American Economic Association - Southern Methodist University
American Economic AssociationDoes Consumer Sentiment Forecast Household Spending? If So, Why?Author(s): Christopher D. Carroll, Jeffrey C. Fuhrer, David W. WilcoxSource: The American Economic Review, Vol. 84, No. 5 (Dec., 1994), pp. 1397-1408Published by: American Economic AssociationStable URL: http://www.jstor.org/stable/2117779Accessed: 06/10/2010 18:08Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available rms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained herCode aea.Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to TheAmerican Economic Review.http://www.jstor.org
Does Consumer Sentiment Forecast Household Spending?IfSo, Why?By CHRISTOPHER D. CARROLL, JEFFREY C. FUHRER,AND DAVID W. WILCOX*In the three months following the Iraqiinvasionof Kuwait,the Universityof Michigan's Index of Consumer Sentiment (ICS)fell an unprecedented24.3 index points, toits lowest level since the 1981-1982 recession.' This collapse in householdconfidencebecame the focus of a great deal of economic commentaryand, indeed, frequentlywas cited as an important-if not the leading-cause of the economic slowdownthatensued.Concern was fueled by the well-knowncontemporaneouscorrelation between theICS and the growthof household spending.Figure 1 shows quarterly averages of theindex, 1978-1993, together with the quarterly growth in real personal consumptionexpenditures as measured in the nationalincome accounts (Bureau of EconomicAnalysis).The correlationis impressive.Of course, it is not surprisingthat sentiment and the growthof spending are positively correlated.This correlationmay simply reflect that, when economic prospectsare poor, householdscurtail their spendingand also give gloomyresponsesto interviewers. Thus, the contemporaneouscorrelationbetween sentiment and spending does not*Carroll:Division of Research and Statistics,Stop80, Federal Reserve Board, Washington,DC 20551:Fuhrer:ResearchDepartment,Federal Reserve Bankof Boston, Boston, MA 02106: Wilcox: Division ofMonetaryAffairs, Stop 71, Federal Reserve Board,Washington,DC 20551. We have benefited from theresearchassistanceof StephenHelwigand ChristopherGeczy and the comments of an anonymousreferee.The views expressed in this paper are those of theauthors and not of the Federal Reserve Board, theFederal Reserve Bank of Boston, or the other members of the staffof either institution.IThe ConferenceBoard'sConsumerConfidenceIndex also plungedat the same time.1397refute traditional life-cycle or permanentincome models of consumption.Nor does itnecessarily make the job of forecastingchanges in consumption any easier. Fromthe point of view of an economicforecaster,the questions of interest are first, whetheran index of consumer sentiment has anypredictive power on its own for futurechanges in consumptionspending,and second, whether it contains informationaboutfuture changes in consumerspending asidefrom the information contained in otheravailableindicators.In Section I, we present evidencethat theanswerto the firstquestionis a clearyes: wefind that laggedvalues of the ICS, taken ontheir own, explain about 14 percent of thevariationin the growthof total real personalconsumption expenditures over the post1954period.Furtherinvestigationshowsthatthe answerto the second questionis probably yes as well, though here the margin isnarrowerand the evidencemore murky.TheICS contributesabout3 percentto the R2 ofa simple reduced-formequation for totalpersonal consumptionexpendituresin thelonger of the two sample periodswe examine, but nothingin the shortersampleperiod(thoughthe latterresultis heavilyinfluencedby the observationfor 1980:2).For the majorsubcategoriesof spending,the contributiongenerally ranges between 1 percent and 8percent. Overall, we read the evidence aspointingtowardat least some significantincrementalexplanatorypower.Therefore, we take as given for the remainder of the paper that sentiment forecasts spending,and we turn to the issue ofhow that statistical relationshipshould beinterpreted.One possible interpretationisthat sentiment is an independent drivingfactor in the economy, and that changes in
1398THE AMERICAN ECONOMIC REVIEWDECEMBER 1994Percent change, compound annual rateIndex points11510Consumer sentiment (left scale)9558;075&28 {l0;WPersonal consumption expenditures (right scale)35II19791981III 19831985II1987II1989IIII19911993FIGURE 1. CONSUMER SENTIMENT AND THE GROWTH IN CONSUMER SPENDINGsentiment not only forecast changes inspending,but also cause them. An alterna-tive interpretationis that sentiment forecasts spendingbecause it reflectsthe overalloutlook for the economy:when consumersare optimistic about the outlook for theeconomy, they give upbeat responses to interviewers.On average,those optimisticexpectations are substantiated,and spendingeventually increases as foreshadowed bysentiment.This alternative account of the role ofsentiment must involve some violation ofthe simplest certainty-equivalenceversionsof the life-cycleand permanent-incometheories; otherwise, current spending wouldfully reflect the optimism or pessimism ofhouseholds about future prospects for theeconomy,and sentimentwould have no predictive power for future changes in spending. In Section II, we introducesuch a violation using the mechanismproposedby JohnY. Campbelland N. GregoryMankiw(1989,1990, 1991). Specifically, Campbell andMankiw posit that some households arestrictlife-cyclerswhile others follow a "ruleof thumb" and set consumption equal toincome. In an economy containing thosetwo types of consumers, sentiment mightwell forecast spendingwithout being an in-dependent driving factor: when prospectsfor the real economy are bright, forwardlookinglife-cyclerswill give optimisticreadings on consumer sentiment. On average,their optimism will be borne out, and income will rise. When it does, spending ofrule-of-thumberswill increase.Thus, in thisaccount, the survey responses of forwardlooking householdspredict the spendingofrule-of-thumbhouseholds.The Campbell-Mankiwmodel is useful inthis context because it delivers a testableimplication-that sentiment should appearin the predictionequation for consumptiononly indirectly,in its role as a predictorofincome. Our objective in Section III is todevelop evidence on the validityof this explanationof the predictivecontent of sentiment.In answer to the first question posed inthe title, we conclude that consumersentiment does indeed forecastfuturechangesinhousehold spending. Whether sentimentshouldbe characterizedas helping"a little"or "a lot" is a more difficultquestion.Giventhe amountof attentionthat has been paidto these indexes recently, their predictiveabilityseems underwhelming.Fromthe perspective of the recent academic literatureon consumption, however, their perfor-
VOL. 84 NO. 5CARROLLETAL.: CONSUMER SENTIMENTAND SPENDINGmance is impressive because it stacks uprelativelywell against the track record ofother variables that previously have beennoted to have some predictive power forpersonalconsumptionexpenditures(e.g., interest rates, stock prices, the unemploymentrate).In answer to the second question, weconclude that the Campbell-Mankiwmodeldoes not provide an adequate explanationfor the predictive power of sentiment forchangesin householdspending.Part of thatpredictivepowerappearsto operatethrougha direct channel (or channels), rather thanthe indirectincome channel allowedfor under the rule-of-thumbaccount we outlinedabove. In the concludingsection, we speculate about other possible explanationsforthat predictivepower.I. SomeReduced-FormRegressionsA simple way to judge the near-termpredictive abilityof the ICS is to examine theR2,Sfrom regressionsof the growthof various measures of household spending onlaggedvalues of the ICS:NA log(C,) ao E iS-i1399period, lagged values of the ICS taken ontheir own explain about 14 percent of theone-quarter-aheadvariation in the growthof total real personalconsumptionexpenditures (PCE) (row 1). The probabilitythatthis explanatory power was generatedmerely by chance is estimated to be essentiallynil (row 1, numberin parentheses).Asshown in rows 2-4, sentiment taken alonehas the most explanatorypower for onequarter-aheadgrowthin PCE for goods excluding motor vehicles (17 percent of thevariationexplained)and the least explanatory power for PCE for motor vehicles(4 percent of the variationexplained).The second columnreportsresultsfor theperiod since 1978, during which time theSurvey Research Center has conducted itssurveyof consumerson a monthlybasis (seefootnote 2). Unfortunately,the results arerather sensitive to this change in sampleperiod. Using the post-1978 data, we findthat lagged values of the sentiment indexexplain only 5 percent of the variation inthe growth of real PCE, 2 percent of thevariationin PCE for services, and none ofthe variationin PCE for motor vehicles. Inthe case of goods excludingmotor vehicles,however,the R2 is even a bit higherthan it et.i 1Note that this procedureamountsto a testof Robert E. Hall's (1978) random-walkhypothesis;if the 8's are significantlydifferentfrom zero, that hypothesisis rejected.Table1 reports the results of implementingthisprocedure at the quarterlyfrequencyusingfour lags of the ICS.2 Over this sample2The left-hand-sidevariable in each regressionisthe log difference of the indicated category of realhousehold spending.The startingdate of the longersample period was chosen to exclude data from theKorean War era. The starting date of the shortersampleperiodwas chosento coincidewith the movebythe Universityof Michiganto administertheir Surveyof Consumerson a monthlybasis. (Priorto 1978, thesurveyusuallywas taken only once each quarter,andoccasionallynot even that frequently.For the periodsince 1978, the quarterlyobservationis defined to bethe averageof the monthlyobservations.For the period prior to 1978,the quarterlyobservationis definedto be the monthlyreadingif a surveywas taken duringthat quarter,and the readingfromthe precedingquarter if one was not.) We ended both sampleperiodsin1992:3to exclude a very dramaticfluctuationin wageand salaryincome in 1992:4and 1993:1reflectingtaxmotivated shifts of income (especially bonuses andcommissions)from 1993into 1992.We partitionPCE into three categories(motorvehicles, other goods, and services)ratherthan the moretraditionaltwo categories(durablesand nondurablesplus services)to better reflect the proceduresused bythe Bureau of Economic Analysis to estimate consumer spending. Briefly, PCE for motor vehicles isestimatedmainlyfrom data on the unit sales of newcars and trucks;PCE for other goods is derivedfromthe monthlyretail sales data; and PCE for servicesisestimatedfrom a varied collection of source data including, among others, employmentin service industries, the stock of occupiedhousingunits as estimatedin the CurrentPopulationSurvey,and paymentsforelectricityand naturalgas as reflectedin the monthlybillingsof utilities. See Wilcox(1992) for furtherdiscussion.The data from the nationalincome accountsreflect the annualrevisionreleased by the CommerceDepartmentin September 1993. RATS code and acompletedata set are availablefrom the authorsuponrequest.
THE AMERICAN ECONOMIC REVIEW1400DECEMBER 1994TABLE 1-REDUCED-FORMEVIDENCE: R2 'S AND INCREMENTALFROM SIMPLE PREDICTION EQUATIONSR2 s4A log(C,) a0 EfiS,ti 'YZt-I Eti lRowCategory ofreal PCE1Total2Motor vehicles3Goods excludingmotor vehiclesServices4Incremental R21955:1-1992:3 1978:1-1992:3 1955:1-1992:3 0)0.03(0.000)0.08(0.000)0.05(0.000)0.01(0.188)- 0.03(0.056)0.03(0.013)0.03(0.001)- 0.07(0.969)Notes: S,t- i i 1.4)are lagged values of the Index of Consumer Sentiment. Zt,is a vector of control variables. The regressions underlying the results reported in thefirst two columns used only the lagged values of sentiment as explanatory variables;the regressions underlying the results reported in the third and fourth columnsincluded the control variables. These controls included four lags of the growth in reallabor income (defined as wages and salaries plus transfers minus personal contributions for social insurance). The numbers in parentheses are p values of the jointsignificance of the lags of sentiment. Hypothesis tests were conducted using aheteroscedasticity- and serial-correlation-robust covariance matrix (allowing serialcorrelation at lags up to 4).was in the longer sample period. In eachcategoryother than motor vehicles, the coefficientson the lags of sentimentare jointlysignificantat the 3-percent level or better.Sensitivityto sample period notwithstanding, we interpretthese results as constituting reasonablystrongsupportfor the proposition that sentiment taken alone has somepredictive power for future changes inhouseholdspending.3We next investigate whether the sentiment index has any predictive ability onceone controls for informationcontained inother variables availableto economic forecasters. We implementthis investigationby3We also estimatedsimple predictionequationsusing lags 2-5 of sentiment,rather than lags 1-4. Theresultswere similarexcept for goods excludingmotorvehicles, in which case the R2 fell to 8.5 percent (pvalue 0.4 percent) in the longer sample period, and4.5 percent (p value 7.3 percent)in the shorterperiod.estimatingequationsof the followingform:NA log(C,) ao EjS,j yZIt-1 Eti l1where Zt is a vector of other variables.Ofcourse, the choice of which other variablesto include in the equation is inherentlysomewhat arbitrary.The third and fourthcolumns of Table 1 present results for aminimal specification of such other variables that includes four lags of the dependent variableand four lags of the growthofreal labor income, defined as wages andsalaries plus transfersminus personal contributionsfor social insurance,all deflatedby the implicitdeflatorfor total PCE. (Theuse of labor income, rather than total disposable income, is motivatedby our investigation in Sections II and III of the rule-ofthumb hypothesisas an explanationfor thepredictive power of sentiment and will bediscussed in greater detail there). In thesetwo columns, the upper entry in each cell
VOL. 84 NO. 5CARROLL ETAL.: CONSUMER SENTIMENTAND SPENDINGrecords the incrementto the adjusted R2providedby the lagged values of consumersentiment,while the lower entry (in parentheses) displaysthe p value from the test ofthe joint hypothesisthat the coefficientsonthe four lagged values of the sentiment index equal zero.Evidently,some-but not all-of the information in the ICS is held in commonwith the control variables.As the first rowof the third column shows, the ICS addsonly 3 percent to the explanatorypower ofthe equation for the growth of total realPCE over the longer sample period (5 percent if the observationsfor 1975:2 [SocialSecuritybonus and income-taxrebate] and1980:2 [credit controls] are omitted).Nonetheless, the coefficients on the fourlags of sentimentare estimatedto be statistically significantat better than the 0.1-percent level. A similarresult holds for goodsexcludingmotorvehicles,with sentimentaccounting for a smaller, but still statisticallysignificant, proportion of the variation inthe growth of spending in the presence ofthe controlvariables.In the case of services,sentimentadds only 1 percent to the R2 ofthe reduced-form equation, and the fourlags are not jointly significantat any of theusual levels. In the motor-vehiclescategory,a counterintuitiveresult holds:the contribution to the R2 from the sentimentvariablesis substantiallylargerwhen the controlvariables are included in the regression thanwhen they are not.The fourthcolumnin Table 1 repeats theexperimentusing the post-1978 data only,no doubt at substantial econometric riskgiven that 13 coefficients are being estimated in a sample of only 59 observations.If all observationsare retained in the sample, the inclusion of sentiment actuallyslightly reduces the A2 of the predictionequation for total PCE. However, if theobservationfor 1980:2is omitted (result notshown in the table), sentiment adds 4 percent to the R2, and the four lags are jointlysignificantat the 0.1-percentlevel. The results for services are dismal (a finding thatappearsto be robustto omission of variousobservations),but the results for motor vehicles and for goods excludingmotor vehi-1401cles are quite strong. For the latter twocategories of spending, the coefficients onthe lags of sentiment remainjointly significant even controllingfor our minimalspecificationof other known information.(If theobservationfor 1980:2is omitted, the k2,Sfor motorvehicles and goods excludingmotor vehicles rise to 6 percent and 7 percent,respectively.)4In sum, we read these results as showingthat sentimenttaken on its own has considerable predictive ability for various measures of householdspending.Further,sentiment likelyhas some (thoughprobablynot agreat deal) of incremental predictive powerrelativeto at least some other indicatorsforthe growthof spending.We turn now to theinterpretationof this finding.II. Interpretingthe Influenceof Sentiment:FrameworkThe Campbell-MankiwIn a series of recent papers, Campbelland Mankiw(1989, 1990, 1991)investigateasimple modificationof the pure life-cycle/permanent-incomehypothesis.They assumethat there are two types of consumers.One type sets spending strictly accordingto a standardlife-cycle/permanent-incomemodel; the other sets spending equal tocurrent income. In the simplest version oftheir model, the consumptiongood is completely nondurable,and the decision periodof consumerscoincidesexactlywith the frequencyof the data. In this case, the changein the consumption of life-cyclers is awhite-noise process (equivalently,the level4Whenwe reestimatethe equationsunderlyingtheresults displayedin the third and fourth columns ofTable 1 laggingall right-hand-sidevariablesan additional period,the resultsare considerablyweaker.Forexample,over the longersampleperiod,the sentimentvariables subtract 0.4 percent from the R2 of theequationfor total PCE ratherthan adding3 percent.Over the shorter period, inclusionof sentiment subtracts from the R2 of the predictionequationfor allfour categoriesof spending.We interpretthese resultsas illustratingthe importanceof retaininguse of explanatoryvariablesat the firstlag, as we do in SectionIII.
1402THE AMERICAN ECONOMIC REVIEWfollows a randomwalk):ACLtEtDECEMBER 1994equation(1) would be modifiedas follows:(2)ACt A}Yt utut 1 MA(1).Because ut is seriallycorrelated,it neednot be orthogonalto variablesdated t -1.Campbell and Mankiw (1989, 1990, 1991)addressthis problemby laggingtheir instruments an extra period (so that all instruments are dated t -2 or before) and bycorrectingtheir test statisticsfor serial correlation in the residuals. They find that,AC AYyReven with the instrumentsdated t -2 orbefore, they have enough power to rejectA crucial assumption in the Campbell- t
American Economic Association Does Consumer Sentiment Forecast Household Spending? If So, Why? Author(s): Christopher D. Carroll, Jeffrey C. Fuhrer, David W. Wilcox Source: The American Economic Review, Vol. 84, No. 5 (Dec., 1994), pp. 1397-1408 Published by: American Economic Association
The United Methodist Publishing House, 2009. The Book of Discipline of the United Methodist Church 2008. Nashville, TN: The United Methodist Publishing House, 2008. The United Methodist Hymnal. Nashville, TN: The United Methodist Publishing House, 1989. The United Methodist Book of Worship. Nashville, TN: Abingdon Press, January, 1996.
American Economic Association U.S. Economic Growth since 1870: One Big Wave? Author(s): Robert J. Gordon Reviewed work(s): Source: The American Economic Review, Vol. 89, No. 2, Papers and Proceedings of the One Hundred Eleventh Annual Meeting of the American Economic Association (May, 1999), pp. 123-128 Published by: American Economic Association
Saint Louis Regional Chamber Alan and Mary Stamborski Herb Standing Gina Stone Richard and Beverly Straub . First United Methodist Church of Odessa Harry and Arden Fisher Florence United Methodist Church United Methodist Women Mary and Nestor Fox . Trinity United Methodist Church of Piedmont, United Methodist Women Jane Tucker William and .
METHODIST CONFERENCE 2003 REPORT Holy Communion in the Methodist Church ‘His presence makes the feast’ CONTENTS Paragraphs 1 - 12 A Summary and conclusions 13 -18 B Introduction 19 -23 C Four ‘snapshots’ of Methodist Communion services D A survey of current practice and beliefs in the Methodist Church 24 - 28 (i) Background
Tennessee — Methodist University Hospital, Methodist North Hospital, Methodist South Hospital, Methodist Germantown Le Bonheur Children's Hospital Type: Finance Facility: System (Replacing S-01-042 and S-01-043) Purpose: The purpose of this policy and the Medical Financial Assistance programs established
The Free Methodist denomination would continue to expand across the U.S. and beyond as Free Methodist missionaries felt called to spread the good news of the gospel overseas. Still today, Free Methodist missionaries travel around the world to encourage thousands of Free Methodist pastors, leaders, and churches around the world!
American Economic Association The Invisible Hand and the Grabbing Hand Author(s): Timothy Frye and Andrei Shleifer Source: The American Economic Review, Vol. 87, No. 2, Papers and Proceedings of the Hundred and Fourth Annual Meeting of the American Economic Association (May, 1997), pp. 354-358 Published by: American Economic Association
Refer to API RP 500 and NFPA 70 for guidance. When loading liquids that can accumulate static charges, refer to the precautions described in the International Safety Guide for Oil Tankers and Terminals, Safety of Life at Sea, API MPMS Ch. 3, and API RP 2003. Care must be taken with all liquid-in-glass thermometers to prevent breakage, which will result in a safety hazard. If the liquid in the .
