Commentary: Housing, Credit and Consumer Expenditure

Transcription

1 Commenary: Housing, Credi and Consumer Expendiure Sydney C. Ludvigson The subjec of his working session is housing and consumer behavior. As emphasized by John Muellbauer, one possible way in which house prices could influence consumer spending is hrough a housing collaeral effec ha is especially imporan for credi consrained households. While we keep in mind Professor Muellbauer s work, I wan o seer he conversaion in a slighly differen direcion, o alk o you oday abou he role of risk premia in he U.S. housing marke and he exen o which flucuaions in such premia migh influence aggregae consumer spending. Indeed, he unprecedened surge in U.S. house prices ha preceded he recen morgage crisis appears, anecdoally, o have been driven by a decline in marke paricipans assessmen of he riskiness of hese asses. Consider Char 1. Char 1 plos he raio of aggregae housing wealh o housing consumpion, for he pos-war period up hrough he end of Houses effecively pay dividends equal o he service flow from he durable housing sock; hence, his is a housing price-dividend raio, denoed P/D in he char. I have spolighed he dramaic run-up in his raio ha began in he lae 1990s. The picure is similar if we look a housing wealh relaive o oal (housing and nonhousing) consumpion (Char 2) or if we look a he price-dividend raio from real esae invesmen russ (REITs) securiies (Char 3). 335

2 336 Sydney C. Ludvigson Char 1 House Prices are High Relaive o Housing Consumpion Raio of Housing Wealh o Housing Consumpion Housing Price-Dividend Raio (P/D) Source: NIPA and U.S. Flow of Funds Char 2 House Prices are High Relaive o Toal Consumpion Raio of Housing Wealh o Toal Consumpion Housing Wealh-Toal Consumpion Raio Source: NIPA,U.S. Flow of Funds and auhor s calculaions

3 Commenary 337 Raio of REITs Price Index o Dividends Char 3 Price-Dividend Raio from REITs REITs Price-Dividend Raio Source: Naional Associaion of Real Esae Invesmen Truss These chars convey wo imporan empirical facs. One is ha he big movemens in housing wealh are really movemens relaive o measures of fundamenal value. A second is ha aggregae house prices relaive o virually any indicaor of fundamenal value have reached unprecedened levels in he las en years. As we ll see, saisical evidence suggess ha a subsanial fracion of he variabiliy in hese valuaion raios is aribuable, hisorically, o movemens in risk premia. Why concern ourselves wih risk premia in his session on consumer behavior? I ll make wo poins. Firs, movemens in risk premia in he U.S. housing marke are large and accoun for a subsanial fracion of variabiliy in house prices relaive o measures of fundamenal value. Second, unlike he direc effecs of credi marke condiions, movemens in risk premia in he aggregae housing marke poenially affec all homeowners, including unconsrained households who comprise a large fracion of aggregae consumpion. By saying his, I do no mean o sugges ha access o credi is a negligible facor in he consumpion decisions of some households. I do sugges, however, ha he direc effec of credi marke condiions on broader aggregae

4 338 Sydney C. Ludvigson consumpion may be smaller han he indirec effec ha hese condiions could have on aggregae consumpion via risk premia. Before elaboraing on hese poins, le s briefly review some oher economic channels hrough which housing wealh migh influence consumer spending. Firs, here migh be a direc wealh effec: The more wealh people have (of any form) he more hey consume. Bu as John Muellbauer poins ou, classic life-cycle consumpion heory suggess ha such a wealh effec should be small or even negaive. Moreover, houses may merely provide a hedge agains ren risk, and herefore flucuaions in house prices could have no real wealh effec (Sinai and Souleles, 2005). Second, house prices may also be correlaed wih consumer spending because of a common macroeconomic facor (for example, income expecaions) ha affecs boh housing and consumpion a he same ime. Professor Muellbauer s analysis explicily conrols for his possibiliy by including a measure of expeced income growh in he empirical model for consumpion. Third, housing wealh may be relaed o consumer spending because of a direc housing collaeral effec. For households ha are borrowing consrained, an increase in house prices means a higher collaeral value of heir home, enabling hem o borrow and spend more. This laer channel wha I ll call he housing collaeral channel forms he focus of John Muellbauer s work. To my knowledge, his work provides one of he few empirical invesigaions of credi marke condiions and aggregae consumer spending, and is herefore mos welcome. Prior o considering his evidence, however, I find i useful o pause for a momen in order o muse on he heoreical basis for linking credi marke condiions o broader aggregae consumpion, keeping in mind he general equilibrium complicaions. For example, in some economic frameworks, such as in work by Hanno Lusig of he Universiy of California, Los Angeles, and Sijn Van Nieuwerburgh of New York Universiy (Lusig and Van Nieuwerburgh, 2006, 2007), flucuaions in housing collaeral affec households abiliy o share risks wih one anoher, and herefore affec he cross-secional disribuion of consumpion, bu do no affec he size of he overall consumpion pie, or aggregae consumpion. Movemens in housing

5 Commenary 339 collaeral do affec risk premia in he housing marke, however, which are linked o aggregae consumpion. My reading of Professor Muellbauer s resuls is in fac quie consisen wih a relaively modes role for credi marke condiions in direcly affecing broader aggregae consumpion. Looking a his Tables 2 and 3, we see ha alhough he credi condiions index is found o be saisically significan in some regressions, measures of consumer credi condiions add lile o he explained variaion in quarerly aggregae consumpion growh. For example, in esimaions on U.S. daa, including measures of credi marke condiions in he regression enables he empirical model o explain jus 3.8% more of he quarerly variaion in consumpion growh han wha can be explained by a benchmark specificaion ha excludes hese indicaors. Boh he benchmark specificaion in Professor Muellbauer s sudy as well as he specificaions including he credi condiions index conain a freely esimaed sochasic rend o accoun for facors such as demographic rends, evoluionary changes in he inequaliy of income and wealh and changes in social securiy and pensions sysems, cohor-specific evoluionary shifs in aiudes in ime preferences and risk, as well as long-erm shifs in credi condiions. While i is possible ha changes in credi condiions have had an imporan effec on consumpion hrough heir effec on his rend erm, we canno ascerain wheher his is he case because i is impossible o idenify he relaive imporance of he many poenial economic forces ha may have conribued o movemens in a freely esimaed sochasic rend. Wha we do observe from hese resuls is ha he direc effec of credi marke condiions on broader aggregae consumpion appears o be small relaive o he effec of oher fundamenal deerminans, for example, he change in he unemploymen rae and he log income-consumpion raio. Wih ha, I wan o reurn o he quesion of housing risk premia, and bring your aenion o several more chars. From Char 1 we saw ha up unil abou 1998, he housing price-dividend raio flucuaed wihin cerain ranges. If we accep for he momen ha housing valuaion raios will coninue o flucuae wihin heir old hisorical ranges, hen a high housing price-dividend raio mus

6 340 Sydney C. Ludvigson foreell some combinaion of unusual increases in housing consumpion or decreases in housing prices in order o bring he raio back in line wih hisorical norms. So le s look a hose hisorical relaionships. Char 4 shows he relaionship, in daa up o 1998, beween he log housing price-dividend raio, on he horizonal axis, and he growh in housing consumpion hree years ahead, on he verical axis. I is clear from he cloud of daa poins in Char 4 ha here is no evidence of a saisically reliable relaionship beween hese wo variables. In fac, he slope of a fied line is almos zero, bu if anyhing is slighly negaive, suggesing ha high price-dividend raios forecas lower fuure housing consumpion, no higher. High house prices do no forecas higher housing consumpion. High house prices do forecas lower fuure house prices, however. Char 5 shows ha a high price-dividend raio for he U.S. housing marke is associaed wih slower and someimes negaive real housing wealh growh over a subsequen hree-year horizon. How do we link hese findings back o movemens in risk premia? We jus saw ha high house prices do no forecas higher housing consumpion. Bu one reason house prices could be high relaive o housing consumpion is ha he rae a which fuure housing consumpion is discouned may have been driven down by unusually low risk premia. We can measure movemens in risk premia in he housing marke as forecasable movemens in excess housing reurns. The nex wo chars show ha, hisorically, much of he variaion in housing wealh relaive o measures of fundamenal value has been driven by movemens in risk premia. Char 6 shows he hisorical relaionship beween he price-dividend raio for he U.S. housing marke, on he horizonal axis, and he hreeyear-ahead log housing reurn in excess of a shor-erm Treasury bill rae, on he verical axis. The slope of a fied line is negaive and large in absolue value, indicaing ha a high price-dividend raio forecass sharply lower excess housing reurns over he nex hree years. In fac, his simple relaionship explains 56% of he hree-year-ahead variaion in he excess housing reurn. I should noe ha hese findings are similar

7 Commenary 341 Char 4 Do High Price-Dividend Raios Forecas Higher Housing Consumpion? 3-Year Growh Forecas Horizon: 3 Years 3-Year Growh Housing Consumpion Growh Slope Log Price-Dividend Raio, U.S. Housing Marke Noe: Hisorically, high P/D does no forecas high fuure dividends, D. Wha abou fuure house prices, P? Source: NIPA, U.S. Flow of Funds, 1952:Q1-1998:Q1, auhor s calculaions Char 5 Do Higher Price-Dividend Raios Forecas House Price Growh? Year Growh Forecas Horizon: 3 Years 3-Year Growh 0.25 House Price Growh 0.15 Slope Log Price-Dividend Raio, U.S. Housing Marke Noe: Hisorically, high P/D forecass lower house prices, P. Source: NIPA, U.S. Flow of Funds, 1952:Q1-1998:Q1, auhor s calculaions

8 342 Sydney C. Ludvigson Char 6 High Price-Dividend Raios Forecas Low Housing Reurns 3-Year Log Excess Reurn 3-Year Log Excess Reurn Housing Reurn 3 mo Treasury Bill Slope Adj Rsq = 56% Log Price-Dividend Raio, U.S. Housing Marke The housing P/D is high when risk premia are low. Source: NIPA, U.S. Flow of Funds, 1952:Q1-1998:Q1, auhor s calculaions o hose repored in a recen working paper by Federal Reserve Board economiss Sean Campbell and coauhors (Campbell, Davis, Gallin and Marin, 2007). Similarly, Char 7 shows ha when he housing wealh-oal consumpion raio is high (as recenly, recall Char 2), excess housing reurns are again forecas o be sharply lower. Given ha movemens in risk premia consiue a quaniaively imporan source of variabiliy in he U.S. housing marke, he quesion I would like o see addressed is wheher such movemens have any influence on aggregae consumer spending. This is an empirical quesion; economic heory provides lile guidance as o he magniude of any such relaion. In an iniial aemp o answer his quesion, Sijn Van Nieuwerburgh of he Sern School of Business a NYU and I recenly esimaed a dynamic model of opimal consumpion choice ha explicily accouns for he role of risk-premia in asse markes. For your reference, he main feaures of his model and some empirical resuls are summarized in Exhibi A. Le me direc you o he empirical resuls in he able. The resuls sugges ha changes in aggregae consumpion are affeced conemporaneously by flucuaions in housing risk premia, bu modesly so. Such movemens comprise an esimaed 9.5% of he quarerly variaion in aggregae consumpion growh. Flucuaions in curren and expeced fuure labor income growh (human wealh), and, o a lesser

9 Commenary 343 Char 7 High Housing Toal Consumpion Forecass Low Housing Reurns 3-Year Log Excess Reurn 3-Year Log Excess Reurn Housing Reurn 3 mo Treasury Bill Slope Adj Rsq = 67% Log Housing Wealh-Toal Consumpion Raio Noe: The housing wealh-consumpion raio is high when housing risk premia are low. Source: NIPA, U.S. Flow of Funds, 1952:Q1-1998:Q1, auhor s calculaions exen, curren and expeced fuure reurns on financial asses, play a more imporan role in conemporaneous consumpion decisions. Bu i is imporan o disinguish he quesion of how flucuaions in housing risk premia may be relaed o conemporaneous consumpion, from he quesion of wha, if anyhing, such flucuaions porend abou he fuure pah of consumer spending. The las char in your handou, Char 8, shows he hisorical relaionship beween he log housing wealh-oal consumpion raio, on he horizonal axis, and he growh in oal aggregae consumpion, hree years ahead, on he verical axis. There is lile relaionship beween he wo variables; he slope of a fied line is close o zero. Thus, even hough flucuaions in housing risk premia may have some effec on conemporaneous consumpion, high house prices relaive o oal consumpion conain lile informaion abou fuure consumer spending. Of course, hese conclusions are based on hisorical relaionships up o The imporan quesion going forward is wheher he unusual behavior in he housing marke since 1998 represens a break oward a fundamenally differen long-run relaionship beween housing wealh and consumer spending, or wheher i is simply a very large oulier, a more exreme version of a familiar hisorical paern. I is oo early o know, bu a leas he hisorical daa give

10 344 Sydney C. Ludvigson Char 8 Housing Wealh-Toal Consumpion Does No Forecas Toal Consumpion Growh 3-Year Growh 3-Year Growh Toal Consumpion Growh Slope Adj Rsq = 5.6% Log Housing Wealh-Toal Consumpion Raio Noe: High house prices relaive oal consumpion conain lile informaion abou fuure consumpion. Source: NIPA, U.S. Flow of Funds, 1952:Q1-1998:Q1, auhor s calculaions us some reason o expec ha even a large decline in housing wealh may have only a modes impac on aggregae consumpion, provided ha such a decline does no have imporan spillover effecs on oher deerminans of consumer spending. I wan o close my commens by making a general observaion on he esimaion of wealh effecs in aggregae consumpion daa. Le lower case leers denoe log variables, e.g., log C = c. To esimae wealh effecs, a ypical empirical specificaion is a regression of log consumpion, c, on log housing wealh, h, log nonhousing wealh, s (including sock marke wealh and oher financial wealh), and log labor income, y : c =a+w s s +w h h +w y y +u. For example, Case, Quigley and Shiller (2005) esimae equaions of his form for several counries. I is commonplace o use such specificaions o esimae he relaive wealh effecs ou of differen forms of wealh (e.g., housing versus nonhousing wealh); ha is o esimae differenial marginal propensiies o consume ou of housing and nonhousing wealh. A larger coefficien on h han on s is inerpreed as a greaer marginal propensiy o consume ou of housing wealh han nonhousing wealh.

11 Commenary 345 In almos any represenaive agen model of consumer spending, however, a loglinear approximaion of he household budge consrain implies an equaion of he form above, in which he coefficiens w s, w h and w y have he inerpreaion of wealh shares, wih w s he share of nonsockmarke wealh in aggregae (human and nonhuman) wealh, w h he share of housing wealh in aggregae wealh and w y he share of human wealh in aggregae wealh (Leau and Ludvigson, 2001). In addiion, a firs-order approximaion of he budge consrain implies ha log consumpion, log wealh (housing and nonhousing) and log labor income are coinegraed, wih he wealh shares w s, w h and w y equal o he coinegraing coefficiens, which may be esimaed from daa. An advanage of his approach is ha he esimaed coefficiens are superconsisen and herefore robus o regressor endogeneiy in large samples. Bu because he esimaed regression coefficiens have he inerpreaion of wealh shares, hey canno be used o reveal he relaive imporance of housing versus nonhousing wealh in consumpion flucuaions. To see his, noe ha he marginal propensiy o consume ou of nonhousing wealh is given by dc dlog C / C C = ws, ds dlog S / S S while he marginal propensiy o consumpion ou of housing wealh is given by dc dlog C / C C = wh. dh dlog H / H H In U.S. daa, housing wealh, H, comprises a larger fracion of household ne worh han nonhousing wealh, S, hus he housing wealh share is larger han he nonhousing wealh share, w h >w s. A he same ime, he consumpion-nonhousing wealh raio is necessarily larger han he consumpion-housing wealh raio. C /S > C / H. I follows ha marginal propensiies o consume ou of differen forms of wealh may be similar even if wealh shares differ. Raher han giving an indicaion of he relaive imporance of differen forms of wealh in aggregae consumpion flucuaions, regressions

12 346 Sydney C. Ludvigson of he form above may merely ell us ha housing wealh is a larger share of aggregae wealh han is sock marke wealh. An alernaive approach o esimaing wealh effecs, one more akin o ha aken in Professor Muellbauer s paper, is he Euler equaioninspired mehodology in which esimaion is carried ou in growh raes and/or raios of variables. For example, he empirical specificaions in Professor Muellbauer s paper are regressions of consumpion growh (log differences in consumpion) on a variey of oher variables, including measures of consumer credi condiions, which are plausibly covariance saionary. The regression coefficiens in such specificaions no longer have he inerpreaion of wealh shares, so we escape he difficulies wih inerpreaion jus discussed. Bu esimaion in growh raes inroduces a new se of difficulies: There is no longer reason o expec ha he esimaed coefficiens will be robus o regressor endogeneiy. The righ-hand-side variables in hese regressions are endogenous, and he esimaed parameers may herefore fail o reveal he rue empirical relaionships ha he researcher seeks o evaluae. Geing around hese difficulies may require a more srucural approach o modeling and esimaing he wealh effecs on consumer spending.

13 Commenary 347 Exhibi A: Model of Aggregae Consumpion 1 : Opimizing households derive uiliy from nonhousing and housing consumpion. Allows for changing expeced reurns (risk premia and shor raes) on boh housing wealh and financial wealh. Allows for curren and expeced fuure labor income (human wealh). Innovaions (unpredicable movemens) in aggregae consumpion deermined by: Revisions in expeced fuure reurns o housing wealh, financial wealh, curren and expeced fuure labor income (human wealh). Revisions in expeced changes o he share of expendiure on nonhousing consumpion in oal consumpion. Esimaes from U.S. aggregae daa. Fracion of variance in nonhousing consumpion aribuable o: Nonhousing Aggregae Consumpion U.S. Daa, 1952:Q1-2006:Q4 Curren and expeced fuure housing wealh reurns (incl. risk premia) 9.5% Curren and expeced fuure labor earnings (human wealh) 54% Curren and expeced fuure financial wealh reurns (socks, bonds, nonhousing durables) 26% Covariance erms 10.5% Source: Sydney C. Ludvigson and Sijn Van Nieuwerburgh, New York Universiy. 1 Under complee markes, he relaionship beween aggregae consumpion and reurns can be represened by he preferences of a ficiious represenaive agen who can rade he aggregae housing reurn. The per-period uiliy funcion is a consanelasiciy-of-consumpion aggregaor over housing and nonhousing consumpion, while ineremporal preferences are modeled using he objecive funcion proposed

14 348 Sydney C. Ludvigson by Epsein and Zin (1989, 1991) and Weil (1989). The model implies ha he innovaion in nonhousing consumpion, c nh, is given by +1 nh nh s 1 c + 1 Ec + 1 = r+ 1 Er + 1+ ( 1 s)( E+ 1 E) κ j j r+ 1+ j + ( E+ 1 E) k s j, e 1 j= 1 j= 0 where E is he expecaion operaor, condiional on informaion a ime, he log reurn r +1 is a porfolio weighed average of he reurns on financial wealh, housing wealh, and curren and expeced fuure labor income (human wealh), s +1 is he log expendiure share on nonhousing consumpion in oal consumpion, and s and e are preference parameers. A vecor auoregression is used o esimae condiional expecaions, as in Campbell (1991).

15 Commenary 349 References Campbell, John Y. A Variance Decomposiion for Sock Reurns. Economic Journal, 1991, 101, pp Campbell, Sean D., and Morris A. Davis and Joshua Gallin and Rober F. Marin. Wha Moves Housing Markes: A Variance Decomposiion of he Ren- Price Raio. Working Paper, Board of Governors of he Federal Reserve Sysem, July Case, Karl E., and John M. Quigley and Rober J. Shiller. Comparing Wealh Effecs: The Sock Marke vs. he Housing Marke. Advances in Macroeconomics, 2005, 5(1), pp Epsein, Larry G., and Sanley E. Zin. Subsiuion, Risk Aversion, and he Temporal Behavior of Consumpion and Asse Reurns: A Theoreical Framework. Economerica, 1989, 57, pp Epsein, Larry G., and Sanley E. Zin. Subsiuion, Risk Aversion, and he Temporal Behavior of Consumpion and Asse Reurns: An Empirical Analysis. Journal of Poliical Economy, 1991, 99, pp Leau, Marin, and Sydney C. Ludvigson. Consumpion, Aggregae Wealh and Expeced Sock Reurns. Journal of Finance, 2001, 56(3), pp Lusig, Hanno, and Sijn Van Nieuwerburgh. How Much Does Housing Collaeral Explain Regional Risk Sharing? Working Paper, New York Universiy, Sepember Lusig, Hanno, and Sijn Van Nieuwerburgh. Can Housing Collaeral Explain Long Run Swings in Asse Reurns? Working paper, New York Universiy, June Sinai, Todd, and Nicholas Souleles. Owner Occupied Housing as a Hedge Agains Ren Risk. Quarerly Journal of Economics, 2005, 120, pp Weil, Philippe. The Equiy Premium Puzzle and he Risk-Free Rae Puzzle. Journal of Moneary Economics, 1989, 24, pp