The Labor Supply and Retirement Behavior of China s Older Workers and Elderly in Comparative Perspective

Transcription

1 D I S C U S S I O N P A P E R S E R I E S IZA DP No. 688 The Labor Supply and Retrement Behavor of Chna s Older Workers and Elderly n Comparatve Perspectve John Gles Dewen Wang We Ca October 2 Forschungsnsttut zur Zukunft der Arbet Insttute for the Study of Labor

2 The Labor Supply and Retrement Behavor of Chna s Older Workers and Elderly n Comparatve Perspectve John Gles World Bank and IZA Dewen Wang World Bank Bejng Offce We Ca World Bank Dscusson Paper No. 688 October 2 IZA P.O. Box Bonn Germany Phone: Fax: E-mal: za@za.org Any opnons expressed here are those of the author(s) and not those of IZA. Research publshed n ths seres may nclude vews on polcy, but the nsttute tself takes no nsttutonal polcy postons. The Insttute for the Study of Labor (IZA) n Bonn s a local and vrtual nternatonal research center and a place of communcaton between scence, poltcs and busness. IZA s an ndependent nonproft organzaton supported by Deutsche Post Foundaton. The center s assocated wth the Unversty of Bonn and offers a stmulatng research envronment through ts nternatonal network, workshops and conferences, data servce, project support, research vsts and doctoral program. IZA engs n () orgnal and nternatonally compettve research n all felds of labor economcs, () development of polcy concepts, and () dssemnaton of research results and concepts to the nterested publc. IZA Dscusson Papers often represent prelmnary work and are crculated to encour dscusson. Ctaton of such a paper should account for ts provsonal character. A revsed verson may be avalable drectly from the author.

3 IZA Dscusson Paper No. 688 October 2 ABSTRACT The Labor Supply and Retrement Behavor of Chna s Older Workers and Elderly n Comparatve Perspectve * Ths paper hghlghts the employment patterns of Chna s over-45 populaton and, for perspectve, places them n the context of work and retrement patterns n Indonesa, Korea, the Unted States, and the Unted Kngdom. As s common n many developng countres, Chna can be characterzed as havng two retrement systems: a formal system, under whch urban employees receve generous pensons and face mandatory retrement by 6, and an nformal system, under whch rural resdents and ndvduals n the nformal sector rely on famly support n old and have much longer workng lves. Gender dfferences n of ext from work are shown to be much greater n urban Chna than n rural areas, and also greater than observed n Korea and Indonesa. Descrptve evdence s presented suggestng that penson elgble workers are far more lkely to cease productve actvty at a relatvely young. A strong relatonshp between health status and labor supply n rural areas s observed, ndcatng the potental role that mprovements n access to health care may play n extendng workng lves and also provdng some bass for a common percepton that older rural resdents tend to work as long as they are physcally capable. The paper concludes wth a dscusson of measures that may facltate longer workng lves as Chna s populaton s. JEL Classfcaton: J26, J4, O5, O7, O57 Keywords: retrement, populaton agng, labor supply, pensons, Chna, Indonesa, Korea Correspondng author: John Gles Development Research Group The World Bank, MSN H St NW Washngton, DC 2433 USA E-mal: jgles@worldbank.org * Ths paper has beneftted from conversatons wth Fang Ca, Xaoyan Le, Phlp O Keefe, Albert Park, James Smth, John Strauss, Frman Wtoelar, Kyeongwon Yoo, Xaoqng Yu and Yaohu Zhao, and benefted from comments of Davd Wse, Yaohu Zhao and other partcpants n the Conference on Agng n Asa, sponsored by the US Natonal Academy of Scences and Chnese Academy of Socal Scences (Bejng December 9-, 2). We are grateful for fnancal support for ths work from two sources at the World Bank: the Gender Acton Program of the PREM Network and the Knowledge for Change Program mand by DEC. The fndngs, nterpretatons, and conclusons expressed n ths paper are entrely those of the authors. They do not necessarly represent the vews of the Internatonal Bank for Reconstructon and Development/World Bank and ts afflated organzatons, or those of the Executve Drectors of the World Bank or the governments they represent.

4 Introducton There s keen awareness across developed and mddle ncome countres of the developng world that ncreased longevty and agng populatons wll place sgnfcant and growng burdens on workng adults n the relatvely near future. In the US and other economes wth pay-as-yougo socal securty systems, these burdens wll be transmtted through fscal systems. Even where publc transfer mechansms are not as mportant, workng adults may nonetheless face ncreasng burdens assocated wth supportng the elderly through both fnancal and n-knd transfers. Increasng the retrement s frequently vewed as one feasble means of easng burdens on workng populatons, and yet t s lkely that exts from productve actvty are shaped by household wealth and ndvdual preferences as well as nsttutons and polcy. Alternatvely, contnued partcpaton n the workforce may reflect the ablty of workers to learn new sklls and to reman productve nto older. Wth an eye toward provdng nsght nto the retrement decson n East Asa, ths paper presents descrptve evdence on retrement and labor supply patterns n Chna, Indonesa and Korea. Whle Chna s rapd demographc transton s frequently hghlghted n news accounts because of the sheer sze of ts agng populaton, Korea and Indonesa are also confrontng rapdly agng populatons. 2 In contrast to most developed countres, however, rural and urban populatons face sgnfcantly dfferent retrement systems. Dfferences across rural and urban areas n both retrement patterns and access to fnancal support s most extreme n Chna, where most longterm resdents n urban areas have had formal w employment, retre at a relatvely young and receve substantal support from pensons. Rural resdents, by contrast, have lacked penson support and may expect to work n farmng or other agrculture-related actvtes untl relatvely late n ther lves. 3 In ths sense, urban resdents of Chna wth formal sector employment face Lee, Mason et al (2) hghlght the mplcatons of populaton agng for sustanablty of publc and prvate transfer systems across Afrca, Asa, Europe, Latn Amerca and the Unted States. 2 Recent research by demographers at the US Census Bureau suggests that the old dependency rates n 22 wll reach 22, 9 and 3 percent, for Korea, Chna and Indonesa, respectvely, and by 24 these rates wll rse to 53, 4 and 25 percent (Knsella and He, 29). A prelmnary release from Chna s 2 census nforms us that 3.3 percent of Chna s populaton s now over 6 as opposed to.3 percent n 2, whle the sze of the future workforce has dwndled, wth ndvduals under 4 accountng for 6 percent of the populaton, down from 23 percent n 2 (Natonal Bureau of Statstcs, 2). 3 New ntatves are currently underway n rural Chna. A government-subsdzed contrbutory rural penson ploted n 29 wll be rolled out to cover all rural countes over the next three years. In ctes, a new penson scheme, 2

5 retrement decsons that are more smlar to those of resdents n developed countres. Resdents of Chna s rural areas, by contrast, share more n common wth resdents of other developng countres, and make labor supply decsons n the absence of both penson avalablty and the constrant mposed by a mandatory of retrement from the formal sector. Ths paper brngs together nformaton from several data sources to hghlght dfferences n labor supply of older workers across urban and rural Chna, Indonesa and Korea, and to revew patterns and trends n the context of nsttutonal dfferences across these three countres and between urban and rural areas. For perspectve on retrement patterns n East Asan economes, we then place the retrement decson n Chna, Indonesa and Korea n the context of employment patterns of older workers n the Unted States (US) and the Unted Kngdom (UK). In common wth fndngs from the retrement lterature focusng on developed economes, the descrptve evdence presented n ths paper s suggestve of the role that ablty to collect a penson (or socal securty beneft) plays n the retrement decson. 4 Mandatory retrement provsons n each of these East Asan economes, however, condton decsons of when and how to ext from productve actvty. Whle sgnfcant numbers of retrees return to work n self-employed actvtes or nformal work after reachng mandatory retrement, the types of work that retrees are able to fnd may be unattractve for some older workers. Dfferences n the mandatory retrement for men and women n Chna lkely contrbutes to dfferences across genders n partcpaton n work later n lfe, wth mportant consequences for relatve penson wealth and relatve fnancal securty of older men and women. After revewng descrptve trends, we lay out an emprcal model to examne correlates of labor supply wth own and spouse elgblty to receve a penson, own and spouse health status and proxes for household wealth. We next revew data sources and correlates of employment separately for Chna, Indonesa and Korea. The paper presents comparatve descrptve evdence modeled on the rural penson program, was frst ntroduced n July 2 wth the am of provdng fnancal protecton n old to non-workng urban resdents and nformal sector workers. 4 Blau (994) suggests that socal securty elgblty contrbutes to relatvely hgh ext from the labor force at 65; Krueger and Pschke (992) explot desgn features of the US socal securty system to demonstrate the effects of benefts on labor force partcpaton. Gruber and Wse (999, 24) present evdence from OECD economes of the effects of socal securty and publc penson systems on labor supply decsons of older workers. 3

6 from East Asa and hghlghts mportant questons on retrement behavor n developng countres that may be addressed from new panel data ntatves currently underway. Employment Patterns and the Retrement of Older Workers In Chna, long-term urban resdents wth formal sector employment can expect to receve a penson upon retrement, but face mandatory retrement at a relatvely young. 5 Where urban employed men confront mandatory retrement at 6, women n blue collar occupatons are frequently requred to retre at 5, those n whte collar occupatons at 55, wth women n some categores (e.g., unversty professors) able to work untl 6. Among current retrement- resdents of urban areas, a large share s receves relatvely generous penson support. By contrast, rural elderly, who had lower ncomes durng ther workng lves and less accumulated wealth than ther urban counterparts (Kanbur and Zhang, 999; Ravallon and Chen, 27), do not typcally have penson ncome. Accordng to the 25 one percent populaton sample, 45 percent of urban resdents over 6 report penson ncome as ther most mport source of fnancal support, but only 4 percent of rural resdents note an mportant role for penson ncome. Instead, 38 percent of rural respondents over 6 report that ncome from own labor s ther most mportant source of support. 6 The stark dfference n employment rates of rural and urban resdents reflects dfferences n both penson wealth and mandatory retrement provsons across urban and rural areas. 7 As evdent n Panel A of Fgure below, whch presents locally weghted regresson (LOWESS) estmates of employment rates by, Chna s rural resdents are far more lkely to be employed well after the mandatory retrement s faced by urban resdents. From the Chna Health and Retrement 5 We defne a long-term urban resdent as an urban dweller wth an urban (non-agrcultural) resdental regstraton (hukou) status. Whle consderable efforts have been made recently to extend socal nsurance benefts to mgrants lvng n the cty, mgrants are much less lkely to have employment contracts and to have employers who are makng mandated contrbutons to penson, health and dsablty nsurance programs (Gles, Park and Wang, 2). 6 Addtonal descrptve statstcs on sources of support from the 25 one percent populaton subsample are reported n Ca et al (2) and Gles, Wang and Zhao (2). 7 In defnng employment n ths paper, we nclude w employment n the nformal sector, casual work, selfemployed actvtes and unpad work n famly run enterprses, all of whch may be mportant for older workers n these economes. We focus on employment as opposed to labor force partcpaton, per se, for two reasons. Job search s often not well-documented, and where t s (e.g., the CHARLS data for Chna) there are a vanshngly small number of respondents (fve n CHARLS) over 45 who are not employed but report actve searches for work. We have no doubt that a search process exsts for older workers who wsh to work, but t s dffcult to capture, and ths s partcularly true when large shares of older workers are self-employed. 4

7 Longtudnal Study (CHARLS) plot conducted n 28, we note 45 percent of urban men d 6 to 64 were stll employed at least one hour per week (Panel A), but n rural areas nearly 86 percent of men n ths range were stll workng. The dfference n employment of urban and rural women d 6 to 64 was even wder, wth only 6 percent workng n urban areas and nearly 57 percent stll employed n rural areas. If anythng the CHARLS plot, wth relatvely small sample szes n urban Zhejang and Gansu Provnces and representatve of only two provnces, may overstate the employment rates of older men and women. Also presented n Panel A of Fgure are the 99 and 29 estmates from the Chna Health Nutrton Survey (CHNS), whch show a substantally lower employment rate of 3 percent for urban men n the 6 to 64 range. When comparng employment rates across the dstrbuton over tme usng the CHNS, one observes declnes for both men and women n urban areas, but less pronounced declnes n employment rates n rural areas. 8 Rural women between 45 and 65 are somewhat less lkely to be workng n 29 than n 99, but ths does not appear to be true for older women or for rural men. 9 One of the sharper changes from 99 to 29, as vewed from the CHNS, lay n the declne n employment rates of women over 45 n urban Chna. Revewng representatve evdence from census data (Fgure A.) ths effect appears strongest among women from 45 to 55, as the populaton census and one-percent sample suggest modest ncreases n the employment of women over 55. Examnng further evdence from three waves of the Chna Urban Labor Survey (CULS), whch allows one to carefully dstngush employment (under our defnton) from labor force partcpaton, t s clear that Chna s boomng economy and a tght labor market has ncreased women s employment across the dstrbuton for women under 45, but that after 45 employment rates drop sharply (Fgure A, Panel A). Moreover, when we turn to labor force partcpaton (Fgure A, Panel B) t s evdent that n 2 and 25 many of the urban women 8 We revew evdence from the CHNS as t s the publcly avalable data source that researchers used (pre-charls) to study labor supply, health status and retrement of older workers n Chna (e.g., Benjamn et al, 23; Dong, 2). As t s based on a panel of households whch does not enumerate complete nformaton on famly members who have splt off of households, one should be concerned that later waves of the CHNS over-represent those who reman n the households. As we are nterested n the over-45 populaton, ths s less of a problem than f we were consderng the employment decsons of ndvduals who were chldren n 99 and lkely to have moved out of households. 9 In Appendx Fgure A., we present statstcs from the 99 and 2 populaton census and the 25 one percent populaton sample. Employment rates are somewhat lower n 25 for older men and somewhat hgher than one observes n the CHNS. 5

8 who were not employed were nonetheless lookng for work and n the labor force, so that labor force partcpaton rates were qute smlar n 2, 25 and 2 for women under 45. By 2, however, the CULS suggests a sharper drop n labor force partcpaton among urban women over 45 than we observed n earler years. The declne n older women s labor force partcpaton rases an mportant queston for labor research n Chna: Does the declne n women s employment reflect a resurgence of gender dscrmnaton n post-reform Chna or the effects of ncreases n household wealth and the ablty of women to ext the labor force at a younger? Dfferences across genders n mandatory retrement s lkely creates an nsttutonal bas aganst women s employment. Even n the absence of dscrmnaton, the employment decson of older women n urban Chna reflects a constraned choce. Women may return to work as consultants or n self-employed actvtes after reachng mandatory retrement, but retred women are frequently recevng pensons, whch rases reservaton ws for new employment. Those urban women unnterested n workng n typcal self-employed actvtes held by blue collar workers (e.g., nannes or housekeepers), may choose to stay out of work f ther penson ncomes are suffcent. Earler research on labor force partcpaton n Chna has noted the drop n employment rates of urban resdents, and urban women n partcular (Ca et al, 28; Maurer-Fazo et al, 2) and attrbuted the drop to the effects of state sector restructurng after 997. Restructurng reduced the state sector through two mechansms: for younger state-sector workers, a specal one-tme urban layoff (xagang) program provded up to three years of lvng subsdes (and contnued penson and health nsurance contrbutons), as well as tranng and job placement assstance through reemployment centers. Workers wthn fve years of retrement were encourd to take early retrement wth penson benefts. As women were much closer to retrement than men, partcularly blue collar workers facng retrement at 5, a far hgher share of women took early retrement than men. After losng work durng state sector restructurng, men, the young and the well-educated, generally faced a shorter duraton out of work (Appleton et al, 22; Gles et al, 26a; Maurer- It should be remembered that the CULS sample used for these comparsons are representatve samples from fve large ctes (Fuzhou, Shangha, Shenyang, Wuhan and Xan), and may not be representatve of work decsons n medum and small ctes. 6

9 Fazo, 27). Moreover, some researchers found that women s decson to reenter the workforce was affected not only by permanent and relatvely generous pensons, but also by famly crcumstances (Gles et al 26b). Gven support avalable through pensons to relatvely young workers durng economc restructurng and potental bases aganst hrng dsplaced workers relatvely close to retrement, one would expect to see sharp drops n employment of urban women from 99 to 997 and 2, as s shown n Panel B of Fgure. If reduced employment rates of urban women over 45 were smply the effect of restructurng n the late 99s, however, we would expect that employment of women n the 45 to 55 range would return to hgher pre-997 levels by 29. The fact that older workng women, who have not yet reached mandatory retrement, contnue to have lower employment rates, even after labor markets tghtened durng the 2s, rases the possblty that exts from the labor force may be a choce facltated by hgher wealth or by ncreasng demands on tme for non-market actvtes, such as carng for chldren, elderly or other ll famly members. Exts from employment among older workers as they approach penson elgblty are not unusual n more developed economes such as the UK and the US, but ext rates n the years before retrement are not typcally as hgh as one observes n Chna. Fgure 2 hghlghts dfferences n employment rates across fve countres and shows that 68 and 7 percent of women 5 to 54 are stll employed n the US and UK, respectvely. Employment rates of women n ths range are somewhat lower n urban Indonesa at 63 percent, but above the 38 and 3 percent employment rates wtnessed n urban Korea and Chna, respectvely. Relatve employment patterns of men n the 55 to 59 range follow a smlar pattern: 68 and 67 percent of urban men n ths range n Chna and Korea, respectvely, are employed, whle 82 percent of urban Indonesa men of ths are stll workng. Mandatory Retrement Provsons and Retrement Patterns. Incentves created by gender dfferences n the mandatory retrement may encour early exts from the labor force for women, partcularly for women who face the prospect of job search n ther 4s. Career changes and job changes later n workng lfe can be dffcult n developed and developng economes Note that ths s the urban employment rate for women usng the CHARLS sample, n the CHNS sample the rate s hgher at 42 percent. 7

10 alke. As begnnng a new job requres learnng processes, technology and culture of the new workplace, a new employee wll not reach peak productvty n a poston mmedately upon beng hred. An employer may be less lkely to consder hrng a worker who s close to mandatory retrement smply because there s not suffcent tme to earn a return on ntal tranng and start-up costs relatve to a younger worker. In the US, where many workers start to leave the workforce upon elgblty for socal securty benefts, dffcultes fndng new employment among older dsplaced workers are well-documented. Research usng the US Health and Retrement Study (US-HRS), for example, has demonstrated that of workers who lose jobs after 55, only 6 and 55 percent of men and women are employed agan wthn two years, whle 8 percent of non-dsplaced workers are employed (Chan and Stevens, 2). In Chna, and to some extent Indonesa and Korea, a mandatory retrement for some occupatons and types of employers s even more bndng than socal securty elgblty n the US. Dfferences n mandatory retrement s of men and women n Chna may have a sgnfcant mpact on how employers vew the relatve returns to hrng male and female employees who are n ther 4s and older. 2 In Korea and Indonesa, retrement s are not mandated by the law, but employment laws allow for frms to set mandatory s, and government employees and cvl servants face mandatory retrement. Explct ratonales for mandatory retrement provsons and legal support for them n Korea s based recognton that frms have long promoted worker allegance by payng employees below ther margnal product early n ther careers and rewardng them wth a premum after sgnfcant tenure. As workers later n ther careers earn sgnfcantly more than ther reservaton ws and thus have a dsncentve to retre, mandatory retrement provsons mplemented by frms and the government cvl servce provde a means of guaranteeng that less productve older workers wll eventually leave the frm, and facltate promotons for younger employees. 3 The Korean Supreme Court passed a rulng n 978 whch permtted employers to adopt a mandatory retrement, 4 and frms have responded by settng 55 as the medan mandatory retrement, wth some exemptons allowng later retrement for senor executves. 2 Of course, employers may already perceve older workers to be less productve (Chan and Stevens, 2; Dalen et al, 2), but ths s a problem faced by both older men and women when lookng for work. 3 Lazear (979) presents a theoretcal model whch suggests that mandatory retrement solves an mportant ncy problem by allowng frms to shed unproductve older workers. 4 Rulng No. 78da 46, 2 September

11 After the East Asan Fnancal Crss n 997/8, frms whch had not mplemented mandatory retrement started to do so as a way of slowng w ncreases assocated wth senorty-based w systems (Cho and Km, 25). In Indonesa, government employees and cvl servants face mandatory retrement at 55, and ths s followed by some large prvate employers as well (Brodjonegoro and Smanjuntak, 22; Leechor, 996). From 995, the normal retrement was consdered to be 55 wth a maxmum of 6, wth these retrement s applyng to larger prvate sector employers as well as the government. 5 Apart from the normal retrement, however, there are numerous exceptons made through presdental and mnsteral decrees, that rase the retrement s for some occupatons. Doctors and other hgh sklled professonal cvl servants may retre at 6, more senor levels of cvl servce are able to retre as late as 65 and unversty professors face mandatory retrement at 7. Employees of smaller scale enterprses and the self-employed, who comprse the urban nformal sector, do not face mandatory retrement. The hgher employment rates of older urban men and women n Indonesa and Korea (Fgure 2) suggest that mandatory retrement from some occupatons and types of employers does not mechancally lead to permanent ext from productve employment. Exstng research on older workers n both countres suggests that an mpendng retrement creates ncentves for forwardlookng workers to leave employers preemptvely, ether to start ther own busnesses or to start second careers workng for smaller prvate sector employers. 6 One sgnfcant dfference across workers n urban and rural areas of Chna, Indonesa and Korea les n the share of the retrement workforce wth access to pensons. In Chna, evdence from Chna Health and Retrement Longtudnal Study (CHARLS) plot suggests that, of urban resdents over 6, 79 and 54 percent of men and women, respectvely, have access to penson 5 Government Regulaton No 32 (979) specfes 55 as the retrement for cvl servants and became the benchmark for the normal retrement followed n the prvate sector as well. Mnstry of Labor Regulaton No 2 (995) stpulates that the normal retrement s 55 and the maxmum s 6, and these are also the s at whch elgble retrees are able to start collectng pensons. 6 McKee (26) fnds that n Indonesa, half of government workers move nto ether the prvate sector or nto selfemployment, and 6 percent of workers who leave ther prvate sector jobs move nto self-employment. In Korea, self-employment was one response to lay-offs from larger employers n the wake of the 997/8 fnancal crss (Sohn, 27), and contnues to be an mportant source of employment for older males (Lee 29). Lee and Lee (2) note dfferences n the retrement s of self-employed and w-salary earners, but do not dscuss the ncentve to move nto self-employment ahead of mandatory retrement among w and salary earners who wsh to contnue workng. 9

12 support. In urban areas of Indonesa and Korea, by contrast, fewer than 35 percent of men and 5 percent of women have access to pensons. 7 Another unque feature of penson elgblty n Korea les wth an mperfect correspondence between mandatory retrement and at whch penson-elgble retrees may start recevng penson benefts. The Natonal Penson scheme, the largest of the three man sources of pensons n Korea, does not begn payng benefts untl 6, yet a sgnfcant share of employees face mandatory retrement at 55 (Cho and Km, 25). Ths mperfect correspondence lkely ncreases ncentves for those employees facng mandatory retrement to look for new career opportuntes, ncludng n the self-employed sector. Across rural areas of Chna, Indonesa and Korea, both men and women reman actvely employed untl much later n ther lves than urban resdents. In rural areas of all three economes, agrcultural producton on the famly farm contnues to be a sgnfcant source of employment for older workers, and ths s necesstated by the fact that rural resdents tend to accumulate less wealth over ther workng lfetmes. In addton, older workers n rural Chna and Indonesa are far less lkely to have access to pensons than ther urban counterparts. Both the CHARLS and the 25 Populaton Census suggest that, of rural resdents over 6, roughly 5 percent of men and less than one percent of women have penson support. Smlarly, n rural Indonesa, the Indonesa Famly Lfe Survey (IFLS) shows that only 8 and 3 percent of men and women over 6 have access to pensons. Rural Korea s a somewhat dfferent matter. Efforts to brng rural resdents nto the Natonal Penson scheme establshed n 988 have led to penson cover rates n rural areas that do not dffer sgnfcantly from urban areas, ndeed 34 percent of older rural men n the 26 KLoSA wave report recevng pensons. Nonetheless the contnued labor supply of rural men s vewed as an mportant factor contrbutng to hgh rates of economc actvty among older Koreans (Lee 29). In spte of avalablty of pensons, levels of support are not suffcent to permt retrement of the rural elderly. Lee (29) suggests that the out-mgraton of the young has left elderly who reman behnd wth both a lack of young labor and nsuffcent wealth to cease productve 7 More specfcally, n Korea, 34 percent of urban men have access to pensons and 5 percent of urban women, and n Indonesa, 27 percent of urban men have access to pensons and percent of urban women. Fgures on penson cover are drawn from the 26 Wave of the Korean Longtudnal Study of Agng (KLoSA) and the 27 wave of the Indonesa Famly Lfe Survey (IFLS).

13 actvty. Gven that the hstory of rapd urbanzaton n Korea mrrors the process of rural to urban mgraton takng place n Chna, one mght be concerned that ncdence of delayed retrement of Chna s rural farmers, who are less affluent and have lacked penson support, may follow a smlar pattern to Korea. Phased or Gradual Retrement? The lterature on retrement n developed countres has emphaszed that retrement s often a gradual process: workers may reduce work hours at present employers, move to work arrangements requrng fewer hours per week or even transton n and out of retrement. 8 Usng the US Health and Retrement Study (HRS), Gustman and Stenmeer (2) fnd that twenty-two percent of the populaton reported beng partally retred at some pont, and that a ffth of the populaton had partally retred by 65. Exts from employment, or the labor force, are also not always permanent: Gustman and Stenmeer s study fnds that over four waves, 7 percent of the sample experenced a reversal n whch they moved from lessntensve work to more work. The partcular set of ncentves nfluencng the labor supply decson of older workers n urban Chna, both before and after reachng mandatory retrement, are lkely to lead to smlar patterns of gradual retrement as observed n other countres. In urban areas, low unemployment rates and shorts of sklled labor n a rapdly growng economy offer opportunty for older sklled workers. On the other hand, older long-term urban resdents who own resdences have experenced a sharp ncrease n ther wealth over the last decade (Park and Porter, 2), whch may rase reservaton ws and reduce the lkelhood of returnng to work. In order to gan a sense of the mportance of gradual, or phased retrement, we show aver hours of work per week across the dstrbuton n Fgure 3 below. The frst panel shows aver work hours uncondtonal on work status, wth the labor hours of respondents who are not workng equal to zero. In panel B, aver hours are reported only for those respondents who 8 Gustman and Stenmeer (984) found that self-reports of partal retrement among US men d 58 to 69 was qute common, partcularly partal retrement nto a job dfferent than the one held at 55. Blau (994) emphaszes that older workers transton n and out the labor force ( retrement ) wth consderable frequency and that these transtons are often not pcked up n annual data.

14 are workng. 9 We use the CHNS to contrast aver work hours n 99 wth 29, and also nclude data from the 28 CHARLS plot. From the uncondtonal plots n Panel A of Fgure 3, a gradual reducton n aver work hours after 45 s apparent for both men and women n urban areas. Also evdent s a steeper declne n work hours for women between 45 and 55 and for men between 55 and 65. Turnng our attenton to Panel B, a gradual declne n work hours s evdent for men and women who contnue workng beyond 55, but t s not steep. By 29, urban men between 55 and 6 who are employed, report spendng an aver of 45 hours per week workng. By 7, urban men who contnue n productve employment are spendng just 2 to 3 hours per week employed whle ther employed rural counterparts of the same are stll workng more than 4 hours per week. The dfference across urban and rural areas n the work-ntensty of those workng beyond mandatory retrement suggests that urban resdents may have the ablty to retre gradually, whle rural resdents, f they have the physcal capacty, may ndeed contnue to work full-tme nto ther 7s. 2 When comparng work-ntensty of older workers across countres, t s evdent that Chna s resdents over 6 work fewer hours than ther respectve urban and rural counterparts n Indonesa and Korea (Fgure 4). Condtonal on workng, aver workng hours for urban men and women (Fgure 4, Panel B) are more smlar to the UK, where hours drop off more quckly wth, than for Indonesa, Korea and the USA. Retrement n Chna, Indonesa and Korea The descrptve patterns shown above make use of data from early sts of longtudnal studes from Chna and Korea whch are modeled on the Health and Retrement Study (HRS) from the US and the Englsh Longtudnal Study on Agng (UK). New survey efforts, lke CHARLS n Chna, promse to facltate the study of retrement and labor supply behavor n regons where penson and socal securty systems are not well establshed, and populaton agng s occurrng at a rapd pace. In the analytcal models that we estmate, we make use of the cross-sectons from 9 As earler, we use non-parametrc locally weghted regresson (LOWESS) to smooth the avers across the dstrbuton. 2 Ths has long been true of Chna s rural elderly. A classc study of the rural elderly n the 6s and 7s referred to ther lves as one of ceaseless tol (Davs-Fredman, 99), and Pang et al (24) characterze the retrement decson n rural Chna as workng untl droppng. 2

15 Chna, Indonesa and Korea used for basc descrptve statstcs presented above. Summary statstcs for mportant varables used n the analyss can be found n Table. Below, we provde some addtonal dscusson of the analyss sample used for each country: Chna. The Chna Health and Retrement Longtudnal Study (CHARLS), drected by Yaohu Zhao at the Natonal School of Development at Pekng Unversty, was developed and mplemented by an accomplshed team of collaborators from Chna and abroad. The sample used n ths analyss s from the plot survey conducted n 28 n Gansu and Zhejang provnces, and s representatve of adults over 45 years of n these provnces. Gansu, n the Northwest s a relatvely poor provnce and Zhejang, on the coast south of Shangha, s relatvely affluent. CHARLS follows the model of the HRS, ELSA and KLoSA n the US, UK and Korea, and enumerates nformaton on respondent and spouse soco-economc status, health condtons, work and employment, ncomes and pensons, and also ncludes household and communty level nformaton. In the plot sample used n the analyss below, there are 57 households and 2685 elgble respondents. 2 Indonesa. The Indonesan Famly Lfe Survey (IFLS) s an ongong project conducted by RAND Corporaton and unverstes n both Indonesa and the US. To date, four waves have been conducted n 993, 997/98, 2, and 27/8. The survey covers ndvduals, households, communtes, and the communty-level health and educatonal facltes that respondent households access n ther daly lfe. Whle t s not a formal member of the HRS famly of agng surveys, the data are comprehensve and nclude nformaton on respondent health status, employment and labor allocaton, penson partcpaton and household and famly characterstcs, all of whch are crucal for study of retrement and labor supply decsons of the elderly. The orgnal survey was representatve of 83% of the Indonesan populaton and 3 of 27 provnces. As t tracks the movers and splt-offs from the orgnal sample, the IFLS was a poneer among household surveys n the developng world. Trackng s mportant when workng wth the 27/8 round of the IFLS, because t allows us to avod usng a selected sample of 2 The CHARLS project has fundng for natonally representatve surveys n 2 and 23, and feldwork on the frst natonal wave wll begn n July 2. Publc use data should be avalable for the research communty by June 22. 3

16 ndvduals who have not moved. 22 The most recent round has 5583 observatons and 357 households. The analytcal work below uses detaled nformaton collected from 7283 drect respondents d 45 and over for whom nformaton on penson elgblty was enumerated. Korea. The Korean Longtudnal Study of Agng (KLoSA) s a natonally representatve survey conducted by the Korea Labor Insttute to nvestgate the overall condtons of people d 45 or older n South Korea (Excludng Jeju sland). KLoSA belongs to the same famly of agng surveys as CHARLS, HRS and ELSA. The frst baselne survey was conducted at 26, and t was followed by a job hstory survey n 27. Whle KLoSA s a longtudnal survey, only the baselne survey s publcally avalable. In common wth CHARLS and ELSA, but dfferng from the HRS, the frst baselne survey doesn t cover the nsttutonalzed elderly. Overall, 67 households and 254 ndvduals partcpated n the baselne survey, and there are 3 observatons wth complete nformaton. A Model of the Employment of Older Adults Below, we examne determnants of the labor supply (employment) of older workers, recognzng that mportant correlates, such as access to pensons, may be systematcally related to both the ablty to retre and unobservable characterstcs (e.g., ablty). Whle these models should be vewed as provdng descrptve evdence, we choose measures of health status, and proxes for wealth and famly characterstcs wth the am of mnmzng endogenety bases. As a framework for understandng the retrement and labor supply decsons of the elderly, assume that ndvduals (or households) maxmze utlty subject to a famly budget constrant, whch s a functon of wealth, ncome, avalable tme, health status and non-labor ncome of household members. From the constraned utlty maxmzaton decson, we conceptualze a general model of labor supply: L S h w nw f W, I, I, H, T, X, V ) () ( j 22 Even though older resdents of both ths survey, as well as the CHNS n Chna, do not move or splt off wth the same frequency as household members who were under 25 durng the ntal wave, one mght nonetheless be concerned about ths potental source of bas. 4

17 where labor supply (or employment) of ndvdual, ncome from work of all household members, status, H, an ndvdual s tme endowment, characterstcs reflectng preferences, S h L, s a functon of household wealth, W, w I, ncome unrelated to current work, nw I, health T, a vector of ndvdual and household X whch nclude own and demographc characterstcs of household members. Gven lkely varaton n opportuntes and returns to labor across geographc regons (Jalan and Ravallon, 22), we control for potental unobserved county-level characterstcs affectng labor supply wth a vector of county (or cty) level ndcators, V j. Identfyng effects of each of these varables are complcated by three factors whch ntroduce bas nto our estmates: some are mperfectly observed, there are functonal relatonshps among mportant varables (e.g., health status may affect ncome through productvty, avalable tme, and avalable household wealth), and labor supply of an elderly ndvdual may be smultaneously determned wth the labor supply decsons of other famly members, partcularly a spouse. In order to reduce such bas, we estmate reduced form models wth proxes for longer-term determnants of (permanent) ncome and wealth. Frst, we use housng wealth as one proxy for household wealth ( W ). Unlke lqud components of wealth, housng wealth wll vary less wth current shocks to both ncome and health that smultaneously determne labor supply. 23 Second, current labor ncome ( I h w ) of a household wll also be systematcally related to own and famly member labor supply decsons. Recognzng that educatonal attanment of the elderly themselves s lkely to be assocated wth lfetme earnngs and accumulated wealth of the household, we nclude the respondents own educatonal attanment as an addtonal proxy for wealth. In addton, the aver educatonal attanment of a spouse and adult chldren are usd as a proxy for both the value of a spouse s labor and the value of potental support from adult chldren (whether or not they are n the household). Health status also affects productvty and ablty to earn ncome through ts mpact on capacty for work. Elderly who are ll or suffer physcal lmtatons may be unable to provde much labor for farm actvtes, and so we make use of self-assessments of ablty to perform actvtes of daly lvng (ADLs) as proxes for health status. In sum, our complete set of proxes for h W, w I 23 Vent and Wse (24) show that housng wealth s not treated as a lqud asset n the US. 5

18 and H n the reduced form are quadratcs n educatonal attanment and, measures of health status of the respondent (ADL z-scores, descrbed below), and the aver educatonal attanment of a spouse and adult chldren of the household. As some older workers may fnd that ther tme s best utlzed n provson of care to relatves and that ths may nfluence employment decsons, we also nclude numbers of grandchldren and lvng parents, respectvely, of the household head and spouse. The three data sources used to examne employment outcomes n Chna (CHARLS 28), Indonesa (IFLS 27) and Korea (KLOSA 26) provde roughly comparable nformaton on labor supply, health status and educaton of both the respondent and spouse, and nformaton on parents and chldren regardless of current resdence n the household. 24 We estmate the reduced form labor supply model n each country: L S E X γ V u (2) 2 E 2E 3Pen 4Pen ADLβ5 6 j where labor supply, S L, s a bnary ndcator of whether ndvdual worked for one hour or more durng the prevous week. We expect that hgher values of educatonal attanment of elderly, E, wll be assocated wth hgher wealth and savngs and, as lesure s a normal good, may be negatvely related to elderly labor supply. 25 Smlarly, we expect that access to own penson ncome, employment. Pen, and penson ncome of a spouse, Pen, wll be negatvely related to The health of older workers and elderly are measured usng two z-scores calculated from responses to a set of actvtes of daly lvng questons, ADL. 26 For each of the three data 24 An mportant beneft of knowng the numbers of grandchldren and lvng parents ndependent of whether they are members of the household s that we are able to control for potental demand for a famly care-provder wthout ntroducng bases related to endogenous household composton, as would occur f relyng on nformaton about household members alone. 25 Of course, an ndvdual wth more educaton may also be able to earn sgnfcantly hgher returns, and so the coeffcent on educaton wll reflect the net effect of returns and accumulated wealth on employment. 26 Bound (99) cautons that general health status questons are lkely to be correlated wth unobservable ndvdual characterstcs, and further, that they may suffer from justfcaton bas. Several studes (e.g. Bound, 999; Dwyer and Mtchell, 999) have suggested that proxes constructed from ADLs do not suffer from such serous bas. Bound et al (2) cautons that fnancal wealth may affect ADL outcomes, and that even proxes developed from ADLs may lead us to underestmate the negatve effects of poor health on labor supply. As we do 6

19 sources, we make use of questons related to actvtes of daly lvng to construct two ndces: one s based on the number of actvtes that the respondent may perform wth dffculty (Dffculty), and the second s based on the number of actvtes that a respondent cannot perform (Unable). Because each survey has some dfferences n actvtes and asks a dfferent number of actvtes, we use these responses to construct wthn country z-scores (respondent count aver count)/(standard devaton of count). 27 Increases n each of these two z-scores reflect declnng physcal ablty and worsenng general health status. A partcular value of the Unable- ADL z-score wll be assocated wth more severe lmtatons than the same value of the Dffculty-ADL z-score. We expect that declnng health wll have a negatve mpact on work actvty, partcularly for those workers n occupatons n whch physcal strength and moblty are mportant (Bound, 999). As the educaton of a spouse or adult chldren wll be related to addtonal sources of ncome (from spouse ncome or transfers from chldren), we nclude aver educaton of these famly members, E. Fnally, we control for a vector of ndvdual and household characterstcs, X, whch nclude and -squared and are assocated wth own productvty, numbers of grandchldren and lvng parents of the head and spouse, whch are assocated wth preferences for employment, and the ln(+per capta value of the household dwellng), as a proxy for household wealth. Wthn the retrement lterature n the US, recent research has focused on the mportant roles of spouse employment and spouse health status n labor supply and retrement decsons. Frst, the retrement decsons of husbands and wves may be nterdependent. Structural models (Blau, 996; Gustman and Stenmeer, 24) suggest that labor supply decsons of older couples reflect preferences for shared retrement. Second, work decsons may be affected by the health status of a spouse. One may plausbly observe an added worker effect, n whch a spouse s health shock leads to ncreased labor supply so as to nsure ncome aganst the earnngs-loss assocated wth the health shock (e.g., Cole, 24a), or alternatvely fnd that spouse care requrements wll not yet have approprate panel data for the three countres n ths study, we are not able to control for dynamc relatonshps between health and wealth. 27 CHARLS asks 9, IFLS asks 8 and KLoSa asks 7. Each of the surveys asks questons related to functonal abltes such as: ablty to walk for a klometer, ablty to go up and down stars, dffcultes lftng or 2 lbs, draw water from a well, ablty to perform household chores, bath oneself or take medcaton. 7

20 requre ext from the labor force (e.g., McGeary, 29). To gauge the potental mportance of these factors n Chna, Indonesa and Korea we next estmate model (3) below: L S 2 E 2E 3Pen 4Pen ADLβ5 6E ADL β7 8 L S X γ V j u (3) where S L s an ndcator or whether or not a spouse s employed, and ADL are measures of spouse health status. As the labor supply decsons of husbands and wves are lkely to be determned jontly and have a dynamc relatonshp wth health and shocks to health and employment, we vew these models as purely descrptve but nformatve of the extent to whch jont labor supply decsons may affect the tmng of retrement. As the HRS-type surveys mature nto full panel studes, t wll be feasble to control for unobservables and to unlock drectons of causalty among these varables. 28 Determnants of Retrement n Chna, Indonesa and Korea Employment and Penson Avalablty. Descrptve patterns of employment n productve actvty across urban and rural areas of Chna, Indonesa and Korea (Fgure 5) are suggestve of the role that elgblty for a penson may play n retrement of older workers and elderly. In these fgures, the lne shows the share of populaton n two-year cohorts that s employed n some productve actvty (salared work, nformal sector w labor, casual work, self-employed actvtes or unpad employment n famly-run enterprses) and the bars represent the penson elgble share. In urban Chna, the employed share decreases sharply wth ncreases n the penson-elgble share. In rural areas of Chna, one does not observe a smlar declne n employment, and shares of workers wth access to pensons are qute low. In Indonesa and Korea, where penson cover rates at young s are lower than for urban Chna, one observes 28 In estmatng (2) and (3), t should be noted that for each country there are subsamples of the populaton for whch there s no spouse or for whch spouse nformaton s not avalable, and ths may lead us to concerns about bases ntroduced by selecton nto marr. In the models presented n ths paper, we have handled ths problem by ncludng ndcator varables for martal status and for absence of spouse nformaton, and then set spouse nformaton to zero for those observatons n whch no data on a spouse are avalable (for ADL z-scores, of course, ths corresponds to the mean). We have also estmated these models on the subset of the data for whch we have nformaton on both spouses. We take comfort n the fact that there s no apprecable dfference n the coeffcents of nterest across models usng the full sample and the one estmated on marred couples wth spouse nformaton. 8

21 a much more gradual declne n the employment of men n urban areas, and a far less pronounced declne n rural areas of both countres. In Table 2, below, we show results from estmatng the labor supply model presented n specfcaton (2) above. 29 The results are suggestve of the role that penson ncome plays n decsons to ext from employment and the labor force. After controllng for, educaton, health status of respondent and spouse, proxes for wealth and famly demographc characterstcs, men and women n urban Chna are 5 and 8.3 percent less lkely to be workng f elgble for a penson. Among the rural populaton, we observe a margnally sgnfcant negatve relatonshp between penson elgblty and employment for men, but no effect on employment of women. In Indonesa, where cover by pensons s most common among retred cvl servants, access to pensons s also negatvely assocated wth work actvty for both genders n rural and urban areas. In Korea, by contrast, access to a penson only affects the employment status of men n urban areas. Ths may not be surprsng f penson cover or ncome s partcularly low n Korea, or s earned by wealther households whose labor force partcpaton s affected by other factors. Lee (29) has noted that although cover expanson to rural areas was an objectve of the Natonal Penson Scheme after 988, wealth dstrbuton n rural areas remans skewed and that the majorty of rural households lacked suffcent penson ncome or wealth to retre. Ablty to retre, of course, not only depends on penson access, but on whether a household has suffcent accumulated wealth to allow for retrement. A respondent s years of schoolng wll be assocated both wth lfetme wealth, and wth potental current returns avalable n the labor market. In the models reported n Table 2, we observe a non-lnear relatonshp between years of educaton for urban and rural women and ther probablty of workng. For urban Chnese women, probablty of workng declnes wth each addtonal year of schoolng through completon of mddle school, and then starts to ncrease wth hgher levels of educaton. We also observe that ncreases n housng wealth per capta n urban Chna are assocated wth reduced probablty of employment. In rural Chna, the probablty of male and female employment declnes wth educaton through eleven years of schoolng. As educatonal attanment s not very 29 We present lnear probablty models by gender and resdence locaton (urban and rural). Whle magntudes dffer somewhat, margnal effects usng probt models do not lead to sgnfcant qualtatve dfferences n results. 9