Vesile Kutlu-Koc The Retirement-Consumption Puzzle and Unretirement DP 11/

Transcription

1 Vesile Kulu-Koc The Reiremen-Consumpion Puzzle and Unreiremen DP 11/

2 The Reiremen-Consumpion Puzzle and Unreiremen a Vesile Kulu-Koc b November 24, 2014 Absrac Using a panel sample of older Americans I invesigae he drop in consumpion a reiremen, i.e. he so-called reiremen-consumpion puzzle, by aking ino accoun ha individuals may re-ener he labor force afer being reired. In conras wih previous sudies I look a how household consumpion responds o he s in boh males and females labor marke saus. This paper has wo imporan findings. Firsly, he unreiremen decision is mainly deermined by pre-reiremen expecaions of work and financial facors such as he amoun of individuals accumulaed savings a he ime of reiremen and having an occupaional pension plan. Secondly, in a model where reiremen and unreiremen are insrumened wih individuals reiremen expecaions, consumpion does no respond o reiremen or unreiremen in line wih he predicions of he life-cycle model. Overall, he findings of his paper sugges ha individuals are able o smooh heir consumpion around reiremen. a I wish o hank Rob Alessie, Adriaan Kalwij, Karen van der Wiel and seminar paricipans a he Nespar Pension Day (Urech, November 2013), he Nespar workshop (Leiden, November 2013), he MEA Seminar (Munich, November 2014) for valuable commens and discussions. Financial suppor has been provided by he Nework for Sudies on Pensions, Aging and Reiremen (Nespar). b Corresponding auhor. kulukoc@mea.mpisoc.mpg.de, Munich Cener for he Economics of Aging (MEA), Nework for Sudies on Pensions, Aging and Reiremen (Nespar).

3 1 Inroducion According o he sandard life cycle heory of consumpion wih quadraic preferences, anicipaed income s do no affec individuals consumpion profiles because raional agens smooh heir consumpion over he life cycle using heir savings. For example, if reiremen is associaed wih an anicipaed decline in income, in he cerainyequivalence model proposed by Browning and Lusardi (1996), he opimal response of individuals is o save before reiremen o avoid a consumpion decline during reiremen. 1 As a resul, individuals level of consumpion should no a reiremen. Banks e al. (1998) exend he cerainy-equivalence model by allowing for non-quadraic preferences and possible nonseparabiliies beween leisure and consumpion and hey find ha marginal uiliy of consumpion is no smoohed around reiremen when individuals reire expecedly. This empirical observaion conradics wih he predicions of a life cycle model of consumpion and is referred o in he lieraure as he reiremen-consumpion puzzle. Looking a consumpion a reiremen migh help o undersand wheher individuals have saved enough for heir reiremen. This is of ineres from a policy perspecive as in many indusrialized counries, like in he Unied Saes; he social securiy budge is under srain due o populaion aging. Therefore, he governmens of he indusrialized counries adop some pension reforms o alleviae he problem of aging populaion. These reforms include reducions in public pensions which require individuals o save more for heir reiremen. If individuals are forced o reduce heir spending a reiremen, his may sugges ha hey have no saved enough for reiremen. In his case he reforms aiming o reduce public pensions migh be ill-advised. Earlier research in his field mainains he assumpion ha reiremen is an absorbing sae which means ha re-enering he labor force afer being reired is no possible. As a resul, hey compare prereiremen consumpion wih pos reiremen consumpion, assuming ha reiremen is a one-ime even. In his paper I relax his assumpion and invesigae consumpion behavior of reirees also afer hey re-ener he labor force. 1 The cerainy-equivalence model proposed by Browning and Lusardi (1996) predics ha anicipaed s in income have no impac on he level consumpion. The assumpions of his model are ha agens have ineremporally addiive uiliy funcions, capial markes are perfec, preferences are quadraic, agens have raional expecaions, and rae of ime preference is equal o real ineres rae. 1

4 Unreiremen is defined as reurning o he labor force afer reiremen and has become more popular in some counries in recen decades. (Maesas, 2010; Peersson, 2011; Kanabar 2012). For example, Maesas (2010) shows ha a leas 26 percen of Americans re-ener he labor force following a reiremen spell. Focusing on consumpion drop a reiremen only may be misleading while unreiremen among reirees is so prevalen. For example, hose who reired earlier han expeced due o unemploymen may experience a negaive income shock a reiremen and herefore may choose o re-ener he labor force o finance heir consumpion during unreiremen. This paper re-addresses he reiremenconsumpion puzzle by aking ino accoun ha individuals may re-ener he labor force afer being reired. For his purpose I use nine waves of he Healh and Reiremen Sudy (HRS) which represens he populaion of Americans over age 50 and heir spouses. When looking a he reiremen-consumpion puzzle, i is imporan o disinguish wheher individuals have reired expecedly or unexpecedly because a consumpion drop a reiremen when reiremen occurred as a resul of an unexpeced shock does no conradic wih he predicions of he life cycle heory of consumpion. For example, he sudy by Smih (2006) finds ha Briish workers reduce heir consumpion when reiremen is involunary as a resul of a healh shock or redundancy, whereas hey do no heir consumpion when reiremen is volunary. 2 In his paper, I use respondens subjecive reiremen expecaions as an insrumen o disinguish beween expeced and unexpeced reiremens. The HRS survey conains a quesion abou individuals self-repored probabiliy of working afer age 65. I use his probabiliy o es wheher consumpion s a expeced labor marke exis. The life cycle heory also predics ha consumpion does no wih unreiremen if individuals unreiremen occurs as planned. Maesas (2010) finds ha he majoriy of he HRS respondens who unreire planned his ransiion before reiremen. She shows ha unreiremen is an anicipaed even for he HRS respondens and i is no a resul of financial shocks or insufficien savings. In his paper I also esimae a model of unreiremen o invesigae he characerisics of he respondens who reenered he labor force afer being reired. I find ha individuals expecaions abou working during reiremen in he baseline year of he survey srongly predic heir fuure ransiions o unreiremen. Then I use hese expecaions as an insrumen o disinguish beween expeced and unexpeced unreiremens. 2 Smih (2006) defines individuals as volunary reirees if hey reired direcly from working and as involunary reirees if hey were observed as unemployed or long-erm sick/disabled before hey reired. 2

5 This paper is closely linked o Haider and Sephens (2007) in he sense ha i looks a consumpion a reiremen. Haider and Sephens (2007) invesigae reiremenconsumpion puzzle using quesions abou reiremen expecaions in he HRS and he Reiremen Hisory Survey (RHS). 3 They insrumen individuals curren reiremen saus wih heir planned year of reiremen and find ha, for he RHS sample, he household food consumpion drops by 7 percen o 11 percen for male workers who reired as expeced. On he oher hand, using he firs hree waves of he HRS survey, hey find ha reiremen is no associaed wih a significan consumpion decline. This paper is differen han heir sudy in wo ways. Firs, in conras wih heir sudy, I also ake ino accoun he fac ha individuals may go back o work afer reiremen and I examine wheher consumpion responds o unreiremen. Second, Haider and Sephens (2007) focus on male-headed households only and hey do no conrol for he labor marke saus of he spouse for couple households. Household consumpion of couples migh be sensiive o he s in he labor marke saus of boh he head and he spouse. For insance, household consumpion may no drop subsanially wih he head s reiremen if his spouse is sill working. In his paper I analyze he s in household consumpion wih boh males and females reiremen and unreiremen. This mehod also allows me o incorporae single females ino he analysis who are ofen lef ou in he previous sudies (See, e.g., Banks e al. 1998, Smih 2006). The main findings of his paper are as follows: The unreiremen decision is mainly deermined by pre-reiremen expecaions of work and financial facors such as he amoun of individuals accumulaed savings a he ime of reiremen and having an occupaional pension plan. In conras wih he earlier sudies in he lieraure, his paper does no find a significan drop in food consumpion a reiremen when reiremen is fully anicipaed. 4 I also find ha men s observed reiremen is no associaed wih a significan drop in food consumpion, ye household food consumpion decreases by abou 7 percen when women become reired expecedly or unexpecedly. The drop in consumpion wih women s reiremen does no necessarily conradic wih he predicion of he life-cycle model since 3 The HRS and he RHS surveys are boh longiudinal and inerview older Americans. The RHS was conduced from 1969 o 1979 and included men and unmarried women born beween 1905 and The HRS is an ongoing survey and began in The iniial HRS sample includes respondens who were born beween 1931 and 1941 and heir spouses (irrespecive of heir age). 4 Alhough here is a disincion beween consumpion and spending, I use hem inerably hroughou he analysis. For nondurable goods consumpion is usually he same as spending because consumpion and spending occur almos a he same ime. 3

6 in his model one canno disinguish wheher reiremen was anicipaed or unanicipaed. Moreover, a significan drop in household food consumpion a he women s reiremen can be explained by women s conribuion o home producion. Household food spending may decrease during women s reiremen since women spend more ime on meal preparaion and shopping han men. Anoher finding of his paper is ha in line wih he predicions of he life-cycle model consumpion does no respond o unreiremen if i is fully anicipaed. One of he reasons for his finding could be ha pos-reiremen jobs pay much less han pre-reiremen jobs and herefore individuals income does no increase significanly when hey unreire. Overall, he findings of his paper sugges ha individuals are able o smooh heir consumpion around reiremen. The paper is srucured as follows: Secion 2 describes he daa and he descripive saisics. Secion 3 presens he esimaion resuls for a model of unreiremen. Secion 4 oulines he consumpion model and he mehods for is esimaion. Secion 5 presens he main esimaion resuls as well as he robusness checks, and Secion 6 offers some concluding remarks. 2 Daa I use nine waves of he HRS survey for he period The sample consiss of he original HRS cohor who were born beween 1931 and 1941 and heir spouses (irrespecive of heir age). The HRS is a biennial panel survey of Americans and is respondens were firs inerviewed in 1992 (Juser and Suzman, 1995). The HRS is well-suied for he purpose of his sudy since i is a large sample of elderly populaion and i includes deailed informaion on employmen saus, food spending, individuals expecaions of work during reiremen, subjecive reiremen expecaions, healh, wealh, and marial saus of he respondens. I selec individuals from he iniial HRS cohor ha enered in 1992 because he quesion abou individuals expecaions of work during reiremen was no asked o he oher cohors ha enered in he subsequen waves. The HRS survey in 1992 covers abou 7,700 households wih a leas one member aged beween 51 and 61. For he analysis I exclude 222 households in which he responden married or remarried during he 5 Quesions abou food spending were no asked in he year The daa is drawn from he RAND HRS daabase which is a cleaned and user-friendly version of he HRS raw files. Variables ha are no available in he RAND daase were aken from he RAND-Enhanced Fa Files. 4

7 observaion period The reason behind his resricion is ha he quesion abou he responden and his/her spouse s expecaions of work during reiremen was only asked in I use hese baseline expecaions o predic he responden and his/her spouse s fuure unreiremen ransiions. If he responden was single in 1992 and was observed as a couple in he subsequen waves, he/she will no be in he sample because his/her wife expecaions of work are no available in he subsequen waves. If he responden remarried during he observaion period, he baseline expecaion of his/her spouse in 1992 is no relevan o he unreiremen decision of his/her curren spouse. Nex I resric he sample o 7,351 households wih posiive household income in Household income is beforeaxes and i consiss of he responden s income plus his spouse s income for couples. The componens of he household income are wage/salary income, capial income, unemploymen and disabiliy benefis, social securiy income, pensions and annuiies, income from veeran benefis, welfare and food samps, alimony, oher income, and lump sums from insurance, pension, and inheriance. 7,286 households also have non-missing informaion on household food spending in I furher resric he sample o 7,049 households in which he responden and, if presen, his/her spouse have non-missing informaion on heir labor marke saus. Nex I focus on 3,450 households in which he responden and, if presen, his/her spouse were working full ime or par-ime in This resricion is imposed because working respondens and heir spouses (a) can poenially become reired during he survey period , and (b) are eligible o answer he quesion on subjecive probabiliy of working afer age 65 and expecaion of work during reiremen. 7 Among 3,450 households in which he responden and, if presen, his/her spouse were working full ime or par-ime in 1992, 2,933 of hem also have non-missing informaion on he quesion abou he responden s and his/her spouse s expecaions of work during reiremen. 2,865 households also have non-missing informaion on he quesion abou he responden s and his/her spouse s probabiliy of working afer age A his sage I lose 127 households (1.7 percen) because he household income is equal o zero for hese households. The main reason why he household income is equal o zero is ha boh he head of he household and he spouse (if presen) were no working in In oher words, non-working respondens are excluded as hey canno reire and do no answer he quesion on subjecive probabiliy of working afer age 65 and expecaion of work during reiremen. This exclusion also concerns couple households in which he husband was working bu he wife was ou of he labor force in I find ha in couple households in which he husband was working in 1992, 65 percen of he wives was also working. Alhough he sample selecion may seem resricive, I sill include 65 percen of he wives due o high raes of female labor force paricipaion in he HRS sample. 5

8 The final sample includes 2,461 households which are also observed in a leas wo consecuive waves. I use he labor force saus informaion of he respondens and heir spouses o define reiremen and unreiremen. Labor force saus measure in he RAND HRS daa files classifies respondens and heir spouses a each wave as working full-ime, working parime, unemployed, parially reired, fully reired, disabled, or no in he labor force. RAND derives his measure by combining objecive and subjecive informaion available in each wave. Working full-ime is defined as working 35+ hours per week, and 36+ weeks per year, and working par-ime is defined as working less han 35 hours per week or less han 36 weeks per year. Respondens and heir spouses are classified as fully reired if hey are no working for pay and describe hemselves as reired and hey are defined as parially reired if hey are working par-ime and describe hemselves as reired. The difference beween parially reired and par-ime working is ha here is menion of reiremen in he case of being parially reired. In addiion o hese definiions and following earlier sudies such as Maesas (2010) and Hurd and Rohwedder (2008), I also define individuals who are unemployed, disabled, or no in he labor force as fully reired. I rea individuals as working if hey are working eiher par-ime or full-ime, and as reired if hey are fully or parially reired. Following Maesas (2010), individuals are defined as unreired if hey ransiioned from full reiremen o full-ime/par-ime employmen or parial reiremen or if hey ransiioned from parial reiremen o full-ime or par-ime employmen. The fac ha unreiremen ransiions are idenified based on a wave-o-wave s may lead o undersaing he imporance of unreiremen since his mehod ignores shor unreiremen spells ha may occur beween waves. Maesas (2010) uses he deailed job hisory informaion o idenify shor unreiremen spells beween waves and she finds ha only 5 percen of he reirees reener and exi he labor force beween waves. Moreover, she poins ou ha hese individuals have a very shor duraion of unreiremen and heir annual earnings do no subsanially due o unreiremen. In he face of Maesas (2010) evidence, I do no include beween-wave unreiremen spells in he analysis. 6

9 Reiremen expecaions are measured wih a quesion abou subjecive probabiliy of working pas This quesion is asked in each wave if he responden and his/her spouse are younger han age 65 and working. The exac wording of he quesion is as follows: On he same scale from 0 o 100 where 0 means absoluely no chance and 100 means absoluely cerain (Thinking abou work in general and no jus your presen job,) wha do you hink he chances are ha you will be working full-ime afer you reach age 65? I use individuals reiremen expecaions available in he las year of heir employmen (one wave before reiremen) o insrumen heir reiremen ransiion in he nex period. For hose who did no ransiion o reiremen during he observaion period, I use heir reiremen expecaions in he curren period. In his paper, I also invesigae he characerisics of he respondens who reenered he labor force afer being reired. For his analysis I use only he 2,541 individuals who were observed as reired (fully or parially) in he previous period (in wave -1) and provided non-missing informaion on he variables used in he analysis. Then I explain he probabiliy of becoming unreired in he curren period (in wave ) using he informaion from he ime of reiremen (in wave -1) and las employmen (before wave -1). The main variable of ineres in his analysis is individuals self-repored expecaions of work during reiremen. A he baseline year, 1992, he HRS asked survey paricipans he following quesion abou heir plans of work during reiremen: Some people wan o sop paid work enirely when hey reire, while ohers would like o coninue doing some paid work. Wha abou you? Survey paricipans repored eiher hey will sop paid work enirely or hey will coninue some paid work during reiremen. As i is shown in Secion 3 individuals expecaions abou working during reiremen in he baseline year of he survey srongly predic heir fuure ransiions o unreiremen. In he consumpion model (secion 4) I use hese expecaions as an insrumen o disinguish beween expeced and unexpeced unreiremens. The HRS colleced informaion abou household food expendiures in all waves excep he fourh wave (in 1998). In heory, households uiliy is affeced no only by food expendiures bu also expendiures spen on oher nondurable goods consumed by he household. In pracice, food expendiures have been used as a proxy for nondurable 8 Noe ha he HRS survey also asks abou he subjecive probabiliy of working pas 62. I did no use his variable in he analysis since his variable is missing for individuals who are aged 62 and older. 7

10 expendiures by prior researchers (Haider and Sephens, 2007; Smih, 2006). Alhough i is no a perfec measure, food is also a nondurable good, herefore, s in food expendiures are closely relaed o s in household uiliy. 9 Food expendiures are available a he household level and he quesions on food expendiures were answered by he financial responden. 10 These expendiures consis of hree separae caegories: food a he sore, food delivered o he door, and food eaing ou (See Appendix for he exac wording of hese quesions). For he households who are receiving governmen food samps las monh, he survey asks abou he quaniy of food samps received and he addiional amoun of money spen on hese hree caegories. For hose who are no receiving food samps, he survey asks abou expendiures on he same caegories. Based on hese quesions I calculae oal annual food expendiures for each household. 2.1 Descripive Saisics The sample which is used in he consumpion regressions includes 2,461 households as inroduced in Secion 2. Among hese households 1,343 were couples and 1,118 were singles in 1992, for a oal of 3,804 working individuals. 11 Among hese individuals 2,323 of hem have become reired during he observaion period (61.07 percen). 735 individuals have re-enered he labor force afer a reiremen spell (31.64 percen). Among hese unreirees 434 of hem re-enered he labor force wo years afer reiremen, which is he nex wave following reiremen (59.05 percen). This finding may sugges ha he majoriy of he individuals who go back o work do so wihin a shor period following reiremen. Maesas (2010) finds ha 26 percen of he reirees in he HRS survey reverse heir reiremen decision during he observaion period from The unreiremen 9 In 2001, he CAMS, a supplemenal survey o he HRS, began o collec daa on households oal spending including nondurable expendiures. However, his survey was inerviewed a subsample of original HRS respondens, herefore i conains smaller number of observaions. In his subsample here are a few individuals who re-enered he labor force afer reiremen. Secion 5 repors some descripive saisics and esimaion resuls based on his subsample as well. 10 In couple households, household level quesions abou finances are answered by one individual called he financial responden. In single households, he only responden is he financial responden. 11 In couple households he responden and his/her spouse are married or have a parnership. In single households he responden is single, never married, divorced, separaed or widowed. In single households and couple households, he household size can be greaer han one and wo, respecively, because hese households can live wih heir children or parens. 8

11 rae in his sudy, which is abou 32 percen, is higher compared o he unreiremen rae in Maesas (2010) probably because I use four addiional waves of he HRS survey which allows me o observe more ransiions o unreiremen. Table 1 illusraes median oal food spending and income around he ime of reiremen and unreiremen of men and women in couple households. Similar saisics for single households are repored in Table 2. The median food spending is repored for wo differen groups. The firs group consiss of reirees who became reired in wave zero and sayed reired in he nex wave (in wave one). The second group includes unreirees who became unreired in wave zero and also observed as unreired in he nex wave. Household food expendiures and income are divided by he OECD-modified equivalence scale. 12 Boh food spending and income are measured in 2004 dollars. According o Table 1, for couples, he husband s and he wife s reiremens are associaed wih a drop in household income which is abou 19 percen and 18 percen, respecively. On he oher hand, household oal food spending falls wih he husband s reiremen by 5 percen whereas i falls wih wife s reiremen by 9 percen. A smaller drop in household consumpion compared o ha in household income may sugges ha couples have enough savings o smooh heir food consumpion around reiremen. Women s conribuion o home producion migh explain a larger drop in household food spending wih he wife s reiremen. Women may spend more ime on home producion and shopping han men during reiremen which reduces expendiures on food eaing ou and bough a he sore. Using a cross secion sample of American households from he CAMS survey Hurd and Rohwedder (2005) find ha, year-old reired men and women spen 1.3 and 2.8 hours more on shopping and meal preparaion han men and women who were no reired, respecively. Table 1 also shows ha household income is 10 percen lower when he husband becomes unreired whereas he wife s unreiremen is associaed wih an increase in income which is abou 3 percen. Alhough individuals sar o earn wage income when hey unreire, i seems ha heir wage income does no exceed heir reiremen income subsanially. In he Unied Saes, when individuals reurn o work afer reiremen, hey may no longer receive employer pensions if hey reurn o heir former employer. For 12 See Hagenaars e al.(1994) for he deails of he OECD-modified equivalence scale. Household income is he sum of he responden s income and he spouse s income (if presen) and does no include he income of he oher members of he household. Therefore, household income of he couples is divided by

12 example, an individual may reire before he legal reiremen age, which is 65 for hose born before 1943 and beween 66 and 67 for people born in 1943 and laer, and may sar receiving employer pension benefis. If his person decides o reurn o his/her former employer, he/she may have o give up pension benefis because i was no legal o be an employee and a pensioner of he same firm before he legal reiremen age unil 2007 (Maesas 2010). If his is he case, i is expeced ha individuals unreiremen income is lower han heir reiremen income. I is also likely ha individuals may prefer o suspend heir social securiy benefis afer hey become unreired. In he Unied Saes, individuals may sop benefis wihin 12 monhs of firs claiming by repaying he benefis received during he las 12 monhs. They may also suspend heir benefis afer reaching he legal reiremen age and delay receiving hem unil he age 70. This is an appealing opion since each year of delay corresponds o 5 percen o 8 percen (depending on he year of birh) increase in reiremen benefis. 13 Moreover, Maesas (2010) finds ha he median wage earned on unreiremen jobs are significanly lower han ha on pre-reiremen jobs which may also be reason ha unreiremen is no associaed wih a significan increase in he household income. Table 1 also reveals ha household food consumpion is slighly lower when he husband becomes unreired whereas he wife s unreiremen is associaed wih a 2 percen increase in household food spending. The finding ha household consumpion does no increase subsanially wih unreiremen is expeced given ha unreiremen does no lead o a significan increase in household income. On he oher hand, a small increase in food consumpion wih he wife s unreiremen can be explained by he wife s conribuion o home producion. When he wife becomes unreired home producion may drop which may cause household food expendiures o go up again. According o Table 2, single men s income drops by 30 percen wih reiremen, whereas single women experience a fall in income a he ime of reiremen which is abou 27 percen. On he oher hand, single men s reiremen does no reduce household food spending a all while single women s reiremen is associaed wih a decrease in household food spending by 9 percen. One reason behind his finding can be ha single women do no have sufficien amoun of savings o smooh consumpion a reiremen compared o single men, since women earn less money han men over he life cycle (see Table A.1 in Appendix). Anoher reason can be ha single women s conribuion o home producion may be higher han ha of single men which reduces expendiures on eaing ou. Table 2 13 hp://m.kiplinger.com/aricle/reiremen/t037-c000-s004-paycheck-impac-on-reiremen-incomebenefis.hml 10

13 also suggess ha, income is slighly higher when single men or women become unreired. Regardless of gender, single households do no sar consuming more in he firs year of heir unreiremen () whereas single men s income and consumpion is higher in he hird year of heir unreiremen (in wave 1) han in heir las period of reiremen (in wave -1). Table 3 shows he s in sub-caegories of household food spending around he ime of reiremen and unreiremen of men and women in couple households. Reiremen may reduce expendiures on food eaing ou because reirees may spend more ime on meal preparaion han workers. According o Table 3, median household spending on eaing ou falls wih boh he husband s and he wife s reiremen. Similarly, boh he wife s and he husband s unreiremen are associaed wih an increase in he median spending on eaing ou. Table 3 also shows ha he median spending on food a home drops wih boh he husband s and he wife s reiremen. The increased leisure ime in reiremen may lead o more efficien shopping which may reduce expendiures on food bough a he sore. According o Table 4, median spending on eaing ou is slighly lower when single men become reired whereas i decreases by 16 percen when single women become reired. Single women s expendiures on food a home also fall wih reiremen by 6 percen while single men s expendiures in he same caegory are higher a he ime of reiremen. On he oher hand, boh single men and women do no sar spending more on eaing ou in he firs year of unreiremen, ye single women s spending on eaing ou is higher in he hird year of unreiremen (in wave 1). 3 Model of Unreiremen In his secion, I invesigae he characerisics of he HRS respondens who reenered he labor force afer being reired. As explained in Secion 2, hese resuls are based on an unbalanced panel of individuals who were reired in he previous period. I esimae a Probi model where he dependen variable is a dummy which akes one if individuals ransiioned o unreiremen in period and 0 oherwise. According o Maesas (2007) in a dynamic reiremen model, opimal unreiremen could arise hrough wo channels. Firsly, dynamic preferences for leisure would affec individuals decision o unreire. For example, individuals who work in a sressful job, and feel burned ou, would become reired. Afer having enough ime o recover, work may become aracive again. This is called he burnou and he recovery process. Secondly, he arrival of new informaion afer reiremen regarding financial siuaion, or saisfacion wih reiremen could cause an individual o 11

14 updae his expecaions of work during reiremen. For example, an unexpeced drop in sockholdings due o a sock marke decline may increase he probabiliy of unreiremen. Similarly, individuals who realized ha reiremen was no as enjoyable as hey expeced may go back o work. Following Maesas (2007) I esimae he probabiliy of unreiremen on a se of variables. Table 5 summarizes he esimaion resuls. The variables wih he sar (*) are consruced based on informaion in he las year of employmen and he res of he variables are based on informaion in period 1. The firs row in Table 5 shows ha hose who repored in he baseline year of he survey ha hey would coninue doing some paid work during reiremen are more likely o ransiion o unreiremen in he curren period. Alhough i is a drawback ha hese expecaions are no available in he subsequen waves of he HRS, i is promising ha he baseline expecaions sill srongly predic individuals fuure unreiremen. The esimaed marginal effec a he mean also suggess ha an average responden who planned o work during reiremen in 1992 is abou 3 - poins more likely o become unreired in he curren period compared o an average person who repored ha he/she would sop work enirely during reiremen. Compared o he average rae of becoming unreired in he curren period which is abou 10 percen, he size of he marginal effec is considerable. The second row in Table 5 shows ha someone s probabiliy of becoming unreired in he curren period is higher if he/she repored a higher probabiliy of working full-ime afer age 65 in he las year of employmen. The coefficien on operaors and laborers is no saisically differen han hose on he oher occupaional groups (i.e. managerial/professional specialy, services, ec.) suggesing ha unreiremen raes are he same across differen occupaion groups. Table 5 also reveals ha hose who have a defined benefi (DB) pension plan from he las employmen are less likely o reverse heir reiremen compared o hose who have defined conribuion (DC) pension plan or no pension plan a all. In he DB pension plan employees receive a fixed annual amoun during heir reiremen and his amoun is deermined by heir age, years of service and salary. In he DC pension plan employees conribue a fixed amoun in each year and hese conribuions are invesed, for example, in he sock marke. The pension income under he DC plan is deermined by he reurn on he invesmens which can be posiive or negaive. Therefore, DC plan holders may face greaer uncerainy regarding heir reiremen benefis compared o DB plan holders. As a resul, i is expeced ha DC plan holders are more likely o unreire o finance heir reiremen years compared o DB plan holders. Table 5 also shows ha he coefficien on years since reiremen is negaive and highly significan 12

15 suggesing ha hose who say longer in reiremen are less likely o unreire han hose who say shorer. If individuals realized ha reiremen was no as enjoyable as hey expeced, hey may reurn o work shorly afer reiremen. The fac ha an individual s work skills depreciae if he/she says in reiremen longer may also affec boh he employer s willingness o hire an old worker and he employee s willingness o work. Among he financial facors, household wealh a he ime of reiremen plays a role in he decision of unreiremen. Those who are in he lower wealh quariles (firs and second) are more likely o unreire han hose who are in he highes wealh quarile. On he oher hand, he level of food consumpion a he ime of reiremen does no seem o deermine he decision o become unreired in he nex period. Table 5 also shows ha hose who live in he low income households are less likely o unreire han hose who live in he high income households. The findings also sugges ha age is negaively associaed wih he likelihood of unreiremen. Those who live in couple households are less likely o reverse heir reiremen han hose who live in single households. The healh saus a he ime of reiremen also plays a key role in he decision of unreiremen: The healhier are more likely o reverse heir reiremen. There is also a gender effec in he decision o unreire such ha he probabiliy of unreiremen is higher for males han for females. The resuls regarding race shows ha whie people are less likely o go back o work afer reiremen compared o black people. Overall, he findings of his secion show ha individuals prereiremen expecaions of work play a key role in he decision of unreiremen which may sugges ha mos of he unreiremen ransiions are anicipaed before reiremen. On he oher hand, he probabiliy of unreiremen is also deermined by financial facors such as he amoun of individuals accumulaed savings a he ime of reiremen. This laer finding may imply ha for some individuals unreiremen was no fully anicipaed before reiremen. The arrival of new informaion afer reiremen regarding financial siuaion may cause individuals o updae heir expecaions of work during reiremen. 4 The Consumpion Model Following Smih (2006), I esimae a consumpion model derived from Frisch s demand funcion (Browning e al., 1985). In his model individuals choose consumpion ( C ) and leisure ( L ) o maximize he following value funcion: A 1, max U C, L, V A 1, 1 V X (1) 1 subjec o he budge consrain 13

16 A r A B w T L 1 1 C (2) where, X is a vecor of individual s characerisics, T is he individual s ime, is he individual s rae of ime preference, A is oal wealh, r is he ineres rae, B is nonlabor income, w is he wage rae. This model gives he following firs order condiions for he marginal uiliy of consumpion, marginal uiliy of leisure, and marginal uiliy of wealh V / A ) : U U C L ( C L, X C L,X w, (3), (4) 1 r 1 1 r , or equivalenly 1 where 1 is a forecas error such ha 1 0. conains all informaion from oher periods ha is needed o solve he maximizaion problem in period. Equaion (5) describes he rule for he allocaion of wealh over ime under uncerainy. According o his rule, he individual chooses savings such ha he marginal uiliy of wealh in period is equal o he discouned expeced marginal uiliy of wealh in period 1. In his model unanicipaed shocks will be refleced in he marginal uiliy of wealh over ime. The log-linearizaion of equaion (5) yields: ln b ln v (6) * 1 where 0 C 1 * v, 1 ln 1 ln 1 r ln 1 b. Equaion (3) implies a Frisch demand funcion condiional on labor supply 14 : C, L, X (7) where consumpion demand C is expressed as a funcion of individual s curren characerisics, leisure, and he marginal uiliy of wealh which conains all expeced fuure informaion including he effec of reiremen (or unreiremen) if i is an anicipaed even. (5) 14 Labor supply is difficul o model because of corner soluions. The condiional demand funcion in equaion (7) holds wheher or no he hour variable, T L, is equal o zero. See he discussion in Browning and Meghir (1991). 14

17 By assuming he wihin-period uiliy funcions exhibi consan relaive risk aversion (CRRA), U. can be expressed as: X L 1 exp U C, L, X C (8) 1 where is he coefficien of risk aversion for he household. Using he firs order condiion for marginal uiliy of consumpion and equaion (6), one can derive he consumpion growh equaion as follows: ln C b β ΔX L u (9) * 1 1 where, β θ, 1, u v and 1u 0. In he empirical implemenaion of his model I will esimae he following equaion: ln C α L β X u (10) i i i i where i 1,..., N and 1,..., T, lnc i lnci lnci 1, ln Ci is he naural logarihm of food expendiures, L is vecor of dummies indicaing he labor marke saus of he responden, X includes household size, a quadraic specificaion for age, labor marke saus of he spouse (if presen), and a couple dummy indicaing wheher he household members are married or have a parnership in period and zero oherwise. The labor marke saus of he responden or he spouse a ime migh be correlaed wih he forecas error u since he model described above assumes ha 1u 0. In oher words, u is uncorrelaed wih he variables in he informaion se available o he household a ime 1, ye labor marke saus a ime is no in he informaion se available a ime 1. For example, individuals may be forced o reire due o unexpeced shocks such as unemploymen or an adverse healh even a ime, and as a resul hey may be forced o reduce heir consumpion simulaneously. The drop in consumpion a ime is no necessarily a odds wih he predicions of he life cycle model described above since his effec is a combinaion of expeced and unexpeced reiremens. Therefore, one needs o disinguish beween expeced and unexpeced reiremens (or unreiremens). For his purpose, I esimae he parameers in he equaion (10) wih a Generalized Mehods of Momens esimaor (GMM) inroduced by Hansen (1982). This esimaion mehod consiss of wo sages. In he firs sage reiremen and unreiremen are regressed on a se of insrumens ha are correlaed wih reiremen and unreiremen bu ha are assumed o be uncorrelaed wih he error erm in equaion (10). In he second sage, prediced reiremen and unreiremen are calculaed for each individual using he firs sage resuls and hese 15

18 prediced variables are used o replace he endogenous labor marke saus in equaion (10). Under he raional expecaions hypohesis, i.e. 1u 0, all variables measured a ime 1 and earlier are exogenous and herefore, hey are poenial insrumens. Following Haider and Sephens (2007), I insrumen he responden s and his/her spouse s reiremen saus wih heir expecaions abou reiremen, which are measured wih heir subjecive probabiliies of working afer age 65 in he las year of employmen. 15 Similarly, I use individuals expecaions of work during reiremen available in 1992 o insrumen heir unreiremen saus in period. Following Banks e al. (1998) I also use individuals lagged labor marke saus as addiional insrumens. If he predicions of he model described above hold, I expec o see no a reiremen or unreiremen if hese evens are fully anicipaed, so he esimaed coefficiens based on GMM esimaion should be equal o zero. 5 Esimaion 5.1 Main esimaion resuls This secion presens he esimaion resuls based on a consumpion equaion described in equaion (10). In addiion o he consumpion equaion, I also esimae an income equaion o see how household income is associaed wih s in he labor marke saus of he responden and his/her spouse (if presen). I expec ha reiremen is associaed wih a decline in income since he replacemen rae, defined as he raio of pos-reiremen income o pre-reiremen income, is less han 100 percen in he Unied Saes. I esimae a pooled regression for couple and single households (See secion for a separae analysis). Table 6 repors he esimaion resuls. According o hese resuls, couple households income is 58 percen higher han single households income. Similarly, couples spend 36 percen more on food han singles. Household size is also posiively associaed wih expendiures on food. The firs wo columns in Table 6 show he resuls based on he Ordinary Leas Squares (OLS) esimaion of he consumpion and income equaions. According o hese resuls, household income falls wih men s reiremen by abou 22 percen whereas women s reiremen is associaed wih a drop in income which is abou 16 percen. This finding 15 The insrumen in Haider and Sephens (2007) is based on individuals expeced year of reiremen which is available in Alhough a similar expecaion quesion is asked in he subsequen waves, fewer individuals answered his quesion which does no allow me o use heir mos recen expecaions. (See Haider and Sephens (2007), foonoe 6 for deails.) 16

19 suggess ha women have higher replacemen raes han men. The differences beween men and women may sem from he fac ha men have higher earnings compared o women because hey (a) earn higher wages han women and (b) work more years han women (hey reire laer and/or have less work inerrupions). In he Unied Saes, he social securiy benefi formula is designed o provide progressive benefis in he sense ha i replaces a larger share of pre-reiremen earnings for low earners compared o high earners (Munnell and Soo, 2005). As a resul, replacemen raes are higher for women han men. The firs column in Table 6 shows ha household food consumpion decreases by abou 7 percen when women become reired. On he oher hand, men s reiremen does no household food spending significanly. The drop in consumpion a observed reiremen is no necessarily a odds wih he predicions of he life cycle model described above since his effec is a combinaion of expeced and unexpeced reiremens. The life cycle model only predics ha consumpion does no if reiremen occurs as planned. If unexpeced shocks such as unemploymen or healh shocks are correlaed wih he reiremen saus a ime, consumpion may decline around he ime of reiremen. Therefore, i is imporan o disinguish beween expeced and unexpeced reiremens. Alhough he OLS esimaes in he consumpion equaion are no consisen due o he possible endogeneiy, i is promising ha he drop in household consumpion is much smaller han he drop in household income suggesing ha individuals have enough savings o smooh heir consumpion around he ime of reiremen. A larger significan drop in household food consumpion a he women s reiremen can also be explained by women s conribuion o home producion during reiremen. The resuls regarding unreiremen sugges ha men s unreiremen is no associaed wih a significan increase in household income whereas women who unreired have a 9 percen higher income. On he oher hand, household food consumpion does no respond o eiher men s or women s unreiremen. Alhough in he OLS esimaion I canno disinguish beween expeced and unexpeced unreiremens, insignifican coefficien esimaes may sugges ha individuals smooh heir consumpion around he ime of unreiremen. The hird column in Table 6 presens he resuls from he GMM esimaion. The insrumens used in he firs sage regressions are, for boh men and women, he probabiliies of working afer age 65 based on he las year of employmen, expecaions of work during reiremen available in 1992, one period lag of working and reired dummies. Noe ha I inerac 1992 expecaions of work during reiremen wih one minus working 17

20 dummy in -1 in order o increase he predicive power of his insrumen. Since his insrumen is only available in 1992, i is assumed o be consan over ime. However, in 1994 he unreiremen dummy is zero for each individual in he sample because all respondens and heir spouses are working in 1992 so ha hey may eiher become reired or coninue working in Using he ineracion variable ensures ha expecaions of work during reiremen are irrelevan for hose who were working in he previous period. The null hypohesis of he Hansen s J saisic in Table 6 saes ha overidenifying resricions are valid. Failing o rejec he null hypohesis indicaes ha he insrumens are exogenous in he sense ha hey are no correlaed wih he error erm in he consumpion equaion (Equaion 10). The parial F saisics are quie high which shows he srengh of he relaionship beween he endogenous variables and excluded insrumens in he firs sage regressions. On he oher hand, F saisics from he firs sage regressions may no be valid o es he weakness of he insrumens if here are muliple endogenous variables in he second sage. Angris and Pischke (2009, pp ) propose a mehod o calculae he F saisics when here are muliple endogenous variables. Table 6 repors Angris and Pischke (AP) F es for weak idenificaion for each of he endogenous regressor. The null hypohesis of he AP F es is ha he paricular endogenous variable alone is weakly idenified. The criical values for he AP F es for weak idenificaion are no available bu one can use he rule of humb of Saiger and Sock (1997), which says ha if F saisic is a leas 10, he weak idenificaion is no a problem anymore. Table 6 shows ha he AP F es is greaer han 10 for each of he endogenous regressor which confirms ha he excluded insrumens srongly predic he endogenous variables. The las row in Table 6 repors Wooldridge's (1995) robus score es saisic o deermine wheher he endogenous righ-hand-side variables in he model are in fac exogenous. The rejecion of he null hypohesis indicaes ha he variables are endogenous a 5 percen level of significance, which may sugges ha he GMM esimaion is more reliable han he OLS esimaion. According o he GMM esimaion resuls, regardless of gender, household food consumpion does no significanly a he ime of reiremen or unreiremen. This finding is in line wih he predicion of he life cycle model described above in he sense ha consumpion does no a he ime of reiremen or unreiremen if hese evens are fully anicipaed. In conras wih Banks e al. (1998) and Haider and Sephens (2007), his sudy does no find a drop in consumpion a reiremen if reiremen is fully anicipaed. These aforemenioned sudies which find a reiremen-consumpion puzzle use relaively old daases from he 70s and 80s. The 18

21 replacemen raes for older cohors migh be differen han he replacemen raes for younger cohors. The recen sudy by Smih (2006) which looks a relaively younger cohors, finds ha food spending of Briish workers does no significanly when hey reire volunary, bu here is a significan drop in food spending when reiremen is involunary. Similarly, Haider and Sephens (2007) find ha he HRS respondens do no experience a fall a expeced reiremen whereas he RHS respondens, who are relaively older han he HRS respondens, reduce heir food consumpion abou 7 percen o 11 percen when hey reire expecedly. Overall, he findings of his sudy are in line wih he recen sudies which do no find a significan drop in consumpion a reiremen. 5.2 Robusness Checks Nondurable Consumpion Daa in his secion cover he period from 2001 o 2009 and were aken from he CAMS survey which is a supplemenal survey o he HRS. In 2001 he CAMS survey sen quesionnaires o a subsample of he households who were inerviewed in he HRS 2000 core survey. If household members are married or have a parnership, he quesionnaire was sen o one of he spouses, seleced randomly. In he iniial wave of he CAMS survey 3,866 households answered quesions abou oal nondurable spending based on 26 caegories (see Hurd and Rohwedder, 2008 for deails). This survey has smaller number of households and covers a shorer ime period han he HRS, herefore, I observe fewer reiremen and unreiremen ransiions in his sample han in he HRS sample. The CAMS survey is mached o he mos of he informaion in he previous HRS wave, i.e. CAMS 2001 is mached o he HRS However, informaion on financial variables such as wealh and income can be obained from he nex HRS wave. For example, HRS 2002 collecs informaion on oal income for he year 2001 which coincides wih he informaion on consumpion in CAMS Since HRS 2012 is no available ye, I exclude CAMS 2011 from he analysis. For he analysis, I selec an unbalanced panel sample of households in which he responden and, if presen, his/her spouse were working eiher full-ime or par-ime in The HRS sample inroduced in Secion 2 includes he households in which he responden and he spouse (if presen) were working in The sample in his secion consiss of individuals from he original HRS cohor who were firs inerviewed in

22 and were sill working in 2001 as well as younger cohors who enered he survey in he subsequen waves. 16 Based on he labor force saus informaion repored in he CAMS survey I define reiremen and unreiremen as explained in Secion 2. The final sample consiss of 678 households which also have non-missing informaion on nondurable spending and on he covariaes used in he analysis. Among hese households 369 were couples and 309 were singles in 2001, for a oal of 1047 working individuals. Among hese individuals 563 of hem have become reired during he observaion period (53.77 percen). 97 individuals re-enered he labor force afer a reiremen spell (17.23 percen). The unreiremen rae in his sample is smaller han he unreiremen rae in he HRS sample used in he main esimaion probably because he observaion period is relaively shorer. Table 7 illusraes median oal nondurable spending around he ime of reiremen and unreiremen of men and women used in he analysis. 17 I do no repor separae analysis for couples and singles due o small number of observaions. The caegories of nondurables are home insurance, propery ax, ren, elecriciy, waer, hea, home repair services, phone/cable/inerne, auo insurance, healh insurance, house/yard supplies, home repair supplies and services, food, dining ou, clohing, gasoline, vehicle services, drugs, healh services, medical supplies, vacaions, ickes, hobbies, conribuions, and gifs. According o Table 7 men s and women s reiremens are associaed wih a drop in household income which is abou 33 percen and 30 percen, respecively. On he oher hand, household oal nondurable spending falls wih he men s reiremen by 8 percen whereas i falls wih women s reiremen by 5 percen. The drop in nondurable consumpion is much smaller han he drop in income suggesing ha individuals have enough savings o smooh heir consumpion around reiremen. Table 7 also reveals ha household income is 11 percen higher when men become unreired while i increases by 3 percen when women become unreired. The small increase in income wih unreiremen is 16 The RAND HRS daa consiss of five cohors: Iniial HRS cohor who was born beween 1931 and 1941 was firs inerviewed in AHEAD cohor born before 1924 was firs inerviewed in Children of Depression (CODA) cohor, born 1924 o 1930, War Baby (WB) cohor, born 1942 o 1947, and Early Baby Boomer (EBB) cohor, born 1948 o 1953, were firs inerviewed in 1998, 1998 and, 2004, respecively. 17 The number of caegories of he nondurable spending differs across waves. For example, in wave 2 new caegories of nondurable spending such as personal care producs and gardening and housekeeping services were added o he survey. To compare nondurable spending over ime I use he caegories ha are available in all waves of he CAMS survey. 20