Housing, Consumption, and Credit Constraints

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Housing, Consumption, and Credit Constraints Andreas Lehnert 2004-63 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Housing, Consumption, and Credit Constraints (cid:3) Andreas Lehnert Board ofGovernorsoftheFederal Reserve System Washington,DC20551 (202)452-3325 Andreas.Lehnert@FRB.GOV First Version: Sep. 2002 Last Revised: September29,2004 (cid:3) IthankJoshuaGallin,DarrelCohen,andseminarparticipantsatseveralinstitutionsforvaluablesuggestions. MaryF.DiCarlantonioforexcellentresearchassistance. Anyremainingerrors aremyown. Theopinions,analysisandconclusionsinthispaperaremyownanddonotnecessarilyrepresentthoseoftheBoardofGovernorsoftheFederalReserveSystemoritsstaff.

Housing, Consumption, and Credit Constraints Abstract I test the credit-market effects of housing wealth shocks by estimating the consumptionelasticityofhousepriceshocksamonghouseholdsindifferentagequintiles. Younger households face faster expected income growth and hence would like to borrow more than older households. I estimate consumption elasticities from housing wealth by age quintileto be f 4 ; 0 ; 3 ; 8 ; 3 g percent. As predicted by theory, the youngest group has a higher elasticity of consumption than the next two age quintiles. That the consumption of the age quintile on the verge of retirement is responsive to housing wealth is also not surprising: I show that these households are likeliest to “downsize” their house and thus realize any capital gains. Journalof EconomicLiteratureclassificationnumbers: D12,D91,E21, G21 Keywords: Consumption,wealtheffect, housing,houseprices

1 Introduction In the U.S., housing is an important part of the typical American’s balance sheet. Moreover, house values have continued to climb even as the stock market has stagnated. Thus, it is particularly important to understand the effect of housing wealthon consumption. Housingwealthcanaffectconsumptionthroughmanychannels. Forexample, homeowners can move to cheaper quarters and realize some of the equity from their previous dwelling or they can assume added debt backed by the wealth of their house. Moreover, households may view housing wealth more as a buffer stock of wealth to be used in an emergency or to finance a specific expenditure. Typically, however, most households are not experiencing these sorts of contingencies, that is, they are not movinginto cheaper houses, experiencing economic stress, or paying a child’s college tuition. How then do households react to quotidian increases or decreases in their housing wealth; that is, to housing wealth shocksthatarrivein thenormalcourseoftheirlives? Further,evenifsuchquotidianincreasesinhousingwealthdoaffectconsumption, it is far from clear that this can translate into increased aggregate consumption. As a matter of basic accounting, increases in house prices are exactly offset byanincreaseintheusercostofhousing. Fromthisaccountingview,houseprices can rise indefinitely without affecting the aggregate consumption of non-housing goodsand services. Indeed, inasimplemodelwithafixed supplyofhousingand increasing labor productivity, house prices must rise indefinitely. Without some 1

kind of frictions, in such a model housing wealth shocksper se will be irrelevant to aggregate macroeconomic conditions.1 Yet to claim that housing wealth does not matter at all for aggregate consumption flies in the face of intuition, not to mentionreceived wisdom. Housing wealth could affect aggregate consumption in its role as collateral; that is, agents who own (and have equity in) their homes can more easily pledge to repay loans, hence overcoming commitment problems in credit markets. This idea is exploited by Lustig and Van Nieuwerburgh (2002), who study the effect of rising house values on stock market returns. In this view, increases in housing wealth can cause decreases in theaggregatesavings ratebecause oftheincreased collateralvalueofhousing,anditsassociatedrelaxationofborrowingconstraints. This “credit channel” of housing wealth can have a further effect if one believes that financial developments over the past decade have made borrowing against home equity easier and cheaper. Thus, as a result of financial innovation, wecouldbeinthetransitionfromanoldequilibrium,inwhichhousingwealthdid not matter much, to a new equilibrium in which housing wealth directly relieves borrowing constraints. This hypothesis, however, awaits formal econometric evidence. If housing wealth relaxes borrowing constraints, we would expect its effects to appear first in households’ precautionary saving. Carroll, Dynan, and Krane (1999) find that, among households with higher labor income risk, precautionary 1See Lehnert and Pence (2003) for one such model that emphasizes the role for foreclosure lawonhousingfinancemarketsandhouseprices. 2

savings was higher, but that most of this increase was in the form of housing wealth. One explanation for this could be that housing wealth has become easier totap intheeventofafamilyemergency.2 Another, thoughrelated, explanationis that housingwealthenjoys special tax status and bankruptcy court status. The treatment of home equity in bankruptcy proceedings is relevant to households considering their optimal plan following severeshocks,such as extendedillnessorjobloss.3 Hurst and Stafford (2002) use the PSID to show that mortgage refinancing playsamajorroleinconsumptioninsurance. Asrefinancingcostsfall,thebufferstock role of home equity would presumably be greater. That is, homeowners today may bemoreconfident in theirabilitytoborrow againsttheirhousingwealth (e.g. by refinancing) than in the past. As a result their savings in other, more liquid, assets may decline. The volume of refinancing, in particular the quantity of so-called “cash out” refinancing, is a measure of how actively households are reallocatingtheirportfoliosortappingtheiraccumulated homeequityforaspecific purpose.4 In a refinancing, households are pursuing an active means of realizing their home equity gains. However, households may also react passively to home 2Formoreonthelargebuffer-stocksavingsliterature,seeCarroll(2000,1994,1992,1996,and 1997a,b),CarrollandSamwick(1998),Hubbard,Skinner,andZeldes(1994),andEngen(1993). 3LehnertandMaki(2002)studytheeffectofbankruptcylawonhouseholdconsumption;they findthat–especiallyamonghomeowners–bankruptcylawaffectstheconsumptionresponsetoincomeshocks. 4Thereisnogenerallyagreed-upondefinitionof“cashout”refinancing.However,mostindustryexpertsagreethatitinvolvesanincreasein theoutstandingmortgageprincipalbymorethan 2.5percent. Informally,thehomeownerwalksawayfromtherefinancingwithacheck,whichhe or she then puts into another form of asset (portfolioreallocation)or uses to purchase goodsor services(specificconsumptionpurpose). 3

equity gains. This paper is an attempt to pin down the nature and importance of thispassivechannel. The relationship of house prices to consumption has been an active area of research interest for more than a decade. Several studies have addressed the effect of house price increases on the consumption of young renters who are, one presumes, saving for housedownpayments. Among theseare Sheiner (1995) and Engelhardt (1994). Under one view, renters should increase saving in the face of house price increases in order to better afford a down payment; under the other view, renters give up trying to buy a house and succumb to – in the memorable phrase–“theconsumptionofdespair.” This paper follows the seminal study by Skinner (1996), who also uses the PSID to test for housing wealth effects on consumption. He finds that wealth effects are greatest among young households, and that these younger households decrease theiractivesavingsinresponsetohousingwealth shocks. These studies addressed the effects of housing wealth shocks on individual households. Fewer studies have tackled the problem of estimating the effects of housingwealthshocksonaggregateconsumption. Studiesusingdataonaggregate U.S.consumptiontendtohaveahardtimeseparatingtheeffectofhousingwealth shocksfromstockpriceandotherwealthshocks. Case,QuigleyandShiller(2001) use the extra variation provided by U.S. states, as well as a cross-country panel, to identify theseparate effects of stock and housingwealth. Theauthors find that themarginal propensitytoconsume(MPC) outofhousingwealth farexceeds the 4

MPCoutofstock-marketwealth.5 In this paper, I test whether the consumption impact of house price shocks is greater among credit constrained households than among other households. I identify credit constrained households by age. Young households face steeper incomeincreases than older households, and hence a greater incentiveto borrow. Conversely,theyface adecreased desireforsaving. Specifically,Iusehousehold-leveldatafromthe1968-1993wavesofthePanel Study of Income Dynamics (PSID) to address this question.6 In this study, I exclude households that experienced severe distress (such as divorce or death of thehead)inthepreviousyear;thatchangedcomposition(suchasaddingorlosing a child at home); and, perhaps mostimportantly,that moved. These are precisely those households that might be expected to consume a large portion of any accumulated housing wealth. Instead of studying these households, I study a sample ofhouseholdsthatare stableover(at least)thecurrent and previousyear. AmongthissampleofstablehomeownersIfindanaverageconsumptionelasticity out of housing wealth gains of between 0.04 and 0.05. These elasticities translateintomarginal propensitiesto consumeofbetween 2 and 3 cents perdollar. When Isplitmysampleintogroupsby ageI find strikingdifferences in these estimated effects. The effect of housing wealth on consumption is greatest for 5Case,Quigley,andShiller’sfindingthathousingwealthaffectsconsumptionmorethanstockmarketwealthmaybepreytoPoterba’s(1991)observationthathousepricescapitalizeexpected future gains in labor income. Thus, households in a region receive good economic news and simultaneouslyincreasetheirnon-durableconsumptionwhilebiddinguplocalhouseprices. 6ThesewavesconstitutethecurrentfinalreleaseofthePSID;datafrom1994–1999areavailable only in “early release” form. Early release data do not contain accuracy codes or sample weights. 5

households in late middle age (ages 52 through 62); however, the next-most affected age group are the youngest households. Other age groups show smaller effects. The fact that younger households react so strongly to housing wealth gains might appear, at first, to be something of a puzzle. After all, young homeowners are most likely to move into a bigger house as their family size increases. House priceappreciationis notnecessarilygoodnews forthisgroup. There are two reasons for younger households to consume a greater share of theirhousingwealth gains. First, young householdsare morelikely to movethan older households; as a result, they are more likely to realize their housing wealth gains. Second, young households may be acting as buffer-stock or liquidity constrained consumers. Youngerhouseholdsface faster-growingpermanent incomes than do older households. As a result they would like to borrow against their future income gains, or, failing that, at least not save very much. Thus, they are more likely to consume shocks to any form of wealth, including housing wealth. As evidence for this view, Gourinchas and Parker (2002) show that households generallyact likebufferstockconsumersuntilaround age40,when theybeginto act moreliketraditionallife-cycleconsumers. Finally, it’s important to note that the estimated elasticity among younger households declines if I do not throw households that move or change family composition out of my dataset. This is probably because changing family composition and moving affect consumption in ways that cannot easily be controlled 6

for. Thefact that householdsinlatemiddleage(ages 52–62)are mostsensitiveto housingwealthgainsisnosurprise. Thesehouseholdsareontheeveofretirement and making their post-retirement housing choices. Evidence from the SCF and CPSshowsthatbeyondthisage,householdsgenerallyliveinlessvaluablehouses. In addition, they ought to have largely completed saving for retirement, so that additionstohousingwealthtranslateintoexpectedrealized wealth gains. Researchersagreethathouseholdsarelikelytospendlargeportionsofrealized home equity gains. However, little was known about households’ propensity to consumeoutofpassivehousingwealthgains(asopposedtoactivehousingwealth gains, which are realized in sales or in a refinancing). In this paper I find reasonableestimatesofthemarginalpropensitytoconsumeoutofhousingwealthgains. Further, I find a strongly non-monotoneage pattern to these estimates. However, theage patternalso fitswitheconomictheory. The rest of this paper is organized as follows: In section 2 I briefly review the aggregateimportance of housing wealth; in section 3 I present evidence from the PSID and the CPS about the effect of age on housing demand; in section 4 I presentestimatedconsumptionelasticitiesofhousingwealth;in section5 Itranslate these elasticities into marginal propensities to consume using weights from the Consumer Expenditure Survey (CE); finally, I briefly conclude and describe ongoingrefinements tothisresearch insection 6. Iincludeseveralappendicestothispaper. InappendixAIreviewthePSIDand present precisely the steps I used to arrive at my sample of “stable households.” 7

In appendix B I present samplestatisticsfrom my dataset. Finally, in appendix C I present exhaustive results from alternative specifications and different samples from the PSID. Almost all of my results are robust to all specification changes and sample selections. However, my finding that younger households have high consumption elasticities can be weakened if one does not eliminate observations fromthesampleinwhichhouseholdsmoveorexperiencecompositionalchanges. 2 Aggregate Trends in Housing Wealth In the past decade housing wealth has grown as a share of household portfolios; moreover, these gains have occurred against the backdrop of a historic rise in the rate of home ownership. Beginning in late 2000 and continuing through the present, residential real estate has appreciated faster than stock prices. Indeed, gains in housing wealth have been credited with propping up consumption even as thestock markethas slumped. Figure1showstherelativeperformanceofresidentialrealestateandthestock market. Although over the longer horizon, equities have heavily outperformed residential real estate, over the period 2000–2004 house prices have risen at a historically rapid pace while equity prices have slumped. The only comparable periodisthemid-tolate-1970s,whenhouseprices outperformedstockpricesfor aboutfiveyears. Figure2showstherelativeimportanceofequitiesandresidentialrealestatein thehouseholdsector’saggregatebalancesheet. Realestatehastraditionallymade 8

upthebulkofthehouseholdsector’sassets;thelargeshareaccountedforbyequitiesinthesecondhalfof1990swassomethingofahistoricalanomaly. Inthepast year, as the prices of the two asset classes moved in opposite directions, housing hasonceagain becomethedominantasset onthehouseholdbalancesheet. 3 Housing and Income Growth by Age House size and income growth are closely related to household age. Younger householdsarelikeliertoanticipatefastincomegrowthandarealsolikeliertoanticipateincreased demand for housingservices. In this section I present evidence ontheempiricalrelationshipbetween housingconsumptionand ageand between incomegrowthandage. 3.1 Housing Consumption and Age Housingchoicesarecloselytiedtohouseholdformationandsize;younger,childless, couples will demand a different mix of housing services than older couples with children. One mightimaginethat youngerhouseholdswould be more likely to increase their housing stock (to “upsize”) and older couples more likely to decrease their housing stock (to “downsize”); however, this demand-driven trend is somewhat obscured by the fact that younger households are much more likely to movethanolderhouseholds. SheinerandWeil(1992)conductedthefirststudyof the housing wealth of the old; they found that older households were more likely to downsize after a severe shock, such as sudden medical expenses or death of a 9

spouse. Theyfound that thisdownsizingoccurred relativelygradually. I also find that older households are quite likely to remain owner-occupiers; however, I do find that older households are much more likely to be downsizers than younger households,conditionalon moving. I used all waves of the PSID (1968–1999) to construct a dataset of the tenure status of household heads. The PSID attempts to follow households over time, so it was possible for most households to determine whether, once a household moved,itincreasedordecreasedthevalueofitshouse. Also,Iclassedasupsizers (downsizers) those households that moved into rented quarters (owned quarters) after having previously owned (rented) their home. I then determined the probability of moving by age and the probability of upsizing (downsizing) by age conditionalon havingmoved. Asameasureofthedemandforhousingservices,figure3presentstheaverage and median number of rooms for each household by the household head’s age. Dwellingsizeincreases withage untilage45, when itstartstodecline. As shown in the top panel of figure 4, households in lowest age quintile (between25and34)haveroughlytwicetheprobabilityofmovingperyearashouseholds in older quintiles. Conditional on moving, however, as shown by the lower paneloffigure4,youngerhouseholdsaretwotothreetimesmorelikelytoupsize thanolderhouseholds. My sample of PSID household-heads does not include household heads that moved in with their children or otherwise ceased to head their household. If an elderlyparentmovesinwithhisson,thesonisconsideredtheheadandtheparent 10

drops out of my sample. Thus I might understate the true decrease in housing demand by age. I turned to the 1976–2001 waves of the March Current Population Survey to get a representative picture of housing demand by person’s age. Controllingfor cohort, year and regional effects, as a person ages the conditional probabilityofbeingahouseholdhead(orspouse)andahomeownerbeginstofall off. Thepeak occurs around age 65; after this point peopleare less likelyto head their household and to own their homes. However, it is worth noting that at age 90,peoplewereas likelytobehomeowner/headsas 45year-olds. 4 Consumption Elasticities for Homeowners ThePSIDcollectsinformationabouthouseholdexpendituresonfood,housevalue (ifowned)andhouseholdincomesources. Inaddition,thePSIDtypicallyfollows households for several years. In this section I describe how I use the PSID to estimate consumption elasticities out of housing wealth. I find economically and statisticallysignificantvalues;moreover,Ifindthatthesevaluesdifferdramatically by age group, with the youngest and the oldest households having the highest elasticitiesand middle-agedhouseholdshavingthelowestelasticities. I constructed a sample of “stable households” from the PSID as my base dataset. AppendixAhasmoreinformationonthedataandsampleselectionprocedure,appendixBpresentsaseriesofsamplestatisticsforthedatasetandappendix Cdescribes resultsusingalternativespecifications. 11

4.1 Specification Let C i;t denote the real food consumption of household i in year t , V i;t denote the real value of the house and Z i;t denote a vector of additional household-level variables. Thespecificationofinterestis: (cid:1) l o g ( C i;t ) = (cid:11) (cid:1) l o g ( V i;t ) + Z i;t (cid:0) + u i;t : (1) Theestimatedparameter (cid:11) givestheelasticityoffoodconsumptionoutofhousing wealth. Although most of my additional controls are standard (household composition, head’s age, year fixed effects, change in family income, etc), three are relativelynoveland meritspecial mention. First, I include measures of permanent labor income (more precisely, permanent income growth). Because the PSID allows several observations of a single household,I can determinewhether a given incomeshock is transitory or permanent. Household consumption ought to react more to permanent than transitory shocks. Let ‘ i;t denote the log real labor income of the household head; for each household i I constructatimeseries ofgrowthrates: L i (cid:17) (cid:8) ‘ i;t (cid:0) ‘ (cid:0) i;t 1 (cid:9) T i =t 2 : I then construct the sample average, L i , for each household, as well as indicator variables for growth rates that are in the top and bottom percentile of the sample. Households with faster-growing permanent incomes, here proxied by higher val- 12

ues of L i , should have faster-growing consumption as well. As Poterba (1991) noted,houseprices,likeallassetprices,respondquicklytonewsabouteconomic conditions. Controlling for the household’s permanent income growth prevents housepricegainsduetoimprovinglocallabormarketconditionsfromspuriously affectingconsumptiongrowth. Second,Iincludedfixedeffectsforthehousehold’sstateofresidenceinyear t . Thesestate-leveleffectsremovecommonelementsinconsumptiongrowthdueto geography;in addition, theycontrol for peculiaritiesdue to conditionsin unusual states with few residents. For example, the final dataset contained only three observationsfrom Montana; eight other states were represented by fewer than 50 observations.7 Third, I included the state-level growth rate of home prices. These data are only available from 1976; thus including this variable required throwing out several early years of my sample. For this reason I generally report results with and withouttheextravariable. I refer to the specification withoutany ofthe extracontrols as the“base specification.” I then added the three extra sets of control variables in all possible combinations,exceptthatIalwaysincludedthestatefixedeffectswithstatehouse pricegrowth. 7Inascendingorder:NorthDakota(14)Wyoming(19),Idaho(22),Hawaii(26),RhodeIsland (33),andNewMexico(36). 13

4.2 Results Table 1 reports the estimated elasticity, (cid:11) b , and the standard error of the estimate for all specifications. The estimated elasticity does not vary with the inclusion of state fixed effects or permanent income controls, remaining constant at 3.94 percent. Thisestimateisdifferentfromzeroatthe0.1percentlevelofsignificance (or better). When the sample is constrained to 1976 and forward, and state-level housepricegrowthisincluded,theestimatedelasticitygrowstoabout4.7percent. Because I am interested in how these elasticities vary by age, I divided the sample into quintiles by age and reestimated the models. I show the point estimates and 90 percent confidence regions for the base specification and the specification withallpossiblecontrolsinfigure7. Notethestrikingpattern: Theelasticityofconsumptionfortheyoungestquintileishigherthanforthenexttwoolderquintiles. Thisisconsistentwithstandard economic theories in two respects. First, younger households are more likely to be liquidity constrained and thus use wealth purely as a buffer stock, while older households use wealth for life-cycle reasons. Gourinchas and Parker (2002) find evidence that younger households do indeed act as buffer-stock consumers while olderhouseholdsact moreas life-cycleconsumers. A second reason for the higher elasticity of younger consumers is that, as showninfigure4,youngerhouseholdsaremuchmorelikelytomove,thuspotentiallyrealizingsomeoftheirhousingwealth gains. Consistently, though, the highest elasticity of consumption is among the second oldest quintile, the group aged 52 through 62; examining figures 4 and 5 it 14

is precisely this age group which will begin downsizing and realizing someof its housingwealth gains. 5 Marginal Propensities to Consume out of Housing Wealth ThePSIDsurveyinstrumentsaskaboutseveralmeasuresofconsumption,includingfoodathome,foodawayfromhome,rent,housevalueandutilities. Although these are important components of total expenditures, they are far from the total. Moreover, there may be systematic differences in the translation from these components to total expenditures by household cohort and demographics. The most popular method of imputing total expenditures was first proposed by Skinner(1987). Morerecently,Blundell,Pistaferri, andPreston(2002)haveproposed invertingestimated Engel curves. Here, I follow the Skinner procedure, although Ihavealso usedtheBlundellet alprocedure, withvery similarresults. Skinner (1987) suggests using the limited information on consumption in the PSID to infer total expenditures for each household. The Consumer Expenditure Survey (CE) tracks spending on a variety of goods and services, including those also tracked by the PSID. I deviate from the standard procedure by not using information on house value in the PSID to impute total spending. Doing so would introduce a mechanical dependence between house values and consumption. Table2 givestheregressioncoefficients fromthemodifiedSkinnerprocedure. As can be seen, excluding house values degrades the fit substantially. On the 15

otherhand,thesedata(whicharefromrepeatedcross-sections)implyanelasticity ofconsumptionoutofhousingwealthofbetween0.22and0.25;eliminatingsuch amechanical resultisworththepricein precision. One might wonder about the value of including other observed household characteristics, such as the age of the head, number of children and female labor force participation. All of these variables (and others) in fact do enter Skinnerstylespecificationssignificantly;howevertheydonotimprovethefitoftheregressionmeasurably. Further, becauseIam interestedintheresponseofconsumption at different ages, I avoid using household information that is correlated with age (such as female labor participation or the number of children living at home) to constructmymeasures ofconsumption. Usingtheweightsfromtable2Iconstructed,foreachhouseholdineachyear, imputed levels of non-durable and total consumption. I then re-ran the specifications from section 4 using these imputed levels of consumption. The results (alongwithotherpertinentinformation)are shownintable3. The estimated elasticities based on the imputed consumption measures show thesamegeneralpatternasthosebasedonfoodconsumption. Theyoungestquintile has an elasticity of between 3 and 4 percent, the second-oldest quintile (ages 52-62) has the highest elasticity, of between 3.4 and 4.7 percent and other quintiles have lower elasticities. However, the age pattern of these elasticities is less pronouncedthanthoseestimatedusingfood consumption. The marginal propensity to consume is just the estimated elasticity times the average ratio of consumption to house value. For age quintile j the MPC can be 16

computedas: MPC j = (cid:11) b j 1 N N X =i j 1 C V i i : The average ratio of consumption to house value for each quintile is also shown intable3. Noticethatolderhouseholdsconsumelessrelativetothevalueoftheir house than do younger households; this is because younger households (as we have discussed) typically live in cheaper houses than do older households. Thus older households will typically spend less of an extra dollar in housing wealth thanyoungerhouseholds. TheestimatedMPCsshownintable3reflectthesetwoage-dependenteffects. The youngest households have the highest elasticities and the ratios of consumption to house value; thus they have the highest MPC. Households in the next age quintile(age35–42)havezeroelasticitiesofconsumptionwithrespecttohousing wealth, so their MPCs are also essentially zero. The last three age quintiles all have about the same MPC out of housing wealth, somewhere between 2 and 3.9 percent. 6 Conclusion InthispaperIusedthePSIDtoestimatetheeffectofpassivehousingwealthgains on consumption. I explicitlyexcluded thoseobservationsin which the household moved, as well as other observations in which consumption growth would react 17

tohousehold-levelchanges. Amongthisgroupofstablehouseholds,Iestimateda total sample MPC out of housing wealth of between 1.9 and 3.1 cents per dollar. Thisvalue,though,concealedvariationsacrossagequintiles. Thisvariationcould beexplainedbyreasonableeconomicforces: youngerhouseholdsarebothliquidity constrained and more likely to movethan older households. Households with the highest sensitivity to housing wealth gains (those aged 52–62) are precisely thosepreparing toretireand thuslikeliestto moveintosmallerhouses. References Blundell, R., L. Pistaferri, and I. Preston (2002). Partial insurance, information, and consumption dynamics. Manuscript, Department of Economics, Stanford University,Palo AltoCA. Carroll, C. D. (1992). Buffer stock saving: Some macroeconomic evidence. BrookingsPaperson EconomicActivity1992(2),61–156. Carroll, C. D. (1994, February). How does future income affect current consumption? QuarterlyJournalof Economics109(1),111–47. Carroll, C. D. (1996). Buffer stock saving: Some theory. Manuscript, DepartmentofEconomics,TheJohnHopkinsUniversity. Carroll, C. D. (1997a). Buffer-stock saving and the life-cycle/permanent incomehypothesis.QuarterlyJournalofEconomics112(1), 1–56. Carroll, C. D. (1997b). Unemployment expectations, jumping ( S ; s ) triggers, and household balance sheets. In B. S. Bernanke and J. Rotemberg (Eds.), NBER MacroeconomicsAnnual,1997.Cambridge, MA:MIT Press. Carroll, C. D. (2000). Requiem for the representative consumer? Aggregate implicationsofmicroeconomicconsumptionbehavior.AmericanEconomic Review90(2),110–15. Carroll, C. D., K. E. Dynan, and S. D. Krane (1999). Unemployment risk and precautionary wealth: Evidence from households’ balance sheets. Finance 18

and EconomicsDiscussionSeries 1999-15,Federal ReserveBoard. Carroll, C. D. and A. A. Samwick (1998). How important is precautionary saving? Reviewof EconomicsandStatistics80(3),410–19. Case, K. E., R. J. Shiller, and J. M. Quigley (2001, November). Comparing wealth effects: The stock market versus the housing market. NBER WorkingPaper 8606,NationalBureau ofEconomicResearch. Engelhardt,G.V.(1994,September).Housepricesandthedecisiontosavefor downpayments.JournalofUrbanEconomics 36(2),209–37. Engen, E. (1993). Consumptionand saving in alife-cyclemodel withstochastic earnings and uncertain lifespan. Manuscript, Federal Reserve Board, WashingtonD.C. Gourinchas, P.-O. and J. A. Parker (2002). Consumption over the life cycle. Econometrica70(1), 47–89. Hubbard, R. G., J. Skinner, and S. P. Zeldes (1994, May). Expanding the lifecycle model: Precautionary saving and public policy. American Economic Review84(2),174–79. Hurst,E.andF.Stafford(2002,August).Homeiswheretheequityis: Liquidity constraints,refinancing and consumption.Graduate School ofBusiness, UniversityofChicago, ChicagoIL. Lehnert, A. and D. M. Maki (2002). Consumption, debt, and portfolio choice: Testing the effects of bankruptcy law. Finance and Economics Discussion Series 2002-14,Federal ReserveBoard, Washington,D.C. Lehnert, A. and K. M. Pence (2003). The price of protection: Foreclosure law and house prices. Unpublished manuscript, Federal Reserve Board, WashingtonD.C. Lustig, H. and S. Van Nieuwerburgh (2002, November). Housing collateral, consumption insurance and risk premia. Manuscript, University of Chicago, DepartmentofEconomics,Chicago IL. Poterba, J. M. (1991). House price dynamics: The role of tax policy and demography.BrookingsPaperson EconomicActivity1991(2), 143–83. Sheiner, L. (1995, July). Housing prices and the savings of renters. Journal of UrbanEconomics38(1), 94–125. Sheiner,L.andD.N.Weil(1992,July).Thehousingwealthoftheaged.NBER WorkingPaper 4115,NationalBureau ofEconomicResearch. 19

Skinner,J.S.(1987).AsuperiormeasureofconsumptionfromthePanelStudy ofIncomeDynamics.EconomicsLetters23, 213–16. Skinner, J. S. (1996). Is housing wealth a sideshow? In Advances in the Economics ofAging,NationalBureau ofEconomicResearch Report, pp.241– 268.Chicago, IL: UniversityofChicago Press. 20

TABLE 1: EstimatedConsumptionElasticities Specification Estimate Standard Error WithoutPermanentIncomeControls Base.................................... 0 : 0 3 9 4 0 : 0 1 0 5 StateDummies........................... 0 : 0 3 9 0 0 : 0 1 0 5 StateDummies+StateHPI Growth......... 0 : 0 4 7 1 0 : 0 1 1 8 WithPermanentIncomeControls Base.................................... 0 : 0 3 9 4 0 : 0 1 0 5 StateDummies........................... 0 : 0 3 9 0 0 : 0 1 0 5 StateDummies+StateHPI Growth......... 0 : 0 4 7 2 0 : 0 1 1 8 NOTE. TablegivesOLSestimatesoftheelasticityofconsumption, (cid:11) b ,withrespect to real house value gains among a sample of stable homeowners. Without state house price controls there were 19,316 observations; with the state house price controlstherewere15,988observations. 21

TABLE 2: ConsumptionWeightsfromtheConsumerExpenditureSurvey(CE) ConsumptionMeasure Regressors Total Non-Durable Logfood awayfrom home. ( 0 0 : : 2 0 1 0 5 3 6 1 ) ( 0 0 : : 2 0 7 0 4 3 6 0 ) ( 0 0 : : 1 0 8 0 4 2 9 3 ) ( 0 0 : : 2 0 4 0 6 2 6 5 ) Logfood athome ......... ( 0 0 : : 2 0 9 0 2 6 4 3 ) ( 0 0 : : 3 0 6 0 1 6 2 5 ) ( 0 0 : : 3 0 1 0 0 4 1 7 ) ( 0 0 : : 3 0 9 0 2 5 1 5 ) Logrent equivalent........ ( 0 0 : : 2 0 2 0 4 3 6 4 ) ( 0 0 : : 2 0 5 0 0 2 4 5 ) Constant ................. R 2 ( 4 0 : : 4 0 9 5 2 2 1 2 ) ( 5 0 : : 4 0 2 5 9 2 4 8 ) ( 4 0 : : 1 0 0 0 9 3 7 9 ) 5 0 : : 1 0 4 4 6 5 3 0 ....................... 0 : 6 1 7 9 0 : 5 0 2 1 0 : 7 4 2 0 0 : 5 6 0 8 NOTE. TablegivescoefficientsfromanOLSregressionofthelogoftheindicated consumption measure on the indicated regressors; all values are real, deflated by the appropriate deflators. Standard errors are in parentheses. Data are from the ConsumerExpenditureSurvey,1984–2001. 22

TABLE 3: EstimatedMPCs outofHousingWealth AgeQuintiles All 25–34 35–42 43–51 52–62 63–95 TotalConsumption Base Elasticity(percent) C = V 2 0 : : 6 9 4 1 3 1 : : 0 0 5 4 0 0 : : 0 9 0 2 2 0 : : 9 8 1 8 3 0 : : 3 8 6 3 2 0 : : 9 8 0 7 MPC(percent) 2 : 4 1 3 : 1 8 0 : 0 0 2 : 5 7 2 : 8 0 2 : 5 2 PermanentIncomeControls+StateDummies+StateHousePriceGains Elasticity(percent) C = V 3 0 : : 4 9 3 1 3 1 : : 7 0 5 4 0 0 : : 0 9 1 2 3 0 : : 9 8 3 8 4 0 : : 6 8 7 3 2 0 : : 9 8 7 7 MPC(percent) 3 : 1 3 3 : 9 1 0 : 0 1 3 : 4 7 3 : 8 9 2 : 5 7 Non-DurableConsumption Base Elasticity(percent) C = V 2 0 : : 6 7 5 0 3 0 : : 0 8 6 0 0 0 : : 0 7 0 1 2 0 : : 9 6 2 8 3 0 : : 3 6 7 4 2 0 : : 9 6 2 7 MPC(percent) 1 : 8 7 2 : 4 6 0 : 0 0 1 : 9 9 2 : 1 7 1 : 9 5 PermanentIncomeControls+StateDummies+StateHousePriceGains Elasticity(percent) C = V 3 0 : : 4 7 4 0 3 0 : : 7 8 7 0 0 0 : : 0 7 1 1 3 0 : : 9 6 5 8 4 0 : : 6 6 9 4 2 0 : : 9 6 8 7 MPC(percent) 2 : 4 1 3 : 0 2 0 : 0 0 2 : 6 8 3 : 0 0 1 : 9 9 NOTE. Table gives data necessary to computeMPCs out of housing wealth from the PSID. The top panel gives results using imputed total consumption and the bottom panel gives results using imputed non-durable consumption. Rows give the estimated elasticities, the average ratio of (imputed) consumption to house valueand theirproduct (theMPC). 23

FIGURE 1: Equityand ResidentialReal EstatePrices in theU.S. 1975–2004 20 15 10 5 0 1975 1980 1985 1990 1995 2000 2005 Date 1=1Q:6791 :xednI S&P 500 HPI 50 40 30 20 10 0 −10 −20 −30 1975 1980 1985 1990 1995 2000 2005 Date tnecreP Four−quarter Percent Change S&P 500 HPI SOURCE. OfficeofFederalHousingEnterpriseOversight(OFHEO)andStandard &Poors. 24

FIGURE 2: AggregateU.S. HouseholdPortfolios1960–2004 60 50 40 30 20 10 0 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Date snoillirT Total Corporate Equities Real Estate 40 35 30 25 20 15 10 5 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 Date stessA dlohesuoH fo tnecreP Real Estate Corporate Equities SOURCE. Flowoffundsaccounts oftheU.S. 25

FIGURE 3: DemandForHousingServices byAge 7 6 5 4 25 45 50 75 Age of Household Head smooR fo rebmuN Mean Median SOURCE. PanelStudy ofIncomeDynamics,1968–1999. 26

FIGURE 4: HomeownershipDecisionsBy Head’sAge 60 50 40 30 20 10 0 30 40 50 60 70 80 Age of Household Head tnecreP Probability of Moving by Age 100 75 50 25 0 30 40 50 60 70 80 Age of Household Head tnecreP Probability Conditional on Moving Upsizing Downsizing NOTE. Figures give the probability of moving (top panel) and the probability of upsizingor downsizingconditional on moving(bottompanel). Data are from the PSID 1968–1999; graphs show linear probability models fit against sixth-order Chebyshev polynomialsin the head’s age. Regressions also included a full set of year, cohort andstatefixed-effects. 27

FIGURE 5: Head/HomeownershipRates By Age 70 60 50 40 30 20 10 30 40 50 60 70 80 Age tnecreP Head/Home Ownership Rates by Age NOTE. Figure gives the probability of being a household head (or spouse) and a homeownerconditionalon age; data are from the March CPS, 1976–2001. Plots show the results of probit regressions on a full set of cohort, year and regional fixedeffects and asixth-orderChebyshevpolynomialin thehead’s age. 28

FIGURE 6: Incomeby Age 50 40 30 20 10 25 30 35 40 45 50 55 60 65 70 AgeofHead srallod 6991 ,sdnasuohT Mean Total Family Income 29

FIGURE 7: ElasticitiesbyAgeQuintile 0.12 0.1 0.08 0.06 0.04 0.02 0 −0.02 −0.04 30 40 50 60 70 80 Age Quintile yticitsalE Base Specification 0.12 0.1 0.08 0.06 0.04 0.02 0 −0.02 −0.04 30 40 50 60 70 80 Age Quintile yticitsalE Permanent Income + State HPI Growth Controls NOTE. Figures give the consumption elasticities estimated via OLS for each age quintile. Thetoppanelgivesresultsfrom thebasespecification,thebottompanel from the specification containing all possible controls. The lighter lines give the 90percent confidence interval. 30

A Data Construction I constructed the sample for use in my analysis in two steps. In the first step, I picked a sample of “stable households” out of the PSID. For these households I haveinformationon food consumption(except in certain years, seebelow), labor and non-labor income, and house value or rental payments. In the second step, I picked which years for each stable household I could use in my regressions. There were a few reasons to throw out particular years for a given household; if the household had moved the previous year, if its size had changed, or, most importantly, if consumption information had not been collected in the previous year. A.1 Overview of Methodology Ibeganwithboththefinalrelease(1968–1993waves)andthecurrentearlyrelease of the PSID (1994–1999 waves). I use informationfrom the early release only to correct missing information from the final release. The PSID began as a representative sample of households in 1968; it has attempted to follow all split-offs (stemming from new household formation by now-adult children of the original sample,divorcesamongtheoriginalsampleandsoon). Ifollowthestandardpractice in restricting the sample to household heads and their families only. Heads neednotbemale;theyaretheprimaryearner. Ialsoeliminatethespecialpoverty subsample(thisis alsostandard practice). The results of this initial draw is a sample of 113,659 responses from 10,412 families that completed the PSID annual survey anywhere from one to 31 times (thePSID wasnotadministeredin1998). Familiescan enterand exitthePSID in any year; themostcommon pattern (705 families)is to answer each PSID survey instrument(i.e. tobeinthesampleforthefull31possibleyears). Thisisthebasic samplethatIused toconstructmy final dataset. Mysampleselectionworkedin thisorder: 1. Idroppedanumberofhousehold-yearsforreasonsoutlinedbelow;however, if a given household violated the criteria in a given year, other observationsfrom thehouseholdwere kept. 2. Ithenrequiredthathouseholdsremaininmysampleforatleastsixconsecutive years. Each household could remain in my sample only once. I call thismysampleofstablehouseholds. 31

3. Of all possible remaining household-years, I determined which ones were associatedwithvalidgrowthrates in consumptionand housevalue. NoticethatIusedtwotypesofcriteria: Thefirsttype(usedinstepone)excludeda particularhousehold-yearfrom consideration. This could then in turn preventthe household from entering the final sample (if for example it broke up the household’ssampleintoblocksoffiveorfewerobservations). Thesecondsetofcriteria (used in step 3) could eliminateany number of observations from a stable household. A.2 Initial Selection Theinitialsampleselectioneliminatedhousehold-yearsthatwouldingeneralrender the observation unusable. Note that although I am ultimately interested only in homeowners, I can use information from the same household in years when it was a renter. For example, certain variables are not collected each year, but are unlikelyto changeovertime(suchas therespondent’srace). Table A.1 gives the steps in my initial sample selection; the table shows how many observations (defined as household year combinations) were dropped at each step and how many separate households remained. I dropped households who neither owned nor rented their dwelling place (mainly students), those with household heads aged 24 or below, those who were not married, those with total family income before taxes of zero (this is an invalid code), and those who livedabroad. Intheend,Irequiredhouseholdstohavecompletedatleastsixconsecutive valid interviews in the period covered by the final release, 1968–1993. Variables of interest from the early release have not been checked for accuracy, and thePSID warns againsttheiruse. A.3 Selecting Household-Year Combinations BasicRestrictions TheselectionproceduredescribedintableA.1leftmewithanunbalancedpanelof 39,642potentialhousehold-yearobservationson2,832stablehouseholds. Notall ofthesehousehold-yearcombinationscanbeusedintheconsumptionregressions. The PSID did not collect consumption information in its 1968, 1973, 1988 and 1989 waves, nor did it collect information on utilities paid in 1973–1976, 1982, 1988-1993. 32

The PSID collects retrospective information on income, hours worked and employment status, so that answers included in the 1990 survey responses refer to the household’s experience in 1989. In constructing my final dataset, I shifted alloftheseretrospectivevariablesbackonecalendaryear. Thefoodconsumption questions may or may not be retrospective in nature. Most researchers who use thePSID haveconcludedthattheyrefertoaveragefoodexpenditureinthemonth orquarterinwhich thesurveywas given. Ifollowthispractice. Let v i;t denotetherealvalueofhousehold i ’shouseasreportedintheinterview year t and c i;t denotetherealvalueoffoodconsumptionby i inyear t . Here“food consumption”isdefinedastheannualizedfoodexpendituresonfoodathomeand food away from home, plus the net benefit from food stamps, if any. Nominal food expenditures are deflated by the relevant deflator at the interview month. Theprimaryrelationshipofinterestis between (cid:1) c i;t and (cid:1) v i;t ,defined as: (cid:1) (cid:1) c v i;t i;t = = l l o o g g (cid:0) (cid:0) c v i;t i;t (cid:1) (cid:1) (cid:0) (cid:0) l o l o g g (cid:0) (cid:0) c v (cid:0) i;t (cid:0) i;t 1 1 (cid:1) (cid:1) ; : Thus we must eliminate not only those years in which the PSID does not collect informationonconsumption,wehavetoeliminatethenextyear’sobservationsas well. Additional Restrictions The restrictionsdescribed aboveare the minimumnecessary to produce a dataset inwhicheveryobservationhasavalueforconsumptiongrowthandforhousevalue growth. I propose an additional set of restrictions designed (1) to isolate the effect of house value changes among households that do not move; (2) to eliminate households that might change consumption for reasons unrelated to house appreciation, such as changes in family size; and (3) to improvethe fit of a linear modelbyeliminatingoutliers. In my remaining sample, 1,246 observations were homeowners who moved to another owned home in the previous year. Because the present study is designed to ignore the role of seller’s equity extraction, it seems reasonable to remove these observations. A further 3,970 household-years represented households that changed composition in some important way. The reported change in consumption of these families will likely reflect in large part these changes; further, controlling for such changes involves essentially taking out the conditional 33

mean change in consumption growth for households that experienced the same change. Withsofew changes ineach category,theseconditionalmeansare likely tobebadlyestimated. Finally, my specifications involve a linear relationship between consumption growth and house value growth. The top and bottom percentile of the sample represent households who reported log changes in consumption and house value larger than 1; this corresponds to increases of about 170 percent and decreases of about 63 percent. Such huge values in the left- and right-hand side variables can distortOLSestimatesofalinearrelationship.8 8Itisworthnotingthattheobservationswiththelargesthousevaluegrowth(decline)alsohave thelargestconsumptiongrowth(decline);theseoutliersareespeciallydangerousbecauseofthis positiverelationship. 34

TABLE A.1: ChoosingStableFamilies Observations Selection Criterion Dropped Remaining Households Initialsample..................... 113,659 10,017 Neitherownednorrented .......... 5,402 108,257 9,858 Unmarried........................ 36,285 71,972 6,125 Head aged 24 oryounger........... 4,666 67,306 5,726 Zero/invalidfamilyincome......... 28 67,278 5,725 Livedabroad...................... 188 67,090 5,721 Zero food consumption ............ 252 66,838 5,713 Invalidhousinginformationa........ 3,748 63,090 5,335 Fewer than5 obs. in1968–1993b ... 6,195 56,895 3,312 Notin sample6 consecutiveyears... 11,776 45,119 2,960 Eliminateearlyreleaseobservations Year (cid:21) 1994 ....................... 4,983 40,136 2,960 Fewer than6 observations.......... 494 39,642 2,832 NOTE. Tabledescribestheselectionofthebasesampleofstablehouseholds. The maincriteriaforinclusionarethatthehouseholdconsistofamarriedcouple,have aheadolderthan24,andrespondtoatleastsixconsecutivePSID questionnaires. aForhomeowners,areportedhousevalueof0or999,999(aninvalidcode),orlessthan$1,000. Forrenters,annualreportedrentthatismissing,zero,orgreaterthan$60,000;in1988and1989 rentinformationwasnotcollected,theseobservationswerenotdeleted. bNotethatthisrequirementknocksoutmanyoftheextrasampleofLatinofamiliesintroduced intheearly1990s. 35

TABLE A.2: SampleSelection WithinStableHouseholds Observations Criterion Dropped Remaining Baserestrictions Bad InterviewYeara ...................... 10,249 29,393 Missingincomeinformation............... 1,813 27,580 Renters (previousyear) ................... 2,140 25,440 Renters (interviewyear)................... 116 25,324 Additionalrestrictions Movedin pastyear....................... 1,246 24,078 New husband/wife........................ 338 23,740 Changein geocodeforStateb.............. 48 23,692 Familycompositionchangec............... 4 23,688 Familysizechanged...................... 3,344 20,344 Adults 6= 2 .............................. 236 20,108 Bottom1% offoodconsumptiongrowth.... 202 19,906 Top 1%offood consumptiongrowth....... 202 19,704 Bottom1% ofhousevaluegrowth ......... 193 19,511 Top 1%ofhousevaluegrowth............. 195 19,316 NOTE. Thetableshowsthebaseselectionandtheadditionallyrestrictedselection ofhousehold-yearsfromthesampleofstablefamilies. Thebaseselectioncriteria aretheminimumnecessarycriteriatoruntheregressionofinterest;theadditional restrictions eliminateoutliers and households that might change consumptionfor reasonsunrelated tohousevalueappreciation. aYears in which consumptiongrowth cannotbe determined: 1968, 1969, 1973, 1974, 1988, 1989,and1990 bInprinciple,householdsthatdidnotmoveshouldipsofactonotchangeStates;mostofthese observationswerechangesininvalid/missingState geocodes. Ratherthanimputeothervalues,I choosetosimplydroptheseobservations. cOtherthannewhusband/wifeornewchildren 36

B Sample Statistics Figures B.1 and B.2 present sample statistics of the income and consumption of households,respectively. Notethat householdsusedin thesamplegenerally have higherincomesandconsumemorethanthebroadersampleofstablefamilies,especially in the early years of the dataset. Recall that to construct growth rates of consumptionand house values for each family we must discard the first observation. Thusfamiliesinthesamplearegenerallyolder,moreexperiencedandsoon. Also notice that reported food expenditures at home have been declining while food expendituresaway from homehavebeen increasing; thisis probably related to the increasing female labor force participation rates over the sample. Figure B.3 shows the mean and median reported housevalue for both groups; again, the levelsareslightlydifferent(especiallyintheearlyyears)butthepathsaresimilar. FigureB.4presentsthedistributionoftheobservedgrowthinhousevalueand totalfoodconsumption;thefigurescanbethoughtofassmoothedhistograms. As can be seen on the figures, the one-year reported changes in both food expendituresand housevaluestend tobequitelarge, andvary agreat deal. TABLE B.1: SampleMeans AgeRanges 25–34 35–42 43–51 52–62 63–95 Variable All FamilySize ( 3 0 : : 3 8 8 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) Age, Head ( 4 1 8 4 : : 0 4 3 4 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) Age, Wife ( 4 1 5 4 : : 0 1 4 4 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) Hours. Heada ( 1 1 : : 8 0 5 3 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) Hours. Wifeb ( 0 0 : : 8 8 2 9 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) ( 0 0 : : 0 0 0 0 ) aThousandsofhoursworkedperyear. bThousandsofhoursworkedperyear. 37

FIGURE B.1: SampleIncomeStatistics 40 38 36 34 32 1970 1975 1980 1985 1990 Year )laer( sdnasuohT Median Total Family Income All Stable Families Sample Only 24 23 22 21 1970 1975 1980 1985 1990 Year )laer( sdnasuohT Median Labor Income of Head All Stable Families Sample Only 2 1.9 1.8 1.7 1970 1975 1980 1985 1990 Year )laer( sdnasuohT Mean Work Hours of Head All Stable Families Sample Only 38

FIGURE B.2: SampleConsumptionStatistics 4.2 4 3.8 3.6 3.4 3.2 3 1970 1975 1980 1985 1990 Year )laer( sdnasuohT Median Food Expenditures at Home All Stable Families Sample Only 8 7 6 5 4 1970 1975 1980 1985 1990 Year )laer( sderdnuH Median Food Expenditures Away From Home All Stable Families Sample Only 39

FIGURE B.3: HouseValues 85 80 75 70 65 60 55 50 1970 1975 1980 1985 1990 Year )laer( sdnasuohT Mean House Value All Stable Families Sample Only 60 55 50 45 1970 1975 1980 1985 1990 Year )laer( sdnasuohT Median House Value All Stable Families Sample Only 40

FIGURE B.4: Densities 100 80 60 40 20 0 −20 −10 0 10 20 30 40 50 60 Percent Change ytilibaborP House Values 1 All Possible Observations Sample Only 0.8 0.6 0.4 0.2 0 −50 −25 0 25 50 Percent Change (a)HouseValueGrowth ytilibaborP Food Consumption All Possible Observations Sample Only (b) ConsumptionGrowth 41

C Alternate Specifications I now present estimated elasticities under every possible combination of dataset and specification. In addition, I report results for different age groups; I divided thesampleintotwogroupsat themedian andintoterciles. Themaindeviationfromtheresultspresentedinthemainbodyofthepaperis that,whenusingtheunrestrictedorfullsampledescribedintableA.1(thatis,the larger sample before the additional restrictions of table A.2 have been imposed), the consumption elasticity for the lowest age quintile drops to zero. The reason forthisisthathouseholdsinthelowestagequintilearethelikeliestto experience changes in family composition and to move. As I argued in appendix A households that move or that change composition are likely to change their reported food consumption for reasons unrelated to housing wealth, and in ways that are quitedifficulttocontrol. The other main insight is that, by breaking the sample into terciles, it is the middletercile that has the largest elasticity, the oldest tercile that has the secondlargestelasticityandtheyoungesttercilethathasthelowestelasticity. Inallcases, though,theestimatedelasticitiesare statisticallydifferentfrom zero. Tables C.1 and C.2 give results using the unrestricted (i.e. the “full”) sample of stable households, described in table A.1. The tables present the estimated elasticities under all six different specifications for all ages, ages divided by the median,intotercilesand intoquintiles. Tables C.3 and C.4 give results using the restricted sample of stable households, described in table A.2. The tables present the estimated elasticities under allsixdifferentspecificationsforallages,agesdividedbythemedian,intoterciles and intoquintiles. 42

TABLE C.1: ConsumptionElasticitiesUsingFullSample AgeQuintiles All 25–34 35–42 43–51 52–62 63–95 Base ( 0 0 : : 0 0 2 1 7 0 6 0 ) ( 0 0 : : 0 0 0 2 9 1 7 6 ) ( 0 0 : : 0 0 0 2 7 3 2 2 ) ( 0 0 : : 0 0 0 2 1 4 5 1 ) ( 0 0 : : 0 0 7 2 5 2 7 3 ) ( 0 0 : : 0 0 2 2 2 2 2 5 ) StateDummies ( 0 0 : : 0 0 2 1 7 0 4 0 ) ( 0 0 : : 0 0 0 2 7 1 1 7 ) ( 0 0 : : 0 0 1 2 0 3 3 4 ) ( 0 0 : : 0 0 0 2 2 4 4 3 ) ( 0 0 : : 0 0 7 2 4 2 6 4 ) ( 0 0 : : 0 0 2 2 1 2 7 6 ) StateDummiesandStateHousePriceGains ( 0 0 : : 0 0 3 1 0 1 7 5 ) ( 0 0 : : 0 0 2 2 1 4 0 5 ) ( 0 0 : : 0 0 0 2 4 5 2 9 ) (cid:0) ( 0 0 : : 0 0 0 2 5 9 4 1 ) ( 0 0 : : 0 0 8 2 0 6 4 4 ) ( 0 0 : : 0 0 3 2 0 6 6 0 ) PermanentIncomeControls ( 0 0 : : 0 0 2 1 7 0 6 0 ) ( 0 0 : : 0 0 1 2 0 1 4 6 ) ( 0 0 : : 0 0 0 2 7 3 2 2 ) ( 0 0 : : 0 0 0 2 1 4 6 1 ) ( 0 0 : : 0 0 7 2 5 2 9 3 ) ( 0 0 : : 0 0 2 2 2 2 4 5 ) PermanentIncomeControlsandStateDummies ( 0 0 : : 0 0 2 1 7 0 4 0 ) ( 0 0 : : 0 0 0 2 7 1 7 7 ) ( 0 0 : : 0 0 1 2 0 3 3 4 ) ( 0 0 : : 0 0 0 2 2 4 7 3 ) ( 0 0 : : 0 0 7 2 4 2 6 4 ) ( 0 0 : : 0 0 2 2 1 2 8 6 ) PermanentIncomeControls+StateDummies+StateHousePriceGains ( 0 0 : : 0 0 3 1 0 1 7 5 ) ( 0 0 : : 0 0 2 2 1 4 9 5 ) ( 0 0 : : 0 0 0 2 4 6 2 0 ) (cid:0) ( 0 0 : : 0 0 0 2 5 9 5 1 ) ( 0 0 : : 0 0 8 2 0 6 4 4 ) ( 0 0 : : 0 0 3 2 0 6 7 0 ) NOTE. TablegivesOLSestimatesoftheconsumptionelasticityofhousingwealth gainsusingthefullsample;standarderrorsareinparentheses. Resultsarereported forallages and foreach agequintileseparately. 43

TABLE C.2: ConsumptionElasticitiesUsingtheFull Sample AgeRanges Median Terciles 25–46 47–95 25–39 40–54 55–95 Base (cid:0) ( 0 0 : : 0 0 0 1 0 4 1 5 ) ( 0 0 : : 0 0 4 1 6 4 0 0 ) ( 0 0 : : 0 0 1 1 2 6 5 7 ) ( 0 0 : : 0 0 3 1 9 8 3 0 ) ( 0 0 : : 0 0 2 1 7 7 6 7 ) StateDummies (cid:0) ( 0 0 : : 0 0 0 1 0 4 2 5 ) ( 0 0 : : 0 0 4 1 5 4 8 0 ) ( 0 0 : : 0 0 1 1 1 6 1 8 ) ( 0 0 : : 0 0 4 1 0 8 4 1 ) ( 0 0 : : 0 0 2 1 7 7 5 8 ) StateDummiesandStateHousePriceGains ( 0 0 : : 0 0 0 1 4 6 5 4 ) ( 0 0 : : 0 0 4 1 9 6 5 4 ) ( 0 0 : : 0 0 1 1 9 8 9 7 ) ( 0 0 : : 0 0 4 2 0 1 1 4 ) ( 0 0 : : 0 0 2 2 7 0 9 5 ) PermanentIncomeControls (cid:0) ( 0 0 : : 0 0 0 1 0 4 1 5 ) ( 0 0 : : 0 0 4 1 5 4 9 0 ) ( 0 0 : : 0 0 1 1 2 6 8 7 ) ( 0 0 : : 0 0 3 1 9 8 4 0 ) ( 0 0 : : 0 0 2 1 7 7 6 7 ) PermanentIncomeControlsandStateDummies (cid:0) ( 0 0 : : 0 0 0 1 0 4 2 5 ) ( 0 0 : : 0 0 4 1 5 4 8 0 ) ( 0 0 : : 0 0 1 1 1 6 4 8 ) ( 0 0 : : 0 0 4 1 0 8 7 1 ) ( 0 0 : : 0 0 2 1 7 7 5 8 ) PermanentIncomeControls+StateDummies+StateHousePriceGains ( 0 0 : : 0 0 0 1 4 6 5 4 ) ( 0 0 : : 0 0 4 1 9 6 5 4 ) ( 0 0 : : 0 0 2 1 0 8 3 7 ) ( 0 0 : : 0 0 3 2 9 1 9 4 ) ( 0 0 : : 0 0 2 2 7 0 9 5 ) NOTE. TablegivesOLSestimatesoftheconsumptionelasticityofhousingwealth gains using the unrestricted (or full) sample; standard errors are in parentheses. Resultsarereportedforagesaboveandbelowthemedianandforeach agetercile separately. 44

TABLE C.3: ConsumptionElasticitiesUsingRestricted Sample AgeQuintiles All 25–34 35–42 43–51 52–62 63–95 Base ( 0 0 : : 0 0 3 1 9 0 4 5 ) ( 0 0 : : 0 0 4 2 1 5 9 1 ) ( 0 0 : : 0 0 0 2 8 4 4 7 ) ( 0 0 : : 0 0 2 2 8 6 3 3 ) ( 0 0 : : 0 0 7 2 3 2 4 0 ) ( 0 0 : : 0 0 3 2 0 1 5 6 ) StateDummies ( 0 0 : : 0 0 3 1 9 0 0 5 ) ( 0 0 : : 0 0 3 2 7 5 3 4 ) ( 0 0 : : 0 0 0 2 8 4 9 9 ) ( 0 0 : : 0 0 2 2 9 6 5 5 ) ( 0 0 : : 0 0 7 2 1 2 8 3 ) ( 0 0 : : 0 0 2 2 8 1 1 7 ) StateDummiesandStateHousePriceGains ( 0 0 : : 0 0 4 1 7 1 1 8 ) ( 0 0 : : 0 0 4 2 4 8 8 0 ) ( 0 0 : : 0 0 1 2 3 8 6 0 ) ( 0 0 : : 0 0 3 3 6 1 1 0 ) ( 0 0 : : 0 0 9 2 1 5 5 5 ) ( 0 0 : : 0 0 2 2 7 3 7 8 ) PermanentIncomeControls ( 0 0 : : 0 0 3 1 9 0 4 5 ) ( 0 0 : : 0 0 4 2 1 5 9 1 ) ( 0 0 : : 0 0 0 2 9 4 1 7 ) ( 0 0 : : 0 0 2 2 8 6 4 3 ) ( 0 0 : : 0 0 7 2 3 2 2 0 ) ( 0 0 : : 0 0 3 2 0 1 8 6 ) PermanentIncomeControlsandStateDummies ( 0 0 : : 0 0 3 1 9 0 0 5 ) ( 0 0 : : 0 0 3 2 7 5 3 4 ) ( 0 0 : : 0 0 0 2 9 4 4 9 ) ( 0 0 : : 0 0 2 2 9 6 6 5 ) ( 0 0 : : 0 0 7 2 1 2 3 3 ) ( 0 0 : : 0 0 2 2 8 1 2 7 ) PermanentIncomeControls+StateDummies+StateHousePriceGains ( 0 0 : : 0 0 4 1 7 1 2 8 ) ( 0 0 : : 0 0 4 2 5 8 0 0 ) ( 0 0 : : 0 0 1 2 4 8 5 0 ) ( 0 0 : : 0 0 3 3 5 1 9 0 ) ( 0 0 : : 0 0 9 2 1 5 6 5 ) ( 0 0 : : 0 0 2 2 8 3 2 8 ) NOTE. TablegivesOLSestimatesoftheconsumptionelasticityofhousingwealth gains using the restricted sample; standard errors are in parentheses. Results are reported forallages and foreach age quintileseparately. 45

TABLE C.4: ConsumptionElasticitiesUsingRestricted Sample AgeRanges Median Terciles 25–46 47–95 25–39 40–54 55–95 Base ( 0 0 : : 0 0 2 1 3 6 7 0 ) ( 0 0 : : 0 0 4 1 9 4 0 0 ) ( 0 0 : : 0 0 1 1 3 9 5 1 ) ( 0 0 : : 0 0 6 1 7 9 7 0 ) ( 0 0 : : 0 0 3 1 6 7 6 1 ) StateDummies ( 0 0 : : 0 0 2 1 3 6 8 1 ) ( 0 0 : : 0 0 4 1 8 4 3 1 ) ( 0 0 : : 0 0 1 1 0 9 4 2 ) ( 0 0 : : 0 0 7 1 0 9 2 0 ) ( 0 0 : : 0 0 3 1 6 7 0 2 ) StateDummiesandStateHousePriceGains ( 0 0 : : 0 0 3 1 2 8 0 1 ) ( 0 0 : : 0 0 5 1 6 5 1 8 ) ( 0 0 : : 0 0 1 2 5 1 2 1 ) ( 0 0 : : 0 0 9 2 2 2 5 3 ) ( 0 0 : : 0 0 3 1 5 9 8 0 ) PermanentIncomeControls ( 0 0 : : 0 0 2 1 3 6 8 0 ) ( 0 0 : : 0 0 4 1 9 4 0 0 ) ( 0 0 : : 0 0 1 1 3 9 8 1 ) ( 0 0 : : 0 0 6 1 7 9 8 0 ) ( 0 0 : : 0 0 3 1 6 7 6 1 ) PermanentIncomeControlsandStateDummies ( 0 0 : : 0 0 2 1 3 6 8 1 ) ( 0 0 : : 0 0 4 1 8 4 2 1 ) ( 0 0 : : 0 0 1 1 0 9 7 2 ) ( 0 0 : : 0 0 7 1 0 9 4 1 ) ( 0 0 : : 0 0 3 1 5 7 9 2 ) PermanentIncomeControls+StateDummies+StateHousePriceGains ( 0 0 : : 0 0 3 1 2 8 3 1 ) ( 0 0 : : 0 0 5 1 6 5 4 8 ) ( 0 0 : : 0 0 1 2 5 1 9 2 ) ( 0 0 : : 0 0 9 2 2 2 6 3 ) ( 0 0 : : 0 0 3 1 6 9 1 1 ) NOTE. TablegivesOLSestimatesoftheconsumptionelasticityofhousingwealth gains using the restricted sample; standard errors are in parentheses. Results are reported forages aboveand belowthemedianand foreach agetercileseparately. 46