion of this index, see Victor Zarnowitzand CharlotteBoschan, "Cyclical Indicators:An Evaluationand New Leading Indexes," BusinessConiditionis Digest (May 1975), pp. v-xix. 4. Saul H. Hymans, "On the Use of Leading Indicatorsto PredictCyclical Turning Points,"BPEA, 2:1973, p. 346. 0~~~~~~~~~~~~~~~~~ 00 ~

C- 0 C's

O..4~~~~~~~~~~. tn ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ 0

aN~ ~ ~~ ~ ~~~~~~~~~~~~~~~~~~~O

0 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ 0

*0

0-~~~~~~.~

- .*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~. **~~~~~~~~~~~~~~~~~~~~~~~~1

0 ~~~~~~~~~~~~~~~~~:

0Aw

0z

gi .P.4~~~~~~~~~~~~~~~~~~~~~~~~~~~O ) 4k~~~~~~~~~~~~~~~~~~~~~~~~~~~~0f 260 BrookingsPapers on EconomicActivity, 2:1975 this relationshipdiffer in their emphasison two distinctmechanisms by whichthe stock marketmight affect demand. The first,which is closelyidentified with the life-cyclemodel of consumerbehavior, empha- sizesthe effectof changesin wealth.The seconduses the stock marketas a proxymeasure of optimismor pessimism(so-called "consumer sentiment"). While in many applicationsthe distinctionbetween these two mecha- nismsis not important,it is centralto an evaluationof the impacton con- sumptionof changesin monetarypolicy. The demandfor corporatestock is commonlyviewed as dependentupon (1) expectationsof corporateearn- ings, and (2) the rate of returnon alternativeassets. can, therefore,affect the priceof corporatestock through altering rates on com- petingassets, without changing expectations of profitsor economicactivity in anyregular, predictable fashion. Thus, if the correlationbetween changes in stock pricesand consumptionfound in previousstudies reflects the role of the stock marketas a crudeproxy for consumersentiment, monetary policywould have no directcausal impact on .If, on the other hand, the observedcorrelation represents a wealtheffect, an explanation for the behaviorof stock prices that includesa link to monetarypolicy becomescrucial.5 Any estimateof stock-marketeffects must rest on an evaluationof the importanceof wealthin consumerbehavior. If one acceptsthe view that consumptionreflects an attemptto maximizeintertemporal subject to the constraintof total availableresources, current earned income is a poor measureof that constraint.Individuals can both save for and borrow againstthe future;and they receiveincome claims from sources other than currentproduction-for example, inheritances,gifts, and capital gains. Wealth,however, is highlycollinear with permanent or normalincome; and motivesrelated to estatebuilding complicate the interpretationof the effect of wealthaccumulation on consumption.Moreover, a conceptualcompli- cationarises because the definitionof incomeused in the nationalincome accounts(NIA) alreadyreflects all earningson wealth-exclusiveof capital gains. Capitalgains on corporatestock raise other,specific, issues. Corporate- stockholdings are highlyconcentrated among the rich,and one mightwell

5. Changesin stock pricesthat result from changesin other rates may or may not be associatedwith a change in expectationsor sentiment.The significantpoint is that there may be no stable relationship,so that differentassumptions can lead to sharplydifferent conclusionsabout the impact of variationsin monetarypolicy. BarryBosworth 261 be dubiousthat unrealizedcapital gains and losses have a constrainingin- fluenceon the consumptiondecisions of such individuals.For example,in 1971,51 percentof all stock-marketassets were held by those in the top 1 percentof the incomedistribution, and 74 percentby the top 10 percent.6 In addition,capital gains and losses on suchassets are highly transitory and only a small proportionare actually realized. Data developedby Kul Bhatiaimply that realizedgains accountfor only a fifthof accruedgains.7 Moreover,the two componentsof capitalgains are not closelycorrelated. Realizedcapital gains are not includedin the NIA definitionof income althoughthey do incuran increasein and thus a declinein measured disposableincome.

A WEALTH VIEW OF COMMON It is convenientto use the consumptionequation of the MIT-Pennsyl- vania-SocialScience Research Council (MPS) model of the as a vehiclefor examiningsome of the issues.The equation fitted to 1954-70 relatesthe quarterlyflow of real per capita consumptionservices (CON) dividedby population(N) to currentand previousvalues of disposablein- come (Y), householdnet worthexclusive of corporatestock (W - S), and an averageof recentvalues of corporate-stockholdings (S):

_ (1) CON bi-) +0054 N )

+ ci (N) +0.74ut-1, where

bi = 0.66 ; Ec = 0.054. (12.7) Here and in subsequentequations the numbersin parenthesesare t-ratios and u is an errorterm. The sum of the coefficientson corporatestock is constrainedto equalthe coefficienton wealth,so that separatemeasures of

6. Marshall E. Blume, Jean Crockett, and Irwin Friend, "Stockownershipin the United States: Characteristicsand Trends,"Survey of CurrentBusiness, vol. 54 (Novem- ber 1974), p. 17. 7. Kul B. Bhatia, "Capital Gains and the Aggregate Consumption Function," AmericanEconomic Review, vol. 62 (December 1972), p. 869. 262 Brookings Papers on Economic Activity, 2:1975 significanceare not available;but the overall term is highly significant.8 Since the originalformulation indicated autocorrelation of the residuals, the equationis estimatedwith an autoregressivetransformation of the data. The serial correlationparameter of 0.74-the coefficienton the previous period'serror (u,)-implies that the problemis severe. This is an appealingequation as it stands.The long-runcoefficients on income and wealth are in close agreementwith earlier estimates by FrancoModigliani, Albert Ando, and RichardBrumberg. Also, if wealth is a significantdeterminant of consumption,one can argueplausibly that wealthaccumulated through capital gains on corporatestock has long-run effectsno differentfrom those of wealthin otherforms. The lag on stock- marketassets can thus be interpretedas a smoothingor discountingof transitoryvariations in stockvalues with an imposedconstraint of long-run equalitybetween the effectsof corporatestock and of othercomponents of wealth.The equationdoes implya majorimpact on consumptionof varia- tions in stock prices.For example,the 1972-73 declinein stock values, if permanent,would have depressedconsumer demand directly by $30 billion, or a 3 percentincrease in the savingrate. The majorissues posed by the equationare whether wealth belongs as an argumentin the consumptionfunction, and, if so, how long it takesbefore capitalgains and losses are fully incorporatedinto individuals'evaluation of theirwealth positions. More specifically,why shouldthe lag on capital gains and losses in the stock marketbe shorterthan that for income in general(eight versus twelve quarters)? If currentincome is a poor measure of resourcesavailable for consumptionand lifetime resources form the rele- vant measure,transitory variations in assetvalues should have minimal in- fluenceson consumption;but this is not the implicationof the MPS ver- sion. Sincethe primaryimpact of transitorychanges in incomeand wealth shouldbe reflectedin ,an extraeffect of gainsand losses in the stock 8. The MPS concept of consumptiontreats the purchase of durablegoods, CD, as investmentand such outlays enter into the estimateof consumptiononly when they are "consumed."In the national income accounts, durablesare included in consumerex- pendituresat the time of purchase.The empiricalrelationship between the two concepts is given by CON = CNIA - 0.935CD + 0.26KDt-1, where KD is the stock of durables.In essence, this formulationreplaces purchasesof durablegoods with a smoothed average of past purchases.Thus, in terms of cyclical variance,the series is most nearly comparableto previous studies of consumption of nondurablesplus services. Barry Bosworth 263 marketmight show up in the durables-purchasesequation (defined as sav- ing) of the model; but it does not. The implicationfor consumptionspending of variationsin stock-market priceshas been examinedin studiesby Arena and Bhatia.9However, be- causeboth of theseauthors included the marketvalue of corporatestock in the definitionof initialwealth, they tested only the contemporaneouseffect. In the equationsestimated by Arena,the wealthcoefficients ranged from 0.03 to 0.09 and were frequentlysignificant, but the contemporaneous capital-gainscoefficients were less than half as largeand neversignificant. One can rejectneither the hypothesisthat capitalgains have no contem- poraneouseffect nor the alternativeextreme that the capital-gainscoeffi- cient was equal to that of the wealthcomponents. Equations specified by Bhatia, which used a permanent-incomeconcept in place of wealth, re- sultedin marginalsignificance for a measureof realizedexpected gains, but insignificantresults for accruedcapital gains. A second form of equation constrainedthe sum of the coefficientson currentand past capitalgains to equal that of total wealthwith lags extendingover five years. Again, be- causethe definitionof beginning-of-periodwealth included corporate stock at marketprices, the equationimplies only a hypothesisabout expected contemporaneouscapital gains ratherthan providinga test of the con- sumptionresponse to short-termfluctuations in pricesof capitalassets.

VIEWS EXCLUDING WEALTH AND CAPITAL GAINS Despite its popularity,the wealthversion of the consumptionfunction has not been closelycompared with the specificationsof otherresearchers. Certainly,all wouldagree that a simplemoving average of incomeis an in- adequatepredictor of cyclicalchanges in consumerdemand. The alterna- tive approach,which excludesany specific wealth variable,emphasizes some disaggregation;a purchaserather than a consumptionconcept of durablegoods; and the inclusionof specialvariables in the individualequa- tionssuch as laggedstocks, measures of anticipatedand unanticipated infla- tion, an index of consumersentiment, complementary demands-particu- larly in the area of housing-and the unemploymentrate. If it uses stock pricesat all, it views them as a proxy for consumersentiment. 9. See John J. Arena, "PostwarStock Market Changes and ConsumerSpending," Review of Econonmicsand Statistics, vol. 47 (November 1965), pp. 379-91; Arena, "Capi- tal Gains and the 'Life Cycle'Hypothesis of Saving,"American Economic Review, vol. 54 (March 1964), pp. 107-11; and Bhatia, "CapitalGains." 264 Brookings Papers on Economic Activity, 2:1975 PreviousBPEA papers on consumptionall rely on empiricalequations thatdo not providea role for wealth.'0Hymans obtained a smalland insig- nificanteffect on automobiledemand in his firstpaper and concludedthat "thenet worthvariable is clearlyno panacea.It has shiftedthe datingof the errors,but theypersist.... If a wealthvariable ultimately proves itself, it is most likely to be in the form of a marginaladdition. ..." The absenceof a wealthvariable in thesestudies stands in sharpcontrast to its significancein the MPS equation;so far, neitherapproach has dem- onstratedits superioritysatisfactorily. The comparisonis complicated by severestatistical problems introduced by the presenceof strongserial correlationin all of the time-seriesdata. Also, the wealthversion uses a measureof consumptionthat includesan estimateof the rentalor of the stock of consumerdurables rather than the directexpenditure on durablesused in the nationalincome accounts and most otherstudies. The followingsections attempt to explainthe sourceof this disagreement overthe role of wealthand examinethe evidenceon the role of short-term capitalgains and losses.

AN EMPIRICAL COMPARISON OF ALTERNATIVES Becausethe alternativeapproach, which omits wealth,is disaggregated by componentsof consumption,the MPS equationwas reestimatedunder a varietyof specificationsin orderto obtaincomparable equations. These are shown in table 1 for the period 1954-72.12 First, the removal of the con- straintof long-runequality between the effectsof corporatestock and other

10. Saul H. Hymans, "Consumption:New Data and Old Puzzles," BPEA, 1:1970, pp. 117-26, and "ConsumerDurable Spending: Explanationand Prediction,"BPEA, 2:1970, pp. 173-99; F. ThomasJuster and Paul Wachtel," and the Consumer," BPEA, 1:1972, pp. 71-114; Arthur M. Okun, "The PersonalTax Surchargeand Con- sumerDemand, 1968-70," BPEA, 1:1971, pp. 167-204; and Lester D. Taylor, "Saving out of DifferentTypes of Income,"BPEA, 2:1971, pp. 383-407. 11. Hymans, "Consumption,"p. 126. 12. The data are primarilyfrom the national income accounts, published in the Surveyof CurrentBusiness, or the data of the MPS model. The serviceconcept of consumption, worth, income, and the corporate-stockvariable are from the model equationbook. The definitionof disposableincome is slightlydifferent from that of the nationalincome accounts because it excludesconsumer payments and includes an imputationfor income from the stock of durables,and because the timing of liabilitiesis computeddifferently. The measureof corporatestock is createdby dividing dividendpayments of the national income accounts by Standardand Poor's dividend- priceratio. All the variablesare measuredin 1958 prices. Barry Bosworth 265 Table1. WealthFormulations of Consumption-DemandEquations, 1954-72a Annual rates in billions of 1958 dollars Variable Nondurables Consumer Consumer and Total plus durable durable summary consumption services servicesb purchasesb statistic (1) (2) (3) (4) Variable Constant 2.511 29.4 -26.47 -9.0 (0.6) (6.5) (9.8) (0. 6) Nonstock wealth 0.069 0.061 0.002 -0.037 (3.8) (3.2) (0.7) (1.6) Corporatestocko 0.046 0.059 0.007 -0.008 (4.7) (6.0) (2.4) (0.6) Incomed 0.612 0.418 0.201 0.204 (10.4) (7.0) (16.8) (1.9) Lagged stock of durables ...... 0.201 (1.6) Summarystatistic R2 0.999 0.999 0.999 0.993 Standarderror 1.34 1.35 0.23 1.87 Rhoe 0.73 0.7 0.99 0.5 Dynamic standarderror 22.5 16.7 44.0 4.5 Sources: See text note 12, where detailed information is also given. a. The numbers in parenthesesare t-ratios. b. The services concept of consumer durables is that of the MPS model. Purchases are as reported in the national income accounts. c. The coefficient is a sum of an eight-quarterlag series. d. The coefficient is a sum of a twelve-quarterlag series. e. Autocorrelation coefficient, estimated by an interactive search routine to minimize the standard error of estimate.

wealth(column 1) altersthe earlierresults only slightly."3The long-runin- come effect is reduced,that of wealthis raised,and the corporate-stock coefficientis about67 percentof that on wealth.More important, all of the coefficientsare highlysignificant in a statisticalsense. The parameteresti- matesare also verystable when the equationis refittedfor a varietyof sub- periods.The evidencesuggests, however, that the assumptionthat the serial correlationof the errorterm follows a first-orderscheme is not realistic.14

13. These equationsalso differfrom (1) above by excludingthe deflationfor popula- tion growth.This exclusion has very little impact on the estimatesand facilitatesinter- pretationof the standarderrors. 14. This assumptioncan be tested by generatinga predictedseries for the dependent variable that uses lagged estimated (instead of actual) values for the autoregressive transformation.A sharp deteriorationin the R2 and standarderror of estimate would imply a more complicated scheme for the error term. For the above equation, the standarderror of the dynamic predictionsincreases by a factor of 17. 266 Brookings Papers on Economic Activity, 2:1975 Second,dividing total consumptioninto (1) nondurablesplus services and(2) durablesreveals that the effectsof the capital-gainsand wealth terms are heavilyconcentrated in the consumptionof the former.Evidently, the creationof a rentalor service-flowconcept of durablesfrom the outstanding stock has produceda data series that depictslittle more than a secular trend.Also, a few problemsbecome apparent. For nondurablesalone, the lag coefficientson incomebecome negative after a fewperiods, an effect that persistseven if the overalllag is shortened.Without the correctionfor auto- correlation,corporate stock has a highlysignificant negative impact in the durable-servicesequation. In both cases, the autocorrelationis very high and has a pronouncedimpact on the point estimatesof the parameters. For purposes of comparison,equations that excludedany wealth or stock-marketvariable were estimated for severalcomponents of consump- tion. The specificationof theseequations draws heavily upon previouspub- lishedpapers and the majoreconometric models."5 Although these models usually have a greaterdegree of disaggregation,the results that follow apply to the consumptionof nondurablesplus servicesand purchasesof durablegoods. Furtherdisaggregation did not alterthe conclusionsabout the role of wealthand the stock market. Nondurablesplus services.For the nondurablesand servicescomponent, alternativeformulations have relied heavily upon currentand lagged values of income. The major additionalvariables have been the housing stock (measuringthe complementarity of utilitiesservices and the stockof homes), population,and relativeprices. In orderto parallelthe durablesequation presentedbelow, the unemploymentrate, the indexof consumersentiment, and pricechanges were tried in the initialformulation but provedinsignifi- cant.A three-periodaverage of the changein incomewas includedto reflect the factthat somenondurables display the stock-adjustmentattributes nor- mallyassigned to durables."6The finalequation for the 1954-72period had the followingform: 2 (2) CNS =-35.14+ 0.33 KHt_1+ 0.11 Z YDt-i (5.2) (11.5) (2.8) i=O 11 + 0.44 E wi YDt- + 0.61ut-1, (22.3) i=O R = 0.999; standarderror = 1.28. 15. See, for example,Hymans, "ConsumerDurable Spending." 16. The equationformulation can also be justifiedby referenceto the dynamicmodel in H. S. Houthakkerand Lester D. Taylor, ConsumerDemand in the UnitedStates: Analysesand Projections (2d ed., HarvardUniversity Press, 1970). Barry Bosworth 267 where

CNS = consumptionof nondurablesplus services,1958 dollars KH = stock of housing, 1958 dollars YD = disposable income, 1958 dollars. wi= weights on income estimatedfrom second-degreepolynomial with sum of unity.

This versionof the equationfits the data as well as the equationwith a wealthvariable does, and there has been some reductionin the extent of autocorrelationin the residuals.However, the coefficienton the housing stock is about twice the value that one would expect from comparing housing-relatedservices to the value of the stock. Since this variableis highlycorrelated with population,permanent income, and total wealth,it is likelyto reflectfactors other than the link betweena home and the con- sumptionservices that are associatedwith it. As a test of a wealtheffect, a secondequation was estimatedby adding the corporate-stockand wealthvariables to equation(2). A two-yearlag structurewas includedfor the corporate-stockterm, as in the MPS equa- tion. Both the wealthand corporate-stockterms were highly significant in a statisticalsense:

(3) CNS -5.74 + 0.177 KHt- + 0.11 I2 A YDt- (0.6) (4.1) (3.1) i-O

11 + 0.385 E wi YDt_i + 0.041(W -S)t_ (8.8) i=O (2.5)

8 + 0.030 E vi St-, + 0.49 uti1. (3.3) i'

RI = 0.999; standarderror = 1.17.

In addition,the coefficienton the housingstock has a far more plausible valueand othercoefficients in the equationhave not changedsignificantly. The steady declinein the weightsfor laggedvalues of corporatestock is consistentwith a notion of smoothingor discountingof transitoryfluctua- tions in assetprices. When the equationis refittedto varioussubperiods, it is evidentthat all of the significanceof the corporate-stockvariable arises 268 Brookings Papers on Economic Activity, 2:1975 after1965. For periodsending prior to 1966,the corporate-stockcoefficient is small and clearlyinsignificant. Conswnerdurables. The need for a distinctionbetween the purchaseof durablesand theirconsumption has long been recognized;in fact, it is in- dispensableto stock-adjustmentformulations of investmentin durable . However,the MPS modeltransforms the stock datato a flow basis priorto estimation,whereas most otherstudies have embeddedthe stock- flow distinction within a model of investmentdemand. Regardlessof whetherpurchases of durablesare definedas consumption,as in the na- tional incomeaccounts, or as saving,as in the MPS model, it is important to determinewhether they are influenced by the stockmarket. The equation reportedin column(3) of table 1 relatesto the serviceflow from durable goods and thus does not directlyanswer the questionof whetherfluctua- tions in wealthand stock pricesaffect purchases of such goods. The model does includean equationfor durablepurchases that is not dissimilarfrom that of otherstudies, but it includesno directrole for either wealth or fluctuationsin the stock market.Instead, total consumptionis used to scale all variablesin the equation:in effect,the equationexplains the ratio of purchasesof durablegoods to total consumption.As a result, changesin the stockmarket have only an indirecteffect through the change in total consumption.A directtest of the capital-gainsand wealthterms in a durables-expenditureequation is shown in column (4) of table 1. The lagged stock of durableswas includedto capturethe stock-adjustment effects,but it was highly collinearwith wealth and both are insignificant and of the wrongsign. The lags on the incomeand corporate-stockterms wereboth significantlynegative after a few periods.'7This is a highlysim- plifiedpurchase model, but it does indicatethat the wealth formulation does not apply directlyto durablepurchases. The alternativeequation for consumerdurables relates the desiredstock (K*)to incomeand the housingstock (as a measureof complementaryde- mand for householdfurnishings). Because of the statisticalproblems of estimatingan equation with a lagged dependentvariable, net purchases (NCD) are relatedto currentand previouschanges in the desiredstock:

17. Becauseof the presenceof a lagged-stockterm the equationwas estimatedby an instrumental-variabletechnique to obtain a consistent estimate of the degree of auto- correlation.The multicollinearityamong the variables greatly increasedthe standard errorsand leaves one with little faith in any of the point estimates. BarryBosworth 269

(4) NCD =f (EW, AKt_).18A

In addition,recent studies of the demandfor durableshave includedthe unemploymentrate for males, an index of consumersentiment, and the rate of priceinflation. These variables could be interpretedas elementsof the desiredstock (implying that they wouldenter with lags on the changes) or as cyclicalvariables that affectthe timingof the adjustmentprocess (in whichcase they would enter in level form).In practice,the specificform was determinedby experimentation.Finally, the level of income was also in- cluded in the equationbecause not all items in the categoryof durable- goodspurchases can be regardedas additionsto stocks.Many are more like nondurables,for whichpurchase and consumption are nearly simultaneous. The finalequation for the 1954-72period is

10 (5) NCD = -8.2 + 0.69 E wi -vYD* i-1.69 RUMt- (3.2) (2.6) i-O (5.3) + 0.57 ICRt - 1.19 CPI + 0.025 YD*t (5.2) (2.8) (3.9) + 0.62 uti, RI = 0.96; standard error = 1.0. where NCD = consumer-durablepurchases less depreciation,1958 dollars YD* = disposableincome less transferpayments, 1958 dollars RUM = rate for males ICR = two-quarteraverage of residentialconstruction, 1958 dollars CPI = two-quarterchange in the consumerprice index.'9 18. This is a simple transformation,with a more flexible lag, of the more standard stock-adjustmentformulation: NCDt = -(Kt-Kt-1). The problem of a lagged dependent variable with autocorrelatederrors can be seen easily by adding Kt-1 to both sides of this equation, Kt = yK* + (1 -)Kt-1, recognizingthat NCD is the currentchange in that stock. 19. The data on expenditure,income, and residential construction are from the national income accounts, published in Survey of CurrentBusiness. Depreciation of consumer durables is from the Board of Governors of the Federal Reserve System, "Flow of Funds Accounts,"various issues. The filteredindex of consumersentiment, as describedin Justerand Wachtel,"Inflation and the Consumer,"pp. 110-12, was initially includedin the equation, but was of very low significancein this final version. Quarters containingserious strikeswere eliminatedfrom the estimationperiod. 270 Brookings Papers on Economic Activity, 2:1975 In concept,relative prices should also appearin the equation;but, at this level of aggregation,their influencewas small and insignificantwith no apparenteffect on the coefficientsof othervariables. An attemptwas made to includea surveymeasure of expectedprices, but it was also insignificant. The negativesign on the actualinflation rate accordswith severalrecent studiesof demandfor durablegoods. The coefficienton new construction (measuringchanges in housingstock) seemsrather large, but is confirmed by other durables-purchasesequations, such as that of the MPS model. Attemptswere made to includesome measures of creditstringency (reflect- ing a beliefthat housing may be a proxyfor suchfactors), but theywere not significant.20 As withnondurables, net worthand the changein the valuesof corporate stock wereadded to equation(5). In this case, both variableswere clearly insignificantin the equation,with the coefficienton wealthnearly identical to zero.Variations in the lag structuremade it possibleto obtaina positive coefficientfor corporatestock, but it never approachedstatistical signifi- cance.The coefficientsof the othervariables in the equationwere not sig- nificantlyaffected. The equationwas also estimatedwithout the housing variable,but this changemade no differencefor the effectof wealthor the stock market.At least for purchasesof durablegoods, it is difficultto find evidence in the aggregatetime-series data of a significantinfluence of fluctuationsin the stock market. The resultsfor nondurablesand servicesoffer very strong supportfor a stock-marketeffect on consumption.But one would not anticipatea priorithe extremeresult obtainedin these equationsof a positive wealth and stock-marketeffect on nondurablesand servicesand no effectof either on durablespurchases. Purchases of durablesshould be postponableand thus responsiveto changesin stock values. In addition, if durablepur- chasesare very sensitiveto cyclicalchanges in income,one would expecta similarresponse to cyclicalchanges in wealth.Viewed through the concept of net worth,the resultsmay seemmore plausible:the rich spendmore on housingand otherservices. But these differencesdo not seemgreat enough to accountfor the failureof wealthand stock-marketfluctuations to have any effect on durablepurchases while exertinga strong impact on other consumptioncomponents.

20. The housing variablehas a coefficienttoo large to be simply a reflectionof com- plementarydemand. There is also a correlationbetween autos and housing. Thus, the housingvariable must be reflectingcredit or expectationalfactors that cannot be cap- turedwith more direct measures. BarryBosworth 271

THE 1973-74 STOCK-MARKET DECLINE The depth of the declinein the stock marketin the years 1973-74 pro- videsan opportunityto evaluatethe equationsoutside the periodof estima- tion. The resultsof using the previousequations to forecast1973-74 are shownin table 2 by half-years.The most strikingaspect of the table is the enormousoverprediction of consumerexpenditures on nondurablesand servicesduring the period,particularly 1973. This erroris reflectedin the large rise in the personal saving rate between 1972 and 1973, from 6.6 per- cent to 8.2 percent.In essence,none of this increasewas predictedby equa- tion (2), whichexcludes the wealthand stock-marketvariables. The inclu- sion of the wealthvariable in equation(3) significantlyreduces the error, but the residualsremain far larger than wouldhave been expected from the standarderror of estimateover the data period. Column(3) of the tableshows that, eventhough they remain consistently negative,the predictionerrors for the nondurablesand servicesequation are sharplyreduced when the period of estimationis extendedthrough 1974.As mightbe expected,the coefficientson wealthand corporatestock riseby abouttwo-thirds and the sum of the weightson incomedrops. Much of the improvementin the residualsresults from a rise in the autocorrela- tion parameter from 0.49 to 0.71. In addition, the reestimated equation consistentlyunderpredicts consumption in the stock-marketdeclines of 1970, 1966, and 1962. As mentionedearlier, the MPS formulation-equation(1)-constrains the effectof capitalgains to equalthat of wealthand thus it impliesa larger effect of the stock-marketdecline. As shown in column(4), this equation also overestimatesconsumption, but by a smalleramount than the equa- tion without the wealth and stock-marketterms. The projectionsbenefit somewhatfrom the higherestimate of the autocorrelationparameter. The errorsfor the durables-purchasesformulation-equation (5), which has no stock-marketvariables-display no distinct pattern (column 5). They offset the errorsin other consumptionuntil the oil embargoof late 1973.Actual expenditures run above the predictedamounts throughout the middleof 1974,a patternthat mightbe explainedin partby the announce- ment of the huge priceincrease on the 1975automobile models. About half of the $14 billion dropin expenditureson durablesin the fourthquarter of 1974is predictedby the equation. The overallresults of the predictionstend generallyto supportthe earlier 272 Brookings Papers on Economic Activity, 2:1975 Table2. PredictionErrorsa for ConsumerExpenditures, Second Half 1972-1974 Annual rates in billions of 1958 dollars Nondurables Nondurablesand services and services with wealthvariable without MPS Year wealth 1972 1974 model Consumer and variable equationb equation" equation durables half (1) (2) (3) (4) (5) 1972:second 0.1 -0.2 -0.1 0.9 -0.4 1973:first -4.1 -3.2 -1.1 -0.3 2.1 second -5.6 -5.4 -2.1 -3.0 -1.5 1974:first -7.1 -5.7 -0.9 -1.6 2.5 second -5.0 -2.7 -0.1 1.7 -1.4 Source: Calculated by the author from text equations (1), (2), (3), and (5). The indicated residuals include a correction for autocorrelation based on the estimated equations. As such, they should be interpreted as averages of one-quarter forecasts. a. Actual minus predicted. b. Text equation (3). c. Text equation (3) reestimated with data for 1954-74. conclusions.There is no evidencethat durablesoutlays are lower than pre- dicted duringthe period of declinein stock prices.On the other hand, a majorproportion, but not all, of the shortfallin outlays on nondurables plus servicescan be accountedfor by referenceto the stock marketand to the deteriorationof wealth,particularly on the basis of the 1964equation.

CROSS-SECTION RESULTS

Potentially,surveys of individualfamily units offer a much richerbase for testinghypotheses about consumer behavior because their data provide far wider independentvariation. Unfortunately, they introducenew sta- tisticalproblems as well. For one thing,very few surveyshave collected the balance-sheetdata required to computea directmeasure of wealth.Second, the problemof measurementerror is likelyto be more severethan it is for time-seriesdata becausethe surveysmust usuallyrely on the memory of interviewees.Third, the effect of economicvariables is difficultto disen- tangle from that of tastes in comparingindividuals in differentcircum- stances. That two individualsin differenteconomic positions consume differentlydoes not meanthat, shouldtheir economic positions become the same, they would then adopt the same consumptionpatterns. BarryBosworth 273 An extensivesurvey of the financialcharacteristics and income of indi- vidualswas conductedby the FederalReserve Board in 1963.21Respon- dentswere reinterviewed in 1964to obtaindata on theirincome and saving in 1963.22The directmeasurement of net worth in these surveyshas per- mittedseveral investigators to use it to test the life-cyclehypothesis-that the consumerunit saves so that its total lifetimeresources are distributed overthe life cyclein the most favorablepattern of consumption.Regressing savingon incomeand net worthfor severalage categoriesand comparing the resultswith valuespreviously hypothesized by Modigliani,Brumberg, and Ando,28Projector concluded from the differencesbetween the survey's coefficientson wealthand the hypothesizedvalues that the surveyevidence was unfavorablefor a narrowinterpretation of the hypothesis.24However, more importantfor the purposeshere, in explainingsaving she did obtain significantnegative coefficients on net worthin half of the age classesand a highlysignificant overall effect. A somewhatmore elaborate test withthe samedata was done by Robert Rasche.25He adjustedthe incomeand wealthdata prior to estimationto re- flect differentstages of the life cycleand used a two-yearaverage of income to approximatenormal income. Saving was then relatedto income and beginning-of-periodassets. His aggregateequation supportsthe role of net worth in the saving function,for he obtains a highly significantand plausiblecoefficient. Unfortunately, disaggregation of the datainto five age classesresulted in extremelywide variationin the net-worthcoefficient (to the extentthat a significantpositive coefficient was obtainedfor individuals under35 yearsof age).Thus, he concludedthat the resultswere mixed and inconclusive. Whilethese studies are only indirectly related to the issueof stock-market effectson consumptionbecause they test the broaderissue of the role of

21. Dorothy S. Projectorand GertrudeS. Weiss, Surveyof FinancialCharacteristics of (Board of Governorsof the FederalReserve System, 1966). 22. Dorothy S. Projector,Survey of Chlangesin FamilyFinances (Board of Governors of the FederalReserve System, 1968). 23. Albert Ando and Franco Modigliani,"The 'Life Cycle' Hypothesis of Saving," AmericanEconomic Review, vol. 53 (March 1963), pp. 55-84; Modiglianiand Richard Brumberg,"Utility Analysis and the Consumption Function: An Interpretationof Cross-SectionData," in Kenneth K. Kurihara,ed., Post-KeynesianEconomics (Rutgers UniversityPress, 1954). 24. Projector,Survey of Chianges,pp. 1-2, 17-23, 87. 25. Robert H. Rasche, "Impactof the Stock Marketon PrivateDemand," American EconomicReview, vol. 62 (), pp. 224-28. 274 Brookings Papers on Economic Activity, 2:1975 wealth,a recentstudy by IrwinFriend and CharlesLieberman includes a directtest of the capital-gainseffect.26 From detaileddata aboutindividual stockholdings in the FRB survey,they computed the marketvalue of stock at the beginningand at the end of 1963,as reportedon the stockexchanges, and thus were able to estimatecapital gains duringthe year. Friendand Liebermanregressed saving on normalincome (two-year aver- age), transitoryincome, beginning-of-periodnet worth, their estimateof capitalgains, and severalage, occupation,and family-sizevariables. The resultsof their test offer evidencethat capital gains and losses do have a contemporaneouseffect on saving.The coefficientsrange between -0.015 and -0.041 and are frequentlysignificant. Since this is a one-year effectonly, the resultsare quiteconsistent with the MPS time-seriesequa- tion. This studyis a majorstep in reducingthe uncertaintyabout the effects of capitalgains and losses. As is alwaysthe case, however,some doubtsarise. First, contraryto the outcomeof Rasche'sstudy, the effectof net worthis smalland generallyin- significant.Good reasonssuggest that the wealthcoefficient will be biased toward zero in a cross-sectionanalysis (principally, its correlationwith "tastes"),but it is difficultto reconcilesuch divergent results for two studies basedon the samedata. An effectof capitalgains without a corresponding role for wealthagain introducesthe problemof distinguishingbetween a wealthand a "sentiment"mechanism. Second, when the data were disag- gregatedinto five age categories,the capital-gainscoefficients ranged be- tween -0.165 and -0.003 and werenever significant (though perhaps this was a problemof small samplesize). In addition,the wealth coefficients changedsigns in some casesand weresignificant for only one age category. Finally,1963 was a year of majoradvance for stock prices(the composite indexof the New York StockExchange rose 18 percent).Thus, to some ex- tent,using capital gains for one yearwill identifythose who havelarge stock holdings,and thus intensifythe problemof correlationwith "tastes." The resultsof the Friendand Liebermanstudy are sufficientlystrong to justifyan attemptto verifythe resultswith othersurvey data. A four-year panelstudy of 1,400families between 1967 and 1970,conducted by the Sur- vey ResearchCenter of the Universityof Michigan,included the required dataon incomeand balance-sheet items, as well as informationon automo- 26. Irwin Friend and Charles Lieberman,"Short-Run Asset Effects on Household Saving and Consumption:The Cross-SectionEvidence," AmericanEconomic Review, vol. 65 (September1975), pp. 624-33. Barry Bosworth 275 bilesand otherdurable-goods purchases.27 While the qualityof the data on net worthis not as highas that of the FRB survey,the panelstudy offers the majoradvantage of followingthe sameindividual over fourinterviews. Estimatesof capitalgains are limitedto two years, 1968 and 1969, be- causeof changesin the formof the questions.These estimates are based on respondents'answers about the marketvalue of theirholdings rather than actualmarket prices, as in the Friendand Liebermanstudy. These estimates may be wrong,but they seem to be most relevantto the questionof what impactcapital gains have on consumptionbehavior. The total samplewas limitedto 191 familiesbecause most of the inter- vieweesdid not own stock, otherswould not answerall of the questions, and some cases containedobvious errors.This sampleis smallerthan the 303 cases of householdsthat owned stock, availableto Friendand Liber- man, but sincethe dataspan two years,the numberof observationsis sub- stantiallylarger. In addition, 1968 was a year of sharp gains in average stock values,whereas prices fell in 1969. From the data,a measureof savingcan be constructedthat excludesthe effectsof all revaluationsof asset prices,while includingnet purchasesof durablegoods. As in the Friendand Liebermanstudy, saving is definedto includean estimateof depreciationon owner-occupiedhomes. The savingestimate is, however,subject to seriousmeasurement error. If capitalgains are disregarded,saving is equalto the changein net worth: St = NWt - NWt-1. In an equation in which beginning-of-periodnet worthis usedto explainsaving, there will be an automaticnegative correla- tion betweensaving and net worthas a resultof any measurementerror in reportingnet worth.The problemwas minimizedfor some asset itemsbe- causerespondents were asked about net purchases,but for most debtitems, only beginning-and end-of-periodestimates are possible. Friend and Liebermanobserved that tasteeffects would bias the expectednegative net- worthcoefficient toward zero: householdswith a high propensityto save will have high net worth. The measurementerror, however, will bias the coefficienton net worthin the oppositedirection. The resultscan be summarizedsimply: none of the tests yieldeda signifi- cant coefficientfor net worthor capitalgains. But in some casesthe coeffi-

27. More detailed informationabout the nature of the survey is contained in Gary Hendricks and Kenwood C. Youmans, ConsumerDurables and InstallmentDebt: A Study of AmericanHouseh0olds (University of Michigan, Institute for Research, 1973). 276 Brookings Papers on Economic Activity, 2:1975 cients on capital gains had the expectedsign even though they did not achievecommon standards of statisticalsignificance. Most of the equations for durablesexpenditures were fitted in the followingmodel: (6) EXPt=a + b,YN + b2A Y + b3NWt-,+ b4CGt+ b5Kt_j+ DV, where Y = after-tax income YN = normalincome, definedas the four-yearaverage of after-tax income AY = incomegrowth, defined as the averageincome of 1968-69mi- nus that of 1966-67 Kt-, = beginning-of-periodstock of durables NWt-, = beginning-of-periodnet worth CG, = capitalgains on corporatestock, definedas the changein mar- ket value minusnet purchasesduring the year DV = demographicvariables-age of the head of familyand family size. Directly measuredexpenditures included net purchasesof automobiles, householddurables, recreational equipment, and vacation expenses.Re- gressionswere estimated for the total as well as individualcomponents; the years 1968 and 1969 were analyzedseparately, combined as a two-year sample,and stackedas a singleseries of observations.28 Themost favorableresults in supportof a capital-gainseffect on durable- typeexpenditures were obtained by poolingthe datafor 1968and 1969into one sampleand dividingall variablesby YN:

EXP AY NWt__ (7) Nt=o0.143 + 0.021 Y 0.0015 NW (4.7) (1.2) (0.4) CG Kt, + 0.023+ YN -0.123 YN + DV. (1.4) (1.6) This equationexplains about 23 percentof the variationin the total of the four expenditurecategories. The capital-gainscoefficient has the expected positivesign, but it is of marginalsignificance and the coefficienton net worthis negative.

28. Some equations were also correctedfor heteroskedasticityby using YN as the scale variable,but with little effect. Barry Bosworth 277 The correspondingequation for saving (SAVE) of the combinedtwo- yearsample yielded a largercoefficient on capitalgains, but againit was not significant,and the coefficienton net worthwas not of the expectedsign: (8) SA VEt = 1553.24- 0.244 YN + 0.499 A Y (2.0) (1.9) (4.6) + 0.011 NWt_-0.038 CGt + DV. (1.3) (0.9) R = 0.18. For the regressionsbased on individualyears and individualexpenditure components,the capital-gainscoefficients were smaller than those reported aboveand sometimesof the oppositesign. An indexof consumersentiment was also constructedfor eachindividual, using the questionsasked in com- piling the aggregateindex of the MichiganSurvey Research Center. This variablewas neversignificant and did not affectthe size of the capital-gains coefficient.Thus, the statisticalresults do not stronglysupport a capital- gains effect,although the magnitudeof the coefficientsis withinthe same range as those reportedby Friend and Lieberman.29Also, the reported equationsyielded the most positiveresults that wereobtained in favor of a stock-marketeffect. The resultsfor net worth were never significant,and the capital-gainscoefficient sometimes bore the unexpectedsign. This surveyof the empiricalevidence is disappointingin that it does not permita firm acceptanceor rejectionof the hypothesisthat stock-market fluctuationsaffect consumer expenditures. Nevertheless, I believethat some tentativeconclusions emerge. First, the weightof the evidencesupports a positive impact of the stock marketon consumption.The impact is not heavy,but extremefluctuations in stockprices can causeit to be significant in some periods. Second, the results indicate that fluctuationsin stock pricesdo not help explainthe erraticshort-term behavior of durable-goods purchases.An improvedunderstanding of the behaviorof this component probablyrequires the incorporationof other factors.

IMPLICATIONS FOR MONETARY POLICY Accordingto the MPSmodel, the effectthat capital gains and losses have on consumptionthrough the wealthmechanism is a majorchannel through 29. One might argue that relying on respondents'estimates of capital gains results in a larger measurementerror than does the method used by Friend and Lieberman. Thus, the standarderror in the coefficientwould be expectedto be largerand the problem is simply that my sample was too small. 278 Brookings Papers on Economic Activity, 2:1975 whichmonetary policy altersaggregate demand. This mechanismis com- pleted with the specificationof a relationshipbetween rates of returnon corporatestock and other assets. By changingmarket rates of interest, monetarypolicy can change stock prices, and thus create and destroy wealth.30However, this notion that interest-ratechanges lead to a chainof changesin stock prices, wealth, and consumptionhas some conceptual problems,because it combinesthe effects on consumptionof changesin wealthand changesin interestrates even though they may have substan- tially differentbehavioral implications. Stock priceschange for a varietyof reasons:they may declinebecause individuals'expectations of total future incomes have declined,because they expecta smallershare of this incometo accrueto ownersof corpora- tions, or becauserates of returnon other assets have risen. When stock values declinebecause of lower expectationsof income,consumption can be expectedto fall in the currentand futureperiods.3' This is an unam- biguousincome effect. If the declinein stock pricesresults from a rise in interestrates, however,there is a negativesubstitution effect on current consumption(which was not presentin the previouscase) plus an income effectfrom higherinterest payments in futureperiods. The sign of this in- come effect on currentconsumption will dependupon whetherthe indi- vidualintended to be a saveror a borrowerbefore the interest-ratechange.32 Althoughthe substitutioneffect arisingfrom an increasein interestrates depressescurrent consumption, its magnitudeis unrelatedto the income effectof the firstcase, in whichexpected future income fell. Second,given a negativesubstitution effect and an indeterminateincome effect, the net im- pactof a changein interestrates on currentconsumption is uncertain.Thus, the implicationsfor consumptionof changesin stock prices spurredby shifts in earningsexpectations on the one hand and interestrates on the otherhave little in common. 30. This issue is discussedin considerabledetail in Franco Modigliani,"Monetary Policy and Consumption,"in ConsumerSpending and MonetaryPolicy: The Linkages, MonetaryConference Series 5 (FederalReserve Bank of Boston, 1971), pp. 9-84. 31. It is assumedthat consumptionis not an inferiorgood. 32. The effect of interest-ratechanges on consumptionis discussedin greaterdetail in articlesby M. S. Feldstein and S. C. Tsiang, "The InterestRate, Taxation, and the PersonalSavings Incentive," QuarterlyJournal of Economics,vol. 82 (August 1968), pp. 419-34; M. J. Farrell, "The Magnitudeof 'Rate-of-Growth'Effects on Aggregate ,"Economic Journal, vol. 80 (December1970), pp. 873-94; and JamesTobin and Walter Dolde, "Wealth, Liquidity and Consumption," in ConsumerSpending and MonetaryPolicy, pp. 104-15. BarryBosworth 279 Certainly,over the postwarperiod changes in stockprices have reflected primarilychanges in expectationsof earnings-if for no otherreason than that sharpvariations in interestrates are a relativelyrecent phenomenon. Thus,it is questionablethat any observedcorrelation between stock prices and consumptionover the postwarperiod can be used to deducethe effect on consumptionof monetarypolicy. Since the influenceof monetarypolicy involvesa substitutioneffect between current and futureconsumption, the focus shouldbe upon consumptionequations that includean interest-rate variable. This reasoningconflicts with some formulationsof the life-cyclehypoth- esis of consumption.That hypothesis is commonlyinterpreted to implythat currentconsumption is a functionof currentwealth and the presentdis- counted value of currentand future nonpropertyincome. Since current wealthis simplythe presentvalue of futureproperty income, current con- sumptionis a functionof the presentvalue of currentand futureincome: Ct= k(PV). If interestrates go up, the presentvalue of futureincome declines and the hypothesispredicts that consumptionin the currentperiod will fall. This predictionappears to contradictthe previousstatement-that the income effectof interest-rateincreases can be positive.The contradictionreflects a misinterpretationof the originalhypothesis. The originalbudget constraint of Modiglianiand Brumbergrelated the presentvalue of currentand future nonproperty income, Y, plusbeginning- of-periodnet assets,at, to the presentvalue of currentand futureconsump- tion, C, over the remainderof an individual'slifetime, L, plus any be- quests.33If bequestsare ignored,this relationcan be expressedas

L Y7 L C7 (9) at at++ =(1 +:+'-+ + r)_+l-_t= r)T+t(t' where r is the rate of interest. If currentconsumption and income are separated,the relationshipis

Y,- (10)(10) Ct = at(l + r) + Yt + ~~~~~~~~~LE l C7)

However,the hypothesishas frequentlybeen used in a form that relates currentconsumption to the presentvalue of currentand futureincome and 33. "Utility Analysis and the ConsumptionFunction," p. 391. 280 Brookings Papers on Economic Activity, 2:1975 ignoresthe presentvalue of futureconsumption. In equation(10), a rise in interestrates will changeboth the presentvalue of currentwealth, at, and the presentvalue of futuresaving, (Y, - C1).If at is positive,a risein r will reduceits presentvalue; but if the individualhas positivecurrent wealth, futureexpected saving must be negativeand a rise in r will increaseits presentvalue and thus offsetthe dropin wealth.The conversewill hold for a debtor,who must be planningto have positivefuture saving. A rise in stock values that resultsfrom revisedexpectations of future earningswill havea positiveeffect on currentconsumption. An equalrise in valuethat resultsfrom lower interest rates will also have a wealtheffect on currentconsumption, but this will be partiallyoffset by a risein the present value of futureconsumption. In addition,the interest-ratechange will in- duce some substitutionbetween currentand future consumption. Thus,the two mechanismsby whichstock pricescan changehave different effectson currentconsumption and one cannotinfer an interest-rateeffect on consumptionfrom evidenceof correlationbetween such expenditures and changesin stock values. In summary,the coefficienton wealthin a consumptionequation mea- sures an averageeffect of (1) changes in earningsexpectations, and (2) changesin interestrates. There is reasonto expectthat the firsteffect will be the largerand that the coefficienton wealth overestimatesthe impact of changesin interestrates. If an interestrate were includeddirectly in the equation,some of the ambiguitieswould be resolved,but past efforts(in- cludingmy own) to do so have been unsuccessful.In the meantime,the currentformulation is of limitedvalue in measuringthe impact on con- sumptionof monetarypolicy.

Investment

The stockmarket and investment behavior are intimately bound together sincefirms invest to earnprofits, and activityin the stockmarket represents an attemptby investorsto evaluatethe magnitudesof that streamof profits. But any attemptto go beyondthat obviousstatement to determinea causal relationshipinvolves many issues that remain controversial.Several of thesecan be illustratedby referenceto the neoclassicalmodel of investment demand,which has been used in many empiricalstudies. Accordingto the neoclassicalmodel, the competitivefirm attemptsto Barry Bosworth 281 maximizethe presentvalue of its futureincome stream.If there exists a resalemarket for capital,and if constantreturns to scalein productionis a close approximationto reality, the firms' desiredcapital stock in equi- librium,K*, is givenby

(11) =K*-Al Q, where A' = a constantgiven by the specificproduction function and competi- tive conditionsin the productmarket and factormarkets Pq = price of

Pk = Pk(r + d- ) = rentalprice of capital Pk = price of capital r = discount rate ( of capital) d = depreciation rate a = elasticityof factor substitution Q = output. The logic of the rental-priceconcept is simplythat the cost of holdinga unit of capitalfor one periodis equalto the interestcost, plus depreciation, less any capitalgain arisingfrom pricechanges in the resalemarket.34 For such a model the cost of capitalcan be measuredby a weightedav- erageof the cost of differentsources of financing.35If the valueof the firmis the sum of bonds and stocks outstanding,the discountrate is represented by the weightedaverage of the returnson thesetwo assets.Thus, the stock marketseems to affectthe firm'sinvestment decisions through its influence on the cost of financingor the appropriatediscount rate. In empiricalwork the modelembodied in equation(11) must be expanded to includethe effectson the rentalprice of capital of numeroustax mea-

34. A more detailed developmentof the model can be found in Dale W. Jorgenson, "The Theory of InvestmentBehavior," in Robert Ferber, ed., Determinantsof Invest- ment Behavior(Columbia University Press for the National Bureau of Economic Re- search, 1967); Robert E. Hall and Dale W. Jorgenson,"Application of the Theory of Optimum ," in Gary Fromm, ed., Tax Incentivesand Capital Spending(Brookings Institution, 1971); and the referencesincluded in both studies. 35. See, for example,Franco Modiglianiand MertonH. Miller,"The Cost of Capital, CorporationFinance and the Theory of Investment,"American Economic Review, vol. 48 (June 1958), pp. 261-97. 282 BrookingsPapers on EconomicActivity, 2:1975 sures,and to allowfor somemechanism by whichthe actualcapital stock is broughtto the desiredlevel. Many estimateshave excludedthe yield on stockfrom the cost of capitaland have used eithera constantdiscount rate or the bond rate alone. Thus, they do not provide directevidence of the effectof changesin stock prices. The simplifiedmodel can be used to highlightthe issues.First, the range of substitutionpossible between capital and labor in the productionprocess playsa crucialrole. If this elasticityis zero, as some investigatorshave ar- gued, relativeprices play no role in the model and stock-marketfluctua- tions are irrelevant.This is the basis of the acceleratormodel, in whichthe desiredstock is relatedonly to expectedoutput. The argumentfor a low elasticityof substitutionrests on threedebatable assertions: (1) alternative productionmethods do not exist exceptin the presenceof largechanges in relativeprices; (2) firmsdo not have full a prioriknowledge of the alterna- tive techniquesavailable and of futureprices; or (3) firmsdo not maximize profits.Jorgenson and his associates,on the otherhand, have typicallyas- sumedan elasticityof one in theirstudies and cite supportingevidence from specificstudies of industryproduction functions.36 Second,the modelassumes, obviously contrary to fact, that capitalhas a full resalemarket so that it becomesa variablerather than a fixedfactor of production.It is this assumptionthat is responsiblefor the simplemyopic rulethat reliessolely on currentvalues of the economicdeterminants of the desiredcapital stock; there is no needto forecastoutput and relativeprices. Third,the measureof the cost of capital,r, as a weightedaverage of bond and stockyields serves merely as a definitionfor the individualfirm. Unless one can specifyhow the returnon stock (expectedcapital gains as well as dividends)is determined,the relationshiphas no behavioralimplications. Suchspecification is empiricallydifficult because expected capital gains are not observable;yet they are a key elementof the returnon stock. Manage- mentmust be viewedas approvingthose projectsexpected to increasethe firm'smarket value. The weighted-averagemeasure of the cost of firmsis 36. This issue still has not been completely resolved, but the argumentsare well representedin a series of articles in Reviewof Economicsand Statistics: Robert Eisner and M. I. Nadiri, "InvestmentBehavior and Neo-classical Theory," vol. 50 (August 1968),pp. 369-82; Dale W. Jorgensonand JamesA. Stephenson,"Issues in the Develop- ment of the Neoclassical Theory of InvestmentBehavior"; and Charles W. Bischoff, "HypothesisTesting and the Demand for Capital Goods," vol. 51 (August 1969), pp. 346-53, 354-68, respectively. BarryBosworth 283 simplyan ex post estimateof the accuracyof those expectations.37In addi- tion, the measurementof the cost of capitalis not so simplewhen taxes, transactioncosts, and bankruptcyrisk are introducedinto the model. In this case the marketvalue of the firm,and thus its cost of capital,may re- flect its financialas well as its productiondecisions.

THE STOCK MARKET AND THE COST OF CAPITAL Bischoffhas done considerableempirical work with an investmentequa- tion that is a variantof the neoclassicalformulation. As appliedto business investmentin equipment,this equation definesthe cost of capital as a weightedaverage of the bond rate(adjusted for inflationexpectations) and the dividend-priceratio. The weightassigned to the dividend-priceratio is abouthalf of thatfor the bondrate. This measure of the returnon corporate stock ignores the expectedcapital gains, but the inclusion of the stock- marketvariable seems to improvethe performanceof the equation.38 The equationimplies that the elasticityof investmentspending in equi- libriumwith respectto a changein the dividend-priceratio is about 0.12 (withrelative prices and tax ratesof 1972).This wouldtranslate to a decline of about $3.5 billion(1972 dollars)in investmentspending for a rise of 1 percentagepoint in the dividend-priceratio. Even this effectwould require considerabletime to haveits full impactbecause of the lag betweenchanges in the determinantsof investment,the placementof new orders,and pro- duction.At a constantlevel of output,about one-halfof the impactwould be felt afterone yearand the full adjustmentwould be stretchedover about fouryears.

37. If all risk accruesto stockholders,the bond rate can be viewed as approximately exogenous to the firm, but the return on stock reflects the personal-riskcharacteristics of the firm relativeto others and expectationsfor the specific investmentprojects that it undertakes.To the extent that the risks associatedwith individualfirms are uncorre- lated, diversificationby investorsor mergersof firms can reducethe importanceof the risk component. 38. For the formulationof the equationand a discussionof its properties,see Charles W. Bischoff, "The Effect of AlternativeLag Distributions,"in Fromm, ed., Tax Incen- tives and CapitalSpending, pp. 61-125. The version used in the MPS model differsin a few minorrespects from that reportedin the Bischoffchapter, and is publishedin Albert K. Ando and others, "On the Role of Expectationsof Price and TechnologicalChange in an InvestmentFunction," InternationialEconomic Review, vol. 15 (), pp. 384-414. 284 Brookings Papers on Economic Activity, 2:1975 The recentdecline in the stock marketraised the dividend-priceratio by 2.7 percentagepoints from fourth quarter 1972 to fourthquarter 1974. Had the declinebeen permanent, the equationwould project a $9 billionshrink- age in investment.This is not a trivialamount but, becauseof the response lags, it cannotbe a majorsource of the depressiveforces at work in 1973 and 1974. The MPS model uses a similarequation to explainnonresidential con- struction.The elasticityof substitutionis estimatedto be 0.45 ratherthan the 1.0 of the equipmentequation; but some offsetcomes from a largerrole for the dividend-priceratio in the cost-of-capitalterm. As a result,the elas- ticitywith respectto changesin the stockmarket is nearlyidentical to that of equipment,0.12. The dollar magnitudeis smaller since outlays for structuresare about one-halfto one-thirdof those for equipment.Again, the lag extendsover three years with one-third of the effectfelt by the end of the firstyear. A permanentdecline of stockvalues to the levelsof the fourth quarterof 1974would reduce total investment(equipment plus structures) by $15 billion.

THE SECURITIES-VALUATION MODEL An alternativeapproach to investigatingthe link betweenthe stockmar- ket and investment,the securities-valuationmodel, places greater emphasis upon the stock market as a determinantof investmentdemand. James Tobin and WilliamBrainard, in particular,have used a modelin whichin- vestmentis determinedby the ratio of the marketvalue of the firmto the replacementcost of its physicalassets.39 When this ratio,q, is greaterthan unity,the value of capitalin the marketis higherthan the cost of producing it, and investmentis stimulated.In Tobin'smodel of the financialprocess, q providesthe link betweenthe real and financialsectors since it represents the ratioof the net returnon realassets to the returnon equity.40Monetary policy can influencereal activityby affectingthe rate of returnon equity. An increasein the quantityof reducesshort-term bond rates; and throughportfolio-balance adjustments the public'sdemand for equitiesis increased,lowering the requiredreturn and raisingq.

39. William C. Brainardand James Tobin, "Pitfalls in Financial Model Building," AmericanEconomic Review, vol. 58 (May 1968), pp. 99-122. 40. See, for example,James Tobin, "A GeneralEquilibrium Approach to Monetary Theory,"Journal of Money, Creditand Banking,vol. 1 (February1969), pp. 15-29. Barry Bosworth 285 Althoughit appearsmuch different, the securities-valuationmodel is con- ceptuallyidentical to the versionof the neoclassicalmodel used in most em- piricalstudies. Both are basedon the first-ordercondition for profitmaxi- mizationby whichthe marginalproduct of capitalmust equalthe ratio of its serviceprice to the priceof output.In termsof the notationof equation (11), this is (12) 8F Pk(r + d)-Pk 8K ~pq In most of the work with the neoclassicalformulation, as reflectedin the studiesby Jorgensonand others,a specificform of the productionfunction is assumedand the expressionfor the marginalphysical of capital is insertedinto equation(12). The solutionyields a desiredcapital stock in terms of expected output, relative prices, and the discount rate. The securities-valuationapproach, on the other hand, is derivedby rewriting the equilibriumcondition as

Pq25F Pk (13) Pk K Pd+ The termon the left-handside is simplythe net marginalrevenue product of capitalrelative to its price.If q is definedas the ratio of this returnto the discountrate, r, a valueof q greaterthan unity implies a disequilibriumcon- ditionin whichthe rate of returnis greaterthan the discountrate (cost of capital)and furtherinvestment is profitable. The securities-valuationformulation offers the advantageof not requir- ing the explicitmeasurement of the effect of taxes, expectedoutput, and expectedprices needed in the neoclassicalversion. This simplificationre- sults fromthe assumptionthat the marketcorrectly values the futureearn- ing capacityof the firm.If the firmseeks to maximizethe discountedvalue of its futureincome, and investorshave the sameinformation as the firm's managers,the two groupswill reachidentical conclusions. In otherwords, an estimateof the discrepancybetween the actualand desiredcapital stock of the neoclassicalmodel is availableby comparingthe marketvalue of the firmwith the replacementcost of its currentcapital stock. The model introducessome new problems,however. First, the ratio of the marketvalue of the firmto the replacementcost of its capitalis a mea- sure of the averagerather than the marginalexpected return on capital. Manyinstances can be given of divergencebetween these two measures.If a full resalemarket for capitalexisted, one could, in principle,adjust the 286 BrookingsPapers on EconomicActivity, 2:1975 existingcapital stock to reflectthe effectsof suddenobsolescence; but, in fact, currentreplacement value must be obtainedby assuminga deprecia- tion rateand valuing the stock withprices for new capital.Also, the riskof the potentialincome from an incrementto new investmentdiffers from that associatedwith existing capital. Moreover, the valueof a firmreflects assets other than its physicalcapital, and at the empiricallevel the model en- countersdifficulty in accuratelyvaluing them. Some of theseassets, such as patentsand knowledgeof the market,constitute barriers to the entry of otherfirms. The most seriousproblem with the approachas a vehiclefor understand- ing investmentbehavior is that it shifts the focus from what determinesa firm'sinvestment to what determinesvalues in the stock market.It does not seempractical to focus upon responsesin the stock marketto measure the impacton investmentof a changein tax law. Nor does it seemreason- ableto believethat the presentvalue of expectedcorporate income actually fell in 1973-74by the magnitudesimplied by the stock-marketdecline of that period,when q declinedby 50 percent.Of course,an equilibriumrela- tionshipmust exist between the marketvalue of a firmand the replacement cost of its capital.But it is quiteanother thing to infera causalmechanism fromthis relationshipand to allegethat changesin stockprices reflect only revisedevaluations of the discountedvalue of prospectsfor corporateearn- ings.As long as managementis concernedabout long-run market value and believesthat this value reflects"fundamentals," it wouldnot scrapinvest- ment plans in responseto the highly volatile short-runchanges in stock prices. Despite the fact that the securities-valuationmodel leaves the basic de- terminantsof investmentin a blackbox, it may have some value as a fore- castingdevice if othermodels cannot accurately reflect the complexforces that drive investmentdemand in the aggregate.An empiricalversion has been estimatedin studiesby Bischoffand by Ciccolo.4'Bischoff estimated the model for total businessinvestment for the 1953-68period. The model didnot fit the dataas well as a simpleaccelerator model or the MPSversion of the neoclassicalmodel, and an extremelyhigh level of autocorrelation requiredthat the equationbe estimatedin first-differenceform. In termsof averageestimation errors, the differencesamong the equationswere not 41. Charles W. Bischoff, "Business Investment in the : A Comparison of Models,"BPEA, 1:1971, pp. 13-58; and John H. Ciccolo, Jr., "Four Essays on Mone- tary Policy" (Ph.D. dissertation,Yale University,1975). BarryBosworth 287 great,however: the standarderrors of estimatefor the securities-valuation equationswere about 20 percentgreater than those of the MPS versionof the neoclassicalmodel. The resultsof forecastingbeyond the period of fit weremore clearly unfavorable to the securities-valuationmodel: the aver- age forecasterror for 1969-70was abouttwice that of eitherthe accelerator or the MPS equation. Ciccolodeveloped a somewhatmore refined measure of q basedon data for nonfinancialcorporations. He findsa significantrelation between total investment,I, deflatedby the gross stock of capital,and an eight-quarter lag structureon q for the period 1953-73:42

8 (14) 100 It/Kt- = 7.75 - 12.16(500/Kt-1) + 5.25 E wiqt-i + 0.966ut-i. (3.46) (4.2) RI = 0.98; standarderror = 0.15. The equationreflects a very strongeffect of changesin q on net invest- ment;but the autocorrelationcontinues to be severe,as in the equationes- timatedby Bischoff.The equilibriumcoefficient of 5.25 on q implies an elasticityof investmentwith respectto q of 0.77. With this ,the changein the dividend-priceratio betweenthe fourthquarter of 1972and the fourthquarter of 1974had a permanentimpact on investmentof $35 billion, comparedwith an estimated$15 billion reductionin the MPS model. Thus, the two modelsyield substantiallydifferent estimated magnitudes of stock-marketeffects. The MPS model emphasizesthe cyclicalimpor- tance of output and assignsthe stock marketa secondaryrole, operating throughthe cost of capital.The alternativeassigns all of the cyclicalchanges in investmentto the stock market. The MPS equationsfit the historicaldata with a smallerstandard error andwith less evidenceof autocorrelationin the residuals.For thesereasons, thatversion of an investmentequation warrants . It is also a more interestingway to estimatethe effectson investmentof changesin tax and monetarypolicy. However, the securities-valuationmodel does fit the data nearlyas well, andhas the advantageof beinga verysimple formulation.

42. The serieson investmentdata is gross privatedomestic nonresidential investment, as reportedin the nationalincome accounts.The gross capitalstock is publishedin John C. Musgrave,"New Estimatesof Fixed NonresidentialBusiness Capital in the United States, 1925-73," Surveyof CurrentBusiness, vol. 54 (March 1974), pp. 23-27. The esti- mate of the q ratio is publishedas an appendixto Ciccolo, "Four Essays." 288 Brookings Papers on Economic Activity, 2:1975 Table 3. PredictionErrorsa for Business Investment,Second Half 1972-1974 Annual rates in billions of 1958 dollars Adjustedfor autocorrelation Unadjustedforecast errors MPS Securities- MPS Securities- Year model valuation model valuation and equation equation equation equation half (1) (2) (3) (4) 1972:second -0.9 0.0 -1.1 1.7 1973:first 0.4 1.2 0.4 -1.3 second -0.3 0.2 -0.2 -1.3 1974:first 0.0 1.1 -0. 1 -3.0 second -1.1 -1.0 -1.3 -2.4 Source: Calculated by the author. The errors of columns (1) and (2) are adjusted by the estimated auto- correlation coefficients for each equation. These were 0.98 for the securities-valuation model and 0.4 and 0.6 for the equipment and structures equations, respectively, of the MPS model. As such, the errors should be interpreted as averages from one-quarter forecasts. Columns (3) and (4) show the forecast errors with no correction for autocorrelation. a. Actual minus predicted.

SOMEFORECAST RESULTS As with those for consumerexpenditures, these equationscan be com- paredwith regard to the accuracyof theirforecasts for the 1973-74period. An equationsimilar to that of Ciccolowas estimated to excludethe 1973-74 period.The equipmentand structuresequations of the MPS model were estimatedwith data that extendedonly throughmid-1968, so that the fore- cast is well outside its own period. The predictionerrors are shown in table 3, both with a correctionbased on the autocorrelationparameter of the estimatedequations and with no adjustmentfor autocorrelation. The MPS equations have very small forecast errors throughoutthe period.Furthermore, they show no obvious tendencyto over- or under- predictthe actual values. However, the semi-annualpresentation of the data does not fullyreflect the failureof the equipmentequation to forecast the collapseof new ordersin late 1974 and early 1975. Ordersfell 14 per- cent in the fourthquarter of 1974 and an additional12 percentin the first quarterof 1975, but the equation forecast a decline of only half this magnitude. The securities-valuationmodel also does verywell in the one-periodfore- castsof column(2), exceptfor a noticeabletendency to underpredictinvest- mentin the earlyperiod. The equationtends to performslightly better than the MPSequations during the largedeclines in investmentin late 1972and BarryBosworth 289 early1975 (not shownin the table).It predictsa largedecline in investment prematurely,but ultimatelyis provedright. This equationalso benefitsin a comparisonwith the MPS equationsbecause the latterare based on new ordersfor equipment,which is a morevolatile data seriesthan that for ex- penditures.However, the securities-valuationmodel proves useful as an alternativemethod of forecastinginvestment demand.

AggregateDemand and the 1973-74 Declinein the Stock Market

Previousstudies have soughtthe causesof the currentrecession in mone- taryand fiscalpolicy, the rise in worldenergy prices, and shortagesof food and raw materials.Certainly, the stock marketcannot be consideredas an additive,independent factor, because it reflectedmany of these other fac- tors. But to what extent did these depressiveinfluences on aggregatede- mand operatethrough the stock market?Although some questionswere raisedin earliersections of this paper about the estimatedsize of stock- marketeffects within the MPSmodel, it doesprovide an orderof magnitude for evaluatingthe importanceof the effectson consumptionand investment expenditures. The resultsof a simulationof the MPSmodel in whichthe dividend-price ratio was held at its late 1972value are reportedin table4. This simulation measuresnot only the directeffects of the declinein stock pricesbut also the secondarymultiplier effects. The total depressiveeffect on gross na- tionalproduct is verymodest throughout most of 1973,but it rapidlybuilds to $28 billion(1958 prices) by the firstquarter of 1975.Most of the direct effectsare concentratedin consumerexpenditures, which account for half of the total declineand whichin turn accountfor much of the drop in in- vestmentexpenditures as a secondaryeffect. The equipmentequation has long lags on changesin relativeprices so thatless thanhalf of the reduction is due to directeffects of the stock market. In orderto placethe simulateddecline in someperspective, the finalcol- umn of table 4 shows the total reductionin GNP from a level consistent with the same ratio of actual to potentialGNP that was achievedin the fourthquarter of 1972.Thus, the MPS model impliesthat approximately one-fourthof the depressiveeffects on the economyover the 1973-74period operatedthrough the stock market. On balance,the resultsof the first section of this paperimply that the 290 Brookings Papers on Economic Activity, 2:1975 Table4. SimulatedEffects on AggregateDemand of Stock-Market Decline,Quarterly, 1973-1975 :1 Annual rates in billions of 1958 dollars Predictedchange in demand Year Producers' Non- Total gross Total and Consumption durable residential national recession quarter expenditures equipment construction product gapa 1973:1 0.0 0.0 0.0 0.0 10.5 2 -0.2 0.0 0.0 -0.3 7.0 3 -1.3 -0.1 -0.1 -2.0 2.2 4 -3.3 -0.3 -0.2 -5.3 -1.1 1974:1 -5.8 -1.0 -0.6 -9.8 -24.6 2 -8.6 -2.0 -1.2 -15.1 -36.5 3 -11.3 -3.3 -2.1 -19.9 -48.9 4 -13.7 -4.9 -3.4 -24.4 -76.7 1975:1 -14.9 -6.4 -4.9 -28.1 -109.4 Source: Simulation results of the MPS model with dividend-priceratio held at the value for 1972:4. The values in the table indicate the estimated decline in demand due to the drop in the dividend-priceratio below the value for 1972:4. a. Reduction in GNP from the 1972:4 level, maintaining a constant ratio of actual to potential GNP. total effecton consumptionof stock-marketchanges may be overestimated becausethe effectis constrainedin the model to equal that of changesin otherforms of wealth.But the estimatedeffects on investmentdo not seem to be excessive.Thus, the significantsize of the simulatedimpact on GNP offers a strong justificationfor furtherattempts to integratethe stock marketmore fully into models of the aggregateeconomy. Comments andDiscussion

Saul Hymans:One of Bosworth'sfindings perplexes me, and I would like to focus on it. Assumingthat the stock marketaffects aggregate consumer demandby a wealtheffect, I would expect it to operateprimarily on the purchaseof durables.Thus, an increasein stock-marketwealth should per- mit a narrowingof the discrepancybetween the actual and the "perma- nent,"or "desired,"flow of consumptionservices from durable goods. But the impactdoes not workthrough durables, either in Bosworth'sown equa- tions or in his disaggregationof the MPS equation.Rather, it operates throughnondurables and services,and that strikesme as a paradoxical result. Bosworthmay provide a clue to the paradoxwhen he points out that the stock marketis significantonly if post-1965data are includedin the equa- tion. I wonderwhether, in fact,it is mainlythe 1968and 1969data that give the stockmarket its explanatorypower. According to Okun'spost mortem of the tax surcharge(BPEA, 1:1971), nondurables and servicesmay have overadjustedto the surchargein 1968-69. In this period stock pricesfell, and hence help to explainnondurables consumption in 1968-69. Further, the paper shows that, when 1973-74 data are included,the stock-market movementhelps even more to accountfor the weaknessof nondurablesand servicesin that period,although it may enlargethe errorsin earlierperiods. These instancesillustrate a generaltendency that I discussedin my first paperfor BPEA,1:1970: wealth variables can be used very successfullyas devicesto shiftthe big errorsof consumptionequations from one periodto another. On the otherhand, according to Okun'sstudy, auto demandmoved per- verselyin the period of the tax surcharge.Thus, if the stock markethelps the nondurablesequation because of the 1968-69episode, it almostsurely 291 292 BrookingsPapers on EconomicActivity, 2:1975 has to be a detrimentwhen includedin the durablesequation. The pecu- liaritiesof 1968-69may thus explainthe paradox. As Bosworthpoints out and as I indicatedin my paperon durablesand consumersentiment (BPEA, 2:1970), the stock-marketvariable may oper- ate throughits effecton consumersentiment or its effecton wealth.If it is operatingas a sentimentmeasure, then it is not terriblysurprising that the durablesequation is not aided by the stock-marketvariable, because that equationcontains the unemploymentrate, the rate of inflation,residential building,income changes, and the like. Thoseare apparentlysufficient indi- catorsof the consumer'sframe of mind; even the Juster-Wachtelfilter got filteredout of the equationwhen they were includedin it. In my 1970 paper I found that stock prices helped explain consumer- durablespurchases, but only with a marginallysignificant coefficient and onlyin the absenceof a measureof consumersentiment. When such a mea- sure was included,especially in a filteredform, the stock-pricevariable becametotally insignificant. It was on that basisthat I conjecturedin 1970 that the stock-pricevariable was servingas an incompletemeasure of con- sumersentiment, rather than playinga wealthrole, in the durablesequa- tion. Bosworthhas repeatedthat experimentby includingthe majordeter- minantsof the consumersentiment index rather than the indexitself; when he puts in the index as well as those determinants,it is insignificant. Bosworthcannot eliminateour empiricalignorance, but he can explain its existenceby invokingthe offsettingincome and substitutioneffects that follow interest-ratechanges, according to the neoclassicaltheory of con- sumerdemand. Given that reasoning,some interestrates must be included if the equationsare to operatethe way neoclassicaldemand theory says they should.Yet he can find no role for interestrates in the finalconsump- tion equations.That may be a serious omission, and interestrates may flunkthe statisticaltests becausetheir influence may be reflectedimplicitly throughother variables in the consumptionequation. The durablesequa- tion has a residential-buildingvariable that will stronglyreflect differentials betweenlong and short interestrates. The nondurablesequation includes wealthin formsother than corporatestocks, a majorcomponent of which is liquidassets, which reflects the cash-balanceratio and henceis movedby interestrates. Therefore,interest rates are indirectlyin the equationsand are intermingledwith the wealthvariables. Withall these pitfallsin mind, Bosworthrealized that the only way the complexbehavioral hypotheses could really be testedis withmasses of solid Barry Bosworth 293 microeconomicdata. Unfortunately,the data he had were of small mass and not too solid, and so they could not deliverunambiguous verdicts. In the discussionof businessfixed investment, I was astoundedby table3, which contains the predictionerrors of that componentfor alternative equations,with and without adjustmentsfor autocorrelation.The MPS equation,fitted only throughmid-1968, appears to explainbusiness fixed investmentfrom mid-1972 through mid-1974 with a degreeof precisionthat is usuallyreserved for forecastsof corporatedividends, capital consump- tion allowances,and personaltransfers to foreigners.Should I believethat the MPSmodel has reallyunlocked all the mysteriesof capitalspending? In any case, the lags in the equationare so long that the stock-marketeffects on businessfixed investmentin 1973and 1974are minimal. BecauseI believethat the main disputesabout stock-marketeffects on both consumptionand investmentremain unresolved despite Bosworth's diligentand capableefforts, I am skepticalof the final calculation,which simulatesthe MPS model to concludethat the stock marketaccounts for about 25 percentof the recentrecession.

FrancoModigliani: Barry Bosworth has providedus with a veryuseful re- view of presentknowledge concerning the impact of the stock marketon aggregatedemand, at leastin the UnitedStates. I willconcentrate my atten- tion on the portionof his paperdealing with the more controversialeffect via consumerexpenditure. First, a good deal of evidenceboth from time-seriesand cross-sectional studiesnow confirmsan importanteffect of wealthon consumption.Bos- worthhas limitedhimself to the Americanevidence, but similarevidence is accumulatingfor othercountries.' However,the centralissue for Bosworthis the specificinfluence of the part of wealththat is embodiedin corporatestocks. Both becauseit is sub- ject to largefluctuations through capital gains and losses, and becauseits ownershiptends to be highly concentrated,this portion could have a smallereffect on consumption.When the consumptionfunction of the MPS model was estimatedusing data for the years 1954-70,it was found that stock-marketwealth had prettymuch the same final effect on con- sumptionas the rest of wealth.In the finalestimation the coefficientsof the 1. Some of this is reviewed in my article, "The Life Cycle Hypothesis of Saving TwentyYears Later," in MichaelParkin and A. R. Nobay, eds., ContemporaryIssues in Economics(Manchester University Press, 1975), pp. 2-35. 294 BrookingsPapers on EconomicActivity, 2:1975 two componentsof wealthwere constrained to be equalsince this equality is impliedby the standardversion of the life-cyclehypothesis (although differentcoefficients would be consistentwith the generalizedversion allow- ing for a bequestmotive). However,other wealth exertsits effect imme- diately,whereas the estimatedeffect of stock wealth is spread over two years,implying that capital gains or losses are only graduallyrecognized and incorporatedinto perceivedwealth. Bosworth'sreestimation of the equationthrough 1972 yields a stock-marketcoefficient only moderately lower than that for other wealth.Furthermore, table 2 shows that when the MPS and the reestimatedBosworth equations are extrapolatedto 1973 and 1974,the MPS, with the constrainedcoefficient, performs better. Nonetheless,Bosworth remains skeptical about the effect of the stock marketon consumptionprimarily because, despite his attempts,he was unableto find much evidenceof an effectof capitalgains on the purchase of consumerdurables. He suggeststhat if wealth-especially capitalgains -has any effectat all on consumerexpenditure, it must surelyhave it on expenditurefor durables.Hence, he concludesby questioningthe credi- bility of the "resultobtained in these equationsof a positive wealth and stock-marketeffect on nondurablesand servicesand no effectof eitheron durablespurchases." This conclusionis not warranted.In the first place, Bosworth'sresults are not altogetherinconsistent with the model underlyingthe MPS equa- tion for durablesexpenditure which, he acknowledges,is "not dissimilar from that of other studies."That model does not imply a "directrole for eitherwealth or fluctuationsin the stock market."It impliesonly an indi- recteffect, in that expendituredepends on the gap betweenthe desiredand the laggedstock, the desiredstock dependson life resourcesapproximated by consumptionand, finally,consumption depends on wealth.But this in- directeffect is reallynegligible, since the coefficientof consumptionin the durablesequation can be put at around0.15, and the coefficientof wealth on consumptionat but 0.05. Thus,if consumptionis eliminated,as in Bos- worth'sequation in table 1, column(4), the coefficientof wealthshould be 0.15 X 0.05, or less than0.01, and the coefficientsof the distributedlag for capitalgains should be but a fractionof this figure.Thus, the poor results obtainedfor wealthin table 1 are not seriouslyinconsistent with the impli- cationsof the MPSmodel, especially allowing for the highmulticollinearity that clearlyplagues the coefficientestimates in column(4). Bosworth'ssuggestion to the contraryseems to rest on the consideration BarryBosworth 295 that the MPS formulation,as well as many othermodels, hypothesize and find a separateeffect on durablespurchases coming from some measureof transitoryincome. This result is usually attributedto the fact that most transitoryincome will not be spent on currentconsumption, but instead will be saved, and that durablesmay be a favoriteform of investmentof this transitorysaving, or windfalladdition to wealth. Since capital gains also representa windfall,should they not have a similareffect on durables purchases?I can see threereasons for castingdoubt on this conclusion. In the firstplace, whilethe savingout of transitoryincome will typically be in the form of money,which can be investedin any asset, capitalgains will accrueand be embodiedin some specificasset: to investthem in dura- bles wouldrequire liquidation of someassets and shiftsinto others.Second, the interpretationof the strong effect of transientincome on durablesas evidencethat windfalladditions to wealthtend to be investedin durablesto an unusualextent is open to some question.The fact that a variablelike unemploymentcan effectivelytake the placeof transitoryincome2 suggests an alternativeexplanation in termsof cyclicalvariations in consumers'con- fidence and perceptionof .Finally, even if there were some tendencyfor additionsto wealthfrom transitoryincome to stimulatepur- chases of durables,it may not operate,at least to any comparableextent, for additionsresulting from capitalgains. The reasonis that transitoryin- come will tendto be distributedfar more widely than capital gains from the stockmarket; in particular,a good portionof transitoryincome will accrue to youngerhouseholds, which typicallyinvest most of their net worth in durablesand would own a yet largerstock except for their lack of net worth.For thesepeople, an additionto net worth,whether windfall or not, is likelyto be investedlargely in durables.On the otherhand, the stock of durablesof those who receivethe bulk of capital gains is unlikelyto be limitedby net worth.Hence, one wouldnot expectthem to respondby in- vestingin durablesexcept to the extent that their permanentincome has been lifted-which is essentiallythe rathersmall effect discussedearlier. Theseconsiderations do not supportthe viewthat wealthor capitalgains shouldplay a major role in explainingpurchases of durables.It remainsdis- appointingthat Bosworthfound no role in his time-seriesanalysis and only a verylimited one in his cross-sectionresults. On the otherhand, Frederic 2. See, for example,F. ThomasJuster and Paul Wachtel,"Anticipatory and Objective Models of Durable Goods Demand," AmericanEconomic Review, vol. 62 (),pp. 564-79. 296 BrookingsPapers on EconomicActivity, 2:1975 Mishkin,in a yet unpublishedpaper3 in whichhe allows for the effect of defaultrisk on the desiredstock of durables,finds a very strongpositive effectof (financial)assets and corporatestock in particular,although only when he simultaneouslyallows for the negativeeffect of consumerdebt. In conclusion,I suggestthat Bosworthboth considerablyoverstates the case for a strongeffect of wealthon durablesand probablyunderstates the actualeffect. On both grounds,his lack of successin establishingan em- piricallysignificant effect of wealth on durablesdoes not seem to justify rejectingor seriouslyquestioning the substantialevidence on the effect of wealthon consumption. The secondmajor issue raisedby Bosworthis that "one cannot infer an interest-rateeffect on consumptionfrom evidenceof correlationbetween suchexpenditure and changesin stock values."Stated more operationally, to estimatethe effectof a risein interestrates on consumptionas the decline in the marketvalue of wealthdue to the risein capitalizationrates (holding the anticipatedprofit flow constant)multiplied by the coefficientof wealth, as is typicallydone-in simulationsof the MPS, for example-is unwar- ranted,and apt greatlyto overestimatethe negativeeffect. This is an im- portantpoint, and as I see it, logicallyvalid, even though Bosworthmay exaggerateits empiricalimportance. Unfortunately, it is not easy to clarify the issuesin a few lines.Briefly, the life-cyclehypothesis implies that, in the neighborhoodof a steadygrowth path, consumptionis linearand homoge- neousin permanentlabor income (YL) and wealth(A), but withcoefficients dependingon the long-runreal rate of returnor capitalizationrate (r). If the coefficientof the wealthcomponent is approximatedby a linearfunc- tion of r, then one can write

(1) C = aYL +? A +? rA. Note that in the last term, rA is simplyexpected permanent property in- come. Also ,u can be shown to be closely (and negatively)related to the strengthof the substitutioneffect. In this sense, the life-cyclehypothesis does allow for interest-rateeffects which operate through time preference. In the MPS it is furtherassumed that, to a first approximation,,ut a, so that the last term can be lumpedwith the first and becomespermanent income.This assumptionis a strongone, but is extremelyconvenient, as it avoidsthe necessityof allocatingtaxes between labor and propertyincome

3. "Illiquidity,Consumer Durable Expenditureand Monetary Policy" (Massachu- setts Instituteof Technology,; processed). BarryBosworth 297 and is at least vaguely consistentwith other evidence(although further work is in progressto test it directly). Now supposea one-timechange occurs in the capitalizationrate, r: can its effectbe inferredfrom (1)? Unfortunately, the answeris verydifferent in termsof short-and long-runeffects. Given time enough, so that all existing assetsare held by peoplewho saved on the basis of the new r, equation(1) will hold again,and at the sametime the value of assetswill have changed to a new steady-statepath.4 One can show that a rise in r will raise the assetpath and certainlyincrease consumption, although it may eitherraise or lower the savingratio. But the short-run,impact, effect is radicallydif- ferent,and Bosworth is correctin assertingthat it cannotbe readilyinferred fromthe coefficientsof (1). Given an unchangedexpectation of the stream of returnsfrom existing (final) assets of society,the risein r will producea fall in A. However,this need not producean unfavorable"income effect" becausethe presentvalue of futureplanned consumption, against which those assetswere held, will also decline.The outcomefor an individualwill depend on how well his portfoliois hedged:a retiredperson with assets promisinga streamexactly matching his consumptionplan wouldsuffer no income effect; a nonretiredperson whose unhedgedconsumption was to come from later savingwould have a favorableincome effect. For society as a whole,the net outcomewould depend on whetherthe declinein A plus the declinein the presentvalue of expectedincome exceeded the reduction in the presentvalue of consumptionplanned by those presentat the time. I do not at this time know of any way to assess this net outcome. My hunchis that, on balance,the incomeeffect is likelyto be negative,in part becausemany assetsare long-livedrelative to consumptionhorizons, and in partbecause wealth holders may not attributethe declinein wealthfully to the change in r, and hence may fail to discount future consumption appropriately.The problem is furthercomplicated by the effect that a changein interestrates may have on the riskpremium intervening between real long-terminterest rates and capitalizationrates for equity.But even if the overalleffect (including the substitutioneffect) is negative,it is most likely to fall short of 8. On the other hand, the wealth coefficientof the MPS consumptionfunction was estimatedover a periodin whichsome of the changesin A werethe resultof changesin r; to this extent,that coeffi- cient is likely also to underestimatethe parameter8 of (1), thoughit may still overestimatethe true effectof a changein r.

4. See the results in "Life Cycle Hypothesisof Saving." 298 Brookings Papers on Economic Activity, 2:1975 In the light of Bosworth'scriticism and the above analysis,I wouldcon- cludethat (1) furtherwork needs to be done in disentanglingthe short-run and long-runeffects of changesin r; and (2) in the meantime,users of the MPS model should be aware that estimatesof the impact of monetary policy throughconsumption must be taken with extra caution, and are likelyto be biasedupward, in partalso becauseof relatedobjections raised by Tobin and Dolde in the work cited by Bosworth.

BarryBosworth: I stand somewherebetween Hymans and Modiglianiin theirreactions to the absenceof a measuredstock-market effect on durables consumption.I cannot fully accept Hymans'demonstration that it is an overwhelmingpuzzle, nor can I acceptModigliani's assurances that it is no puzzleat all. After devotinga lot of time tryingto includeinterest rates in these equationsand taking out some of the other variablesthat might be correlatedwith interestrates, I am convincedthat the answerdoes not lie in interest-rateeffects or liquidityeffects, as Hymansbelieves. Hymans' argumentabout the role of the consumersentiment index makessense for the aggregatetime series.But the Michiganpanel study providedan attitudeindex for eachperson; following the individualhouse- hold throughtime, I can find no correlationbetween attitudes and either purchasesof durablesor saving.The cross-sectionresults cannot be dis- missedon the basis that the stock marketreflects attitudes. The fact that people own widelyvarying amounts of stock and thus experiencesharply differentamounts of capitalgains should be enoughto breakthe correla- tion with overallattitudes. But thenmy own resultsfrom cross-section data weremuch weakerthan those of Friendand Lieberman. Finally,I wouldlike to reiteratemy conclusionthat it is inappropriateto use the MPSmodel to measurethe impactof monetarypolicy on consump- tion. Modiglianiseems to agree with my conclusion,but presentssome argumentsfor believingthat the true effect may not be sharplydifferent fromthe MPS estimate.

GeneralDiscussion

A numberof participantsexpressed reservations about Franco Modi- gliani'sexplanation of the lack of a significantstock-market effect upon durablesconsumption. Thomas Juster argued that uncertaintytended both Barry Bosworth 299 to shortenthe horizonsof decisionsand to encourageportfolio diversifica- tion. He and ArthurOkun felt that, throughthe first effect,there was as muchreason to expecttransitory wealth effects on durablesas to expectthe effectsfrom transitory cyclical changes in income that make durablesso cyclicallysensitive. Nor could Modigliani'scase rest on low income elas- ticitiesfor durables,asserted both Okun and MartinFeldstein: cross-sec- tion studiesreveal income elasticities of durablesdemand at least as high as those of nondurables.James Pierce joined Justerin stressingthe port- folio aspectof gainsor losses in stock-marketwealth, through which hold- ings of durableconsumer goods should be expectedto respond. Pierce suggestedthat relativelyminor speedupsor slowdownsof purchasesof autos and householddurables could show up as substantialeffects. Modiglianireiterated his expectationof some stock-marketeffect on durablesbut was not surprisedthat it was smallenough to be missedby the equation.One possibleexplanation for the empiricalpuzzle might lie in the distributionof capital gains from the stock market: they accrueto wealthy, older individuals,who are not likely to accumulatesubstantial amountsof durables.In this case, remarkedR. A. Gordon, it would be usefulto disaggregatethe durablesequation by incomeand age class. Feldstein,John Shoven,R. A. Gordon,and Justersuggested ways that consumersentiment and expectationsmight be handledmore effectively. Feldsteinwas concernedabout the simultaneous-equationbias that might arise in regressionsfitted by ordinaryleast squares.Shoven elaborated on that point, questioningwhether movements in consumptionand invest- mentcould be validlyattributed to the stockmarket when the stockmarket itself is strongly influencedby economic forces. Stock prices must be treatedas endogenousvariables, he concluded.Juster also suggestedexperi- mentationwith longerlags on the index of consumersentiment and with other measuresof consumersentiment, such as the varianceof expected pricechanges. Finally, Juster saw a possibledefect in the regressionsarising from the calculationof wealthother than corporateequities. Much of that is the housingstock, whichnecessarily is almostperfectly correlated with imputedconsumption services from housing. By using that as an "inde- pendentvariable," Bosworth explains consumption by a variablethat in- cludesconsumption. R. A. Gordonurged the use of tests that allowedfor the possibilitythat fluctuations of the stockmarket influenced buying senti- ment indirectlyeven for those who had no stock-marketwealth. Severalparticipants commented, as Saul Hymanshad, on the apparent 300 Brookings Papers on Economic Activity, 2:1975 discontinuity of stock-market effects in different time periods. Lawrence Klein was struck by the 1962 experience, when consumption and GNP kept advancing despite a sharp decline in the stock market. Klein found a similar lack of correlation between the stock market and the real economy during the 1930s, although Jan Tinbergen had found a strong connection during the 1920s. Klein cautioned the panel that the Friend and Lieberman study, though very carefully executed, applied to the rising stock market of 1963 and did not necessarily depict a down-side experience like 1962. Bosworth and Modigliani agreed that the closer connection of the stock market and real activity after 1965 pointed out by Hymans could be explained by the dominance of monetary policy in the swings of stock-market prices since that time. William Nordhaus cited research evidence that the stock market had exerted an important independent influence on the 1929-32 collapse of the economy. Robert Solomon added that the widespread purchase of stock on margin in the 1920s and the subsequent margin calls of the 1930s may have sharpened the impact of the stock market on consumption. That phenomenon is unlikely to recur because of the of margin credit. Both Michael Wachter and R. A. Gordon emphasized the changing pat- terns of common-stock holdings in postwar America. An increasing amount of stock is held in pension funds and trusts. In order for some people to change their buying in response to stock-market fluctuations, they have to know what is happening to the value of their annuities and even then their responses might be different from what they would be if they were able to realize capital gains.